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> | |
174cd4b1 | 32 | #include <linux/sched/signal.h> |
b95f1b31 | 33 | #include <linux/export.h> |
853ac43a | 34 | #include <linux/swap.h> |
e2e40f2c | 35 | #include <linux/uio.h> |
f3f0e1d2 | 36 | #include <linux/khugepaged.h> |
749df87b | 37 | #include <linux/hugetlb.h> |
853ac43a | 38 | |
95cc09d6 AA |
39 | #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */ |
40 | ||
853ac43a MM |
41 | static struct vfsmount *shm_mnt; |
42 | ||
43 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
44 | /* |
45 | * This virtual memory filesystem is heavily based on the ramfs. It | |
46 | * extends ramfs by the ability to use swap and honor resource limits | |
47 | * which makes it a completely usable filesystem. | |
48 | */ | |
49 | ||
39f0247d | 50 | #include <linux/xattr.h> |
a5694255 | 51 | #include <linux/exportfs.h> |
1c7c474c | 52 | #include <linux/posix_acl.h> |
feda821e | 53 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 54 | #include <linux/mman.h> |
1da177e4 LT |
55 | #include <linux/string.h> |
56 | #include <linux/slab.h> | |
57 | #include <linux/backing-dev.h> | |
58 | #include <linux/shmem_fs.h> | |
1da177e4 | 59 | #include <linux/writeback.h> |
1da177e4 | 60 | #include <linux/blkdev.h> |
bda97eab | 61 | #include <linux/pagevec.h> |
41ffe5d5 | 62 | #include <linux/percpu_counter.h> |
83e4fa9c | 63 | #include <linux/falloc.h> |
708e3508 | 64 | #include <linux/splice.h> |
1da177e4 LT |
65 | #include <linux/security.h> |
66 | #include <linux/swapops.h> | |
67 | #include <linux/mempolicy.h> | |
68 | #include <linux/namei.h> | |
b00dc3ad | 69 | #include <linux/ctype.h> |
304dbdb7 | 70 | #include <linux/migrate.h> |
c1f60a5a | 71 | #include <linux/highmem.h> |
680d794b | 72 | #include <linux/seq_file.h> |
92562927 | 73 | #include <linux/magic.h> |
9183df25 | 74 | #include <linux/syscalls.h> |
40e041a2 | 75 | #include <linux/fcntl.h> |
9183df25 | 76 | #include <uapi/linux/memfd.h> |
cfda0526 | 77 | #include <linux/userfaultfd_k.h> |
4c27fe4c | 78 | #include <linux/rmap.h> |
2b4db796 | 79 | #include <linux/uuid.h> |
304dbdb7 | 80 | |
7c0f6ba6 | 81 | #include <linux/uaccess.h> |
1da177e4 LT |
82 | #include <asm/pgtable.h> |
83 | ||
dd56b046 MG |
84 | #include "internal.h" |
85 | ||
09cbfeaf KS |
86 | #define BLOCKS_PER_PAGE (PAGE_SIZE/512) |
87 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT) | |
1da177e4 | 88 | |
1da177e4 LT |
89 | /* Pretend that each entry is of this size in directory's i_size */ |
90 | #define BOGO_DIRENT_SIZE 20 | |
91 | ||
69f07ec9 HD |
92 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
93 | #define SHORT_SYMLINK_LEN 128 | |
94 | ||
1aac1400 | 95 | /* |
f00cdc6d HD |
96 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
97 | * inode->i_private (with i_mutex making sure that it has only one user at | |
98 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
99 | */ |
100 | struct shmem_falloc { | |
8e205f77 | 101 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
102 | pgoff_t start; /* start of range currently being fallocated */ |
103 | pgoff_t next; /* the next page offset to be fallocated */ | |
104 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
105 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
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 | 124 | struct page **pagep, enum sgp_type sgp, |
cfda0526 MR |
125 | gfp_t gfp, struct vm_area_struct *vma, |
126 | struct vm_fault *vmf, int *fault_type); | |
68da9f05 | 127 | |
f3f0e1d2 | 128 | int shmem_getpage(struct inode *inode, pgoff_t index, |
9e18eb29 | 129 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
130 | { |
131 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
cfda0526 | 132 | mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL); |
68da9f05 | 133 | } |
1da177e4 | 134 | |
1da177e4 LT |
135 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
136 | { | |
137 | return sb->s_fs_info; | |
138 | } | |
139 | ||
140 | /* | |
141 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
142 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
143 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
144 | * consistent with the pre-accounting of private mappings ... | |
145 | */ | |
146 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
147 | { | |
0b0a0806 | 148 | return (flags & VM_NORESERVE) ? |
191c5424 | 149 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
150 | } |
151 | ||
152 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
153 | { | |
0b0a0806 | 154 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
155 | vm_unacct_memory(VM_ACCT(size)); |
156 | } | |
157 | ||
77142517 KK |
158 | static inline int shmem_reacct_size(unsigned long flags, |
159 | loff_t oldsize, loff_t newsize) | |
160 | { | |
161 | if (!(flags & VM_NORESERVE)) { | |
162 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
163 | return security_vm_enough_memory_mm(current->mm, | |
164 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
165 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
166 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
167 | } | |
168 | return 0; | |
169 | } | |
170 | ||
1da177e4 LT |
171 | /* |
172 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 173 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
174 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
175 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
176 | */ | |
800d8c63 | 177 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 178 | { |
800d8c63 KS |
179 | if (!(flags & VM_NORESERVE)) |
180 | return 0; | |
181 | ||
182 | return security_vm_enough_memory_mm(current->mm, | |
183 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
184 | } |
185 | ||
186 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
187 | { | |
0b0a0806 | 188 | if (flags & VM_NORESERVE) |
09cbfeaf | 189 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
190 | } |
191 | ||
0f079694 MR |
192 | static inline bool shmem_inode_acct_block(struct inode *inode, long pages) |
193 | { | |
194 | struct shmem_inode_info *info = SHMEM_I(inode); | |
195 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
196 | ||
197 | if (shmem_acct_block(info->flags, pages)) | |
198 | return false; | |
199 | ||
200 | if (sbinfo->max_blocks) { | |
201 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
202 | sbinfo->max_blocks - pages) > 0) | |
203 | goto unacct; | |
204 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
205 | } | |
206 | ||
207 | return true; | |
208 | ||
209 | unacct: | |
210 | shmem_unacct_blocks(info->flags, pages); | |
211 | return false; | |
212 | } | |
213 | ||
214 | static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages) | |
215 | { | |
216 | struct shmem_inode_info *info = SHMEM_I(inode); | |
217 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
218 | ||
219 | if (sbinfo->max_blocks) | |
220 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
221 | shmem_unacct_blocks(info->flags, pages); | |
222 | } | |
223 | ||
759b9775 | 224 | static const struct super_operations shmem_ops; |
f5e54d6e | 225 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 226 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
227 | static const struct inode_operations shmem_inode_operations; |
228 | static const struct inode_operations shmem_dir_inode_operations; | |
229 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 230 | static const struct vm_operations_struct shmem_vm_ops; |
779750d2 | 231 | static struct file_system_type shmem_fs_type; |
1da177e4 | 232 | |
b0506e48 MR |
233 | bool vma_is_shmem(struct vm_area_struct *vma) |
234 | { | |
235 | return vma->vm_ops == &shmem_vm_ops; | |
236 | } | |
237 | ||
1da177e4 | 238 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 239 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 240 | |
5b04c689 PE |
241 | static int shmem_reserve_inode(struct super_block *sb) |
242 | { | |
243 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
244 | if (sbinfo->max_inodes) { | |
245 | spin_lock(&sbinfo->stat_lock); | |
246 | if (!sbinfo->free_inodes) { | |
247 | spin_unlock(&sbinfo->stat_lock); | |
248 | return -ENOSPC; | |
249 | } | |
250 | sbinfo->free_inodes--; | |
251 | spin_unlock(&sbinfo->stat_lock); | |
252 | } | |
253 | return 0; | |
254 | } | |
255 | ||
256 | static void shmem_free_inode(struct super_block *sb) | |
257 | { | |
258 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
259 | if (sbinfo->max_inodes) { | |
260 | spin_lock(&sbinfo->stat_lock); | |
261 | sbinfo->free_inodes++; | |
262 | spin_unlock(&sbinfo->stat_lock); | |
263 | } | |
264 | } | |
265 | ||
46711810 | 266 | /** |
41ffe5d5 | 267 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
268 | * @inode: inode to recalc |
269 | * | |
270 | * We have to calculate the free blocks since the mm can drop | |
271 | * undirtied hole pages behind our back. | |
272 | * | |
273 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
274 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
275 | * | |
276 | * It has to be called with the spinlock held. | |
277 | */ | |
278 | static void shmem_recalc_inode(struct inode *inode) | |
279 | { | |
280 | struct shmem_inode_info *info = SHMEM_I(inode); | |
281 | long freed; | |
282 | ||
283 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
284 | if (freed > 0) { | |
285 | info->alloced -= freed; | |
54af6042 | 286 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
0f079694 | 287 | shmem_inode_unacct_blocks(inode, freed); |
1da177e4 LT |
288 | } |
289 | } | |
290 | ||
800d8c63 KS |
291 | bool shmem_charge(struct inode *inode, long pages) |
292 | { | |
293 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 294 | unsigned long flags; |
800d8c63 | 295 | |
0f079694 | 296 | if (!shmem_inode_acct_block(inode, pages)) |
800d8c63 | 297 | return false; |
b1cc94ab | 298 | |
4595ef88 | 299 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
300 | info->alloced += pages; |
301 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
302 | shmem_recalc_inode(inode); | |
4595ef88 | 303 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
304 | inode->i_mapping->nrpages += pages; |
305 | ||
800d8c63 KS |
306 | return true; |
307 | } | |
308 | ||
309 | void shmem_uncharge(struct inode *inode, long pages) | |
310 | { | |
311 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 312 | unsigned long flags; |
800d8c63 | 313 | |
4595ef88 | 314 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
315 | info->alloced -= pages; |
316 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
317 | shmem_recalc_inode(inode); | |
4595ef88 | 318 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 | 319 | |
0f079694 | 320 | shmem_inode_unacct_blocks(inode, pages); |
800d8c63 KS |
321 | } |
322 | ||
7a5d0fbb HD |
323 | /* |
324 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
325 | */ | |
326 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
327 | pgoff_t index, void *expected, void *replacement) | |
328 | { | |
f7942430 | 329 | struct radix_tree_node *node; |
5b9c98f3 | 330 | void __rcu **pslot; |
6dbaf22c | 331 | void *item; |
7a5d0fbb HD |
332 | |
333 | VM_BUG_ON(!expected); | |
6dbaf22c | 334 | VM_BUG_ON(!replacement); |
b93b0163 | 335 | item = __radix_tree_lookup(&mapping->i_pages, index, &node, &pslot); |
f7942430 | 336 | if (!item) |
6dbaf22c | 337 | return -ENOENT; |
7a5d0fbb HD |
338 | if (item != expected) |
339 | return -ENOENT; | |
b93b0163 | 340 | __radix_tree_replace(&mapping->i_pages, node, pslot, |
c7df8ad2 | 341 | replacement, NULL); |
7a5d0fbb HD |
342 | return 0; |
343 | } | |
344 | ||
d1899228 HD |
345 | /* |
346 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
347 | * that an entry was not already brought back from swap by a racing thread. | |
348 | * | |
349 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
350 | * might be reused, and again be SwapCache, using the same swap as before. | |
351 | */ | |
352 | static bool shmem_confirm_swap(struct address_space *mapping, | |
353 | pgoff_t index, swp_entry_t swap) | |
354 | { | |
355 | void *item; | |
356 | ||
357 | rcu_read_lock(); | |
b93b0163 | 358 | item = radix_tree_lookup(&mapping->i_pages, index); |
d1899228 HD |
359 | rcu_read_unlock(); |
360 | return item == swp_to_radix_entry(swap); | |
361 | } | |
362 | ||
5a6e75f8 KS |
363 | /* |
364 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
365 | * | |
366 | * SHMEM_HUGE_NEVER: | |
367 | * disables huge pages for the mount; | |
368 | * SHMEM_HUGE_ALWAYS: | |
369 | * enables huge pages for the mount; | |
370 | * SHMEM_HUGE_WITHIN_SIZE: | |
371 | * only allocate huge pages if the page will be fully within i_size, | |
372 | * also respect fadvise()/madvise() hints; | |
373 | * SHMEM_HUGE_ADVISE: | |
374 | * only allocate huge pages if requested with fadvise()/madvise(); | |
375 | */ | |
376 | ||
377 | #define SHMEM_HUGE_NEVER 0 | |
378 | #define SHMEM_HUGE_ALWAYS 1 | |
379 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
380 | #define SHMEM_HUGE_ADVISE 3 | |
381 | ||
382 | /* | |
383 | * Special values. | |
384 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
385 | * | |
386 | * SHMEM_HUGE_DENY: | |
387 | * disables huge on shm_mnt and all mounts, for emergency use; | |
388 | * SHMEM_HUGE_FORCE: | |
389 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
390 | * | |
391 | */ | |
392 | #define SHMEM_HUGE_DENY (-1) | |
393 | #define SHMEM_HUGE_FORCE (-2) | |
394 | ||
e496cf3d | 395 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
396 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
397 | ||
5b9c98f3 | 398 | static int shmem_huge __read_mostly; |
5a6e75f8 | 399 | |
f1f5929c | 400 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
401 | static int shmem_parse_huge(const char *str) |
402 | { | |
403 | if (!strcmp(str, "never")) | |
404 | return SHMEM_HUGE_NEVER; | |
405 | if (!strcmp(str, "always")) | |
406 | return SHMEM_HUGE_ALWAYS; | |
407 | if (!strcmp(str, "within_size")) | |
408 | return SHMEM_HUGE_WITHIN_SIZE; | |
409 | if (!strcmp(str, "advise")) | |
410 | return SHMEM_HUGE_ADVISE; | |
411 | if (!strcmp(str, "deny")) | |
412 | return SHMEM_HUGE_DENY; | |
413 | if (!strcmp(str, "force")) | |
414 | return SHMEM_HUGE_FORCE; | |
415 | return -EINVAL; | |
416 | } | |
417 | ||
418 | static const char *shmem_format_huge(int huge) | |
419 | { | |
420 | switch (huge) { | |
421 | case SHMEM_HUGE_NEVER: | |
422 | return "never"; | |
423 | case SHMEM_HUGE_ALWAYS: | |
424 | return "always"; | |
425 | case SHMEM_HUGE_WITHIN_SIZE: | |
426 | return "within_size"; | |
427 | case SHMEM_HUGE_ADVISE: | |
428 | return "advise"; | |
429 | case SHMEM_HUGE_DENY: | |
430 | return "deny"; | |
431 | case SHMEM_HUGE_FORCE: | |
432 | return "force"; | |
433 | default: | |
434 | VM_BUG_ON(1); | |
435 | return "bad_val"; | |
436 | } | |
437 | } | |
f1f5929c | 438 | #endif |
5a6e75f8 | 439 | |
779750d2 KS |
440 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
441 | struct shrink_control *sc, unsigned long nr_to_split) | |
442 | { | |
443 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 444 | LIST_HEAD(to_remove); |
779750d2 KS |
445 | struct inode *inode; |
446 | struct shmem_inode_info *info; | |
447 | struct page *page; | |
448 | unsigned long batch = sc ? sc->nr_to_scan : 128; | |
449 | int removed = 0, split = 0; | |
450 | ||
451 | if (list_empty(&sbinfo->shrinklist)) | |
452 | return SHRINK_STOP; | |
453 | ||
454 | spin_lock(&sbinfo->shrinklist_lock); | |
455 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
456 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
457 | ||
458 | /* pin the inode */ | |
459 | inode = igrab(&info->vfs_inode); | |
460 | ||
461 | /* inode is about to be evicted */ | |
462 | if (!inode) { | |
463 | list_del_init(&info->shrinklist); | |
464 | removed++; | |
465 | goto next; | |
466 | } | |
467 | ||
468 | /* Check if there's anything to gain */ | |
469 | if (round_up(inode->i_size, PAGE_SIZE) == | |
470 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 471 | list_move(&info->shrinklist, &to_remove); |
779750d2 | 472 | removed++; |
779750d2 KS |
473 | goto next; |
474 | } | |
475 | ||
476 | list_move(&info->shrinklist, &list); | |
477 | next: | |
478 | if (!--batch) | |
479 | break; | |
480 | } | |
481 | spin_unlock(&sbinfo->shrinklist_lock); | |
482 | ||
253fd0f0 KS |
483 | list_for_each_safe(pos, next, &to_remove) { |
484 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
485 | inode = &info->vfs_inode; | |
486 | list_del_init(&info->shrinklist); | |
487 | iput(inode); | |
488 | } | |
489 | ||
779750d2 KS |
490 | list_for_each_safe(pos, next, &list) { |
491 | int ret; | |
492 | ||
493 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
494 | inode = &info->vfs_inode; | |
495 | ||
b3cd54b2 KS |
496 | if (nr_to_split && split >= nr_to_split) |
497 | goto leave; | |
779750d2 | 498 | |
b3cd54b2 | 499 | page = find_get_page(inode->i_mapping, |
779750d2 KS |
500 | (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT); |
501 | if (!page) | |
502 | goto drop; | |
503 | ||
b3cd54b2 | 504 | /* No huge page at the end of the file: nothing to split */ |
779750d2 | 505 | if (!PageTransHuge(page)) { |
779750d2 KS |
506 | put_page(page); |
507 | goto drop; | |
508 | } | |
509 | ||
b3cd54b2 KS |
510 | /* |
511 | * Leave the inode on the list if we failed to lock | |
512 | * the page at this time. | |
513 | * | |
514 | * Waiting for the lock may lead to deadlock in the | |
515 | * reclaim path. | |
516 | */ | |
517 | if (!trylock_page(page)) { | |
518 | put_page(page); | |
519 | goto leave; | |
520 | } | |
521 | ||
779750d2 KS |
522 | ret = split_huge_page(page); |
523 | unlock_page(page); | |
524 | put_page(page); | |
525 | ||
b3cd54b2 KS |
526 | /* If split failed leave the inode on the list */ |
527 | if (ret) | |
528 | goto leave; | |
779750d2 KS |
529 | |
530 | split++; | |
531 | drop: | |
532 | list_del_init(&info->shrinklist); | |
533 | removed++; | |
b3cd54b2 | 534 | leave: |
779750d2 KS |
535 | iput(inode); |
536 | } | |
537 | ||
538 | spin_lock(&sbinfo->shrinklist_lock); | |
539 | list_splice_tail(&list, &sbinfo->shrinklist); | |
540 | sbinfo->shrinklist_len -= removed; | |
541 | spin_unlock(&sbinfo->shrinklist_lock); | |
542 | ||
543 | return split; | |
544 | } | |
545 | ||
546 | static long shmem_unused_huge_scan(struct super_block *sb, | |
547 | struct shrink_control *sc) | |
548 | { | |
549 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
550 | ||
551 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
552 | return SHRINK_STOP; | |
553 | ||
554 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
555 | } | |
556 | ||
557 | static long shmem_unused_huge_count(struct super_block *sb, | |
558 | struct shrink_control *sc) | |
559 | { | |
560 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
561 | return READ_ONCE(sbinfo->shrinklist_len); | |
562 | } | |
e496cf3d | 563 | #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 KS |
564 | |
565 | #define shmem_huge SHMEM_HUGE_DENY | |
566 | ||
779750d2 KS |
567 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
568 | struct shrink_control *sc, unsigned long nr_to_split) | |
569 | { | |
570 | return 0; | |
571 | } | |
e496cf3d | 572 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 573 | |
46f65ec1 HD |
574 | /* |
575 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
576 | */ | |
577 | static int shmem_add_to_page_cache(struct page *page, | |
578 | struct address_space *mapping, | |
fed400a1 | 579 | pgoff_t index, void *expected) |
46f65ec1 | 580 | { |
800d8c63 | 581 | int error, nr = hpage_nr_pages(page); |
46f65ec1 | 582 | |
800d8c63 KS |
583 | VM_BUG_ON_PAGE(PageTail(page), page); |
584 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
585 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
586 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 587 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 588 | |
800d8c63 | 589 | page_ref_add(page, nr); |
b065b432 HD |
590 | page->mapping = mapping; |
591 | page->index = index; | |
592 | ||
b93b0163 | 593 | xa_lock_irq(&mapping->i_pages); |
800d8c63 KS |
594 | if (PageTransHuge(page)) { |
595 | void __rcu **results; | |
596 | pgoff_t idx; | |
597 | int i; | |
598 | ||
599 | error = 0; | |
b93b0163 | 600 | if (radix_tree_gang_lookup_slot(&mapping->i_pages, |
800d8c63 KS |
601 | &results, &idx, index, 1) && |
602 | idx < index + HPAGE_PMD_NR) { | |
603 | error = -EEXIST; | |
604 | } | |
605 | ||
606 | if (!error) { | |
607 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
b93b0163 | 608 | error = radix_tree_insert(&mapping->i_pages, |
800d8c63 KS |
609 | index + i, page + i); |
610 | VM_BUG_ON(error); | |
611 | } | |
612 | count_vm_event(THP_FILE_ALLOC); | |
613 | } | |
614 | } else if (!expected) { | |
b93b0163 | 615 | error = radix_tree_insert(&mapping->i_pages, index, page); |
800d8c63 | 616 | } else { |
b065b432 HD |
617 | error = shmem_radix_tree_replace(mapping, index, expected, |
618 | page); | |
800d8c63 KS |
619 | } |
620 | ||
46f65ec1 | 621 | if (!error) { |
800d8c63 KS |
622 | mapping->nrpages += nr; |
623 | if (PageTransHuge(page)) | |
11fb9989 MG |
624 | __inc_node_page_state(page, NR_SHMEM_THPS); |
625 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); | |
626 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr); | |
b93b0163 | 627 | xa_unlock_irq(&mapping->i_pages); |
b065b432 HD |
628 | } else { |
629 | page->mapping = NULL; | |
b93b0163 | 630 | xa_unlock_irq(&mapping->i_pages); |
800d8c63 | 631 | page_ref_sub(page, nr); |
46f65ec1 | 632 | } |
46f65ec1 HD |
633 | return error; |
634 | } | |
635 | ||
6922c0c7 HD |
636 | /* |
637 | * Like delete_from_page_cache, but substitutes swap for page. | |
638 | */ | |
639 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
640 | { | |
641 | struct address_space *mapping = page->mapping; | |
642 | int error; | |
643 | ||
800d8c63 KS |
644 | VM_BUG_ON_PAGE(PageCompound(page), page); |
645 | ||
b93b0163 | 646 | xa_lock_irq(&mapping->i_pages); |
6922c0c7 HD |
647 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); |
648 | page->mapping = NULL; | |
649 | mapping->nrpages--; | |
11fb9989 MG |
650 | __dec_node_page_state(page, NR_FILE_PAGES); |
651 | __dec_node_page_state(page, NR_SHMEM); | |
b93b0163 | 652 | xa_unlock_irq(&mapping->i_pages); |
09cbfeaf | 653 | put_page(page); |
6922c0c7 HD |
654 | BUG_ON(error); |
655 | } | |
656 | ||
7a5d0fbb HD |
657 | /* |
658 | * Remove swap entry from radix tree, free the swap and its page cache. | |
659 | */ | |
660 | static int shmem_free_swap(struct address_space *mapping, | |
661 | pgoff_t index, void *radswap) | |
662 | { | |
6dbaf22c | 663 | void *old; |
7a5d0fbb | 664 | |
b93b0163 MW |
665 | xa_lock_irq(&mapping->i_pages); |
666 | old = radix_tree_delete_item(&mapping->i_pages, index, radswap); | |
667 | xa_unlock_irq(&mapping->i_pages); | |
6dbaf22c JW |
668 | if (old != radswap) |
669 | return -ENOENT; | |
670 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
671 | return 0; | |
7a5d0fbb HD |
672 | } |
673 | ||
6a15a370 VB |
674 | /* |
675 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 676 | * given offsets are swapped out. |
6a15a370 | 677 | * |
b93b0163 | 678 | * This is safe to call without i_mutex or the i_pages lock thanks to RCU, |
6a15a370 VB |
679 | * as long as the inode doesn't go away and racy results are not a problem. |
680 | */ | |
48131e03 VB |
681 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
682 | pgoff_t start, pgoff_t end) | |
6a15a370 | 683 | { |
6a15a370 | 684 | struct radix_tree_iter iter; |
5b9c98f3 | 685 | void __rcu **slot; |
6a15a370 | 686 | struct page *page; |
48131e03 | 687 | unsigned long swapped = 0; |
6a15a370 VB |
688 | |
689 | rcu_read_lock(); | |
690 | ||
b93b0163 | 691 | radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) { |
6a15a370 VB |
692 | if (iter.index >= end) |
693 | break; | |
694 | ||
695 | page = radix_tree_deref_slot(slot); | |
696 | ||
2cf938aa MW |
697 | if (radix_tree_deref_retry(page)) { |
698 | slot = radix_tree_iter_retry(&iter); | |
699 | continue; | |
700 | } | |
6a15a370 VB |
701 | |
702 | if (radix_tree_exceptional_entry(page)) | |
703 | swapped++; | |
704 | ||
705 | if (need_resched()) { | |
148deab2 | 706 | slot = radix_tree_iter_resume(slot, &iter); |
6a15a370 | 707 | cond_resched_rcu(); |
6a15a370 VB |
708 | } |
709 | } | |
710 | ||
711 | rcu_read_unlock(); | |
712 | ||
713 | return swapped << PAGE_SHIFT; | |
714 | } | |
715 | ||
48131e03 VB |
716 | /* |
717 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
718 | * given vma is swapped out. | |
719 | * | |
b93b0163 | 720 | * This is safe to call without i_mutex or the i_pages lock thanks to RCU, |
48131e03 VB |
721 | * as long as the inode doesn't go away and racy results are not a problem. |
722 | */ | |
723 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
724 | { | |
725 | struct inode *inode = file_inode(vma->vm_file); | |
726 | struct shmem_inode_info *info = SHMEM_I(inode); | |
727 | struct address_space *mapping = inode->i_mapping; | |
728 | unsigned long swapped; | |
729 | ||
730 | /* Be careful as we don't hold info->lock */ | |
731 | swapped = READ_ONCE(info->swapped); | |
732 | ||
733 | /* | |
734 | * The easier cases are when the shmem object has nothing in swap, or | |
735 | * the vma maps it whole. Then we can simply use the stats that we | |
736 | * already track. | |
737 | */ | |
738 | if (!swapped) | |
739 | return 0; | |
740 | ||
741 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
742 | return swapped << PAGE_SHIFT; | |
743 | ||
744 | /* Here comes the more involved part */ | |
745 | return shmem_partial_swap_usage(mapping, | |
746 | linear_page_index(vma, vma->vm_start), | |
747 | linear_page_index(vma, vma->vm_end)); | |
748 | } | |
749 | ||
24513264 HD |
750 | /* |
751 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
752 | */ | |
753 | void shmem_unlock_mapping(struct address_space *mapping) | |
754 | { | |
755 | struct pagevec pvec; | |
756 | pgoff_t indices[PAGEVEC_SIZE]; | |
757 | pgoff_t index = 0; | |
758 | ||
86679820 | 759 | pagevec_init(&pvec); |
24513264 HD |
760 | /* |
761 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
762 | */ | |
763 | while (!mapping_unevictable(mapping)) { | |
764 | /* | |
765 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
766 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
767 | */ | |
0cd6144a JW |
768 | pvec.nr = find_get_entries(mapping, index, |
769 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
770 | if (!pvec.nr) |
771 | break; | |
772 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 773 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
774 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
775 | pagevec_release(&pvec); | |
776 | cond_resched(); | |
777 | } | |
7a5d0fbb HD |
778 | } |
779 | ||
780 | /* | |
781 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 782 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 783 | */ |
1635f6a7 HD |
784 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
785 | bool unfalloc) | |
1da177e4 | 786 | { |
285b2c4f | 787 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 788 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
789 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
790 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
791 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
792 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 793 | struct pagevec pvec; |
7a5d0fbb HD |
794 | pgoff_t indices[PAGEVEC_SIZE]; |
795 | long nr_swaps_freed = 0; | |
285b2c4f | 796 | pgoff_t index; |
bda97eab HD |
797 | int i; |
798 | ||
83e4fa9c HD |
799 | if (lend == -1) |
800 | end = -1; /* unsigned, so actually very big */ | |
bda97eab | 801 | |
86679820 | 802 | pagevec_init(&pvec); |
bda97eab | 803 | index = start; |
83e4fa9c | 804 | while (index < end) { |
0cd6144a JW |
805 | pvec.nr = find_get_entries(mapping, index, |
806 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
807 | pvec.pages, indices); | |
7a5d0fbb HD |
808 | if (!pvec.nr) |
809 | break; | |
bda97eab HD |
810 | for (i = 0; i < pagevec_count(&pvec); i++) { |
811 | struct page *page = pvec.pages[i]; | |
812 | ||
7a5d0fbb | 813 | index = indices[i]; |
83e4fa9c | 814 | if (index >= end) |
bda97eab HD |
815 | break; |
816 | ||
7a5d0fbb | 817 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
818 | if (unfalloc) |
819 | continue; | |
7a5d0fbb HD |
820 | nr_swaps_freed += !shmem_free_swap(mapping, |
821 | index, page); | |
bda97eab | 822 | continue; |
7a5d0fbb HD |
823 | } |
824 | ||
800d8c63 KS |
825 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
826 | ||
7a5d0fbb | 827 | if (!trylock_page(page)) |
bda97eab | 828 | continue; |
800d8c63 KS |
829 | |
830 | if (PageTransTail(page)) { | |
831 | /* Middle of THP: zero out the page */ | |
832 | clear_highpage(page); | |
833 | unlock_page(page); | |
834 | continue; | |
835 | } else if (PageTransHuge(page)) { | |
836 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
837 | /* | |
838 | * Range ends in the middle of THP: | |
839 | * zero out the page | |
840 | */ | |
841 | clear_highpage(page); | |
842 | unlock_page(page); | |
843 | continue; | |
844 | } | |
845 | index += HPAGE_PMD_NR - 1; | |
846 | i += HPAGE_PMD_NR - 1; | |
847 | } | |
848 | ||
1635f6a7 | 849 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
850 | VM_BUG_ON_PAGE(PageTail(page), page); |
851 | if (page_mapping(page) == mapping) { | |
309381fe | 852 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
853 | truncate_inode_page(mapping, page); |
854 | } | |
bda97eab | 855 | } |
bda97eab HD |
856 | unlock_page(page); |
857 | } | |
0cd6144a | 858 | pagevec_remove_exceptionals(&pvec); |
24513264 | 859 | pagevec_release(&pvec); |
bda97eab HD |
860 | cond_resched(); |
861 | index++; | |
862 | } | |
1da177e4 | 863 | |
83e4fa9c | 864 | if (partial_start) { |
bda97eab | 865 | struct page *page = NULL; |
9e18eb29 | 866 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 867 | if (page) { |
09cbfeaf | 868 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
869 | if (start > end) { |
870 | top = partial_end; | |
871 | partial_end = 0; | |
872 | } | |
873 | zero_user_segment(page, partial_start, top); | |
874 | set_page_dirty(page); | |
875 | unlock_page(page); | |
09cbfeaf | 876 | put_page(page); |
83e4fa9c HD |
877 | } |
878 | } | |
879 | if (partial_end) { | |
880 | struct page *page = NULL; | |
9e18eb29 | 881 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
882 | if (page) { |
883 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
884 | set_page_dirty(page); |
885 | unlock_page(page); | |
09cbfeaf | 886 | put_page(page); |
bda97eab HD |
887 | } |
888 | } | |
83e4fa9c HD |
889 | if (start >= end) |
890 | return; | |
bda97eab HD |
891 | |
892 | index = start; | |
b1a36650 | 893 | while (index < end) { |
bda97eab | 894 | cond_resched(); |
0cd6144a JW |
895 | |
896 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 897 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 898 | pvec.pages, indices); |
7a5d0fbb | 899 | if (!pvec.nr) { |
b1a36650 HD |
900 | /* If all gone or hole-punch or unfalloc, we're done */ |
901 | if (index == start || end != -1) | |
bda97eab | 902 | break; |
b1a36650 | 903 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
904 | index = start; |
905 | continue; | |
906 | } | |
bda97eab HD |
907 | for (i = 0; i < pagevec_count(&pvec); i++) { |
908 | struct page *page = pvec.pages[i]; | |
909 | ||
7a5d0fbb | 910 | index = indices[i]; |
83e4fa9c | 911 | if (index >= end) |
bda97eab HD |
912 | break; |
913 | ||
7a5d0fbb | 914 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
915 | if (unfalloc) |
916 | continue; | |
b1a36650 HD |
917 | if (shmem_free_swap(mapping, index, page)) { |
918 | /* Swap was replaced by page: retry */ | |
919 | index--; | |
920 | break; | |
921 | } | |
922 | nr_swaps_freed++; | |
7a5d0fbb HD |
923 | continue; |
924 | } | |
925 | ||
bda97eab | 926 | lock_page(page); |
800d8c63 KS |
927 | |
928 | if (PageTransTail(page)) { | |
929 | /* Middle of THP: zero out the page */ | |
930 | clear_highpage(page); | |
931 | unlock_page(page); | |
932 | /* | |
933 | * Partial thp truncate due 'start' in middle | |
934 | * of THP: don't need to look on these pages | |
935 | * again on !pvec.nr restart. | |
936 | */ | |
937 | if (index != round_down(end, HPAGE_PMD_NR)) | |
938 | start++; | |
939 | continue; | |
940 | } else if (PageTransHuge(page)) { | |
941 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
942 | /* | |
943 | * Range ends in the middle of THP: | |
944 | * zero out the page | |
945 | */ | |
946 | clear_highpage(page); | |
947 | unlock_page(page); | |
948 | continue; | |
949 | } | |
950 | index += HPAGE_PMD_NR - 1; | |
951 | i += HPAGE_PMD_NR - 1; | |
952 | } | |
953 | ||
1635f6a7 | 954 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
955 | VM_BUG_ON_PAGE(PageTail(page), page); |
956 | if (page_mapping(page) == mapping) { | |
309381fe | 957 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 958 | truncate_inode_page(mapping, page); |
b1a36650 HD |
959 | } else { |
960 | /* Page was replaced by swap: retry */ | |
961 | unlock_page(page); | |
962 | index--; | |
963 | break; | |
1635f6a7 | 964 | } |
7a5d0fbb | 965 | } |
bda97eab HD |
966 | unlock_page(page); |
967 | } | |
0cd6144a | 968 | pagevec_remove_exceptionals(&pvec); |
24513264 | 969 | pagevec_release(&pvec); |
bda97eab HD |
970 | index++; |
971 | } | |
94c1e62d | 972 | |
4595ef88 | 973 | spin_lock_irq(&info->lock); |
7a5d0fbb | 974 | info->swapped -= nr_swaps_freed; |
1da177e4 | 975 | shmem_recalc_inode(inode); |
4595ef88 | 976 | spin_unlock_irq(&info->lock); |
1635f6a7 | 977 | } |
1da177e4 | 978 | |
1635f6a7 HD |
979 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
980 | { | |
981 | shmem_undo_range(inode, lstart, lend, false); | |
078cd827 | 982 | inode->i_ctime = inode->i_mtime = current_time(inode); |
1da177e4 | 983 | } |
94c1e62d | 984 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 985 | |
a528d35e DH |
986 | static int shmem_getattr(const struct path *path, struct kstat *stat, |
987 | u32 request_mask, unsigned int query_flags) | |
44a30220 | 988 | { |
a528d35e | 989 | struct inode *inode = path->dentry->d_inode; |
44a30220 YZ |
990 | struct shmem_inode_info *info = SHMEM_I(inode); |
991 | ||
d0424c42 | 992 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
4595ef88 | 993 | spin_lock_irq(&info->lock); |
d0424c42 | 994 | shmem_recalc_inode(inode); |
4595ef88 | 995 | spin_unlock_irq(&info->lock); |
d0424c42 | 996 | } |
44a30220 | 997 | generic_fillattr(inode, stat); |
44a30220 YZ |
998 | return 0; |
999 | } | |
1000 | ||
94c1e62d | 1001 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 1002 | { |
75c3cfa8 | 1003 | struct inode *inode = d_inode(dentry); |
40e041a2 | 1004 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1005 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 LT |
1006 | int error; |
1007 | ||
31051c85 | 1008 | error = setattr_prepare(dentry, attr); |
db78b877 CH |
1009 | if (error) |
1010 | return error; | |
1011 | ||
94c1e62d HD |
1012 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
1013 | loff_t oldsize = inode->i_size; | |
1014 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 1015 | |
40e041a2 DH |
1016 | /* protected by i_mutex */ |
1017 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
1018 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
1019 | return -EPERM; | |
1020 | ||
94c1e62d | 1021 | if (newsize != oldsize) { |
77142517 KK |
1022 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
1023 | oldsize, newsize); | |
1024 | if (error) | |
1025 | return error; | |
94c1e62d | 1026 | i_size_write(inode, newsize); |
078cd827 | 1027 | inode->i_ctime = inode->i_mtime = current_time(inode); |
94c1e62d | 1028 | } |
afa2db2f | 1029 | if (newsize <= oldsize) { |
94c1e62d | 1030 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
1031 | if (oldsize > holebegin) |
1032 | unmap_mapping_range(inode->i_mapping, | |
1033 | holebegin, 0, 1); | |
1034 | if (info->alloced) | |
1035 | shmem_truncate_range(inode, | |
1036 | newsize, (loff_t)-1); | |
94c1e62d | 1037 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
1038 | if (oldsize > holebegin) |
1039 | unmap_mapping_range(inode->i_mapping, | |
1040 | holebegin, 0, 1); | |
779750d2 KS |
1041 | |
1042 | /* | |
1043 | * Part of the huge page can be beyond i_size: subject | |
1044 | * to shrink under memory pressure. | |
1045 | */ | |
1046 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { | |
1047 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1048 | /* |
1049 | * _careful to defend against unlocked access to | |
1050 | * ->shrink_list in shmem_unused_huge_shrink() | |
1051 | */ | |
1052 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1053 | list_add_tail(&info->shrinklist, |
1054 | &sbinfo->shrinklist); | |
1055 | sbinfo->shrinklist_len++; | |
1056 | } | |
1057 | spin_unlock(&sbinfo->shrinklist_lock); | |
1058 | } | |
94c1e62d | 1059 | } |
1da177e4 LT |
1060 | } |
1061 | ||
db78b877 | 1062 | setattr_copy(inode, attr); |
db78b877 | 1063 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 1064 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
1065 | return error; |
1066 | } | |
1067 | ||
1f895f75 | 1068 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1069 | { |
1da177e4 | 1070 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1071 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 | 1072 | |
3889e6e7 | 1073 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
1074 | shmem_unacct_size(info->flags, inode->i_size); |
1075 | inode->i_size = 0; | |
3889e6e7 | 1076 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1077 | if (!list_empty(&info->shrinklist)) { |
1078 | spin_lock(&sbinfo->shrinklist_lock); | |
1079 | if (!list_empty(&info->shrinklist)) { | |
1080 | list_del_init(&info->shrinklist); | |
1081 | sbinfo->shrinklist_len--; | |
1082 | } | |
1083 | spin_unlock(&sbinfo->shrinklist_lock); | |
1084 | } | |
1da177e4 | 1085 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 1086 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 1087 | list_del_init(&info->swaplist); |
cb5f7b9a | 1088 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1089 | } |
3ed47db3 | 1090 | } |
b09e0fa4 | 1091 | |
38f38657 | 1092 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 1093 | WARN_ON(inode->i_blocks); |
5b04c689 | 1094 | shmem_free_inode(inode->i_sb); |
dbd5768f | 1095 | clear_inode(inode); |
1da177e4 LT |
1096 | } |
1097 | ||
478922e2 MW |
1098 | static unsigned long find_swap_entry(struct radix_tree_root *root, void *item) |
1099 | { | |
1100 | struct radix_tree_iter iter; | |
5b9c98f3 | 1101 | void __rcu **slot; |
478922e2 MW |
1102 | unsigned long found = -1; |
1103 | unsigned int checked = 0; | |
1104 | ||
1105 | rcu_read_lock(); | |
1106 | radix_tree_for_each_slot(slot, root, &iter, 0) { | |
5b9c98f3 MK |
1107 | void *entry = radix_tree_deref_slot(slot); |
1108 | ||
1109 | if (radix_tree_deref_retry(entry)) { | |
1110 | slot = radix_tree_iter_retry(&iter); | |
1111 | continue; | |
1112 | } | |
1113 | if (entry == item) { | |
478922e2 MW |
1114 | found = iter.index; |
1115 | break; | |
1116 | } | |
1117 | checked++; | |
1118 | if ((checked % 4096) != 0) | |
1119 | continue; | |
1120 | slot = radix_tree_iter_resume(slot, &iter); | |
1121 | cond_resched_rcu(); | |
1122 | } | |
1123 | ||
1124 | rcu_read_unlock(); | |
1125 | return found; | |
1126 | } | |
1127 | ||
46f65ec1 HD |
1128 | /* |
1129 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1130 | */ | |
41ffe5d5 | 1131 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 1132 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 1133 | { |
285b2c4f | 1134 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 1135 | void *radswap; |
41ffe5d5 | 1136 | pgoff_t index; |
bde05d1c HD |
1137 | gfp_t gfp; |
1138 | int error = 0; | |
1da177e4 | 1139 | |
46f65ec1 | 1140 | radswap = swp_to_radix_entry(swap); |
b93b0163 | 1141 | index = find_swap_entry(&mapping->i_pages, radswap); |
46f65ec1 | 1142 | if (index == -1) |
00501b53 | 1143 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 1144 | |
1b1b32f2 HD |
1145 | /* |
1146 | * Move _head_ to start search for next from here. | |
1f895f75 | 1147 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 1148 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 1149 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
1150 | */ |
1151 | if (shmem_swaplist.next != &info->swaplist) | |
1152 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 1153 | |
bde05d1c HD |
1154 | gfp = mapping_gfp_mask(mapping); |
1155 | if (shmem_should_replace_page(*pagep, gfp)) { | |
1156 | mutex_unlock(&shmem_swaplist_mutex); | |
1157 | error = shmem_replace_page(pagep, gfp, info, index); | |
1158 | mutex_lock(&shmem_swaplist_mutex); | |
1159 | /* | |
1160 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
1161 | * allocation, but the inode might have been freed while we |
1162 | * dropped it: although a racing shmem_evict_inode() cannot | |
1163 | * complete without emptying the radix_tree, our page lock | |
1164 | * on this swapcache page is not enough to prevent that - | |
1165 | * free_swap_and_cache() of our swap entry will only | |
1166 | * trylock_page(), removing swap from radix_tree whatever. | |
1167 | * | |
1168 | * We must not proceed to shmem_add_to_page_cache() if the | |
1169 | * inode has been freed, but of course we cannot rely on | |
1170 | * inode or mapping or info to check that. However, we can | |
1171 | * safely check if our swap entry is still in use (and here | |
1172 | * it can't have got reused for another page): if it's still | |
1173 | * in use, then the inode cannot have been freed yet, and we | |
1174 | * can safely proceed (if it's no longer in use, that tells | |
1175 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
1176 | */ |
1177 | if (!page_swapcount(*pagep)) | |
1178 | error = -ENOENT; | |
1179 | } | |
1180 | ||
d13d1443 | 1181 | /* |
778dd893 HD |
1182 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
1183 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
1184 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 1185 | */ |
bde05d1c HD |
1186 | if (!error) |
1187 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 1188 | radswap); |
48f170fb | 1189 | if (error != -ENOMEM) { |
46f65ec1 HD |
1190 | /* |
1191 | * Truncation and eviction use free_swap_and_cache(), which | |
1192 | * only does trylock page: if we raced, best clean up here. | |
1193 | */ | |
bde05d1c HD |
1194 | delete_from_swap_cache(*pagep); |
1195 | set_page_dirty(*pagep); | |
46f65ec1 | 1196 | if (!error) { |
4595ef88 | 1197 | spin_lock_irq(&info->lock); |
46f65ec1 | 1198 | info->swapped--; |
4595ef88 | 1199 | spin_unlock_irq(&info->lock); |
46f65ec1 HD |
1200 | swap_free(swap); |
1201 | } | |
1da177e4 | 1202 | } |
2e0e26c7 | 1203 | return error; |
1da177e4 LT |
1204 | } |
1205 | ||
1206 | /* | |
46f65ec1 | 1207 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 1208 | */ |
41ffe5d5 | 1209 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 1210 | { |
41ffe5d5 | 1211 | struct list_head *this, *next; |
1da177e4 | 1212 | struct shmem_inode_info *info; |
00501b53 | 1213 | struct mem_cgroup *memcg; |
bde05d1c HD |
1214 | int error = 0; |
1215 | ||
1216 | /* | |
1217 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 1218 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 1219 | */ |
0142ef6c | 1220 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 1221 | goto out; |
778dd893 HD |
1222 | |
1223 | /* | |
1224 | * Charge page using GFP_KERNEL while we can wait, before taking | |
1225 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
1226 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 1227 | */ |
f627c2f5 KS |
1228 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg, |
1229 | false); | |
778dd893 HD |
1230 | if (error) |
1231 | goto out; | |
46f65ec1 | 1232 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 1233 | error = -EAGAIN; |
1da177e4 | 1234 | |
cb5f7b9a | 1235 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
1236 | list_for_each_safe(this, next, &shmem_swaplist) { |
1237 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 1238 | if (info->swapped) |
00501b53 | 1239 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
1240 | else |
1241 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1242 | cond_resched(); |
00501b53 | 1243 | if (error != -EAGAIN) |
778dd893 | 1244 | break; |
00501b53 | 1245 | /* found nothing in this: move on to search the next */ |
1da177e4 | 1246 | } |
cb5f7b9a | 1247 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1248 | |
00501b53 JW |
1249 | if (error) { |
1250 | if (error != -ENOMEM) | |
1251 | error = 0; | |
f627c2f5 | 1252 | mem_cgroup_cancel_charge(page, memcg, false); |
00501b53 | 1253 | } else |
f627c2f5 | 1254 | mem_cgroup_commit_charge(page, memcg, true, false); |
778dd893 | 1255 | out: |
aaa46865 | 1256 | unlock_page(page); |
09cbfeaf | 1257 | put_page(page); |
778dd893 | 1258 | return error; |
1da177e4 LT |
1259 | } |
1260 | ||
1261 | /* | |
1262 | * Move the page from the page cache to the swap cache. | |
1263 | */ | |
1264 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1265 | { | |
1266 | struct shmem_inode_info *info; | |
1da177e4 | 1267 | struct address_space *mapping; |
1da177e4 | 1268 | struct inode *inode; |
6922c0c7 HD |
1269 | swp_entry_t swap; |
1270 | pgoff_t index; | |
1da177e4 | 1271 | |
800d8c63 | 1272 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1273 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1274 | mapping = page->mapping; |
1275 | index = page->index; | |
1276 | inode = mapping->host; | |
1277 | info = SHMEM_I(inode); | |
1278 | if (info->flags & VM_LOCKED) | |
1279 | goto redirty; | |
d9fe526a | 1280 | if (!total_swap_pages) |
1da177e4 LT |
1281 | goto redirty; |
1282 | ||
d9fe526a | 1283 | /* |
97b713ba CH |
1284 | * Our capabilities prevent regular writeback or sync from ever calling |
1285 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1286 | * its underlying filesystem, in which case tmpfs should write out to | |
1287 | * swap only in response to memory pressure, and not for the writeback | |
1288 | * threads or sync. | |
d9fe526a | 1289 | */ |
48f170fb HD |
1290 | if (!wbc->for_reclaim) { |
1291 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1292 | goto redirty; | |
1293 | } | |
1635f6a7 HD |
1294 | |
1295 | /* | |
1296 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1297 | * value into swapfile.c, the only way we can correctly account for a | |
1298 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1299 | * |
1300 | * That's okay for a page already fallocated earlier, but if we have | |
1301 | * not yet completed the fallocation, then (a) we want to keep track | |
1302 | * of this page in case we have to undo it, and (b) it may not be a | |
1303 | * good idea to continue anyway, once we're pushing into swap. So | |
1304 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1305 | */ |
1306 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1307 | if (inode->i_private) { |
1308 | struct shmem_falloc *shmem_falloc; | |
1309 | spin_lock(&inode->i_lock); | |
1310 | shmem_falloc = inode->i_private; | |
1311 | if (shmem_falloc && | |
8e205f77 | 1312 | !shmem_falloc->waitq && |
1aac1400 HD |
1313 | index >= shmem_falloc->start && |
1314 | index < shmem_falloc->next) | |
1315 | shmem_falloc->nr_unswapped++; | |
1316 | else | |
1317 | shmem_falloc = NULL; | |
1318 | spin_unlock(&inode->i_lock); | |
1319 | if (shmem_falloc) | |
1320 | goto redirty; | |
1321 | } | |
1635f6a7 HD |
1322 | clear_highpage(page); |
1323 | flush_dcache_page(page); | |
1324 | SetPageUptodate(page); | |
1325 | } | |
1326 | ||
38d8b4e6 | 1327 | swap = get_swap_page(page); |
48f170fb HD |
1328 | if (!swap.val) |
1329 | goto redirty; | |
d9fe526a | 1330 | |
b1dea800 HD |
1331 | /* |
1332 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1333 | * if it's not already there. Do it now before the page is |
1334 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1335 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1336 | * we've incremented swapped, because shmem_unuse_inode() will |
1337 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1338 | */ |
48f170fb HD |
1339 | mutex_lock(&shmem_swaplist_mutex); |
1340 | if (list_empty(&info->swaplist)) | |
1341 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 1342 | |
48f170fb | 1343 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
4595ef88 | 1344 | spin_lock_irq(&info->lock); |
6922c0c7 | 1345 | shmem_recalc_inode(inode); |
267a4c76 | 1346 | info->swapped++; |
4595ef88 | 1347 | spin_unlock_irq(&info->lock); |
6922c0c7 | 1348 | |
267a4c76 HD |
1349 | swap_shmem_alloc(swap); |
1350 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1351 | ||
6922c0c7 | 1352 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1353 | BUG_ON(page_mapped(page)); |
9fab5619 | 1354 | swap_writepage(page, wbc); |
1da177e4 LT |
1355 | return 0; |
1356 | } | |
1357 | ||
6922c0c7 | 1358 | mutex_unlock(&shmem_swaplist_mutex); |
75f6d6d2 | 1359 | put_swap_page(page, swap); |
1da177e4 LT |
1360 | redirty: |
1361 | set_page_dirty(page); | |
d9fe526a HD |
1362 | if (wbc->for_reclaim) |
1363 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1364 | unlock_page(page); | |
1365 | return 0; | |
1da177e4 LT |
1366 | } |
1367 | ||
75edd345 | 1368 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1369 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1370 | { |
095f1fc4 | 1371 | char buffer[64]; |
680d794b | 1372 | |
71fe804b | 1373 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1374 | return; /* show nothing */ |
680d794b | 1375 | |
a7a88b23 | 1376 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1377 | |
1378 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1379 | } |
71fe804b LS |
1380 | |
1381 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1382 | { | |
1383 | struct mempolicy *mpol = NULL; | |
1384 | if (sbinfo->mpol) { | |
1385 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1386 | mpol = sbinfo->mpol; | |
1387 | mpol_get(mpol); | |
1388 | spin_unlock(&sbinfo->stat_lock); | |
1389 | } | |
1390 | return mpol; | |
1391 | } | |
75edd345 HD |
1392 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1393 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1394 | { | |
1395 | } | |
1396 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1397 | { | |
1398 | return NULL; | |
1399 | } | |
1400 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1401 | #ifndef CONFIG_NUMA | |
1402 | #define vm_policy vm_private_data | |
1403 | #endif | |
680d794b | 1404 | |
800d8c63 KS |
1405 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1406 | struct shmem_inode_info *info, pgoff_t index) | |
1407 | { | |
1408 | /* Create a pseudo vma that just contains the policy */ | |
1409 | vma->vm_start = 0; | |
1410 | /* Bias interleave by inode number to distribute better across nodes */ | |
1411 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
1412 | vma->vm_ops = NULL; | |
1413 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); | |
1414 | } | |
1415 | ||
1416 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1417 | { | |
1418 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1419 | mpol_cond_put(vma->vm_policy); | |
1420 | } | |
1421 | ||
41ffe5d5 HD |
1422 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1423 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1424 | { |
1da177e4 | 1425 | struct vm_area_struct pvma; |
18a2f371 | 1426 | struct page *page; |
e9e9b7ec | 1427 | struct vm_fault vmf; |
52cd3b07 | 1428 | |
800d8c63 | 1429 | shmem_pseudo_vma_init(&pvma, info, index); |
e9e9b7ec MK |
1430 | vmf.vma = &pvma; |
1431 | vmf.address = 0; | |
1432 | page = swap_cluster_readahead(swap, gfp, &vmf); | |
800d8c63 | 1433 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1434 | |
800d8c63 KS |
1435 | return page; |
1436 | } | |
1437 | ||
1438 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1439 | struct shmem_inode_info *info, pgoff_t index) | |
1440 | { | |
1441 | struct vm_area_struct pvma; | |
1442 | struct inode *inode = &info->vfs_inode; | |
1443 | struct address_space *mapping = inode->i_mapping; | |
4620a06e | 1444 | pgoff_t idx, hindex; |
800d8c63 KS |
1445 | void __rcu **results; |
1446 | struct page *page; | |
1447 | ||
e496cf3d | 1448 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1449 | return NULL; |
1450 | ||
4620a06e | 1451 | hindex = round_down(index, HPAGE_PMD_NR); |
800d8c63 | 1452 | rcu_read_lock(); |
b93b0163 | 1453 | if (radix_tree_gang_lookup_slot(&mapping->i_pages, &results, &idx, |
800d8c63 KS |
1454 | hindex, 1) && idx < hindex + HPAGE_PMD_NR) { |
1455 | rcu_read_unlock(); | |
1456 | return NULL; | |
1457 | } | |
1458 | rcu_read_unlock(); | |
18a2f371 | 1459 | |
800d8c63 KS |
1460 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1461 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
1462 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); | |
1463 | shmem_pseudo_vma_destroy(&pvma); | |
1464 | if (page) | |
1465 | prep_transhuge_page(page); | |
18a2f371 | 1466 | return page; |
1da177e4 LT |
1467 | } |
1468 | ||
02098fea | 1469 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1470 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1471 | { |
1472 | struct vm_area_struct pvma; | |
18a2f371 | 1473 | struct page *page; |
1da177e4 | 1474 | |
800d8c63 KS |
1475 | shmem_pseudo_vma_init(&pvma, info, index); |
1476 | page = alloc_page_vma(gfp, &pvma, 0); | |
1477 | shmem_pseudo_vma_destroy(&pvma); | |
1478 | ||
1479 | return page; | |
1480 | } | |
1481 | ||
1482 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
0f079694 | 1483 | struct inode *inode, |
800d8c63 KS |
1484 | pgoff_t index, bool huge) |
1485 | { | |
0f079694 | 1486 | struct shmem_inode_info *info = SHMEM_I(inode); |
800d8c63 KS |
1487 | struct page *page; |
1488 | int nr; | |
1489 | int err = -ENOSPC; | |
52cd3b07 | 1490 | |
e496cf3d | 1491 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1492 | huge = false; |
1493 | nr = huge ? HPAGE_PMD_NR : 1; | |
1494 | ||
0f079694 | 1495 | if (!shmem_inode_acct_block(inode, nr)) |
800d8c63 | 1496 | goto failed; |
800d8c63 KS |
1497 | |
1498 | if (huge) | |
1499 | page = shmem_alloc_hugepage(gfp, info, index); | |
1500 | else | |
1501 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1502 | if (page) { |
1503 | __SetPageLocked(page); | |
1504 | __SetPageSwapBacked(page); | |
800d8c63 | 1505 | return page; |
75edd345 | 1506 | } |
18a2f371 | 1507 | |
800d8c63 | 1508 | err = -ENOMEM; |
0f079694 | 1509 | shmem_inode_unacct_blocks(inode, nr); |
800d8c63 KS |
1510 | failed: |
1511 | return ERR_PTR(err); | |
1da177e4 | 1512 | } |
71fe804b | 1513 | |
bde05d1c HD |
1514 | /* |
1515 | * When a page is moved from swapcache to shmem filecache (either by the | |
1516 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1517 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1518 | * ignorance of the mapping it belongs to. If that mapping has special | |
1519 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1520 | * we may need to copy to a suitable page before moving to filecache. | |
1521 | * | |
1522 | * In a future release, this may well be extended to respect cpuset and | |
1523 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1524 | * but for now it is a simple matter of zone. | |
1525 | */ | |
1526 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1527 | { | |
1528 | return page_zonenum(page) > gfp_zone(gfp); | |
1529 | } | |
1530 | ||
1531 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1532 | struct shmem_inode_info *info, pgoff_t index) | |
1533 | { | |
1534 | struct page *oldpage, *newpage; | |
1535 | struct address_space *swap_mapping; | |
1536 | pgoff_t swap_index; | |
1537 | int error; | |
1538 | ||
1539 | oldpage = *pagep; | |
1540 | swap_index = page_private(oldpage); | |
1541 | swap_mapping = page_mapping(oldpage); | |
1542 | ||
1543 | /* | |
1544 | * We have arrived here because our zones are constrained, so don't | |
1545 | * limit chance of success by further cpuset and node constraints. | |
1546 | */ | |
1547 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1548 | newpage = shmem_alloc_page(gfp, info, index); | |
1549 | if (!newpage) | |
1550 | return -ENOMEM; | |
bde05d1c | 1551 | |
09cbfeaf | 1552 | get_page(newpage); |
bde05d1c | 1553 | copy_highpage(newpage, oldpage); |
0142ef6c | 1554 | flush_dcache_page(newpage); |
bde05d1c | 1555 | |
9956edf3 HD |
1556 | __SetPageLocked(newpage); |
1557 | __SetPageSwapBacked(newpage); | |
bde05d1c | 1558 | SetPageUptodate(newpage); |
bde05d1c | 1559 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1560 | SetPageSwapCache(newpage); |
1561 | ||
1562 | /* | |
1563 | * Our caller will very soon move newpage out of swapcache, but it's | |
1564 | * a nice clean interface for us to replace oldpage by newpage there. | |
1565 | */ | |
b93b0163 | 1566 | xa_lock_irq(&swap_mapping->i_pages); |
bde05d1c HD |
1567 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, |
1568 | newpage); | |
0142ef6c | 1569 | if (!error) { |
11fb9989 MG |
1570 | __inc_node_page_state(newpage, NR_FILE_PAGES); |
1571 | __dec_node_page_state(oldpage, NR_FILE_PAGES); | |
0142ef6c | 1572 | } |
b93b0163 | 1573 | xa_unlock_irq(&swap_mapping->i_pages); |
bde05d1c | 1574 | |
0142ef6c HD |
1575 | if (unlikely(error)) { |
1576 | /* | |
1577 | * Is this possible? I think not, now that our callers check | |
1578 | * both PageSwapCache and page_private after getting page lock; | |
1579 | * but be defensive. Reverse old to newpage for clear and free. | |
1580 | */ | |
1581 | oldpage = newpage; | |
1582 | } else { | |
6a93ca8f | 1583 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1584 | lru_cache_add_anon(newpage); |
1585 | *pagep = newpage; | |
1586 | } | |
bde05d1c HD |
1587 | |
1588 | ClearPageSwapCache(oldpage); | |
1589 | set_page_private(oldpage, 0); | |
1590 | ||
1591 | unlock_page(oldpage); | |
09cbfeaf KS |
1592 | put_page(oldpage); |
1593 | put_page(oldpage); | |
0142ef6c | 1594 | return error; |
bde05d1c HD |
1595 | } |
1596 | ||
1da177e4 | 1597 | /* |
68da9f05 | 1598 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1599 | * |
1600 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1601 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1602 | * entry since a page cannot live in both the swap and page cache. |
1603 | * | |
1604 | * fault_mm and fault_type are only supplied by shmem_fault: | |
1605 | * otherwise they are NULL. | |
1da177e4 | 1606 | */ |
41ffe5d5 | 1607 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 1608 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
cfda0526 | 1609 | struct vm_area_struct *vma, struct vm_fault *vmf, int *fault_type) |
1da177e4 LT |
1610 | { |
1611 | struct address_space *mapping = inode->i_mapping; | |
23f919d4 | 1612 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 | 1613 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1614 | struct mm_struct *charge_mm; |
00501b53 | 1615 | struct mem_cgroup *memcg; |
27ab7006 | 1616 | struct page *page; |
1da177e4 | 1617 | swp_entry_t swap; |
657e3038 | 1618 | enum sgp_type sgp_huge = sgp; |
800d8c63 | 1619 | pgoff_t hindex = index; |
1da177e4 | 1620 | int error; |
54af6042 | 1621 | int once = 0; |
1635f6a7 | 1622 | int alloced = 0; |
1da177e4 | 1623 | |
09cbfeaf | 1624 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1625 | return -EFBIG; |
657e3038 KS |
1626 | if (sgp == SGP_NOHUGE || sgp == SGP_HUGE) |
1627 | sgp = SGP_CACHE; | |
1da177e4 | 1628 | repeat: |
54af6042 | 1629 | swap.val = 0; |
0cd6144a | 1630 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1631 | if (radix_tree_exceptional_entry(page)) { |
1632 | swap = radix_to_swp_entry(page); | |
1633 | page = NULL; | |
1634 | } | |
1635 | ||
75edd345 | 1636 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1637 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 1638 | error = -EINVAL; |
267a4c76 | 1639 | goto unlock; |
54af6042 HD |
1640 | } |
1641 | ||
66d2f4d2 HD |
1642 | if (page && sgp == SGP_WRITE) |
1643 | mark_page_accessed(page); | |
1644 | ||
1635f6a7 HD |
1645 | /* fallocated page? */ |
1646 | if (page && !PageUptodate(page)) { | |
1647 | if (sgp != SGP_READ) | |
1648 | goto clear; | |
1649 | unlock_page(page); | |
09cbfeaf | 1650 | put_page(page); |
1635f6a7 HD |
1651 | page = NULL; |
1652 | } | |
54af6042 | 1653 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1654 | *pagep = page; |
1655 | return 0; | |
27ab7006 HD |
1656 | } |
1657 | ||
1658 | /* | |
54af6042 HD |
1659 | * Fast cache lookup did not find it: |
1660 | * bring it back from swap or allocate. | |
27ab7006 | 1661 | */ |
54af6042 | 1662 | sbinfo = SHMEM_SB(inode->i_sb); |
cfda0526 | 1663 | charge_mm = vma ? vma->vm_mm : current->mm; |
1da177e4 | 1664 | |
1da177e4 LT |
1665 | if (swap.val) { |
1666 | /* Look it up and read it in.. */ | |
ec560175 | 1667 | page = lookup_swap_cache(swap, NULL, 0); |
27ab7006 | 1668 | if (!page) { |
9e18eb29 ALC |
1669 | /* Or update major stats only when swapin succeeds?? */ |
1670 | if (fault_type) { | |
68da9f05 | 1671 | *fault_type |= VM_FAULT_MAJOR; |
9e18eb29 | 1672 | count_vm_event(PGMAJFAULT); |
2262185c | 1673 | count_memcg_event_mm(charge_mm, PGMAJFAULT); |
9e18eb29 ALC |
1674 | } |
1675 | /* Here we actually start the io */ | |
41ffe5d5 | 1676 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1677 | if (!page) { |
54af6042 HD |
1678 | error = -ENOMEM; |
1679 | goto failed; | |
1da177e4 | 1680 | } |
1da177e4 LT |
1681 | } |
1682 | ||
1683 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1684 | lock_page(page); |
0142ef6c | 1685 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1686 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1687 | error = -EEXIST; /* try again */ |
d1899228 | 1688 | goto unlock; |
bde05d1c | 1689 | } |
27ab7006 | 1690 | if (!PageUptodate(page)) { |
1da177e4 | 1691 | error = -EIO; |
54af6042 | 1692 | goto failed; |
1da177e4 | 1693 | } |
54af6042 HD |
1694 | wait_on_page_writeback(page); |
1695 | ||
bde05d1c HD |
1696 | if (shmem_should_replace_page(page, gfp)) { |
1697 | error = shmem_replace_page(&page, gfp, info, index); | |
1698 | if (error) | |
1699 | goto failed; | |
1da177e4 | 1700 | } |
27ab7006 | 1701 | |
9e18eb29 | 1702 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
f627c2f5 | 1703 | false); |
d1899228 | 1704 | if (!error) { |
aa3b1895 | 1705 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1706 | swp_to_radix_entry(swap)); |
215c02bc HD |
1707 | /* |
1708 | * We already confirmed swap under page lock, and make | |
1709 | * no memory allocation here, so usually no possibility | |
1710 | * of error; but free_swap_and_cache() only trylocks a | |
1711 | * page, so it is just possible that the entry has been | |
1712 | * truncated or holepunched since swap was confirmed. | |
1713 | * shmem_undo_range() will have done some of the | |
1714 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1715 | * the rest. |
215c02bc HD |
1716 | * Reset swap.val? No, leave it so "failed" goes back to |
1717 | * "repeat": reading a hole and writing should succeed. | |
1718 | */ | |
00501b53 | 1719 | if (error) { |
f627c2f5 | 1720 | mem_cgroup_cancel_charge(page, memcg, false); |
215c02bc | 1721 | delete_from_swap_cache(page); |
00501b53 | 1722 | } |
d1899228 | 1723 | } |
54af6042 HD |
1724 | if (error) |
1725 | goto failed; | |
1726 | ||
f627c2f5 | 1727 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1728 | |
4595ef88 | 1729 | spin_lock_irq(&info->lock); |
285b2c4f | 1730 | info->swapped--; |
54af6042 | 1731 | shmem_recalc_inode(inode); |
4595ef88 | 1732 | spin_unlock_irq(&info->lock); |
54af6042 | 1733 | |
66d2f4d2 HD |
1734 | if (sgp == SGP_WRITE) |
1735 | mark_page_accessed(page); | |
1736 | ||
54af6042 | 1737 | delete_from_swap_cache(page); |
27ab7006 HD |
1738 | set_page_dirty(page); |
1739 | swap_free(swap); | |
1740 | ||
54af6042 | 1741 | } else { |
cfda0526 MR |
1742 | if (vma && userfaultfd_missing(vma)) { |
1743 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
1744 | return 0; | |
1745 | } | |
1746 | ||
800d8c63 KS |
1747 | /* shmem_symlink() */ |
1748 | if (mapping->a_ops != &shmem_aops) | |
1749 | goto alloc_nohuge; | |
657e3038 | 1750 | if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE) |
800d8c63 KS |
1751 | goto alloc_nohuge; |
1752 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1753 | goto alloc_huge; | |
1754 | switch (sbinfo->huge) { | |
1755 | loff_t i_size; | |
1756 | pgoff_t off; | |
1757 | case SHMEM_HUGE_NEVER: | |
1758 | goto alloc_nohuge; | |
1759 | case SHMEM_HUGE_WITHIN_SIZE: | |
1760 | off = round_up(index, HPAGE_PMD_NR); | |
1761 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1762 | if (i_size >= HPAGE_PMD_SIZE && | |
1763 | i_size >> PAGE_SHIFT >= off) | |
1764 | goto alloc_huge; | |
1765 | /* fallthrough */ | |
1766 | case SHMEM_HUGE_ADVISE: | |
657e3038 KS |
1767 | if (sgp_huge == SGP_HUGE) |
1768 | goto alloc_huge; | |
1769 | /* TODO: implement fadvise() hints */ | |
800d8c63 | 1770 | goto alloc_nohuge; |
54af6042 | 1771 | } |
1da177e4 | 1772 | |
800d8c63 | 1773 | alloc_huge: |
0f079694 | 1774 | page = shmem_alloc_and_acct_page(gfp, inode, index, true); |
800d8c63 | 1775 | if (IS_ERR(page)) { |
0f079694 | 1776 | alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, inode, |
800d8c63 | 1777 | index, false); |
1da177e4 | 1778 | } |
800d8c63 | 1779 | if (IS_ERR(page)) { |
779750d2 | 1780 | int retry = 5; |
800d8c63 KS |
1781 | error = PTR_ERR(page); |
1782 | page = NULL; | |
779750d2 KS |
1783 | if (error != -ENOSPC) |
1784 | goto failed; | |
1785 | /* | |
1786 | * Try to reclaim some spece by splitting a huge page | |
1787 | * beyond i_size on the filesystem. | |
1788 | */ | |
1789 | while (retry--) { | |
1790 | int ret; | |
1791 | ret = shmem_unused_huge_shrink(sbinfo, NULL, 1); | |
1792 | if (ret == SHRINK_STOP) | |
1793 | break; | |
1794 | if (ret) | |
1795 | goto alloc_nohuge; | |
1796 | } | |
800d8c63 KS |
1797 | goto failed; |
1798 | } | |
1799 | ||
1800 | if (PageTransHuge(page)) | |
1801 | hindex = round_down(index, HPAGE_PMD_NR); | |
1802 | else | |
1803 | hindex = index; | |
1804 | ||
66d2f4d2 | 1805 | if (sgp == SGP_WRITE) |
eb39d618 | 1806 | __SetPageReferenced(page); |
66d2f4d2 | 1807 | |
9e18eb29 | 1808 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
800d8c63 | 1809 | PageTransHuge(page)); |
54af6042 | 1810 | if (error) |
800d8c63 KS |
1811 | goto unacct; |
1812 | error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK, | |
1813 | compound_order(page)); | |
b065b432 | 1814 | if (!error) { |
800d8c63 | 1815 | error = shmem_add_to_page_cache(page, mapping, hindex, |
fed400a1 | 1816 | NULL); |
b065b432 HD |
1817 | radix_tree_preload_end(); |
1818 | } | |
1819 | if (error) { | |
800d8c63 KS |
1820 | mem_cgroup_cancel_charge(page, memcg, |
1821 | PageTransHuge(page)); | |
1822 | goto unacct; | |
b065b432 | 1823 | } |
800d8c63 KS |
1824 | mem_cgroup_commit_charge(page, memcg, false, |
1825 | PageTransHuge(page)); | |
54af6042 HD |
1826 | lru_cache_add_anon(page); |
1827 | ||
4595ef88 | 1828 | spin_lock_irq(&info->lock); |
800d8c63 KS |
1829 | info->alloced += 1 << compound_order(page); |
1830 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); | |
54af6042 | 1831 | shmem_recalc_inode(inode); |
4595ef88 | 1832 | spin_unlock_irq(&info->lock); |
1635f6a7 | 1833 | alloced = true; |
54af6042 | 1834 | |
779750d2 KS |
1835 | if (PageTransHuge(page) && |
1836 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < | |
1837 | hindex + HPAGE_PMD_NR - 1) { | |
1838 | /* | |
1839 | * Part of the huge page is beyond i_size: subject | |
1840 | * to shrink under memory pressure. | |
1841 | */ | |
1842 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1843 | /* |
1844 | * _careful to defend against unlocked access to | |
1845 | * ->shrink_list in shmem_unused_huge_shrink() | |
1846 | */ | |
1847 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1848 | list_add_tail(&info->shrinklist, |
1849 | &sbinfo->shrinklist); | |
1850 | sbinfo->shrinklist_len++; | |
1851 | } | |
1852 | spin_unlock(&sbinfo->shrinklist_lock); | |
1853 | } | |
1854 | ||
ec9516fb | 1855 | /* |
1635f6a7 HD |
1856 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1857 | */ | |
1858 | if (sgp == SGP_FALLOC) | |
1859 | sgp = SGP_WRITE; | |
1860 | clear: | |
1861 | /* | |
1862 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1863 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1864 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb | 1865 | */ |
800d8c63 KS |
1866 | if (sgp != SGP_WRITE && !PageUptodate(page)) { |
1867 | struct page *head = compound_head(page); | |
1868 | int i; | |
1869 | ||
1870 | for (i = 0; i < (1 << compound_order(head)); i++) { | |
1871 | clear_highpage(head + i); | |
1872 | flush_dcache_page(head + i); | |
1873 | } | |
1874 | SetPageUptodate(head); | |
ec9516fb | 1875 | } |
1da177e4 | 1876 | } |
bde05d1c | 1877 | |
54af6042 | 1878 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1879 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1880 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1881 | if (alloced) { |
1882 | ClearPageDirty(page); | |
1883 | delete_from_page_cache(page); | |
4595ef88 | 1884 | spin_lock_irq(&info->lock); |
267a4c76 | 1885 | shmem_recalc_inode(inode); |
4595ef88 | 1886 | spin_unlock_irq(&info->lock); |
267a4c76 | 1887 | } |
54af6042 | 1888 | error = -EINVAL; |
267a4c76 | 1889 | goto unlock; |
e83c32e8 | 1890 | } |
800d8c63 | 1891 | *pagep = page + index - hindex; |
54af6042 | 1892 | return 0; |
1da177e4 | 1893 | |
59a16ead | 1894 | /* |
54af6042 | 1895 | * Error recovery. |
59a16ead | 1896 | */ |
54af6042 | 1897 | unacct: |
0f079694 | 1898 | shmem_inode_unacct_blocks(inode, 1 << compound_order(page)); |
800d8c63 KS |
1899 | |
1900 | if (PageTransHuge(page)) { | |
1901 | unlock_page(page); | |
1902 | put_page(page); | |
1903 | goto alloc_nohuge; | |
1904 | } | |
54af6042 | 1905 | failed: |
267a4c76 | 1906 | if (swap.val && !shmem_confirm_swap(mapping, index, swap)) |
d1899228 HD |
1907 | error = -EEXIST; |
1908 | unlock: | |
27ab7006 | 1909 | if (page) { |
54af6042 | 1910 | unlock_page(page); |
09cbfeaf | 1911 | put_page(page); |
54af6042 HD |
1912 | } |
1913 | if (error == -ENOSPC && !once++) { | |
4595ef88 | 1914 | spin_lock_irq(&info->lock); |
54af6042 | 1915 | shmem_recalc_inode(inode); |
4595ef88 | 1916 | spin_unlock_irq(&info->lock); |
27ab7006 | 1917 | goto repeat; |
ff36b801 | 1918 | } |
d1899228 | 1919 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1920 | goto repeat; |
1921 | return error; | |
1da177e4 LT |
1922 | } |
1923 | ||
10d20bd2 LT |
1924 | /* |
1925 | * This is like autoremove_wake_function, but it removes the wait queue | |
1926 | * entry unconditionally - even if something else had already woken the | |
1927 | * target. | |
1928 | */ | |
ac6424b9 | 1929 | static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
10d20bd2 LT |
1930 | { |
1931 | int ret = default_wake_function(wait, mode, sync, key); | |
2055da97 | 1932 | list_del_init(&wait->entry); |
10d20bd2 LT |
1933 | return ret; |
1934 | } | |
1935 | ||
11bac800 | 1936 | static int shmem_fault(struct vm_fault *vmf) |
1da177e4 | 1937 | { |
11bac800 | 1938 | struct vm_area_struct *vma = vmf->vma; |
496ad9aa | 1939 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 1940 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
657e3038 | 1941 | enum sgp_type sgp; |
1da177e4 | 1942 | int error; |
68da9f05 | 1943 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1944 | |
f00cdc6d HD |
1945 | /* |
1946 | * Trinity finds that probing a hole which tmpfs is punching can | |
1947 | * prevent the hole-punch from ever completing: which in turn | |
1948 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1949 | * faulting pages into the hole while it's being punched. Although |
1950 | * shmem_undo_range() does remove the additions, it may be unable to | |
1951 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1952 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1953 | * | |
1954 | * It does not matter if we sometimes reach this check just before the | |
1955 | * hole-punch begins, so that one fault then races with the punch: | |
1956 | * we just need to make racing faults a rare case. | |
1957 | * | |
1958 | * The implementation below would be much simpler if we just used a | |
1959 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1960 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1961 | */ |
1962 | if (unlikely(inode->i_private)) { | |
1963 | struct shmem_falloc *shmem_falloc; | |
1964 | ||
1965 | spin_lock(&inode->i_lock); | |
1966 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1967 | if (shmem_falloc && |
1968 | shmem_falloc->waitq && | |
1969 | vmf->pgoff >= shmem_falloc->start && | |
1970 | vmf->pgoff < shmem_falloc->next) { | |
1971 | wait_queue_head_t *shmem_falloc_waitq; | |
10d20bd2 | 1972 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); |
8e205f77 HD |
1973 | |
1974 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1975 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1976 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1977 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1978 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1979 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1980 | } |
8e205f77 HD |
1981 | |
1982 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1983 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1984 | TASK_UNINTERRUPTIBLE); | |
1985 | spin_unlock(&inode->i_lock); | |
1986 | schedule(); | |
1987 | ||
1988 | /* | |
1989 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1990 | * stack of the hole-punching task: shmem_falloc_waitq | |
1991 | * is usually invalid by the time we reach here, but | |
1992 | * finish_wait() does not dereference it in that case; | |
1993 | * though i_lock needed lest racing with wake_up_all(). | |
1994 | */ | |
1995 | spin_lock(&inode->i_lock); | |
1996 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1997 | spin_unlock(&inode->i_lock); | |
1998 | return ret; | |
f00cdc6d | 1999 | } |
8e205f77 | 2000 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
2001 | } |
2002 | ||
657e3038 | 2003 | sgp = SGP_CACHE; |
18600332 MH |
2004 | |
2005 | if ((vma->vm_flags & VM_NOHUGEPAGE) || | |
2006 | test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) | |
657e3038 | 2007 | sgp = SGP_NOHUGE; |
18600332 MH |
2008 | else if (vma->vm_flags & VM_HUGEPAGE) |
2009 | sgp = SGP_HUGE; | |
657e3038 KS |
2010 | |
2011 | error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp, | |
cfda0526 | 2012 | gfp, vma, vmf, &ret); |
d0217ac0 NP |
2013 | if (error) |
2014 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 2015 | return ret; |
1da177e4 LT |
2016 | } |
2017 | ||
c01d5b30 HD |
2018 | unsigned long shmem_get_unmapped_area(struct file *file, |
2019 | unsigned long uaddr, unsigned long len, | |
2020 | unsigned long pgoff, unsigned long flags) | |
2021 | { | |
2022 | unsigned long (*get_area)(struct file *, | |
2023 | unsigned long, unsigned long, unsigned long, unsigned long); | |
2024 | unsigned long addr; | |
2025 | unsigned long offset; | |
2026 | unsigned long inflated_len; | |
2027 | unsigned long inflated_addr; | |
2028 | unsigned long inflated_offset; | |
2029 | ||
2030 | if (len > TASK_SIZE) | |
2031 | return -ENOMEM; | |
2032 | ||
2033 | get_area = current->mm->get_unmapped_area; | |
2034 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2035 | ||
e496cf3d | 2036 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
c01d5b30 HD |
2037 | return addr; |
2038 | if (IS_ERR_VALUE(addr)) | |
2039 | return addr; | |
2040 | if (addr & ~PAGE_MASK) | |
2041 | return addr; | |
2042 | if (addr > TASK_SIZE - len) | |
2043 | return addr; | |
2044 | ||
2045 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2046 | return addr; | |
2047 | if (len < HPAGE_PMD_SIZE) | |
2048 | return addr; | |
2049 | if (flags & MAP_FIXED) | |
2050 | return addr; | |
2051 | /* | |
2052 | * Our priority is to support MAP_SHARED mapped hugely; | |
2053 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
2054 | * But if caller specified an address hint, respect that as before. | |
2055 | */ | |
2056 | if (uaddr) | |
2057 | return addr; | |
2058 | ||
2059 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2060 | struct super_block *sb; | |
2061 | ||
2062 | if (file) { | |
2063 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2064 | sb = file_inode(file)->i_sb; | |
2065 | } else { | |
2066 | /* | |
2067 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2068 | * for "/dev/zero", to create a shared anonymous object. | |
2069 | */ | |
2070 | if (IS_ERR(shm_mnt)) | |
2071 | return addr; | |
2072 | sb = shm_mnt->mnt_sb; | |
2073 | } | |
3089bf61 | 2074 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2075 | return addr; |
2076 | } | |
2077 | ||
2078 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2079 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2080 | return addr; | |
2081 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2082 | return addr; | |
2083 | ||
2084 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2085 | if (inflated_len > TASK_SIZE) | |
2086 | return addr; | |
2087 | if (inflated_len < len) | |
2088 | return addr; | |
2089 | ||
2090 | inflated_addr = get_area(NULL, 0, inflated_len, 0, flags); | |
2091 | if (IS_ERR_VALUE(inflated_addr)) | |
2092 | return addr; | |
2093 | if (inflated_addr & ~PAGE_MASK) | |
2094 | return addr; | |
2095 | ||
2096 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2097 | inflated_addr += offset - inflated_offset; | |
2098 | if (inflated_offset > offset) | |
2099 | inflated_addr += HPAGE_PMD_SIZE; | |
2100 | ||
2101 | if (inflated_addr > TASK_SIZE - len) | |
2102 | return addr; | |
2103 | return inflated_addr; | |
2104 | } | |
2105 | ||
1da177e4 | 2106 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2107 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2108 | { |
496ad9aa | 2109 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2110 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2111 | } |
2112 | ||
d8dc74f2 AB |
2113 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
2114 | unsigned long addr) | |
1da177e4 | 2115 | { |
496ad9aa | 2116 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2117 | pgoff_t index; |
1da177e4 | 2118 | |
41ffe5d5 HD |
2119 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2120 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
2121 | } |
2122 | #endif | |
2123 | ||
2124 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
2125 | { | |
496ad9aa | 2126 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2127 | struct shmem_inode_info *info = SHMEM_I(inode); |
2128 | int retval = -ENOMEM; | |
2129 | ||
4595ef88 | 2130 | spin_lock_irq(&info->lock); |
1da177e4 LT |
2131 | if (lock && !(info->flags & VM_LOCKED)) { |
2132 | if (!user_shm_lock(inode->i_size, user)) | |
2133 | goto out_nomem; | |
2134 | info->flags |= VM_LOCKED; | |
89e004ea | 2135 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
2136 | } |
2137 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
2138 | user_shm_unlock(inode->i_size, user); | |
2139 | info->flags &= ~VM_LOCKED; | |
89e004ea | 2140 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2141 | } |
2142 | retval = 0; | |
89e004ea | 2143 | |
1da177e4 | 2144 | out_nomem: |
4595ef88 | 2145 | spin_unlock_irq(&info->lock); |
1da177e4 LT |
2146 | return retval; |
2147 | } | |
2148 | ||
9b83a6a8 | 2149 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
2150 | { |
2151 | file_accessed(file); | |
2152 | vma->vm_ops = &shmem_vm_ops; | |
e496cf3d | 2153 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
2154 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
2155 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
2156 | khugepaged_enter(vma, vma->vm_flags); | |
2157 | } | |
1da177e4 LT |
2158 | return 0; |
2159 | } | |
2160 | ||
454abafe | 2161 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 2162 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
2163 | { |
2164 | struct inode *inode; | |
2165 | struct shmem_inode_info *info; | |
2166 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
2167 | ||
5b04c689 PE |
2168 | if (shmem_reserve_inode(sb)) |
2169 | return NULL; | |
1da177e4 LT |
2170 | |
2171 | inode = new_inode(sb); | |
2172 | if (inode) { | |
85fe4025 | 2173 | inode->i_ino = get_next_ino(); |
454abafe | 2174 | inode_init_owner(inode, dir, mode); |
1da177e4 | 2175 | inode->i_blocks = 0; |
078cd827 | 2176 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
91828a40 | 2177 | inode->i_generation = get_seconds(); |
1da177e4 LT |
2178 | info = SHMEM_I(inode); |
2179 | memset(info, 0, (char *)inode - (char *)info); | |
2180 | spin_lock_init(&info->lock); | |
40e041a2 | 2181 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 2182 | info->flags = flags & VM_NORESERVE; |
779750d2 | 2183 | INIT_LIST_HEAD(&info->shrinklist); |
1da177e4 | 2184 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 2185 | simple_xattrs_init(&info->xattrs); |
72c04902 | 2186 | cache_no_acl(inode); |
1da177e4 LT |
2187 | |
2188 | switch (mode & S_IFMT) { | |
2189 | default: | |
39f0247d | 2190 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
2191 | init_special_inode(inode, mode, dev); |
2192 | break; | |
2193 | case S_IFREG: | |
14fcc23f | 2194 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
2195 | inode->i_op = &shmem_inode_operations; |
2196 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
2197 | mpol_shared_policy_init(&info->policy, |
2198 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
2199 | break; |
2200 | case S_IFDIR: | |
d8c76e6f | 2201 | inc_nlink(inode); |
1da177e4 LT |
2202 | /* Some things misbehave if size == 0 on a directory */ |
2203 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2204 | inode->i_op = &shmem_dir_inode_operations; | |
2205 | inode->i_fop = &simple_dir_operations; | |
2206 | break; | |
2207 | case S_IFLNK: | |
2208 | /* | |
2209 | * Must not load anything in the rbtree, | |
2210 | * mpol_free_shared_policy will not be called. | |
2211 | */ | |
71fe804b | 2212 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
2213 | break; |
2214 | } | |
5b04c689 PE |
2215 | } else |
2216 | shmem_free_inode(sb); | |
1da177e4 LT |
2217 | return inode; |
2218 | } | |
2219 | ||
0cd6144a JW |
2220 | bool shmem_mapping(struct address_space *mapping) |
2221 | { | |
f8005451 | 2222 | return mapping->a_ops == &shmem_aops; |
0cd6144a JW |
2223 | } |
2224 | ||
8d103963 MR |
2225 | static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm, |
2226 | pmd_t *dst_pmd, | |
2227 | struct vm_area_struct *dst_vma, | |
2228 | unsigned long dst_addr, | |
2229 | unsigned long src_addr, | |
2230 | bool zeropage, | |
2231 | struct page **pagep) | |
4c27fe4c MR |
2232 | { |
2233 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2234 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4c27fe4c MR |
2235 | struct address_space *mapping = inode->i_mapping; |
2236 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2237 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
2238 | struct mem_cgroup *memcg; | |
2239 | spinlock_t *ptl; | |
2240 | void *page_kaddr; | |
2241 | struct page *page; | |
2242 | pte_t _dst_pte, *dst_pte; | |
2243 | int ret; | |
2244 | ||
cb658a45 | 2245 | ret = -ENOMEM; |
0f079694 | 2246 | if (!shmem_inode_acct_block(inode, 1)) |
cb658a45 | 2247 | goto out; |
4c27fe4c | 2248 | |
cb658a45 | 2249 | if (!*pagep) { |
4c27fe4c MR |
2250 | page = shmem_alloc_page(gfp, info, pgoff); |
2251 | if (!page) | |
0f079694 | 2252 | goto out_unacct_blocks; |
4c27fe4c | 2253 | |
8d103963 MR |
2254 | if (!zeropage) { /* mcopy_atomic */ |
2255 | page_kaddr = kmap_atomic(page); | |
2256 | ret = copy_from_user(page_kaddr, | |
2257 | (const void __user *)src_addr, | |
2258 | PAGE_SIZE); | |
2259 | kunmap_atomic(page_kaddr); | |
2260 | ||
2261 | /* fallback to copy_from_user outside mmap_sem */ | |
2262 | if (unlikely(ret)) { | |
2263 | *pagep = page; | |
2264 | shmem_inode_unacct_blocks(inode, 1); | |
2265 | /* don't free the page */ | |
2266 | return -EFAULT; | |
2267 | } | |
2268 | } else { /* mfill_zeropage_atomic */ | |
2269 | clear_highpage(page); | |
4c27fe4c MR |
2270 | } |
2271 | } else { | |
2272 | page = *pagep; | |
2273 | *pagep = NULL; | |
2274 | } | |
2275 | ||
9cc90c66 AA |
2276 | VM_BUG_ON(PageLocked(page) || PageSwapBacked(page)); |
2277 | __SetPageLocked(page); | |
2278 | __SetPageSwapBacked(page); | |
a425d358 | 2279 | __SetPageUptodate(page); |
9cc90c66 | 2280 | |
4c27fe4c MR |
2281 | ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false); |
2282 | if (ret) | |
2283 | goto out_release; | |
2284 | ||
2285 | ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK); | |
2286 | if (!ret) { | |
2287 | ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL); | |
2288 | radix_tree_preload_end(); | |
2289 | } | |
2290 | if (ret) | |
2291 | goto out_release_uncharge; | |
2292 | ||
2293 | mem_cgroup_commit_charge(page, memcg, false, false); | |
2294 | ||
2295 | _dst_pte = mk_pte(page, dst_vma->vm_page_prot); | |
2296 | if (dst_vma->vm_flags & VM_WRITE) | |
2297 | _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte)); | |
2298 | ||
2299 | ret = -EEXIST; | |
2300 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); | |
2301 | if (!pte_none(*dst_pte)) | |
2302 | goto out_release_uncharge_unlock; | |
2303 | ||
4c27fe4c MR |
2304 | lru_cache_add_anon(page); |
2305 | ||
2306 | spin_lock(&info->lock); | |
2307 | info->alloced++; | |
2308 | inode->i_blocks += BLOCKS_PER_PAGE; | |
2309 | shmem_recalc_inode(inode); | |
2310 | spin_unlock(&info->lock); | |
2311 | ||
2312 | inc_mm_counter(dst_mm, mm_counter_file(page)); | |
2313 | page_add_file_rmap(page, false); | |
2314 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
2315 | ||
2316 | /* No need to invalidate - it was non-present before */ | |
2317 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
2318 | unlock_page(page); | |
2319 | pte_unmap_unlock(dst_pte, ptl); | |
2320 | ret = 0; | |
2321 | out: | |
2322 | return ret; | |
2323 | out_release_uncharge_unlock: | |
2324 | pte_unmap_unlock(dst_pte, ptl); | |
2325 | out_release_uncharge: | |
2326 | mem_cgroup_cancel_charge(page, memcg, false); | |
2327 | out_release: | |
9cc90c66 | 2328 | unlock_page(page); |
4c27fe4c | 2329 | put_page(page); |
4c27fe4c | 2330 | out_unacct_blocks: |
0f079694 | 2331 | shmem_inode_unacct_blocks(inode, 1); |
4c27fe4c MR |
2332 | goto out; |
2333 | } | |
2334 | ||
8d103963 MR |
2335 | int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm, |
2336 | pmd_t *dst_pmd, | |
2337 | struct vm_area_struct *dst_vma, | |
2338 | unsigned long dst_addr, | |
2339 | unsigned long src_addr, | |
2340 | struct page **pagep) | |
2341 | { | |
2342 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2343 | dst_addr, src_addr, false, pagep); | |
2344 | } | |
2345 | ||
2346 | int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm, | |
2347 | pmd_t *dst_pmd, | |
2348 | struct vm_area_struct *dst_vma, | |
2349 | unsigned long dst_addr) | |
2350 | { | |
2351 | struct page *page = NULL; | |
2352 | ||
2353 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2354 | dst_addr, 0, true, &page); | |
2355 | } | |
2356 | ||
1da177e4 | 2357 | #ifdef CONFIG_TMPFS |
92e1d5be | 2358 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2359 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2360 | |
6d9d88d0 JS |
2361 | #ifdef CONFIG_TMPFS_XATTR |
2362 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2363 | #else | |
2364 | #define shmem_initxattrs NULL | |
2365 | #endif | |
2366 | ||
1da177e4 | 2367 | static int |
800d15a5 NP |
2368 | shmem_write_begin(struct file *file, struct address_space *mapping, |
2369 | loff_t pos, unsigned len, unsigned flags, | |
2370 | struct page **pagep, void **fsdata) | |
1da177e4 | 2371 | { |
800d15a5 | 2372 | struct inode *inode = mapping->host; |
40e041a2 | 2373 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2374 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
2375 | |
2376 | /* i_mutex is held by caller */ | |
3f472cc9 | 2377 | if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) { |
40e041a2 DH |
2378 | if (info->seals & F_SEAL_WRITE) |
2379 | return -EPERM; | |
2380 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2381 | return -EPERM; | |
2382 | } | |
2383 | ||
9e18eb29 | 2384 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
2385 | } |
2386 | ||
2387 | static int | |
2388 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2389 | loff_t pos, unsigned len, unsigned copied, | |
2390 | struct page *page, void *fsdata) | |
2391 | { | |
2392 | struct inode *inode = mapping->host; | |
2393 | ||
d3602444 HD |
2394 | if (pos + copied > inode->i_size) |
2395 | i_size_write(inode, pos + copied); | |
2396 | ||
ec9516fb | 2397 | if (!PageUptodate(page)) { |
800d8c63 KS |
2398 | struct page *head = compound_head(page); |
2399 | if (PageTransCompound(page)) { | |
2400 | int i; | |
2401 | ||
2402 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2403 | if (head + i == page) | |
2404 | continue; | |
2405 | clear_highpage(head + i); | |
2406 | flush_dcache_page(head + i); | |
2407 | } | |
2408 | } | |
09cbfeaf KS |
2409 | if (copied < PAGE_SIZE) { |
2410 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2411 | zero_user_segments(page, 0, from, |
09cbfeaf | 2412 | from + copied, PAGE_SIZE); |
ec9516fb | 2413 | } |
800d8c63 | 2414 | SetPageUptodate(head); |
ec9516fb | 2415 | } |
800d15a5 | 2416 | set_page_dirty(page); |
6746aff7 | 2417 | unlock_page(page); |
09cbfeaf | 2418 | put_page(page); |
800d15a5 | 2419 | |
800d15a5 | 2420 | return copied; |
1da177e4 LT |
2421 | } |
2422 | ||
2ba5bbed | 2423 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2424 | { |
6e58e79d AV |
2425 | struct file *file = iocb->ki_filp; |
2426 | struct inode *inode = file_inode(file); | |
1da177e4 | 2427 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2428 | pgoff_t index; |
2429 | unsigned long offset; | |
a0ee5ec5 | 2430 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2431 | int error = 0; |
cb66a7a1 | 2432 | ssize_t retval = 0; |
6e58e79d | 2433 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2434 | |
2435 | /* | |
2436 | * Might this read be for a stacking filesystem? Then when reading | |
2437 | * holes of a sparse file, we actually need to allocate those pages, | |
2438 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2439 | */ | |
777eda2c | 2440 | if (!iter_is_iovec(to)) |
75edd345 | 2441 | sgp = SGP_CACHE; |
1da177e4 | 2442 | |
09cbfeaf KS |
2443 | index = *ppos >> PAGE_SHIFT; |
2444 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2445 | |
2446 | for (;;) { | |
2447 | struct page *page = NULL; | |
41ffe5d5 HD |
2448 | pgoff_t end_index; |
2449 | unsigned long nr, ret; | |
1da177e4 LT |
2450 | loff_t i_size = i_size_read(inode); |
2451 | ||
09cbfeaf | 2452 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2453 | if (index > end_index) |
2454 | break; | |
2455 | if (index == end_index) { | |
09cbfeaf | 2456 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2457 | if (nr <= offset) |
2458 | break; | |
2459 | } | |
2460 | ||
9e18eb29 | 2461 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2462 | if (error) { |
2463 | if (error == -EINVAL) | |
2464 | error = 0; | |
1da177e4 LT |
2465 | break; |
2466 | } | |
75edd345 HD |
2467 | if (page) { |
2468 | if (sgp == SGP_CACHE) | |
2469 | set_page_dirty(page); | |
d3602444 | 2470 | unlock_page(page); |
75edd345 | 2471 | } |
1da177e4 LT |
2472 | |
2473 | /* | |
2474 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2475 | * are called without i_mutex protection against truncate |
1da177e4 | 2476 | */ |
09cbfeaf | 2477 | nr = PAGE_SIZE; |
1da177e4 | 2478 | i_size = i_size_read(inode); |
09cbfeaf | 2479 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2480 | if (index == end_index) { |
09cbfeaf | 2481 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2482 | if (nr <= offset) { |
2483 | if (page) | |
09cbfeaf | 2484 | put_page(page); |
1da177e4 LT |
2485 | break; |
2486 | } | |
2487 | } | |
2488 | nr -= offset; | |
2489 | ||
2490 | if (page) { | |
2491 | /* | |
2492 | * If users can be writing to this page using arbitrary | |
2493 | * virtual addresses, take care about potential aliasing | |
2494 | * before reading the page on the kernel side. | |
2495 | */ | |
2496 | if (mapping_writably_mapped(mapping)) | |
2497 | flush_dcache_page(page); | |
2498 | /* | |
2499 | * Mark the page accessed if we read the beginning. | |
2500 | */ | |
2501 | if (!offset) | |
2502 | mark_page_accessed(page); | |
b5810039 | 2503 | } else { |
1da177e4 | 2504 | page = ZERO_PAGE(0); |
09cbfeaf | 2505 | get_page(page); |
b5810039 | 2506 | } |
1da177e4 LT |
2507 | |
2508 | /* | |
2509 | * Ok, we have the page, and it's up-to-date, so | |
2510 | * now we can copy it to user space... | |
1da177e4 | 2511 | */ |
2ba5bbed | 2512 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2513 | retval += ret; |
1da177e4 | 2514 | offset += ret; |
09cbfeaf KS |
2515 | index += offset >> PAGE_SHIFT; |
2516 | offset &= ~PAGE_MASK; | |
1da177e4 | 2517 | |
09cbfeaf | 2518 | put_page(page); |
2ba5bbed | 2519 | if (!iov_iter_count(to)) |
1da177e4 | 2520 | break; |
6e58e79d AV |
2521 | if (ret < nr) { |
2522 | error = -EFAULT; | |
2523 | break; | |
2524 | } | |
1da177e4 LT |
2525 | cond_resched(); |
2526 | } | |
2527 | ||
09cbfeaf | 2528 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2529 | file_accessed(file); |
2530 | return retval ? retval : error; | |
1da177e4 LT |
2531 | } |
2532 | ||
220f2ac9 HD |
2533 | /* |
2534 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
2535 | */ | |
2536 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2537 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2538 | { |
2539 | struct page *page; | |
2540 | struct pagevec pvec; | |
2541 | pgoff_t indices[PAGEVEC_SIZE]; | |
2542 | bool done = false; | |
2543 | int i; | |
2544 | ||
86679820 | 2545 | pagevec_init(&pvec); |
220f2ac9 HD |
2546 | pvec.nr = 1; /* start small: we may be there already */ |
2547 | while (!done) { | |
0cd6144a | 2548 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2549 | pvec.nr, pvec.pages, indices); |
2550 | if (!pvec.nr) { | |
965c8e59 | 2551 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2552 | index = end; |
2553 | break; | |
2554 | } | |
2555 | for (i = 0; i < pvec.nr; i++, index++) { | |
2556 | if (index < indices[i]) { | |
965c8e59 | 2557 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2558 | done = true; |
2559 | break; | |
2560 | } | |
2561 | index = indices[i]; | |
2562 | } | |
2563 | page = pvec.pages[i]; | |
2564 | if (page && !radix_tree_exceptional_entry(page)) { | |
2565 | if (!PageUptodate(page)) | |
2566 | page = NULL; | |
2567 | } | |
2568 | if (index >= end || | |
965c8e59 AM |
2569 | (page && whence == SEEK_DATA) || |
2570 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2571 | done = true; |
2572 | break; | |
2573 | } | |
2574 | } | |
0cd6144a | 2575 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2576 | pagevec_release(&pvec); |
2577 | pvec.nr = PAGEVEC_SIZE; | |
2578 | cond_resched(); | |
2579 | } | |
2580 | return index; | |
2581 | } | |
2582 | ||
965c8e59 | 2583 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2584 | { |
2585 | struct address_space *mapping = file->f_mapping; | |
2586 | struct inode *inode = mapping->host; | |
2587 | pgoff_t start, end; | |
2588 | loff_t new_offset; | |
2589 | ||
965c8e59 AM |
2590 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2591 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2592 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2593 | inode_lock(inode); |
220f2ac9 HD |
2594 | /* We're holding i_mutex so we can access i_size directly */ |
2595 | ||
2596 | if (offset < 0) | |
2597 | offset = -EINVAL; | |
2598 | else if (offset >= inode->i_size) | |
2599 | offset = -ENXIO; | |
2600 | else { | |
09cbfeaf KS |
2601 | start = offset >> PAGE_SHIFT; |
2602 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2603 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2604 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2605 | if (new_offset > offset) { |
2606 | if (new_offset < inode->i_size) | |
2607 | offset = new_offset; | |
965c8e59 | 2608 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2609 | offset = -ENXIO; |
2610 | else | |
2611 | offset = inode->i_size; | |
2612 | } | |
2613 | } | |
2614 | ||
387aae6f HD |
2615 | if (offset >= 0) |
2616 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2617 | inode_unlock(inode); |
220f2ac9 HD |
2618 | return offset; |
2619 | } | |
2620 | ||
05f65b5c DH |
2621 | /* |
2622 | * We need a tag: a new tag would expand every radix_tree_node by 8 bytes, | |
2623 | * so reuse a tag which we firmly believe is never set or cleared on shmem. | |
2624 | */ | |
2625 | #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE | |
2626 | #define LAST_SCAN 4 /* about 150ms max */ | |
2627 | ||
2628 | static void shmem_tag_pins(struct address_space *mapping) | |
2629 | { | |
2630 | struct radix_tree_iter iter; | |
5b9c98f3 | 2631 | void __rcu **slot; |
05f65b5c DH |
2632 | pgoff_t start; |
2633 | struct page *page; | |
2634 | ||
2635 | lru_add_drain(); | |
2636 | start = 0; | |
2637 | rcu_read_lock(); | |
2638 | ||
b93b0163 | 2639 | radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) { |
05f65b5c DH |
2640 | page = radix_tree_deref_slot(slot); |
2641 | if (!page || radix_tree_exception(page)) { | |
2cf938aa MW |
2642 | if (radix_tree_deref_retry(page)) { |
2643 | slot = radix_tree_iter_retry(&iter); | |
2644 | continue; | |
2645 | } | |
05f65b5c | 2646 | } else if (page_count(page) - page_mapcount(page) > 1) { |
b93b0163 MW |
2647 | xa_lock_irq(&mapping->i_pages); |
2648 | radix_tree_tag_set(&mapping->i_pages, iter.index, | |
05f65b5c | 2649 | SHMEM_TAG_PINNED); |
b93b0163 | 2650 | xa_unlock_irq(&mapping->i_pages); |
05f65b5c DH |
2651 | } |
2652 | ||
2653 | if (need_resched()) { | |
148deab2 | 2654 | slot = radix_tree_iter_resume(slot, &iter); |
05f65b5c | 2655 | cond_resched_rcu(); |
05f65b5c DH |
2656 | } |
2657 | } | |
2658 | rcu_read_unlock(); | |
2659 | } | |
2660 | ||
2661 | /* | |
2662 | * Setting SEAL_WRITE requires us to verify there's no pending writer. However, | |
2663 | * via get_user_pages(), drivers might have some pending I/O without any active | |
2664 | * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages | |
2665 | * and see whether it has an elevated ref-count. If so, we tag them and wait for | |
2666 | * them to be dropped. | |
2667 | * The caller must guarantee that no new user will acquire writable references | |
2668 | * to those pages to avoid races. | |
2669 | */ | |
40e041a2 DH |
2670 | static int shmem_wait_for_pins(struct address_space *mapping) |
2671 | { | |
05f65b5c | 2672 | struct radix_tree_iter iter; |
5b9c98f3 | 2673 | void __rcu **slot; |
05f65b5c DH |
2674 | pgoff_t start; |
2675 | struct page *page; | |
2676 | int error, scan; | |
2677 | ||
2678 | shmem_tag_pins(mapping); | |
2679 | ||
2680 | error = 0; | |
2681 | for (scan = 0; scan <= LAST_SCAN; scan++) { | |
b93b0163 | 2682 | if (!radix_tree_tagged(&mapping->i_pages, SHMEM_TAG_PINNED)) |
05f65b5c DH |
2683 | break; |
2684 | ||
2685 | if (!scan) | |
2686 | lru_add_drain_all(); | |
2687 | else if (schedule_timeout_killable((HZ << scan) / 200)) | |
2688 | scan = LAST_SCAN; | |
2689 | ||
2690 | start = 0; | |
2691 | rcu_read_lock(); | |
b93b0163 | 2692 | radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter, |
05f65b5c DH |
2693 | start, SHMEM_TAG_PINNED) { |
2694 | ||
2695 | page = radix_tree_deref_slot(slot); | |
2696 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2697 | if (radix_tree_deref_retry(page)) { |
2698 | slot = radix_tree_iter_retry(&iter); | |
2699 | continue; | |
2700 | } | |
05f65b5c DH |
2701 | |
2702 | page = NULL; | |
2703 | } | |
2704 | ||
2705 | if (page && | |
2706 | page_count(page) - page_mapcount(page) != 1) { | |
2707 | if (scan < LAST_SCAN) | |
2708 | goto continue_resched; | |
2709 | ||
2710 | /* | |
2711 | * On the last scan, we clean up all those tags | |
2712 | * we inserted; but make a note that we still | |
2713 | * found pages pinned. | |
2714 | */ | |
2715 | error = -EBUSY; | |
2716 | } | |
2717 | ||
b93b0163 MW |
2718 | xa_lock_irq(&mapping->i_pages); |
2719 | radix_tree_tag_clear(&mapping->i_pages, | |
05f65b5c | 2720 | iter.index, SHMEM_TAG_PINNED); |
b93b0163 | 2721 | xa_unlock_irq(&mapping->i_pages); |
05f65b5c DH |
2722 | continue_resched: |
2723 | if (need_resched()) { | |
148deab2 | 2724 | slot = radix_tree_iter_resume(slot, &iter); |
05f65b5c | 2725 | cond_resched_rcu(); |
05f65b5c DH |
2726 | } |
2727 | } | |
2728 | rcu_read_unlock(); | |
2729 | } | |
2730 | ||
2731 | return error; | |
40e041a2 DH |
2732 | } |
2733 | ||
47b9012e MAL |
2734 | static unsigned int *memfd_file_seals_ptr(struct file *file) |
2735 | { | |
2736 | if (file->f_op == &shmem_file_operations) | |
2737 | return &SHMEM_I(file_inode(file))->seals; | |
2738 | ||
2739 | #ifdef CONFIG_HUGETLBFS | |
2740 | if (file->f_op == &hugetlbfs_file_operations) | |
2741 | return &HUGETLBFS_I(file_inode(file))->seals; | |
2742 | #endif | |
2743 | ||
2744 | return NULL; | |
2745 | } | |
2746 | ||
40e041a2 DH |
2747 | #define F_ALL_SEALS (F_SEAL_SEAL | \ |
2748 | F_SEAL_SHRINK | \ | |
2749 | F_SEAL_GROW | \ | |
2750 | F_SEAL_WRITE) | |
2751 | ||
5aadc431 | 2752 | static int memfd_add_seals(struct file *file, unsigned int seals) |
40e041a2 DH |
2753 | { |
2754 | struct inode *inode = file_inode(file); | |
47b9012e | 2755 | unsigned int *file_seals; |
40e041a2 DH |
2756 | int error; |
2757 | ||
2758 | /* | |
2759 | * SEALING | |
2760 | * Sealing allows multiple parties to share a shmem-file but restrict | |
2761 | * access to a specific subset of file operations. Seals can only be | |
2762 | * added, but never removed. This way, mutually untrusted parties can | |
2763 | * share common memory regions with a well-defined policy. A malicious | |
2764 | * peer can thus never perform unwanted operations on a shared object. | |
2765 | * | |
2766 | * Seals are only supported on special shmem-files and always affect | |
2767 | * the whole underlying inode. Once a seal is set, it may prevent some | |
2768 | * kinds of access to the file. Currently, the following seals are | |
2769 | * defined: | |
2770 | * SEAL_SEAL: Prevent further seals from being set on this file | |
2771 | * SEAL_SHRINK: Prevent the file from shrinking | |
2772 | * SEAL_GROW: Prevent the file from growing | |
2773 | * SEAL_WRITE: Prevent write access to the file | |
2774 | * | |
2775 | * As we don't require any trust relationship between two parties, we | |
2776 | * must prevent seals from being removed. Therefore, sealing a file | |
2777 | * only adds a given set of seals to the file, it never touches | |
2778 | * existing seals. Furthermore, the "setting seals"-operation can be | |
2779 | * sealed itself, which basically prevents any further seal from being | |
2780 | * added. | |
2781 | * | |
2782 | * Semantics of sealing are only defined on volatile files. Only | |
2783 | * anonymous shmem files support sealing. More importantly, seals are | |
2784 | * never written to disk. Therefore, there's no plan to support it on | |
2785 | * other file types. | |
2786 | */ | |
2787 | ||
40e041a2 DH |
2788 | if (!(file->f_mode & FMODE_WRITE)) |
2789 | return -EPERM; | |
2790 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
2791 | return -EINVAL; | |
2792 | ||
5955102c | 2793 | inode_lock(inode); |
40e041a2 | 2794 | |
47b9012e MAL |
2795 | file_seals = memfd_file_seals_ptr(file); |
2796 | if (!file_seals) { | |
2797 | error = -EINVAL; | |
2798 | goto unlock; | |
2799 | } | |
2800 | ||
2801 | if (*file_seals & F_SEAL_SEAL) { | |
40e041a2 DH |
2802 | error = -EPERM; |
2803 | goto unlock; | |
2804 | } | |
2805 | ||
47b9012e | 2806 | if ((seals & F_SEAL_WRITE) && !(*file_seals & F_SEAL_WRITE)) { |
40e041a2 DH |
2807 | error = mapping_deny_writable(file->f_mapping); |
2808 | if (error) | |
2809 | goto unlock; | |
2810 | ||
2811 | error = shmem_wait_for_pins(file->f_mapping); | |
2812 | if (error) { | |
2813 | mapping_allow_writable(file->f_mapping); | |
2814 | goto unlock; | |
2815 | } | |
2816 | } | |
2817 | ||
47b9012e | 2818 | *file_seals |= seals; |
40e041a2 DH |
2819 | error = 0; |
2820 | ||
2821 | unlock: | |
5955102c | 2822 | inode_unlock(inode); |
40e041a2 DH |
2823 | return error; |
2824 | } | |
40e041a2 | 2825 | |
5aadc431 | 2826 | static int memfd_get_seals(struct file *file) |
40e041a2 | 2827 | { |
47b9012e | 2828 | unsigned int *seals = memfd_file_seals_ptr(file); |
40e041a2 | 2829 | |
47b9012e | 2830 | return seals ? *seals : -EINVAL; |
40e041a2 | 2831 | } |
40e041a2 | 2832 | |
5aadc431 | 2833 | long memfd_fcntl(struct file *file, unsigned int cmd, unsigned long arg) |
40e041a2 DH |
2834 | { |
2835 | long error; | |
2836 | ||
2837 | switch (cmd) { | |
2838 | case F_ADD_SEALS: | |
2839 | /* disallow upper 32bit */ | |
2840 | if (arg > UINT_MAX) | |
2841 | return -EINVAL; | |
2842 | ||
5aadc431 | 2843 | error = memfd_add_seals(file, arg); |
40e041a2 DH |
2844 | break; |
2845 | case F_GET_SEALS: | |
5aadc431 | 2846 | error = memfd_get_seals(file); |
40e041a2 DH |
2847 | break; |
2848 | default: | |
2849 | error = -EINVAL; | |
2850 | break; | |
2851 | } | |
2852 | ||
2853 | return error; | |
2854 | } | |
2855 | ||
83e4fa9c HD |
2856 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2857 | loff_t len) | |
2858 | { | |
496ad9aa | 2859 | struct inode *inode = file_inode(file); |
e2d12e22 | 2860 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2861 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2862 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2863 | pgoff_t start, index, end; |
2864 | int error; | |
83e4fa9c | 2865 | |
13ace4d0 HD |
2866 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2867 | return -EOPNOTSUPP; | |
2868 | ||
5955102c | 2869 | inode_lock(inode); |
83e4fa9c HD |
2870 | |
2871 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2872 | struct address_space *mapping = file->f_mapping; | |
2873 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2874 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2875 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2876 | |
40e041a2 DH |
2877 | /* protected by i_mutex */ |
2878 | if (info->seals & F_SEAL_WRITE) { | |
2879 | error = -EPERM; | |
2880 | goto out; | |
2881 | } | |
2882 | ||
8e205f77 | 2883 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
2884 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
2885 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
2886 | spin_lock(&inode->i_lock); | |
2887 | inode->i_private = &shmem_falloc; | |
2888 | spin_unlock(&inode->i_lock); | |
2889 | ||
83e4fa9c HD |
2890 | if ((u64)unmap_end > (u64)unmap_start) |
2891 | unmap_mapping_range(mapping, unmap_start, | |
2892 | 1 + unmap_end - unmap_start, 0); | |
2893 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2894 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2895 | |
2896 | spin_lock(&inode->i_lock); | |
2897 | inode->i_private = NULL; | |
2898 | wake_up_all(&shmem_falloc_waitq); | |
2055da97 | 2899 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head)); |
8e205f77 | 2900 | spin_unlock(&inode->i_lock); |
83e4fa9c | 2901 | error = 0; |
8e205f77 | 2902 | goto out; |
e2d12e22 HD |
2903 | } |
2904 | ||
2905 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2906 | error = inode_newsize_ok(inode, offset + len); | |
2907 | if (error) | |
2908 | goto out; | |
2909 | ||
40e041a2 DH |
2910 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2911 | error = -EPERM; | |
2912 | goto out; | |
2913 | } | |
2914 | ||
09cbfeaf KS |
2915 | start = offset >> PAGE_SHIFT; |
2916 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2917 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2918 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2919 | error = -ENOSPC; | |
2920 | goto out; | |
83e4fa9c HD |
2921 | } |
2922 | ||
8e205f77 | 2923 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2924 | shmem_falloc.start = start; |
2925 | shmem_falloc.next = start; | |
2926 | shmem_falloc.nr_falloced = 0; | |
2927 | shmem_falloc.nr_unswapped = 0; | |
2928 | spin_lock(&inode->i_lock); | |
2929 | inode->i_private = &shmem_falloc; | |
2930 | spin_unlock(&inode->i_lock); | |
2931 | ||
e2d12e22 HD |
2932 | for (index = start; index < end; index++) { |
2933 | struct page *page; | |
2934 | ||
2935 | /* | |
2936 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2937 | * been interrupted because we are using up too much memory. | |
2938 | */ | |
2939 | if (signal_pending(current)) | |
2940 | error = -EINTR; | |
1aac1400 HD |
2941 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2942 | error = -ENOMEM; | |
e2d12e22 | 2943 | else |
9e18eb29 | 2944 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2945 | if (error) { |
1635f6a7 | 2946 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2947 | if (index > start) { |
2948 | shmem_undo_range(inode, | |
2949 | (loff_t)start << PAGE_SHIFT, | |
2950 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2951 | } | |
1aac1400 | 2952 | goto undone; |
e2d12e22 HD |
2953 | } |
2954 | ||
1aac1400 HD |
2955 | /* |
2956 | * Inform shmem_writepage() how far we have reached. | |
2957 | * No need for lock or barrier: we have the page lock. | |
2958 | */ | |
2959 | shmem_falloc.next++; | |
2960 | if (!PageUptodate(page)) | |
2961 | shmem_falloc.nr_falloced++; | |
2962 | ||
e2d12e22 | 2963 | /* |
1635f6a7 HD |
2964 | * If !PageUptodate, leave it that way so that freeable pages |
2965 | * can be recognized if we need to rollback on error later. | |
2966 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2967 | * than free the pages we are allocating (and SGP_CACHE pages |
2968 | * might still be clean: we now need to mark those dirty too). | |
2969 | */ | |
2970 | set_page_dirty(page); | |
2971 | unlock_page(page); | |
09cbfeaf | 2972 | put_page(page); |
e2d12e22 HD |
2973 | cond_resched(); |
2974 | } | |
2975 | ||
2976 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2977 | i_size_write(inode, offset + len); | |
078cd827 | 2978 | inode->i_ctime = current_time(inode); |
1aac1400 HD |
2979 | undone: |
2980 | spin_lock(&inode->i_lock); | |
2981 | inode->i_private = NULL; | |
2982 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2983 | out: |
5955102c | 2984 | inode_unlock(inode); |
83e4fa9c HD |
2985 | return error; |
2986 | } | |
2987 | ||
726c3342 | 2988 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2989 | { |
726c3342 | 2990 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2991 | |
2992 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2993 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2994 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2995 | if (sbinfo->max_blocks) { |
1da177e4 | 2996 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2997 | buf->f_bavail = |
2998 | buf->f_bfree = sbinfo->max_blocks - | |
2999 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
3000 | } |
3001 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
3002 | buf->f_files = sbinfo->max_inodes; |
3003 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
3004 | } |
3005 | /* else leave those fields 0 like simple_statfs */ | |
3006 | return 0; | |
3007 | } | |
3008 | ||
3009 | /* | |
3010 | * File creation. Allocate an inode, and we're done.. | |
3011 | */ | |
3012 | static int | |
1a67aafb | 3013 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3014 | { |
0b0a0806 | 3015 | struct inode *inode; |
1da177e4 LT |
3016 | int error = -ENOSPC; |
3017 | ||
454abafe | 3018 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 3019 | if (inode) { |
feda821e CH |
3020 | error = simple_acl_create(dir, inode); |
3021 | if (error) | |
3022 | goto out_iput; | |
2a7dba39 | 3023 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 3024 | &dentry->d_name, |
6d9d88d0 | 3025 | shmem_initxattrs, NULL); |
feda821e CH |
3026 | if (error && error != -EOPNOTSUPP) |
3027 | goto out_iput; | |
37ec43cd | 3028 | |
718deb6b | 3029 | error = 0; |
1da177e4 | 3030 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3031 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3032 | d_instantiate(dentry, inode); |
3033 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
3034 | } |
3035 | return error; | |
feda821e CH |
3036 | out_iput: |
3037 | iput(inode); | |
3038 | return error; | |
1da177e4 LT |
3039 | } |
3040 | ||
60545d0d AV |
3041 | static int |
3042 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
3043 | { | |
3044 | struct inode *inode; | |
3045 | int error = -ENOSPC; | |
3046 | ||
3047 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
3048 | if (inode) { | |
3049 | error = security_inode_init_security(inode, dir, | |
3050 | NULL, | |
3051 | shmem_initxattrs, NULL); | |
feda821e CH |
3052 | if (error && error != -EOPNOTSUPP) |
3053 | goto out_iput; | |
3054 | error = simple_acl_create(dir, inode); | |
3055 | if (error) | |
3056 | goto out_iput; | |
60545d0d AV |
3057 | d_tmpfile(dentry, inode); |
3058 | } | |
3059 | return error; | |
feda821e CH |
3060 | out_iput: |
3061 | iput(inode); | |
3062 | return error; | |
60545d0d AV |
3063 | } |
3064 | ||
18bb1db3 | 3065 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
3066 | { |
3067 | int error; | |
3068 | ||
3069 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
3070 | return error; | |
d8c76e6f | 3071 | inc_nlink(dir); |
1da177e4 LT |
3072 | return 0; |
3073 | } | |
3074 | ||
4acdaf27 | 3075 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 3076 | bool excl) |
1da177e4 LT |
3077 | { |
3078 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
3079 | } | |
3080 | ||
3081 | /* | |
3082 | * Link a file.. | |
3083 | */ | |
3084 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
3085 | { | |
75c3cfa8 | 3086 | struct inode *inode = d_inode(old_dentry); |
5b04c689 | 3087 | int ret; |
1da177e4 LT |
3088 | |
3089 | /* | |
3090 | * No ordinary (disk based) filesystem counts links as inodes; | |
3091 | * but each new link needs a new dentry, pinning lowmem, and | |
3092 | * tmpfs dentries cannot be pruned until they are unlinked. | |
3093 | */ | |
5b04c689 PE |
3094 | ret = shmem_reserve_inode(inode->i_sb); |
3095 | if (ret) | |
3096 | goto out; | |
1da177e4 LT |
3097 | |
3098 | dir->i_size += BOGO_DIRENT_SIZE; | |
078cd827 | 3099 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
d8c76e6f | 3100 | inc_nlink(inode); |
7de9c6ee | 3101 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
3102 | dget(dentry); /* Extra pinning count for the created dentry */ |
3103 | d_instantiate(dentry, inode); | |
5b04c689 PE |
3104 | out: |
3105 | return ret; | |
1da177e4 LT |
3106 | } |
3107 | ||
3108 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
3109 | { | |
75c3cfa8 | 3110 | struct inode *inode = d_inode(dentry); |
1da177e4 | 3111 | |
5b04c689 PE |
3112 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
3113 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
3114 | |
3115 | dir->i_size -= BOGO_DIRENT_SIZE; | |
078cd827 | 3116 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
9a53c3a7 | 3117 | drop_nlink(inode); |
1da177e4 LT |
3118 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
3119 | return 0; | |
3120 | } | |
3121 | ||
3122 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
3123 | { | |
3124 | if (!simple_empty(dentry)) | |
3125 | return -ENOTEMPTY; | |
3126 | ||
75c3cfa8 | 3127 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 3128 | drop_nlink(dir); |
1da177e4 LT |
3129 | return shmem_unlink(dir, dentry); |
3130 | } | |
3131 | ||
37456771 MS |
3132 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
3133 | { | |
e36cb0b8 DH |
3134 | bool old_is_dir = d_is_dir(old_dentry); |
3135 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
3136 | |
3137 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
3138 | if (old_is_dir) { | |
3139 | drop_nlink(old_dir); | |
3140 | inc_nlink(new_dir); | |
3141 | } else { | |
3142 | drop_nlink(new_dir); | |
3143 | inc_nlink(old_dir); | |
3144 | } | |
3145 | } | |
3146 | old_dir->i_ctime = old_dir->i_mtime = | |
3147 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 | 3148 | d_inode(old_dentry)->i_ctime = |
078cd827 | 3149 | d_inode(new_dentry)->i_ctime = current_time(old_dir); |
37456771 MS |
3150 | |
3151 | return 0; | |
3152 | } | |
3153 | ||
46fdb794 MS |
3154 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
3155 | { | |
3156 | struct dentry *whiteout; | |
3157 | int error; | |
3158 | ||
3159 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
3160 | if (!whiteout) | |
3161 | return -ENOMEM; | |
3162 | ||
3163 | error = shmem_mknod(old_dir, whiteout, | |
3164 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
3165 | dput(whiteout); | |
3166 | if (error) | |
3167 | return error; | |
3168 | ||
3169 | /* | |
3170 | * Cheat and hash the whiteout while the old dentry is still in | |
3171 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
3172 | * | |
3173 | * d_lookup() will consistently find one of them at this point, | |
3174 | * not sure which one, but that isn't even important. | |
3175 | */ | |
3176 | d_rehash(whiteout); | |
3177 | return 0; | |
3178 | } | |
3179 | ||
1da177e4 LT |
3180 | /* |
3181 | * The VFS layer already does all the dentry stuff for rename, | |
3182 | * we just have to decrement the usage count for the target if | |
3183 | * it exists so that the VFS layer correctly free's it when it | |
3184 | * gets overwritten. | |
3185 | */ | |
3b69ff51 | 3186 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 3187 | { |
75c3cfa8 | 3188 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
3189 | int they_are_dirs = S_ISDIR(inode->i_mode); |
3190 | ||
46fdb794 | 3191 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
3192 | return -EINVAL; |
3193 | ||
37456771 MS |
3194 | if (flags & RENAME_EXCHANGE) |
3195 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
3196 | ||
1da177e4 LT |
3197 | if (!simple_empty(new_dentry)) |
3198 | return -ENOTEMPTY; | |
3199 | ||
46fdb794 MS |
3200 | if (flags & RENAME_WHITEOUT) { |
3201 | int error; | |
3202 | ||
3203 | error = shmem_whiteout(old_dir, old_dentry); | |
3204 | if (error) | |
3205 | return error; | |
3206 | } | |
3207 | ||
75c3cfa8 | 3208 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 3209 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 3210 | if (they_are_dirs) { |
75c3cfa8 | 3211 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 3212 | drop_nlink(old_dir); |
b928095b | 3213 | } |
1da177e4 | 3214 | } else if (they_are_dirs) { |
9a53c3a7 | 3215 | drop_nlink(old_dir); |
d8c76e6f | 3216 | inc_nlink(new_dir); |
1da177e4 LT |
3217 | } |
3218 | ||
3219 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
3220 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
3221 | old_dir->i_ctime = old_dir->i_mtime = | |
3222 | new_dir->i_ctime = new_dir->i_mtime = | |
078cd827 | 3223 | inode->i_ctime = current_time(old_dir); |
1da177e4 LT |
3224 | return 0; |
3225 | } | |
3226 | ||
3227 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
3228 | { | |
3229 | int error; | |
3230 | int len; | |
3231 | struct inode *inode; | |
9276aad6 | 3232 | struct page *page; |
1da177e4 LT |
3233 | |
3234 | len = strlen(symname) + 1; | |
09cbfeaf | 3235 | if (len > PAGE_SIZE) |
1da177e4 LT |
3236 | return -ENAMETOOLONG; |
3237 | ||
454abafe | 3238 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
3239 | if (!inode) |
3240 | return -ENOSPC; | |
3241 | ||
9d8f13ba | 3242 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3243 | shmem_initxattrs, NULL); |
570bc1c2 SS |
3244 | if (error) { |
3245 | if (error != -EOPNOTSUPP) { | |
3246 | iput(inode); | |
3247 | return error; | |
3248 | } | |
3249 | error = 0; | |
3250 | } | |
3251 | ||
1da177e4 | 3252 | inode->i_size = len-1; |
69f07ec9 | 3253 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3254 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3255 | if (!inode->i_link) { | |
69f07ec9 HD |
3256 | iput(inode); |
3257 | return -ENOMEM; | |
3258 | } | |
3259 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3260 | } else { |
e8ecde25 | 3261 | inode_nohighmem(inode); |
9e18eb29 | 3262 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
3263 | if (error) { |
3264 | iput(inode); | |
3265 | return error; | |
3266 | } | |
14fcc23f | 3267 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 3268 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 3269 | memcpy(page_address(page), symname, len); |
ec9516fb | 3270 | SetPageUptodate(page); |
1da177e4 | 3271 | set_page_dirty(page); |
6746aff7 | 3272 | unlock_page(page); |
09cbfeaf | 3273 | put_page(page); |
1da177e4 | 3274 | } |
1da177e4 | 3275 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3276 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3277 | d_instantiate(dentry, inode); |
3278 | dget(dentry); | |
3279 | return 0; | |
3280 | } | |
3281 | ||
fceef393 | 3282 | static void shmem_put_link(void *arg) |
1da177e4 | 3283 | { |
fceef393 AV |
3284 | mark_page_accessed(arg); |
3285 | put_page(arg); | |
1da177e4 LT |
3286 | } |
3287 | ||
6b255391 | 3288 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
3289 | struct inode *inode, |
3290 | struct delayed_call *done) | |
1da177e4 | 3291 | { |
1da177e4 | 3292 | struct page *page = NULL; |
6b255391 | 3293 | int error; |
6a6c9904 AV |
3294 | if (!dentry) { |
3295 | page = find_get_page(inode->i_mapping, 0); | |
3296 | if (!page) | |
3297 | return ERR_PTR(-ECHILD); | |
3298 | if (!PageUptodate(page)) { | |
3299 | put_page(page); | |
3300 | return ERR_PTR(-ECHILD); | |
3301 | } | |
3302 | } else { | |
9e18eb29 | 3303 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3304 | if (error) |
3305 | return ERR_PTR(error); | |
3306 | unlock_page(page); | |
3307 | } | |
fceef393 | 3308 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3309 | return page_address(page); |
1da177e4 LT |
3310 | } |
3311 | ||
b09e0fa4 | 3312 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3313 | /* |
b09e0fa4 EP |
3314 | * Superblocks without xattr inode operations may get some security.* xattr |
3315 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3316 | * like ACLs, we also need to implement the security.* handlers at |
3317 | * filesystem level, though. | |
3318 | */ | |
3319 | ||
6d9d88d0 JS |
3320 | /* |
3321 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3322 | */ | |
3323 | static int shmem_initxattrs(struct inode *inode, | |
3324 | const struct xattr *xattr_array, | |
3325 | void *fs_info) | |
3326 | { | |
3327 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3328 | const struct xattr *xattr; | |
38f38657 | 3329 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3330 | size_t len; |
3331 | ||
3332 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3333 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3334 | if (!new_xattr) |
3335 | return -ENOMEM; | |
3336 | ||
3337 | len = strlen(xattr->name) + 1; | |
3338 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3339 | GFP_KERNEL); | |
3340 | if (!new_xattr->name) { | |
3341 | kfree(new_xattr); | |
3342 | return -ENOMEM; | |
3343 | } | |
3344 | ||
3345 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3346 | XATTR_SECURITY_PREFIX_LEN); | |
3347 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3348 | xattr->name, len); | |
3349 | ||
38f38657 | 3350 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3351 | } |
3352 | ||
3353 | return 0; | |
3354 | } | |
3355 | ||
aa7c5241 | 3356 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3357 | struct dentry *unused, struct inode *inode, |
3358 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3359 | { |
b296821a | 3360 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3361 | |
aa7c5241 | 3362 | name = xattr_full_name(handler, name); |
38f38657 | 3363 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3364 | } |
3365 | ||
aa7c5241 | 3366 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
59301226 AV |
3367 | struct dentry *unused, struct inode *inode, |
3368 | const char *name, const void *value, | |
3369 | size_t size, int flags) | |
b09e0fa4 | 3370 | { |
59301226 | 3371 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3372 | |
aa7c5241 | 3373 | name = xattr_full_name(handler, name); |
38f38657 | 3374 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
3375 | } |
3376 | ||
aa7c5241 AG |
3377 | static const struct xattr_handler shmem_security_xattr_handler = { |
3378 | .prefix = XATTR_SECURITY_PREFIX, | |
3379 | .get = shmem_xattr_handler_get, | |
3380 | .set = shmem_xattr_handler_set, | |
3381 | }; | |
b09e0fa4 | 3382 | |
aa7c5241 AG |
3383 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3384 | .prefix = XATTR_TRUSTED_PREFIX, | |
3385 | .get = shmem_xattr_handler_get, | |
3386 | .set = shmem_xattr_handler_set, | |
3387 | }; | |
b09e0fa4 | 3388 | |
aa7c5241 AG |
3389 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3390 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3391 | &posix_acl_access_xattr_handler, | |
3392 | &posix_acl_default_xattr_handler, | |
3393 | #endif | |
3394 | &shmem_security_xattr_handler, | |
3395 | &shmem_trusted_xattr_handler, | |
3396 | NULL | |
3397 | }; | |
b09e0fa4 EP |
3398 | |
3399 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3400 | { | |
75c3cfa8 | 3401 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3402 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3403 | } |
3404 | #endif /* CONFIG_TMPFS_XATTR */ | |
3405 | ||
69f07ec9 | 3406 | static const struct inode_operations shmem_short_symlink_operations = { |
6b255391 | 3407 | .get_link = simple_get_link, |
b09e0fa4 | 3408 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3409 | .listxattr = shmem_listxattr, |
b09e0fa4 EP |
3410 | #endif |
3411 | }; | |
3412 | ||
3413 | static const struct inode_operations shmem_symlink_inode_operations = { | |
6b255391 | 3414 | .get_link = shmem_get_link, |
b09e0fa4 | 3415 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3416 | .listxattr = shmem_listxattr, |
39f0247d | 3417 | #endif |
b09e0fa4 | 3418 | }; |
39f0247d | 3419 | |
91828a40 DG |
3420 | static struct dentry *shmem_get_parent(struct dentry *child) |
3421 | { | |
3422 | return ERR_PTR(-ESTALE); | |
3423 | } | |
3424 | ||
3425 | static int shmem_match(struct inode *ino, void *vfh) | |
3426 | { | |
3427 | __u32 *fh = vfh; | |
3428 | __u64 inum = fh[2]; | |
3429 | inum = (inum << 32) | fh[1]; | |
3430 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
3431 | } | |
3432 | ||
480b116c CH |
3433 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3434 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3435 | { |
91828a40 | 3436 | struct inode *inode; |
480b116c | 3437 | struct dentry *dentry = NULL; |
35c2a7f4 | 3438 | u64 inum; |
480b116c CH |
3439 | |
3440 | if (fh_len < 3) | |
3441 | return NULL; | |
91828a40 | 3442 | |
35c2a7f4 HD |
3443 | inum = fid->raw[2]; |
3444 | inum = (inum << 32) | fid->raw[1]; | |
3445 | ||
480b116c CH |
3446 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
3447 | shmem_match, fid->raw); | |
91828a40 | 3448 | if (inode) { |
480b116c | 3449 | dentry = d_find_alias(inode); |
91828a40 DG |
3450 | iput(inode); |
3451 | } | |
3452 | ||
480b116c | 3453 | return dentry; |
91828a40 DG |
3454 | } |
3455 | ||
b0b0382b AV |
3456 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3457 | struct inode *parent) | |
91828a40 | 3458 | { |
5fe0c237 AK |
3459 | if (*len < 3) { |
3460 | *len = 3; | |
94e07a75 | 3461 | return FILEID_INVALID; |
5fe0c237 | 3462 | } |
91828a40 | 3463 | |
1d3382cb | 3464 | if (inode_unhashed(inode)) { |
91828a40 DG |
3465 | /* Unfortunately insert_inode_hash is not idempotent, |
3466 | * so as we hash inodes here rather than at creation | |
3467 | * time, we need a lock to ensure we only try | |
3468 | * to do it once | |
3469 | */ | |
3470 | static DEFINE_SPINLOCK(lock); | |
3471 | spin_lock(&lock); | |
1d3382cb | 3472 | if (inode_unhashed(inode)) |
91828a40 DG |
3473 | __insert_inode_hash(inode, |
3474 | inode->i_ino + inode->i_generation); | |
3475 | spin_unlock(&lock); | |
3476 | } | |
3477 | ||
3478 | fh[0] = inode->i_generation; | |
3479 | fh[1] = inode->i_ino; | |
3480 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
3481 | ||
3482 | *len = 3; | |
3483 | return 1; | |
3484 | } | |
3485 | ||
39655164 | 3486 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3487 | .get_parent = shmem_get_parent, |
91828a40 | 3488 | .encode_fh = shmem_encode_fh, |
480b116c | 3489 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3490 | }; |
3491 | ||
680d794b | 3492 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
3493 | bool remount) | |
1da177e4 LT |
3494 | { |
3495 | char *this_char, *value, *rest; | |
49cd0a5c | 3496 | struct mempolicy *mpol = NULL; |
8751e039 EB |
3497 | uid_t uid; |
3498 | gid_t gid; | |
1da177e4 | 3499 | |
b00dc3ad HD |
3500 | while (options != NULL) { |
3501 | this_char = options; | |
3502 | for (;;) { | |
3503 | /* | |
3504 | * NUL-terminate this option: unfortunately, | |
3505 | * mount options form a comma-separated list, | |
3506 | * but mpol's nodelist may also contain commas. | |
3507 | */ | |
3508 | options = strchr(options, ','); | |
3509 | if (options == NULL) | |
3510 | break; | |
3511 | options++; | |
3512 | if (!isdigit(*options)) { | |
3513 | options[-1] = '\0'; | |
3514 | break; | |
3515 | } | |
3516 | } | |
1da177e4 LT |
3517 | if (!*this_char) |
3518 | continue; | |
3519 | if ((value = strchr(this_char,'=')) != NULL) { | |
3520 | *value++ = 0; | |
3521 | } else { | |
1170532b JP |
3522 | pr_err("tmpfs: No value for mount option '%s'\n", |
3523 | this_char); | |
49cd0a5c | 3524 | goto error; |
1da177e4 LT |
3525 | } |
3526 | ||
3527 | if (!strcmp(this_char,"size")) { | |
3528 | unsigned long long size; | |
3529 | size = memparse(value,&rest); | |
3530 | if (*rest == '%') { | |
3531 | size <<= PAGE_SHIFT; | |
3532 | size *= totalram_pages; | |
3533 | do_div(size, 100); | |
3534 | rest++; | |
3535 | } | |
3536 | if (*rest) | |
3537 | goto bad_val; | |
680d794b | 3538 | sbinfo->max_blocks = |
09cbfeaf | 3539 | DIV_ROUND_UP(size, PAGE_SIZE); |
1da177e4 | 3540 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 3541 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
3542 | if (*rest) |
3543 | goto bad_val; | |
3544 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 3545 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
3546 | if (*rest) |
3547 | goto bad_val; | |
3548 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 3549 | if (remount) |
1da177e4 | 3550 | continue; |
680d794b | 3551 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
3552 | if (*rest) |
3553 | goto bad_val; | |
3554 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 3555 | if (remount) |
1da177e4 | 3556 | continue; |
8751e039 | 3557 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3558 | if (*rest) |
3559 | goto bad_val; | |
8751e039 EB |
3560 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
3561 | if (!uid_valid(sbinfo->uid)) | |
3562 | goto bad_val; | |
1da177e4 | 3563 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 3564 | if (remount) |
1da177e4 | 3565 | continue; |
8751e039 | 3566 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3567 | if (*rest) |
3568 | goto bad_val; | |
8751e039 EB |
3569 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
3570 | if (!gid_valid(sbinfo->gid)) | |
3571 | goto bad_val; | |
e496cf3d | 3572 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3573 | } else if (!strcmp(this_char, "huge")) { |
3574 | int huge; | |
3575 | huge = shmem_parse_huge(value); | |
3576 | if (huge < 0) | |
3577 | goto bad_val; | |
3578 | if (!has_transparent_hugepage() && | |
3579 | huge != SHMEM_HUGE_NEVER) | |
3580 | goto bad_val; | |
3581 | sbinfo->huge = huge; | |
3582 | #endif | |
3583 | #ifdef CONFIG_NUMA | |
7339ff83 | 3584 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
3585 | mpol_put(mpol); |
3586 | mpol = NULL; | |
3587 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 3588 | goto bad_val; |
5a6e75f8 | 3589 | #endif |
1da177e4 | 3590 | } else { |
1170532b | 3591 | pr_err("tmpfs: Bad mount option %s\n", this_char); |
49cd0a5c | 3592 | goto error; |
1da177e4 LT |
3593 | } |
3594 | } | |
49cd0a5c | 3595 | sbinfo->mpol = mpol; |
1da177e4 LT |
3596 | return 0; |
3597 | ||
3598 | bad_val: | |
1170532b | 3599 | pr_err("tmpfs: Bad value '%s' for mount option '%s'\n", |
1da177e4 | 3600 | value, this_char); |
49cd0a5c GT |
3601 | error: |
3602 | mpol_put(mpol); | |
1da177e4 LT |
3603 | return 1; |
3604 | ||
3605 | } | |
3606 | ||
3607 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
3608 | { | |
3609 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 3610 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
3611 | unsigned long inodes; |
3612 | int error = -EINVAL; | |
3613 | ||
5f00110f | 3614 | config.mpol = NULL; |
680d794b | 3615 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 3616 | return error; |
1da177e4 | 3617 | |
0edd73b3 | 3618 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3619 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 3620 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 3621 | goto out; |
680d794b | 3622 | if (config.max_inodes < inodes) |
0edd73b3 HD |
3623 | goto out; |
3624 | /* | |
54af6042 | 3625 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
3626 | * but we must separately disallow unlimited->limited, because |
3627 | * in that case we have no record of how much is already in use. | |
3628 | */ | |
680d794b | 3629 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 3630 | goto out; |
680d794b | 3631 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
3632 | goto out; |
3633 | ||
3634 | error = 0; | |
5a6e75f8 | 3635 | sbinfo->huge = config.huge; |
680d794b | 3636 | sbinfo->max_blocks = config.max_blocks; |
680d794b | 3637 | sbinfo->max_inodes = config.max_inodes; |
3638 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 3639 | |
5f00110f GT |
3640 | /* |
3641 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3642 | */ | |
3643 | if (config.mpol) { | |
3644 | mpol_put(sbinfo->mpol); | |
3645 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
3646 | } | |
0edd73b3 HD |
3647 | out: |
3648 | spin_unlock(&sbinfo->stat_lock); | |
3649 | return error; | |
1da177e4 | 3650 | } |
680d794b | 3651 | |
34c80b1d | 3652 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3653 | { |
34c80b1d | 3654 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b | 3655 | |
3656 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3657 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3658 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 3659 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
3660 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
3661 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 3662 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3663 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3664 | seq_printf(seq, ",uid=%u", | |
3665 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3666 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3667 | seq_printf(seq, ",gid=%u", | |
3668 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
e496cf3d | 3669 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3670 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
3671 | if (sbinfo->huge) | |
3672 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3673 | #endif | |
71fe804b | 3674 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 3675 | return 0; |
3676 | } | |
9183df25 DH |
3677 | |
3678 | #define MFD_NAME_PREFIX "memfd:" | |
3679 | #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1) | |
3680 | #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN) | |
3681 | ||
749df87b | 3682 | #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB) |
9183df25 DH |
3683 | |
3684 | SYSCALL_DEFINE2(memfd_create, | |
3685 | const char __user *, uname, | |
3686 | unsigned int, flags) | |
3687 | { | |
47b9012e | 3688 | unsigned int *file_seals; |
9183df25 DH |
3689 | struct file *file; |
3690 | int fd, error; | |
3691 | char *name; | |
3692 | long len; | |
3693 | ||
749df87b MK |
3694 | if (!(flags & MFD_HUGETLB)) { |
3695 | if (flags & ~(unsigned int)MFD_ALL_FLAGS) | |
3696 | return -EINVAL; | |
3697 | } else { | |
749df87b MK |
3698 | /* Allow huge page size encoding in flags. */ |
3699 | if (flags & ~(unsigned int)(MFD_ALL_FLAGS | | |
3700 | (MFD_HUGE_MASK << MFD_HUGE_SHIFT))) | |
3701 | return -EINVAL; | |
3702 | } | |
9183df25 DH |
3703 | |
3704 | /* length includes terminating zero */ | |
3705 | len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1); | |
3706 | if (len <= 0) | |
3707 | return -EFAULT; | |
3708 | if (len > MFD_NAME_MAX_LEN + 1) | |
3709 | return -EINVAL; | |
3710 | ||
0ee931c4 | 3711 | name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_KERNEL); |
9183df25 DH |
3712 | if (!name) |
3713 | return -ENOMEM; | |
3714 | ||
3715 | strcpy(name, MFD_NAME_PREFIX); | |
3716 | if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) { | |
3717 | error = -EFAULT; | |
3718 | goto err_name; | |
3719 | } | |
3720 | ||
3721 | /* terminating-zero may have changed after strnlen_user() returned */ | |
3722 | if (name[len + MFD_NAME_PREFIX_LEN - 1]) { | |
3723 | error = -EFAULT; | |
3724 | goto err_name; | |
3725 | } | |
3726 | ||
3727 | fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0); | |
3728 | if (fd < 0) { | |
3729 | error = fd; | |
3730 | goto err_name; | |
3731 | } | |
3732 | ||
749df87b MK |
3733 | if (flags & MFD_HUGETLB) { |
3734 | struct user_struct *user = NULL; | |
3735 | ||
3736 | file = hugetlb_file_setup(name, 0, VM_NORESERVE, &user, | |
3737 | HUGETLB_ANONHUGE_INODE, | |
3738 | (flags >> MFD_HUGE_SHIFT) & | |
3739 | MFD_HUGE_MASK); | |
3740 | } else | |
3741 | file = shmem_file_setup(name, 0, VM_NORESERVE); | |
9183df25 DH |
3742 | if (IS_ERR(file)) { |
3743 | error = PTR_ERR(file); | |
3744 | goto err_fd; | |
3745 | } | |
9183df25 DH |
3746 | file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; |
3747 | file->f_flags |= O_RDWR | O_LARGEFILE; | |
749df87b MK |
3748 | |
3749 | if (flags & MFD_ALLOW_SEALING) { | |
47b9012e MAL |
3750 | file_seals = memfd_file_seals_ptr(file); |
3751 | *file_seals &= ~F_SEAL_SEAL; | |
749df87b | 3752 | } |
9183df25 DH |
3753 | |
3754 | fd_install(fd, file); | |
3755 | kfree(name); | |
3756 | return fd; | |
3757 | ||
3758 | err_fd: | |
3759 | put_unused_fd(fd); | |
3760 | err_name: | |
3761 | kfree(name); | |
3762 | return error; | |
3763 | } | |
3764 | ||
680d794b | 3765 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3766 | |
3767 | static void shmem_put_super(struct super_block *sb) | |
3768 | { | |
602586a8 HD |
3769 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3770 | ||
3771 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 3772 | mpol_put(sbinfo->mpol); |
602586a8 | 3773 | kfree(sbinfo); |
1da177e4 LT |
3774 | sb->s_fs_info = NULL; |
3775 | } | |
3776 | ||
2b2af54a | 3777 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3778 | { |
3779 | struct inode *inode; | |
0edd73b3 | 3780 | struct shmem_sb_info *sbinfo; |
680d794b | 3781 | int err = -ENOMEM; |
3782 | ||
3783 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3784 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 3785 | L1_CACHE_BYTES), GFP_KERNEL); |
3786 | if (!sbinfo) | |
3787 | return -ENOMEM; | |
3788 | ||
680d794b | 3789 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
3790 | sbinfo->uid = current_fsuid(); |
3791 | sbinfo->gid = current_fsgid(); | |
680d794b | 3792 | sb->s_fs_info = sbinfo; |
1da177e4 | 3793 | |
0edd73b3 | 3794 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3795 | /* |
3796 | * Per default we only allow half of the physical ram per | |
3797 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3798 | * but the internal instance is left unlimited. | |
3799 | */ | |
1751e8a6 | 3800 | if (!(sb->s_flags & SB_KERNMOUNT)) { |
680d794b | 3801 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3802 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3803 | if (shmem_parse_options(data, sbinfo, false)) { | |
3804 | err = -EINVAL; | |
3805 | goto failed; | |
3806 | } | |
ca4e0519 | 3807 | } else { |
1751e8a6 | 3808 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 3809 | } |
91828a40 | 3810 | sb->s_export_op = &shmem_export_ops; |
1751e8a6 | 3811 | sb->s_flags |= SB_NOSEC; |
1da177e4 | 3812 | #else |
1751e8a6 | 3813 | sb->s_flags |= SB_NOUSER; |
1da177e4 LT |
3814 | #endif |
3815 | ||
0edd73b3 | 3816 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3817 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3818 | goto failed; |
680d794b | 3819 | sbinfo->free_inodes = sbinfo->max_inodes; |
779750d2 KS |
3820 | spin_lock_init(&sbinfo->shrinklist_lock); |
3821 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 3822 | |
285b2c4f | 3823 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3824 | sb->s_blocksize = PAGE_SIZE; |
3825 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3826 | sb->s_magic = TMPFS_MAGIC; |
3827 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3828 | sb->s_time_gran = 1; |
b09e0fa4 | 3829 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3830 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3831 | #endif |
3832 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
1751e8a6 | 3833 | sb->s_flags |= SB_POSIXACL; |
39f0247d | 3834 | #endif |
2b4db796 | 3835 | uuid_gen(&sb->s_uuid); |
0edd73b3 | 3836 | |
454abafe | 3837 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3838 | if (!inode) |
3839 | goto failed; | |
680d794b | 3840 | inode->i_uid = sbinfo->uid; |
3841 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3842 | sb->s_root = d_make_root(inode); |
3843 | if (!sb->s_root) | |
48fde701 | 3844 | goto failed; |
1da177e4 LT |
3845 | return 0; |
3846 | ||
1da177e4 LT |
3847 | failed: |
3848 | shmem_put_super(sb); | |
3849 | return err; | |
3850 | } | |
3851 | ||
fcc234f8 | 3852 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3853 | |
3854 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3855 | { | |
41ffe5d5 HD |
3856 | struct shmem_inode_info *info; |
3857 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3858 | if (!info) | |
1da177e4 | 3859 | return NULL; |
41ffe5d5 | 3860 | return &info->vfs_inode; |
1da177e4 LT |
3861 | } |
3862 | ||
41ffe5d5 | 3863 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3864 | { |
3865 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84e710da AV |
3866 | if (S_ISLNK(inode->i_mode)) |
3867 | kfree(inode->i_link); | |
fa0d7e3d NP |
3868 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3869 | } | |
3870 | ||
1da177e4 LT |
3871 | static void shmem_destroy_inode(struct inode *inode) |
3872 | { | |
09208d15 | 3873 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3874 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3875 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3876 | } |
3877 | ||
41ffe5d5 | 3878 | static void shmem_init_inode(void *foo) |
1da177e4 | 3879 | { |
41ffe5d5 HD |
3880 | struct shmem_inode_info *info = foo; |
3881 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3882 | } |
3883 | ||
9a8ec03e | 3884 | static void shmem_init_inodecache(void) |
1da177e4 LT |
3885 | { |
3886 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3887 | sizeof(struct shmem_inode_info), | |
5d097056 | 3888 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3889 | } |
3890 | ||
41ffe5d5 | 3891 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3892 | { |
1a1d92c1 | 3893 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3894 | } |
3895 | ||
f5e54d6e | 3896 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3897 | .writepage = shmem_writepage, |
76719325 | 3898 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3899 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3900 | .write_begin = shmem_write_begin, |
3901 | .write_end = shmem_write_end, | |
1da177e4 | 3902 | #endif |
1c93923c | 3903 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3904 | .migratepage = migrate_page, |
1c93923c | 3905 | #endif |
aa261f54 | 3906 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3907 | }; |
3908 | ||
15ad7cdc | 3909 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3910 | .mmap = shmem_mmap, |
c01d5b30 | 3911 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3912 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3913 | .llseek = shmem_file_llseek, |
2ba5bbed | 3914 | .read_iter = shmem_file_read_iter, |
8174202b | 3915 | .write_iter = generic_file_write_iter, |
1b061d92 | 3916 | .fsync = noop_fsync, |
82c156f8 | 3917 | .splice_read = generic_file_splice_read, |
f6cb85d0 | 3918 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3919 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3920 | #endif |
3921 | }; | |
3922 | ||
92e1d5be | 3923 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3924 | .getattr = shmem_getattr, |
94c1e62d | 3925 | .setattr = shmem_setattr, |
b09e0fa4 | 3926 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3927 | .listxattr = shmem_listxattr, |
feda821e | 3928 | .set_acl = simple_set_acl, |
b09e0fa4 | 3929 | #endif |
1da177e4 LT |
3930 | }; |
3931 | ||
92e1d5be | 3932 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3933 | #ifdef CONFIG_TMPFS |
3934 | .create = shmem_create, | |
3935 | .lookup = simple_lookup, | |
3936 | .link = shmem_link, | |
3937 | .unlink = shmem_unlink, | |
3938 | .symlink = shmem_symlink, | |
3939 | .mkdir = shmem_mkdir, | |
3940 | .rmdir = shmem_rmdir, | |
3941 | .mknod = shmem_mknod, | |
2773bf00 | 3942 | .rename = shmem_rename2, |
60545d0d | 3943 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3944 | #endif |
b09e0fa4 | 3945 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3946 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3947 | #endif |
39f0247d | 3948 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3949 | .setattr = shmem_setattr, |
feda821e | 3950 | .set_acl = simple_set_acl, |
39f0247d AG |
3951 | #endif |
3952 | }; | |
3953 | ||
92e1d5be | 3954 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3955 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3956 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3957 | #endif |
39f0247d | 3958 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3959 | .setattr = shmem_setattr, |
feda821e | 3960 | .set_acl = simple_set_acl, |
39f0247d | 3961 | #endif |
1da177e4 LT |
3962 | }; |
3963 | ||
759b9775 | 3964 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
3965 | .alloc_inode = shmem_alloc_inode, |
3966 | .destroy_inode = shmem_destroy_inode, | |
3967 | #ifdef CONFIG_TMPFS | |
3968 | .statfs = shmem_statfs, | |
3969 | .remount_fs = shmem_remount_fs, | |
680d794b | 3970 | .show_options = shmem_show_options, |
1da177e4 | 3971 | #endif |
1f895f75 | 3972 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3973 | .drop_inode = generic_delete_inode, |
3974 | .put_super = shmem_put_super, | |
779750d2 KS |
3975 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
3976 | .nr_cached_objects = shmem_unused_huge_count, | |
3977 | .free_cached_objects = shmem_unused_huge_scan, | |
3978 | #endif | |
1da177e4 LT |
3979 | }; |
3980 | ||
f0f37e2f | 3981 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3982 | .fault = shmem_fault, |
d7c17551 | 3983 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3984 | #ifdef CONFIG_NUMA |
3985 | .set_policy = shmem_set_policy, | |
3986 | .get_policy = shmem_get_policy, | |
3987 | #endif | |
3988 | }; | |
3989 | ||
3c26ff6e AV |
3990 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
3991 | int flags, const char *dev_name, void *data) | |
1da177e4 | 3992 | { |
3c26ff6e | 3993 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
3994 | } |
3995 | ||
41ffe5d5 | 3996 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3997 | .owner = THIS_MODULE, |
3998 | .name = "tmpfs", | |
3c26ff6e | 3999 | .mount = shmem_mount, |
1da177e4 | 4000 | .kill_sb = kill_litter_super, |
2b8576cb | 4001 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 4002 | }; |
1da177e4 | 4003 | |
41ffe5d5 | 4004 | int __init shmem_init(void) |
1da177e4 LT |
4005 | { |
4006 | int error; | |
4007 | ||
16203a7a RL |
4008 | /* If rootfs called this, don't re-init */ |
4009 | if (shmem_inode_cachep) | |
4010 | return 0; | |
4011 | ||
9a8ec03e | 4012 | shmem_init_inodecache(); |
1da177e4 | 4013 | |
41ffe5d5 | 4014 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 4015 | if (error) { |
1170532b | 4016 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
4017 | goto out2; |
4018 | } | |
95dc112a | 4019 | |
ca4e0519 | 4020 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
4021 | if (IS_ERR(shm_mnt)) { |
4022 | error = PTR_ERR(shm_mnt); | |
1170532b | 4023 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
4024 | goto out1; |
4025 | } | |
5a6e75f8 | 4026 | |
e496cf3d | 4027 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
435c0b87 | 4028 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4029 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4030 | else | |
4031 | shmem_huge = 0; /* just in case it was patched */ | |
4032 | #endif | |
1da177e4 LT |
4033 | return 0; |
4034 | ||
4035 | out1: | |
41ffe5d5 | 4036 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 4037 | out2: |
41ffe5d5 | 4038 | shmem_destroy_inodecache(); |
1da177e4 LT |
4039 | shm_mnt = ERR_PTR(error); |
4040 | return error; | |
4041 | } | |
853ac43a | 4042 | |
e496cf3d | 4043 | #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS) |
5a6e75f8 KS |
4044 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
4045 | struct kobj_attribute *attr, char *buf) | |
4046 | { | |
4047 | int values[] = { | |
4048 | SHMEM_HUGE_ALWAYS, | |
4049 | SHMEM_HUGE_WITHIN_SIZE, | |
4050 | SHMEM_HUGE_ADVISE, | |
4051 | SHMEM_HUGE_NEVER, | |
4052 | SHMEM_HUGE_DENY, | |
4053 | SHMEM_HUGE_FORCE, | |
4054 | }; | |
4055 | int i, count; | |
4056 | ||
4057 | for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) { | |
4058 | const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s "; | |
4059 | ||
4060 | count += sprintf(buf + count, fmt, | |
4061 | shmem_format_huge(values[i])); | |
4062 | } | |
4063 | buf[count - 1] = '\n'; | |
4064 | return count; | |
4065 | } | |
4066 | ||
4067 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
4068 | struct kobj_attribute *attr, const char *buf, size_t count) | |
4069 | { | |
4070 | char tmp[16]; | |
4071 | int huge; | |
4072 | ||
4073 | if (count + 1 > sizeof(tmp)) | |
4074 | return -EINVAL; | |
4075 | memcpy(tmp, buf, count); | |
4076 | tmp[count] = '\0'; | |
4077 | if (count && tmp[count - 1] == '\n') | |
4078 | tmp[count - 1] = '\0'; | |
4079 | ||
4080 | huge = shmem_parse_huge(tmp); | |
4081 | if (huge == -EINVAL) | |
4082 | return -EINVAL; | |
4083 | if (!has_transparent_hugepage() && | |
4084 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
4085 | return -EINVAL; | |
4086 | ||
4087 | shmem_huge = huge; | |
435c0b87 | 4088 | if (shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4089 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4090 | return count; | |
4091 | } | |
4092 | ||
4093 | struct kobj_attribute shmem_enabled_attr = | |
4094 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
3b33719c | 4095 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */ |
f3f0e1d2 | 4096 | |
3b33719c | 4097 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
f3f0e1d2 KS |
4098 | bool shmem_huge_enabled(struct vm_area_struct *vma) |
4099 | { | |
4100 | struct inode *inode = file_inode(vma->vm_file); | |
4101 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4102 | loff_t i_size; | |
4103 | pgoff_t off; | |
4104 | ||
4105 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
4106 | return true; | |
4107 | if (shmem_huge == SHMEM_HUGE_DENY) | |
4108 | return false; | |
4109 | switch (sbinfo->huge) { | |
4110 | case SHMEM_HUGE_NEVER: | |
4111 | return false; | |
4112 | case SHMEM_HUGE_ALWAYS: | |
4113 | return true; | |
4114 | case SHMEM_HUGE_WITHIN_SIZE: | |
4115 | off = round_up(vma->vm_pgoff, HPAGE_PMD_NR); | |
4116 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
4117 | if (i_size >= HPAGE_PMD_SIZE && | |
4118 | i_size >> PAGE_SHIFT >= off) | |
4119 | return true; | |
c8402871 | 4120 | /* fall through */ |
f3f0e1d2 KS |
4121 | case SHMEM_HUGE_ADVISE: |
4122 | /* TODO: implement fadvise() hints */ | |
4123 | return (vma->vm_flags & VM_HUGEPAGE); | |
4124 | default: | |
4125 | VM_BUG_ON(1); | |
4126 | return false; | |
4127 | } | |
4128 | } | |
3b33719c | 4129 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 4130 | |
853ac43a MM |
4131 | #else /* !CONFIG_SHMEM */ |
4132 | ||
4133 | /* | |
4134 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
4135 | * | |
4136 | * This is intended for small system where the benefits of the full | |
4137 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4138 | * their complexity. On systems without swap this code should be | |
4139 | * effectively equivalent, but much lighter weight. | |
4140 | */ | |
4141 | ||
41ffe5d5 | 4142 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4143 | .name = "tmpfs", |
3c26ff6e | 4144 | .mount = ramfs_mount, |
853ac43a | 4145 | .kill_sb = kill_litter_super, |
2b8576cb | 4146 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4147 | }; |
4148 | ||
41ffe5d5 | 4149 | int __init shmem_init(void) |
853ac43a | 4150 | { |
41ffe5d5 | 4151 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4152 | |
41ffe5d5 | 4153 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
4154 | BUG_ON(IS_ERR(shm_mnt)); |
4155 | ||
4156 | return 0; | |
4157 | } | |
4158 | ||
41ffe5d5 | 4159 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
4160 | { |
4161 | return 0; | |
4162 | } | |
4163 | ||
3f96b79a HD |
4164 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
4165 | { | |
4166 | return 0; | |
4167 | } | |
4168 | ||
24513264 HD |
4169 | void shmem_unlock_mapping(struct address_space *mapping) |
4170 | { | |
4171 | } | |
4172 | ||
c01d5b30 HD |
4173 | #ifdef CONFIG_MMU |
4174 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4175 | unsigned long addr, unsigned long len, | |
4176 | unsigned long pgoff, unsigned long flags) | |
4177 | { | |
4178 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4179 | } | |
4180 | #endif | |
4181 | ||
41ffe5d5 | 4182 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4183 | { |
41ffe5d5 | 4184 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4185 | } |
4186 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4187 | ||
0b0a0806 HD |
4188 | #define shmem_vm_ops generic_file_vm_ops |
4189 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 4190 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
4191 | #define shmem_acct_size(flags, size) 0 |
4192 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
4193 | |
4194 | #endif /* CONFIG_SHMEM */ | |
4195 | ||
4196 | /* common code */ | |
1da177e4 | 4197 | |
19938e35 | 4198 | static const struct dentry_operations anon_ops = { |
118b2302 | 4199 | .d_dname = simple_dname |
3451538a AV |
4200 | }; |
4201 | ||
703321b6 | 4202 | static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size, |
c7277090 | 4203 | unsigned long flags, unsigned int i_flags) |
1da177e4 | 4204 | { |
6b4d0b27 | 4205 | struct file *res; |
1da177e4 | 4206 | struct inode *inode; |
2c48b9c4 | 4207 | struct path path; |
3451538a | 4208 | struct super_block *sb; |
1da177e4 LT |
4209 | struct qstr this; |
4210 | ||
703321b6 MA |
4211 | if (IS_ERR(mnt)) |
4212 | return ERR_CAST(mnt); | |
1da177e4 | 4213 | |
285b2c4f | 4214 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4215 | return ERR_PTR(-EINVAL); |
4216 | ||
4217 | if (shmem_acct_size(flags, size)) | |
4218 | return ERR_PTR(-ENOMEM); | |
4219 | ||
6b4d0b27 | 4220 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
4221 | this.name = name; |
4222 | this.len = strlen(name); | |
4223 | this.hash = 0; /* will go */ | |
703321b6 MA |
4224 | sb = mnt->mnt_sb; |
4225 | path.mnt = mntget(mnt); | |
3451538a | 4226 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 4227 | if (!path.dentry) |
1da177e4 | 4228 | goto put_memory; |
3451538a | 4229 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 4230 | |
6b4d0b27 | 4231 | res = ERR_PTR(-ENOSPC); |
3451538a | 4232 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 4233 | if (!inode) |
66ee4b88 | 4234 | goto put_memory; |
1da177e4 | 4235 | |
c7277090 | 4236 | inode->i_flags |= i_flags; |
2c48b9c4 | 4237 | d_instantiate(path.dentry, inode); |
1da177e4 | 4238 | inode->i_size = size; |
6d6b77f1 | 4239 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
4240 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
4241 | if (IS_ERR(res)) | |
66ee4b88 | 4242 | goto put_path; |
4b42af81 | 4243 | |
6b4d0b27 | 4244 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 4245 | &shmem_file_operations); |
6b4d0b27 | 4246 | if (IS_ERR(res)) |
66ee4b88 | 4247 | goto put_path; |
4b42af81 | 4248 | |
6b4d0b27 | 4249 | return res; |
1da177e4 | 4250 | |
1da177e4 LT |
4251 | put_memory: |
4252 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
4253 | put_path: |
4254 | path_put(&path); | |
6b4d0b27 | 4255 | return res; |
1da177e4 | 4256 | } |
c7277090 EP |
4257 | |
4258 | /** | |
4259 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4260 | * kernel internal. There will be NO LSM permission checks against the | |
4261 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4262 | * higher layer. The users are the big_key and shm implementations. LSM |
4263 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4264 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4265 | * @size: size to be set for the file | |
4266 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4267 | */ | |
4268 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4269 | { | |
703321b6 | 4270 | return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE); |
c7277090 EP |
4271 | } |
4272 | ||
4273 | /** | |
4274 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4275 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4276 | * @size: size to be set for the file | |
4277 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4278 | */ | |
4279 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4280 | { | |
703321b6 | 4281 | return __shmem_file_setup(shm_mnt, name, size, flags, 0); |
c7277090 | 4282 | } |
395e0ddc | 4283 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4284 | |
703321b6 MA |
4285 | /** |
4286 | * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs | |
4287 | * @mnt: the tmpfs mount where the file will be created | |
4288 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4289 | * @size: size to be set for the file | |
4290 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4291 | */ | |
4292 | struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name, | |
4293 | loff_t size, unsigned long flags) | |
4294 | { | |
4295 | return __shmem_file_setup(mnt, name, size, flags, 0); | |
4296 | } | |
4297 | EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt); | |
4298 | ||
46711810 | 4299 | /** |
1da177e4 | 4300 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
4301 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
4302 | */ | |
4303 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4304 | { | |
4305 | struct file *file; | |
4306 | loff_t size = vma->vm_end - vma->vm_start; | |
4307 | ||
66fc1303 HD |
4308 | /* |
4309 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
4310 | * between XFS directory reading and selinux: since this file is only | |
4311 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4312 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4313 | */ | |
703321b6 | 4314 | file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags); |
1da177e4 LT |
4315 | if (IS_ERR(file)) |
4316 | return PTR_ERR(file); | |
4317 | ||
4318 | if (vma->vm_file) | |
4319 | fput(vma->vm_file); | |
4320 | vma->vm_file = file; | |
4321 | vma->vm_ops = &shmem_vm_ops; | |
f3f0e1d2 | 4322 | |
e496cf3d | 4323 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
4324 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
4325 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
4326 | khugepaged_enter(vma, vma->vm_flags); | |
4327 | } | |
4328 | ||
1da177e4 LT |
4329 | return 0; |
4330 | } | |
d9d90e5e HD |
4331 | |
4332 | /** | |
4333 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
4334 | * @mapping: the page's address_space | |
4335 | * @index: the page index | |
4336 | * @gfp: the page allocator flags to use if allocating | |
4337 | * | |
4338 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4339 | * with any new page allocations done using the specified allocation flags. | |
4340 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
4341 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
4342 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4343 | * | |
68da9f05 HD |
4344 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4345 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4346 | */ |
4347 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4348 | pgoff_t index, gfp_t gfp) | |
4349 | { | |
68da9f05 HD |
4350 | #ifdef CONFIG_SHMEM |
4351 | struct inode *inode = mapping->host; | |
9276aad6 | 4352 | struct page *page; |
68da9f05 HD |
4353 | int error; |
4354 | ||
4355 | BUG_ON(mapping->a_ops != &shmem_aops); | |
9e18eb29 | 4356 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
cfda0526 | 4357 | gfp, NULL, NULL, NULL); |
68da9f05 HD |
4358 | if (error) |
4359 | page = ERR_PTR(error); | |
4360 | else | |
4361 | unlock_page(page); | |
4362 | return page; | |
4363 | #else | |
4364 | /* | |
4365 | * The tiny !SHMEM case uses ramfs without swap | |
4366 | */ | |
d9d90e5e | 4367 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4368 | #endif |
d9d90e5e HD |
4369 | } |
4370 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |