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 | ||
dd56b046 MG |
76 | #include "internal.h" |
77 | ||
09cbfeaf KS |
78 | #define BLOCKS_PER_PAGE (PAGE_SIZE/512) |
79 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT) | |
1da177e4 | 80 | |
1da177e4 LT |
81 | /* Pretend that each entry is of this size in directory's i_size */ |
82 | #define BOGO_DIRENT_SIZE 20 | |
83 | ||
69f07ec9 HD |
84 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
85 | #define SHORT_SYMLINK_LEN 128 | |
86 | ||
1aac1400 | 87 | /* |
f00cdc6d HD |
88 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
89 | * inode->i_private (with i_mutex making sure that it has only one user at | |
90 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
91 | */ |
92 | struct shmem_falloc { | |
8e205f77 | 93 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
94 | pgoff_t start; /* start of range currently being fallocated */ |
95 | pgoff_t next; /* the next page offset to be fallocated */ | |
96 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
97 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
98 | }; | |
99 | ||
285b2c4f | 100 | /* Flag allocation requirements to shmem_getpage */ |
1da177e4 | 101 | enum sgp_type { |
1da177e4 LT |
102 | SGP_READ, /* don't exceed i_size, don't allocate page */ |
103 | SGP_CACHE, /* don't exceed i_size, may allocate page */ | |
1635f6a7 HD |
104 | SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */ |
105 | SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */ | |
1da177e4 LT |
106 | }; |
107 | ||
b76db735 | 108 | #ifdef CONFIG_TMPFS |
680d794b | 109 | static unsigned long shmem_default_max_blocks(void) |
110 | { | |
111 | return totalram_pages / 2; | |
112 | } | |
113 | ||
114 | static unsigned long shmem_default_max_inodes(void) | |
115 | { | |
116 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
117 | } | |
b76db735 | 118 | #endif |
680d794b | 119 | |
bde05d1c HD |
120 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
121 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
122 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 | 123 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 ALC |
124 | struct page **pagep, enum sgp_type sgp, |
125 | gfp_t gfp, struct mm_struct *fault_mm, int *fault_type); | |
68da9f05 HD |
126 | |
127 | static inline int shmem_getpage(struct inode *inode, pgoff_t index, | |
9e18eb29 | 128 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
129 | { |
130 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
9e18eb29 | 131 | mapping_gfp_mask(inode->i_mapping), NULL, NULL); |
68da9f05 | 132 | } |
1da177e4 | 133 | |
1da177e4 LT |
134 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
135 | { | |
136 | return sb->s_fs_info; | |
137 | } | |
138 | ||
139 | /* | |
140 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
141 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
142 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
143 | * consistent with the pre-accounting of private mappings ... | |
144 | */ | |
145 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
146 | { | |
0b0a0806 | 147 | return (flags & VM_NORESERVE) ? |
191c5424 | 148 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
149 | } |
150 | ||
151 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
152 | { | |
0b0a0806 | 153 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
154 | vm_unacct_memory(VM_ACCT(size)); |
155 | } | |
156 | ||
77142517 KK |
157 | static inline int shmem_reacct_size(unsigned long flags, |
158 | loff_t oldsize, loff_t newsize) | |
159 | { | |
160 | if (!(flags & VM_NORESERVE)) { | |
161 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
162 | return security_vm_enough_memory_mm(current->mm, | |
163 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
164 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
165 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
166 | } | |
167 | return 0; | |
168 | } | |
169 | ||
1da177e4 LT |
170 | /* |
171 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 172 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
173 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
174 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
175 | */ | |
800d8c63 | 176 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 177 | { |
800d8c63 KS |
178 | if (!(flags & VM_NORESERVE)) |
179 | return 0; | |
180 | ||
181 | return security_vm_enough_memory_mm(current->mm, | |
182 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
183 | } |
184 | ||
185 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
186 | { | |
0b0a0806 | 187 | if (flags & VM_NORESERVE) |
09cbfeaf | 188 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
189 | } |
190 | ||
759b9775 | 191 | static const struct super_operations shmem_ops; |
f5e54d6e | 192 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 193 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
194 | static const struct inode_operations shmem_inode_operations; |
195 | static const struct inode_operations shmem_dir_inode_operations; | |
196 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 197 | static const struct vm_operations_struct shmem_vm_ops; |
1da177e4 | 198 | |
1da177e4 | 199 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 200 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 201 | |
5b04c689 PE |
202 | static int shmem_reserve_inode(struct super_block *sb) |
203 | { | |
204 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
205 | if (sbinfo->max_inodes) { | |
206 | spin_lock(&sbinfo->stat_lock); | |
207 | if (!sbinfo->free_inodes) { | |
208 | spin_unlock(&sbinfo->stat_lock); | |
209 | return -ENOSPC; | |
210 | } | |
211 | sbinfo->free_inodes--; | |
212 | spin_unlock(&sbinfo->stat_lock); | |
213 | } | |
214 | return 0; | |
215 | } | |
216 | ||
217 | static void shmem_free_inode(struct super_block *sb) | |
218 | { | |
219 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
220 | if (sbinfo->max_inodes) { | |
221 | spin_lock(&sbinfo->stat_lock); | |
222 | sbinfo->free_inodes++; | |
223 | spin_unlock(&sbinfo->stat_lock); | |
224 | } | |
225 | } | |
226 | ||
46711810 | 227 | /** |
41ffe5d5 | 228 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
229 | * @inode: inode to recalc |
230 | * | |
231 | * We have to calculate the free blocks since the mm can drop | |
232 | * undirtied hole pages behind our back. | |
233 | * | |
234 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
235 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
236 | * | |
237 | * It has to be called with the spinlock held. | |
238 | */ | |
239 | static void shmem_recalc_inode(struct inode *inode) | |
240 | { | |
241 | struct shmem_inode_info *info = SHMEM_I(inode); | |
242 | long freed; | |
243 | ||
244 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
245 | if (freed > 0) { | |
54af6042 HD |
246 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
247 | if (sbinfo->max_blocks) | |
248 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 249 | info->alloced -= freed; |
54af6042 | 250 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 251 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
252 | } |
253 | } | |
254 | ||
800d8c63 KS |
255 | bool shmem_charge(struct inode *inode, long pages) |
256 | { | |
257 | struct shmem_inode_info *info = SHMEM_I(inode); | |
258 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
259 | ||
260 | if (shmem_acct_block(info->flags, pages)) | |
261 | return false; | |
262 | spin_lock(&info->lock); | |
263 | info->alloced += pages; | |
264 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
265 | shmem_recalc_inode(inode); | |
266 | spin_unlock(&info->lock); | |
267 | inode->i_mapping->nrpages += pages; | |
268 | ||
269 | if (!sbinfo->max_blocks) | |
270 | return true; | |
271 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
272 | sbinfo->max_blocks - pages) > 0) { | |
273 | inode->i_mapping->nrpages -= pages; | |
274 | spin_lock(&info->lock); | |
275 | info->alloced -= pages; | |
276 | shmem_recalc_inode(inode); | |
277 | spin_unlock(&info->lock); | |
278 | ||
279 | return false; | |
280 | } | |
281 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
282 | return true; | |
283 | } | |
284 | ||
285 | void shmem_uncharge(struct inode *inode, long pages) | |
286 | { | |
287 | struct shmem_inode_info *info = SHMEM_I(inode); | |
288 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
289 | ||
290 | spin_lock(&info->lock); | |
291 | info->alloced -= pages; | |
292 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
293 | shmem_recalc_inode(inode); | |
294 | spin_unlock(&info->lock); | |
295 | ||
296 | if (sbinfo->max_blocks) | |
297 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
298 | } | |
299 | ||
7a5d0fbb HD |
300 | /* |
301 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
302 | */ | |
303 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
304 | pgoff_t index, void *expected, void *replacement) | |
305 | { | |
306 | void **pslot; | |
6dbaf22c | 307 | void *item; |
7a5d0fbb HD |
308 | |
309 | VM_BUG_ON(!expected); | |
6dbaf22c | 310 | VM_BUG_ON(!replacement); |
7a5d0fbb | 311 | pslot = radix_tree_lookup_slot(&mapping->page_tree, index); |
6dbaf22c JW |
312 | if (!pslot) |
313 | return -ENOENT; | |
314 | item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock); | |
7a5d0fbb HD |
315 | if (item != expected) |
316 | return -ENOENT; | |
6dbaf22c | 317 | radix_tree_replace_slot(pslot, replacement); |
7a5d0fbb HD |
318 | return 0; |
319 | } | |
320 | ||
d1899228 HD |
321 | /* |
322 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
323 | * that an entry was not already brought back from swap by a racing thread. | |
324 | * | |
325 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
326 | * might be reused, and again be SwapCache, using the same swap as before. | |
327 | */ | |
328 | static bool shmem_confirm_swap(struct address_space *mapping, | |
329 | pgoff_t index, swp_entry_t swap) | |
330 | { | |
331 | void *item; | |
332 | ||
333 | rcu_read_lock(); | |
334 | item = radix_tree_lookup(&mapping->page_tree, index); | |
335 | rcu_read_unlock(); | |
336 | return item == swp_to_radix_entry(swap); | |
337 | } | |
338 | ||
5a6e75f8 KS |
339 | /* |
340 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
341 | * | |
342 | * SHMEM_HUGE_NEVER: | |
343 | * disables huge pages for the mount; | |
344 | * SHMEM_HUGE_ALWAYS: | |
345 | * enables huge pages for the mount; | |
346 | * SHMEM_HUGE_WITHIN_SIZE: | |
347 | * only allocate huge pages if the page will be fully within i_size, | |
348 | * also respect fadvise()/madvise() hints; | |
349 | * SHMEM_HUGE_ADVISE: | |
350 | * only allocate huge pages if requested with fadvise()/madvise(); | |
351 | */ | |
352 | ||
353 | #define SHMEM_HUGE_NEVER 0 | |
354 | #define SHMEM_HUGE_ALWAYS 1 | |
355 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
356 | #define SHMEM_HUGE_ADVISE 3 | |
357 | ||
358 | /* | |
359 | * Special values. | |
360 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
361 | * | |
362 | * SHMEM_HUGE_DENY: | |
363 | * disables huge on shm_mnt and all mounts, for emergency use; | |
364 | * SHMEM_HUGE_FORCE: | |
365 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
366 | * | |
367 | */ | |
368 | #define SHMEM_HUGE_DENY (-1) | |
369 | #define SHMEM_HUGE_FORCE (-2) | |
370 | ||
371 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
372 | /* ifdef here to avoid bloating shmem.o when not necessary */ | |
373 | ||
374 | int shmem_huge __read_mostly; | |
375 | ||
376 | static int shmem_parse_huge(const char *str) | |
377 | { | |
378 | if (!strcmp(str, "never")) | |
379 | return SHMEM_HUGE_NEVER; | |
380 | if (!strcmp(str, "always")) | |
381 | return SHMEM_HUGE_ALWAYS; | |
382 | if (!strcmp(str, "within_size")) | |
383 | return SHMEM_HUGE_WITHIN_SIZE; | |
384 | if (!strcmp(str, "advise")) | |
385 | return SHMEM_HUGE_ADVISE; | |
386 | if (!strcmp(str, "deny")) | |
387 | return SHMEM_HUGE_DENY; | |
388 | if (!strcmp(str, "force")) | |
389 | return SHMEM_HUGE_FORCE; | |
390 | return -EINVAL; | |
391 | } | |
392 | ||
393 | static const char *shmem_format_huge(int huge) | |
394 | { | |
395 | switch (huge) { | |
396 | case SHMEM_HUGE_NEVER: | |
397 | return "never"; | |
398 | case SHMEM_HUGE_ALWAYS: | |
399 | return "always"; | |
400 | case SHMEM_HUGE_WITHIN_SIZE: | |
401 | return "within_size"; | |
402 | case SHMEM_HUGE_ADVISE: | |
403 | return "advise"; | |
404 | case SHMEM_HUGE_DENY: | |
405 | return "deny"; | |
406 | case SHMEM_HUGE_FORCE: | |
407 | return "force"; | |
408 | default: | |
409 | VM_BUG_ON(1); | |
410 | return "bad_val"; | |
411 | } | |
412 | } | |
413 | ||
414 | #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ | |
415 | ||
416 | #define shmem_huge SHMEM_HUGE_DENY | |
417 | ||
418 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
419 | ||
46f65ec1 HD |
420 | /* |
421 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
422 | */ | |
423 | static int shmem_add_to_page_cache(struct page *page, | |
424 | struct address_space *mapping, | |
fed400a1 | 425 | pgoff_t index, void *expected) |
46f65ec1 | 426 | { |
800d8c63 | 427 | int error, nr = hpage_nr_pages(page); |
46f65ec1 | 428 | |
800d8c63 KS |
429 | VM_BUG_ON_PAGE(PageTail(page), page); |
430 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
431 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
432 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 433 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 434 | |
800d8c63 | 435 | page_ref_add(page, nr); |
b065b432 HD |
436 | page->mapping = mapping; |
437 | page->index = index; | |
438 | ||
439 | spin_lock_irq(&mapping->tree_lock); | |
800d8c63 KS |
440 | if (PageTransHuge(page)) { |
441 | void __rcu **results; | |
442 | pgoff_t idx; | |
443 | int i; | |
444 | ||
445 | error = 0; | |
446 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, | |
447 | &results, &idx, index, 1) && | |
448 | idx < index + HPAGE_PMD_NR) { | |
449 | error = -EEXIST; | |
450 | } | |
451 | ||
452 | if (!error) { | |
453 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
454 | error = radix_tree_insert(&mapping->page_tree, | |
455 | index + i, page + i); | |
456 | VM_BUG_ON(error); | |
457 | } | |
458 | count_vm_event(THP_FILE_ALLOC); | |
459 | } | |
460 | } else if (!expected) { | |
b065b432 | 461 | error = radix_tree_insert(&mapping->page_tree, index, page); |
800d8c63 | 462 | } else { |
b065b432 HD |
463 | error = shmem_radix_tree_replace(mapping, index, expected, |
464 | page); | |
800d8c63 KS |
465 | } |
466 | ||
46f65ec1 | 467 | if (!error) { |
800d8c63 KS |
468 | mapping->nrpages += nr; |
469 | if (PageTransHuge(page)) | |
470 | __inc_zone_page_state(page, NR_SHMEM_THPS); | |
471 | __mod_zone_page_state(page_zone(page), NR_FILE_PAGES, nr); | |
472 | __mod_zone_page_state(page_zone(page), NR_SHMEM, nr); | |
b065b432 HD |
473 | spin_unlock_irq(&mapping->tree_lock); |
474 | } else { | |
475 | page->mapping = NULL; | |
476 | spin_unlock_irq(&mapping->tree_lock); | |
800d8c63 | 477 | page_ref_sub(page, nr); |
46f65ec1 | 478 | } |
46f65ec1 HD |
479 | return error; |
480 | } | |
481 | ||
6922c0c7 HD |
482 | /* |
483 | * Like delete_from_page_cache, but substitutes swap for page. | |
484 | */ | |
485 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
486 | { | |
487 | struct address_space *mapping = page->mapping; | |
488 | int error; | |
489 | ||
800d8c63 KS |
490 | VM_BUG_ON_PAGE(PageCompound(page), page); |
491 | ||
6922c0c7 HD |
492 | spin_lock_irq(&mapping->tree_lock); |
493 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
494 | page->mapping = NULL; | |
495 | mapping->nrpages--; | |
496 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
497 | __dec_zone_page_state(page, NR_SHMEM); | |
498 | spin_unlock_irq(&mapping->tree_lock); | |
09cbfeaf | 499 | put_page(page); |
6922c0c7 HD |
500 | BUG_ON(error); |
501 | } | |
502 | ||
7a5d0fbb HD |
503 | /* |
504 | * Remove swap entry from radix tree, free the swap and its page cache. | |
505 | */ | |
506 | static int shmem_free_swap(struct address_space *mapping, | |
507 | pgoff_t index, void *radswap) | |
508 | { | |
6dbaf22c | 509 | void *old; |
7a5d0fbb HD |
510 | |
511 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 512 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 513 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
514 | if (old != radswap) |
515 | return -ENOENT; | |
516 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
517 | return 0; | |
7a5d0fbb HD |
518 | } |
519 | ||
6a15a370 VB |
520 | /* |
521 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 522 | * given offsets are swapped out. |
6a15a370 VB |
523 | * |
524 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
525 | * as long as the inode doesn't go away and racy results are not a problem. | |
526 | */ | |
48131e03 VB |
527 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
528 | pgoff_t start, pgoff_t end) | |
6a15a370 | 529 | { |
6a15a370 VB |
530 | struct radix_tree_iter iter; |
531 | void **slot; | |
532 | struct page *page; | |
48131e03 | 533 | unsigned long swapped = 0; |
6a15a370 VB |
534 | |
535 | rcu_read_lock(); | |
536 | ||
6a15a370 VB |
537 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
538 | if (iter.index >= end) | |
539 | break; | |
540 | ||
541 | page = radix_tree_deref_slot(slot); | |
542 | ||
2cf938aa MW |
543 | if (radix_tree_deref_retry(page)) { |
544 | slot = radix_tree_iter_retry(&iter); | |
545 | continue; | |
546 | } | |
6a15a370 VB |
547 | |
548 | if (radix_tree_exceptional_entry(page)) | |
549 | swapped++; | |
550 | ||
551 | if (need_resched()) { | |
552 | cond_resched_rcu(); | |
7165092f | 553 | slot = radix_tree_iter_next(&iter); |
6a15a370 VB |
554 | } |
555 | } | |
556 | ||
557 | rcu_read_unlock(); | |
558 | ||
559 | return swapped << PAGE_SHIFT; | |
560 | } | |
561 | ||
48131e03 VB |
562 | /* |
563 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
564 | * given vma is swapped out. | |
565 | * | |
566 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
567 | * as long as the inode doesn't go away and racy results are not a problem. | |
568 | */ | |
569 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
570 | { | |
571 | struct inode *inode = file_inode(vma->vm_file); | |
572 | struct shmem_inode_info *info = SHMEM_I(inode); | |
573 | struct address_space *mapping = inode->i_mapping; | |
574 | unsigned long swapped; | |
575 | ||
576 | /* Be careful as we don't hold info->lock */ | |
577 | swapped = READ_ONCE(info->swapped); | |
578 | ||
579 | /* | |
580 | * The easier cases are when the shmem object has nothing in swap, or | |
581 | * the vma maps it whole. Then we can simply use the stats that we | |
582 | * already track. | |
583 | */ | |
584 | if (!swapped) | |
585 | return 0; | |
586 | ||
587 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
588 | return swapped << PAGE_SHIFT; | |
589 | ||
590 | /* Here comes the more involved part */ | |
591 | return shmem_partial_swap_usage(mapping, | |
592 | linear_page_index(vma, vma->vm_start), | |
593 | linear_page_index(vma, vma->vm_end)); | |
594 | } | |
595 | ||
24513264 HD |
596 | /* |
597 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
598 | */ | |
599 | void shmem_unlock_mapping(struct address_space *mapping) | |
600 | { | |
601 | struct pagevec pvec; | |
602 | pgoff_t indices[PAGEVEC_SIZE]; | |
603 | pgoff_t index = 0; | |
604 | ||
605 | pagevec_init(&pvec, 0); | |
606 | /* | |
607 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
608 | */ | |
609 | while (!mapping_unevictable(mapping)) { | |
610 | /* | |
611 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
612 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
613 | */ | |
0cd6144a JW |
614 | pvec.nr = find_get_entries(mapping, index, |
615 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
616 | if (!pvec.nr) |
617 | break; | |
618 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 619 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
620 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
621 | pagevec_release(&pvec); | |
622 | cond_resched(); | |
623 | } | |
7a5d0fbb HD |
624 | } |
625 | ||
626 | /* | |
627 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 628 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 629 | */ |
1635f6a7 HD |
630 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
631 | bool unfalloc) | |
1da177e4 | 632 | { |
285b2c4f | 633 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 634 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
635 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
636 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
637 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
638 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 639 | struct pagevec pvec; |
7a5d0fbb HD |
640 | pgoff_t indices[PAGEVEC_SIZE]; |
641 | long nr_swaps_freed = 0; | |
285b2c4f | 642 | pgoff_t index; |
bda97eab HD |
643 | int i; |
644 | ||
83e4fa9c HD |
645 | if (lend == -1) |
646 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
647 | |
648 | pagevec_init(&pvec, 0); | |
649 | index = start; | |
83e4fa9c | 650 | while (index < end) { |
0cd6144a JW |
651 | pvec.nr = find_get_entries(mapping, index, |
652 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
653 | pvec.pages, indices); | |
7a5d0fbb HD |
654 | if (!pvec.nr) |
655 | break; | |
bda97eab HD |
656 | for (i = 0; i < pagevec_count(&pvec); i++) { |
657 | struct page *page = pvec.pages[i]; | |
658 | ||
7a5d0fbb | 659 | index = indices[i]; |
83e4fa9c | 660 | if (index >= end) |
bda97eab HD |
661 | break; |
662 | ||
7a5d0fbb | 663 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
664 | if (unfalloc) |
665 | continue; | |
7a5d0fbb HD |
666 | nr_swaps_freed += !shmem_free_swap(mapping, |
667 | index, page); | |
bda97eab | 668 | continue; |
7a5d0fbb HD |
669 | } |
670 | ||
800d8c63 KS |
671 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
672 | ||
7a5d0fbb | 673 | if (!trylock_page(page)) |
bda97eab | 674 | continue; |
800d8c63 KS |
675 | |
676 | if (PageTransTail(page)) { | |
677 | /* Middle of THP: zero out the page */ | |
678 | clear_highpage(page); | |
679 | unlock_page(page); | |
680 | continue; | |
681 | } else if (PageTransHuge(page)) { | |
682 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
683 | /* | |
684 | * Range ends in the middle of THP: | |
685 | * zero out the page | |
686 | */ | |
687 | clear_highpage(page); | |
688 | unlock_page(page); | |
689 | continue; | |
690 | } | |
691 | index += HPAGE_PMD_NR - 1; | |
692 | i += HPAGE_PMD_NR - 1; | |
693 | } | |
694 | ||
1635f6a7 | 695 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
696 | VM_BUG_ON_PAGE(PageTail(page), page); |
697 | if (page_mapping(page) == mapping) { | |
309381fe | 698 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
699 | truncate_inode_page(mapping, page); |
700 | } | |
bda97eab | 701 | } |
bda97eab HD |
702 | unlock_page(page); |
703 | } | |
0cd6144a | 704 | pagevec_remove_exceptionals(&pvec); |
24513264 | 705 | pagevec_release(&pvec); |
bda97eab HD |
706 | cond_resched(); |
707 | index++; | |
708 | } | |
1da177e4 | 709 | |
83e4fa9c | 710 | if (partial_start) { |
bda97eab | 711 | struct page *page = NULL; |
9e18eb29 | 712 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 713 | if (page) { |
09cbfeaf | 714 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
715 | if (start > end) { |
716 | top = partial_end; | |
717 | partial_end = 0; | |
718 | } | |
719 | zero_user_segment(page, partial_start, top); | |
720 | set_page_dirty(page); | |
721 | unlock_page(page); | |
09cbfeaf | 722 | put_page(page); |
83e4fa9c HD |
723 | } |
724 | } | |
725 | if (partial_end) { | |
726 | struct page *page = NULL; | |
9e18eb29 | 727 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
728 | if (page) { |
729 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
730 | set_page_dirty(page); |
731 | unlock_page(page); | |
09cbfeaf | 732 | put_page(page); |
bda97eab HD |
733 | } |
734 | } | |
83e4fa9c HD |
735 | if (start >= end) |
736 | return; | |
bda97eab HD |
737 | |
738 | index = start; | |
b1a36650 | 739 | while (index < end) { |
bda97eab | 740 | cond_resched(); |
0cd6144a JW |
741 | |
742 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 743 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 744 | pvec.pages, indices); |
7a5d0fbb | 745 | if (!pvec.nr) { |
b1a36650 HD |
746 | /* If all gone or hole-punch or unfalloc, we're done */ |
747 | if (index == start || end != -1) | |
bda97eab | 748 | break; |
b1a36650 | 749 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
750 | index = start; |
751 | continue; | |
752 | } | |
bda97eab HD |
753 | for (i = 0; i < pagevec_count(&pvec); i++) { |
754 | struct page *page = pvec.pages[i]; | |
755 | ||
7a5d0fbb | 756 | index = indices[i]; |
83e4fa9c | 757 | if (index >= end) |
bda97eab HD |
758 | break; |
759 | ||
7a5d0fbb | 760 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
761 | if (unfalloc) |
762 | continue; | |
b1a36650 HD |
763 | if (shmem_free_swap(mapping, index, page)) { |
764 | /* Swap was replaced by page: retry */ | |
765 | index--; | |
766 | break; | |
767 | } | |
768 | nr_swaps_freed++; | |
7a5d0fbb HD |
769 | continue; |
770 | } | |
771 | ||
bda97eab | 772 | lock_page(page); |
800d8c63 KS |
773 | |
774 | if (PageTransTail(page)) { | |
775 | /* Middle of THP: zero out the page */ | |
776 | clear_highpage(page); | |
777 | unlock_page(page); | |
778 | /* | |
779 | * Partial thp truncate due 'start' in middle | |
780 | * of THP: don't need to look on these pages | |
781 | * again on !pvec.nr restart. | |
782 | */ | |
783 | if (index != round_down(end, HPAGE_PMD_NR)) | |
784 | start++; | |
785 | continue; | |
786 | } else if (PageTransHuge(page)) { | |
787 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
788 | /* | |
789 | * Range ends in the middle of THP: | |
790 | * zero out the page | |
791 | */ | |
792 | clear_highpage(page); | |
793 | unlock_page(page); | |
794 | continue; | |
795 | } | |
796 | index += HPAGE_PMD_NR - 1; | |
797 | i += HPAGE_PMD_NR - 1; | |
798 | } | |
799 | ||
1635f6a7 | 800 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
801 | VM_BUG_ON_PAGE(PageTail(page), page); |
802 | if (page_mapping(page) == mapping) { | |
309381fe | 803 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 804 | truncate_inode_page(mapping, page); |
b1a36650 HD |
805 | } else { |
806 | /* Page was replaced by swap: retry */ | |
807 | unlock_page(page); | |
808 | index--; | |
809 | break; | |
1635f6a7 | 810 | } |
7a5d0fbb | 811 | } |
bda97eab HD |
812 | unlock_page(page); |
813 | } | |
0cd6144a | 814 | pagevec_remove_exceptionals(&pvec); |
24513264 | 815 | pagevec_release(&pvec); |
bda97eab HD |
816 | index++; |
817 | } | |
94c1e62d | 818 | |
1da177e4 | 819 | spin_lock(&info->lock); |
7a5d0fbb | 820 | info->swapped -= nr_swaps_freed; |
1da177e4 LT |
821 | shmem_recalc_inode(inode); |
822 | spin_unlock(&info->lock); | |
1635f6a7 | 823 | } |
1da177e4 | 824 | |
1635f6a7 HD |
825 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
826 | { | |
827 | shmem_undo_range(inode, lstart, lend, false); | |
285b2c4f | 828 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1da177e4 | 829 | } |
94c1e62d | 830 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 831 | |
44a30220 YZ |
832 | static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry, |
833 | struct kstat *stat) | |
834 | { | |
835 | struct inode *inode = dentry->d_inode; | |
836 | struct shmem_inode_info *info = SHMEM_I(inode); | |
837 | ||
d0424c42 HD |
838 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
839 | spin_lock(&info->lock); | |
840 | shmem_recalc_inode(inode); | |
841 | spin_unlock(&info->lock); | |
842 | } | |
44a30220 | 843 | generic_fillattr(inode, stat); |
44a30220 YZ |
844 | return 0; |
845 | } | |
846 | ||
94c1e62d | 847 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 848 | { |
75c3cfa8 | 849 | struct inode *inode = d_inode(dentry); |
40e041a2 | 850 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 LT |
851 | int error; |
852 | ||
db78b877 CH |
853 | error = inode_change_ok(inode, attr); |
854 | if (error) | |
855 | return error; | |
856 | ||
94c1e62d HD |
857 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
858 | loff_t oldsize = inode->i_size; | |
859 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 860 | |
40e041a2 DH |
861 | /* protected by i_mutex */ |
862 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
863 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
864 | return -EPERM; | |
865 | ||
94c1e62d | 866 | if (newsize != oldsize) { |
77142517 KK |
867 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
868 | oldsize, newsize); | |
869 | if (error) | |
870 | return error; | |
94c1e62d HD |
871 | i_size_write(inode, newsize); |
872 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
873 | } | |
afa2db2f | 874 | if (newsize <= oldsize) { |
94c1e62d | 875 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
876 | if (oldsize > holebegin) |
877 | unmap_mapping_range(inode->i_mapping, | |
878 | holebegin, 0, 1); | |
879 | if (info->alloced) | |
880 | shmem_truncate_range(inode, | |
881 | newsize, (loff_t)-1); | |
94c1e62d | 882 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
883 | if (oldsize > holebegin) |
884 | unmap_mapping_range(inode->i_mapping, | |
885 | holebegin, 0, 1); | |
94c1e62d | 886 | } |
1da177e4 LT |
887 | } |
888 | ||
db78b877 | 889 | setattr_copy(inode, attr); |
db78b877 | 890 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 891 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
892 | return error; |
893 | } | |
894 | ||
1f895f75 | 895 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 896 | { |
1da177e4 LT |
897 | struct shmem_inode_info *info = SHMEM_I(inode); |
898 | ||
3889e6e7 | 899 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
900 | shmem_unacct_size(info->flags, inode->i_size); |
901 | inode->i_size = 0; | |
3889e6e7 | 902 | shmem_truncate_range(inode, 0, (loff_t)-1); |
1da177e4 | 903 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 904 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 905 | list_del_init(&info->swaplist); |
cb5f7b9a | 906 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 907 | } |
3ed47db3 | 908 | } |
b09e0fa4 | 909 | |
38f38657 | 910 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 911 | WARN_ON(inode->i_blocks); |
5b04c689 | 912 | shmem_free_inode(inode->i_sb); |
dbd5768f | 913 | clear_inode(inode); |
1da177e4 LT |
914 | } |
915 | ||
46f65ec1 HD |
916 | /* |
917 | * If swap found in inode, free it and move page from swapcache to filecache. | |
918 | */ | |
41ffe5d5 | 919 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 920 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 921 | { |
285b2c4f | 922 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 923 | void *radswap; |
41ffe5d5 | 924 | pgoff_t index; |
bde05d1c HD |
925 | gfp_t gfp; |
926 | int error = 0; | |
1da177e4 | 927 | |
46f65ec1 | 928 | radswap = swp_to_radix_entry(swap); |
e504f3fd | 929 | index = radix_tree_locate_item(&mapping->page_tree, radswap); |
46f65ec1 | 930 | if (index == -1) |
00501b53 | 931 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 932 | |
1b1b32f2 HD |
933 | /* |
934 | * Move _head_ to start search for next from here. | |
1f895f75 | 935 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 936 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 937 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
938 | */ |
939 | if (shmem_swaplist.next != &info->swaplist) | |
940 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 941 | |
bde05d1c HD |
942 | gfp = mapping_gfp_mask(mapping); |
943 | if (shmem_should_replace_page(*pagep, gfp)) { | |
944 | mutex_unlock(&shmem_swaplist_mutex); | |
945 | error = shmem_replace_page(pagep, gfp, info, index); | |
946 | mutex_lock(&shmem_swaplist_mutex); | |
947 | /* | |
948 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
949 | * allocation, but the inode might have been freed while we |
950 | * dropped it: although a racing shmem_evict_inode() cannot | |
951 | * complete without emptying the radix_tree, our page lock | |
952 | * on this swapcache page is not enough to prevent that - | |
953 | * free_swap_and_cache() of our swap entry will only | |
954 | * trylock_page(), removing swap from radix_tree whatever. | |
955 | * | |
956 | * We must not proceed to shmem_add_to_page_cache() if the | |
957 | * inode has been freed, but of course we cannot rely on | |
958 | * inode or mapping or info to check that. However, we can | |
959 | * safely check if our swap entry is still in use (and here | |
960 | * it can't have got reused for another page): if it's still | |
961 | * in use, then the inode cannot have been freed yet, and we | |
962 | * can safely proceed (if it's no longer in use, that tells | |
963 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
964 | */ |
965 | if (!page_swapcount(*pagep)) | |
966 | error = -ENOENT; | |
967 | } | |
968 | ||
d13d1443 | 969 | /* |
778dd893 HD |
970 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
971 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
972 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 973 | */ |
bde05d1c HD |
974 | if (!error) |
975 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 976 | radswap); |
48f170fb | 977 | if (error != -ENOMEM) { |
46f65ec1 HD |
978 | /* |
979 | * Truncation and eviction use free_swap_and_cache(), which | |
980 | * only does trylock page: if we raced, best clean up here. | |
981 | */ | |
bde05d1c HD |
982 | delete_from_swap_cache(*pagep); |
983 | set_page_dirty(*pagep); | |
46f65ec1 HD |
984 | if (!error) { |
985 | spin_lock(&info->lock); | |
986 | info->swapped--; | |
987 | spin_unlock(&info->lock); | |
988 | swap_free(swap); | |
989 | } | |
1da177e4 | 990 | } |
2e0e26c7 | 991 | return error; |
1da177e4 LT |
992 | } |
993 | ||
994 | /* | |
46f65ec1 | 995 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 996 | */ |
41ffe5d5 | 997 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 998 | { |
41ffe5d5 | 999 | struct list_head *this, *next; |
1da177e4 | 1000 | struct shmem_inode_info *info; |
00501b53 | 1001 | struct mem_cgroup *memcg; |
bde05d1c HD |
1002 | int error = 0; |
1003 | ||
1004 | /* | |
1005 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 1006 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 1007 | */ |
0142ef6c | 1008 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 1009 | goto out; |
778dd893 HD |
1010 | |
1011 | /* | |
1012 | * Charge page using GFP_KERNEL while we can wait, before taking | |
1013 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
1014 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 1015 | */ |
f627c2f5 KS |
1016 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg, |
1017 | false); | |
778dd893 HD |
1018 | if (error) |
1019 | goto out; | |
46f65ec1 | 1020 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 1021 | error = -EAGAIN; |
1da177e4 | 1022 | |
cb5f7b9a | 1023 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
1024 | list_for_each_safe(this, next, &shmem_swaplist) { |
1025 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 1026 | if (info->swapped) |
00501b53 | 1027 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
1028 | else |
1029 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1030 | cond_resched(); |
00501b53 | 1031 | if (error != -EAGAIN) |
778dd893 | 1032 | break; |
00501b53 | 1033 | /* found nothing in this: move on to search the next */ |
1da177e4 | 1034 | } |
cb5f7b9a | 1035 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1036 | |
00501b53 JW |
1037 | if (error) { |
1038 | if (error != -ENOMEM) | |
1039 | error = 0; | |
f627c2f5 | 1040 | mem_cgroup_cancel_charge(page, memcg, false); |
00501b53 | 1041 | } else |
f627c2f5 | 1042 | mem_cgroup_commit_charge(page, memcg, true, false); |
778dd893 | 1043 | out: |
aaa46865 | 1044 | unlock_page(page); |
09cbfeaf | 1045 | put_page(page); |
778dd893 | 1046 | return error; |
1da177e4 LT |
1047 | } |
1048 | ||
1049 | /* | |
1050 | * Move the page from the page cache to the swap cache. | |
1051 | */ | |
1052 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1053 | { | |
1054 | struct shmem_inode_info *info; | |
1da177e4 | 1055 | struct address_space *mapping; |
1da177e4 | 1056 | struct inode *inode; |
6922c0c7 HD |
1057 | swp_entry_t swap; |
1058 | pgoff_t index; | |
1da177e4 | 1059 | |
800d8c63 | 1060 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1061 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1062 | mapping = page->mapping; |
1063 | index = page->index; | |
1064 | inode = mapping->host; | |
1065 | info = SHMEM_I(inode); | |
1066 | if (info->flags & VM_LOCKED) | |
1067 | goto redirty; | |
d9fe526a | 1068 | if (!total_swap_pages) |
1da177e4 LT |
1069 | goto redirty; |
1070 | ||
d9fe526a | 1071 | /* |
97b713ba CH |
1072 | * Our capabilities prevent regular writeback or sync from ever calling |
1073 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1074 | * its underlying filesystem, in which case tmpfs should write out to | |
1075 | * swap only in response to memory pressure, and not for the writeback | |
1076 | * threads or sync. | |
d9fe526a | 1077 | */ |
48f170fb HD |
1078 | if (!wbc->for_reclaim) { |
1079 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1080 | goto redirty; | |
1081 | } | |
1635f6a7 HD |
1082 | |
1083 | /* | |
1084 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1085 | * value into swapfile.c, the only way we can correctly account for a | |
1086 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1087 | * |
1088 | * That's okay for a page already fallocated earlier, but if we have | |
1089 | * not yet completed the fallocation, then (a) we want to keep track | |
1090 | * of this page in case we have to undo it, and (b) it may not be a | |
1091 | * good idea to continue anyway, once we're pushing into swap. So | |
1092 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1093 | */ |
1094 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1095 | if (inode->i_private) { |
1096 | struct shmem_falloc *shmem_falloc; | |
1097 | spin_lock(&inode->i_lock); | |
1098 | shmem_falloc = inode->i_private; | |
1099 | if (shmem_falloc && | |
8e205f77 | 1100 | !shmem_falloc->waitq && |
1aac1400 HD |
1101 | index >= shmem_falloc->start && |
1102 | index < shmem_falloc->next) | |
1103 | shmem_falloc->nr_unswapped++; | |
1104 | else | |
1105 | shmem_falloc = NULL; | |
1106 | spin_unlock(&inode->i_lock); | |
1107 | if (shmem_falloc) | |
1108 | goto redirty; | |
1109 | } | |
1635f6a7 HD |
1110 | clear_highpage(page); |
1111 | flush_dcache_page(page); | |
1112 | SetPageUptodate(page); | |
1113 | } | |
1114 | ||
48f170fb HD |
1115 | swap = get_swap_page(); |
1116 | if (!swap.val) | |
1117 | goto redirty; | |
d9fe526a | 1118 | |
37e84351 VD |
1119 | if (mem_cgroup_try_charge_swap(page, swap)) |
1120 | goto free_swap; | |
1121 | ||
b1dea800 HD |
1122 | /* |
1123 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1124 | * if it's not already there. Do it now before the page is |
1125 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1126 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1127 | * we've incremented swapped, because shmem_unuse_inode() will |
1128 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1129 | */ |
48f170fb HD |
1130 | mutex_lock(&shmem_swaplist_mutex); |
1131 | if (list_empty(&info->swaplist)) | |
1132 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 1133 | |
48f170fb | 1134 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
6922c0c7 | 1135 | spin_lock(&info->lock); |
6922c0c7 | 1136 | shmem_recalc_inode(inode); |
267a4c76 | 1137 | info->swapped++; |
826267cf | 1138 | spin_unlock(&info->lock); |
6922c0c7 | 1139 | |
267a4c76 HD |
1140 | swap_shmem_alloc(swap); |
1141 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1142 | ||
6922c0c7 | 1143 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1144 | BUG_ON(page_mapped(page)); |
9fab5619 | 1145 | swap_writepage(page, wbc); |
1da177e4 LT |
1146 | return 0; |
1147 | } | |
1148 | ||
6922c0c7 | 1149 | mutex_unlock(&shmem_swaplist_mutex); |
37e84351 | 1150 | free_swap: |
0a31bc97 | 1151 | swapcache_free(swap); |
1da177e4 LT |
1152 | redirty: |
1153 | set_page_dirty(page); | |
d9fe526a HD |
1154 | if (wbc->for_reclaim) |
1155 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1156 | unlock_page(page); | |
1157 | return 0; | |
1da177e4 LT |
1158 | } |
1159 | ||
75edd345 | 1160 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1161 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1162 | { |
095f1fc4 | 1163 | char buffer[64]; |
680d794b | 1164 | |
71fe804b | 1165 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1166 | return; /* show nothing */ |
680d794b | 1167 | |
a7a88b23 | 1168 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1169 | |
1170 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1171 | } |
71fe804b LS |
1172 | |
1173 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1174 | { | |
1175 | struct mempolicy *mpol = NULL; | |
1176 | if (sbinfo->mpol) { | |
1177 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1178 | mpol = sbinfo->mpol; | |
1179 | mpol_get(mpol); | |
1180 | spin_unlock(&sbinfo->stat_lock); | |
1181 | } | |
1182 | return mpol; | |
1183 | } | |
75edd345 HD |
1184 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1185 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1186 | { | |
1187 | } | |
1188 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1189 | { | |
1190 | return NULL; | |
1191 | } | |
1192 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1193 | #ifndef CONFIG_NUMA | |
1194 | #define vm_policy vm_private_data | |
1195 | #endif | |
680d794b | 1196 | |
800d8c63 KS |
1197 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1198 | struct shmem_inode_info *info, pgoff_t index) | |
1199 | { | |
1200 | /* Create a pseudo vma that just contains the policy */ | |
1201 | vma->vm_start = 0; | |
1202 | /* Bias interleave by inode number to distribute better across nodes */ | |
1203 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
1204 | vma->vm_ops = NULL; | |
1205 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); | |
1206 | } | |
1207 | ||
1208 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1209 | { | |
1210 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1211 | mpol_cond_put(vma->vm_policy); | |
1212 | } | |
1213 | ||
41ffe5d5 HD |
1214 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1215 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1216 | { |
1da177e4 | 1217 | struct vm_area_struct pvma; |
18a2f371 | 1218 | struct page *page; |
52cd3b07 | 1219 | |
800d8c63 | 1220 | shmem_pseudo_vma_init(&pvma, info, index); |
18a2f371 | 1221 | page = swapin_readahead(swap, gfp, &pvma, 0); |
800d8c63 | 1222 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1223 | |
800d8c63 KS |
1224 | return page; |
1225 | } | |
1226 | ||
1227 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1228 | struct shmem_inode_info *info, pgoff_t index) | |
1229 | { | |
1230 | struct vm_area_struct pvma; | |
1231 | struct inode *inode = &info->vfs_inode; | |
1232 | struct address_space *mapping = inode->i_mapping; | |
1233 | pgoff_t idx, hindex = round_down(index, HPAGE_PMD_NR); | |
1234 | void __rcu **results; | |
1235 | struct page *page; | |
1236 | ||
1237 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) | |
1238 | return NULL; | |
1239 | ||
1240 | rcu_read_lock(); | |
1241 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx, | |
1242 | hindex, 1) && idx < hindex + HPAGE_PMD_NR) { | |
1243 | rcu_read_unlock(); | |
1244 | return NULL; | |
1245 | } | |
1246 | rcu_read_unlock(); | |
18a2f371 | 1247 | |
800d8c63 KS |
1248 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1249 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
1250 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); | |
1251 | shmem_pseudo_vma_destroy(&pvma); | |
1252 | if (page) | |
1253 | prep_transhuge_page(page); | |
18a2f371 | 1254 | return page; |
1da177e4 LT |
1255 | } |
1256 | ||
02098fea | 1257 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1258 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1259 | { |
1260 | struct vm_area_struct pvma; | |
18a2f371 | 1261 | struct page *page; |
1da177e4 | 1262 | |
800d8c63 KS |
1263 | shmem_pseudo_vma_init(&pvma, info, index); |
1264 | page = alloc_page_vma(gfp, &pvma, 0); | |
1265 | shmem_pseudo_vma_destroy(&pvma); | |
1266 | ||
1267 | return page; | |
1268 | } | |
1269 | ||
1270 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
1271 | struct shmem_inode_info *info, struct shmem_sb_info *sbinfo, | |
1272 | pgoff_t index, bool huge) | |
1273 | { | |
1274 | struct page *page; | |
1275 | int nr; | |
1276 | int err = -ENOSPC; | |
52cd3b07 | 1277 | |
800d8c63 KS |
1278 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) |
1279 | huge = false; | |
1280 | nr = huge ? HPAGE_PMD_NR : 1; | |
1281 | ||
1282 | if (shmem_acct_block(info->flags, nr)) | |
1283 | goto failed; | |
1284 | if (sbinfo->max_blocks) { | |
1285 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
1286 | sbinfo->max_blocks - nr) > 0) | |
1287 | goto unacct; | |
1288 | percpu_counter_add(&sbinfo->used_blocks, nr); | |
1289 | } | |
1290 | ||
1291 | if (huge) | |
1292 | page = shmem_alloc_hugepage(gfp, info, index); | |
1293 | else | |
1294 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1295 | if (page) { |
1296 | __SetPageLocked(page); | |
1297 | __SetPageSwapBacked(page); | |
800d8c63 | 1298 | return page; |
75edd345 | 1299 | } |
18a2f371 | 1300 | |
800d8c63 KS |
1301 | err = -ENOMEM; |
1302 | if (sbinfo->max_blocks) | |
1303 | percpu_counter_add(&sbinfo->used_blocks, -nr); | |
1304 | unacct: | |
1305 | shmem_unacct_blocks(info->flags, nr); | |
1306 | failed: | |
1307 | return ERR_PTR(err); | |
1da177e4 | 1308 | } |
71fe804b | 1309 | |
bde05d1c HD |
1310 | /* |
1311 | * When a page is moved from swapcache to shmem filecache (either by the | |
1312 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1313 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1314 | * ignorance of the mapping it belongs to. If that mapping has special | |
1315 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1316 | * we may need to copy to a suitable page before moving to filecache. | |
1317 | * | |
1318 | * In a future release, this may well be extended to respect cpuset and | |
1319 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1320 | * but for now it is a simple matter of zone. | |
1321 | */ | |
1322 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1323 | { | |
1324 | return page_zonenum(page) > gfp_zone(gfp); | |
1325 | } | |
1326 | ||
1327 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1328 | struct shmem_inode_info *info, pgoff_t index) | |
1329 | { | |
1330 | struct page *oldpage, *newpage; | |
1331 | struct address_space *swap_mapping; | |
1332 | pgoff_t swap_index; | |
1333 | int error; | |
1334 | ||
1335 | oldpage = *pagep; | |
1336 | swap_index = page_private(oldpage); | |
1337 | swap_mapping = page_mapping(oldpage); | |
1338 | ||
1339 | /* | |
1340 | * We have arrived here because our zones are constrained, so don't | |
1341 | * limit chance of success by further cpuset and node constraints. | |
1342 | */ | |
1343 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1344 | newpage = shmem_alloc_page(gfp, info, index); | |
1345 | if (!newpage) | |
1346 | return -ENOMEM; | |
bde05d1c | 1347 | |
09cbfeaf | 1348 | get_page(newpage); |
bde05d1c | 1349 | copy_highpage(newpage, oldpage); |
0142ef6c | 1350 | flush_dcache_page(newpage); |
bde05d1c | 1351 | |
bde05d1c | 1352 | SetPageUptodate(newpage); |
bde05d1c | 1353 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1354 | SetPageSwapCache(newpage); |
1355 | ||
1356 | /* | |
1357 | * Our caller will very soon move newpage out of swapcache, but it's | |
1358 | * a nice clean interface for us to replace oldpage by newpage there. | |
1359 | */ | |
1360 | spin_lock_irq(&swap_mapping->tree_lock); | |
1361 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1362 | newpage); | |
0142ef6c HD |
1363 | if (!error) { |
1364 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
1365 | __dec_zone_page_state(oldpage, NR_FILE_PAGES); | |
1366 | } | |
bde05d1c | 1367 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1368 | |
0142ef6c HD |
1369 | if (unlikely(error)) { |
1370 | /* | |
1371 | * Is this possible? I think not, now that our callers check | |
1372 | * both PageSwapCache and page_private after getting page lock; | |
1373 | * but be defensive. Reverse old to newpage for clear and free. | |
1374 | */ | |
1375 | oldpage = newpage; | |
1376 | } else { | |
6a93ca8f | 1377 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1378 | lru_cache_add_anon(newpage); |
1379 | *pagep = newpage; | |
1380 | } | |
bde05d1c HD |
1381 | |
1382 | ClearPageSwapCache(oldpage); | |
1383 | set_page_private(oldpage, 0); | |
1384 | ||
1385 | unlock_page(oldpage); | |
09cbfeaf KS |
1386 | put_page(oldpage); |
1387 | put_page(oldpage); | |
0142ef6c | 1388 | return error; |
bde05d1c HD |
1389 | } |
1390 | ||
1da177e4 | 1391 | /* |
68da9f05 | 1392 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1393 | * |
1394 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1395 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1396 | * entry since a page cannot live in both the swap and page cache. |
1397 | * | |
1398 | * fault_mm and fault_type are only supplied by shmem_fault: | |
1399 | * otherwise they are NULL. | |
1da177e4 | 1400 | */ |
41ffe5d5 | 1401 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 ALC |
1402 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
1403 | struct mm_struct *fault_mm, int *fault_type) | |
1da177e4 LT |
1404 | { |
1405 | struct address_space *mapping = inode->i_mapping; | |
54af6042 | 1406 | struct shmem_inode_info *info; |
1da177e4 | 1407 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1408 | struct mm_struct *charge_mm; |
00501b53 | 1409 | struct mem_cgroup *memcg; |
27ab7006 | 1410 | struct page *page; |
1da177e4 | 1411 | swp_entry_t swap; |
800d8c63 | 1412 | pgoff_t hindex = index; |
1da177e4 | 1413 | int error; |
54af6042 | 1414 | int once = 0; |
1635f6a7 | 1415 | int alloced = 0; |
1da177e4 | 1416 | |
09cbfeaf | 1417 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1418 | return -EFBIG; |
1da177e4 | 1419 | repeat: |
54af6042 | 1420 | swap.val = 0; |
0cd6144a | 1421 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1422 | if (radix_tree_exceptional_entry(page)) { |
1423 | swap = radix_to_swp_entry(page); | |
1424 | page = NULL; | |
1425 | } | |
1426 | ||
75edd345 | 1427 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1428 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 1429 | error = -EINVAL; |
267a4c76 | 1430 | goto unlock; |
54af6042 HD |
1431 | } |
1432 | ||
66d2f4d2 HD |
1433 | if (page && sgp == SGP_WRITE) |
1434 | mark_page_accessed(page); | |
1435 | ||
1635f6a7 HD |
1436 | /* fallocated page? */ |
1437 | if (page && !PageUptodate(page)) { | |
1438 | if (sgp != SGP_READ) | |
1439 | goto clear; | |
1440 | unlock_page(page); | |
09cbfeaf | 1441 | put_page(page); |
1635f6a7 HD |
1442 | page = NULL; |
1443 | } | |
54af6042 | 1444 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1445 | *pagep = page; |
1446 | return 0; | |
27ab7006 HD |
1447 | } |
1448 | ||
1449 | /* | |
54af6042 HD |
1450 | * Fast cache lookup did not find it: |
1451 | * bring it back from swap or allocate. | |
27ab7006 | 1452 | */ |
54af6042 HD |
1453 | info = SHMEM_I(inode); |
1454 | sbinfo = SHMEM_SB(inode->i_sb); | |
9e18eb29 | 1455 | charge_mm = fault_mm ? : current->mm; |
1da177e4 | 1456 | |
1da177e4 LT |
1457 | if (swap.val) { |
1458 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1459 | page = lookup_swap_cache(swap); |
1460 | if (!page) { | |
9e18eb29 ALC |
1461 | /* Or update major stats only when swapin succeeds?? */ |
1462 | if (fault_type) { | |
68da9f05 | 1463 | *fault_type |= VM_FAULT_MAJOR; |
9e18eb29 ALC |
1464 | count_vm_event(PGMAJFAULT); |
1465 | mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT); | |
1466 | } | |
1467 | /* Here we actually start the io */ | |
41ffe5d5 | 1468 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1469 | if (!page) { |
54af6042 HD |
1470 | error = -ENOMEM; |
1471 | goto failed; | |
1da177e4 | 1472 | } |
1da177e4 LT |
1473 | } |
1474 | ||
1475 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1476 | lock_page(page); |
0142ef6c | 1477 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1478 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1479 | error = -EEXIST; /* try again */ |
d1899228 | 1480 | goto unlock; |
bde05d1c | 1481 | } |
27ab7006 | 1482 | if (!PageUptodate(page)) { |
1da177e4 | 1483 | error = -EIO; |
54af6042 | 1484 | goto failed; |
1da177e4 | 1485 | } |
54af6042 HD |
1486 | wait_on_page_writeback(page); |
1487 | ||
bde05d1c HD |
1488 | if (shmem_should_replace_page(page, gfp)) { |
1489 | error = shmem_replace_page(&page, gfp, info, index); | |
1490 | if (error) | |
1491 | goto failed; | |
1da177e4 | 1492 | } |
27ab7006 | 1493 | |
9e18eb29 | 1494 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
f627c2f5 | 1495 | false); |
d1899228 | 1496 | if (!error) { |
aa3b1895 | 1497 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1498 | swp_to_radix_entry(swap)); |
215c02bc HD |
1499 | /* |
1500 | * We already confirmed swap under page lock, and make | |
1501 | * no memory allocation here, so usually no possibility | |
1502 | * of error; but free_swap_and_cache() only trylocks a | |
1503 | * page, so it is just possible that the entry has been | |
1504 | * truncated or holepunched since swap was confirmed. | |
1505 | * shmem_undo_range() will have done some of the | |
1506 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1507 | * the rest. |
215c02bc HD |
1508 | * Reset swap.val? No, leave it so "failed" goes back to |
1509 | * "repeat": reading a hole and writing should succeed. | |
1510 | */ | |
00501b53 | 1511 | if (error) { |
f627c2f5 | 1512 | mem_cgroup_cancel_charge(page, memcg, false); |
215c02bc | 1513 | delete_from_swap_cache(page); |
00501b53 | 1514 | } |
d1899228 | 1515 | } |
54af6042 HD |
1516 | if (error) |
1517 | goto failed; | |
1518 | ||
f627c2f5 | 1519 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1520 | |
54af6042 | 1521 | spin_lock(&info->lock); |
285b2c4f | 1522 | info->swapped--; |
54af6042 | 1523 | shmem_recalc_inode(inode); |
27ab7006 | 1524 | spin_unlock(&info->lock); |
54af6042 | 1525 | |
66d2f4d2 HD |
1526 | if (sgp == SGP_WRITE) |
1527 | mark_page_accessed(page); | |
1528 | ||
54af6042 | 1529 | delete_from_swap_cache(page); |
27ab7006 HD |
1530 | set_page_dirty(page); |
1531 | swap_free(swap); | |
1532 | ||
54af6042 | 1533 | } else { |
800d8c63 KS |
1534 | /* shmem_symlink() */ |
1535 | if (mapping->a_ops != &shmem_aops) | |
1536 | goto alloc_nohuge; | |
1537 | if (shmem_huge == SHMEM_HUGE_DENY) | |
1538 | goto alloc_nohuge; | |
1539 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1540 | goto alloc_huge; | |
1541 | switch (sbinfo->huge) { | |
1542 | loff_t i_size; | |
1543 | pgoff_t off; | |
1544 | case SHMEM_HUGE_NEVER: | |
1545 | goto alloc_nohuge; | |
1546 | case SHMEM_HUGE_WITHIN_SIZE: | |
1547 | off = round_up(index, HPAGE_PMD_NR); | |
1548 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1549 | if (i_size >= HPAGE_PMD_SIZE && | |
1550 | i_size >> PAGE_SHIFT >= off) | |
1551 | goto alloc_huge; | |
1552 | /* fallthrough */ | |
1553 | case SHMEM_HUGE_ADVISE: | |
1554 | /* TODO: wire up fadvise()/madvise() */ | |
1555 | goto alloc_nohuge; | |
54af6042 | 1556 | } |
1da177e4 | 1557 | |
800d8c63 KS |
1558 | alloc_huge: |
1559 | page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1560 | index, true); | |
1561 | if (IS_ERR(page)) { | |
1562 | alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1563 | index, false); | |
1da177e4 | 1564 | } |
800d8c63 KS |
1565 | if (IS_ERR(page)) { |
1566 | error = PTR_ERR(page); | |
1567 | page = NULL; | |
1568 | goto failed; | |
1569 | } | |
1570 | ||
1571 | if (PageTransHuge(page)) | |
1572 | hindex = round_down(index, HPAGE_PMD_NR); | |
1573 | else | |
1574 | hindex = index; | |
1575 | ||
66d2f4d2 | 1576 | if (sgp == SGP_WRITE) |
eb39d618 | 1577 | __SetPageReferenced(page); |
66d2f4d2 | 1578 | |
9e18eb29 | 1579 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
800d8c63 | 1580 | PageTransHuge(page)); |
54af6042 | 1581 | if (error) |
800d8c63 KS |
1582 | goto unacct; |
1583 | error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK, | |
1584 | compound_order(page)); | |
b065b432 | 1585 | if (!error) { |
800d8c63 | 1586 | error = shmem_add_to_page_cache(page, mapping, hindex, |
fed400a1 | 1587 | NULL); |
b065b432 HD |
1588 | radix_tree_preload_end(); |
1589 | } | |
1590 | if (error) { | |
800d8c63 KS |
1591 | mem_cgroup_cancel_charge(page, memcg, |
1592 | PageTransHuge(page)); | |
1593 | goto unacct; | |
b065b432 | 1594 | } |
800d8c63 KS |
1595 | mem_cgroup_commit_charge(page, memcg, false, |
1596 | PageTransHuge(page)); | |
54af6042 HD |
1597 | lru_cache_add_anon(page); |
1598 | ||
1599 | spin_lock(&info->lock); | |
800d8c63 KS |
1600 | info->alloced += 1 << compound_order(page); |
1601 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); | |
54af6042 | 1602 | shmem_recalc_inode(inode); |
1da177e4 | 1603 | spin_unlock(&info->lock); |
1635f6a7 | 1604 | alloced = true; |
54af6042 | 1605 | |
ec9516fb | 1606 | /* |
1635f6a7 HD |
1607 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1608 | */ | |
1609 | if (sgp == SGP_FALLOC) | |
1610 | sgp = SGP_WRITE; | |
1611 | clear: | |
1612 | /* | |
1613 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1614 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1615 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb | 1616 | */ |
800d8c63 KS |
1617 | if (sgp != SGP_WRITE && !PageUptodate(page)) { |
1618 | struct page *head = compound_head(page); | |
1619 | int i; | |
1620 | ||
1621 | for (i = 0; i < (1 << compound_order(head)); i++) { | |
1622 | clear_highpage(head + i); | |
1623 | flush_dcache_page(head + i); | |
1624 | } | |
1625 | SetPageUptodate(head); | |
ec9516fb | 1626 | } |
1da177e4 | 1627 | } |
bde05d1c | 1628 | |
54af6042 | 1629 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1630 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1631 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1632 | if (alloced) { |
1633 | ClearPageDirty(page); | |
1634 | delete_from_page_cache(page); | |
1635 | spin_lock(&info->lock); | |
1636 | shmem_recalc_inode(inode); | |
1637 | spin_unlock(&info->lock); | |
1638 | } | |
54af6042 | 1639 | error = -EINVAL; |
267a4c76 | 1640 | goto unlock; |
e83c32e8 | 1641 | } |
800d8c63 | 1642 | *pagep = page + index - hindex; |
54af6042 | 1643 | return 0; |
1da177e4 | 1644 | |
59a16ead | 1645 | /* |
54af6042 | 1646 | * Error recovery. |
59a16ead | 1647 | */ |
54af6042 | 1648 | unacct: |
800d8c63 KS |
1649 | if (sbinfo->max_blocks) |
1650 | percpu_counter_sub(&sbinfo->used_blocks, | |
1651 | 1 << compound_order(page)); | |
1652 | shmem_unacct_blocks(info->flags, 1 << compound_order(page)); | |
1653 | ||
1654 | if (PageTransHuge(page)) { | |
1655 | unlock_page(page); | |
1656 | put_page(page); | |
1657 | goto alloc_nohuge; | |
1658 | } | |
54af6042 | 1659 | failed: |
267a4c76 | 1660 | if (swap.val && !shmem_confirm_swap(mapping, index, swap)) |
d1899228 HD |
1661 | error = -EEXIST; |
1662 | unlock: | |
27ab7006 | 1663 | if (page) { |
54af6042 | 1664 | unlock_page(page); |
09cbfeaf | 1665 | put_page(page); |
54af6042 HD |
1666 | } |
1667 | if (error == -ENOSPC && !once++) { | |
1668 | info = SHMEM_I(inode); | |
1669 | spin_lock(&info->lock); | |
1670 | shmem_recalc_inode(inode); | |
1671 | spin_unlock(&info->lock); | |
27ab7006 | 1672 | goto repeat; |
ff36b801 | 1673 | } |
d1899228 | 1674 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1675 | goto repeat; |
1676 | return error; | |
1da177e4 LT |
1677 | } |
1678 | ||
d0217ac0 | 1679 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1680 | { |
496ad9aa | 1681 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 1682 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
1da177e4 | 1683 | int error; |
68da9f05 | 1684 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1685 | |
f00cdc6d HD |
1686 | /* |
1687 | * Trinity finds that probing a hole which tmpfs is punching can | |
1688 | * prevent the hole-punch from ever completing: which in turn | |
1689 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1690 | * faulting pages into the hole while it's being punched. Although |
1691 | * shmem_undo_range() does remove the additions, it may be unable to | |
1692 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1693 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1694 | * | |
1695 | * It does not matter if we sometimes reach this check just before the | |
1696 | * hole-punch begins, so that one fault then races with the punch: | |
1697 | * we just need to make racing faults a rare case. | |
1698 | * | |
1699 | * The implementation below would be much simpler if we just used a | |
1700 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1701 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1702 | */ |
1703 | if (unlikely(inode->i_private)) { | |
1704 | struct shmem_falloc *shmem_falloc; | |
1705 | ||
1706 | spin_lock(&inode->i_lock); | |
1707 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1708 | if (shmem_falloc && |
1709 | shmem_falloc->waitq && | |
1710 | vmf->pgoff >= shmem_falloc->start && | |
1711 | vmf->pgoff < shmem_falloc->next) { | |
1712 | wait_queue_head_t *shmem_falloc_waitq; | |
1713 | DEFINE_WAIT(shmem_fault_wait); | |
1714 | ||
1715 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1716 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1717 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1718 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1719 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1720 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1721 | } |
8e205f77 HD |
1722 | |
1723 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1724 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1725 | TASK_UNINTERRUPTIBLE); | |
1726 | spin_unlock(&inode->i_lock); | |
1727 | schedule(); | |
1728 | ||
1729 | /* | |
1730 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1731 | * stack of the hole-punching task: shmem_falloc_waitq | |
1732 | * is usually invalid by the time we reach here, but | |
1733 | * finish_wait() does not dereference it in that case; | |
1734 | * though i_lock needed lest racing with wake_up_all(). | |
1735 | */ | |
1736 | spin_lock(&inode->i_lock); | |
1737 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1738 | spin_unlock(&inode->i_lock); | |
1739 | return ret; | |
f00cdc6d | 1740 | } |
8e205f77 | 1741 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1742 | } |
1743 | ||
9e18eb29 ALC |
1744 | error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, SGP_CACHE, |
1745 | gfp, vma->vm_mm, &ret); | |
d0217ac0 NP |
1746 | if (error) |
1747 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1748 | return ret; |
1da177e4 LT |
1749 | } |
1750 | ||
c01d5b30 HD |
1751 | unsigned long shmem_get_unmapped_area(struct file *file, |
1752 | unsigned long uaddr, unsigned long len, | |
1753 | unsigned long pgoff, unsigned long flags) | |
1754 | { | |
1755 | unsigned long (*get_area)(struct file *, | |
1756 | unsigned long, unsigned long, unsigned long, unsigned long); | |
1757 | unsigned long addr; | |
1758 | unsigned long offset; | |
1759 | unsigned long inflated_len; | |
1760 | unsigned long inflated_addr; | |
1761 | unsigned long inflated_offset; | |
1762 | ||
1763 | if (len > TASK_SIZE) | |
1764 | return -ENOMEM; | |
1765 | ||
1766 | get_area = current->mm->get_unmapped_area; | |
1767 | addr = get_area(file, uaddr, len, pgoff, flags); | |
1768 | ||
1769 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) | |
1770 | return addr; | |
1771 | if (IS_ERR_VALUE(addr)) | |
1772 | return addr; | |
1773 | if (addr & ~PAGE_MASK) | |
1774 | return addr; | |
1775 | if (addr > TASK_SIZE - len) | |
1776 | return addr; | |
1777 | ||
1778 | if (shmem_huge == SHMEM_HUGE_DENY) | |
1779 | return addr; | |
1780 | if (len < HPAGE_PMD_SIZE) | |
1781 | return addr; | |
1782 | if (flags & MAP_FIXED) | |
1783 | return addr; | |
1784 | /* | |
1785 | * Our priority is to support MAP_SHARED mapped hugely; | |
1786 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
1787 | * But if caller specified an address hint, respect that as before. | |
1788 | */ | |
1789 | if (uaddr) | |
1790 | return addr; | |
1791 | ||
1792 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
1793 | struct super_block *sb; | |
1794 | ||
1795 | if (file) { | |
1796 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
1797 | sb = file_inode(file)->i_sb; | |
1798 | } else { | |
1799 | /* | |
1800 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
1801 | * for "/dev/zero", to create a shared anonymous object. | |
1802 | */ | |
1803 | if (IS_ERR(shm_mnt)) | |
1804 | return addr; | |
1805 | sb = shm_mnt->mnt_sb; | |
1806 | } | |
1807 | if (SHMEM_SB(sb)->huge != SHMEM_HUGE_NEVER) | |
1808 | return addr; | |
1809 | } | |
1810 | ||
1811 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
1812 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
1813 | return addr; | |
1814 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
1815 | return addr; | |
1816 | ||
1817 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
1818 | if (inflated_len > TASK_SIZE) | |
1819 | return addr; | |
1820 | if (inflated_len < len) | |
1821 | return addr; | |
1822 | ||
1823 | inflated_addr = get_area(NULL, 0, inflated_len, 0, flags); | |
1824 | if (IS_ERR_VALUE(inflated_addr)) | |
1825 | return addr; | |
1826 | if (inflated_addr & ~PAGE_MASK) | |
1827 | return addr; | |
1828 | ||
1829 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
1830 | inflated_addr += offset - inflated_offset; | |
1831 | if (inflated_offset > offset) | |
1832 | inflated_addr += HPAGE_PMD_SIZE; | |
1833 | ||
1834 | if (inflated_addr > TASK_SIZE - len) | |
1835 | return addr; | |
1836 | return inflated_addr; | |
1837 | } | |
1838 | ||
1da177e4 | 1839 | #ifdef CONFIG_NUMA |
41ffe5d5 | 1840 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 1841 | { |
496ad9aa | 1842 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1843 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
1844 | } |
1845 | ||
d8dc74f2 AB |
1846 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1847 | unsigned long addr) | |
1da177e4 | 1848 | { |
496ad9aa | 1849 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1850 | pgoff_t index; |
1da177e4 | 1851 | |
41ffe5d5 HD |
1852 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
1853 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
1854 | } |
1855 | #endif | |
1856 | ||
1857 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1858 | { | |
496ad9aa | 1859 | struct inode *inode = file_inode(file); |
1da177e4 LT |
1860 | struct shmem_inode_info *info = SHMEM_I(inode); |
1861 | int retval = -ENOMEM; | |
1862 | ||
1863 | spin_lock(&info->lock); | |
1864 | if (lock && !(info->flags & VM_LOCKED)) { | |
1865 | if (!user_shm_lock(inode->i_size, user)) | |
1866 | goto out_nomem; | |
1867 | info->flags |= VM_LOCKED; | |
89e004ea | 1868 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
1869 | } |
1870 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1871 | user_shm_unlock(inode->i_size, user); | |
1872 | info->flags &= ~VM_LOCKED; | |
89e004ea | 1873 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
1874 | } |
1875 | retval = 0; | |
89e004ea | 1876 | |
1da177e4 LT |
1877 | out_nomem: |
1878 | spin_unlock(&info->lock); | |
1879 | return retval; | |
1880 | } | |
1881 | ||
9b83a6a8 | 1882 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1883 | { |
1884 | file_accessed(file); | |
1885 | vma->vm_ops = &shmem_vm_ops; | |
1886 | return 0; | |
1887 | } | |
1888 | ||
454abafe | 1889 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 1890 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
1891 | { |
1892 | struct inode *inode; | |
1893 | struct shmem_inode_info *info; | |
1894 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1895 | ||
5b04c689 PE |
1896 | if (shmem_reserve_inode(sb)) |
1897 | return NULL; | |
1da177e4 LT |
1898 | |
1899 | inode = new_inode(sb); | |
1900 | if (inode) { | |
85fe4025 | 1901 | inode->i_ino = get_next_ino(); |
454abafe | 1902 | inode_init_owner(inode, dir, mode); |
1da177e4 | 1903 | inode->i_blocks = 0; |
1da177e4 | 1904 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
91828a40 | 1905 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1906 | info = SHMEM_I(inode); |
1907 | memset(info, 0, (char *)inode - (char *)info); | |
1908 | spin_lock_init(&info->lock); | |
40e041a2 | 1909 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 1910 | info->flags = flags & VM_NORESERVE; |
1da177e4 | 1911 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 1912 | simple_xattrs_init(&info->xattrs); |
72c04902 | 1913 | cache_no_acl(inode); |
1da177e4 LT |
1914 | |
1915 | switch (mode & S_IFMT) { | |
1916 | default: | |
39f0247d | 1917 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1918 | init_special_inode(inode, mode, dev); |
1919 | break; | |
1920 | case S_IFREG: | |
14fcc23f | 1921 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1922 | inode->i_op = &shmem_inode_operations; |
1923 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
1924 | mpol_shared_policy_init(&info->policy, |
1925 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
1926 | break; |
1927 | case S_IFDIR: | |
d8c76e6f | 1928 | inc_nlink(inode); |
1da177e4 LT |
1929 | /* Some things misbehave if size == 0 on a directory */ |
1930 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1931 | inode->i_op = &shmem_dir_inode_operations; | |
1932 | inode->i_fop = &simple_dir_operations; | |
1933 | break; | |
1934 | case S_IFLNK: | |
1935 | /* | |
1936 | * Must not load anything in the rbtree, | |
1937 | * mpol_free_shared_policy will not be called. | |
1938 | */ | |
71fe804b | 1939 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
1940 | break; |
1941 | } | |
5b04c689 PE |
1942 | } else |
1943 | shmem_free_inode(sb); | |
1da177e4 LT |
1944 | return inode; |
1945 | } | |
1946 | ||
0cd6144a JW |
1947 | bool shmem_mapping(struct address_space *mapping) |
1948 | { | |
f0774d88 SL |
1949 | if (!mapping->host) |
1950 | return false; | |
1951 | ||
97b713ba | 1952 | return mapping->host->i_sb->s_op == &shmem_ops; |
0cd6144a JW |
1953 | } |
1954 | ||
1da177e4 | 1955 | #ifdef CONFIG_TMPFS |
92e1d5be | 1956 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 1957 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 1958 | |
6d9d88d0 JS |
1959 | #ifdef CONFIG_TMPFS_XATTR |
1960 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
1961 | #else | |
1962 | #define shmem_initxattrs NULL | |
1963 | #endif | |
1964 | ||
1da177e4 | 1965 | static int |
800d15a5 NP |
1966 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1967 | loff_t pos, unsigned len, unsigned flags, | |
1968 | struct page **pagep, void **fsdata) | |
1da177e4 | 1969 | { |
800d15a5 | 1970 | struct inode *inode = mapping->host; |
40e041a2 | 1971 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 1972 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
1973 | |
1974 | /* i_mutex is held by caller */ | |
1975 | if (unlikely(info->seals)) { | |
1976 | if (info->seals & F_SEAL_WRITE) | |
1977 | return -EPERM; | |
1978 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
1979 | return -EPERM; | |
1980 | } | |
1981 | ||
9e18eb29 | 1982 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
1983 | } |
1984 | ||
1985 | static int | |
1986 | shmem_write_end(struct file *file, struct address_space *mapping, | |
1987 | loff_t pos, unsigned len, unsigned copied, | |
1988 | struct page *page, void *fsdata) | |
1989 | { | |
1990 | struct inode *inode = mapping->host; | |
1991 | ||
d3602444 HD |
1992 | if (pos + copied > inode->i_size) |
1993 | i_size_write(inode, pos + copied); | |
1994 | ||
ec9516fb | 1995 | if (!PageUptodate(page)) { |
800d8c63 KS |
1996 | struct page *head = compound_head(page); |
1997 | if (PageTransCompound(page)) { | |
1998 | int i; | |
1999 | ||
2000 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2001 | if (head + i == page) | |
2002 | continue; | |
2003 | clear_highpage(head + i); | |
2004 | flush_dcache_page(head + i); | |
2005 | } | |
2006 | } | |
09cbfeaf KS |
2007 | if (copied < PAGE_SIZE) { |
2008 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2009 | zero_user_segments(page, 0, from, |
09cbfeaf | 2010 | from + copied, PAGE_SIZE); |
ec9516fb | 2011 | } |
800d8c63 | 2012 | SetPageUptodate(head); |
ec9516fb | 2013 | } |
800d15a5 | 2014 | set_page_dirty(page); |
6746aff7 | 2015 | unlock_page(page); |
09cbfeaf | 2016 | put_page(page); |
800d15a5 | 2017 | |
800d15a5 | 2018 | return copied; |
1da177e4 LT |
2019 | } |
2020 | ||
2ba5bbed | 2021 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2022 | { |
6e58e79d AV |
2023 | struct file *file = iocb->ki_filp; |
2024 | struct inode *inode = file_inode(file); | |
1da177e4 | 2025 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2026 | pgoff_t index; |
2027 | unsigned long offset; | |
a0ee5ec5 | 2028 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2029 | int error = 0; |
cb66a7a1 | 2030 | ssize_t retval = 0; |
6e58e79d | 2031 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2032 | |
2033 | /* | |
2034 | * Might this read be for a stacking filesystem? Then when reading | |
2035 | * holes of a sparse file, we actually need to allocate those pages, | |
2036 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2037 | */ | |
777eda2c | 2038 | if (!iter_is_iovec(to)) |
75edd345 | 2039 | sgp = SGP_CACHE; |
1da177e4 | 2040 | |
09cbfeaf KS |
2041 | index = *ppos >> PAGE_SHIFT; |
2042 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2043 | |
2044 | for (;;) { | |
2045 | struct page *page = NULL; | |
41ffe5d5 HD |
2046 | pgoff_t end_index; |
2047 | unsigned long nr, ret; | |
1da177e4 LT |
2048 | loff_t i_size = i_size_read(inode); |
2049 | ||
09cbfeaf | 2050 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2051 | if (index > end_index) |
2052 | break; | |
2053 | if (index == end_index) { | |
09cbfeaf | 2054 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2055 | if (nr <= offset) |
2056 | break; | |
2057 | } | |
2058 | ||
9e18eb29 | 2059 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2060 | if (error) { |
2061 | if (error == -EINVAL) | |
2062 | error = 0; | |
1da177e4 LT |
2063 | break; |
2064 | } | |
75edd345 HD |
2065 | if (page) { |
2066 | if (sgp == SGP_CACHE) | |
2067 | set_page_dirty(page); | |
d3602444 | 2068 | unlock_page(page); |
75edd345 | 2069 | } |
1da177e4 LT |
2070 | |
2071 | /* | |
2072 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2073 | * are called without i_mutex protection against truncate |
1da177e4 | 2074 | */ |
09cbfeaf | 2075 | nr = PAGE_SIZE; |
1da177e4 | 2076 | i_size = i_size_read(inode); |
09cbfeaf | 2077 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2078 | if (index == end_index) { |
09cbfeaf | 2079 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2080 | if (nr <= offset) { |
2081 | if (page) | |
09cbfeaf | 2082 | put_page(page); |
1da177e4 LT |
2083 | break; |
2084 | } | |
2085 | } | |
2086 | nr -= offset; | |
2087 | ||
2088 | if (page) { | |
2089 | /* | |
2090 | * If users can be writing to this page using arbitrary | |
2091 | * virtual addresses, take care about potential aliasing | |
2092 | * before reading the page on the kernel side. | |
2093 | */ | |
2094 | if (mapping_writably_mapped(mapping)) | |
2095 | flush_dcache_page(page); | |
2096 | /* | |
2097 | * Mark the page accessed if we read the beginning. | |
2098 | */ | |
2099 | if (!offset) | |
2100 | mark_page_accessed(page); | |
b5810039 | 2101 | } else { |
1da177e4 | 2102 | page = ZERO_PAGE(0); |
09cbfeaf | 2103 | get_page(page); |
b5810039 | 2104 | } |
1da177e4 LT |
2105 | |
2106 | /* | |
2107 | * Ok, we have the page, and it's up-to-date, so | |
2108 | * now we can copy it to user space... | |
1da177e4 | 2109 | */ |
2ba5bbed | 2110 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2111 | retval += ret; |
1da177e4 | 2112 | offset += ret; |
09cbfeaf KS |
2113 | index += offset >> PAGE_SHIFT; |
2114 | offset &= ~PAGE_MASK; | |
1da177e4 | 2115 | |
09cbfeaf | 2116 | put_page(page); |
2ba5bbed | 2117 | if (!iov_iter_count(to)) |
1da177e4 | 2118 | break; |
6e58e79d AV |
2119 | if (ret < nr) { |
2120 | error = -EFAULT; | |
2121 | break; | |
2122 | } | |
1da177e4 LT |
2123 | cond_resched(); |
2124 | } | |
2125 | ||
09cbfeaf | 2126 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2127 | file_accessed(file); |
2128 | return retval ? retval : error; | |
1da177e4 LT |
2129 | } |
2130 | ||
708e3508 HD |
2131 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, |
2132 | struct pipe_inode_info *pipe, size_t len, | |
2133 | unsigned int flags) | |
2134 | { | |
2135 | struct address_space *mapping = in->f_mapping; | |
71f0e07a | 2136 | struct inode *inode = mapping->host; |
708e3508 HD |
2137 | unsigned int loff, nr_pages, req_pages; |
2138 | struct page *pages[PIPE_DEF_BUFFERS]; | |
2139 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
2140 | struct page *page; | |
2141 | pgoff_t index, end_index; | |
2142 | loff_t isize, left; | |
2143 | int error, page_nr; | |
2144 | struct splice_pipe_desc spd = { | |
2145 | .pages = pages, | |
2146 | .partial = partial, | |
047fe360 | 2147 | .nr_pages_max = PIPE_DEF_BUFFERS, |
708e3508 HD |
2148 | .flags = flags, |
2149 | .ops = &page_cache_pipe_buf_ops, | |
2150 | .spd_release = spd_release_page, | |
2151 | }; | |
2152 | ||
71f0e07a | 2153 | isize = i_size_read(inode); |
708e3508 HD |
2154 | if (unlikely(*ppos >= isize)) |
2155 | return 0; | |
2156 | ||
2157 | left = isize - *ppos; | |
2158 | if (unlikely(left < len)) | |
2159 | len = left; | |
2160 | ||
2161 | if (splice_grow_spd(pipe, &spd)) | |
2162 | return -ENOMEM; | |
2163 | ||
09cbfeaf KS |
2164 | index = *ppos >> PAGE_SHIFT; |
2165 | loff = *ppos & ~PAGE_MASK; | |
2166 | req_pages = (len + loff + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
a786c06d | 2167 | nr_pages = min(req_pages, spd.nr_pages_max); |
708e3508 | 2168 | |
708e3508 HD |
2169 | spd.nr_pages = find_get_pages_contig(mapping, index, |
2170 | nr_pages, spd.pages); | |
2171 | index += spd.nr_pages; | |
708e3508 | 2172 | error = 0; |
708e3508 | 2173 | |
71f0e07a | 2174 | while (spd.nr_pages < nr_pages) { |
9e18eb29 | 2175 | error = shmem_getpage(inode, index, &page, SGP_CACHE); |
71f0e07a HD |
2176 | if (error) |
2177 | break; | |
2178 | unlock_page(page); | |
708e3508 HD |
2179 | spd.pages[spd.nr_pages++] = page; |
2180 | index++; | |
2181 | } | |
2182 | ||
09cbfeaf | 2183 | index = *ppos >> PAGE_SHIFT; |
708e3508 HD |
2184 | nr_pages = spd.nr_pages; |
2185 | spd.nr_pages = 0; | |
71f0e07a | 2186 | |
708e3508 HD |
2187 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { |
2188 | unsigned int this_len; | |
2189 | ||
2190 | if (!len) | |
2191 | break; | |
2192 | ||
09cbfeaf | 2193 | this_len = min_t(unsigned long, len, PAGE_SIZE - loff); |
708e3508 HD |
2194 | page = spd.pages[page_nr]; |
2195 | ||
71f0e07a | 2196 | if (!PageUptodate(page) || page->mapping != mapping) { |
9e18eb29 | 2197 | error = shmem_getpage(inode, index, &page, SGP_CACHE); |
71f0e07a | 2198 | if (error) |
708e3508 | 2199 | break; |
71f0e07a | 2200 | unlock_page(page); |
09cbfeaf | 2201 | put_page(spd.pages[page_nr]); |
71f0e07a | 2202 | spd.pages[page_nr] = page; |
708e3508 | 2203 | } |
71f0e07a HD |
2204 | |
2205 | isize = i_size_read(inode); | |
09cbfeaf | 2206 | end_index = (isize - 1) >> PAGE_SHIFT; |
708e3508 HD |
2207 | if (unlikely(!isize || index > end_index)) |
2208 | break; | |
2209 | ||
708e3508 HD |
2210 | if (end_index == index) { |
2211 | unsigned int plen; | |
2212 | ||
09cbfeaf | 2213 | plen = ((isize - 1) & ~PAGE_MASK) + 1; |
708e3508 HD |
2214 | if (plen <= loff) |
2215 | break; | |
2216 | ||
708e3508 HD |
2217 | this_len = min(this_len, plen - loff); |
2218 | len = this_len; | |
2219 | } | |
2220 | ||
2221 | spd.partial[page_nr].offset = loff; | |
2222 | spd.partial[page_nr].len = this_len; | |
2223 | len -= this_len; | |
2224 | loff = 0; | |
2225 | spd.nr_pages++; | |
2226 | index++; | |
2227 | } | |
2228 | ||
708e3508 | 2229 | while (page_nr < nr_pages) |
09cbfeaf | 2230 | put_page(spd.pages[page_nr++]); |
708e3508 HD |
2231 | |
2232 | if (spd.nr_pages) | |
2233 | error = splice_to_pipe(pipe, &spd); | |
2234 | ||
047fe360 | 2235 | splice_shrink_spd(&spd); |
708e3508 HD |
2236 | |
2237 | if (error > 0) { | |
2238 | *ppos += error; | |
2239 | file_accessed(in); | |
2240 | } | |
2241 | return error; | |
2242 | } | |
2243 | ||
220f2ac9 HD |
2244 | /* |
2245 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
2246 | */ | |
2247 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2248 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2249 | { |
2250 | struct page *page; | |
2251 | struct pagevec pvec; | |
2252 | pgoff_t indices[PAGEVEC_SIZE]; | |
2253 | bool done = false; | |
2254 | int i; | |
2255 | ||
2256 | pagevec_init(&pvec, 0); | |
2257 | pvec.nr = 1; /* start small: we may be there already */ | |
2258 | while (!done) { | |
0cd6144a | 2259 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2260 | pvec.nr, pvec.pages, indices); |
2261 | if (!pvec.nr) { | |
965c8e59 | 2262 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2263 | index = end; |
2264 | break; | |
2265 | } | |
2266 | for (i = 0; i < pvec.nr; i++, index++) { | |
2267 | if (index < indices[i]) { | |
965c8e59 | 2268 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2269 | done = true; |
2270 | break; | |
2271 | } | |
2272 | index = indices[i]; | |
2273 | } | |
2274 | page = pvec.pages[i]; | |
2275 | if (page && !radix_tree_exceptional_entry(page)) { | |
2276 | if (!PageUptodate(page)) | |
2277 | page = NULL; | |
2278 | } | |
2279 | if (index >= end || | |
965c8e59 AM |
2280 | (page && whence == SEEK_DATA) || |
2281 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2282 | done = true; |
2283 | break; | |
2284 | } | |
2285 | } | |
0cd6144a | 2286 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2287 | pagevec_release(&pvec); |
2288 | pvec.nr = PAGEVEC_SIZE; | |
2289 | cond_resched(); | |
2290 | } | |
2291 | return index; | |
2292 | } | |
2293 | ||
965c8e59 | 2294 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2295 | { |
2296 | struct address_space *mapping = file->f_mapping; | |
2297 | struct inode *inode = mapping->host; | |
2298 | pgoff_t start, end; | |
2299 | loff_t new_offset; | |
2300 | ||
965c8e59 AM |
2301 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2302 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2303 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2304 | inode_lock(inode); |
220f2ac9 HD |
2305 | /* We're holding i_mutex so we can access i_size directly */ |
2306 | ||
2307 | if (offset < 0) | |
2308 | offset = -EINVAL; | |
2309 | else if (offset >= inode->i_size) | |
2310 | offset = -ENXIO; | |
2311 | else { | |
09cbfeaf KS |
2312 | start = offset >> PAGE_SHIFT; |
2313 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2314 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2315 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2316 | if (new_offset > offset) { |
2317 | if (new_offset < inode->i_size) | |
2318 | offset = new_offset; | |
965c8e59 | 2319 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2320 | offset = -ENXIO; |
2321 | else | |
2322 | offset = inode->i_size; | |
2323 | } | |
2324 | } | |
2325 | ||
387aae6f HD |
2326 | if (offset >= 0) |
2327 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2328 | inode_unlock(inode); |
220f2ac9 HD |
2329 | return offset; |
2330 | } | |
2331 | ||
05f65b5c DH |
2332 | /* |
2333 | * We need a tag: a new tag would expand every radix_tree_node by 8 bytes, | |
2334 | * so reuse a tag which we firmly believe is never set or cleared on shmem. | |
2335 | */ | |
2336 | #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE | |
2337 | #define LAST_SCAN 4 /* about 150ms max */ | |
2338 | ||
2339 | static void shmem_tag_pins(struct address_space *mapping) | |
2340 | { | |
2341 | struct radix_tree_iter iter; | |
2342 | void **slot; | |
2343 | pgoff_t start; | |
2344 | struct page *page; | |
2345 | ||
2346 | lru_add_drain(); | |
2347 | start = 0; | |
2348 | rcu_read_lock(); | |
2349 | ||
05f65b5c DH |
2350 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
2351 | page = radix_tree_deref_slot(slot); | |
2352 | if (!page || radix_tree_exception(page)) { | |
2cf938aa MW |
2353 | if (radix_tree_deref_retry(page)) { |
2354 | slot = radix_tree_iter_retry(&iter); | |
2355 | continue; | |
2356 | } | |
05f65b5c DH |
2357 | } else if (page_count(page) - page_mapcount(page) > 1) { |
2358 | spin_lock_irq(&mapping->tree_lock); | |
2359 | radix_tree_tag_set(&mapping->page_tree, iter.index, | |
2360 | SHMEM_TAG_PINNED); | |
2361 | spin_unlock_irq(&mapping->tree_lock); | |
2362 | } | |
2363 | ||
2364 | if (need_resched()) { | |
2365 | cond_resched_rcu(); | |
7165092f | 2366 | slot = radix_tree_iter_next(&iter); |
05f65b5c DH |
2367 | } |
2368 | } | |
2369 | rcu_read_unlock(); | |
2370 | } | |
2371 | ||
2372 | /* | |
2373 | * Setting SEAL_WRITE requires us to verify there's no pending writer. However, | |
2374 | * via get_user_pages(), drivers might have some pending I/O without any active | |
2375 | * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages | |
2376 | * and see whether it has an elevated ref-count. If so, we tag them and wait for | |
2377 | * them to be dropped. | |
2378 | * The caller must guarantee that no new user will acquire writable references | |
2379 | * to those pages to avoid races. | |
2380 | */ | |
40e041a2 DH |
2381 | static int shmem_wait_for_pins(struct address_space *mapping) |
2382 | { | |
05f65b5c DH |
2383 | struct radix_tree_iter iter; |
2384 | void **slot; | |
2385 | pgoff_t start; | |
2386 | struct page *page; | |
2387 | int error, scan; | |
2388 | ||
2389 | shmem_tag_pins(mapping); | |
2390 | ||
2391 | error = 0; | |
2392 | for (scan = 0; scan <= LAST_SCAN; scan++) { | |
2393 | if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED)) | |
2394 | break; | |
2395 | ||
2396 | if (!scan) | |
2397 | lru_add_drain_all(); | |
2398 | else if (schedule_timeout_killable((HZ << scan) / 200)) | |
2399 | scan = LAST_SCAN; | |
2400 | ||
2401 | start = 0; | |
2402 | rcu_read_lock(); | |
05f65b5c DH |
2403 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, |
2404 | start, SHMEM_TAG_PINNED) { | |
2405 | ||
2406 | page = radix_tree_deref_slot(slot); | |
2407 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2408 | if (radix_tree_deref_retry(page)) { |
2409 | slot = radix_tree_iter_retry(&iter); | |
2410 | continue; | |
2411 | } | |
05f65b5c DH |
2412 | |
2413 | page = NULL; | |
2414 | } | |
2415 | ||
2416 | if (page && | |
2417 | page_count(page) - page_mapcount(page) != 1) { | |
2418 | if (scan < LAST_SCAN) | |
2419 | goto continue_resched; | |
2420 | ||
2421 | /* | |
2422 | * On the last scan, we clean up all those tags | |
2423 | * we inserted; but make a note that we still | |
2424 | * found pages pinned. | |
2425 | */ | |
2426 | error = -EBUSY; | |
2427 | } | |
2428 | ||
2429 | spin_lock_irq(&mapping->tree_lock); | |
2430 | radix_tree_tag_clear(&mapping->page_tree, | |
2431 | iter.index, SHMEM_TAG_PINNED); | |
2432 | spin_unlock_irq(&mapping->tree_lock); | |
2433 | continue_resched: | |
2434 | if (need_resched()) { | |
2435 | cond_resched_rcu(); | |
7165092f | 2436 | slot = radix_tree_iter_next(&iter); |
05f65b5c DH |
2437 | } |
2438 | } | |
2439 | rcu_read_unlock(); | |
2440 | } | |
2441 | ||
2442 | return error; | |
40e041a2 DH |
2443 | } |
2444 | ||
2445 | #define F_ALL_SEALS (F_SEAL_SEAL | \ | |
2446 | F_SEAL_SHRINK | \ | |
2447 | F_SEAL_GROW | \ | |
2448 | F_SEAL_WRITE) | |
2449 | ||
2450 | int shmem_add_seals(struct file *file, unsigned int seals) | |
2451 | { | |
2452 | struct inode *inode = file_inode(file); | |
2453 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2454 | int error; | |
2455 | ||
2456 | /* | |
2457 | * SEALING | |
2458 | * Sealing allows multiple parties to share a shmem-file but restrict | |
2459 | * access to a specific subset of file operations. Seals can only be | |
2460 | * added, but never removed. This way, mutually untrusted parties can | |
2461 | * share common memory regions with a well-defined policy. A malicious | |
2462 | * peer can thus never perform unwanted operations on a shared object. | |
2463 | * | |
2464 | * Seals are only supported on special shmem-files and always affect | |
2465 | * the whole underlying inode. Once a seal is set, it may prevent some | |
2466 | * kinds of access to the file. Currently, the following seals are | |
2467 | * defined: | |
2468 | * SEAL_SEAL: Prevent further seals from being set on this file | |
2469 | * SEAL_SHRINK: Prevent the file from shrinking | |
2470 | * SEAL_GROW: Prevent the file from growing | |
2471 | * SEAL_WRITE: Prevent write access to the file | |
2472 | * | |
2473 | * As we don't require any trust relationship between two parties, we | |
2474 | * must prevent seals from being removed. Therefore, sealing a file | |
2475 | * only adds a given set of seals to the file, it never touches | |
2476 | * existing seals. Furthermore, the "setting seals"-operation can be | |
2477 | * sealed itself, which basically prevents any further seal from being | |
2478 | * added. | |
2479 | * | |
2480 | * Semantics of sealing are only defined on volatile files. Only | |
2481 | * anonymous shmem files support sealing. More importantly, seals are | |
2482 | * never written to disk. Therefore, there's no plan to support it on | |
2483 | * other file types. | |
2484 | */ | |
2485 | ||
2486 | if (file->f_op != &shmem_file_operations) | |
2487 | return -EINVAL; | |
2488 | if (!(file->f_mode & FMODE_WRITE)) | |
2489 | return -EPERM; | |
2490 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
2491 | return -EINVAL; | |
2492 | ||
5955102c | 2493 | inode_lock(inode); |
40e041a2 DH |
2494 | |
2495 | if (info->seals & F_SEAL_SEAL) { | |
2496 | error = -EPERM; | |
2497 | goto unlock; | |
2498 | } | |
2499 | ||
2500 | if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) { | |
2501 | error = mapping_deny_writable(file->f_mapping); | |
2502 | if (error) | |
2503 | goto unlock; | |
2504 | ||
2505 | error = shmem_wait_for_pins(file->f_mapping); | |
2506 | if (error) { | |
2507 | mapping_allow_writable(file->f_mapping); | |
2508 | goto unlock; | |
2509 | } | |
2510 | } | |
2511 | ||
2512 | info->seals |= seals; | |
2513 | error = 0; | |
2514 | ||
2515 | unlock: | |
5955102c | 2516 | inode_unlock(inode); |
40e041a2 DH |
2517 | return error; |
2518 | } | |
2519 | EXPORT_SYMBOL_GPL(shmem_add_seals); | |
2520 | ||
2521 | int shmem_get_seals(struct file *file) | |
2522 | { | |
2523 | if (file->f_op != &shmem_file_operations) | |
2524 | return -EINVAL; | |
2525 | ||
2526 | return SHMEM_I(file_inode(file))->seals; | |
2527 | } | |
2528 | EXPORT_SYMBOL_GPL(shmem_get_seals); | |
2529 | ||
2530 | long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg) | |
2531 | { | |
2532 | long error; | |
2533 | ||
2534 | switch (cmd) { | |
2535 | case F_ADD_SEALS: | |
2536 | /* disallow upper 32bit */ | |
2537 | if (arg > UINT_MAX) | |
2538 | return -EINVAL; | |
2539 | ||
2540 | error = shmem_add_seals(file, arg); | |
2541 | break; | |
2542 | case F_GET_SEALS: | |
2543 | error = shmem_get_seals(file); | |
2544 | break; | |
2545 | default: | |
2546 | error = -EINVAL; | |
2547 | break; | |
2548 | } | |
2549 | ||
2550 | return error; | |
2551 | } | |
2552 | ||
83e4fa9c HD |
2553 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2554 | loff_t len) | |
2555 | { | |
496ad9aa | 2556 | struct inode *inode = file_inode(file); |
e2d12e22 | 2557 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2558 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2559 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2560 | pgoff_t start, index, end; |
2561 | int error; | |
83e4fa9c | 2562 | |
13ace4d0 HD |
2563 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2564 | return -EOPNOTSUPP; | |
2565 | ||
5955102c | 2566 | inode_lock(inode); |
83e4fa9c HD |
2567 | |
2568 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2569 | struct address_space *mapping = file->f_mapping; | |
2570 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2571 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2572 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2573 | |
40e041a2 DH |
2574 | /* protected by i_mutex */ |
2575 | if (info->seals & F_SEAL_WRITE) { | |
2576 | error = -EPERM; | |
2577 | goto out; | |
2578 | } | |
2579 | ||
8e205f77 | 2580 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
2581 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
2582 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
2583 | spin_lock(&inode->i_lock); | |
2584 | inode->i_private = &shmem_falloc; | |
2585 | spin_unlock(&inode->i_lock); | |
2586 | ||
83e4fa9c HD |
2587 | if ((u64)unmap_end > (u64)unmap_start) |
2588 | unmap_mapping_range(mapping, unmap_start, | |
2589 | 1 + unmap_end - unmap_start, 0); | |
2590 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2591 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2592 | |
2593 | spin_lock(&inode->i_lock); | |
2594 | inode->i_private = NULL; | |
2595 | wake_up_all(&shmem_falloc_waitq); | |
2596 | spin_unlock(&inode->i_lock); | |
83e4fa9c | 2597 | error = 0; |
8e205f77 | 2598 | goto out; |
e2d12e22 HD |
2599 | } |
2600 | ||
2601 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2602 | error = inode_newsize_ok(inode, offset + len); | |
2603 | if (error) | |
2604 | goto out; | |
2605 | ||
40e041a2 DH |
2606 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2607 | error = -EPERM; | |
2608 | goto out; | |
2609 | } | |
2610 | ||
09cbfeaf KS |
2611 | start = offset >> PAGE_SHIFT; |
2612 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2613 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2614 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2615 | error = -ENOSPC; | |
2616 | goto out; | |
83e4fa9c HD |
2617 | } |
2618 | ||
8e205f77 | 2619 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2620 | shmem_falloc.start = start; |
2621 | shmem_falloc.next = start; | |
2622 | shmem_falloc.nr_falloced = 0; | |
2623 | shmem_falloc.nr_unswapped = 0; | |
2624 | spin_lock(&inode->i_lock); | |
2625 | inode->i_private = &shmem_falloc; | |
2626 | spin_unlock(&inode->i_lock); | |
2627 | ||
e2d12e22 HD |
2628 | for (index = start; index < end; index++) { |
2629 | struct page *page; | |
2630 | ||
2631 | /* | |
2632 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2633 | * been interrupted because we are using up too much memory. | |
2634 | */ | |
2635 | if (signal_pending(current)) | |
2636 | error = -EINTR; | |
1aac1400 HD |
2637 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2638 | error = -ENOMEM; | |
e2d12e22 | 2639 | else |
9e18eb29 | 2640 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2641 | if (error) { |
1635f6a7 | 2642 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2643 | if (index > start) { |
2644 | shmem_undo_range(inode, | |
2645 | (loff_t)start << PAGE_SHIFT, | |
2646 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2647 | } | |
1aac1400 | 2648 | goto undone; |
e2d12e22 HD |
2649 | } |
2650 | ||
1aac1400 HD |
2651 | /* |
2652 | * Inform shmem_writepage() how far we have reached. | |
2653 | * No need for lock or barrier: we have the page lock. | |
2654 | */ | |
2655 | shmem_falloc.next++; | |
2656 | if (!PageUptodate(page)) | |
2657 | shmem_falloc.nr_falloced++; | |
2658 | ||
e2d12e22 | 2659 | /* |
1635f6a7 HD |
2660 | * If !PageUptodate, leave it that way so that freeable pages |
2661 | * can be recognized if we need to rollback on error later. | |
2662 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2663 | * than free the pages we are allocating (and SGP_CACHE pages |
2664 | * might still be clean: we now need to mark those dirty too). | |
2665 | */ | |
2666 | set_page_dirty(page); | |
2667 | unlock_page(page); | |
09cbfeaf | 2668 | put_page(page); |
e2d12e22 HD |
2669 | cond_resched(); |
2670 | } | |
2671 | ||
2672 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2673 | i_size_write(inode, offset + len); | |
e2d12e22 | 2674 | inode->i_ctime = CURRENT_TIME; |
1aac1400 HD |
2675 | undone: |
2676 | spin_lock(&inode->i_lock); | |
2677 | inode->i_private = NULL; | |
2678 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2679 | out: |
5955102c | 2680 | inode_unlock(inode); |
83e4fa9c HD |
2681 | return error; |
2682 | } | |
2683 | ||
726c3342 | 2684 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2685 | { |
726c3342 | 2686 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2687 | |
2688 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2689 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2690 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2691 | if (sbinfo->max_blocks) { |
1da177e4 | 2692 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2693 | buf->f_bavail = |
2694 | buf->f_bfree = sbinfo->max_blocks - | |
2695 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2696 | } |
2697 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2698 | buf->f_files = sbinfo->max_inodes; |
2699 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2700 | } |
2701 | /* else leave those fields 0 like simple_statfs */ | |
2702 | return 0; | |
2703 | } | |
2704 | ||
2705 | /* | |
2706 | * File creation. Allocate an inode, and we're done.. | |
2707 | */ | |
2708 | static int | |
1a67aafb | 2709 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2710 | { |
0b0a0806 | 2711 | struct inode *inode; |
1da177e4 LT |
2712 | int error = -ENOSPC; |
2713 | ||
454abafe | 2714 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2715 | if (inode) { |
feda821e CH |
2716 | error = simple_acl_create(dir, inode); |
2717 | if (error) | |
2718 | goto out_iput; | |
2a7dba39 | 2719 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2720 | &dentry->d_name, |
6d9d88d0 | 2721 | shmem_initxattrs, NULL); |
feda821e CH |
2722 | if (error && error != -EOPNOTSUPP) |
2723 | goto out_iput; | |
37ec43cd | 2724 | |
718deb6b | 2725 | error = 0; |
1da177e4 LT |
2726 | dir->i_size += BOGO_DIRENT_SIZE; |
2727 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2728 | d_instantiate(dentry, inode); | |
2729 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2730 | } |
2731 | return error; | |
feda821e CH |
2732 | out_iput: |
2733 | iput(inode); | |
2734 | return error; | |
1da177e4 LT |
2735 | } |
2736 | ||
60545d0d AV |
2737 | static int |
2738 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2739 | { | |
2740 | struct inode *inode; | |
2741 | int error = -ENOSPC; | |
2742 | ||
2743 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
2744 | if (inode) { | |
2745 | error = security_inode_init_security(inode, dir, | |
2746 | NULL, | |
2747 | shmem_initxattrs, NULL); | |
feda821e CH |
2748 | if (error && error != -EOPNOTSUPP) |
2749 | goto out_iput; | |
2750 | error = simple_acl_create(dir, inode); | |
2751 | if (error) | |
2752 | goto out_iput; | |
60545d0d AV |
2753 | d_tmpfile(dentry, inode); |
2754 | } | |
2755 | return error; | |
feda821e CH |
2756 | out_iput: |
2757 | iput(inode); | |
2758 | return error; | |
60545d0d AV |
2759 | } |
2760 | ||
18bb1db3 | 2761 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2762 | { |
2763 | int error; | |
2764 | ||
2765 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
2766 | return error; | |
d8c76e6f | 2767 | inc_nlink(dir); |
1da177e4 LT |
2768 | return 0; |
2769 | } | |
2770 | ||
4acdaf27 | 2771 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2772 | bool excl) |
1da177e4 LT |
2773 | { |
2774 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
2775 | } | |
2776 | ||
2777 | /* | |
2778 | * Link a file.. | |
2779 | */ | |
2780 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2781 | { | |
75c3cfa8 | 2782 | struct inode *inode = d_inode(old_dentry); |
5b04c689 | 2783 | int ret; |
1da177e4 LT |
2784 | |
2785 | /* | |
2786 | * No ordinary (disk based) filesystem counts links as inodes; | |
2787 | * but each new link needs a new dentry, pinning lowmem, and | |
2788 | * tmpfs dentries cannot be pruned until they are unlinked. | |
2789 | */ | |
5b04c689 PE |
2790 | ret = shmem_reserve_inode(inode->i_sb); |
2791 | if (ret) | |
2792 | goto out; | |
1da177e4 LT |
2793 | |
2794 | dir->i_size += BOGO_DIRENT_SIZE; | |
2795 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 2796 | inc_nlink(inode); |
7de9c6ee | 2797 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
2798 | dget(dentry); /* Extra pinning count for the created dentry */ |
2799 | d_instantiate(dentry, inode); | |
5b04c689 PE |
2800 | out: |
2801 | return ret; | |
1da177e4 LT |
2802 | } |
2803 | ||
2804 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
2805 | { | |
75c3cfa8 | 2806 | struct inode *inode = d_inode(dentry); |
1da177e4 | 2807 | |
5b04c689 PE |
2808 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2809 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
2810 | |
2811 | dir->i_size -= BOGO_DIRENT_SIZE; | |
2812 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 2813 | drop_nlink(inode); |
1da177e4 LT |
2814 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
2815 | return 0; | |
2816 | } | |
2817 | ||
2818 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
2819 | { | |
2820 | if (!simple_empty(dentry)) | |
2821 | return -ENOTEMPTY; | |
2822 | ||
75c3cfa8 | 2823 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 2824 | drop_nlink(dir); |
1da177e4 LT |
2825 | return shmem_unlink(dir, dentry); |
2826 | } | |
2827 | ||
37456771 MS |
2828 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
2829 | { | |
e36cb0b8 DH |
2830 | bool old_is_dir = d_is_dir(old_dentry); |
2831 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
2832 | |
2833 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
2834 | if (old_is_dir) { | |
2835 | drop_nlink(old_dir); | |
2836 | inc_nlink(new_dir); | |
2837 | } else { | |
2838 | drop_nlink(new_dir); | |
2839 | inc_nlink(old_dir); | |
2840 | } | |
2841 | } | |
2842 | old_dir->i_ctime = old_dir->i_mtime = | |
2843 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 DH |
2844 | d_inode(old_dentry)->i_ctime = |
2845 | d_inode(new_dentry)->i_ctime = CURRENT_TIME; | |
37456771 MS |
2846 | |
2847 | return 0; | |
2848 | } | |
2849 | ||
46fdb794 MS |
2850 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
2851 | { | |
2852 | struct dentry *whiteout; | |
2853 | int error; | |
2854 | ||
2855 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
2856 | if (!whiteout) | |
2857 | return -ENOMEM; | |
2858 | ||
2859 | error = shmem_mknod(old_dir, whiteout, | |
2860 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
2861 | dput(whiteout); | |
2862 | if (error) | |
2863 | return error; | |
2864 | ||
2865 | /* | |
2866 | * Cheat and hash the whiteout while the old dentry is still in | |
2867 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
2868 | * | |
2869 | * d_lookup() will consistently find one of them at this point, | |
2870 | * not sure which one, but that isn't even important. | |
2871 | */ | |
2872 | d_rehash(whiteout); | |
2873 | return 0; | |
2874 | } | |
2875 | ||
1da177e4 LT |
2876 | /* |
2877 | * The VFS layer already does all the dentry stuff for rename, | |
2878 | * we just have to decrement the usage count for the target if | |
2879 | * it exists so that the VFS layer correctly free's it when it | |
2880 | * gets overwritten. | |
2881 | */ | |
3b69ff51 | 2882 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 2883 | { |
75c3cfa8 | 2884 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
2885 | int they_are_dirs = S_ISDIR(inode->i_mode); |
2886 | ||
46fdb794 | 2887 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
2888 | return -EINVAL; |
2889 | ||
37456771 MS |
2890 | if (flags & RENAME_EXCHANGE) |
2891 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
2892 | ||
1da177e4 LT |
2893 | if (!simple_empty(new_dentry)) |
2894 | return -ENOTEMPTY; | |
2895 | ||
46fdb794 MS |
2896 | if (flags & RENAME_WHITEOUT) { |
2897 | int error; | |
2898 | ||
2899 | error = shmem_whiteout(old_dir, old_dentry); | |
2900 | if (error) | |
2901 | return error; | |
2902 | } | |
2903 | ||
75c3cfa8 | 2904 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 2905 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 2906 | if (they_are_dirs) { |
75c3cfa8 | 2907 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 2908 | drop_nlink(old_dir); |
b928095b | 2909 | } |
1da177e4 | 2910 | } else if (they_are_dirs) { |
9a53c3a7 | 2911 | drop_nlink(old_dir); |
d8c76e6f | 2912 | inc_nlink(new_dir); |
1da177e4 LT |
2913 | } |
2914 | ||
2915 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
2916 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
2917 | old_dir->i_ctime = old_dir->i_mtime = | |
2918 | new_dir->i_ctime = new_dir->i_mtime = | |
2919 | inode->i_ctime = CURRENT_TIME; | |
2920 | return 0; | |
2921 | } | |
2922 | ||
2923 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
2924 | { | |
2925 | int error; | |
2926 | int len; | |
2927 | struct inode *inode; | |
9276aad6 | 2928 | struct page *page; |
1da177e4 LT |
2929 | struct shmem_inode_info *info; |
2930 | ||
2931 | len = strlen(symname) + 1; | |
09cbfeaf | 2932 | if (len > PAGE_SIZE) |
1da177e4 LT |
2933 | return -ENAMETOOLONG; |
2934 | ||
454abafe | 2935 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
2936 | if (!inode) |
2937 | return -ENOSPC; | |
2938 | ||
9d8f13ba | 2939 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 2940 | shmem_initxattrs, NULL); |
570bc1c2 SS |
2941 | if (error) { |
2942 | if (error != -EOPNOTSUPP) { | |
2943 | iput(inode); | |
2944 | return error; | |
2945 | } | |
2946 | error = 0; | |
2947 | } | |
2948 | ||
1da177e4 LT |
2949 | info = SHMEM_I(inode); |
2950 | inode->i_size = len-1; | |
69f07ec9 | 2951 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
2952 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
2953 | if (!inode->i_link) { | |
69f07ec9 HD |
2954 | iput(inode); |
2955 | return -ENOMEM; | |
2956 | } | |
2957 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 2958 | } else { |
e8ecde25 | 2959 | inode_nohighmem(inode); |
9e18eb29 | 2960 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
2961 | if (error) { |
2962 | iput(inode); | |
2963 | return error; | |
2964 | } | |
14fcc23f | 2965 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 2966 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 2967 | memcpy(page_address(page), symname, len); |
ec9516fb | 2968 | SetPageUptodate(page); |
1da177e4 | 2969 | set_page_dirty(page); |
6746aff7 | 2970 | unlock_page(page); |
09cbfeaf | 2971 | put_page(page); |
1da177e4 | 2972 | } |
1da177e4 LT |
2973 | dir->i_size += BOGO_DIRENT_SIZE; |
2974 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2975 | d_instantiate(dentry, inode); | |
2976 | dget(dentry); | |
2977 | return 0; | |
2978 | } | |
2979 | ||
fceef393 | 2980 | static void shmem_put_link(void *arg) |
1da177e4 | 2981 | { |
fceef393 AV |
2982 | mark_page_accessed(arg); |
2983 | put_page(arg); | |
1da177e4 LT |
2984 | } |
2985 | ||
6b255391 | 2986 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
2987 | struct inode *inode, |
2988 | struct delayed_call *done) | |
1da177e4 | 2989 | { |
1da177e4 | 2990 | struct page *page = NULL; |
6b255391 | 2991 | int error; |
6a6c9904 AV |
2992 | if (!dentry) { |
2993 | page = find_get_page(inode->i_mapping, 0); | |
2994 | if (!page) | |
2995 | return ERR_PTR(-ECHILD); | |
2996 | if (!PageUptodate(page)) { | |
2997 | put_page(page); | |
2998 | return ERR_PTR(-ECHILD); | |
2999 | } | |
3000 | } else { | |
9e18eb29 | 3001 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3002 | if (error) |
3003 | return ERR_PTR(error); | |
3004 | unlock_page(page); | |
3005 | } | |
fceef393 | 3006 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3007 | return page_address(page); |
1da177e4 LT |
3008 | } |
3009 | ||
b09e0fa4 | 3010 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3011 | /* |
b09e0fa4 EP |
3012 | * Superblocks without xattr inode operations may get some security.* xattr |
3013 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3014 | * like ACLs, we also need to implement the security.* handlers at |
3015 | * filesystem level, though. | |
3016 | */ | |
3017 | ||
6d9d88d0 JS |
3018 | /* |
3019 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3020 | */ | |
3021 | static int shmem_initxattrs(struct inode *inode, | |
3022 | const struct xattr *xattr_array, | |
3023 | void *fs_info) | |
3024 | { | |
3025 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3026 | const struct xattr *xattr; | |
38f38657 | 3027 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3028 | size_t len; |
3029 | ||
3030 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3031 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3032 | if (!new_xattr) |
3033 | return -ENOMEM; | |
3034 | ||
3035 | len = strlen(xattr->name) + 1; | |
3036 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3037 | GFP_KERNEL); | |
3038 | if (!new_xattr->name) { | |
3039 | kfree(new_xattr); | |
3040 | return -ENOMEM; | |
3041 | } | |
3042 | ||
3043 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3044 | XATTR_SECURITY_PREFIX_LEN); | |
3045 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3046 | xattr->name, len); | |
3047 | ||
38f38657 | 3048 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3049 | } |
3050 | ||
3051 | return 0; | |
3052 | } | |
3053 | ||
aa7c5241 | 3054 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3055 | struct dentry *unused, struct inode *inode, |
3056 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3057 | { |
b296821a | 3058 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3059 | |
aa7c5241 | 3060 | name = xattr_full_name(handler, name); |
38f38657 | 3061 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3062 | } |
3063 | ||
aa7c5241 | 3064 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
59301226 AV |
3065 | struct dentry *unused, struct inode *inode, |
3066 | const char *name, const void *value, | |
3067 | size_t size, int flags) | |
b09e0fa4 | 3068 | { |
59301226 | 3069 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3070 | |
aa7c5241 | 3071 | name = xattr_full_name(handler, name); |
38f38657 | 3072 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
3073 | } |
3074 | ||
aa7c5241 AG |
3075 | static const struct xattr_handler shmem_security_xattr_handler = { |
3076 | .prefix = XATTR_SECURITY_PREFIX, | |
3077 | .get = shmem_xattr_handler_get, | |
3078 | .set = shmem_xattr_handler_set, | |
3079 | }; | |
b09e0fa4 | 3080 | |
aa7c5241 AG |
3081 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3082 | .prefix = XATTR_TRUSTED_PREFIX, | |
3083 | .get = shmem_xattr_handler_get, | |
3084 | .set = shmem_xattr_handler_set, | |
3085 | }; | |
b09e0fa4 | 3086 | |
aa7c5241 AG |
3087 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3088 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3089 | &posix_acl_access_xattr_handler, | |
3090 | &posix_acl_default_xattr_handler, | |
3091 | #endif | |
3092 | &shmem_security_xattr_handler, | |
3093 | &shmem_trusted_xattr_handler, | |
3094 | NULL | |
3095 | }; | |
b09e0fa4 EP |
3096 | |
3097 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3098 | { | |
75c3cfa8 | 3099 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3100 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3101 | } |
3102 | #endif /* CONFIG_TMPFS_XATTR */ | |
3103 | ||
69f07ec9 | 3104 | static const struct inode_operations shmem_short_symlink_operations = { |
b09e0fa4 | 3105 | .readlink = generic_readlink, |
6b255391 | 3106 | .get_link = simple_get_link, |
b09e0fa4 | 3107 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3108 | .setxattr = generic_setxattr, |
3109 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3110 | .listxattr = shmem_listxattr, |
aa7c5241 | 3111 | .removexattr = generic_removexattr, |
b09e0fa4 EP |
3112 | #endif |
3113 | }; | |
3114 | ||
3115 | static const struct inode_operations shmem_symlink_inode_operations = { | |
3116 | .readlink = generic_readlink, | |
6b255391 | 3117 | .get_link = shmem_get_link, |
b09e0fa4 | 3118 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3119 | .setxattr = generic_setxattr, |
3120 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3121 | .listxattr = shmem_listxattr, |
aa7c5241 | 3122 | .removexattr = generic_removexattr, |
39f0247d | 3123 | #endif |
b09e0fa4 | 3124 | }; |
39f0247d | 3125 | |
91828a40 DG |
3126 | static struct dentry *shmem_get_parent(struct dentry *child) |
3127 | { | |
3128 | return ERR_PTR(-ESTALE); | |
3129 | } | |
3130 | ||
3131 | static int shmem_match(struct inode *ino, void *vfh) | |
3132 | { | |
3133 | __u32 *fh = vfh; | |
3134 | __u64 inum = fh[2]; | |
3135 | inum = (inum << 32) | fh[1]; | |
3136 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
3137 | } | |
3138 | ||
480b116c CH |
3139 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3140 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3141 | { |
91828a40 | 3142 | struct inode *inode; |
480b116c | 3143 | struct dentry *dentry = NULL; |
35c2a7f4 | 3144 | u64 inum; |
480b116c CH |
3145 | |
3146 | if (fh_len < 3) | |
3147 | return NULL; | |
91828a40 | 3148 | |
35c2a7f4 HD |
3149 | inum = fid->raw[2]; |
3150 | inum = (inum << 32) | fid->raw[1]; | |
3151 | ||
480b116c CH |
3152 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
3153 | shmem_match, fid->raw); | |
91828a40 | 3154 | if (inode) { |
480b116c | 3155 | dentry = d_find_alias(inode); |
91828a40 DG |
3156 | iput(inode); |
3157 | } | |
3158 | ||
480b116c | 3159 | return dentry; |
91828a40 DG |
3160 | } |
3161 | ||
b0b0382b AV |
3162 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3163 | struct inode *parent) | |
91828a40 | 3164 | { |
5fe0c237 AK |
3165 | if (*len < 3) { |
3166 | *len = 3; | |
94e07a75 | 3167 | return FILEID_INVALID; |
5fe0c237 | 3168 | } |
91828a40 | 3169 | |
1d3382cb | 3170 | if (inode_unhashed(inode)) { |
91828a40 DG |
3171 | /* Unfortunately insert_inode_hash is not idempotent, |
3172 | * so as we hash inodes here rather than at creation | |
3173 | * time, we need a lock to ensure we only try | |
3174 | * to do it once | |
3175 | */ | |
3176 | static DEFINE_SPINLOCK(lock); | |
3177 | spin_lock(&lock); | |
1d3382cb | 3178 | if (inode_unhashed(inode)) |
91828a40 DG |
3179 | __insert_inode_hash(inode, |
3180 | inode->i_ino + inode->i_generation); | |
3181 | spin_unlock(&lock); | |
3182 | } | |
3183 | ||
3184 | fh[0] = inode->i_generation; | |
3185 | fh[1] = inode->i_ino; | |
3186 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
3187 | ||
3188 | *len = 3; | |
3189 | return 1; | |
3190 | } | |
3191 | ||
39655164 | 3192 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3193 | .get_parent = shmem_get_parent, |
91828a40 | 3194 | .encode_fh = shmem_encode_fh, |
480b116c | 3195 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3196 | }; |
3197 | ||
680d794b | 3198 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
3199 | bool remount) | |
1da177e4 LT |
3200 | { |
3201 | char *this_char, *value, *rest; | |
49cd0a5c | 3202 | struct mempolicy *mpol = NULL; |
8751e039 EB |
3203 | uid_t uid; |
3204 | gid_t gid; | |
1da177e4 | 3205 | |
b00dc3ad HD |
3206 | while (options != NULL) { |
3207 | this_char = options; | |
3208 | for (;;) { | |
3209 | /* | |
3210 | * NUL-terminate this option: unfortunately, | |
3211 | * mount options form a comma-separated list, | |
3212 | * but mpol's nodelist may also contain commas. | |
3213 | */ | |
3214 | options = strchr(options, ','); | |
3215 | if (options == NULL) | |
3216 | break; | |
3217 | options++; | |
3218 | if (!isdigit(*options)) { | |
3219 | options[-1] = '\0'; | |
3220 | break; | |
3221 | } | |
3222 | } | |
1da177e4 LT |
3223 | if (!*this_char) |
3224 | continue; | |
3225 | if ((value = strchr(this_char,'=')) != NULL) { | |
3226 | *value++ = 0; | |
3227 | } else { | |
1170532b JP |
3228 | pr_err("tmpfs: No value for mount option '%s'\n", |
3229 | this_char); | |
49cd0a5c | 3230 | goto error; |
1da177e4 LT |
3231 | } |
3232 | ||
3233 | if (!strcmp(this_char,"size")) { | |
3234 | unsigned long long size; | |
3235 | size = memparse(value,&rest); | |
3236 | if (*rest == '%') { | |
3237 | size <<= PAGE_SHIFT; | |
3238 | size *= totalram_pages; | |
3239 | do_div(size, 100); | |
3240 | rest++; | |
3241 | } | |
3242 | if (*rest) | |
3243 | goto bad_val; | |
680d794b | 3244 | sbinfo->max_blocks = |
09cbfeaf | 3245 | DIV_ROUND_UP(size, PAGE_SIZE); |
1da177e4 | 3246 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 3247 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
3248 | if (*rest) |
3249 | goto bad_val; | |
3250 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 3251 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
3252 | if (*rest) |
3253 | goto bad_val; | |
3254 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 3255 | if (remount) |
1da177e4 | 3256 | continue; |
680d794b | 3257 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
3258 | if (*rest) |
3259 | goto bad_val; | |
3260 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 3261 | if (remount) |
1da177e4 | 3262 | continue; |
8751e039 | 3263 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3264 | if (*rest) |
3265 | goto bad_val; | |
8751e039 EB |
3266 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
3267 | if (!uid_valid(sbinfo->uid)) | |
3268 | goto bad_val; | |
1da177e4 | 3269 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 3270 | if (remount) |
1da177e4 | 3271 | continue; |
8751e039 | 3272 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3273 | if (*rest) |
3274 | goto bad_val; | |
8751e039 EB |
3275 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
3276 | if (!gid_valid(sbinfo->gid)) | |
3277 | goto bad_val; | |
5a6e75f8 KS |
3278 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3279 | } else if (!strcmp(this_char, "huge")) { | |
3280 | int huge; | |
3281 | huge = shmem_parse_huge(value); | |
3282 | if (huge < 0) | |
3283 | goto bad_val; | |
3284 | if (!has_transparent_hugepage() && | |
3285 | huge != SHMEM_HUGE_NEVER) | |
3286 | goto bad_val; | |
3287 | sbinfo->huge = huge; | |
3288 | #endif | |
3289 | #ifdef CONFIG_NUMA | |
7339ff83 | 3290 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
3291 | mpol_put(mpol); |
3292 | mpol = NULL; | |
3293 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 3294 | goto bad_val; |
5a6e75f8 | 3295 | #endif |
1da177e4 | 3296 | } else { |
1170532b | 3297 | pr_err("tmpfs: Bad mount option %s\n", this_char); |
49cd0a5c | 3298 | goto error; |
1da177e4 LT |
3299 | } |
3300 | } | |
49cd0a5c | 3301 | sbinfo->mpol = mpol; |
1da177e4 LT |
3302 | return 0; |
3303 | ||
3304 | bad_val: | |
1170532b | 3305 | pr_err("tmpfs: Bad value '%s' for mount option '%s'\n", |
1da177e4 | 3306 | value, this_char); |
49cd0a5c GT |
3307 | error: |
3308 | mpol_put(mpol); | |
1da177e4 LT |
3309 | return 1; |
3310 | ||
3311 | } | |
3312 | ||
3313 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
3314 | { | |
3315 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 3316 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
3317 | unsigned long inodes; |
3318 | int error = -EINVAL; | |
3319 | ||
5f00110f | 3320 | config.mpol = NULL; |
680d794b | 3321 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 3322 | return error; |
1da177e4 | 3323 | |
0edd73b3 | 3324 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3325 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 3326 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 3327 | goto out; |
680d794b | 3328 | if (config.max_inodes < inodes) |
0edd73b3 HD |
3329 | goto out; |
3330 | /* | |
54af6042 | 3331 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
3332 | * but we must separately disallow unlimited->limited, because |
3333 | * in that case we have no record of how much is already in use. | |
3334 | */ | |
680d794b | 3335 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 3336 | goto out; |
680d794b | 3337 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
3338 | goto out; |
3339 | ||
3340 | error = 0; | |
5a6e75f8 | 3341 | sbinfo->huge = config.huge; |
680d794b | 3342 | sbinfo->max_blocks = config.max_blocks; |
680d794b | 3343 | sbinfo->max_inodes = config.max_inodes; |
3344 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 3345 | |
5f00110f GT |
3346 | /* |
3347 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3348 | */ | |
3349 | if (config.mpol) { | |
3350 | mpol_put(sbinfo->mpol); | |
3351 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
3352 | } | |
0edd73b3 HD |
3353 | out: |
3354 | spin_unlock(&sbinfo->stat_lock); | |
3355 | return error; | |
1da177e4 | 3356 | } |
680d794b | 3357 | |
34c80b1d | 3358 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3359 | { |
34c80b1d | 3360 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b | 3361 | |
3362 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3363 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3364 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 3365 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
3366 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
3367 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 3368 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3369 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3370 | seq_printf(seq, ",uid=%u", | |
3371 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3372 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3373 | seq_printf(seq, ",gid=%u", | |
3374 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
5a6e75f8 KS |
3375 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3376 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ | |
3377 | if (sbinfo->huge) | |
3378 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3379 | #endif | |
71fe804b | 3380 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 3381 | return 0; |
3382 | } | |
9183df25 DH |
3383 | |
3384 | #define MFD_NAME_PREFIX "memfd:" | |
3385 | #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1) | |
3386 | #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN) | |
3387 | ||
3388 | #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING) | |
3389 | ||
3390 | SYSCALL_DEFINE2(memfd_create, | |
3391 | const char __user *, uname, | |
3392 | unsigned int, flags) | |
3393 | { | |
3394 | struct shmem_inode_info *info; | |
3395 | struct file *file; | |
3396 | int fd, error; | |
3397 | char *name; | |
3398 | long len; | |
3399 | ||
3400 | if (flags & ~(unsigned int)MFD_ALL_FLAGS) | |
3401 | return -EINVAL; | |
3402 | ||
3403 | /* length includes terminating zero */ | |
3404 | len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1); | |
3405 | if (len <= 0) | |
3406 | return -EFAULT; | |
3407 | if (len > MFD_NAME_MAX_LEN + 1) | |
3408 | return -EINVAL; | |
3409 | ||
3410 | name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY); | |
3411 | if (!name) | |
3412 | return -ENOMEM; | |
3413 | ||
3414 | strcpy(name, MFD_NAME_PREFIX); | |
3415 | if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) { | |
3416 | error = -EFAULT; | |
3417 | goto err_name; | |
3418 | } | |
3419 | ||
3420 | /* terminating-zero may have changed after strnlen_user() returned */ | |
3421 | if (name[len + MFD_NAME_PREFIX_LEN - 1]) { | |
3422 | error = -EFAULT; | |
3423 | goto err_name; | |
3424 | } | |
3425 | ||
3426 | fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0); | |
3427 | if (fd < 0) { | |
3428 | error = fd; | |
3429 | goto err_name; | |
3430 | } | |
3431 | ||
3432 | file = shmem_file_setup(name, 0, VM_NORESERVE); | |
3433 | if (IS_ERR(file)) { | |
3434 | error = PTR_ERR(file); | |
3435 | goto err_fd; | |
3436 | } | |
3437 | info = SHMEM_I(file_inode(file)); | |
3438 | file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; | |
3439 | file->f_flags |= O_RDWR | O_LARGEFILE; | |
3440 | if (flags & MFD_ALLOW_SEALING) | |
3441 | info->seals &= ~F_SEAL_SEAL; | |
3442 | ||
3443 | fd_install(fd, file); | |
3444 | kfree(name); | |
3445 | return fd; | |
3446 | ||
3447 | err_fd: | |
3448 | put_unused_fd(fd); | |
3449 | err_name: | |
3450 | kfree(name); | |
3451 | return error; | |
3452 | } | |
3453 | ||
680d794b | 3454 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3455 | |
3456 | static void shmem_put_super(struct super_block *sb) | |
3457 | { | |
602586a8 HD |
3458 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3459 | ||
3460 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 3461 | mpol_put(sbinfo->mpol); |
602586a8 | 3462 | kfree(sbinfo); |
1da177e4 LT |
3463 | sb->s_fs_info = NULL; |
3464 | } | |
3465 | ||
2b2af54a | 3466 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3467 | { |
3468 | struct inode *inode; | |
0edd73b3 | 3469 | struct shmem_sb_info *sbinfo; |
680d794b | 3470 | int err = -ENOMEM; |
3471 | ||
3472 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3473 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 3474 | L1_CACHE_BYTES), GFP_KERNEL); |
3475 | if (!sbinfo) | |
3476 | return -ENOMEM; | |
3477 | ||
680d794b | 3478 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
3479 | sbinfo->uid = current_fsuid(); |
3480 | sbinfo->gid = current_fsgid(); | |
680d794b | 3481 | sb->s_fs_info = sbinfo; |
1da177e4 | 3482 | |
0edd73b3 | 3483 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3484 | /* |
3485 | * Per default we only allow half of the physical ram per | |
3486 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3487 | * but the internal instance is left unlimited. | |
3488 | */ | |
ca4e0519 | 3489 | if (!(sb->s_flags & MS_KERNMOUNT)) { |
680d794b | 3490 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3491 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3492 | if (shmem_parse_options(data, sbinfo, false)) { | |
3493 | err = -EINVAL; | |
3494 | goto failed; | |
3495 | } | |
ca4e0519 AV |
3496 | } else { |
3497 | sb->s_flags |= MS_NOUSER; | |
1da177e4 | 3498 | } |
91828a40 | 3499 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 3500 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
3501 | #else |
3502 | sb->s_flags |= MS_NOUSER; | |
3503 | #endif | |
3504 | ||
0edd73b3 | 3505 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3506 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3507 | goto failed; |
680d794b | 3508 | sbinfo->free_inodes = sbinfo->max_inodes; |
0edd73b3 | 3509 | |
285b2c4f | 3510 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3511 | sb->s_blocksize = PAGE_SIZE; |
3512 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3513 | sb->s_magic = TMPFS_MAGIC; |
3514 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3515 | sb->s_time_gran = 1; |
b09e0fa4 | 3516 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3517 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3518 | #endif |
3519 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
3520 | sb->s_flags |= MS_POSIXACL; |
3521 | #endif | |
0edd73b3 | 3522 | |
454abafe | 3523 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3524 | if (!inode) |
3525 | goto failed; | |
680d794b | 3526 | inode->i_uid = sbinfo->uid; |
3527 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3528 | sb->s_root = d_make_root(inode); |
3529 | if (!sb->s_root) | |
48fde701 | 3530 | goto failed; |
1da177e4 LT |
3531 | return 0; |
3532 | ||
1da177e4 LT |
3533 | failed: |
3534 | shmem_put_super(sb); | |
3535 | return err; | |
3536 | } | |
3537 | ||
fcc234f8 | 3538 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3539 | |
3540 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3541 | { | |
41ffe5d5 HD |
3542 | struct shmem_inode_info *info; |
3543 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3544 | if (!info) | |
1da177e4 | 3545 | return NULL; |
41ffe5d5 | 3546 | return &info->vfs_inode; |
1da177e4 LT |
3547 | } |
3548 | ||
41ffe5d5 | 3549 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3550 | { |
3551 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84e710da AV |
3552 | if (S_ISLNK(inode->i_mode)) |
3553 | kfree(inode->i_link); | |
fa0d7e3d NP |
3554 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3555 | } | |
3556 | ||
1da177e4 LT |
3557 | static void shmem_destroy_inode(struct inode *inode) |
3558 | { | |
09208d15 | 3559 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3560 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3561 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3562 | } |
3563 | ||
41ffe5d5 | 3564 | static void shmem_init_inode(void *foo) |
1da177e4 | 3565 | { |
41ffe5d5 HD |
3566 | struct shmem_inode_info *info = foo; |
3567 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3568 | } |
3569 | ||
41ffe5d5 | 3570 | static int shmem_init_inodecache(void) |
1da177e4 LT |
3571 | { |
3572 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3573 | sizeof(struct shmem_inode_info), | |
5d097056 | 3574 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3575 | return 0; |
3576 | } | |
3577 | ||
41ffe5d5 | 3578 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3579 | { |
1a1d92c1 | 3580 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3581 | } |
3582 | ||
f5e54d6e | 3583 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3584 | .writepage = shmem_writepage, |
76719325 | 3585 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3586 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3587 | .write_begin = shmem_write_begin, |
3588 | .write_end = shmem_write_end, | |
1da177e4 | 3589 | #endif |
1c93923c | 3590 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3591 | .migratepage = migrate_page, |
1c93923c | 3592 | #endif |
aa261f54 | 3593 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3594 | }; |
3595 | ||
15ad7cdc | 3596 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3597 | .mmap = shmem_mmap, |
c01d5b30 | 3598 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3599 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3600 | .llseek = shmem_file_llseek, |
2ba5bbed | 3601 | .read_iter = shmem_file_read_iter, |
8174202b | 3602 | .write_iter = generic_file_write_iter, |
1b061d92 | 3603 | .fsync = noop_fsync, |
708e3508 | 3604 | .splice_read = shmem_file_splice_read, |
f6cb85d0 | 3605 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3606 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3607 | #endif |
3608 | }; | |
3609 | ||
92e1d5be | 3610 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3611 | .getattr = shmem_getattr, |
94c1e62d | 3612 | .setattr = shmem_setattr, |
b09e0fa4 | 3613 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3614 | .setxattr = generic_setxattr, |
3615 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3616 | .listxattr = shmem_listxattr, |
aa7c5241 | 3617 | .removexattr = generic_removexattr, |
feda821e | 3618 | .set_acl = simple_set_acl, |
b09e0fa4 | 3619 | #endif |
1da177e4 LT |
3620 | }; |
3621 | ||
92e1d5be | 3622 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3623 | #ifdef CONFIG_TMPFS |
3624 | .create = shmem_create, | |
3625 | .lookup = simple_lookup, | |
3626 | .link = shmem_link, | |
3627 | .unlink = shmem_unlink, | |
3628 | .symlink = shmem_symlink, | |
3629 | .mkdir = shmem_mkdir, | |
3630 | .rmdir = shmem_rmdir, | |
3631 | .mknod = shmem_mknod, | |
3b69ff51 | 3632 | .rename2 = shmem_rename2, |
60545d0d | 3633 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3634 | #endif |
b09e0fa4 | 3635 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3636 | .setxattr = generic_setxattr, |
3637 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3638 | .listxattr = shmem_listxattr, |
aa7c5241 | 3639 | .removexattr = generic_removexattr, |
b09e0fa4 | 3640 | #endif |
39f0247d | 3641 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3642 | .setattr = shmem_setattr, |
feda821e | 3643 | .set_acl = simple_set_acl, |
39f0247d AG |
3644 | #endif |
3645 | }; | |
3646 | ||
92e1d5be | 3647 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3648 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3649 | .setxattr = generic_setxattr, |
3650 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3651 | .listxattr = shmem_listxattr, |
aa7c5241 | 3652 | .removexattr = generic_removexattr, |
b09e0fa4 | 3653 | #endif |
39f0247d | 3654 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3655 | .setattr = shmem_setattr, |
feda821e | 3656 | .set_acl = simple_set_acl, |
39f0247d | 3657 | #endif |
1da177e4 LT |
3658 | }; |
3659 | ||
759b9775 | 3660 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
3661 | .alloc_inode = shmem_alloc_inode, |
3662 | .destroy_inode = shmem_destroy_inode, | |
3663 | #ifdef CONFIG_TMPFS | |
3664 | .statfs = shmem_statfs, | |
3665 | .remount_fs = shmem_remount_fs, | |
680d794b | 3666 | .show_options = shmem_show_options, |
1da177e4 | 3667 | #endif |
1f895f75 | 3668 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3669 | .drop_inode = generic_delete_inode, |
3670 | .put_super = shmem_put_super, | |
3671 | }; | |
3672 | ||
f0f37e2f | 3673 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3674 | .fault = shmem_fault, |
d7c17551 | 3675 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3676 | #ifdef CONFIG_NUMA |
3677 | .set_policy = shmem_set_policy, | |
3678 | .get_policy = shmem_get_policy, | |
3679 | #endif | |
3680 | }; | |
3681 | ||
3c26ff6e AV |
3682 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
3683 | int flags, const char *dev_name, void *data) | |
1da177e4 | 3684 | { |
3c26ff6e | 3685 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
3686 | } |
3687 | ||
41ffe5d5 | 3688 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3689 | .owner = THIS_MODULE, |
3690 | .name = "tmpfs", | |
3c26ff6e | 3691 | .mount = shmem_mount, |
1da177e4 | 3692 | .kill_sb = kill_litter_super, |
2b8576cb | 3693 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 3694 | }; |
1da177e4 | 3695 | |
41ffe5d5 | 3696 | int __init shmem_init(void) |
1da177e4 LT |
3697 | { |
3698 | int error; | |
3699 | ||
16203a7a RL |
3700 | /* If rootfs called this, don't re-init */ |
3701 | if (shmem_inode_cachep) | |
3702 | return 0; | |
3703 | ||
41ffe5d5 | 3704 | error = shmem_init_inodecache(); |
1da177e4 LT |
3705 | if (error) |
3706 | goto out3; | |
3707 | ||
41ffe5d5 | 3708 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 3709 | if (error) { |
1170532b | 3710 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
3711 | goto out2; |
3712 | } | |
95dc112a | 3713 | |
ca4e0519 | 3714 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
3715 | if (IS_ERR(shm_mnt)) { |
3716 | error = PTR_ERR(shm_mnt); | |
1170532b | 3717 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
3718 | goto out1; |
3719 | } | |
5a6e75f8 KS |
3720 | |
3721 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3722 | if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY) | |
3723 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; | |
3724 | else | |
3725 | shmem_huge = 0; /* just in case it was patched */ | |
3726 | #endif | |
1da177e4 LT |
3727 | return 0; |
3728 | ||
3729 | out1: | |
41ffe5d5 | 3730 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 3731 | out2: |
41ffe5d5 | 3732 | shmem_destroy_inodecache(); |
1da177e4 LT |
3733 | out3: |
3734 | shm_mnt = ERR_PTR(error); | |
3735 | return error; | |
3736 | } | |
853ac43a | 3737 | |
5a6e75f8 KS |
3738 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS) |
3739 | static ssize_t shmem_enabled_show(struct kobject *kobj, | |
3740 | struct kobj_attribute *attr, char *buf) | |
3741 | { | |
3742 | int values[] = { | |
3743 | SHMEM_HUGE_ALWAYS, | |
3744 | SHMEM_HUGE_WITHIN_SIZE, | |
3745 | SHMEM_HUGE_ADVISE, | |
3746 | SHMEM_HUGE_NEVER, | |
3747 | SHMEM_HUGE_DENY, | |
3748 | SHMEM_HUGE_FORCE, | |
3749 | }; | |
3750 | int i, count; | |
3751 | ||
3752 | for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) { | |
3753 | const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s "; | |
3754 | ||
3755 | count += sprintf(buf + count, fmt, | |
3756 | shmem_format_huge(values[i])); | |
3757 | } | |
3758 | buf[count - 1] = '\n'; | |
3759 | return count; | |
3760 | } | |
3761 | ||
3762 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
3763 | struct kobj_attribute *attr, const char *buf, size_t count) | |
3764 | { | |
3765 | char tmp[16]; | |
3766 | int huge; | |
3767 | ||
3768 | if (count + 1 > sizeof(tmp)) | |
3769 | return -EINVAL; | |
3770 | memcpy(tmp, buf, count); | |
3771 | tmp[count] = '\0'; | |
3772 | if (count && tmp[count - 1] == '\n') | |
3773 | tmp[count - 1] = '\0'; | |
3774 | ||
3775 | huge = shmem_parse_huge(tmp); | |
3776 | if (huge == -EINVAL) | |
3777 | return -EINVAL; | |
3778 | if (!has_transparent_hugepage() && | |
3779 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
3780 | return -EINVAL; | |
3781 | ||
3782 | shmem_huge = huge; | |
3783 | if (shmem_huge < SHMEM_HUGE_DENY) | |
3784 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; | |
3785 | return count; | |
3786 | } | |
3787 | ||
3788 | struct kobj_attribute shmem_enabled_attr = | |
3789 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
3790 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */ | |
3791 | ||
853ac43a MM |
3792 | #else /* !CONFIG_SHMEM */ |
3793 | ||
3794 | /* | |
3795 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
3796 | * | |
3797 | * This is intended for small system where the benefits of the full | |
3798 | * shmem code (swap-backed and resource-limited) are outweighed by | |
3799 | * their complexity. On systems without swap this code should be | |
3800 | * effectively equivalent, but much lighter weight. | |
3801 | */ | |
3802 | ||
41ffe5d5 | 3803 | static struct file_system_type shmem_fs_type = { |
853ac43a | 3804 | .name = "tmpfs", |
3c26ff6e | 3805 | .mount = ramfs_mount, |
853ac43a | 3806 | .kill_sb = kill_litter_super, |
2b8576cb | 3807 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
3808 | }; |
3809 | ||
41ffe5d5 | 3810 | int __init shmem_init(void) |
853ac43a | 3811 | { |
41ffe5d5 | 3812 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 3813 | |
41ffe5d5 | 3814 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
3815 | BUG_ON(IS_ERR(shm_mnt)); |
3816 | ||
3817 | return 0; | |
3818 | } | |
3819 | ||
41ffe5d5 | 3820 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
3821 | { |
3822 | return 0; | |
3823 | } | |
3824 | ||
3f96b79a HD |
3825 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
3826 | { | |
3827 | return 0; | |
3828 | } | |
3829 | ||
24513264 HD |
3830 | void shmem_unlock_mapping(struct address_space *mapping) |
3831 | { | |
3832 | } | |
3833 | ||
c01d5b30 HD |
3834 | #ifdef CONFIG_MMU |
3835 | unsigned long shmem_get_unmapped_area(struct file *file, | |
3836 | unsigned long addr, unsigned long len, | |
3837 | unsigned long pgoff, unsigned long flags) | |
3838 | { | |
3839 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
3840 | } | |
3841 | #endif | |
3842 | ||
41ffe5d5 | 3843 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 3844 | { |
41ffe5d5 | 3845 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
3846 | } |
3847 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
3848 | ||
0b0a0806 HD |
3849 | #define shmem_vm_ops generic_file_vm_ops |
3850 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 3851 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
3852 | #define shmem_acct_size(flags, size) 0 |
3853 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
3854 | |
3855 | #endif /* CONFIG_SHMEM */ | |
3856 | ||
3857 | /* common code */ | |
1da177e4 | 3858 | |
3451538a | 3859 | static struct dentry_operations anon_ops = { |
118b2302 | 3860 | .d_dname = simple_dname |
3451538a AV |
3861 | }; |
3862 | ||
c7277090 EP |
3863 | static struct file *__shmem_file_setup(const char *name, loff_t size, |
3864 | unsigned long flags, unsigned int i_flags) | |
1da177e4 | 3865 | { |
6b4d0b27 | 3866 | struct file *res; |
1da177e4 | 3867 | struct inode *inode; |
2c48b9c4 | 3868 | struct path path; |
3451538a | 3869 | struct super_block *sb; |
1da177e4 LT |
3870 | struct qstr this; |
3871 | ||
3872 | if (IS_ERR(shm_mnt)) | |
6b4d0b27 | 3873 | return ERR_CAST(shm_mnt); |
1da177e4 | 3874 | |
285b2c4f | 3875 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
3876 | return ERR_PTR(-EINVAL); |
3877 | ||
3878 | if (shmem_acct_size(flags, size)) | |
3879 | return ERR_PTR(-ENOMEM); | |
3880 | ||
6b4d0b27 | 3881 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
3882 | this.name = name; |
3883 | this.len = strlen(name); | |
3884 | this.hash = 0; /* will go */ | |
3451538a | 3885 | sb = shm_mnt->mnt_sb; |
66ee4b88 | 3886 | path.mnt = mntget(shm_mnt); |
3451538a | 3887 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 3888 | if (!path.dentry) |
1da177e4 | 3889 | goto put_memory; |
3451538a | 3890 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 3891 | |
6b4d0b27 | 3892 | res = ERR_PTR(-ENOSPC); |
3451538a | 3893 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 3894 | if (!inode) |
66ee4b88 | 3895 | goto put_memory; |
1da177e4 | 3896 | |
c7277090 | 3897 | inode->i_flags |= i_flags; |
2c48b9c4 | 3898 | d_instantiate(path.dentry, inode); |
1da177e4 | 3899 | inode->i_size = size; |
6d6b77f1 | 3900 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
3901 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
3902 | if (IS_ERR(res)) | |
66ee4b88 | 3903 | goto put_path; |
4b42af81 | 3904 | |
6b4d0b27 | 3905 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 3906 | &shmem_file_operations); |
6b4d0b27 | 3907 | if (IS_ERR(res)) |
66ee4b88 | 3908 | goto put_path; |
4b42af81 | 3909 | |
6b4d0b27 | 3910 | return res; |
1da177e4 | 3911 | |
1da177e4 LT |
3912 | put_memory: |
3913 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
3914 | put_path: |
3915 | path_put(&path); | |
6b4d0b27 | 3916 | return res; |
1da177e4 | 3917 | } |
c7277090 EP |
3918 | |
3919 | /** | |
3920 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
3921 | * kernel internal. There will be NO LSM permission checks against the | |
3922 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
3923 | * higher layer. The users are the big_key and shm implementations. LSM |
3924 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
3925 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
3926 | * @size: size to be set for the file | |
3927 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3928 | */ | |
3929 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
3930 | { | |
3931 | return __shmem_file_setup(name, size, flags, S_PRIVATE); | |
3932 | } | |
3933 | ||
3934 | /** | |
3935 | * shmem_file_setup - get an unlinked file living in tmpfs | |
3936 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
3937 | * @size: size to be set for the file | |
3938 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3939 | */ | |
3940 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
3941 | { | |
3942 | return __shmem_file_setup(name, size, flags, 0); | |
3943 | } | |
395e0ddc | 3944 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 3945 | |
46711810 | 3946 | /** |
1da177e4 | 3947 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
3948 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
3949 | */ | |
3950 | int shmem_zero_setup(struct vm_area_struct *vma) | |
3951 | { | |
3952 | struct file *file; | |
3953 | loff_t size = vma->vm_end - vma->vm_start; | |
3954 | ||
66fc1303 HD |
3955 | /* |
3956 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
3957 | * between XFS directory reading and selinux: since this file is only | |
3958 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
3959 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
3960 | */ | |
3961 | file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE); | |
1da177e4 LT |
3962 | if (IS_ERR(file)) |
3963 | return PTR_ERR(file); | |
3964 | ||
3965 | if (vma->vm_file) | |
3966 | fput(vma->vm_file); | |
3967 | vma->vm_file = file; | |
3968 | vma->vm_ops = &shmem_vm_ops; | |
3969 | return 0; | |
3970 | } | |
d9d90e5e HD |
3971 | |
3972 | /** | |
3973 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
3974 | * @mapping: the page's address_space | |
3975 | * @index: the page index | |
3976 | * @gfp: the page allocator flags to use if allocating | |
3977 | * | |
3978 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
3979 | * with any new page allocations done using the specified allocation flags. | |
3980 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
3981 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
3982 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
3983 | * | |
68da9f05 HD |
3984 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
3985 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
3986 | */ |
3987 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
3988 | pgoff_t index, gfp_t gfp) | |
3989 | { | |
68da9f05 HD |
3990 | #ifdef CONFIG_SHMEM |
3991 | struct inode *inode = mapping->host; | |
9276aad6 | 3992 | struct page *page; |
68da9f05 HD |
3993 | int error; |
3994 | ||
3995 | BUG_ON(mapping->a_ops != &shmem_aops); | |
9e18eb29 ALC |
3996 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
3997 | gfp, NULL, NULL); | |
68da9f05 HD |
3998 | if (error) |
3999 | page = ERR_PTR(error); | |
4000 | else | |
4001 | unlock_page(page); | |
4002 | return page; | |
4003 | #else | |
4004 | /* | |
4005 | * The tiny !SHMEM case uses ramfs without swap | |
4006 | */ | |
d9d90e5e | 4007 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4008 | #endif |
d9d90e5e HD |
4009 | } |
4010 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |