]>
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 AM |
109 | static unsigned long shmem_default_max_blocks(void) |
110 | { | |
111 | return totalram_pages / 2; | |
112 | } | |
113 | ||
0006ebb4 | 114 | static int shmem_default_max_inodes(void) |
680d794b | 115 | { |
0006ebb4 SF |
116 | unsigned long ul; |
117 | ||
118 | ul = INT_MAX; | |
119 | ul = min3(ul, totalram_pages - totalhigh_pages, totalram_pages / 2); | |
120 | return ul; | |
680d794b | 121 | } |
b76db735 | 122 | #endif |
680d794b | 123 | |
bde05d1c HD |
124 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
125 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
126 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 | 127 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 128 | struct page **pagep, enum sgp_type sgp, |
cfda0526 MR |
129 | gfp_t gfp, struct vm_area_struct *vma, |
130 | struct vm_fault *vmf, int *fault_type); | |
68da9f05 | 131 | |
f3f0e1d2 | 132 | int shmem_getpage(struct inode *inode, pgoff_t index, |
9e18eb29 | 133 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
134 | { |
135 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
cfda0526 | 136 | mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL); |
68da9f05 | 137 | } |
1da177e4 | 138 | |
1da177e4 LT |
139 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
140 | { | |
141 | return sb->s_fs_info; | |
142 | } | |
143 | ||
144 | /* | |
145 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
146 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
147 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
148 | * consistent with the pre-accounting of private mappings ... | |
149 | */ | |
150 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
151 | { | |
0b0a0806 | 152 | return (flags & VM_NORESERVE) ? |
191c5424 | 153 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
154 | } |
155 | ||
156 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
157 | { | |
0b0a0806 | 158 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
159 | vm_unacct_memory(VM_ACCT(size)); |
160 | } | |
161 | ||
77142517 KK |
162 | static inline int shmem_reacct_size(unsigned long flags, |
163 | loff_t oldsize, loff_t newsize) | |
164 | { | |
165 | if (!(flags & VM_NORESERVE)) { | |
166 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
167 | return security_vm_enough_memory_mm(current->mm, | |
168 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
169 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
170 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
171 | } | |
172 | return 0; | |
173 | } | |
174 | ||
1da177e4 LT |
175 | /* |
176 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 177 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
178 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
179 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
180 | */ | |
800d8c63 | 181 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 182 | { |
800d8c63 KS |
183 | if (!(flags & VM_NORESERVE)) |
184 | return 0; | |
185 | ||
186 | return security_vm_enough_memory_mm(current->mm, | |
187 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
188 | } |
189 | ||
190 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
191 | { | |
0b0a0806 | 192 | if (flags & VM_NORESERVE) |
09cbfeaf | 193 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
194 | } |
195 | ||
0f079694 MR |
196 | static inline bool shmem_inode_acct_block(struct inode *inode, long pages) |
197 | { | |
198 | struct shmem_inode_info *info = SHMEM_I(inode); | |
199 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
200 | ||
201 | if (shmem_acct_block(info->flags, pages)) | |
202 | return false; | |
203 | ||
204 | if (sbinfo->max_blocks) { | |
205 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
206 | sbinfo->max_blocks - pages) > 0) | |
207 | goto unacct; | |
208 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
209 | } | |
210 | ||
211 | return true; | |
212 | ||
213 | unacct: | |
214 | shmem_unacct_blocks(info->flags, pages); | |
215 | return false; | |
216 | } | |
217 | ||
218 | static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages) | |
219 | { | |
220 | struct shmem_inode_info *info = SHMEM_I(inode); | |
221 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
222 | ||
223 | if (sbinfo->max_blocks) | |
224 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
225 | shmem_unacct_blocks(info->flags, pages); | |
226 | } | |
227 | ||
759b9775 | 228 | static const struct super_operations shmem_ops; |
f5e54d6e | 229 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 230 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
231 | static const struct inode_operations shmem_inode_operations; |
232 | static const struct inode_operations shmem_dir_inode_operations; | |
233 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 234 | static const struct vm_operations_struct shmem_vm_ops; |
779750d2 | 235 | static struct file_system_type shmem_fs_type; |
1da177e4 | 236 | |
b0506e48 MR |
237 | bool vma_is_shmem(struct vm_area_struct *vma) |
238 | { | |
239 | return vma->vm_ops == &shmem_vm_ops; | |
240 | } | |
241 | ||
1da177e4 | 242 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 243 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 244 | |
5b04c689 PE |
245 | static int shmem_reserve_inode(struct super_block *sb) |
246 | { | |
247 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
248 | if (sbinfo->max_inodes) { | |
249 | spin_lock(&sbinfo->stat_lock); | |
250 | if (!sbinfo->free_inodes) { | |
251 | spin_unlock(&sbinfo->stat_lock); | |
252 | return -ENOSPC; | |
253 | } | |
254 | sbinfo->free_inodes--; | |
255 | spin_unlock(&sbinfo->stat_lock); | |
256 | } | |
257 | return 0; | |
258 | } | |
259 | ||
260 | static void shmem_free_inode(struct super_block *sb) | |
261 | { | |
262 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
263 | if (sbinfo->max_inodes) { | |
264 | spin_lock(&sbinfo->stat_lock); | |
265 | sbinfo->free_inodes++; | |
266 | spin_unlock(&sbinfo->stat_lock); | |
267 | } | |
268 | } | |
269 | ||
46711810 | 270 | /** |
41ffe5d5 | 271 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
272 | * @inode: inode to recalc |
273 | * | |
274 | * We have to calculate the free blocks since the mm can drop | |
275 | * undirtied hole pages behind our back. | |
276 | * | |
277 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
278 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
279 | * | |
280 | * It has to be called with the spinlock held. | |
281 | */ | |
282 | static void shmem_recalc_inode(struct inode *inode) | |
283 | { | |
284 | struct shmem_inode_info *info = SHMEM_I(inode); | |
285 | long freed; | |
286 | ||
287 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
288 | if (freed > 0) { | |
289 | info->alloced -= freed; | |
54af6042 | 290 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
0f079694 | 291 | shmem_inode_unacct_blocks(inode, freed); |
1da177e4 LT |
292 | } |
293 | } | |
294 | ||
800d8c63 KS |
295 | bool shmem_charge(struct inode *inode, long pages) |
296 | { | |
297 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 298 | unsigned long flags; |
800d8c63 | 299 | |
0f079694 | 300 | if (!shmem_inode_acct_block(inode, pages)) |
800d8c63 | 301 | return false; |
b1cc94ab | 302 | |
4595ef88 | 303 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
304 | info->alloced += pages; |
305 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
306 | shmem_recalc_inode(inode); | |
4595ef88 | 307 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
308 | inode->i_mapping->nrpages += pages; |
309 | ||
800d8c63 KS |
310 | return true; |
311 | } | |
312 | ||
313 | void shmem_uncharge(struct inode *inode, long pages) | |
314 | { | |
315 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 316 | unsigned long flags; |
800d8c63 | 317 | |
4595ef88 | 318 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
319 | info->alloced -= pages; |
320 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
321 | shmem_recalc_inode(inode); | |
4595ef88 | 322 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 | 323 | |
0f079694 | 324 | shmem_inode_unacct_blocks(inode, pages); |
800d8c63 KS |
325 | } |
326 | ||
7a5d0fbb HD |
327 | /* |
328 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
329 | */ | |
330 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
331 | pgoff_t index, void *expected, void *replacement) | |
332 | { | |
f7942430 | 333 | struct radix_tree_node *node; |
7a5d0fbb | 334 | void **pslot; |
6dbaf22c | 335 | void *item; |
7a5d0fbb HD |
336 | |
337 | VM_BUG_ON(!expected); | |
6dbaf22c | 338 | VM_BUG_ON(!replacement); |
f7942430 JW |
339 | item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot); |
340 | if (!item) | |
6dbaf22c | 341 | return -ENOENT; |
7a5d0fbb HD |
342 | if (item != expected) |
343 | return -ENOENT; | |
4d693d08 | 344 | __radix_tree_replace(&mapping->page_tree, node, pslot, |
c7df8ad2 | 345 | replacement, NULL); |
7a5d0fbb HD |
346 | return 0; |
347 | } | |
348 | ||
d1899228 HD |
349 | /* |
350 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
351 | * that an entry was not already brought back from swap by a racing thread. | |
352 | * | |
353 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
354 | * might be reused, and again be SwapCache, using the same swap as before. | |
355 | */ | |
356 | static bool shmem_confirm_swap(struct address_space *mapping, | |
357 | pgoff_t index, swp_entry_t swap) | |
358 | { | |
359 | void *item; | |
360 | ||
361 | rcu_read_lock(); | |
362 | item = radix_tree_lookup(&mapping->page_tree, index); | |
363 | rcu_read_unlock(); | |
364 | return item == swp_to_radix_entry(swap); | |
365 | } | |
366 | ||
5a6e75f8 KS |
367 | /* |
368 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
369 | * | |
370 | * SHMEM_HUGE_NEVER: | |
371 | * disables huge pages for the mount; | |
372 | * SHMEM_HUGE_ALWAYS: | |
373 | * enables huge pages for the mount; | |
374 | * SHMEM_HUGE_WITHIN_SIZE: | |
375 | * only allocate huge pages if the page will be fully within i_size, | |
376 | * also respect fadvise()/madvise() hints; | |
377 | * SHMEM_HUGE_ADVISE: | |
378 | * only allocate huge pages if requested with fadvise()/madvise(); | |
379 | */ | |
380 | ||
381 | #define SHMEM_HUGE_NEVER 0 | |
382 | #define SHMEM_HUGE_ALWAYS 1 | |
383 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
384 | #define SHMEM_HUGE_ADVISE 3 | |
385 | ||
386 | /* | |
387 | * Special values. | |
388 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
389 | * | |
390 | * SHMEM_HUGE_DENY: | |
391 | * disables huge on shm_mnt and all mounts, for emergency use; | |
392 | * SHMEM_HUGE_FORCE: | |
393 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
394 | * | |
395 | */ | |
396 | #define SHMEM_HUGE_DENY (-1) | |
397 | #define SHMEM_HUGE_FORCE (-2) | |
398 | ||
e496cf3d | 399 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
400 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
401 | ||
402 | int shmem_huge __read_mostly; | |
403 | ||
f1f5929c | 404 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
405 | static int shmem_parse_huge(const char *str) |
406 | { | |
407 | if (!strcmp(str, "never")) | |
408 | return SHMEM_HUGE_NEVER; | |
409 | if (!strcmp(str, "always")) | |
410 | return SHMEM_HUGE_ALWAYS; | |
411 | if (!strcmp(str, "within_size")) | |
412 | return SHMEM_HUGE_WITHIN_SIZE; | |
413 | if (!strcmp(str, "advise")) | |
414 | return SHMEM_HUGE_ADVISE; | |
415 | if (!strcmp(str, "deny")) | |
416 | return SHMEM_HUGE_DENY; | |
417 | if (!strcmp(str, "force")) | |
418 | return SHMEM_HUGE_FORCE; | |
419 | return -EINVAL; | |
420 | } | |
421 | ||
422 | static const char *shmem_format_huge(int huge) | |
423 | { | |
424 | switch (huge) { | |
425 | case SHMEM_HUGE_NEVER: | |
426 | return "never"; | |
427 | case SHMEM_HUGE_ALWAYS: | |
428 | return "always"; | |
429 | case SHMEM_HUGE_WITHIN_SIZE: | |
430 | return "within_size"; | |
431 | case SHMEM_HUGE_ADVISE: | |
432 | return "advise"; | |
433 | case SHMEM_HUGE_DENY: | |
434 | return "deny"; | |
435 | case SHMEM_HUGE_FORCE: | |
436 | return "force"; | |
437 | default: | |
438 | VM_BUG_ON(1); | |
439 | return "bad_val"; | |
440 | } | |
441 | } | |
f1f5929c | 442 | #endif |
5a6e75f8 | 443 | |
779750d2 KS |
444 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
445 | struct shrink_control *sc, unsigned long nr_to_split) | |
446 | { | |
447 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 448 | LIST_HEAD(to_remove); |
779750d2 KS |
449 | struct inode *inode; |
450 | struct shmem_inode_info *info; | |
451 | struct page *page; | |
452 | unsigned long batch = sc ? sc->nr_to_scan : 128; | |
453 | int removed = 0, split = 0; | |
454 | ||
455 | if (list_empty(&sbinfo->shrinklist)) | |
456 | return SHRINK_STOP; | |
457 | ||
458 | spin_lock(&sbinfo->shrinklist_lock); | |
459 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
460 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
461 | ||
462 | /* pin the inode */ | |
463 | inode = igrab(&info->vfs_inode); | |
464 | ||
465 | /* inode is about to be evicted */ | |
466 | if (!inode) { | |
467 | list_del_init(&info->shrinklist); | |
468 | removed++; | |
469 | goto next; | |
470 | } | |
471 | ||
472 | /* Check if there's anything to gain */ | |
473 | if (round_up(inode->i_size, PAGE_SIZE) == | |
474 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 475 | list_move(&info->shrinklist, &to_remove); |
779750d2 | 476 | removed++; |
779750d2 KS |
477 | goto next; |
478 | } | |
479 | ||
480 | list_move(&info->shrinklist, &list); | |
481 | next: | |
482 | if (!--batch) | |
483 | break; | |
484 | } | |
485 | spin_unlock(&sbinfo->shrinklist_lock); | |
486 | ||
253fd0f0 KS |
487 | list_for_each_safe(pos, next, &to_remove) { |
488 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
489 | inode = &info->vfs_inode; | |
490 | list_del_init(&info->shrinklist); | |
491 | iput(inode); | |
492 | } | |
493 | ||
779750d2 KS |
494 | list_for_each_safe(pos, next, &list) { |
495 | int ret; | |
496 | ||
497 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
498 | inode = &info->vfs_inode; | |
499 | ||
500 | if (nr_to_split && split >= nr_to_split) { | |
501 | iput(inode); | |
502 | continue; | |
503 | } | |
504 | ||
505 | page = find_lock_page(inode->i_mapping, | |
506 | (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT); | |
507 | if (!page) | |
508 | goto drop; | |
509 | ||
510 | if (!PageTransHuge(page)) { | |
511 | unlock_page(page); | |
512 | put_page(page); | |
513 | goto drop; | |
514 | } | |
515 | ||
516 | ret = split_huge_page(page); | |
517 | unlock_page(page); | |
518 | put_page(page); | |
519 | ||
520 | if (ret) { | |
521 | /* split failed: leave it on the list */ | |
522 | iput(inode); | |
523 | continue; | |
524 | } | |
525 | ||
526 | split++; | |
527 | drop: | |
528 | list_del_init(&info->shrinklist); | |
529 | removed++; | |
530 | iput(inode); | |
531 | } | |
532 | ||
533 | spin_lock(&sbinfo->shrinklist_lock); | |
534 | list_splice_tail(&list, &sbinfo->shrinklist); | |
535 | sbinfo->shrinklist_len -= removed; | |
536 | spin_unlock(&sbinfo->shrinklist_lock); | |
537 | ||
538 | return split; | |
539 | } | |
540 | ||
541 | static long shmem_unused_huge_scan(struct super_block *sb, | |
542 | struct shrink_control *sc) | |
543 | { | |
544 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
545 | ||
546 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
547 | return SHRINK_STOP; | |
548 | ||
549 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
550 | } | |
551 | ||
552 | static long shmem_unused_huge_count(struct super_block *sb, | |
553 | struct shrink_control *sc) | |
554 | { | |
555 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
556 | return READ_ONCE(sbinfo->shrinklist_len); | |
557 | } | |
e496cf3d | 558 | #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 KS |
559 | |
560 | #define shmem_huge SHMEM_HUGE_DENY | |
561 | ||
779750d2 KS |
562 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
563 | struct shrink_control *sc, unsigned long nr_to_split) | |
564 | { | |
565 | return 0; | |
566 | } | |
e496cf3d | 567 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 568 | |
46f65ec1 HD |
569 | /* |
570 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
571 | */ | |
572 | static int shmem_add_to_page_cache(struct page *page, | |
573 | struct address_space *mapping, | |
fed400a1 | 574 | pgoff_t index, void *expected) |
46f65ec1 | 575 | { |
800d8c63 | 576 | int error, nr = hpage_nr_pages(page); |
46f65ec1 | 577 | |
800d8c63 KS |
578 | VM_BUG_ON_PAGE(PageTail(page), page); |
579 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
580 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
581 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 582 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 583 | |
800d8c63 | 584 | page_ref_add(page, nr); |
b065b432 HD |
585 | page->mapping = mapping; |
586 | page->index = index; | |
587 | ||
588 | spin_lock_irq(&mapping->tree_lock); | |
800d8c63 KS |
589 | if (PageTransHuge(page)) { |
590 | void __rcu **results; | |
591 | pgoff_t idx; | |
592 | int i; | |
593 | ||
594 | error = 0; | |
595 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, | |
596 | &results, &idx, index, 1) && | |
597 | idx < index + HPAGE_PMD_NR) { | |
598 | error = -EEXIST; | |
599 | } | |
600 | ||
601 | if (!error) { | |
602 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
603 | error = radix_tree_insert(&mapping->page_tree, | |
604 | index + i, page + i); | |
605 | VM_BUG_ON(error); | |
606 | } | |
607 | count_vm_event(THP_FILE_ALLOC); | |
608 | } | |
609 | } else if (!expected) { | |
b065b432 | 610 | error = radix_tree_insert(&mapping->page_tree, index, page); |
800d8c63 | 611 | } else { |
b065b432 HD |
612 | error = shmem_radix_tree_replace(mapping, index, expected, |
613 | page); | |
800d8c63 KS |
614 | } |
615 | ||
46f65ec1 | 616 | if (!error) { |
800d8c63 KS |
617 | mapping->nrpages += nr; |
618 | if (PageTransHuge(page)) | |
11fb9989 MG |
619 | __inc_node_page_state(page, NR_SHMEM_THPS); |
620 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); | |
621 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr); | |
b065b432 HD |
622 | spin_unlock_irq(&mapping->tree_lock); |
623 | } else { | |
624 | page->mapping = NULL; | |
625 | spin_unlock_irq(&mapping->tree_lock); | |
800d8c63 | 626 | page_ref_sub(page, nr); |
46f65ec1 | 627 | } |
46f65ec1 HD |
628 | return error; |
629 | } | |
630 | ||
6922c0c7 HD |
631 | /* |
632 | * Like delete_from_page_cache, but substitutes swap for page. | |
633 | */ | |
634 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
635 | { | |
636 | struct address_space *mapping = page->mapping; | |
637 | int error; | |
638 | ||
800d8c63 KS |
639 | VM_BUG_ON_PAGE(PageCompound(page), page); |
640 | ||
6922c0c7 HD |
641 | spin_lock_irq(&mapping->tree_lock); |
642 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
643 | page->mapping = NULL; | |
644 | mapping->nrpages--; | |
11fb9989 MG |
645 | __dec_node_page_state(page, NR_FILE_PAGES); |
646 | __dec_node_page_state(page, NR_SHMEM); | |
6922c0c7 | 647 | spin_unlock_irq(&mapping->tree_lock); |
09cbfeaf | 648 | put_page(page); |
6922c0c7 HD |
649 | BUG_ON(error); |
650 | } | |
651 | ||
7a5d0fbb HD |
652 | /* |
653 | * Remove swap entry from radix tree, free the swap and its page cache. | |
654 | */ | |
655 | static int shmem_free_swap(struct address_space *mapping, | |
656 | pgoff_t index, void *radswap) | |
657 | { | |
6dbaf22c | 658 | void *old; |
7a5d0fbb HD |
659 | |
660 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 661 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 662 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
663 | if (old != radswap) |
664 | return -ENOENT; | |
665 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
666 | return 0; | |
7a5d0fbb HD |
667 | } |
668 | ||
6a15a370 VB |
669 | /* |
670 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 671 | * given offsets are swapped out. |
6a15a370 VB |
672 | * |
673 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
674 | * as long as the inode doesn't go away and racy results are not a problem. | |
675 | */ | |
48131e03 VB |
676 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
677 | pgoff_t start, pgoff_t end) | |
6a15a370 | 678 | { |
6a15a370 VB |
679 | struct radix_tree_iter iter; |
680 | void **slot; | |
681 | struct page *page; | |
48131e03 | 682 | unsigned long swapped = 0; |
6a15a370 VB |
683 | |
684 | rcu_read_lock(); | |
685 | ||
6a15a370 VB |
686 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
687 | if (iter.index >= end) | |
688 | break; | |
689 | ||
690 | page = radix_tree_deref_slot(slot); | |
691 | ||
2cf938aa MW |
692 | if (radix_tree_deref_retry(page)) { |
693 | slot = radix_tree_iter_retry(&iter); | |
694 | continue; | |
695 | } | |
6a15a370 VB |
696 | |
697 | if (radix_tree_exceptional_entry(page)) | |
698 | swapped++; | |
699 | ||
700 | if (need_resched()) { | |
148deab2 | 701 | slot = radix_tree_iter_resume(slot, &iter); |
6a15a370 | 702 | cond_resched_rcu(); |
6a15a370 VB |
703 | } |
704 | } | |
705 | ||
706 | rcu_read_unlock(); | |
707 | ||
708 | return swapped << PAGE_SHIFT; | |
709 | } | |
710 | ||
48131e03 VB |
711 | /* |
712 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
713 | * given vma is swapped out. | |
714 | * | |
715 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
716 | * as long as the inode doesn't go away and racy results are not a problem. | |
717 | */ | |
718 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
719 | { | |
720 | struct inode *inode = file_inode(vma->vm_file); | |
721 | struct shmem_inode_info *info = SHMEM_I(inode); | |
722 | struct address_space *mapping = inode->i_mapping; | |
723 | unsigned long swapped; | |
724 | ||
725 | /* Be careful as we don't hold info->lock */ | |
726 | swapped = READ_ONCE(info->swapped); | |
727 | ||
728 | /* | |
729 | * The easier cases are when the shmem object has nothing in swap, or | |
730 | * the vma maps it whole. Then we can simply use the stats that we | |
731 | * already track. | |
732 | */ | |
733 | if (!swapped) | |
734 | return 0; | |
735 | ||
736 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
737 | return swapped << PAGE_SHIFT; | |
738 | ||
739 | /* Here comes the more involved part */ | |
740 | return shmem_partial_swap_usage(mapping, | |
741 | linear_page_index(vma, vma->vm_start), | |
742 | linear_page_index(vma, vma->vm_end)); | |
743 | } | |
744 | ||
24513264 HD |
745 | /* |
746 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
747 | */ | |
748 | void shmem_unlock_mapping(struct address_space *mapping) | |
749 | { | |
750 | struct pagevec pvec; | |
751 | pgoff_t indices[PAGEVEC_SIZE]; | |
752 | pgoff_t index = 0; | |
753 | ||
86679820 | 754 | pagevec_init(&pvec); |
24513264 HD |
755 | /* |
756 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
757 | */ | |
758 | while (!mapping_unevictable(mapping)) { | |
759 | /* | |
760 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
761 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
762 | */ | |
0cd6144a JW |
763 | pvec.nr = find_get_entries(mapping, index, |
764 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
765 | if (!pvec.nr) |
766 | break; | |
767 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 768 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
769 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
770 | pagevec_release(&pvec); | |
771 | cond_resched(); | |
772 | } | |
7a5d0fbb HD |
773 | } |
774 | ||
775 | /* | |
776 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 777 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 778 | */ |
1635f6a7 HD |
779 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
780 | bool unfalloc) | |
1da177e4 | 781 | { |
285b2c4f | 782 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 783 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
784 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
785 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
786 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
787 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 788 | struct pagevec pvec; |
7a5d0fbb HD |
789 | pgoff_t indices[PAGEVEC_SIZE]; |
790 | long nr_swaps_freed = 0; | |
285b2c4f | 791 | pgoff_t index; |
bda97eab HD |
792 | int i; |
793 | ||
83e4fa9c HD |
794 | if (lend == -1) |
795 | end = -1; /* unsigned, so actually very big */ | |
bda97eab | 796 | |
86679820 | 797 | pagevec_init(&pvec); |
bda97eab | 798 | index = start; |
83e4fa9c | 799 | while (index < end) { |
0cd6144a JW |
800 | pvec.nr = find_get_entries(mapping, index, |
801 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
802 | pvec.pages, indices); | |
7a5d0fbb HD |
803 | if (!pvec.nr) |
804 | break; | |
bda97eab HD |
805 | for (i = 0; i < pagevec_count(&pvec); i++) { |
806 | struct page *page = pvec.pages[i]; | |
807 | ||
7a5d0fbb | 808 | index = indices[i]; |
83e4fa9c | 809 | if (index >= end) |
bda97eab HD |
810 | break; |
811 | ||
7a5d0fbb | 812 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
813 | if (unfalloc) |
814 | continue; | |
7a5d0fbb HD |
815 | nr_swaps_freed += !shmem_free_swap(mapping, |
816 | index, page); | |
bda97eab | 817 | continue; |
7a5d0fbb HD |
818 | } |
819 | ||
800d8c63 KS |
820 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
821 | ||
7a5d0fbb | 822 | if (!trylock_page(page)) |
bda97eab | 823 | continue; |
800d8c63 KS |
824 | |
825 | if (PageTransTail(page)) { | |
826 | /* Middle of THP: zero out the page */ | |
827 | clear_highpage(page); | |
828 | unlock_page(page); | |
829 | continue; | |
830 | } else if (PageTransHuge(page)) { | |
831 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
832 | /* | |
833 | * Range ends in the middle of THP: | |
834 | * zero out the page | |
835 | */ | |
836 | clear_highpage(page); | |
837 | unlock_page(page); | |
838 | continue; | |
839 | } | |
840 | index += HPAGE_PMD_NR - 1; | |
841 | i += HPAGE_PMD_NR - 1; | |
842 | } | |
843 | ||
1635f6a7 | 844 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
845 | VM_BUG_ON_PAGE(PageTail(page), page); |
846 | if (page_mapping(page) == mapping) { | |
309381fe | 847 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
848 | truncate_inode_page(mapping, page); |
849 | } | |
bda97eab | 850 | } |
bda97eab HD |
851 | unlock_page(page); |
852 | } | |
0cd6144a | 853 | pagevec_remove_exceptionals(&pvec); |
24513264 | 854 | pagevec_release(&pvec); |
bda97eab HD |
855 | cond_resched(); |
856 | index++; | |
857 | } | |
1da177e4 | 858 | |
83e4fa9c | 859 | if (partial_start) { |
bda97eab | 860 | struct page *page = NULL; |
9e18eb29 | 861 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 862 | if (page) { |
09cbfeaf | 863 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
864 | if (start > end) { |
865 | top = partial_end; | |
866 | partial_end = 0; | |
867 | } | |
868 | zero_user_segment(page, partial_start, top); | |
869 | set_page_dirty(page); | |
870 | unlock_page(page); | |
09cbfeaf | 871 | put_page(page); |
83e4fa9c HD |
872 | } |
873 | } | |
874 | if (partial_end) { | |
875 | struct page *page = NULL; | |
9e18eb29 | 876 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
877 | if (page) { |
878 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
879 | set_page_dirty(page); |
880 | unlock_page(page); | |
09cbfeaf | 881 | put_page(page); |
bda97eab HD |
882 | } |
883 | } | |
83e4fa9c HD |
884 | if (start >= end) |
885 | return; | |
bda97eab HD |
886 | |
887 | index = start; | |
b1a36650 | 888 | while (index < end) { |
bda97eab | 889 | cond_resched(); |
0cd6144a JW |
890 | |
891 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 892 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 893 | pvec.pages, indices); |
7a5d0fbb | 894 | if (!pvec.nr) { |
b1a36650 HD |
895 | /* If all gone or hole-punch or unfalloc, we're done */ |
896 | if (index == start || end != -1) | |
bda97eab | 897 | break; |
b1a36650 | 898 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
899 | index = start; |
900 | continue; | |
901 | } | |
bda97eab HD |
902 | for (i = 0; i < pagevec_count(&pvec); i++) { |
903 | struct page *page = pvec.pages[i]; | |
904 | ||
7a5d0fbb | 905 | index = indices[i]; |
83e4fa9c | 906 | if (index >= end) |
bda97eab HD |
907 | break; |
908 | ||
7a5d0fbb | 909 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
910 | if (unfalloc) |
911 | continue; | |
b1a36650 HD |
912 | if (shmem_free_swap(mapping, index, page)) { |
913 | /* Swap was replaced by page: retry */ | |
914 | index--; | |
915 | break; | |
916 | } | |
917 | nr_swaps_freed++; | |
7a5d0fbb HD |
918 | continue; |
919 | } | |
920 | ||
bda97eab | 921 | lock_page(page); |
800d8c63 KS |
922 | |
923 | if (PageTransTail(page)) { | |
924 | /* Middle of THP: zero out the page */ | |
925 | clear_highpage(page); | |
926 | unlock_page(page); | |
927 | /* | |
928 | * Partial thp truncate due 'start' in middle | |
929 | * of THP: don't need to look on these pages | |
930 | * again on !pvec.nr restart. | |
931 | */ | |
932 | if (index != round_down(end, HPAGE_PMD_NR)) | |
933 | start++; | |
934 | continue; | |
935 | } else if (PageTransHuge(page)) { | |
936 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
937 | /* | |
938 | * Range ends in the middle of THP: | |
939 | * zero out the page | |
940 | */ | |
941 | clear_highpage(page); | |
942 | unlock_page(page); | |
943 | continue; | |
944 | } | |
945 | index += HPAGE_PMD_NR - 1; | |
946 | i += HPAGE_PMD_NR - 1; | |
947 | } | |
948 | ||
1635f6a7 | 949 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
950 | VM_BUG_ON_PAGE(PageTail(page), page); |
951 | if (page_mapping(page) == mapping) { | |
309381fe | 952 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 953 | truncate_inode_page(mapping, page); |
b1a36650 HD |
954 | } else { |
955 | /* Page was replaced by swap: retry */ | |
956 | unlock_page(page); | |
957 | index--; | |
958 | break; | |
1635f6a7 | 959 | } |
7a5d0fbb | 960 | } |
bda97eab HD |
961 | unlock_page(page); |
962 | } | |
0cd6144a | 963 | pagevec_remove_exceptionals(&pvec); |
24513264 | 964 | pagevec_release(&pvec); |
bda97eab HD |
965 | index++; |
966 | } | |
94c1e62d | 967 | |
4595ef88 | 968 | spin_lock_irq(&info->lock); |
7a5d0fbb | 969 | info->swapped -= nr_swaps_freed; |
1da177e4 | 970 | shmem_recalc_inode(inode); |
4595ef88 | 971 | spin_unlock_irq(&info->lock); |
1635f6a7 | 972 | } |
1da177e4 | 973 | |
1635f6a7 HD |
974 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
975 | { | |
976 | shmem_undo_range(inode, lstart, lend, false); | |
078cd827 | 977 | inode->i_ctime = inode->i_mtime = current_time(inode); |
1da177e4 | 978 | } |
94c1e62d | 979 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 980 | |
a528d35e DH |
981 | static int shmem_getattr(const struct path *path, struct kstat *stat, |
982 | u32 request_mask, unsigned int query_flags) | |
44a30220 | 983 | { |
a528d35e | 984 | struct inode *inode = path->dentry->d_inode; |
44a30220 YZ |
985 | struct shmem_inode_info *info = SHMEM_I(inode); |
986 | ||
d0424c42 | 987 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
4595ef88 | 988 | spin_lock_irq(&info->lock); |
d0424c42 | 989 | shmem_recalc_inode(inode); |
4595ef88 | 990 | spin_unlock_irq(&info->lock); |
d0424c42 | 991 | } |
44a30220 | 992 | generic_fillattr(inode, stat); |
44a30220 YZ |
993 | return 0; |
994 | } | |
995 | ||
94c1e62d | 996 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 997 | { |
75c3cfa8 | 998 | struct inode *inode = d_inode(dentry); |
40e041a2 | 999 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1000 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 LT |
1001 | int error; |
1002 | ||
31051c85 | 1003 | error = setattr_prepare(dentry, attr); |
db78b877 CH |
1004 | if (error) |
1005 | return error; | |
1006 | ||
94c1e62d HD |
1007 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
1008 | loff_t oldsize = inode->i_size; | |
1009 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 1010 | |
40e041a2 DH |
1011 | /* protected by i_mutex */ |
1012 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
1013 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
1014 | return -EPERM; | |
1015 | ||
94c1e62d | 1016 | if (newsize != oldsize) { |
77142517 KK |
1017 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
1018 | oldsize, newsize); | |
1019 | if (error) | |
1020 | return error; | |
94c1e62d | 1021 | i_size_write(inode, newsize); |
078cd827 | 1022 | inode->i_ctime = inode->i_mtime = current_time(inode); |
94c1e62d | 1023 | } |
afa2db2f | 1024 | if (newsize <= oldsize) { |
94c1e62d | 1025 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
1026 | if (oldsize > holebegin) |
1027 | unmap_mapping_range(inode->i_mapping, | |
1028 | holebegin, 0, 1); | |
1029 | if (info->alloced) | |
1030 | shmem_truncate_range(inode, | |
1031 | newsize, (loff_t)-1); | |
94c1e62d | 1032 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
1033 | if (oldsize > holebegin) |
1034 | unmap_mapping_range(inode->i_mapping, | |
1035 | holebegin, 0, 1); | |
779750d2 KS |
1036 | |
1037 | /* | |
1038 | * Part of the huge page can be beyond i_size: subject | |
1039 | * to shrink under memory pressure. | |
1040 | */ | |
1041 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { | |
1042 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1043 | /* |
1044 | * _careful to defend against unlocked access to | |
1045 | * ->shrink_list in shmem_unused_huge_shrink() | |
1046 | */ | |
1047 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1048 | list_add_tail(&info->shrinklist, |
1049 | &sbinfo->shrinklist); | |
1050 | sbinfo->shrinklist_len++; | |
1051 | } | |
1052 | spin_unlock(&sbinfo->shrinklist_lock); | |
1053 | } | |
94c1e62d | 1054 | } |
1da177e4 LT |
1055 | } |
1056 | ||
db78b877 | 1057 | setattr_copy(inode, attr); |
db78b877 | 1058 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 1059 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
1060 | return error; |
1061 | } | |
1062 | ||
1f895f75 | 1063 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1064 | { |
1da177e4 | 1065 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1066 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 | 1067 | |
3889e6e7 | 1068 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
1069 | shmem_unacct_size(info->flags, inode->i_size); |
1070 | inode->i_size = 0; | |
3889e6e7 | 1071 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1072 | if (!list_empty(&info->shrinklist)) { |
1073 | spin_lock(&sbinfo->shrinklist_lock); | |
1074 | if (!list_empty(&info->shrinklist)) { | |
1075 | list_del_init(&info->shrinklist); | |
1076 | sbinfo->shrinklist_len--; | |
1077 | } | |
1078 | spin_unlock(&sbinfo->shrinklist_lock); | |
1079 | } | |
1da177e4 | 1080 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 1081 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 1082 | list_del_init(&info->swaplist); |
cb5f7b9a | 1083 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1084 | } |
3ed47db3 | 1085 | } |
b09e0fa4 | 1086 | |
38f38657 | 1087 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 1088 | WARN_ON(inode->i_blocks); |
0006ebb4 SF |
1089 | if (!sbinfo->idr_nouse && inode->i_ino) { |
1090 | mutex_lock(&sbinfo->idr_lock); | |
1091 | idr_remove(&sbinfo->idr, inode->i_ino); | |
1092 | mutex_unlock(&sbinfo->idr_lock); | |
1093 | } | |
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; | |
1101 | void **slot; | |
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) { | |
1107 | if (*slot == item) { | |
1108 | found = iter.index; | |
1109 | break; | |
1110 | } | |
1111 | checked++; | |
1112 | if ((checked % 4096) != 0) | |
1113 | continue; | |
1114 | slot = radix_tree_iter_resume(slot, &iter); | |
1115 | cond_resched_rcu(); | |
1116 | } | |
1117 | ||
1118 | rcu_read_unlock(); | |
1119 | return found; | |
1120 | } | |
1121 | ||
46f65ec1 HD |
1122 | /* |
1123 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1124 | */ | |
41ffe5d5 | 1125 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 1126 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 1127 | { |
285b2c4f | 1128 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 1129 | void *radswap; |
41ffe5d5 | 1130 | pgoff_t index; |
bde05d1c HD |
1131 | gfp_t gfp; |
1132 | int error = 0; | |
1da177e4 | 1133 | |
46f65ec1 | 1134 | radswap = swp_to_radix_entry(swap); |
478922e2 | 1135 | index = find_swap_entry(&mapping->page_tree, radswap); |
46f65ec1 | 1136 | if (index == -1) |
00501b53 | 1137 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 1138 | |
1b1b32f2 HD |
1139 | /* |
1140 | * Move _head_ to start search for next from here. | |
1f895f75 | 1141 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 1142 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 1143 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
1144 | */ |
1145 | if (shmem_swaplist.next != &info->swaplist) | |
1146 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 1147 | |
bde05d1c HD |
1148 | gfp = mapping_gfp_mask(mapping); |
1149 | if (shmem_should_replace_page(*pagep, gfp)) { | |
1150 | mutex_unlock(&shmem_swaplist_mutex); | |
1151 | error = shmem_replace_page(pagep, gfp, info, index); | |
1152 | mutex_lock(&shmem_swaplist_mutex); | |
1153 | /* | |
1154 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
1155 | * allocation, but the inode might have been freed while we |
1156 | * dropped it: although a racing shmem_evict_inode() cannot | |
1157 | * complete without emptying the radix_tree, our page lock | |
1158 | * on this swapcache page is not enough to prevent that - | |
1159 | * free_swap_and_cache() of our swap entry will only | |
1160 | * trylock_page(), removing swap from radix_tree whatever. | |
1161 | * | |
1162 | * We must not proceed to shmem_add_to_page_cache() if the | |
1163 | * inode has been freed, but of course we cannot rely on | |
1164 | * inode or mapping or info to check that. However, we can | |
1165 | * safely check if our swap entry is still in use (and here | |
1166 | * it can't have got reused for another page): if it's still | |
1167 | * in use, then the inode cannot have been freed yet, and we | |
1168 | * can safely proceed (if it's no longer in use, that tells | |
1169 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
1170 | */ |
1171 | if (!page_swapcount(*pagep)) | |
1172 | error = -ENOENT; | |
1173 | } | |
1174 | ||
d13d1443 | 1175 | /* |
778dd893 HD |
1176 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
1177 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
1178 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 1179 | */ |
bde05d1c HD |
1180 | if (!error) |
1181 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 1182 | radswap); |
48f170fb | 1183 | if (error != -ENOMEM) { |
46f65ec1 HD |
1184 | /* |
1185 | * Truncation and eviction use free_swap_and_cache(), which | |
1186 | * only does trylock page: if we raced, best clean up here. | |
1187 | */ | |
bde05d1c HD |
1188 | delete_from_swap_cache(*pagep); |
1189 | set_page_dirty(*pagep); | |
46f65ec1 | 1190 | if (!error) { |
4595ef88 | 1191 | spin_lock_irq(&info->lock); |
46f65ec1 | 1192 | info->swapped--; |
4595ef88 | 1193 | spin_unlock_irq(&info->lock); |
46f65ec1 HD |
1194 | swap_free(swap); |
1195 | } | |
1da177e4 | 1196 | } |
2e0e26c7 | 1197 | return error; |
1da177e4 LT |
1198 | } |
1199 | ||
1200 | /* | |
46f65ec1 | 1201 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 1202 | */ |
41ffe5d5 | 1203 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 1204 | { |
41ffe5d5 | 1205 | struct list_head *this, *next; |
1da177e4 | 1206 | struct shmem_inode_info *info; |
00501b53 | 1207 | struct mem_cgroup *memcg; |
bde05d1c HD |
1208 | int error = 0; |
1209 | ||
1210 | /* | |
1211 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 1212 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 1213 | */ |
0142ef6c | 1214 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 1215 | goto out; |
778dd893 HD |
1216 | |
1217 | /* | |
1218 | * Charge page using GFP_KERNEL while we can wait, before taking | |
1219 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
1220 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 1221 | */ |
f627c2f5 KS |
1222 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg, |
1223 | false); | |
778dd893 HD |
1224 | if (error) |
1225 | goto out; | |
46f65ec1 | 1226 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 1227 | error = -EAGAIN; |
1da177e4 | 1228 | |
cb5f7b9a | 1229 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
1230 | list_for_each_safe(this, next, &shmem_swaplist) { |
1231 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 1232 | if (info->swapped) |
00501b53 | 1233 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
1234 | else |
1235 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1236 | cond_resched(); |
00501b53 | 1237 | if (error != -EAGAIN) |
778dd893 | 1238 | break; |
00501b53 | 1239 | /* found nothing in this: move on to search the next */ |
1da177e4 | 1240 | } |
cb5f7b9a | 1241 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1242 | |
00501b53 JW |
1243 | if (error) { |
1244 | if (error != -ENOMEM) | |
1245 | error = 0; | |
f627c2f5 | 1246 | mem_cgroup_cancel_charge(page, memcg, false); |
00501b53 | 1247 | } else |
f627c2f5 | 1248 | mem_cgroup_commit_charge(page, memcg, true, false); |
778dd893 | 1249 | out: |
aaa46865 | 1250 | unlock_page(page); |
09cbfeaf | 1251 | put_page(page); |
778dd893 | 1252 | return error; |
1da177e4 LT |
1253 | } |
1254 | ||
1255 | /* | |
1256 | * Move the page from the page cache to the swap cache. | |
1257 | */ | |
1258 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1259 | { | |
1260 | struct shmem_inode_info *info; | |
1da177e4 | 1261 | struct address_space *mapping; |
1da177e4 | 1262 | struct inode *inode; |
6922c0c7 HD |
1263 | swp_entry_t swap; |
1264 | pgoff_t index; | |
1da177e4 | 1265 | |
800d8c63 | 1266 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1267 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1268 | mapping = page->mapping; |
1269 | index = page->index; | |
1270 | inode = mapping->host; | |
1271 | info = SHMEM_I(inode); | |
1272 | if (info->flags & VM_LOCKED) | |
1273 | goto redirty; | |
d9fe526a | 1274 | if (!total_swap_pages) |
1da177e4 LT |
1275 | goto redirty; |
1276 | ||
d9fe526a | 1277 | /* |
97b713ba CH |
1278 | * Our capabilities prevent regular writeback or sync from ever calling |
1279 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1280 | * its underlying filesystem, in which case tmpfs should write out to | |
1281 | * swap only in response to memory pressure, and not for the writeback | |
1282 | * threads or sync. | |
d9fe526a | 1283 | */ |
48f170fb HD |
1284 | if (!wbc->for_reclaim) { |
1285 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1286 | goto redirty; | |
1287 | } | |
1635f6a7 HD |
1288 | |
1289 | /* | |
1290 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1291 | * value into swapfile.c, the only way we can correctly account for a | |
1292 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1293 | * |
1294 | * That's okay for a page already fallocated earlier, but if we have | |
1295 | * not yet completed the fallocation, then (a) we want to keep track | |
1296 | * of this page in case we have to undo it, and (b) it may not be a | |
1297 | * good idea to continue anyway, once we're pushing into swap. So | |
1298 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1299 | */ |
1300 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1301 | if (inode->i_private) { |
1302 | struct shmem_falloc *shmem_falloc; | |
1303 | spin_lock(&inode->i_lock); | |
1304 | shmem_falloc = inode->i_private; | |
1305 | if (shmem_falloc && | |
8e205f77 | 1306 | !shmem_falloc->waitq && |
1aac1400 HD |
1307 | index >= shmem_falloc->start && |
1308 | index < shmem_falloc->next) | |
1309 | shmem_falloc->nr_unswapped++; | |
1310 | else | |
1311 | shmem_falloc = NULL; | |
1312 | spin_unlock(&inode->i_lock); | |
1313 | if (shmem_falloc) | |
1314 | goto redirty; | |
1315 | } | |
1635f6a7 HD |
1316 | clear_highpage(page); |
1317 | flush_dcache_page(page); | |
1318 | SetPageUptodate(page); | |
1319 | } | |
1320 | ||
38d8b4e6 | 1321 | swap = get_swap_page(page); |
48f170fb HD |
1322 | if (!swap.val) |
1323 | goto redirty; | |
d9fe526a | 1324 | |
37e84351 VD |
1325 | if (mem_cgroup_try_charge_swap(page, swap)) |
1326 | goto free_swap; | |
1327 | ||
b1dea800 HD |
1328 | /* |
1329 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1330 | * if it's not already there. Do it now before the page is |
1331 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1332 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1333 | * we've incremented swapped, because shmem_unuse_inode() will |
1334 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1335 | */ |
48f170fb HD |
1336 | mutex_lock(&shmem_swaplist_mutex); |
1337 | if (list_empty(&info->swaplist)) | |
1338 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 1339 | |
48f170fb | 1340 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
4595ef88 | 1341 | spin_lock_irq(&info->lock); |
6922c0c7 | 1342 | shmem_recalc_inode(inode); |
267a4c76 | 1343 | info->swapped++; |
4595ef88 | 1344 | spin_unlock_irq(&info->lock); |
6922c0c7 | 1345 | |
267a4c76 HD |
1346 | swap_shmem_alloc(swap); |
1347 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1348 | ||
6922c0c7 | 1349 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1350 | BUG_ON(page_mapped(page)); |
9fab5619 | 1351 | swap_writepage(page, wbc); |
1da177e4 LT |
1352 | return 0; |
1353 | } | |
1354 | ||
6922c0c7 | 1355 | mutex_unlock(&shmem_swaplist_mutex); |
37e84351 | 1356 | free_swap: |
75f6d6d2 | 1357 | put_swap_page(page, swap); |
1da177e4 LT |
1358 | redirty: |
1359 | set_page_dirty(page); | |
d9fe526a HD |
1360 | if (wbc->for_reclaim) |
1361 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1362 | unlock_page(page); | |
1363 | return 0; | |
1da177e4 LT |
1364 | } |
1365 | ||
75edd345 | 1366 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1367 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1368 | { |
095f1fc4 | 1369 | char buffer[64]; |
680d794b | 1370 | |
71fe804b | 1371 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1372 | return; /* show nothing */ |
680d794b | 1373 | |
a7a88b23 | 1374 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1375 | |
1376 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1377 | } |
71fe804b LS |
1378 | |
1379 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1380 | { | |
1381 | struct mempolicy *mpol = NULL; | |
1382 | if (sbinfo->mpol) { | |
1383 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1384 | mpol = sbinfo->mpol; | |
1385 | mpol_get(mpol); | |
1386 | spin_unlock(&sbinfo->stat_lock); | |
1387 | } | |
1388 | return mpol; | |
1389 | } | |
75edd345 HD |
1390 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1391 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1392 | { | |
1393 | } | |
1394 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1395 | { | |
1396 | return NULL; | |
1397 | } | |
1398 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1399 | #ifndef CONFIG_NUMA | |
1400 | #define vm_policy vm_private_data | |
1401 | #endif | |
680d794b | 1402 | |
800d8c63 KS |
1403 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1404 | struct shmem_inode_info *info, pgoff_t index) | |
1405 | { | |
1406 | /* Create a pseudo vma that just contains the policy */ | |
1407 | vma->vm_start = 0; | |
1408 | /* Bias interleave by inode number to distribute better across nodes */ | |
1409 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
1410 | vma->vm_ops = NULL; | |
1411 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); | |
1412 | } | |
1413 | ||
1414 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1415 | { | |
1416 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1417 | mpol_cond_put(vma->vm_policy); | |
1418 | } | |
1419 | ||
41ffe5d5 HD |
1420 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1421 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1422 | { |
1da177e4 | 1423 | struct vm_area_struct pvma; |
18a2f371 | 1424 | struct page *page; |
52cd3b07 | 1425 | |
800d8c63 | 1426 | shmem_pseudo_vma_init(&pvma, info, index); |
18a2f371 | 1427 | page = swapin_readahead(swap, gfp, &pvma, 0); |
800d8c63 | 1428 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1429 | |
800d8c63 KS |
1430 | return page; |
1431 | } | |
1432 | ||
1433 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1434 | struct shmem_inode_info *info, pgoff_t index) | |
1435 | { | |
1436 | struct vm_area_struct pvma; | |
1437 | struct inode *inode = &info->vfs_inode; | |
1438 | struct address_space *mapping = inode->i_mapping; | |
4620a06e | 1439 | pgoff_t idx, hindex; |
800d8c63 KS |
1440 | void __rcu **results; |
1441 | struct page *page; | |
1442 | ||
e496cf3d | 1443 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1444 | return NULL; |
1445 | ||
4620a06e | 1446 | hindex = round_down(index, HPAGE_PMD_NR); |
800d8c63 KS |
1447 | rcu_read_lock(); |
1448 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx, | |
1449 | hindex, 1) && idx < hindex + HPAGE_PMD_NR) { | |
1450 | rcu_read_unlock(); | |
1451 | return NULL; | |
1452 | } | |
1453 | rcu_read_unlock(); | |
18a2f371 | 1454 | |
800d8c63 KS |
1455 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1456 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
1457 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); | |
1458 | shmem_pseudo_vma_destroy(&pvma); | |
1459 | if (page) | |
1460 | prep_transhuge_page(page); | |
18a2f371 | 1461 | return page; |
1da177e4 LT |
1462 | } |
1463 | ||
02098fea | 1464 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1465 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1466 | { |
1467 | struct vm_area_struct pvma; | |
18a2f371 | 1468 | struct page *page; |
1da177e4 | 1469 | |
800d8c63 KS |
1470 | shmem_pseudo_vma_init(&pvma, info, index); |
1471 | page = alloc_page_vma(gfp, &pvma, 0); | |
1472 | shmem_pseudo_vma_destroy(&pvma); | |
1473 | ||
1474 | return page; | |
1475 | } | |
1476 | ||
1477 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
0f079694 | 1478 | struct inode *inode, |
800d8c63 KS |
1479 | pgoff_t index, bool huge) |
1480 | { | |
0f079694 | 1481 | struct shmem_inode_info *info = SHMEM_I(inode); |
800d8c63 KS |
1482 | struct page *page; |
1483 | int nr; | |
1484 | int err = -ENOSPC; | |
52cd3b07 | 1485 | |
e496cf3d | 1486 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1487 | huge = false; |
1488 | nr = huge ? HPAGE_PMD_NR : 1; | |
1489 | ||
0f079694 | 1490 | if (!shmem_inode_acct_block(inode, nr)) |
800d8c63 | 1491 | goto failed; |
800d8c63 KS |
1492 | |
1493 | if (huge) | |
1494 | page = shmem_alloc_hugepage(gfp, info, index); | |
1495 | else | |
1496 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1497 | if (page) { |
1498 | __SetPageLocked(page); | |
1499 | __SetPageSwapBacked(page); | |
800d8c63 | 1500 | return page; |
75edd345 | 1501 | } |
18a2f371 | 1502 | |
800d8c63 | 1503 | err = -ENOMEM; |
0f079694 | 1504 | shmem_inode_unacct_blocks(inode, nr); |
800d8c63 KS |
1505 | failed: |
1506 | return ERR_PTR(err); | |
1da177e4 | 1507 | } |
71fe804b | 1508 | |
bde05d1c HD |
1509 | /* |
1510 | * When a page is moved from swapcache to shmem filecache (either by the | |
1511 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1512 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1513 | * ignorance of the mapping it belongs to. If that mapping has special | |
1514 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1515 | * we may need to copy to a suitable page before moving to filecache. | |
1516 | * | |
1517 | * In a future release, this may well be extended to respect cpuset and | |
1518 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1519 | * but for now it is a simple matter of zone. | |
1520 | */ | |
1521 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1522 | { | |
1523 | return page_zonenum(page) > gfp_zone(gfp); | |
1524 | } | |
1525 | ||
1526 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1527 | struct shmem_inode_info *info, pgoff_t index) | |
1528 | { | |
1529 | struct page *oldpage, *newpage; | |
1530 | struct address_space *swap_mapping; | |
1531 | pgoff_t swap_index; | |
1532 | int error; | |
1533 | ||
1534 | oldpage = *pagep; | |
1535 | swap_index = page_private(oldpage); | |
1536 | swap_mapping = page_mapping(oldpage); | |
1537 | ||
1538 | /* | |
1539 | * We have arrived here because our zones are constrained, so don't | |
1540 | * limit chance of success by further cpuset and node constraints. | |
1541 | */ | |
1542 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1543 | newpage = shmem_alloc_page(gfp, info, index); | |
1544 | if (!newpage) | |
1545 | return -ENOMEM; | |
bde05d1c | 1546 | |
09cbfeaf | 1547 | get_page(newpage); |
bde05d1c | 1548 | copy_highpage(newpage, oldpage); |
0142ef6c | 1549 | flush_dcache_page(newpage); |
bde05d1c | 1550 | |
9956edf3 HD |
1551 | __SetPageLocked(newpage); |
1552 | __SetPageSwapBacked(newpage); | |
bde05d1c | 1553 | SetPageUptodate(newpage); |
bde05d1c | 1554 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1555 | SetPageSwapCache(newpage); |
1556 | ||
1557 | /* | |
1558 | * Our caller will very soon move newpage out of swapcache, but it's | |
1559 | * a nice clean interface for us to replace oldpage by newpage there. | |
1560 | */ | |
1561 | spin_lock_irq(&swap_mapping->tree_lock); | |
1562 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1563 | newpage); | |
0142ef6c | 1564 | if (!error) { |
11fb9989 MG |
1565 | __inc_node_page_state(newpage, NR_FILE_PAGES); |
1566 | __dec_node_page_state(oldpage, NR_FILE_PAGES); | |
0142ef6c | 1567 | } |
bde05d1c | 1568 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1569 | |
0142ef6c HD |
1570 | if (unlikely(error)) { |
1571 | /* | |
1572 | * Is this possible? I think not, now that our callers check | |
1573 | * both PageSwapCache and page_private after getting page lock; | |
1574 | * but be defensive. Reverse old to newpage for clear and free. | |
1575 | */ | |
1576 | oldpage = newpage; | |
1577 | } else { | |
6a93ca8f | 1578 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1579 | lru_cache_add_anon(newpage); |
1580 | *pagep = newpage; | |
1581 | } | |
bde05d1c HD |
1582 | |
1583 | ClearPageSwapCache(oldpage); | |
1584 | set_page_private(oldpage, 0); | |
1585 | ||
1586 | unlock_page(oldpage); | |
09cbfeaf KS |
1587 | put_page(oldpage); |
1588 | put_page(oldpage); | |
0142ef6c | 1589 | return error; |
bde05d1c HD |
1590 | } |
1591 | ||
1da177e4 | 1592 | /* |
68da9f05 | 1593 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1594 | * |
1595 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1596 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1597 | * entry since a page cannot live in both the swap and page cache. |
1598 | * | |
1599 | * fault_mm and fault_type are only supplied by shmem_fault: | |
1600 | * otherwise they are NULL. | |
1da177e4 | 1601 | */ |
41ffe5d5 | 1602 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 1603 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
cfda0526 | 1604 | struct vm_area_struct *vma, struct vm_fault *vmf, int *fault_type) |
1da177e4 LT |
1605 | { |
1606 | struct address_space *mapping = inode->i_mapping; | |
23f919d4 | 1607 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 | 1608 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1609 | struct mm_struct *charge_mm; |
00501b53 | 1610 | struct mem_cgroup *memcg; |
27ab7006 | 1611 | struct page *page; |
1da177e4 | 1612 | swp_entry_t swap; |
657e3038 | 1613 | enum sgp_type sgp_huge = sgp; |
800d8c63 | 1614 | pgoff_t hindex = index; |
1da177e4 | 1615 | int error; |
54af6042 | 1616 | int once = 0; |
1635f6a7 | 1617 | int alloced = 0; |
1da177e4 | 1618 | |
09cbfeaf | 1619 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1620 | return -EFBIG; |
657e3038 KS |
1621 | if (sgp == SGP_NOHUGE || sgp == SGP_HUGE) |
1622 | sgp = SGP_CACHE; | |
1da177e4 | 1623 | repeat: |
54af6042 | 1624 | swap.val = 0; |
0cd6144a | 1625 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1626 | if (radix_tree_exceptional_entry(page)) { |
1627 | swap = radix_to_swp_entry(page); | |
1628 | page = NULL; | |
1629 | } | |
1630 | ||
75edd345 | 1631 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1632 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 1633 | error = -EINVAL; |
267a4c76 | 1634 | goto unlock; |
54af6042 HD |
1635 | } |
1636 | ||
66d2f4d2 HD |
1637 | if (page && sgp == SGP_WRITE) |
1638 | mark_page_accessed(page); | |
1639 | ||
1635f6a7 HD |
1640 | /* fallocated page? */ |
1641 | if (page && !PageUptodate(page)) { | |
1642 | if (sgp != SGP_READ) | |
1643 | goto clear; | |
1644 | unlock_page(page); | |
09cbfeaf | 1645 | put_page(page); |
1635f6a7 HD |
1646 | page = NULL; |
1647 | } | |
54af6042 | 1648 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1649 | *pagep = page; |
1650 | return 0; | |
27ab7006 HD |
1651 | } |
1652 | ||
1653 | /* | |
54af6042 HD |
1654 | * Fast cache lookup did not find it: |
1655 | * bring it back from swap or allocate. | |
27ab7006 | 1656 | */ |
54af6042 | 1657 | sbinfo = SHMEM_SB(inode->i_sb); |
cfda0526 | 1658 | charge_mm = vma ? vma->vm_mm : current->mm; |
1da177e4 | 1659 | |
1da177e4 LT |
1660 | if (swap.val) { |
1661 | /* Look it up and read it in.. */ | |
ec560175 | 1662 | page = lookup_swap_cache(swap, NULL, 0); |
27ab7006 | 1663 | if (!page) { |
9e18eb29 ALC |
1664 | /* Or update major stats only when swapin succeeds?? */ |
1665 | if (fault_type) { | |
68da9f05 | 1666 | *fault_type |= VM_FAULT_MAJOR; |
9e18eb29 | 1667 | count_vm_event(PGMAJFAULT); |
2262185c | 1668 | count_memcg_event_mm(charge_mm, PGMAJFAULT); |
9e18eb29 ALC |
1669 | } |
1670 | /* Here we actually start the io */ | |
41ffe5d5 | 1671 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1672 | if (!page) { |
54af6042 HD |
1673 | error = -ENOMEM; |
1674 | goto failed; | |
1da177e4 | 1675 | } |
1da177e4 LT |
1676 | } |
1677 | ||
1678 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1679 | lock_page(page); |
0142ef6c | 1680 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1681 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1682 | error = -EEXIST; /* try again */ |
d1899228 | 1683 | goto unlock; |
bde05d1c | 1684 | } |
27ab7006 | 1685 | if (!PageUptodate(page)) { |
1da177e4 | 1686 | error = -EIO; |
54af6042 | 1687 | goto failed; |
1da177e4 | 1688 | } |
54af6042 HD |
1689 | wait_on_page_writeback(page); |
1690 | ||
bde05d1c HD |
1691 | if (shmem_should_replace_page(page, gfp)) { |
1692 | error = shmem_replace_page(&page, gfp, info, index); | |
1693 | if (error) | |
1694 | goto failed; | |
1da177e4 | 1695 | } |
27ab7006 | 1696 | |
9e18eb29 | 1697 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
f627c2f5 | 1698 | false); |
d1899228 | 1699 | if (!error) { |
aa3b1895 | 1700 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1701 | swp_to_radix_entry(swap)); |
215c02bc HD |
1702 | /* |
1703 | * We already confirmed swap under page lock, and make | |
1704 | * no memory allocation here, so usually no possibility | |
1705 | * of error; but free_swap_and_cache() only trylocks a | |
1706 | * page, so it is just possible that the entry has been | |
1707 | * truncated or holepunched since swap was confirmed. | |
1708 | * shmem_undo_range() will have done some of the | |
1709 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1710 | * the rest. |
215c02bc HD |
1711 | * Reset swap.val? No, leave it so "failed" goes back to |
1712 | * "repeat": reading a hole and writing should succeed. | |
1713 | */ | |
00501b53 | 1714 | if (error) { |
f627c2f5 | 1715 | mem_cgroup_cancel_charge(page, memcg, false); |
215c02bc | 1716 | delete_from_swap_cache(page); |
00501b53 | 1717 | } |
d1899228 | 1718 | } |
54af6042 HD |
1719 | if (error) |
1720 | goto failed; | |
1721 | ||
f627c2f5 | 1722 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1723 | |
4595ef88 | 1724 | spin_lock_irq(&info->lock); |
285b2c4f | 1725 | info->swapped--; |
54af6042 | 1726 | shmem_recalc_inode(inode); |
4595ef88 | 1727 | spin_unlock_irq(&info->lock); |
54af6042 | 1728 | |
66d2f4d2 HD |
1729 | if (sgp == SGP_WRITE) |
1730 | mark_page_accessed(page); | |
1731 | ||
54af6042 | 1732 | delete_from_swap_cache(page); |
27ab7006 HD |
1733 | set_page_dirty(page); |
1734 | swap_free(swap); | |
1735 | ||
54af6042 | 1736 | } else { |
cfda0526 MR |
1737 | if (vma && userfaultfd_missing(vma)) { |
1738 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
1739 | return 0; | |
1740 | } | |
1741 | ||
800d8c63 KS |
1742 | /* shmem_symlink() */ |
1743 | if (mapping->a_ops != &shmem_aops) | |
1744 | goto alloc_nohuge; | |
657e3038 | 1745 | if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE) |
800d8c63 KS |
1746 | goto alloc_nohuge; |
1747 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1748 | goto alloc_huge; | |
1749 | switch (sbinfo->huge) { | |
1750 | loff_t i_size; | |
1751 | pgoff_t off; | |
1752 | case SHMEM_HUGE_NEVER: | |
1753 | goto alloc_nohuge; | |
1754 | case SHMEM_HUGE_WITHIN_SIZE: | |
1755 | off = round_up(index, HPAGE_PMD_NR); | |
1756 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1757 | if (i_size >= HPAGE_PMD_SIZE && | |
1758 | i_size >> PAGE_SHIFT >= off) | |
1759 | goto alloc_huge; | |
1760 | /* fallthrough */ | |
1761 | case SHMEM_HUGE_ADVISE: | |
657e3038 KS |
1762 | if (sgp_huge == SGP_HUGE) |
1763 | goto alloc_huge; | |
1764 | /* TODO: implement fadvise() hints */ | |
800d8c63 | 1765 | goto alloc_nohuge; |
54af6042 | 1766 | } |
1da177e4 | 1767 | |
800d8c63 | 1768 | alloc_huge: |
0f079694 | 1769 | page = shmem_alloc_and_acct_page(gfp, inode, index, true); |
800d8c63 | 1770 | if (IS_ERR(page)) { |
0f079694 | 1771 | alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, inode, |
800d8c63 | 1772 | index, false); |
1da177e4 | 1773 | } |
800d8c63 | 1774 | if (IS_ERR(page)) { |
779750d2 | 1775 | int retry = 5; |
800d8c63 KS |
1776 | error = PTR_ERR(page); |
1777 | page = NULL; | |
779750d2 KS |
1778 | if (error != -ENOSPC) |
1779 | goto failed; | |
1780 | /* | |
1781 | * Try to reclaim some spece by splitting a huge page | |
1782 | * beyond i_size on the filesystem. | |
1783 | */ | |
1784 | while (retry--) { | |
1785 | int ret; | |
1786 | ret = shmem_unused_huge_shrink(sbinfo, NULL, 1); | |
1787 | if (ret == SHRINK_STOP) | |
1788 | break; | |
1789 | if (ret) | |
1790 | goto alloc_nohuge; | |
1791 | } | |
800d8c63 KS |
1792 | goto failed; |
1793 | } | |
1794 | ||
1795 | if (PageTransHuge(page)) | |
1796 | hindex = round_down(index, HPAGE_PMD_NR); | |
1797 | else | |
1798 | hindex = index; | |
1799 | ||
66d2f4d2 | 1800 | if (sgp == SGP_WRITE) |
eb39d618 | 1801 | __SetPageReferenced(page); |
66d2f4d2 | 1802 | |
9e18eb29 | 1803 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
800d8c63 | 1804 | PageTransHuge(page)); |
54af6042 | 1805 | if (error) |
800d8c63 KS |
1806 | goto unacct; |
1807 | error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK, | |
1808 | compound_order(page)); | |
b065b432 | 1809 | if (!error) { |
800d8c63 | 1810 | error = shmem_add_to_page_cache(page, mapping, hindex, |
fed400a1 | 1811 | NULL); |
b065b432 HD |
1812 | radix_tree_preload_end(); |
1813 | } | |
1814 | if (error) { | |
800d8c63 KS |
1815 | mem_cgroup_cancel_charge(page, memcg, |
1816 | PageTransHuge(page)); | |
1817 | goto unacct; | |
b065b432 | 1818 | } |
800d8c63 KS |
1819 | mem_cgroup_commit_charge(page, memcg, false, |
1820 | PageTransHuge(page)); | |
54af6042 HD |
1821 | lru_cache_add_anon(page); |
1822 | ||
4595ef88 | 1823 | spin_lock_irq(&info->lock); |
800d8c63 KS |
1824 | info->alloced += 1 << compound_order(page); |
1825 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); | |
54af6042 | 1826 | shmem_recalc_inode(inode); |
4595ef88 | 1827 | spin_unlock_irq(&info->lock); |
1635f6a7 | 1828 | alloced = true; |
54af6042 | 1829 | |
779750d2 KS |
1830 | if (PageTransHuge(page) && |
1831 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < | |
1832 | hindex + HPAGE_PMD_NR - 1) { | |
1833 | /* | |
1834 | * Part of the huge page is beyond i_size: subject | |
1835 | * to shrink under memory pressure. | |
1836 | */ | |
1837 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1838 | /* |
1839 | * _careful to defend against unlocked access to | |
1840 | * ->shrink_list in shmem_unused_huge_shrink() | |
1841 | */ | |
1842 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1843 | list_add_tail(&info->shrinklist, |
1844 | &sbinfo->shrinklist); | |
1845 | sbinfo->shrinklist_len++; | |
1846 | } | |
1847 | spin_unlock(&sbinfo->shrinklist_lock); | |
1848 | } | |
1849 | ||
ec9516fb | 1850 | /* |
1635f6a7 HD |
1851 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1852 | */ | |
1853 | if (sgp == SGP_FALLOC) | |
1854 | sgp = SGP_WRITE; | |
1855 | clear: | |
1856 | /* | |
1857 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1858 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1859 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb | 1860 | */ |
800d8c63 KS |
1861 | if (sgp != SGP_WRITE && !PageUptodate(page)) { |
1862 | struct page *head = compound_head(page); | |
1863 | int i; | |
1864 | ||
1865 | for (i = 0; i < (1 << compound_order(head)); i++) { | |
1866 | clear_highpage(head + i); | |
1867 | flush_dcache_page(head + i); | |
1868 | } | |
1869 | SetPageUptodate(head); | |
ec9516fb | 1870 | } |
1da177e4 | 1871 | } |
bde05d1c | 1872 | |
54af6042 | 1873 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1874 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1875 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1876 | if (alloced) { |
1877 | ClearPageDirty(page); | |
1878 | delete_from_page_cache(page); | |
4595ef88 | 1879 | spin_lock_irq(&info->lock); |
267a4c76 | 1880 | shmem_recalc_inode(inode); |
4595ef88 | 1881 | spin_unlock_irq(&info->lock); |
267a4c76 | 1882 | } |
54af6042 | 1883 | error = -EINVAL; |
267a4c76 | 1884 | goto unlock; |
e83c32e8 | 1885 | } |
800d8c63 | 1886 | *pagep = page + index - hindex; |
54af6042 | 1887 | return 0; |
1da177e4 | 1888 | |
59a16ead | 1889 | /* |
54af6042 | 1890 | * Error recovery. |
59a16ead | 1891 | */ |
54af6042 | 1892 | unacct: |
0f079694 | 1893 | shmem_inode_unacct_blocks(inode, 1 << compound_order(page)); |
800d8c63 KS |
1894 | |
1895 | if (PageTransHuge(page)) { | |
1896 | unlock_page(page); | |
1897 | put_page(page); | |
1898 | goto alloc_nohuge; | |
1899 | } | |
54af6042 | 1900 | failed: |
267a4c76 | 1901 | if (swap.val && !shmem_confirm_swap(mapping, index, swap)) |
d1899228 HD |
1902 | error = -EEXIST; |
1903 | unlock: | |
27ab7006 | 1904 | if (page) { |
54af6042 | 1905 | unlock_page(page); |
09cbfeaf | 1906 | put_page(page); |
54af6042 HD |
1907 | } |
1908 | if (error == -ENOSPC && !once++) { | |
4595ef88 | 1909 | spin_lock_irq(&info->lock); |
54af6042 | 1910 | shmem_recalc_inode(inode); |
4595ef88 | 1911 | spin_unlock_irq(&info->lock); |
27ab7006 | 1912 | goto repeat; |
ff36b801 | 1913 | } |
d1899228 | 1914 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1915 | goto repeat; |
1916 | return error; | |
1da177e4 LT |
1917 | } |
1918 | ||
10d20bd2 LT |
1919 | /* |
1920 | * This is like autoremove_wake_function, but it removes the wait queue | |
1921 | * entry unconditionally - even if something else had already woken the | |
1922 | * target. | |
1923 | */ | |
ac6424b9 | 1924 | static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
10d20bd2 LT |
1925 | { |
1926 | int ret = default_wake_function(wait, mode, sync, key); | |
2055da97 | 1927 | list_del_init(&wait->entry); |
10d20bd2 LT |
1928 | return ret; |
1929 | } | |
1930 | ||
11bac800 | 1931 | static int shmem_fault(struct vm_fault *vmf) |
1da177e4 | 1932 | { |
11bac800 | 1933 | struct vm_area_struct *vma = vmf->vma; |
496ad9aa | 1934 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 1935 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
657e3038 | 1936 | enum sgp_type sgp; |
1da177e4 | 1937 | int error; |
68da9f05 | 1938 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1939 | |
f00cdc6d HD |
1940 | /* |
1941 | * Trinity finds that probing a hole which tmpfs is punching can | |
1942 | * prevent the hole-punch from ever completing: which in turn | |
1943 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1944 | * faulting pages into the hole while it's being punched. Although |
1945 | * shmem_undo_range() does remove the additions, it may be unable to | |
1946 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1947 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1948 | * | |
1949 | * It does not matter if we sometimes reach this check just before the | |
1950 | * hole-punch begins, so that one fault then races with the punch: | |
1951 | * we just need to make racing faults a rare case. | |
1952 | * | |
1953 | * The implementation below would be much simpler if we just used a | |
1954 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1955 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1956 | */ |
1957 | if (unlikely(inode->i_private)) { | |
1958 | struct shmem_falloc *shmem_falloc; | |
1959 | ||
1960 | spin_lock(&inode->i_lock); | |
1961 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1962 | if (shmem_falloc && |
1963 | shmem_falloc->waitq && | |
1964 | vmf->pgoff >= shmem_falloc->start && | |
1965 | vmf->pgoff < shmem_falloc->next) { | |
1966 | wait_queue_head_t *shmem_falloc_waitq; | |
10d20bd2 | 1967 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); |
8e205f77 HD |
1968 | |
1969 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1970 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1971 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1972 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1973 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1974 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1975 | } |
8e205f77 HD |
1976 | |
1977 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1978 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1979 | TASK_UNINTERRUPTIBLE); | |
1980 | spin_unlock(&inode->i_lock); | |
1981 | schedule(); | |
1982 | ||
1983 | /* | |
1984 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1985 | * stack of the hole-punching task: shmem_falloc_waitq | |
1986 | * is usually invalid by the time we reach here, but | |
1987 | * finish_wait() does not dereference it in that case; | |
1988 | * though i_lock needed lest racing with wake_up_all(). | |
1989 | */ | |
1990 | spin_lock(&inode->i_lock); | |
1991 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1992 | spin_unlock(&inode->i_lock); | |
1993 | return ret; | |
f00cdc6d | 1994 | } |
8e205f77 | 1995 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1996 | } |
1997 | ||
657e3038 | 1998 | sgp = SGP_CACHE; |
18600332 MH |
1999 | |
2000 | if ((vma->vm_flags & VM_NOHUGEPAGE) || | |
2001 | test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) | |
657e3038 | 2002 | sgp = SGP_NOHUGE; |
18600332 MH |
2003 | else if (vma->vm_flags & VM_HUGEPAGE) |
2004 | sgp = SGP_HUGE; | |
657e3038 KS |
2005 | |
2006 | error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp, | |
cfda0526 | 2007 | gfp, vma, vmf, &ret); |
d0217ac0 NP |
2008 | if (error) |
2009 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 2010 | return ret; |
1da177e4 LT |
2011 | } |
2012 | ||
c01d5b30 HD |
2013 | unsigned long shmem_get_unmapped_area(struct file *file, |
2014 | unsigned long uaddr, unsigned long len, | |
2015 | unsigned long pgoff, unsigned long flags) | |
2016 | { | |
2017 | unsigned long (*get_area)(struct file *, | |
2018 | unsigned long, unsigned long, unsigned long, unsigned long); | |
2019 | unsigned long addr; | |
2020 | unsigned long offset; | |
2021 | unsigned long inflated_len; | |
2022 | unsigned long inflated_addr; | |
2023 | unsigned long inflated_offset; | |
2024 | ||
2025 | if (len > TASK_SIZE) | |
2026 | return -ENOMEM; | |
2027 | ||
2028 | get_area = current->mm->get_unmapped_area; | |
2029 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2030 | ||
e496cf3d | 2031 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
c01d5b30 HD |
2032 | return addr; |
2033 | if (IS_ERR_VALUE(addr)) | |
2034 | return addr; | |
2035 | if (addr & ~PAGE_MASK) | |
2036 | return addr; | |
2037 | if (addr > TASK_SIZE - len) | |
2038 | return addr; | |
2039 | ||
2040 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2041 | return addr; | |
2042 | if (len < HPAGE_PMD_SIZE) | |
2043 | return addr; | |
2044 | if (flags & MAP_FIXED) | |
2045 | return addr; | |
2046 | /* | |
2047 | * Our priority is to support MAP_SHARED mapped hugely; | |
2048 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
2049 | * But if caller specified an address hint, respect that as before. | |
2050 | */ | |
2051 | if (uaddr) | |
2052 | return addr; | |
2053 | ||
2054 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2055 | struct super_block *sb; | |
2056 | ||
2057 | if (file) { | |
2058 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2059 | sb = file_inode(file)->i_sb; | |
2060 | } else { | |
2061 | /* | |
2062 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2063 | * for "/dev/zero", to create a shared anonymous object. | |
2064 | */ | |
2065 | if (IS_ERR(shm_mnt)) | |
2066 | return addr; | |
2067 | sb = shm_mnt->mnt_sb; | |
2068 | } | |
3089bf61 | 2069 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2070 | return addr; |
2071 | } | |
2072 | ||
2073 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2074 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2075 | return addr; | |
2076 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2077 | return addr; | |
2078 | ||
2079 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2080 | if (inflated_len > TASK_SIZE) | |
2081 | return addr; | |
2082 | if (inflated_len < len) | |
2083 | return addr; | |
2084 | ||
2085 | inflated_addr = get_area(NULL, 0, inflated_len, 0, flags); | |
2086 | if (IS_ERR_VALUE(inflated_addr)) | |
2087 | return addr; | |
2088 | if (inflated_addr & ~PAGE_MASK) | |
2089 | return addr; | |
2090 | ||
2091 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2092 | inflated_addr += offset - inflated_offset; | |
2093 | if (inflated_offset > offset) | |
2094 | inflated_addr += HPAGE_PMD_SIZE; | |
2095 | ||
2096 | if (inflated_addr > TASK_SIZE - len) | |
2097 | return addr; | |
2098 | return inflated_addr; | |
2099 | } | |
2100 | ||
1da177e4 | 2101 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2102 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2103 | { |
496ad9aa | 2104 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2105 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2106 | } |
2107 | ||
d8dc74f2 AB |
2108 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
2109 | unsigned long addr) | |
1da177e4 | 2110 | { |
496ad9aa | 2111 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2112 | pgoff_t index; |
1da177e4 | 2113 | |
41ffe5d5 HD |
2114 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2115 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
2116 | } |
2117 | #endif | |
2118 | ||
2119 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
2120 | { | |
496ad9aa | 2121 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2122 | struct shmem_inode_info *info = SHMEM_I(inode); |
2123 | int retval = -ENOMEM; | |
2124 | ||
4595ef88 | 2125 | spin_lock_irq(&info->lock); |
1da177e4 LT |
2126 | if (lock && !(info->flags & VM_LOCKED)) { |
2127 | if (!user_shm_lock(inode->i_size, user)) | |
2128 | goto out_nomem; | |
2129 | info->flags |= VM_LOCKED; | |
89e004ea | 2130 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
2131 | } |
2132 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
2133 | user_shm_unlock(inode->i_size, user); | |
2134 | info->flags &= ~VM_LOCKED; | |
89e004ea | 2135 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2136 | } |
2137 | retval = 0; | |
89e004ea | 2138 | |
1da177e4 | 2139 | out_nomem: |
4595ef88 | 2140 | spin_unlock_irq(&info->lock); |
1da177e4 LT |
2141 | return retval; |
2142 | } | |
2143 | ||
9b83a6a8 | 2144 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
2145 | { |
2146 | file_accessed(file); | |
2147 | vma->vm_ops = &shmem_vm_ops; | |
e496cf3d | 2148 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
2149 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
2150 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
2151 | khugepaged_enter(vma, vma->vm_flags); | |
2152 | } | |
1da177e4 LT |
2153 | return 0; |
2154 | } | |
2155 | ||
454abafe | 2156 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 2157 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
2158 | { |
2159 | struct inode *inode; | |
2160 | struct shmem_inode_info *info; | |
2161 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
0006ebb4 | 2162 | int ino; |
1da177e4 | 2163 | |
5b04c689 PE |
2164 | if (shmem_reserve_inode(sb)) |
2165 | return NULL; | |
1da177e4 LT |
2166 | |
2167 | inode = new_inode(sb); | |
2168 | if (inode) { | |
454abafe | 2169 | inode_init_owner(inode, dir, mode); |
1da177e4 | 2170 | inode->i_blocks = 0; |
078cd827 | 2171 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
91828a40 | 2172 | inode->i_generation = get_seconds(); |
1da177e4 LT |
2173 | info = SHMEM_I(inode); |
2174 | memset(info, 0, (char *)inode - (char *)info); | |
2175 | spin_lock_init(&info->lock); | |
40e041a2 | 2176 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 2177 | info->flags = flags & VM_NORESERVE; |
779750d2 | 2178 | INIT_LIST_HEAD(&info->shrinklist); |
1da177e4 | 2179 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 2180 | simple_xattrs_init(&info->xattrs); |
72c04902 | 2181 | cache_no_acl(inode); |
1da177e4 LT |
2182 | |
2183 | switch (mode & S_IFMT) { | |
2184 | default: | |
39f0247d | 2185 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
2186 | init_special_inode(inode, mode, dev); |
2187 | break; | |
2188 | case S_IFREG: | |
14fcc23f | 2189 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
2190 | inode->i_op = &shmem_inode_operations; |
2191 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
2192 | mpol_shared_policy_init(&info->policy, |
2193 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
2194 | break; |
2195 | case S_IFDIR: | |
d8c76e6f | 2196 | inc_nlink(inode); |
1da177e4 LT |
2197 | /* Some things misbehave if size == 0 on a directory */ |
2198 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2199 | inode->i_op = &shmem_dir_inode_operations; | |
2200 | inode->i_fop = &simple_dir_operations; | |
2201 | break; | |
2202 | case S_IFLNK: | |
2203 | /* | |
2204 | * Must not load anything in the rbtree, | |
2205 | * mpol_free_shared_policy will not be called. | |
2206 | */ | |
71fe804b | 2207 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
2208 | break; |
2209 | } | |
0006ebb4 SF |
2210 | |
2211 | if (!sbinfo->idr_nouse) { | |
2212 | /* inum 0 and 1 are unused */ | |
2213 | mutex_lock(&sbinfo->idr_lock); | |
2214 | ino = idr_alloc(&sbinfo->idr, inode, 2, INT_MAX, | |
2215 | GFP_NOFS); | |
2216 | if (ino > 0) { | |
2217 | inode->i_ino = ino; | |
2218 | mutex_unlock(&sbinfo->idr_lock); | |
2219 | __insert_inode_hash(inode, inode->i_ino); | |
2220 | } else { | |
2221 | inode->i_ino = 0; | |
2222 | mutex_unlock(&sbinfo->idr_lock); | |
2223 | iput(inode); | |
2224 | /* shmem_free_inode() will be called */ | |
2225 | inode = NULL; | |
2226 | } | |
2227 | } else | |
2228 | inode->i_ino = get_next_ino(); | |
5b04c689 PE |
2229 | } else |
2230 | shmem_free_inode(sb); | |
1da177e4 LT |
2231 | return inode; |
2232 | } | |
2233 | ||
0cd6144a JW |
2234 | bool shmem_mapping(struct address_space *mapping) |
2235 | { | |
f8005451 | 2236 | return mapping->a_ops == &shmem_aops; |
0cd6144a JW |
2237 | } |
2238 | ||
8d103963 MR |
2239 | static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm, |
2240 | pmd_t *dst_pmd, | |
2241 | struct vm_area_struct *dst_vma, | |
2242 | unsigned long dst_addr, | |
2243 | unsigned long src_addr, | |
2244 | bool zeropage, | |
2245 | struct page **pagep) | |
4c27fe4c MR |
2246 | { |
2247 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2248 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4c27fe4c MR |
2249 | struct address_space *mapping = inode->i_mapping; |
2250 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2251 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
2252 | struct mem_cgroup *memcg; | |
2253 | spinlock_t *ptl; | |
2254 | void *page_kaddr; | |
2255 | struct page *page; | |
2256 | pte_t _dst_pte, *dst_pte; | |
2257 | int ret; | |
2258 | ||
cb658a45 | 2259 | ret = -ENOMEM; |
0f079694 | 2260 | if (!shmem_inode_acct_block(inode, 1)) |
cb658a45 | 2261 | goto out; |
4c27fe4c | 2262 | |
cb658a45 | 2263 | if (!*pagep) { |
4c27fe4c MR |
2264 | page = shmem_alloc_page(gfp, info, pgoff); |
2265 | if (!page) | |
0f079694 | 2266 | goto out_unacct_blocks; |
4c27fe4c | 2267 | |
8d103963 MR |
2268 | if (!zeropage) { /* mcopy_atomic */ |
2269 | page_kaddr = kmap_atomic(page); | |
2270 | ret = copy_from_user(page_kaddr, | |
2271 | (const void __user *)src_addr, | |
2272 | PAGE_SIZE); | |
2273 | kunmap_atomic(page_kaddr); | |
2274 | ||
2275 | /* fallback to copy_from_user outside mmap_sem */ | |
2276 | if (unlikely(ret)) { | |
2277 | *pagep = page; | |
2278 | shmem_inode_unacct_blocks(inode, 1); | |
2279 | /* don't free the page */ | |
2280 | return -EFAULT; | |
2281 | } | |
2282 | } else { /* mfill_zeropage_atomic */ | |
2283 | clear_highpage(page); | |
4c27fe4c MR |
2284 | } |
2285 | } else { | |
2286 | page = *pagep; | |
2287 | *pagep = NULL; | |
2288 | } | |
2289 | ||
9cc90c66 AA |
2290 | VM_BUG_ON(PageLocked(page) || PageSwapBacked(page)); |
2291 | __SetPageLocked(page); | |
2292 | __SetPageSwapBacked(page); | |
a425d358 | 2293 | __SetPageUptodate(page); |
9cc90c66 | 2294 | |
4c27fe4c MR |
2295 | ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false); |
2296 | if (ret) | |
2297 | goto out_release; | |
2298 | ||
2299 | ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK); | |
2300 | if (!ret) { | |
2301 | ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL); | |
2302 | radix_tree_preload_end(); | |
2303 | } | |
2304 | if (ret) | |
2305 | goto out_release_uncharge; | |
2306 | ||
2307 | mem_cgroup_commit_charge(page, memcg, false, false); | |
2308 | ||
2309 | _dst_pte = mk_pte(page, dst_vma->vm_page_prot); | |
2310 | if (dst_vma->vm_flags & VM_WRITE) | |
2311 | _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte)); | |
2312 | ||
2313 | ret = -EEXIST; | |
2314 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); | |
2315 | if (!pte_none(*dst_pte)) | |
2316 | goto out_release_uncharge_unlock; | |
2317 | ||
4c27fe4c MR |
2318 | lru_cache_add_anon(page); |
2319 | ||
2320 | spin_lock(&info->lock); | |
2321 | info->alloced++; | |
2322 | inode->i_blocks += BLOCKS_PER_PAGE; | |
2323 | shmem_recalc_inode(inode); | |
2324 | spin_unlock(&info->lock); | |
2325 | ||
2326 | inc_mm_counter(dst_mm, mm_counter_file(page)); | |
2327 | page_add_file_rmap(page, false); | |
2328 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
2329 | ||
2330 | /* No need to invalidate - it was non-present before */ | |
2331 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
2332 | unlock_page(page); | |
2333 | pte_unmap_unlock(dst_pte, ptl); | |
2334 | ret = 0; | |
2335 | out: | |
2336 | return ret; | |
2337 | out_release_uncharge_unlock: | |
2338 | pte_unmap_unlock(dst_pte, ptl); | |
2339 | out_release_uncharge: | |
2340 | mem_cgroup_cancel_charge(page, memcg, false); | |
2341 | out_release: | |
9cc90c66 | 2342 | unlock_page(page); |
4c27fe4c | 2343 | put_page(page); |
4c27fe4c | 2344 | out_unacct_blocks: |
0f079694 | 2345 | shmem_inode_unacct_blocks(inode, 1); |
4c27fe4c MR |
2346 | goto out; |
2347 | } | |
2348 | ||
8d103963 MR |
2349 | int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm, |
2350 | pmd_t *dst_pmd, | |
2351 | struct vm_area_struct *dst_vma, | |
2352 | unsigned long dst_addr, | |
2353 | unsigned long src_addr, | |
2354 | struct page **pagep) | |
2355 | { | |
2356 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2357 | dst_addr, src_addr, false, pagep); | |
2358 | } | |
2359 | ||
2360 | int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm, | |
2361 | pmd_t *dst_pmd, | |
2362 | struct vm_area_struct *dst_vma, | |
2363 | unsigned long dst_addr) | |
2364 | { | |
2365 | struct page *page = NULL; | |
2366 | ||
2367 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2368 | dst_addr, 0, true, &page); | |
2369 | } | |
2370 | ||
1da177e4 | 2371 | #ifdef CONFIG_TMPFS |
92e1d5be | 2372 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2373 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2374 | |
6d9d88d0 JS |
2375 | #ifdef CONFIG_TMPFS_XATTR |
2376 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2377 | #else | |
2378 | #define shmem_initxattrs NULL | |
2379 | #endif | |
2380 | ||
1da177e4 | 2381 | static int |
800d15a5 NP |
2382 | shmem_write_begin(struct file *file, struct address_space *mapping, |
2383 | loff_t pos, unsigned len, unsigned flags, | |
2384 | struct page **pagep, void **fsdata) | |
1da177e4 | 2385 | { |
800d15a5 | 2386 | struct inode *inode = mapping->host; |
40e041a2 | 2387 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2388 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
2389 | |
2390 | /* i_mutex is held by caller */ | |
3f472cc9 | 2391 | if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) { |
40e041a2 DH |
2392 | if (info->seals & F_SEAL_WRITE) |
2393 | return -EPERM; | |
2394 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2395 | return -EPERM; | |
2396 | } | |
2397 | ||
9e18eb29 | 2398 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
2399 | } |
2400 | ||
2401 | static int | |
2402 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2403 | loff_t pos, unsigned len, unsigned copied, | |
2404 | struct page *page, void *fsdata) | |
2405 | { | |
2406 | struct inode *inode = mapping->host; | |
2407 | ||
d3602444 HD |
2408 | if (pos + copied > inode->i_size) |
2409 | i_size_write(inode, pos + copied); | |
2410 | ||
ec9516fb | 2411 | if (!PageUptodate(page)) { |
800d8c63 KS |
2412 | struct page *head = compound_head(page); |
2413 | if (PageTransCompound(page)) { | |
2414 | int i; | |
2415 | ||
2416 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2417 | if (head + i == page) | |
2418 | continue; | |
2419 | clear_highpage(head + i); | |
2420 | flush_dcache_page(head + i); | |
2421 | } | |
2422 | } | |
09cbfeaf KS |
2423 | if (copied < PAGE_SIZE) { |
2424 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2425 | zero_user_segments(page, 0, from, |
09cbfeaf | 2426 | from + copied, PAGE_SIZE); |
ec9516fb | 2427 | } |
800d8c63 | 2428 | SetPageUptodate(head); |
ec9516fb | 2429 | } |
800d15a5 | 2430 | set_page_dirty(page); |
6746aff7 | 2431 | unlock_page(page); |
09cbfeaf | 2432 | put_page(page); |
800d15a5 | 2433 | |
800d15a5 | 2434 | return copied; |
1da177e4 LT |
2435 | } |
2436 | ||
2ba5bbed | 2437 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2438 | { |
6e58e79d AV |
2439 | struct file *file = iocb->ki_filp; |
2440 | struct inode *inode = file_inode(file); | |
1da177e4 | 2441 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2442 | pgoff_t index; |
2443 | unsigned long offset; | |
a0ee5ec5 | 2444 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2445 | int error = 0; |
cb66a7a1 | 2446 | ssize_t retval = 0; |
6e58e79d | 2447 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2448 | |
2449 | /* | |
2450 | * Might this read be for a stacking filesystem? Then when reading | |
2451 | * holes of a sparse file, we actually need to allocate those pages, | |
2452 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2453 | */ | |
777eda2c | 2454 | if (!iter_is_iovec(to)) |
75edd345 | 2455 | sgp = SGP_CACHE; |
1da177e4 | 2456 | |
09cbfeaf KS |
2457 | index = *ppos >> PAGE_SHIFT; |
2458 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2459 | |
2460 | for (;;) { | |
2461 | struct page *page = NULL; | |
41ffe5d5 HD |
2462 | pgoff_t end_index; |
2463 | unsigned long nr, ret; | |
1da177e4 LT |
2464 | loff_t i_size = i_size_read(inode); |
2465 | ||
09cbfeaf | 2466 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2467 | if (index > end_index) |
2468 | break; | |
2469 | if (index == end_index) { | |
09cbfeaf | 2470 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2471 | if (nr <= offset) |
2472 | break; | |
2473 | } | |
2474 | ||
9e18eb29 | 2475 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2476 | if (error) { |
2477 | if (error == -EINVAL) | |
2478 | error = 0; | |
1da177e4 LT |
2479 | break; |
2480 | } | |
75edd345 HD |
2481 | if (page) { |
2482 | if (sgp == SGP_CACHE) | |
2483 | set_page_dirty(page); | |
d3602444 | 2484 | unlock_page(page); |
75edd345 | 2485 | } |
1da177e4 LT |
2486 | |
2487 | /* | |
2488 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2489 | * are called without i_mutex protection against truncate |
1da177e4 | 2490 | */ |
09cbfeaf | 2491 | nr = PAGE_SIZE; |
1da177e4 | 2492 | i_size = i_size_read(inode); |
09cbfeaf | 2493 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2494 | if (index == end_index) { |
09cbfeaf | 2495 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2496 | if (nr <= offset) { |
2497 | if (page) | |
09cbfeaf | 2498 | put_page(page); |
1da177e4 LT |
2499 | break; |
2500 | } | |
2501 | } | |
2502 | nr -= offset; | |
2503 | ||
2504 | if (page) { | |
2505 | /* | |
2506 | * If users can be writing to this page using arbitrary | |
2507 | * virtual addresses, take care about potential aliasing | |
2508 | * before reading the page on the kernel side. | |
2509 | */ | |
2510 | if (mapping_writably_mapped(mapping)) | |
2511 | flush_dcache_page(page); | |
2512 | /* | |
2513 | * Mark the page accessed if we read the beginning. | |
2514 | */ | |
2515 | if (!offset) | |
2516 | mark_page_accessed(page); | |
b5810039 | 2517 | } else { |
1da177e4 | 2518 | page = ZERO_PAGE(0); |
09cbfeaf | 2519 | get_page(page); |
b5810039 | 2520 | } |
1da177e4 LT |
2521 | |
2522 | /* | |
2523 | * Ok, we have the page, and it's up-to-date, so | |
2524 | * now we can copy it to user space... | |
1da177e4 | 2525 | */ |
2ba5bbed | 2526 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2527 | retval += ret; |
1da177e4 | 2528 | offset += ret; |
09cbfeaf KS |
2529 | index += offset >> PAGE_SHIFT; |
2530 | offset &= ~PAGE_MASK; | |
1da177e4 | 2531 | |
09cbfeaf | 2532 | put_page(page); |
2ba5bbed | 2533 | if (!iov_iter_count(to)) |
1da177e4 | 2534 | break; |
6e58e79d AV |
2535 | if (ret < nr) { |
2536 | error = -EFAULT; | |
2537 | break; | |
2538 | } | |
1da177e4 LT |
2539 | cond_resched(); |
2540 | } | |
2541 | ||
09cbfeaf | 2542 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2543 | file_accessed(file); |
2544 | return retval ? retval : error; | |
1da177e4 LT |
2545 | } |
2546 | ||
220f2ac9 HD |
2547 | /* |
2548 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
2549 | */ | |
2550 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2551 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2552 | { |
2553 | struct page *page; | |
2554 | struct pagevec pvec; | |
2555 | pgoff_t indices[PAGEVEC_SIZE]; | |
2556 | bool done = false; | |
2557 | int i; | |
2558 | ||
86679820 | 2559 | pagevec_init(&pvec); |
220f2ac9 HD |
2560 | pvec.nr = 1; /* start small: we may be there already */ |
2561 | while (!done) { | |
0cd6144a | 2562 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2563 | pvec.nr, pvec.pages, indices); |
2564 | if (!pvec.nr) { | |
965c8e59 | 2565 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2566 | index = end; |
2567 | break; | |
2568 | } | |
2569 | for (i = 0; i < pvec.nr; i++, index++) { | |
2570 | if (index < indices[i]) { | |
965c8e59 | 2571 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2572 | done = true; |
2573 | break; | |
2574 | } | |
2575 | index = indices[i]; | |
2576 | } | |
2577 | page = pvec.pages[i]; | |
2578 | if (page && !radix_tree_exceptional_entry(page)) { | |
2579 | if (!PageUptodate(page)) | |
2580 | page = NULL; | |
2581 | } | |
2582 | if (index >= end || | |
965c8e59 AM |
2583 | (page && whence == SEEK_DATA) || |
2584 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2585 | done = true; |
2586 | break; | |
2587 | } | |
2588 | } | |
0cd6144a | 2589 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2590 | pagevec_release(&pvec); |
2591 | pvec.nr = PAGEVEC_SIZE; | |
2592 | cond_resched(); | |
2593 | } | |
2594 | return index; | |
2595 | } | |
2596 | ||
965c8e59 | 2597 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2598 | { |
2599 | struct address_space *mapping = file->f_mapping; | |
2600 | struct inode *inode = mapping->host; | |
2601 | pgoff_t start, end; | |
2602 | loff_t new_offset; | |
2603 | ||
965c8e59 AM |
2604 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2605 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2606 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2607 | inode_lock(inode); |
220f2ac9 HD |
2608 | /* We're holding i_mutex so we can access i_size directly */ |
2609 | ||
2610 | if (offset < 0) | |
2611 | offset = -EINVAL; | |
2612 | else if (offset >= inode->i_size) | |
2613 | offset = -ENXIO; | |
2614 | else { | |
09cbfeaf KS |
2615 | start = offset >> PAGE_SHIFT; |
2616 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2617 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2618 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2619 | if (new_offset > offset) { |
2620 | if (new_offset < inode->i_size) | |
2621 | offset = new_offset; | |
965c8e59 | 2622 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2623 | offset = -ENXIO; |
2624 | else | |
2625 | offset = inode->i_size; | |
2626 | } | |
2627 | } | |
2628 | ||
387aae6f HD |
2629 | if (offset >= 0) |
2630 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2631 | inode_unlock(inode); |
220f2ac9 HD |
2632 | return offset; |
2633 | } | |
2634 | ||
05f65b5c DH |
2635 | /* |
2636 | * We need a tag: a new tag would expand every radix_tree_node by 8 bytes, | |
2637 | * so reuse a tag which we firmly believe is never set or cleared on shmem. | |
2638 | */ | |
2639 | #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE | |
2640 | #define LAST_SCAN 4 /* about 150ms max */ | |
2641 | ||
2642 | static void shmem_tag_pins(struct address_space *mapping) | |
2643 | { | |
2644 | struct radix_tree_iter iter; | |
2645 | void **slot; | |
2646 | pgoff_t start; | |
2647 | struct page *page; | |
2648 | ||
2649 | lru_add_drain(); | |
2650 | start = 0; | |
2651 | rcu_read_lock(); | |
2652 | ||
05f65b5c DH |
2653 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
2654 | page = radix_tree_deref_slot(slot); | |
2655 | if (!page || radix_tree_exception(page)) { | |
2cf938aa MW |
2656 | if (radix_tree_deref_retry(page)) { |
2657 | slot = radix_tree_iter_retry(&iter); | |
2658 | continue; | |
2659 | } | |
05f65b5c DH |
2660 | } else if (page_count(page) - page_mapcount(page) > 1) { |
2661 | spin_lock_irq(&mapping->tree_lock); | |
2662 | radix_tree_tag_set(&mapping->page_tree, iter.index, | |
2663 | SHMEM_TAG_PINNED); | |
2664 | spin_unlock_irq(&mapping->tree_lock); | |
2665 | } | |
2666 | ||
2667 | if (need_resched()) { | |
148deab2 | 2668 | slot = radix_tree_iter_resume(slot, &iter); |
05f65b5c | 2669 | cond_resched_rcu(); |
05f65b5c DH |
2670 | } |
2671 | } | |
2672 | rcu_read_unlock(); | |
2673 | } | |
2674 | ||
2675 | /* | |
2676 | * Setting SEAL_WRITE requires us to verify there's no pending writer. However, | |
2677 | * via get_user_pages(), drivers might have some pending I/O without any active | |
2678 | * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages | |
2679 | * and see whether it has an elevated ref-count. If so, we tag them and wait for | |
2680 | * them to be dropped. | |
2681 | * The caller must guarantee that no new user will acquire writable references | |
2682 | * to those pages to avoid races. | |
2683 | */ | |
40e041a2 DH |
2684 | static int shmem_wait_for_pins(struct address_space *mapping) |
2685 | { | |
05f65b5c DH |
2686 | struct radix_tree_iter iter; |
2687 | void **slot; | |
2688 | pgoff_t start; | |
2689 | struct page *page; | |
2690 | int error, scan; | |
2691 | ||
2692 | shmem_tag_pins(mapping); | |
2693 | ||
2694 | error = 0; | |
2695 | for (scan = 0; scan <= LAST_SCAN; scan++) { | |
2696 | if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED)) | |
2697 | break; | |
2698 | ||
2699 | if (!scan) | |
2700 | lru_add_drain_all(); | |
2701 | else if (schedule_timeout_killable((HZ << scan) / 200)) | |
2702 | scan = LAST_SCAN; | |
2703 | ||
2704 | start = 0; | |
2705 | rcu_read_lock(); | |
05f65b5c DH |
2706 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, |
2707 | start, SHMEM_TAG_PINNED) { | |
2708 | ||
2709 | page = radix_tree_deref_slot(slot); | |
2710 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2711 | if (radix_tree_deref_retry(page)) { |
2712 | slot = radix_tree_iter_retry(&iter); | |
2713 | continue; | |
2714 | } | |
05f65b5c DH |
2715 | |
2716 | page = NULL; | |
2717 | } | |
2718 | ||
2719 | if (page && | |
2720 | page_count(page) - page_mapcount(page) != 1) { | |
2721 | if (scan < LAST_SCAN) | |
2722 | goto continue_resched; | |
2723 | ||
2724 | /* | |
2725 | * On the last scan, we clean up all those tags | |
2726 | * we inserted; but make a note that we still | |
2727 | * found pages pinned. | |
2728 | */ | |
2729 | error = -EBUSY; | |
2730 | } | |
2731 | ||
2732 | spin_lock_irq(&mapping->tree_lock); | |
2733 | radix_tree_tag_clear(&mapping->page_tree, | |
2734 | iter.index, SHMEM_TAG_PINNED); | |
2735 | spin_unlock_irq(&mapping->tree_lock); | |
2736 | continue_resched: | |
2737 | if (need_resched()) { | |
148deab2 | 2738 | slot = radix_tree_iter_resume(slot, &iter); |
05f65b5c | 2739 | cond_resched_rcu(); |
05f65b5c DH |
2740 | } |
2741 | } | |
2742 | rcu_read_unlock(); | |
2743 | } | |
2744 | ||
2745 | return error; | |
40e041a2 DH |
2746 | } |
2747 | ||
2748 | #define F_ALL_SEALS (F_SEAL_SEAL | \ | |
2749 | F_SEAL_SHRINK | \ | |
2750 | F_SEAL_GROW | \ | |
2751 | F_SEAL_WRITE) | |
2752 | ||
2753 | int shmem_add_seals(struct file *file, unsigned int seals) | |
2754 | { | |
2755 | struct inode *inode = file_inode(file); | |
2756 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2757 | int error; | |
2758 | ||
2759 | /* | |
2760 | * SEALING | |
2761 | * Sealing allows multiple parties to share a shmem-file but restrict | |
2762 | * access to a specific subset of file operations. Seals can only be | |
2763 | * added, but never removed. This way, mutually untrusted parties can | |
2764 | * share common memory regions with a well-defined policy. A malicious | |
2765 | * peer can thus never perform unwanted operations on a shared object. | |
2766 | * | |
2767 | * Seals are only supported on special shmem-files and always affect | |
2768 | * the whole underlying inode. Once a seal is set, it may prevent some | |
2769 | * kinds of access to the file. Currently, the following seals are | |
2770 | * defined: | |
2771 | * SEAL_SEAL: Prevent further seals from being set on this file | |
2772 | * SEAL_SHRINK: Prevent the file from shrinking | |
2773 | * SEAL_GROW: Prevent the file from growing | |
2774 | * SEAL_WRITE: Prevent write access to the file | |
2775 | * | |
2776 | * As we don't require any trust relationship between two parties, we | |
2777 | * must prevent seals from being removed. Therefore, sealing a file | |
2778 | * only adds a given set of seals to the file, it never touches | |
2779 | * existing seals. Furthermore, the "setting seals"-operation can be | |
2780 | * sealed itself, which basically prevents any further seal from being | |
2781 | * added. | |
2782 | * | |
2783 | * Semantics of sealing are only defined on volatile files. Only | |
2784 | * anonymous shmem files support sealing. More importantly, seals are | |
2785 | * never written to disk. Therefore, there's no plan to support it on | |
2786 | * other file types. | |
2787 | */ | |
2788 | ||
2789 | if (file->f_op != &shmem_file_operations) | |
2790 | return -EINVAL; | |
2791 | if (!(file->f_mode & FMODE_WRITE)) | |
2792 | return -EPERM; | |
2793 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
2794 | return -EINVAL; | |
2795 | ||
5955102c | 2796 | inode_lock(inode); |
40e041a2 DH |
2797 | |
2798 | if (info->seals & F_SEAL_SEAL) { | |
2799 | error = -EPERM; | |
2800 | goto unlock; | |
2801 | } | |
2802 | ||
2803 | if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) { | |
2804 | error = mapping_deny_writable(file->f_mapping); | |
2805 | if (error) | |
2806 | goto unlock; | |
2807 | ||
2808 | error = shmem_wait_for_pins(file->f_mapping); | |
2809 | if (error) { | |
2810 | mapping_allow_writable(file->f_mapping); | |
2811 | goto unlock; | |
2812 | } | |
2813 | } | |
2814 | ||
2815 | info->seals |= seals; | |
2816 | error = 0; | |
2817 | ||
2818 | unlock: | |
5955102c | 2819 | inode_unlock(inode); |
40e041a2 DH |
2820 | return error; |
2821 | } | |
2822 | EXPORT_SYMBOL_GPL(shmem_add_seals); | |
2823 | ||
2824 | int shmem_get_seals(struct file *file) | |
2825 | { | |
2826 | if (file->f_op != &shmem_file_operations) | |
2827 | return -EINVAL; | |
2828 | ||
2829 | return SHMEM_I(file_inode(file))->seals; | |
2830 | } | |
2831 | EXPORT_SYMBOL_GPL(shmem_get_seals); | |
2832 | ||
2833 | long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg) | |
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 | ||
2843 | error = shmem_add_seals(file, arg); | |
2844 | break; | |
2845 | case F_GET_SEALS: | |
2846 | error = shmem_get_seals(file); | |
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; | |
0006ebb4 | 3428 | __u64 inum = fh[1]; |
91828a40 DG |
3429 | return ino->i_ino == inum && fh[0] == ino->i_generation; |
3430 | } | |
3431 | ||
480b116c CH |
3432 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3433 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3434 | { |
91828a40 | 3435 | struct inode *inode; |
480b116c | 3436 | struct dentry *dentry = NULL; |
35c2a7f4 | 3437 | u64 inum; |
480b116c | 3438 | |
0006ebb4 | 3439 | if (fh_len < 2) |
480b116c | 3440 | return NULL; |
91828a40 | 3441 | |
0006ebb4 SF |
3442 | inum = fid->raw[1]; |
3443 | inode = ilookup5(sb, inum, shmem_match, fid->raw); | |
91828a40 | 3444 | if (inode) { |
480b116c | 3445 | dentry = d_find_alias(inode); |
91828a40 DG |
3446 | iput(inode); |
3447 | } | |
3448 | ||
480b116c | 3449 | return dentry; |
91828a40 DG |
3450 | } |
3451 | ||
b0b0382b AV |
3452 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3453 | struct inode *parent) | |
91828a40 | 3454 | { |
0006ebb4 SF |
3455 | if (*len < 2) { |
3456 | *len = 2; | |
94e07a75 | 3457 | return FILEID_INVALID; |
5fe0c237 | 3458 | } |
91828a40 | 3459 | |
91828a40 DG |
3460 | fh[0] = inode->i_generation; |
3461 | fh[1] = inode->i_ino; | |
91828a40 | 3462 | |
0006ebb4 | 3463 | *len = 2; |
91828a40 DG |
3464 | return 1; |
3465 | } | |
3466 | ||
39655164 | 3467 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3468 | .get_parent = shmem_get_parent, |
91828a40 | 3469 | .encode_fh = shmem_encode_fh, |
480b116c | 3470 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3471 | }; |
3472 | ||
680d794b AM |
3473 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
3474 | bool remount) | |
1da177e4 LT |
3475 | { |
3476 | char *this_char, *value, *rest; | |
49cd0a5c | 3477 | struct mempolicy *mpol = NULL; |
8751e039 EB |
3478 | uid_t uid; |
3479 | gid_t gid; | |
1da177e4 | 3480 | |
b00dc3ad HD |
3481 | while (options != NULL) { |
3482 | this_char = options; | |
3483 | for (;;) { | |
3484 | /* | |
3485 | * NUL-terminate this option: unfortunately, | |
3486 | * mount options form a comma-separated list, | |
3487 | * but mpol's nodelist may also contain commas. | |
3488 | */ | |
3489 | options = strchr(options, ','); | |
3490 | if (options == NULL) | |
3491 | break; | |
3492 | options++; | |
3493 | if (!isdigit(*options)) { | |
3494 | options[-1] = '\0'; | |
3495 | break; | |
3496 | } | |
3497 | } | |
1da177e4 LT |
3498 | if (!*this_char) |
3499 | continue; | |
3500 | if ((value = strchr(this_char,'=')) != NULL) { | |
3501 | *value++ = 0; | |
3502 | } else { | |
1170532b JP |
3503 | pr_err("tmpfs: No value for mount option '%s'\n", |
3504 | this_char); | |
49cd0a5c | 3505 | goto error; |
1da177e4 LT |
3506 | } |
3507 | ||
3508 | if (!strcmp(this_char,"size")) { | |
3509 | unsigned long long size; | |
3510 | size = memparse(value,&rest); | |
3511 | if (*rest == '%') { | |
3512 | size <<= PAGE_SHIFT; | |
3513 | size *= totalram_pages; | |
3514 | do_div(size, 100); | |
3515 | rest++; | |
3516 | } | |
3517 | if (*rest) | |
3518 | goto bad_val; | |
680d794b | 3519 | sbinfo->max_blocks = |
09cbfeaf | 3520 | DIV_ROUND_UP(size, PAGE_SIZE); |
1da177e4 | 3521 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 3522 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
3523 | if (*rest) |
3524 | goto bad_val; | |
3525 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 3526 | sbinfo->max_inodes = memparse(value, &rest); |
0006ebb4 | 3527 | if (*rest || sbinfo->max_inodes < 2) |
1da177e4 LT |
3528 | goto bad_val; |
3529 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 3530 | if (remount) |
1da177e4 | 3531 | continue; |
680d794b | 3532 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
3533 | if (*rest) |
3534 | goto bad_val; | |
3535 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 3536 | if (remount) |
1da177e4 | 3537 | continue; |
8751e039 | 3538 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3539 | if (*rest) |
3540 | goto bad_val; | |
8751e039 EB |
3541 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
3542 | if (!uid_valid(sbinfo->uid)) | |
3543 | goto bad_val; | |
1da177e4 | 3544 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 3545 | if (remount) |
1da177e4 | 3546 | continue; |
8751e039 | 3547 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3548 | if (*rest) |
3549 | goto bad_val; | |
8751e039 EB |
3550 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
3551 | if (!gid_valid(sbinfo->gid)) | |
3552 | goto bad_val; | |
e496cf3d | 3553 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3554 | } else if (!strcmp(this_char, "huge")) { |
3555 | int huge; | |
3556 | huge = shmem_parse_huge(value); | |
3557 | if (huge < 0) | |
3558 | goto bad_val; | |
3559 | if (!has_transparent_hugepage() && | |
3560 | huge != SHMEM_HUGE_NEVER) | |
3561 | goto bad_val; | |
3562 | sbinfo->huge = huge; | |
3563 | #endif | |
3564 | #ifdef CONFIG_NUMA | |
7339ff83 | 3565 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
3566 | mpol_put(mpol); |
3567 | mpol = NULL; | |
3568 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 3569 | goto bad_val; |
5a6e75f8 | 3570 | #endif |
1da177e4 | 3571 | } else { |
1170532b | 3572 | pr_err("tmpfs: Bad mount option %s\n", this_char); |
49cd0a5c | 3573 | goto error; |
1da177e4 LT |
3574 | } |
3575 | } | |
49cd0a5c | 3576 | sbinfo->mpol = mpol; |
1da177e4 LT |
3577 | return 0; |
3578 | ||
3579 | bad_val: | |
1170532b | 3580 | pr_err("tmpfs: Bad value '%s' for mount option '%s'\n", |
1da177e4 | 3581 | value, this_char); |
49cd0a5c GT |
3582 | error: |
3583 | mpol_put(mpol); | |
1da177e4 LT |
3584 | return 1; |
3585 | ||
3586 | } | |
3587 | ||
3588 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
3589 | { | |
3590 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 3591 | struct shmem_sb_info config = *sbinfo; |
0006ebb4 | 3592 | int inodes; |
0edd73b3 HD |
3593 | int error = -EINVAL; |
3594 | ||
5f00110f | 3595 | config.mpol = NULL; |
680d794b | 3596 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 3597 | return error; |
1da177e4 | 3598 | |
0edd73b3 | 3599 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3600 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 3601 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 3602 | goto out; |
680d794b | 3603 | if (config.max_inodes < inodes) |
0edd73b3 HD |
3604 | goto out; |
3605 | /* | |
54af6042 | 3606 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
3607 | * but we must separately disallow unlimited->limited, because |
3608 | * in that case we have no record of how much is already in use. | |
3609 | */ | |
680d794b | 3610 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 3611 | goto out; |
680d794b | 3612 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
3613 | goto out; |
3614 | ||
3615 | error = 0; | |
5a6e75f8 | 3616 | sbinfo->huge = config.huge; |
680d794b | 3617 | sbinfo->max_blocks = config.max_blocks; |
680d794b AM |
3618 | sbinfo->max_inodes = config.max_inodes; |
3619 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 3620 | |
5f00110f GT |
3621 | /* |
3622 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3623 | */ | |
3624 | if (config.mpol) { | |
3625 | mpol_put(sbinfo->mpol); | |
3626 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
3627 | } | |
0edd73b3 HD |
3628 | out: |
3629 | spin_unlock(&sbinfo->stat_lock); | |
3630 | return error; | |
1da177e4 | 3631 | } |
680d794b | 3632 | |
34c80b1d | 3633 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3634 | { |
34c80b1d | 3635 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b AM |
3636 | |
3637 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3638 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3639 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 3640 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
0006ebb4 | 3641 | seq_printf(seq, ",nr_inodes=%d", sbinfo->max_inodes); |
680d794b | 3642 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) |
09208d15 | 3643 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3644 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3645 | seq_printf(seq, ",uid=%u", | |
3646 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3647 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3648 | seq_printf(seq, ",gid=%u", | |
3649 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
e496cf3d | 3650 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3651 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
3652 | if (sbinfo->huge) | |
3653 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3654 | #endif | |
71fe804b | 3655 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b AM |
3656 | return 0; |
3657 | } | |
9183df25 DH |
3658 | |
3659 | #define MFD_NAME_PREFIX "memfd:" | |
3660 | #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1) | |
3661 | #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN) | |
3662 | ||
749df87b | 3663 | #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB) |
9183df25 DH |
3664 | |
3665 | SYSCALL_DEFINE2(memfd_create, | |
3666 | const char __user *, uname, | |
3667 | unsigned int, flags) | |
3668 | { | |
3669 | struct shmem_inode_info *info; | |
3670 | struct file *file; | |
3671 | int fd, error; | |
3672 | char *name; | |
3673 | long len; | |
3674 | ||
749df87b MK |
3675 | if (!(flags & MFD_HUGETLB)) { |
3676 | if (flags & ~(unsigned int)MFD_ALL_FLAGS) | |
3677 | return -EINVAL; | |
3678 | } else { | |
3679 | /* Sealing not supported in hugetlbfs (MFD_HUGETLB) */ | |
3680 | if (flags & MFD_ALLOW_SEALING) | |
3681 | return -EINVAL; | |
3682 | /* Allow huge page size encoding in flags. */ | |
3683 | if (flags & ~(unsigned int)(MFD_ALL_FLAGS | | |
3684 | (MFD_HUGE_MASK << MFD_HUGE_SHIFT))) | |
3685 | return -EINVAL; | |
3686 | } | |
9183df25 DH |
3687 | |
3688 | /* length includes terminating zero */ | |
3689 | len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1); | |
3690 | if (len <= 0) | |
3691 | return -EFAULT; | |
3692 | if (len > MFD_NAME_MAX_LEN + 1) | |
3693 | return -EINVAL; | |
3694 | ||
0ee931c4 | 3695 | name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_KERNEL); |
9183df25 DH |
3696 | if (!name) |
3697 | return -ENOMEM; | |
3698 | ||
3699 | strcpy(name, MFD_NAME_PREFIX); | |
3700 | if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) { | |
3701 | error = -EFAULT; | |
3702 | goto err_name; | |
3703 | } | |
3704 | ||
3705 | /* terminating-zero may have changed after strnlen_user() returned */ | |
3706 | if (name[len + MFD_NAME_PREFIX_LEN - 1]) { | |
3707 | error = -EFAULT; | |
3708 | goto err_name; | |
3709 | } | |
3710 | ||
3711 | fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0); | |
3712 | if (fd < 0) { | |
3713 | error = fd; | |
3714 | goto err_name; | |
3715 | } | |
3716 | ||
749df87b MK |
3717 | if (flags & MFD_HUGETLB) { |
3718 | struct user_struct *user = NULL; | |
3719 | ||
3720 | file = hugetlb_file_setup(name, 0, VM_NORESERVE, &user, | |
3721 | HUGETLB_ANONHUGE_INODE, | |
3722 | (flags >> MFD_HUGE_SHIFT) & | |
3723 | MFD_HUGE_MASK); | |
3724 | } else | |
3725 | file = shmem_file_setup(name, 0, VM_NORESERVE); | |
9183df25 DH |
3726 | if (IS_ERR(file)) { |
3727 | error = PTR_ERR(file); | |
3728 | goto err_fd; | |
3729 | } | |
9183df25 DH |
3730 | file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; |
3731 | file->f_flags |= O_RDWR | O_LARGEFILE; | |
749df87b MK |
3732 | |
3733 | if (flags & MFD_ALLOW_SEALING) { | |
3734 | /* | |
3735 | * flags check at beginning of function ensures | |
3736 | * this is not a hugetlbfs (MFD_HUGETLB) file. | |
3737 | */ | |
3738 | info = SHMEM_I(file_inode(file)); | |
9183df25 | 3739 | info->seals &= ~F_SEAL_SEAL; |
749df87b | 3740 | } |
9183df25 DH |
3741 | |
3742 | fd_install(fd, file); | |
3743 | kfree(name); | |
3744 | return fd; | |
3745 | ||
3746 | err_fd: | |
3747 | put_unused_fd(fd); | |
3748 | err_name: | |
3749 | kfree(name); | |
3750 | return error; | |
3751 | } | |
3752 | ||
680d794b | 3753 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3754 | |
3755 | static void shmem_put_super(struct super_block *sb) | |
3756 | { | |
602586a8 HD |
3757 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3758 | ||
0006ebb4 SF |
3759 | if (!sbinfo->idr_nouse) |
3760 | idr_destroy(&sbinfo->idr); | |
602586a8 | 3761 | percpu_counter_destroy(&sbinfo->used_blocks); |
49cd0a5c | 3762 | mpol_put(sbinfo->mpol); |
602586a8 | 3763 | kfree(sbinfo); |
1da177e4 LT |
3764 | sb->s_fs_info = NULL; |
3765 | } | |
3766 | ||
2b2af54a | 3767 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3768 | { |
3769 | struct inode *inode; | |
0edd73b3 | 3770 | struct shmem_sb_info *sbinfo; |
680d794b AM |
3771 | int err = -ENOMEM; |
3772 | ||
3773 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3774 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b AM |
3775 | L1_CACHE_BYTES), GFP_KERNEL); |
3776 | if (!sbinfo) | |
3777 | return -ENOMEM; | |
3778 | ||
0006ebb4 SF |
3779 | mutex_init(&sbinfo->idr_lock); |
3780 | idr_init(&sbinfo->idr); | |
680d794b | 3781 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
3782 | sbinfo->uid = current_fsuid(); |
3783 | sbinfo->gid = current_fsgid(); | |
680d794b | 3784 | sb->s_fs_info = sbinfo; |
1da177e4 | 3785 | |
0edd73b3 | 3786 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3787 | /* |
3788 | * Per default we only allow half of the physical ram per | |
3789 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3790 | * but the internal instance is left unlimited. | |
3791 | */ | |
1751e8a6 | 3792 | if (!(sb->s_flags & SB_KERNMOUNT)) { |
680d794b AM |
3793 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3794 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3795 | if (shmem_parse_options(data, sbinfo, false)) { | |
3796 | err = -EINVAL; | |
3797 | goto failed; | |
3798 | } | |
ca4e0519 | 3799 | } else { |
1751e8a6 | 3800 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 3801 | } |
91828a40 | 3802 | sb->s_export_op = &shmem_export_ops; |
1751e8a6 | 3803 | sb->s_flags |= SB_NOSEC; |
1da177e4 | 3804 | #else |
1751e8a6 | 3805 | sb->s_flags |= SB_NOUSER; |
1da177e4 LT |
3806 | #endif |
3807 | ||
0edd73b3 | 3808 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3809 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3810 | goto failed; |
680d794b | 3811 | sbinfo->free_inodes = sbinfo->max_inodes; |
779750d2 KS |
3812 | spin_lock_init(&sbinfo->shrinklist_lock); |
3813 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 3814 | |
285b2c4f | 3815 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3816 | sb->s_blocksize = PAGE_SIZE; |
3817 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3818 | sb->s_magic = TMPFS_MAGIC; |
3819 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3820 | sb->s_time_gran = 1; |
b09e0fa4 | 3821 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3822 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3823 | #endif |
3824 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
1751e8a6 | 3825 | sb->s_flags |= SB_POSIXACL; |
39f0247d | 3826 | #endif |
2b4db796 | 3827 | uuid_gen(&sb->s_uuid); |
0edd73b3 | 3828 | |
454abafe | 3829 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3830 | if (!inode) |
3831 | goto failed; | |
680d794b AM |
3832 | inode->i_uid = sbinfo->uid; |
3833 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3834 | sb->s_root = d_make_root(inode); |
3835 | if (!sb->s_root) | |
48fde701 | 3836 | goto failed; |
1da177e4 LT |
3837 | return 0; |
3838 | ||
1da177e4 LT |
3839 | failed: |
3840 | shmem_put_super(sb); | |
3841 | return err; | |
3842 | } | |
3843 | ||
fcc234f8 | 3844 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3845 | |
3846 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3847 | { | |
41ffe5d5 HD |
3848 | struct shmem_inode_info *info; |
3849 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3850 | if (!info) | |
1da177e4 | 3851 | return NULL; |
41ffe5d5 | 3852 | return &info->vfs_inode; |
1da177e4 LT |
3853 | } |
3854 | ||
41ffe5d5 | 3855 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3856 | { |
3857 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84e710da AV |
3858 | if (S_ISLNK(inode->i_mode)) |
3859 | kfree(inode->i_link); | |
fa0d7e3d NP |
3860 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3861 | } | |
3862 | ||
1da177e4 LT |
3863 | static void shmem_destroy_inode(struct inode *inode) |
3864 | { | |
09208d15 | 3865 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3866 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3867 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3868 | } |
3869 | ||
41ffe5d5 | 3870 | static void shmem_init_inode(void *foo) |
1da177e4 | 3871 | { |
41ffe5d5 HD |
3872 | struct shmem_inode_info *info = foo; |
3873 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3874 | } |
3875 | ||
9a8ec03e | 3876 | static void shmem_init_inodecache(void) |
1da177e4 LT |
3877 | { |
3878 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3879 | sizeof(struct shmem_inode_info), | |
5d097056 | 3880 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3881 | } |
3882 | ||
41ffe5d5 | 3883 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3884 | { |
1a1d92c1 | 3885 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3886 | } |
3887 | ||
0006ebb4 SF |
3888 | static __init void shmem_no_idr(struct super_block *sb) |
3889 | { | |
3890 | struct shmem_sb_info *sbinfo; | |
3891 | ||
3892 | sbinfo = SHMEM_SB(sb); | |
3893 | sbinfo->idr_nouse = true; | |
3894 | idr_destroy(&sbinfo->idr); | |
3895 | } | |
3896 | ||
f5e54d6e | 3897 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3898 | .writepage = shmem_writepage, |
76719325 | 3899 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3900 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3901 | .write_begin = shmem_write_begin, |
3902 | .write_end = shmem_write_end, | |
1da177e4 | 3903 | #endif |
1c93923c | 3904 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3905 | .migratepage = migrate_page, |
1c93923c | 3906 | #endif |
aa261f54 | 3907 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3908 | }; |
3909 | ||
15ad7cdc | 3910 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3911 | .mmap = shmem_mmap, |
c01d5b30 | 3912 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3913 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3914 | .llseek = shmem_file_llseek, |
2ba5bbed | 3915 | .read_iter = shmem_file_read_iter, |
8174202b | 3916 | .write_iter = generic_file_write_iter, |
1b061d92 | 3917 | .fsync = noop_fsync, |
82c156f8 | 3918 | .splice_read = generic_file_splice_read, |
f6cb85d0 | 3919 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3920 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3921 | #endif |
3922 | }; | |
3923 | ||
92e1d5be | 3924 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3925 | .getattr = shmem_getattr, |
94c1e62d | 3926 | .setattr = shmem_setattr, |
b09e0fa4 | 3927 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3928 | .listxattr = shmem_listxattr, |
feda821e | 3929 | .set_acl = simple_set_acl, |
b09e0fa4 | 3930 | #endif |
1da177e4 LT |
3931 | }; |
3932 | ||
92e1d5be | 3933 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3934 | #ifdef CONFIG_TMPFS |
3935 | .create = shmem_create, | |
3936 | .lookup = simple_lookup, | |
3937 | .link = shmem_link, | |
3938 | .unlink = shmem_unlink, | |
3939 | .symlink = shmem_symlink, | |
3940 | .mkdir = shmem_mkdir, | |
3941 | .rmdir = shmem_rmdir, | |
3942 | .mknod = shmem_mknod, | |
2773bf00 | 3943 | .rename = shmem_rename2, |
60545d0d | 3944 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3945 | #endif |
b09e0fa4 | 3946 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3947 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3948 | #endif |
39f0247d | 3949 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3950 | .setattr = shmem_setattr, |
feda821e | 3951 | .set_acl = simple_set_acl, |
39f0247d AG |
3952 | #endif |
3953 | }; | |
3954 | ||
92e1d5be | 3955 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3956 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3957 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3958 | #endif |
39f0247d | 3959 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3960 | .setattr = shmem_setattr, |
feda821e | 3961 | .set_acl = simple_set_acl, |
39f0247d | 3962 | #endif |
1da177e4 LT |
3963 | }; |
3964 | ||
759b9775 | 3965 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
3966 | .alloc_inode = shmem_alloc_inode, |
3967 | .destroy_inode = shmem_destroy_inode, | |
3968 | #ifdef CONFIG_TMPFS | |
3969 | .statfs = shmem_statfs, | |
3970 | .remount_fs = shmem_remount_fs, | |
680d794b | 3971 | .show_options = shmem_show_options, |
1da177e4 | 3972 | #endif |
1f895f75 | 3973 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3974 | .drop_inode = generic_delete_inode, |
3975 | .put_super = shmem_put_super, | |
779750d2 KS |
3976 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
3977 | .nr_cached_objects = shmem_unused_huge_count, | |
3978 | .free_cached_objects = shmem_unused_huge_scan, | |
3979 | #endif | |
1da177e4 LT |
3980 | }; |
3981 | ||
f0f37e2f | 3982 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3983 | .fault = shmem_fault, |
d7c17551 | 3984 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3985 | #ifdef CONFIG_NUMA |
3986 | .set_policy = shmem_set_policy, | |
3987 | .get_policy = shmem_get_policy, | |
3988 | #endif | |
3989 | }; | |
3990 | ||
3c26ff6e AV |
3991 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
3992 | int flags, const char *dev_name, void *data) | |
1da177e4 | 3993 | { |
3c26ff6e | 3994 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
3995 | } |
3996 | ||
41ffe5d5 | 3997 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3998 | .owner = THIS_MODULE, |
3999 | .name = "tmpfs", | |
3c26ff6e | 4000 | .mount = shmem_mount, |
1da177e4 | 4001 | .kill_sb = kill_litter_super, |
2b8576cb | 4002 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 4003 | }; |
1da177e4 | 4004 | |
41ffe5d5 | 4005 | int __init shmem_init(void) |
1da177e4 LT |
4006 | { |
4007 | int error; | |
4008 | ||
16203a7a RL |
4009 | /* If rootfs called this, don't re-init */ |
4010 | if (shmem_inode_cachep) | |
4011 | return 0; | |
4012 | ||
9a8ec03e | 4013 | shmem_init_inodecache(); |
1da177e4 | 4014 | |
41ffe5d5 | 4015 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 4016 | if (error) { |
1170532b | 4017 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
4018 | goto out2; |
4019 | } | |
95dc112a | 4020 | |
ca4e0519 | 4021 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
4022 | if (IS_ERR(shm_mnt)) { |
4023 | error = PTR_ERR(shm_mnt); | |
1170532b | 4024 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
4025 | goto out1; |
4026 | } | |
0006ebb4 | 4027 | shmem_no_idr(shm_mnt->mnt_sb); |
5a6e75f8 | 4028 | |
e496cf3d | 4029 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
435c0b87 | 4030 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4031 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4032 | else | |
4033 | shmem_huge = 0; /* just in case it was patched */ | |
4034 | #endif | |
1da177e4 LT |
4035 | return 0; |
4036 | ||
4037 | out1: | |
41ffe5d5 | 4038 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 4039 | out2: |
41ffe5d5 | 4040 | shmem_destroy_inodecache(); |
1da177e4 LT |
4041 | shm_mnt = ERR_PTR(error); |
4042 | return error; | |
4043 | } | |
853ac43a | 4044 | |
e496cf3d | 4045 | #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS) |
5a6e75f8 KS |
4046 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
4047 | struct kobj_attribute *attr, char *buf) | |
4048 | { | |
4049 | int values[] = { | |
4050 | SHMEM_HUGE_ALWAYS, | |
4051 | SHMEM_HUGE_WITHIN_SIZE, | |
4052 | SHMEM_HUGE_ADVISE, | |
4053 | SHMEM_HUGE_NEVER, | |
4054 | SHMEM_HUGE_DENY, | |
4055 | SHMEM_HUGE_FORCE, | |
4056 | }; | |
4057 | int i, count; | |
4058 | ||
4059 | for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) { | |
4060 | const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s "; | |
4061 | ||
4062 | count += sprintf(buf + count, fmt, | |
4063 | shmem_format_huge(values[i])); | |
4064 | } | |
4065 | buf[count - 1] = '\n'; | |
4066 | return count; | |
4067 | } | |
4068 | ||
4069 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
4070 | struct kobj_attribute *attr, const char *buf, size_t count) | |
4071 | { | |
4072 | char tmp[16]; | |
4073 | int huge; | |
4074 | ||
4075 | if (count + 1 > sizeof(tmp)) | |
4076 | return -EINVAL; | |
4077 | memcpy(tmp, buf, count); | |
4078 | tmp[count] = '\0'; | |
4079 | if (count && tmp[count - 1] == '\n') | |
4080 | tmp[count - 1] = '\0'; | |
4081 | ||
4082 | huge = shmem_parse_huge(tmp); | |
4083 | if (huge == -EINVAL) | |
4084 | return -EINVAL; | |
4085 | if (!has_transparent_hugepage() && | |
4086 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
4087 | return -EINVAL; | |
4088 | ||
4089 | shmem_huge = huge; | |
435c0b87 | 4090 | if (shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4091 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4092 | return count; | |
4093 | } | |
4094 | ||
4095 | struct kobj_attribute shmem_enabled_attr = | |
4096 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
3b33719c | 4097 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */ |
f3f0e1d2 | 4098 | |
3b33719c | 4099 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
f3f0e1d2 KS |
4100 | bool shmem_huge_enabled(struct vm_area_struct *vma) |
4101 | { | |
4102 | struct inode *inode = file_inode(vma->vm_file); | |
4103 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4104 | loff_t i_size; | |
4105 | pgoff_t off; | |
4106 | ||
4107 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
4108 | return true; | |
4109 | if (shmem_huge == SHMEM_HUGE_DENY) | |
4110 | return false; | |
4111 | switch (sbinfo->huge) { | |
4112 | case SHMEM_HUGE_NEVER: | |
4113 | return false; | |
4114 | case SHMEM_HUGE_ALWAYS: | |
4115 | return true; | |
4116 | case SHMEM_HUGE_WITHIN_SIZE: | |
4117 | off = round_up(vma->vm_pgoff, HPAGE_PMD_NR); | |
4118 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
4119 | if (i_size >= HPAGE_PMD_SIZE && | |
4120 | i_size >> PAGE_SHIFT >= off) | |
4121 | return true; | |
c8402871 | 4122 | /* fall through */ |
f3f0e1d2 KS |
4123 | case SHMEM_HUGE_ADVISE: |
4124 | /* TODO: implement fadvise() hints */ | |
4125 | return (vma->vm_flags & VM_HUGEPAGE); | |
4126 | default: | |
4127 | VM_BUG_ON(1); | |
4128 | return false; | |
4129 | } | |
4130 | } | |
3b33719c | 4131 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 4132 | |
853ac43a MM |
4133 | #else /* !CONFIG_SHMEM */ |
4134 | ||
4135 | /* | |
4136 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
4137 | * | |
4138 | * This is intended for small system where the benefits of the full | |
4139 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4140 | * their complexity. On systems without swap this code should be | |
4141 | * effectively equivalent, but much lighter weight. | |
4142 | */ | |
4143 | ||
41ffe5d5 | 4144 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4145 | .name = "tmpfs", |
3c26ff6e | 4146 | .mount = ramfs_mount, |
853ac43a | 4147 | .kill_sb = kill_litter_super, |
2b8576cb | 4148 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4149 | }; |
4150 | ||
41ffe5d5 | 4151 | int __init shmem_init(void) |
853ac43a | 4152 | { |
41ffe5d5 | 4153 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4154 | |
41ffe5d5 | 4155 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
4156 | BUG_ON(IS_ERR(shm_mnt)); |
4157 | ||
4158 | return 0; | |
4159 | } | |
4160 | ||
41ffe5d5 | 4161 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
4162 | { |
4163 | return 0; | |
4164 | } | |
4165 | ||
3f96b79a HD |
4166 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
4167 | { | |
4168 | return 0; | |
4169 | } | |
4170 | ||
24513264 HD |
4171 | void shmem_unlock_mapping(struct address_space *mapping) |
4172 | { | |
4173 | } | |
4174 | ||
c01d5b30 HD |
4175 | #ifdef CONFIG_MMU |
4176 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4177 | unsigned long addr, unsigned long len, | |
4178 | unsigned long pgoff, unsigned long flags) | |
4179 | { | |
4180 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4181 | } | |
4182 | #endif | |
4183 | ||
41ffe5d5 | 4184 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4185 | { |
41ffe5d5 | 4186 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4187 | } |
4188 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4189 | ||
0b0a0806 HD |
4190 | #define shmem_vm_ops generic_file_vm_ops |
4191 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 4192 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
4193 | #define shmem_acct_size(flags, size) 0 |
4194 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
4195 | |
4196 | #endif /* CONFIG_SHMEM */ | |
4197 | ||
4198 | /* common code */ | |
1da177e4 | 4199 | |
19938e35 | 4200 | static const struct dentry_operations anon_ops = { |
118b2302 | 4201 | .d_dname = simple_dname |
3451538a AV |
4202 | }; |
4203 | ||
703321b6 | 4204 | static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size, |
c7277090 | 4205 | unsigned long flags, unsigned int i_flags) |
1da177e4 | 4206 | { |
6b4d0b27 | 4207 | struct file *res; |
1da177e4 | 4208 | struct inode *inode; |
2c48b9c4 | 4209 | struct path path; |
3451538a | 4210 | struct super_block *sb; |
1da177e4 LT |
4211 | struct qstr this; |
4212 | ||
703321b6 MA |
4213 | if (IS_ERR(mnt)) |
4214 | return ERR_CAST(mnt); | |
1da177e4 | 4215 | |
285b2c4f | 4216 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4217 | return ERR_PTR(-EINVAL); |
4218 | ||
4219 | if (shmem_acct_size(flags, size)) | |
4220 | return ERR_PTR(-ENOMEM); | |
4221 | ||
6b4d0b27 | 4222 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
4223 | this.name = name; |
4224 | this.len = strlen(name); | |
4225 | this.hash = 0; /* will go */ | |
703321b6 MA |
4226 | sb = mnt->mnt_sb; |
4227 | path.mnt = mntget(mnt); | |
3451538a | 4228 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 4229 | if (!path.dentry) |
1da177e4 | 4230 | goto put_memory; |
3451538a | 4231 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 4232 | |
6b4d0b27 | 4233 | res = ERR_PTR(-ENOSPC); |
3451538a | 4234 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 4235 | if (!inode) |
66ee4b88 | 4236 | goto put_memory; |
1da177e4 | 4237 | |
c7277090 | 4238 | inode->i_flags |= i_flags; |
2c48b9c4 | 4239 | d_instantiate(path.dentry, inode); |
1da177e4 | 4240 | inode->i_size = size; |
6d6b77f1 | 4241 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
4242 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
4243 | if (IS_ERR(res)) | |
66ee4b88 | 4244 | goto put_path; |
4b42af81 | 4245 | |
6b4d0b27 | 4246 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 4247 | &shmem_file_operations); |
6b4d0b27 | 4248 | if (IS_ERR(res)) |
66ee4b88 | 4249 | goto put_path; |
4b42af81 | 4250 | |
6b4d0b27 | 4251 | return res; |
1da177e4 | 4252 | |
1da177e4 LT |
4253 | put_memory: |
4254 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
4255 | put_path: |
4256 | path_put(&path); | |
6b4d0b27 | 4257 | return res; |
1da177e4 | 4258 | } |
c7277090 EP |
4259 | |
4260 | /** | |
4261 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4262 | * kernel internal. There will be NO LSM permission checks against the | |
4263 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4264 | * higher layer. The users are the big_key and shm implementations. LSM |
4265 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4266 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4267 | * @size: size to be set for the file | |
4268 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4269 | */ | |
4270 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4271 | { | |
703321b6 | 4272 | return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE); |
c7277090 EP |
4273 | } |
4274 | ||
4275 | /** | |
4276 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4277 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4278 | * @size: size to be set for the file | |
4279 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4280 | */ | |
4281 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4282 | { | |
703321b6 | 4283 | return __shmem_file_setup(shm_mnt, name, size, flags, 0); |
c7277090 | 4284 | } |
395e0ddc | 4285 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4286 | |
703321b6 MA |
4287 | /** |
4288 | * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs | |
4289 | * @mnt: the tmpfs mount where the file will be created | |
4290 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4291 | * @size: size to be set for the file | |
4292 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4293 | */ | |
4294 | struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name, | |
4295 | loff_t size, unsigned long flags) | |
4296 | { | |
4297 | return __shmem_file_setup(mnt, name, size, flags, 0); | |
4298 | } | |
4299 | EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt); | |
4300 | ||
46711810 | 4301 | /** |
1da177e4 | 4302 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
4303 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
4304 | */ | |
4305 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4306 | { | |
4307 | struct file *file; | |
4308 | loff_t size = vma->vm_end - vma->vm_start; | |
4309 | ||
66fc1303 HD |
4310 | /* |
4311 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
4312 | * between XFS directory reading and selinux: since this file is only | |
4313 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4314 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4315 | */ | |
703321b6 | 4316 | file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags); |
1da177e4 LT |
4317 | if (IS_ERR(file)) |
4318 | return PTR_ERR(file); | |
4319 | ||
4320 | if (vma->vm_file) | |
4321 | fput(vma->vm_file); | |
4322 | vma->vm_file = file; | |
4323 | vma->vm_ops = &shmem_vm_ops; | |
f3f0e1d2 | 4324 | |
e496cf3d | 4325 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
4326 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
4327 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
4328 | khugepaged_enter(vma, vma->vm_flags); | |
4329 | } | |
4330 | ||
1da177e4 LT |
4331 | return 0; |
4332 | } | |
d9d90e5e HD |
4333 | |
4334 | /** | |
4335 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
4336 | * @mapping: the page's address_space | |
4337 | * @index: the page index | |
4338 | * @gfp: the page allocator flags to use if allocating | |
4339 | * | |
4340 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4341 | * with any new page allocations done using the specified allocation flags. | |
4342 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
4343 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
4344 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4345 | * | |
68da9f05 HD |
4346 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4347 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4348 | */ |
4349 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4350 | pgoff_t index, gfp_t gfp) | |
4351 | { | |
68da9f05 HD |
4352 | #ifdef CONFIG_SHMEM |
4353 | struct inode *inode = mapping->host; | |
9276aad6 | 4354 | struct page *page; |
68da9f05 HD |
4355 | int error; |
4356 | ||
4357 | BUG_ON(mapping->a_ops != &shmem_aops); | |
9e18eb29 | 4358 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
cfda0526 | 4359 | gfp, NULL, NULL, NULL); |
68da9f05 HD |
4360 | if (error) |
4361 | page = ERR_PTR(error); | |
4362 | else | |
4363 | unlock_page(page); | |
4364 | return page; | |
4365 | #else | |
4366 | /* | |
4367 | * The tiny !SHMEM case uses ramfs without swap | |
4368 | */ | |
d9d90e5e | 4369 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4370 | #endif |
d9d90e5e HD |
4371 | } |
4372 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |