]>
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 | ||
c088e31d | 114 | static int shmem_default_max_inodes(void) |
680d794b | 115 | { |
c088e31d 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 | |
d9371d92 HD |
303 | /* nrpages adjustment first, then shmem_recalc_inode() when balanced */ |
304 | inode->i_mapping->nrpages += pages; | |
305 | ||
4595ef88 | 306 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
307 | info->alloced += pages; |
308 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
309 | shmem_recalc_inode(inode); | |
4595ef88 | 310 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 | 311 | |
800d8c63 KS |
312 | return true; |
313 | } | |
314 | ||
315 | void shmem_uncharge(struct inode *inode, long pages) | |
316 | { | |
317 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 318 | unsigned long flags; |
800d8c63 | 319 | |
d9371d92 HD |
320 | /* nrpages adjustment done by __delete_from_page_cache() or caller */ |
321 | ||
4595ef88 | 322 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
323 | info->alloced -= pages; |
324 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
325 | shmem_recalc_inode(inode); | |
4595ef88 | 326 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 | 327 | |
0f079694 | 328 | shmem_inode_unacct_blocks(inode, pages); |
800d8c63 KS |
329 | } |
330 | ||
7a5d0fbb HD |
331 | /* |
332 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
333 | */ | |
334 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
335 | pgoff_t index, void *expected, void *replacement) | |
336 | { | |
f7942430 | 337 | struct radix_tree_node *node; |
7a5d0fbb | 338 | void **pslot; |
6dbaf22c | 339 | void *item; |
7a5d0fbb HD |
340 | |
341 | VM_BUG_ON(!expected); | |
6dbaf22c | 342 | VM_BUG_ON(!replacement); |
f7942430 JW |
343 | item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot); |
344 | if (!item) | |
6dbaf22c | 345 | return -ENOENT; |
7a5d0fbb HD |
346 | if (item != expected) |
347 | return -ENOENT; | |
4d693d08 | 348 | __radix_tree_replace(&mapping->page_tree, node, pslot, |
c7df8ad2 | 349 | replacement, NULL); |
7a5d0fbb HD |
350 | return 0; |
351 | } | |
352 | ||
d1899228 HD |
353 | /* |
354 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
355 | * that an entry was not already brought back from swap by a racing thread. | |
356 | * | |
357 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
358 | * might be reused, and again be SwapCache, using the same swap as before. | |
359 | */ | |
360 | static bool shmem_confirm_swap(struct address_space *mapping, | |
361 | pgoff_t index, swp_entry_t swap) | |
362 | { | |
363 | void *item; | |
364 | ||
365 | rcu_read_lock(); | |
366 | item = radix_tree_lookup(&mapping->page_tree, index); | |
367 | rcu_read_unlock(); | |
368 | return item == swp_to_radix_entry(swap); | |
369 | } | |
370 | ||
5a6e75f8 KS |
371 | /* |
372 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
373 | * | |
374 | * SHMEM_HUGE_NEVER: | |
375 | * disables huge pages for the mount; | |
376 | * SHMEM_HUGE_ALWAYS: | |
377 | * enables huge pages for the mount; | |
378 | * SHMEM_HUGE_WITHIN_SIZE: | |
379 | * only allocate huge pages if the page will be fully within i_size, | |
380 | * also respect fadvise()/madvise() hints; | |
381 | * SHMEM_HUGE_ADVISE: | |
382 | * only allocate huge pages if requested with fadvise()/madvise(); | |
383 | */ | |
384 | ||
385 | #define SHMEM_HUGE_NEVER 0 | |
386 | #define SHMEM_HUGE_ALWAYS 1 | |
387 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
388 | #define SHMEM_HUGE_ADVISE 3 | |
389 | ||
390 | /* | |
391 | * Special values. | |
392 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
393 | * | |
394 | * SHMEM_HUGE_DENY: | |
395 | * disables huge on shm_mnt and all mounts, for emergency use; | |
396 | * SHMEM_HUGE_FORCE: | |
397 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
398 | * | |
399 | */ | |
400 | #define SHMEM_HUGE_DENY (-1) | |
401 | #define SHMEM_HUGE_FORCE (-2) | |
402 | ||
e496cf3d | 403 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
404 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
405 | ||
406 | int shmem_huge __read_mostly; | |
407 | ||
f1f5929c | 408 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
409 | static int shmem_parse_huge(const char *str) |
410 | { | |
411 | if (!strcmp(str, "never")) | |
412 | return SHMEM_HUGE_NEVER; | |
413 | if (!strcmp(str, "always")) | |
414 | return SHMEM_HUGE_ALWAYS; | |
415 | if (!strcmp(str, "within_size")) | |
416 | return SHMEM_HUGE_WITHIN_SIZE; | |
417 | if (!strcmp(str, "advise")) | |
418 | return SHMEM_HUGE_ADVISE; | |
419 | if (!strcmp(str, "deny")) | |
420 | return SHMEM_HUGE_DENY; | |
421 | if (!strcmp(str, "force")) | |
422 | return SHMEM_HUGE_FORCE; | |
423 | return -EINVAL; | |
424 | } | |
425 | ||
426 | static const char *shmem_format_huge(int huge) | |
427 | { | |
428 | switch (huge) { | |
429 | case SHMEM_HUGE_NEVER: | |
430 | return "never"; | |
431 | case SHMEM_HUGE_ALWAYS: | |
432 | return "always"; | |
433 | case SHMEM_HUGE_WITHIN_SIZE: | |
434 | return "within_size"; | |
435 | case SHMEM_HUGE_ADVISE: | |
436 | return "advise"; | |
437 | case SHMEM_HUGE_DENY: | |
438 | return "deny"; | |
439 | case SHMEM_HUGE_FORCE: | |
440 | return "force"; | |
441 | default: | |
442 | VM_BUG_ON(1); | |
443 | return "bad_val"; | |
444 | } | |
445 | } | |
f1f5929c | 446 | #endif |
5a6e75f8 | 447 | |
779750d2 KS |
448 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
449 | struct shrink_control *sc, unsigned long nr_to_split) | |
450 | { | |
451 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 452 | LIST_HEAD(to_remove); |
779750d2 KS |
453 | struct inode *inode; |
454 | struct shmem_inode_info *info; | |
455 | struct page *page; | |
456 | unsigned long batch = sc ? sc->nr_to_scan : 128; | |
457 | int removed = 0, split = 0; | |
458 | ||
459 | if (list_empty(&sbinfo->shrinklist)) | |
460 | return SHRINK_STOP; | |
461 | ||
462 | spin_lock(&sbinfo->shrinklist_lock); | |
463 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
464 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
465 | ||
466 | /* pin the inode */ | |
467 | inode = igrab(&info->vfs_inode); | |
468 | ||
469 | /* inode is about to be evicted */ | |
470 | if (!inode) { | |
471 | list_del_init(&info->shrinklist); | |
472 | removed++; | |
473 | goto next; | |
474 | } | |
475 | ||
476 | /* Check if there's anything to gain */ | |
477 | if (round_up(inode->i_size, PAGE_SIZE) == | |
478 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 479 | list_move(&info->shrinklist, &to_remove); |
779750d2 | 480 | removed++; |
779750d2 KS |
481 | goto next; |
482 | } | |
483 | ||
484 | list_move(&info->shrinklist, &list); | |
485 | next: | |
486 | if (!--batch) | |
487 | break; | |
488 | } | |
489 | spin_unlock(&sbinfo->shrinklist_lock); | |
490 | ||
253fd0f0 KS |
491 | list_for_each_safe(pos, next, &to_remove) { |
492 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
493 | inode = &info->vfs_inode; | |
494 | list_del_init(&info->shrinklist); | |
495 | iput(inode); | |
496 | } | |
497 | ||
779750d2 KS |
498 | list_for_each_safe(pos, next, &list) { |
499 | int ret; | |
500 | ||
501 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
502 | inode = &info->vfs_inode; | |
503 | ||
34649251 KS |
504 | if (nr_to_split && split >= nr_to_split) |
505 | goto leave; | |
779750d2 | 506 | |
34649251 | 507 | page = find_get_page(inode->i_mapping, |
779750d2 KS |
508 | (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT); |
509 | if (!page) | |
510 | goto drop; | |
511 | ||
34649251 | 512 | /* No huge page at the end of the file: nothing to split */ |
779750d2 | 513 | if (!PageTransHuge(page)) { |
779750d2 KS |
514 | put_page(page); |
515 | goto drop; | |
516 | } | |
517 | ||
34649251 KS |
518 | /* |
519 | * Leave the inode on the list if we failed to lock | |
520 | * the page at this time. | |
521 | * | |
522 | * Waiting for the lock may lead to deadlock in the | |
523 | * reclaim path. | |
524 | */ | |
525 | if (!trylock_page(page)) { | |
526 | put_page(page); | |
527 | goto leave; | |
528 | } | |
529 | ||
779750d2 KS |
530 | ret = split_huge_page(page); |
531 | unlock_page(page); | |
532 | put_page(page); | |
533 | ||
34649251 KS |
534 | /* If split failed leave the inode on the list */ |
535 | if (ret) | |
536 | goto leave; | |
779750d2 KS |
537 | |
538 | split++; | |
539 | drop: | |
540 | list_del_init(&info->shrinklist); | |
541 | removed++; | |
34649251 | 542 | leave: |
779750d2 KS |
543 | iput(inode); |
544 | } | |
545 | ||
546 | spin_lock(&sbinfo->shrinklist_lock); | |
547 | list_splice_tail(&list, &sbinfo->shrinklist); | |
548 | sbinfo->shrinklist_len -= removed; | |
549 | spin_unlock(&sbinfo->shrinklist_lock); | |
550 | ||
551 | return split; | |
552 | } | |
553 | ||
554 | static long shmem_unused_huge_scan(struct super_block *sb, | |
555 | struct shrink_control *sc) | |
556 | { | |
557 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
558 | ||
559 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
560 | return SHRINK_STOP; | |
561 | ||
562 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
563 | } | |
564 | ||
565 | static long shmem_unused_huge_count(struct super_block *sb, | |
566 | struct shrink_control *sc) | |
567 | { | |
568 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
569 | return READ_ONCE(sbinfo->shrinklist_len); | |
570 | } | |
e496cf3d | 571 | #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 KS |
572 | |
573 | #define shmem_huge SHMEM_HUGE_DENY | |
574 | ||
779750d2 KS |
575 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
576 | struct shrink_control *sc, unsigned long nr_to_split) | |
577 | { | |
578 | return 0; | |
579 | } | |
e496cf3d | 580 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 581 | |
46f65ec1 HD |
582 | /* |
583 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
584 | */ | |
585 | static int shmem_add_to_page_cache(struct page *page, | |
586 | struct address_space *mapping, | |
fed400a1 | 587 | pgoff_t index, void *expected) |
46f65ec1 | 588 | { |
800d8c63 | 589 | int error, nr = hpage_nr_pages(page); |
46f65ec1 | 590 | |
800d8c63 KS |
591 | VM_BUG_ON_PAGE(PageTail(page), page); |
592 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
593 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
594 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 595 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 596 | |
800d8c63 | 597 | page_ref_add(page, nr); |
b065b432 HD |
598 | page->mapping = mapping; |
599 | page->index = index; | |
600 | ||
601 | spin_lock_irq(&mapping->tree_lock); | |
800d8c63 KS |
602 | if (PageTransHuge(page)) { |
603 | void __rcu **results; | |
604 | pgoff_t idx; | |
605 | int i; | |
606 | ||
607 | error = 0; | |
608 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, | |
609 | &results, &idx, index, 1) && | |
610 | idx < index + HPAGE_PMD_NR) { | |
611 | error = -EEXIST; | |
612 | } | |
613 | ||
614 | if (!error) { | |
615 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
616 | error = radix_tree_insert(&mapping->page_tree, | |
617 | index + i, page + i); | |
618 | VM_BUG_ON(error); | |
619 | } | |
620 | count_vm_event(THP_FILE_ALLOC); | |
621 | } | |
622 | } else if (!expected) { | |
b065b432 | 623 | error = radix_tree_insert(&mapping->page_tree, index, page); |
800d8c63 | 624 | } else { |
b065b432 HD |
625 | error = shmem_radix_tree_replace(mapping, index, expected, |
626 | page); | |
800d8c63 KS |
627 | } |
628 | ||
46f65ec1 | 629 | if (!error) { |
800d8c63 KS |
630 | mapping->nrpages += nr; |
631 | if (PageTransHuge(page)) | |
11fb9989 MG |
632 | __inc_node_page_state(page, NR_SHMEM_THPS); |
633 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); | |
634 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr); | |
b065b432 HD |
635 | spin_unlock_irq(&mapping->tree_lock); |
636 | } else { | |
637 | page->mapping = NULL; | |
638 | spin_unlock_irq(&mapping->tree_lock); | |
800d8c63 | 639 | page_ref_sub(page, nr); |
46f65ec1 | 640 | } |
46f65ec1 HD |
641 | return error; |
642 | } | |
643 | ||
6922c0c7 HD |
644 | /* |
645 | * Like delete_from_page_cache, but substitutes swap for page. | |
646 | */ | |
647 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
648 | { | |
649 | struct address_space *mapping = page->mapping; | |
650 | int error; | |
651 | ||
800d8c63 KS |
652 | VM_BUG_ON_PAGE(PageCompound(page), page); |
653 | ||
6922c0c7 HD |
654 | spin_lock_irq(&mapping->tree_lock); |
655 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
656 | page->mapping = NULL; | |
657 | mapping->nrpages--; | |
11fb9989 MG |
658 | __dec_node_page_state(page, NR_FILE_PAGES); |
659 | __dec_node_page_state(page, NR_SHMEM); | |
6922c0c7 | 660 | spin_unlock_irq(&mapping->tree_lock); |
09cbfeaf | 661 | put_page(page); |
6922c0c7 HD |
662 | BUG_ON(error); |
663 | } | |
664 | ||
7a5d0fbb HD |
665 | /* |
666 | * Remove swap entry from radix tree, free the swap and its page cache. | |
667 | */ | |
668 | static int shmem_free_swap(struct address_space *mapping, | |
669 | pgoff_t index, void *radswap) | |
670 | { | |
6dbaf22c | 671 | void *old; |
7a5d0fbb HD |
672 | |
673 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 674 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 675 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
676 | if (old != radswap) |
677 | return -ENOENT; | |
678 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
679 | return 0; | |
7a5d0fbb HD |
680 | } |
681 | ||
6a15a370 VB |
682 | /* |
683 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 684 | * given offsets are swapped out. |
6a15a370 VB |
685 | * |
686 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
687 | * as long as the inode doesn't go away and racy results are not a problem. | |
688 | */ | |
48131e03 VB |
689 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
690 | pgoff_t start, pgoff_t end) | |
6a15a370 | 691 | { |
6a15a370 VB |
692 | struct radix_tree_iter iter; |
693 | void **slot; | |
694 | struct page *page; | |
48131e03 | 695 | unsigned long swapped = 0; |
6a15a370 VB |
696 | |
697 | rcu_read_lock(); | |
698 | ||
6a15a370 VB |
699 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
700 | if (iter.index >= end) | |
701 | break; | |
702 | ||
703 | page = radix_tree_deref_slot(slot); | |
704 | ||
2cf938aa MW |
705 | if (radix_tree_deref_retry(page)) { |
706 | slot = radix_tree_iter_retry(&iter); | |
707 | continue; | |
708 | } | |
6a15a370 VB |
709 | |
710 | if (radix_tree_exceptional_entry(page)) | |
711 | swapped++; | |
712 | ||
713 | if (need_resched()) { | |
148deab2 | 714 | slot = radix_tree_iter_resume(slot, &iter); |
6a15a370 | 715 | cond_resched_rcu(); |
6a15a370 VB |
716 | } |
717 | } | |
718 | ||
719 | rcu_read_unlock(); | |
720 | ||
721 | return swapped << PAGE_SHIFT; | |
722 | } | |
723 | ||
48131e03 VB |
724 | /* |
725 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
726 | * given vma is swapped out. | |
727 | * | |
728 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
729 | * as long as the inode doesn't go away and racy results are not a problem. | |
730 | */ | |
731 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
732 | { | |
733 | struct inode *inode = file_inode(vma->vm_file); | |
734 | struct shmem_inode_info *info = SHMEM_I(inode); | |
735 | struct address_space *mapping = inode->i_mapping; | |
736 | unsigned long swapped; | |
737 | ||
738 | /* Be careful as we don't hold info->lock */ | |
739 | swapped = READ_ONCE(info->swapped); | |
740 | ||
741 | /* | |
742 | * The easier cases are when the shmem object has nothing in swap, or | |
743 | * the vma maps it whole. Then we can simply use the stats that we | |
744 | * already track. | |
745 | */ | |
746 | if (!swapped) | |
747 | return 0; | |
748 | ||
749 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
750 | return swapped << PAGE_SHIFT; | |
751 | ||
752 | /* Here comes the more involved part */ | |
753 | return shmem_partial_swap_usage(mapping, | |
754 | linear_page_index(vma, vma->vm_start), | |
755 | linear_page_index(vma, vma->vm_end)); | |
756 | } | |
757 | ||
24513264 HD |
758 | /* |
759 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
760 | */ | |
761 | void shmem_unlock_mapping(struct address_space *mapping) | |
762 | { | |
763 | struct pagevec pvec; | |
764 | pgoff_t indices[PAGEVEC_SIZE]; | |
765 | pgoff_t index = 0; | |
766 | ||
86679820 | 767 | pagevec_init(&pvec); |
24513264 HD |
768 | /* |
769 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
770 | */ | |
771 | while (!mapping_unevictable(mapping)) { | |
772 | /* | |
773 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
774 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
775 | */ | |
0cd6144a JW |
776 | pvec.nr = find_get_entries(mapping, index, |
777 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
778 | if (!pvec.nr) |
779 | break; | |
780 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 781 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
782 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
783 | pagevec_release(&pvec); | |
784 | cond_resched(); | |
785 | } | |
7a5d0fbb HD |
786 | } |
787 | ||
788 | /* | |
789 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 790 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 791 | */ |
1635f6a7 HD |
792 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
793 | bool unfalloc) | |
1da177e4 | 794 | { |
285b2c4f | 795 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 796 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
797 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
798 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
799 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
800 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 801 | struct pagevec pvec; |
7a5d0fbb HD |
802 | pgoff_t indices[PAGEVEC_SIZE]; |
803 | long nr_swaps_freed = 0; | |
285b2c4f | 804 | pgoff_t index; |
bda97eab HD |
805 | int i; |
806 | ||
83e4fa9c HD |
807 | if (lend == -1) |
808 | end = -1; /* unsigned, so actually very big */ | |
bda97eab | 809 | |
86679820 | 810 | pagevec_init(&pvec); |
bda97eab | 811 | index = start; |
83e4fa9c | 812 | while (index < end) { |
0cd6144a JW |
813 | pvec.nr = find_get_entries(mapping, index, |
814 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
815 | pvec.pages, indices); | |
7a5d0fbb HD |
816 | if (!pvec.nr) |
817 | break; | |
bda97eab HD |
818 | for (i = 0; i < pagevec_count(&pvec); i++) { |
819 | struct page *page = pvec.pages[i]; | |
820 | ||
7a5d0fbb | 821 | index = indices[i]; |
83e4fa9c | 822 | if (index >= end) |
bda97eab HD |
823 | break; |
824 | ||
7a5d0fbb | 825 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
826 | if (unfalloc) |
827 | continue; | |
7a5d0fbb HD |
828 | nr_swaps_freed += !shmem_free_swap(mapping, |
829 | index, page); | |
bda97eab | 830 | continue; |
7a5d0fbb HD |
831 | } |
832 | ||
800d8c63 KS |
833 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
834 | ||
7a5d0fbb | 835 | if (!trylock_page(page)) |
bda97eab | 836 | continue; |
800d8c63 KS |
837 | |
838 | if (PageTransTail(page)) { | |
839 | /* Middle of THP: zero out the page */ | |
840 | clear_highpage(page); | |
841 | unlock_page(page); | |
842 | continue; | |
843 | } else if (PageTransHuge(page)) { | |
844 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
845 | /* | |
846 | * Range ends in the middle of THP: | |
847 | * zero out the page | |
848 | */ | |
849 | clear_highpage(page); | |
850 | unlock_page(page); | |
851 | continue; | |
852 | } | |
853 | index += HPAGE_PMD_NR - 1; | |
854 | i += HPAGE_PMD_NR - 1; | |
855 | } | |
856 | ||
1635f6a7 | 857 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
858 | VM_BUG_ON_PAGE(PageTail(page), page); |
859 | if (page_mapping(page) == mapping) { | |
309381fe | 860 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
861 | truncate_inode_page(mapping, page); |
862 | } | |
bda97eab | 863 | } |
bda97eab HD |
864 | unlock_page(page); |
865 | } | |
0cd6144a | 866 | pagevec_remove_exceptionals(&pvec); |
24513264 | 867 | pagevec_release(&pvec); |
bda97eab HD |
868 | cond_resched(); |
869 | index++; | |
870 | } | |
1da177e4 | 871 | |
83e4fa9c | 872 | if (partial_start) { |
bda97eab | 873 | struct page *page = NULL; |
9e18eb29 | 874 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 875 | if (page) { |
09cbfeaf | 876 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
877 | if (start > end) { |
878 | top = partial_end; | |
879 | partial_end = 0; | |
880 | } | |
881 | zero_user_segment(page, partial_start, top); | |
882 | set_page_dirty(page); | |
883 | unlock_page(page); | |
09cbfeaf | 884 | put_page(page); |
83e4fa9c HD |
885 | } |
886 | } | |
887 | if (partial_end) { | |
888 | struct page *page = NULL; | |
9e18eb29 | 889 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
890 | if (page) { |
891 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
892 | set_page_dirty(page); |
893 | unlock_page(page); | |
09cbfeaf | 894 | put_page(page); |
bda97eab HD |
895 | } |
896 | } | |
83e4fa9c HD |
897 | if (start >= end) |
898 | return; | |
bda97eab HD |
899 | |
900 | index = start; | |
b1a36650 | 901 | while (index < end) { |
bda97eab | 902 | cond_resched(); |
0cd6144a JW |
903 | |
904 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 905 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 906 | pvec.pages, indices); |
7a5d0fbb | 907 | if (!pvec.nr) { |
b1a36650 HD |
908 | /* If all gone or hole-punch or unfalloc, we're done */ |
909 | if (index == start || end != -1) | |
bda97eab | 910 | break; |
b1a36650 | 911 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
912 | index = start; |
913 | continue; | |
914 | } | |
bda97eab HD |
915 | for (i = 0; i < pagevec_count(&pvec); i++) { |
916 | struct page *page = pvec.pages[i]; | |
917 | ||
7a5d0fbb | 918 | index = indices[i]; |
83e4fa9c | 919 | if (index >= end) |
bda97eab HD |
920 | break; |
921 | ||
7a5d0fbb | 922 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
923 | if (unfalloc) |
924 | continue; | |
b1a36650 HD |
925 | if (shmem_free_swap(mapping, index, page)) { |
926 | /* Swap was replaced by page: retry */ | |
927 | index--; | |
928 | break; | |
929 | } | |
930 | nr_swaps_freed++; | |
7a5d0fbb HD |
931 | continue; |
932 | } | |
933 | ||
bda97eab | 934 | lock_page(page); |
800d8c63 KS |
935 | |
936 | if (PageTransTail(page)) { | |
937 | /* Middle of THP: zero out the page */ | |
938 | clear_highpage(page); | |
939 | unlock_page(page); | |
940 | /* | |
941 | * Partial thp truncate due 'start' in middle | |
942 | * of THP: don't need to look on these pages | |
943 | * again on !pvec.nr restart. | |
944 | */ | |
945 | if (index != round_down(end, HPAGE_PMD_NR)) | |
946 | start++; | |
947 | continue; | |
948 | } else if (PageTransHuge(page)) { | |
949 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
950 | /* | |
951 | * Range ends in the middle of THP: | |
952 | * zero out the page | |
953 | */ | |
954 | clear_highpage(page); | |
955 | unlock_page(page); | |
956 | continue; | |
957 | } | |
958 | index += HPAGE_PMD_NR - 1; | |
959 | i += HPAGE_PMD_NR - 1; | |
960 | } | |
961 | ||
1635f6a7 | 962 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
963 | VM_BUG_ON_PAGE(PageTail(page), page); |
964 | if (page_mapping(page) == mapping) { | |
309381fe | 965 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 966 | truncate_inode_page(mapping, page); |
b1a36650 HD |
967 | } else { |
968 | /* Page was replaced by swap: retry */ | |
969 | unlock_page(page); | |
970 | index--; | |
971 | break; | |
1635f6a7 | 972 | } |
7a5d0fbb | 973 | } |
bda97eab HD |
974 | unlock_page(page); |
975 | } | |
0cd6144a | 976 | pagevec_remove_exceptionals(&pvec); |
24513264 | 977 | pagevec_release(&pvec); |
bda97eab HD |
978 | index++; |
979 | } | |
94c1e62d | 980 | |
4595ef88 | 981 | spin_lock_irq(&info->lock); |
7a5d0fbb | 982 | info->swapped -= nr_swaps_freed; |
1da177e4 | 983 | shmem_recalc_inode(inode); |
4595ef88 | 984 | spin_unlock_irq(&info->lock); |
1635f6a7 | 985 | } |
1da177e4 | 986 | |
1635f6a7 HD |
987 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
988 | { | |
989 | shmem_undo_range(inode, lstart, lend, false); | |
078cd827 | 990 | inode->i_ctime = inode->i_mtime = current_time(inode); |
1da177e4 | 991 | } |
94c1e62d | 992 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 993 | |
a528d35e DH |
994 | static int shmem_getattr(const struct path *path, struct kstat *stat, |
995 | u32 request_mask, unsigned int query_flags) | |
44a30220 | 996 | { |
a528d35e | 997 | struct inode *inode = path->dentry->d_inode; |
44a30220 YZ |
998 | struct shmem_inode_info *info = SHMEM_I(inode); |
999 | ||
d0424c42 | 1000 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
4595ef88 | 1001 | spin_lock_irq(&info->lock); |
d0424c42 | 1002 | shmem_recalc_inode(inode); |
4595ef88 | 1003 | spin_unlock_irq(&info->lock); |
d0424c42 | 1004 | } |
44a30220 | 1005 | generic_fillattr(inode, stat); |
44a30220 YZ |
1006 | return 0; |
1007 | } | |
1008 | ||
94c1e62d | 1009 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 1010 | { |
75c3cfa8 | 1011 | struct inode *inode = d_inode(dentry); |
40e041a2 | 1012 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1013 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 LT |
1014 | int error; |
1015 | ||
31051c85 | 1016 | error = setattr_prepare(dentry, attr); |
db78b877 CH |
1017 | if (error) |
1018 | return error; | |
1019 | ||
94c1e62d HD |
1020 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
1021 | loff_t oldsize = inode->i_size; | |
1022 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 1023 | |
40e041a2 DH |
1024 | /* protected by i_mutex */ |
1025 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
1026 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
1027 | return -EPERM; | |
1028 | ||
94c1e62d | 1029 | if (newsize != oldsize) { |
77142517 KK |
1030 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
1031 | oldsize, newsize); | |
1032 | if (error) | |
1033 | return error; | |
94c1e62d | 1034 | i_size_write(inode, newsize); |
078cd827 | 1035 | inode->i_ctime = inode->i_mtime = current_time(inode); |
94c1e62d | 1036 | } |
afa2db2f | 1037 | if (newsize <= oldsize) { |
94c1e62d | 1038 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
1039 | if (oldsize > holebegin) |
1040 | unmap_mapping_range(inode->i_mapping, | |
1041 | holebegin, 0, 1); | |
1042 | if (info->alloced) | |
1043 | shmem_truncate_range(inode, | |
1044 | newsize, (loff_t)-1); | |
94c1e62d | 1045 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
1046 | if (oldsize > holebegin) |
1047 | unmap_mapping_range(inode->i_mapping, | |
1048 | holebegin, 0, 1); | |
779750d2 KS |
1049 | |
1050 | /* | |
1051 | * Part of the huge page can be beyond i_size: subject | |
1052 | * to shrink under memory pressure. | |
1053 | */ | |
1054 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { | |
1055 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1056 | /* |
1057 | * _careful to defend against unlocked access to | |
1058 | * ->shrink_list in shmem_unused_huge_shrink() | |
1059 | */ | |
1060 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1061 | list_add_tail(&info->shrinklist, |
1062 | &sbinfo->shrinklist); | |
1063 | sbinfo->shrinklist_len++; | |
1064 | } | |
1065 | spin_unlock(&sbinfo->shrinklist_lock); | |
1066 | } | |
94c1e62d | 1067 | } |
1da177e4 LT |
1068 | } |
1069 | ||
db78b877 | 1070 | setattr_copy(inode, attr); |
db78b877 | 1071 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 1072 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
1073 | return error; |
1074 | } | |
1075 | ||
1f895f75 | 1076 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1077 | { |
1da177e4 | 1078 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1079 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 | 1080 | |
3889e6e7 | 1081 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
1082 | shmem_unacct_size(info->flags, inode->i_size); |
1083 | inode->i_size = 0; | |
3889e6e7 | 1084 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1085 | if (!list_empty(&info->shrinklist)) { |
1086 | spin_lock(&sbinfo->shrinklist_lock); | |
1087 | if (!list_empty(&info->shrinklist)) { | |
1088 | list_del_init(&info->shrinklist); | |
1089 | sbinfo->shrinklist_len--; | |
1090 | } | |
1091 | spin_unlock(&sbinfo->shrinklist_lock); | |
1092 | } | |
1da177e4 | 1093 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 1094 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 1095 | list_del_init(&info->swaplist); |
cb5f7b9a | 1096 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1097 | } |
3ed47db3 | 1098 | } |
b09e0fa4 | 1099 | |
38f38657 | 1100 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 1101 | WARN_ON(inode->i_blocks); |
c088e31d SF |
1102 | if (!sbinfo->idr_nouse && inode->i_ino) { |
1103 | mutex_lock(&sbinfo->idr_lock); | |
1104 | idr_remove(&sbinfo->idr, inode->i_ino); | |
1105 | mutex_unlock(&sbinfo->idr_lock); | |
1106 | } | |
5b04c689 | 1107 | shmem_free_inode(inode->i_sb); |
dbd5768f | 1108 | clear_inode(inode); |
1da177e4 LT |
1109 | } |
1110 | ||
478922e2 MW |
1111 | static unsigned long find_swap_entry(struct radix_tree_root *root, void *item) |
1112 | { | |
1113 | struct radix_tree_iter iter; | |
1114 | void **slot; | |
1115 | unsigned long found = -1; | |
1116 | unsigned int checked = 0; | |
1117 | ||
1118 | rcu_read_lock(); | |
1119 | radix_tree_for_each_slot(slot, root, &iter, 0) { | |
1120 | if (*slot == item) { | |
1121 | found = iter.index; | |
1122 | break; | |
1123 | } | |
1124 | checked++; | |
1125 | if ((checked % 4096) != 0) | |
1126 | continue; | |
1127 | slot = radix_tree_iter_resume(slot, &iter); | |
1128 | cond_resched_rcu(); | |
1129 | } | |
1130 | ||
1131 | rcu_read_unlock(); | |
1132 | return found; | |
1133 | } | |
1134 | ||
46f65ec1 HD |
1135 | /* |
1136 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1137 | */ | |
41ffe5d5 | 1138 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 1139 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 1140 | { |
285b2c4f | 1141 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 1142 | void *radswap; |
41ffe5d5 | 1143 | pgoff_t index; |
bde05d1c HD |
1144 | gfp_t gfp; |
1145 | int error = 0; | |
1da177e4 | 1146 | |
46f65ec1 | 1147 | radswap = swp_to_radix_entry(swap); |
478922e2 | 1148 | index = find_swap_entry(&mapping->page_tree, radswap); |
46f65ec1 | 1149 | if (index == -1) |
00501b53 | 1150 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 1151 | |
1b1b32f2 HD |
1152 | /* |
1153 | * Move _head_ to start search for next from here. | |
1f895f75 | 1154 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 1155 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 1156 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
1157 | */ |
1158 | if (shmem_swaplist.next != &info->swaplist) | |
1159 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 1160 | |
bde05d1c HD |
1161 | gfp = mapping_gfp_mask(mapping); |
1162 | if (shmem_should_replace_page(*pagep, gfp)) { | |
1163 | mutex_unlock(&shmem_swaplist_mutex); | |
1164 | error = shmem_replace_page(pagep, gfp, info, index); | |
1165 | mutex_lock(&shmem_swaplist_mutex); | |
1166 | /* | |
1167 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
1168 | * allocation, but the inode might have been freed while we |
1169 | * dropped it: although a racing shmem_evict_inode() cannot | |
1170 | * complete without emptying the radix_tree, our page lock | |
1171 | * on this swapcache page is not enough to prevent that - | |
1172 | * free_swap_and_cache() of our swap entry will only | |
1173 | * trylock_page(), removing swap from radix_tree whatever. | |
1174 | * | |
1175 | * We must not proceed to shmem_add_to_page_cache() if the | |
1176 | * inode has been freed, but of course we cannot rely on | |
1177 | * inode or mapping or info to check that. However, we can | |
1178 | * safely check if our swap entry is still in use (and here | |
1179 | * it can't have got reused for another page): if it's still | |
1180 | * in use, then the inode cannot have been freed yet, and we | |
1181 | * can safely proceed (if it's no longer in use, that tells | |
1182 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
1183 | */ |
1184 | if (!page_swapcount(*pagep)) | |
1185 | error = -ENOENT; | |
1186 | } | |
1187 | ||
d13d1443 | 1188 | /* |
778dd893 HD |
1189 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
1190 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
1191 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 1192 | */ |
bde05d1c HD |
1193 | if (!error) |
1194 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 1195 | radswap); |
48f170fb | 1196 | if (error != -ENOMEM) { |
46f65ec1 HD |
1197 | /* |
1198 | * Truncation and eviction use free_swap_and_cache(), which | |
1199 | * only does trylock page: if we raced, best clean up here. | |
1200 | */ | |
bde05d1c HD |
1201 | delete_from_swap_cache(*pagep); |
1202 | set_page_dirty(*pagep); | |
46f65ec1 | 1203 | if (!error) { |
4595ef88 | 1204 | spin_lock_irq(&info->lock); |
46f65ec1 | 1205 | info->swapped--; |
4595ef88 | 1206 | spin_unlock_irq(&info->lock); |
46f65ec1 HD |
1207 | swap_free(swap); |
1208 | } | |
1da177e4 | 1209 | } |
2e0e26c7 | 1210 | return error; |
1da177e4 LT |
1211 | } |
1212 | ||
1213 | /* | |
46f65ec1 | 1214 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 1215 | */ |
41ffe5d5 | 1216 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 1217 | { |
41ffe5d5 | 1218 | struct list_head *this, *next; |
1da177e4 | 1219 | struct shmem_inode_info *info; |
00501b53 | 1220 | struct mem_cgroup *memcg; |
bde05d1c HD |
1221 | int error = 0; |
1222 | ||
1223 | /* | |
1224 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 1225 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 1226 | */ |
0142ef6c | 1227 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 1228 | goto out; |
778dd893 HD |
1229 | |
1230 | /* | |
1231 | * Charge page using GFP_KERNEL while we can wait, before taking | |
1232 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
1233 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 1234 | */ |
f627c2f5 KS |
1235 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg, |
1236 | false); | |
778dd893 HD |
1237 | if (error) |
1238 | goto out; | |
46f65ec1 | 1239 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 1240 | error = -EAGAIN; |
1da177e4 | 1241 | |
cb5f7b9a | 1242 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
1243 | list_for_each_safe(this, next, &shmem_swaplist) { |
1244 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 1245 | if (info->swapped) |
00501b53 | 1246 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
1247 | else |
1248 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1249 | cond_resched(); |
00501b53 | 1250 | if (error != -EAGAIN) |
778dd893 | 1251 | break; |
00501b53 | 1252 | /* found nothing in this: move on to search the next */ |
1da177e4 | 1253 | } |
cb5f7b9a | 1254 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1255 | |
00501b53 JW |
1256 | if (error) { |
1257 | if (error != -ENOMEM) | |
1258 | error = 0; | |
f627c2f5 | 1259 | mem_cgroup_cancel_charge(page, memcg, false); |
00501b53 | 1260 | } else |
f627c2f5 | 1261 | mem_cgroup_commit_charge(page, memcg, true, false); |
778dd893 | 1262 | out: |
aaa46865 | 1263 | unlock_page(page); |
09cbfeaf | 1264 | put_page(page); |
778dd893 | 1265 | return error; |
1da177e4 LT |
1266 | } |
1267 | ||
1268 | /* | |
1269 | * Move the page from the page cache to the swap cache. | |
1270 | */ | |
1271 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1272 | { | |
1273 | struct shmem_inode_info *info; | |
1da177e4 | 1274 | struct address_space *mapping; |
1da177e4 | 1275 | struct inode *inode; |
6922c0c7 HD |
1276 | swp_entry_t swap; |
1277 | pgoff_t index; | |
1da177e4 | 1278 | |
800d8c63 | 1279 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1280 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1281 | mapping = page->mapping; |
1282 | index = page->index; | |
1283 | inode = mapping->host; | |
1284 | info = SHMEM_I(inode); | |
1285 | if (info->flags & VM_LOCKED) | |
1286 | goto redirty; | |
d9fe526a | 1287 | if (!total_swap_pages) |
1da177e4 LT |
1288 | goto redirty; |
1289 | ||
d9fe526a | 1290 | /* |
97b713ba CH |
1291 | * Our capabilities prevent regular writeback or sync from ever calling |
1292 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1293 | * its underlying filesystem, in which case tmpfs should write out to | |
1294 | * swap only in response to memory pressure, and not for the writeback | |
1295 | * threads or sync. | |
d9fe526a | 1296 | */ |
48f170fb HD |
1297 | if (!wbc->for_reclaim) { |
1298 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1299 | goto redirty; | |
1300 | } | |
1635f6a7 HD |
1301 | |
1302 | /* | |
1303 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1304 | * value into swapfile.c, the only way we can correctly account for a | |
1305 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1306 | * |
1307 | * That's okay for a page already fallocated earlier, but if we have | |
1308 | * not yet completed the fallocation, then (a) we want to keep track | |
1309 | * of this page in case we have to undo it, and (b) it may not be a | |
1310 | * good idea to continue anyway, once we're pushing into swap. So | |
1311 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1312 | */ |
1313 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1314 | if (inode->i_private) { |
1315 | struct shmem_falloc *shmem_falloc; | |
1316 | spin_lock(&inode->i_lock); | |
1317 | shmem_falloc = inode->i_private; | |
1318 | if (shmem_falloc && | |
8e205f77 | 1319 | !shmem_falloc->waitq && |
1aac1400 HD |
1320 | index >= shmem_falloc->start && |
1321 | index < shmem_falloc->next) | |
1322 | shmem_falloc->nr_unswapped++; | |
1323 | else | |
1324 | shmem_falloc = NULL; | |
1325 | spin_unlock(&inode->i_lock); | |
1326 | if (shmem_falloc) | |
1327 | goto redirty; | |
1328 | } | |
1635f6a7 HD |
1329 | clear_highpage(page); |
1330 | flush_dcache_page(page); | |
1331 | SetPageUptodate(page); | |
1332 | } | |
1333 | ||
38d8b4e6 | 1334 | swap = get_swap_page(page); |
48f170fb HD |
1335 | if (!swap.val) |
1336 | goto redirty; | |
d9fe526a | 1337 | |
37e84351 VD |
1338 | if (mem_cgroup_try_charge_swap(page, swap)) |
1339 | goto free_swap; | |
1340 | ||
b1dea800 HD |
1341 | /* |
1342 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1343 | * if it's not already there. Do it now before the page is |
1344 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1345 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1346 | * we've incremented swapped, because shmem_unuse_inode() will |
1347 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1348 | */ |
48f170fb HD |
1349 | mutex_lock(&shmem_swaplist_mutex); |
1350 | if (list_empty(&info->swaplist)) | |
1351 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 1352 | |
48f170fb | 1353 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
4595ef88 | 1354 | spin_lock_irq(&info->lock); |
6922c0c7 | 1355 | shmem_recalc_inode(inode); |
267a4c76 | 1356 | info->swapped++; |
4595ef88 | 1357 | spin_unlock_irq(&info->lock); |
6922c0c7 | 1358 | |
267a4c76 HD |
1359 | swap_shmem_alloc(swap); |
1360 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1361 | ||
6922c0c7 | 1362 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1363 | BUG_ON(page_mapped(page)); |
9fab5619 | 1364 | swap_writepage(page, wbc); |
1da177e4 LT |
1365 | return 0; |
1366 | } | |
1367 | ||
6922c0c7 | 1368 | mutex_unlock(&shmem_swaplist_mutex); |
37e84351 | 1369 | free_swap: |
75f6d6d2 | 1370 | put_swap_page(page, swap); |
1da177e4 LT |
1371 | redirty: |
1372 | set_page_dirty(page); | |
d9fe526a HD |
1373 | if (wbc->for_reclaim) |
1374 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1375 | unlock_page(page); | |
1376 | return 0; | |
1da177e4 LT |
1377 | } |
1378 | ||
75edd345 | 1379 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1380 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1381 | { |
095f1fc4 | 1382 | char buffer[64]; |
680d794b | 1383 | |
71fe804b | 1384 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1385 | return; /* show nothing */ |
680d794b | 1386 | |
a7a88b23 | 1387 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1388 | |
1389 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1390 | } |
71fe804b LS |
1391 | |
1392 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1393 | { | |
1394 | struct mempolicy *mpol = NULL; | |
1395 | if (sbinfo->mpol) { | |
1396 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1397 | mpol = sbinfo->mpol; | |
1398 | mpol_get(mpol); | |
1399 | spin_unlock(&sbinfo->stat_lock); | |
1400 | } | |
1401 | return mpol; | |
1402 | } | |
75edd345 HD |
1403 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1404 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1405 | { | |
1406 | } | |
1407 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1408 | { | |
1409 | return NULL; | |
1410 | } | |
1411 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1412 | #ifndef CONFIG_NUMA | |
1413 | #define vm_policy vm_private_data | |
1414 | #endif | |
680d794b | 1415 | |
800d8c63 KS |
1416 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1417 | struct shmem_inode_info *info, pgoff_t index) | |
1418 | { | |
1419 | /* Create a pseudo vma that just contains the policy */ | |
1420 | vma->vm_start = 0; | |
1421 | /* Bias interleave by inode number to distribute better across nodes */ | |
1422 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
1423 | vma->vm_ops = NULL; | |
1424 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); | |
1425 | } | |
1426 | ||
1427 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1428 | { | |
1429 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1430 | mpol_cond_put(vma->vm_policy); | |
1431 | } | |
1432 | ||
41ffe5d5 HD |
1433 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1434 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1435 | { |
1da177e4 | 1436 | struct vm_area_struct pvma; |
18a2f371 | 1437 | struct page *page; |
52cd3b07 | 1438 | |
800d8c63 | 1439 | shmem_pseudo_vma_init(&pvma, info, index); |
18a2f371 | 1440 | page = swapin_readahead(swap, gfp, &pvma, 0); |
800d8c63 | 1441 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1442 | |
800d8c63 KS |
1443 | return page; |
1444 | } | |
1445 | ||
1446 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1447 | struct shmem_inode_info *info, pgoff_t index) | |
1448 | { | |
1449 | struct vm_area_struct pvma; | |
1450 | struct inode *inode = &info->vfs_inode; | |
1451 | struct address_space *mapping = inode->i_mapping; | |
4620a06e | 1452 | pgoff_t idx, hindex; |
800d8c63 KS |
1453 | void __rcu **results; |
1454 | struct page *page; | |
1455 | ||
e496cf3d | 1456 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1457 | return NULL; |
1458 | ||
4620a06e | 1459 | hindex = round_down(index, HPAGE_PMD_NR); |
800d8c63 KS |
1460 | rcu_read_lock(); |
1461 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx, | |
1462 | hindex, 1) && idx < hindex + HPAGE_PMD_NR) { | |
1463 | rcu_read_unlock(); | |
1464 | return NULL; | |
1465 | } | |
1466 | rcu_read_unlock(); | |
18a2f371 | 1467 | |
800d8c63 KS |
1468 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1469 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
1470 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); | |
1471 | shmem_pseudo_vma_destroy(&pvma); | |
1472 | if (page) | |
1473 | prep_transhuge_page(page); | |
18a2f371 | 1474 | return page; |
1da177e4 LT |
1475 | } |
1476 | ||
02098fea | 1477 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1478 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1479 | { |
1480 | struct vm_area_struct pvma; | |
18a2f371 | 1481 | struct page *page; |
1da177e4 | 1482 | |
800d8c63 KS |
1483 | shmem_pseudo_vma_init(&pvma, info, index); |
1484 | page = alloc_page_vma(gfp, &pvma, 0); | |
1485 | shmem_pseudo_vma_destroy(&pvma); | |
1486 | ||
1487 | return page; | |
1488 | } | |
1489 | ||
1490 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
0f079694 | 1491 | struct inode *inode, |
800d8c63 KS |
1492 | pgoff_t index, bool huge) |
1493 | { | |
0f079694 | 1494 | struct shmem_inode_info *info = SHMEM_I(inode); |
800d8c63 KS |
1495 | struct page *page; |
1496 | int nr; | |
1497 | int err = -ENOSPC; | |
52cd3b07 | 1498 | |
e496cf3d | 1499 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1500 | huge = false; |
1501 | nr = huge ? HPAGE_PMD_NR : 1; | |
1502 | ||
0f079694 | 1503 | if (!shmem_inode_acct_block(inode, nr)) |
800d8c63 | 1504 | goto failed; |
800d8c63 KS |
1505 | |
1506 | if (huge) | |
1507 | page = shmem_alloc_hugepage(gfp, info, index); | |
1508 | else | |
1509 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1510 | if (page) { |
1511 | __SetPageLocked(page); | |
1512 | __SetPageSwapBacked(page); | |
800d8c63 | 1513 | return page; |
75edd345 | 1514 | } |
18a2f371 | 1515 | |
800d8c63 | 1516 | err = -ENOMEM; |
0f079694 | 1517 | shmem_inode_unacct_blocks(inode, nr); |
800d8c63 KS |
1518 | failed: |
1519 | return ERR_PTR(err); | |
1da177e4 | 1520 | } |
71fe804b | 1521 | |
bde05d1c HD |
1522 | /* |
1523 | * When a page is moved from swapcache to shmem filecache (either by the | |
1524 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1525 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1526 | * ignorance of the mapping it belongs to. If that mapping has special | |
1527 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1528 | * we may need to copy to a suitable page before moving to filecache. | |
1529 | * | |
1530 | * In a future release, this may well be extended to respect cpuset and | |
1531 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1532 | * but for now it is a simple matter of zone. | |
1533 | */ | |
1534 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1535 | { | |
1536 | return page_zonenum(page) > gfp_zone(gfp); | |
1537 | } | |
1538 | ||
1539 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1540 | struct shmem_inode_info *info, pgoff_t index) | |
1541 | { | |
1542 | struct page *oldpage, *newpage; | |
1543 | struct address_space *swap_mapping; | |
2aca711b | 1544 | swp_entry_t entry; |
bde05d1c HD |
1545 | pgoff_t swap_index; |
1546 | int error; | |
1547 | ||
1548 | oldpage = *pagep; | |
2aca711b YZ |
1549 | entry.val = page_private(oldpage); |
1550 | swap_index = swp_offset(entry); | |
bde05d1c HD |
1551 | swap_mapping = page_mapping(oldpage); |
1552 | ||
1553 | /* | |
1554 | * We have arrived here because our zones are constrained, so don't | |
1555 | * limit chance of success by further cpuset and node constraints. | |
1556 | */ | |
1557 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1558 | newpage = shmem_alloc_page(gfp, info, index); | |
1559 | if (!newpage) | |
1560 | return -ENOMEM; | |
bde05d1c | 1561 | |
09cbfeaf | 1562 | get_page(newpage); |
bde05d1c | 1563 | copy_highpage(newpage, oldpage); |
0142ef6c | 1564 | flush_dcache_page(newpage); |
bde05d1c | 1565 | |
9956edf3 HD |
1566 | __SetPageLocked(newpage); |
1567 | __SetPageSwapBacked(newpage); | |
bde05d1c | 1568 | SetPageUptodate(newpage); |
2aca711b | 1569 | set_page_private(newpage, entry.val); |
bde05d1c HD |
1570 | SetPageSwapCache(newpage); |
1571 | ||
1572 | /* | |
1573 | * Our caller will very soon move newpage out of swapcache, but it's | |
1574 | * a nice clean interface for us to replace oldpage by newpage there. | |
1575 | */ | |
1576 | spin_lock_irq(&swap_mapping->tree_lock); | |
1577 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1578 | newpage); | |
0142ef6c | 1579 | if (!error) { |
11fb9989 MG |
1580 | __inc_node_page_state(newpage, NR_FILE_PAGES); |
1581 | __dec_node_page_state(oldpage, NR_FILE_PAGES); | |
0142ef6c | 1582 | } |
bde05d1c | 1583 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1584 | |
0142ef6c HD |
1585 | if (unlikely(error)) { |
1586 | /* | |
1587 | * Is this possible? I think not, now that our callers check | |
1588 | * both PageSwapCache and page_private after getting page lock; | |
1589 | * but be defensive. Reverse old to newpage for clear and free. | |
1590 | */ | |
1591 | oldpage = newpage; | |
1592 | } else { | |
6a93ca8f | 1593 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1594 | lru_cache_add_anon(newpage); |
1595 | *pagep = newpage; | |
1596 | } | |
bde05d1c HD |
1597 | |
1598 | ClearPageSwapCache(oldpage); | |
1599 | set_page_private(oldpage, 0); | |
1600 | ||
1601 | unlock_page(oldpage); | |
09cbfeaf KS |
1602 | put_page(oldpage); |
1603 | put_page(oldpage); | |
0142ef6c | 1604 | return error; |
bde05d1c HD |
1605 | } |
1606 | ||
1da177e4 | 1607 | /* |
68da9f05 | 1608 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1609 | * |
1610 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1611 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1612 | * entry since a page cannot live in both the swap and page cache. |
1613 | * | |
1614 | * fault_mm and fault_type are only supplied by shmem_fault: | |
1615 | * otherwise they are NULL. | |
1da177e4 | 1616 | */ |
41ffe5d5 | 1617 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 1618 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
cfda0526 | 1619 | struct vm_area_struct *vma, struct vm_fault *vmf, int *fault_type) |
1da177e4 LT |
1620 | { |
1621 | struct address_space *mapping = inode->i_mapping; | |
23f919d4 | 1622 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 | 1623 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1624 | struct mm_struct *charge_mm; |
00501b53 | 1625 | struct mem_cgroup *memcg; |
27ab7006 | 1626 | struct page *page; |
1da177e4 | 1627 | swp_entry_t swap; |
657e3038 | 1628 | enum sgp_type sgp_huge = sgp; |
800d8c63 | 1629 | pgoff_t hindex = index; |
1da177e4 | 1630 | int error; |
54af6042 | 1631 | int once = 0; |
1635f6a7 | 1632 | int alloced = 0; |
1da177e4 | 1633 | |
09cbfeaf | 1634 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1635 | return -EFBIG; |
657e3038 KS |
1636 | if (sgp == SGP_NOHUGE || sgp == SGP_HUGE) |
1637 | sgp = SGP_CACHE; | |
1da177e4 | 1638 | repeat: |
54af6042 | 1639 | swap.val = 0; |
0cd6144a | 1640 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1641 | if (radix_tree_exceptional_entry(page)) { |
1642 | swap = radix_to_swp_entry(page); | |
1643 | page = NULL; | |
1644 | } | |
1645 | ||
75edd345 | 1646 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1647 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 1648 | error = -EINVAL; |
267a4c76 | 1649 | goto unlock; |
54af6042 HD |
1650 | } |
1651 | ||
66d2f4d2 HD |
1652 | if (page && sgp == SGP_WRITE) |
1653 | mark_page_accessed(page); | |
1654 | ||
1635f6a7 HD |
1655 | /* fallocated page? */ |
1656 | if (page && !PageUptodate(page)) { | |
1657 | if (sgp != SGP_READ) | |
1658 | goto clear; | |
1659 | unlock_page(page); | |
09cbfeaf | 1660 | put_page(page); |
1635f6a7 HD |
1661 | page = NULL; |
1662 | } | |
54af6042 | 1663 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1664 | *pagep = page; |
1665 | return 0; | |
27ab7006 HD |
1666 | } |
1667 | ||
1668 | /* | |
54af6042 HD |
1669 | * Fast cache lookup did not find it: |
1670 | * bring it back from swap or allocate. | |
27ab7006 | 1671 | */ |
54af6042 | 1672 | sbinfo = SHMEM_SB(inode->i_sb); |
cfda0526 | 1673 | charge_mm = vma ? vma->vm_mm : current->mm; |
1da177e4 | 1674 | |
1da177e4 LT |
1675 | if (swap.val) { |
1676 | /* Look it up and read it in.. */ | |
ec560175 | 1677 | page = lookup_swap_cache(swap, NULL, 0); |
27ab7006 | 1678 | if (!page) { |
9e18eb29 ALC |
1679 | /* Or update major stats only when swapin succeeds?? */ |
1680 | if (fault_type) { | |
68da9f05 | 1681 | *fault_type |= VM_FAULT_MAJOR; |
9e18eb29 | 1682 | count_vm_event(PGMAJFAULT); |
2262185c | 1683 | count_memcg_event_mm(charge_mm, PGMAJFAULT); |
9e18eb29 ALC |
1684 | } |
1685 | /* Here we actually start the io */ | |
41ffe5d5 | 1686 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1687 | if (!page) { |
54af6042 HD |
1688 | error = -ENOMEM; |
1689 | goto failed; | |
1da177e4 | 1690 | } |
1da177e4 LT |
1691 | } |
1692 | ||
1693 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1694 | lock_page(page); |
0142ef6c | 1695 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1696 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1697 | error = -EEXIST; /* try again */ |
d1899228 | 1698 | goto unlock; |
bde05d1c | 1699 | } |
27ab7006 | 1700 | if (!PageUptodate(page)) { |
1da177e4 | 1701 | error = -EIO; |
54af6042 | 1702 | goto failed; |
1da177e4 | 1703 | } |
54af6042 HD |
1704 | wait_on_page_writeback(page); |
1705 | ||
bde05d1c HD |
1706 | if (shmem_should_replace_page(page, gfp)) { |
1707 | error = shmem_replace_page(&page, gfp, info, index); | |
1708 | if (error) | |
1709 | goto failed; | |
1da177e4 | 1710 | } |
27ab7006 | 1711 | |
9e18eb29 | 1712 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
f627c2f5 | 1713 | false); |
d1899228 | 1714 | if (!error) { |
aa3b1895 | 1715 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1716 | swp_to_radix_entry(swap)); |
215c02bc HD |
1717 | /* |
1718 | * We already confirmed swap under page lock, and make | |
1719 | * no memory allocation here, so usually no possibility | |
1720 | * of error; but free_swap_and_cache() only trylocks a | |
1721 | * page, so it is just possible that the entry has been | |
1722 | * truncated or holepunched since swap was confirmed. | |
1723 | * shmem_undo_range() will have done some of the | |
1724 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1725 | * the rest. |
215c02bc HD |
1726 | * Reset swap.val? No, leave it so "failed" goes back to |
1727 | * "repeat": reading a hole and writing should succeed. | |
1728 | */ | |
00501b53 | 1729 | if (error) { |
f627c2f5 | 1730 | mem_cgroup_cancel_charge(page, memcg, false); |
215c02bc | 1731 | delete_from_swap_cache(page); |
00501b53 | 1732 | } |
d1899228 | 1733 | } |
54af6042 HD |
1734 | if (error) |
1735 | goto failed; | |
1736 | ||
f627c2f5 | 1737 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1738 | |
4595ef88 | 1739 | spin_lock_irq(&info->lock); |
285b2c4f | 1740 | info->swapped--; |
54af6042 | 1741 | shmem_recalc_inode(inode); |
4595ef88 | 1742 | spin_unlock_irq(&info->lock); |
54af6042 | 1743 | |
66d2f4d2 HD |
1744 | if (sgp == SGP_WRITE) |
1745 | mark_page_accessed(page); | |
1746 | ||
54af6042 | 1747 | delete_from_swap_cache(page); |
27ab7006 HD |
1748 | set_page_dirty(page); |
1749 | swap_free(swap); | |
1750 | ||
54af6042 | 1751 | } else { |
cfda0526 MR |
1752 | if (vma && userfaultfd_missing(vma)) { |
1753 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
1754 | return 0; | |
1755 | } | |
1756 | ||
800d8c63 KS |
1757 | /* shmem_symlink() */ |
1758 | if (mapping->a_ops != &shmem_aops) | |
1759 | goto alloc_nohuge; | |
657e3038 | 1760 | if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE) |
800d8c63 KS |
1761 | goto alloc_nohuge; |
1762 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1763 | goto alloc_huge; | |
1764 | switch (sbinfo->huge) { | |
1765 | loff_t i_size; | |
1766 | pgoff_t off; | |
1767 | case SHMEM_HUGE_NEVER: | |
1768 | goto alloc_nohuge; | |
1769 | case SHMEM_HUGE_WITHIN_SIZE: | |
1770 | off = round_up(index, HPAGE_PMD_NR); | |
1771 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1772 | if (i_size >= HPAGE_PMD_SIZE && | |
1773 | i_size >> PAGE_SHIFT >= off) | |
1774 | goto alloc_huge; | |
1775 | /* fallthrough */ | |
1776 | case SHMEM_HUGE_ADVISE: | |
657e3038 KS |
1777 | if (sgp_huge == SGP_HUGE) |
1778 | goto alloc_huge; | |
1779 | /* TODO: implement fadvise() hints */ | |
800d8c63 | 1780 | goto alloc_nohuge; |
54af6042 | 1781 | } |
1da177e4 | 1782 | |
800d8c63 | 1783 | alloc_huge: |
0f079694 | 1784 | page = shmem_alloc_and_acct_page(gfp, inode, index, true); |
800d8c63 | 1785 | if (IS_ERR(page)) { |
0f079694 | 1786 | alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, inode, |
800d8c63 | 1787 | index, false); |
1da177e4 | 1788 | } |
800d8c63 | 1789 | if (IS_ERR(page)) { |
779750d2 | 1790 | int retry = 5; |
800d8c63 KS |
1791 | error = PTR_ERR(page); |
1792 | page = NULL; | |
779750d2 KS |
1793 | if (error != -ENOSPC) |
1794 | goto failed; | |
1795 | /* | |
1796 | * Try to reclaim some spece by splitting a huge page | |
1797 | * beyond i_size on the filesystem. | |
1798 | */ | |
1799 | while (retry--) { | |
1800 | int ret; | |
1801 | ret = shmem_unused_huge_shrink(sbinfo, NULL, 1); | |
1802 | if (ret == SHRINK_STOP) | |
1803 | break; | |
1804 | if (ret) | |
1805 | goto alloc_nohuge; | |
1806 | } | |
800d8c63 KS |
1807 | goto failed; |
1808 | } | |
1809 | ||
1810 | if (PageTransHuge(page)) | |
1811 | hindex = round_down(index, HPAGE_PMD_NR); | |
1812 | else | |
1813 | hindex = index; | |
1814 | ||
66d2f4d2 | 1815 | if (sgp == SGP_WRITE) |
eb39d618 | 1816 | __SetPageReferenced(page); |
66d2f4d2 | 1817 | |
9e18eb29 | 1818 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
800d8c63 | 1819 | PageTransHuge(page)); |
54af6042 | 1820 | if (error) |
800d8c63 KS |
1821 | goto unacct; |
1822 | error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK, | |
1823 | compound_order(page)); | |
b065b432 | 1824 | if (!error) { |
800d8c63 | 1825 | error = shmem_add_to_page_cache(page, mapping, hindex, |
fed400a1 | 1826 | NULL); |
b065b432 HD |
1827 | radix_tree_preload_end(); |
1828 | } | |
1829 | if (error) { | |
800d8c63 KS |
1830 | mem_cgroup_cancel_charge(page, memcg, |
1831 | PageTransHuge(page)); | |
1832 | goto unacct; | |
b065b432 | 1833 | } |
800d8c63 KS |
1834 | mem_cgroup_commit_charge(page, memcg, false, |
1835 | PageTransHuge(page)); | |
54af6042 HD |
1836 | lru_cache_add_anon(page); |
1837 | ||
4595ef88 | 1838 | spin_lock_irq(&info->lock); |
800d8c63 KS |
1839 | info->alloced += 1 << compound_order(page); |
1840 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); | |
54af6042 | 1841 | shmem_recalc_inode(inode); |
4595ef88 | 1842 | spin_unlock_irq(&info->lock); |
1635f6a7 | 1843 | alloced = true; |
54af6042 | 1844 | |
779750d2 KS |
1845 | if (PageTransHuge(page) && |
1846 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < | |
1847 | hindex + HPAGE_PMD_NR - 1) { | |
1848 | /* | |
1849 | * Part of the huge page is beyond i_size: subject | |
1850 | * to shrink under memory pressure. | |
1851 | */ | |
1852 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1853 | /* |
1854 | * _careful to defend against unlocked access to | |
1855 | * ->shrink_list in shmem_unused_huge_shrink() | |
1856 | */ | |
1857 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1858 | list_add_tail(&info->shrinklist, |
1859 | &sbinfo->shrinklist); | |
1860 | sbinfo->shrinklist_len++; | |
1861 | } | |
1862 | spin_unlock(&sbinfo->shrinklist_lock); | |
1863 | } | |
1864 | ||
ec9516fb | 1865 | /* |
1635f6a7 HD |
1866 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1867 | */ | |
1868 | if (sgp == SGP_FALLOC) | |
1869 | sgp = SGP_WRITE; | |
1870 | clear: | |
1871 | /* | |
1872 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1873 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1874 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb | 1875 | */ |
800d8c63 KS |
1876 | if (sgp != SGP_WRITE && !PageUptodate(page)) { |
1877 | struct page *head = compound_head(page); | |
1878 | int i; | |
1879 | ||
1880 | for (i = 0; i < (1 << compound_order(head)); i++) { | |
1881 | clear_highpage(head + i); | |
1882 | flush_dcache_page(head + i); | |
1883 | } | |
1884 | SetPageUptodate(head); | |
ec9516fb | 1885 | } |
1da177e4 | 1886 | } |
bde05d1c | 1887 | |
54af6042 | 1888 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1889 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1890 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1891 | if (alloced) { |
1892 | ClearPageDirty(page); | |
1893 | delete_from_page_cache(page); | |
4595ef88 | 1894 | spin_lock_irq(&info->lock); |
267a4c76 | 1895 | shmem_recalc_inode(inode); |
4595ef88 | 1896 | spin_unlock_irq(&info->lock); |
267a4c76 | 1897 | } |
54af6042 | 1898 | error = -EINVAL; |
267a4c76 | 1899 | goto unlock; |
e83c32e8 | 1900 | } |
800d8c63 | 1901 | *pagep = page + index - hindex; |
54af6042 | 1902 | return 0; |
1da177e4 | 1903 | |
59a16ead | 1904 | /* |
54af6042 | 1905 | * Error recovery. |
59a16ead | 1906 | */ |
54af6042 | 1907 | unacct: |
0f079694 | 1908 | shmem_inode_unacct_blocks(inode, 1 << compound_order(page)); |
800d8c63 KS |
1909 | |
1910 | if (PageTransHuge(page)) { | |
1911 | unlock_page(page); | |
1912 | put_page(page); | |
1913 | goto alloc_nohuge; | |
1914 | } | |
54af6042 | 1915 | failed: |
267a4c76 | 1916 | if (swap.val && !shmem_confirm_swap(mapping, index, swap)) |
d1899228 HD |
1917 | error = -EEXIST; |
1918 | unlock: | |
27ab7006 | 1919 | if (page) { |
54af6042 | 1920 | unlock_page(page); |
09cbfeaf | 1921 | put_page(page); |
54af6042 HD |
1922 | } |
1923 | if (error == -ENOSPC && !once++) { | |
4595ef88 | 1924 | spin_lock_irq(&info->lock); |
54af6042 | 1925 | shmem_recalc_inode(inode); |
4595ef88 | 1926 | spin_unlock_irq(&info->lock); |
27ab7006 | 1927 | goto repeat; |
ff36b801 | 1928 | } |
d1899228 | 1929 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1930 | goto repeat; |
1931 | return error; | |
1da177e4 LT |
1932 | } |
1933 | ||
10d20bd2 LT |
1934 | /* |
1935 | * This is like autoremove_wake_function, but it removes the wait queue | |
1936 | * entry unconditionally - even if something else had already woken the | |
1937 | * target. | |
1938 | */ | |
ac6424b9 | 1939 | static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
10d20bd2 LT |
1940 | { |
1941 | int ret = default_wake_function(wait, mode, sync, key); | |
2055da97 | 1942 | list_del_init(&wait->entry); |
10d20bd2 LT |
1943 | return ret; |
1944 | } | |
1945 | ||
11bac800 | 1946 | static int shmem_fault(struct vm_fault *vmf) |
1da177e4 | 1947 | { |
11bac800 | 1948 | struct vm_area_struct *vma = vmf->vma; |
496ad9aa | 1949 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 1950 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
657e3038 | 1951 | enum sgp_type sgp; |
1da177e4 | 1952 | int error; |
68da9f05 | 1953 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1954 | |
f00cdc6d HD |
1955 | /* |
1956 | * Trinity finds that probing a hole which tmpfs is punching can | |
1957 | * prevent the hole-punch from ever completing: which in turn | |
1958 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1959 | * faulting pages into the hole while it's being punched. Although |
1960 | * shmem_undo_range() does remove the additions, it may be unable to | |
1961 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1962 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1963 | * | |
1964 | * It does not matter if we sometimes reach this check just before the | |
1965 | * hole-punch begins, so that one fault then races with the punch: | |
1966 | * we just need to make racing faults a rare case. | |
1967 | * | |
1968 | * The implementation below would be much simpler if we just used a | |
1969 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1970 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1971 | */ |
1972 | if (unlikely(inode->i_private)) { | |
1973 | struct shmem_falloc *shmem_falloc; | |
1974 | ||
1975 | spin_lock(&inode->i_lock); | |
1976 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1977 | if (shmem_falloc && |
1978 | shmem_falloc->waitq && | |
1979 | vmf->pgoff >= shmem_falloc->start && | |
1980 | vmf->pgoff < shmem_falloc->next) { | |
1981 | wait_queue_head_t *shmem_falloc_waitq; | |
10d20bd2 | 1982 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); |
8e205f77 HD |
1983 | |
1984 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1985 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1986 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1987 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1988 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1989 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1990 | } |
8e205f77 HD |
1991 | |
1992 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1993 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1994 | TASK_UNINTERRUPTIBLE); | |
1995 | spin_unlock(&inode->i_lock); | |
1996 | schedule(); | |
1997 | ||
1998 | /* | |
1999 | * shmem_falloc_waitq points into the shmem_fallocate() | |
2000 | * stack of the hole-punching task: shmem_falloc_waitq | |
2001 | * is usually invalid by the time we reach here, but | |
2002 | * finish_wait() does not dereference it in that case; | |
2003 | * though i_lock needed lest racing with wake_up_all(). | |
2004 | */ | |
2005 | spin_lock(&inode->i_lock); | |
2006 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
2007 | spin_unlock(&inode->i_lock); | |
2008 | return ret; | |
f00cdc6d | 2009 | } |
8e205f77 | 2010 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
2011 | } |
2012 | ||
657e3038 | 2013 | sgp = SGP_CACHE; |
18600332 MH |
2014 | |
2015 | if ((vma->vm_flags & VM_NOHUGEPAGE) || | |
2016 | test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) | |
657e3038 | 2017 | sgp = SGP_NOHUGE; |
18600332 MH |
2018 | else if (vma->vm_flags & VM_HUGEPAGE) |
2019 | sgp = SGP_HUGE; | |
657e3038 KS |
2020 | |
2021 | error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp, | |
cfda0526 | 2022 | gfp, vma, vmf, &ret); |
d0217ac0 NP |
2023 | if (error) |
2024 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 2025 | return ret; |
1da177e4 LT |
2026 | } |
2027 | ||
c01d5b30 HD |
2028 | unsigned long shmem_get_unmapped_area(struct file *file, |
2029 | unsigned long uaddr, unsigned long len, | |
2030 | unsigned long pgoff, unsigned long flags) | |
2031 | { | |
2032 | unsigned long (*get_area)(struct file *, | |
2033 | unsigned long, unsigned long, unsigned long, unsigned long); | |
2034 | unsigned long addr; | |
2035 | unsigned long offset; | |
2036 | unsigned long inflated_len; | |
2037 | unsigned long inflated_addr; | |
2038 | unsigned long inflated_offset; | |
2039 | ||
2040 | if (len > TASK_SIZE) | |
2041 | return -ENOMEM; | |
2042 | ||
2043 | get_area = current->mm->get_unmapped_area; | |
2044 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2045 | ||
e496cf3d | 2046 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
c01d5b30 HD |
2047 | return addr; |
2048 | if (IS_ERR_VALUE(addr)) | |
2049 | return addr; | |
2050 | if (addr & ~PAGE_MASK) | |
2051 | return addr; | |
2052 | if (addr > TASK_SIZE - len) | |
2053 | return addr; | |
2054 | ||
2055 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2056 | return addr; | |
2057 | if (len < HPAGE_PMD_SIZE) | |
2058 | return addr; | |
2059 | if (flags & MAP_FIXED) | |
2060 | return addr; | |
2061 | /* | |
2062 | * Our priority is to support MAP_SHARED mapped hugely; | |
2063 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
da6c5fad KS |
2064 | * But if caller specified an address hint and we allocated area there |
2065 | * successfully, respect that as before. | |
c01d5b30 | 2066 | */ |
da6c5fad | 2067 | if (uaddr == addr) |
c01d5b30 HD |
2068 | return addr; |
2069 | ||
2070 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2071 | struct super_block *sb; | |
2072 | ||
2073 | if (file) { | |
2074 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2075 | sb = file_inode(file)->i_sb; | |
2076 | } else { | |
2077 | /* | |
2078 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2079 | * for "/dev/zero", to create a shared anonymous object. | |
2080 | */ | |
2081 | if (IS_ERR(shm_mnt)) | |
2082 | return addr; | |
2083 | sb = shm_mnt->mnt_sb; | |
2084 | } | |
3089bf61 | 2085 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2086 | return addr; |
2087 | } | |
2088 | ||
2089 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2090 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2091 | return addr; | |
2092 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2093 | return addr; | |
2094 | ||
2095 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2096 | if (inflated_len > TASK_SIZE) | |
2097 | return addr; | |
2098 | if (inflated_len < len) | |
2099 | return addr; | |
2100 | ||
da6c5fad | 2101 | inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags); |
c01d5b30 HD |
2102 | if (IS_ERR_VALUE(inflated_addr)) |
2103 | return addr; | |
2104 | if (inflated_addr & ~PAGE_MASK) | |
2105 | return addr; | |
2106 | ||
2107 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2108 | inflated_addr += offset - inflated_offset; | |
2109 | if (inflated_offset > offset) | |
2110 | inflated_addr += HPAGE_PMD_SIZE; | |
2111 | ||
2112 | if (inflated_addr > TASK_SIZE - len) | |
2113 | return addr; | |
2114 | return inflated_addr; | |
2115 | } | |
2116 | ||
1da177e4 | 2117 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2118 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2119 | { |
496ad9aa | 2120 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2121 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2122 | } |
2123 | ||
d8dc74f2 AB |
2124 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
2125 | unsigned long addr) | |
1da177e4 | 2126 | { |
496ad9aa | 2127 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2128 | pgoff_t index; |
1da177e4 | 2129 | |
41ffe5d5 HD |
2130 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2131 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
2132 | } |
2133 | #endif | |
2134 | ||
2135 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
2136 | { | |
496ad9aa | 2137 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2138 | struct shmem_inode_info *info = SHMEM_I(inode); |
2139 | int retval = -ENOMEM; | |
2140 | ||
4595ef88 | 2141 | spin_lock_irq(&info->lock); |
1da177e4 LT |
2142 | if (lock && !(info->flags & VM_LOCKED)) { |
2143 | if (!user_shm_lock(inode->i_size, user)) | |
2144 | goto out_nomem; | |
2145 | info->flags |= VM_LOCKED; | |
89e004ea | 2146 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
2147 | } |
2148 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
2149 | user_shm_unlock(inode->i_size, user); | |
2150 | info->flags &= ~VM_LOCKED; | |
89e004ea | 2151 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2152 | } |
2153 | retval = 0; | |
89e004ea | 2154 | |
1da177e4 | 2155 | out_nomem: |
4595ef88 | 2156 | spin_unlock_irq(&info->lock); |
1da177e4 LT |
2157 | return retval; |
2158 | } | |
2159 | ||
9b83a6a8 | 2160 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
2161 | { |
2162 | file_accessed(file); | |
2163 | vma->vm_ops = &shmem_vm_ops; | |
e496cf3d | 2164 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
2165 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
2166 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
2167 | khugepaged_enter(vma, vma->vm_flags); | |
2168 | } | |
1da177e4 LT |
2169 | return 0; |
2170 | } | |
2171 | ||
454abafe | 2172 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 2173 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
2174 | { |
2175 | struct inode *inode; | |
2176 | struct shmem_inode_info *info; | |
2177 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
c088e31d | 2178 | int ino; |
1da177e4 | 2179 | |
5b04c689 PE |
2180 | if (shmem_reserve_inode(sb)) |
2181 | return NULL; | |
1da177e4 LT |
2182 | |
2183 | inode = new_inode(sb); | |
2184 | if (inode) { | |
454abafe | 2185 | inode_init_owner(inode, dir, mode); |
1da177e4 | 2186 | inode->i_blocks = 0; |
078cd827 | 2187 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
91828a40 | 2188 | inode->i_generation = get_seconds(); |
1da177e4 LT |
2189 | info = SHMEM_I(inode); |
2190 | memset(info, 0, (char *)inode - (char *)info); | |
2191 | spin_lock_init(&info->lock); | |
40e041a2 | 2192 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 2193 | info->flags = flags & VM_NORESERVE; |
779750d2 | 2194 | INIT_LIST_HEAD(&info->shrinklist); |
1da177e4 | 2195 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 2196 | simple_xattrs_init(&info->xattrs); |
72c04902 | 2197 | cache_no_acl(inode); |
1da177e4 LT |
2198 | |
2199 | switch (mode & S_IFMT) { | |
2200 | default: | |
39f0247d | 2201 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
2202 | init_special_inode(inode, mode, dev); |
2203 | break; | |
2204 | case S_IFREG: | |
14fcc23f | 2205 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
2206 | inode->i_op = &shmem_inode_operations; |
2207 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
2208 | mpol_shared_policy_init(&info->policy, |
2209 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
2210 | break; |
2211 | case S_IFDIR: | |
d8c76e6f | 2212 | inc_nlink(inode); |
1da177e4 LT |
2213 | /* Some things misbehave if size == 0 on a directory */ |
2214 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2215 | inode->i_op = &shmem_dir_inode_operations; | |
2216 | inode->i_fop = &simple_dir_operations; | |
2217 | break; | |
2218 | case S_IFLNK: | |
2219 | /* | |
2220 | * Must not load anything in the rbtree, | |
2221 | * mpol_free_shared_policy will not be called. | |
2222 | */ | |
71fe804b | 2223 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
2224 | break; |
2225 | } | |
c088e31d | 2226 | |
0a496dec JFG |
2227 | lockdep_annotate_inode_mutex_key(inode); |
2228 | ||
c088e31d SF |
2229 | if (!sbinfo->idr_nouse) { |
2230 | /* inum 0 and 1 are unused */ | |
2231 | mutex_lock(&sbinfo->idr_lock); | |
2232 | ino = idr_alloc(&sbinfo->idr, inode, 2, INT_MAX, | |
2233 | GFP_NOFS); | |
2234 | if (ino > 0) { | |
2235 | inode->i_ino = ino; | |
2236 | mutex_unlock(&sbinfo->idr_lock); | |
2237 | __insert_inode_hash(inode, inode->i_ino); | |
2238 | } else { | |
2239 | inode->i_ino = 0; | |
2240 | mutex_unlock(&sbinfo->idr_lock); | |
2241 | iput(inode); | |
2242 | /* shmem_free_inode() will be called */ | |
2243 | inode = NULL; | |
2244 | } | |
2245 | } else | |
2246 | inode->i_ino = get_next_ino(); | |
5b04c689 PE |
2247 | } else |
2248 | shmem_free_inode(sb); | |
1da177e4 LT |
2249 | return inode; |
2250 | } | |
2251 | ||
0cd6144a JW |
2252 | bool shmem_mapping(struct address_space *mapping) |
2253 | { | |
f8005451 | 2254 | return mapping->a_ops == &shmem_aops; |
0cd6144a JW |
2255 | } |
2256 | ||
8d103963 MR |
2257 | static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm, |
2258 | pmd_t *dst_pmd, | |
2259 | struct vm_area_struct *dst_vma, | |
2260 | unsigned long dst_addr, | |
2261 | unsigned long src_addr, | |
2262 | bool zeropage, | |
2263 | struct page **pagep) | |
4c27fe4c MR |
2264 | { |
2265 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2266 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4c27fe4c MR |
2267 | struct address_space *mapping = inode->i_mapping; |
2268 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2269 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
2270 | struct mem_cgroup *memcg; | |
2271 | spinlock_t *ptl; | |
2272 | void *page_kaddr; | |
2273 | struct page *page; | |
2274 | pte_t _dst_pte, *dst_pte; | |
2275 | int ret; | |
fdd65e88 | 2276 | pgoff_t offset, max_off; |
4c27fe4c | 2277 | |
cb658a45 | 2278 | ret = -ENOMEM; |
0f079694 | 2279 | if (!shmem_inode_acct_block(inode, 1)) |
cb658a45 | 2280 | goto out; |
4c27fe4c | 2281 | |
cb658a45 | 2282 | if (!*pagep) { |
4c27fe4c MR |
2283 | page = shmem_alloc_page(gfp, info, pgoff); |
2284 | if (!page) | |
0f079694 | 2285 | goto out_unacct_blocks; |
4c27fe4c | 2286 | |
8d103963 MR |
2287 | if (!zeropage) { /* mcopy_atomic */ |
2288 | page_kaddr = kmap_atomic(page); | |
2289 | ret = copy_from_user(page_kaddr, | |
2290 | (const void __user *)src_addr, | |
2291 | PAGE_SIZE); | |
2292 | kunmap_atomic(page_kaddr); | |
2293 | ||
2294 | /* fallback to copy_from_user outside mmap_sem */ | |
2295 | if (unlikely(ret)) { | |
2296 | *pagep = page; | |
2297 | shmem_inode_unacct_blocks(inode, 1); | |
2298 | /* don't free the page */ | |
c9ddac33 | 2299 | return -ENOENT; |
8d103963 MR |
2300 | } |
2301 | } else { /* mfill_zeropage_atomic */ | |
2302 | clear_highpage(page); | |
4c27fe4c MR |
2303 | } |
2304 | } else { | |
2305 | page = *pagep; | |
2306 | *pagep = NULL; | |
2307 | } | |
2308 | ||
9cc90c66 AA |
2309 | VM_BUG_ON(PageLocked(page) || PageSwapBacked(page)); |
2310 | __SetPageLocked(page); | |
2311 | __SetPageSwapBacked(page); | |
a425d358 | 2312 | __SetPageUptodate(page); |
9cc90c66 | 2313 | |
fdd65e88 AA |
2314 | ret = -EFAULT; |
2315 | offset = linear_page_index(dst_vma, dst_addr); | |
2316 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); | |
2317 | if (unlikely(offset >= max_off)) | |
2318 | goto out_release; | |
2319 | ||
4c27fe4c MR |
2320 | ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false); |
2321 | if (ret) | |
2322 | goto out_release; | |
2323 | ||
2324 | ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK); | |
2325 | if (!ret) { | |
2326 | ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL); | |
2327 | radix_tree_preload_end(); | |
2328 | } | |
2329 | if (ret) | |
2330 | goto out_release_uncharge; | |
2331 | ||
2332 | mem_cgroup_commit_charge(page, memcg, false, false); | |
2333 | ||
2334 | _dst_pte = mk_pte(page, dst_vma->vm_page_prot); | |
2335 | if (dst_vma->vm_flags & VM_WRITE) | |
2336 | _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte)); | |
b616e2d4 AA |
2337 | else { |
2338 | /* | |
2339 | * We don't set the pte dirty if the vma has no | |
2340 | * VM_WRITE permission, so mark the page dirty or it | |
2341 | * could be freed from under us. We could do it | |
2342 | * unconditionally before unlock_page(), but doing it | |
2343 | * only if VM_WRITE is not set is faster. | |
2344 | */ | |
2345 | set_page_dirty(page); | |
2346 | } | |
4c27fe4c | 2347 | |
4c27fe4c | 2348 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); |
fdd65e88 AA |
2349 | |
2350 | ret = -EFAULT; | |
2351 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); | |
2352 | if (unlikely(offset >= max_off)) | |
2353 | goto out_release_uncharge_unlock; | |
2354 | ||
2355 | ret = -EEXIST; | |
4c27fe4c MR |
2356 | if (!pte_none(*dst_pte)) |
2357 | goto out_release_uncharge_unlock; | |
2358 | ||
4c27fe4c MR |
2359 | lru_cache_add_anon(page); |
2360 | ||
2361 | spin_lock(&info->lock); | |
2362 | info->alloced++; | |
2363 | inode->i_blocks += BLOCKS_PER_PAGE; | |
2364 | shmem_recalc_inode(inode); | |
2365 | spin_unlock(&info->lock); | |
2366 | ||
2367 | inc_mm_counter(dst_mm, mm_counter_file(page)); | |
2368 | page_add_file_rmap(page, false); | |
2369 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
2370 | ||
2371 | /* No need to invalidate - it was non-present before */ | |
2372 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
4c27fe4c | 2373 | pte_unmap_unlock(dst_pte, ptl); |
fdd65e88 | 2374 | unlock_page(page); |
4c27fe4c MR |
2375 | ret = 0; |
2376 | out: | |
2377 | return ret; | |
2378 | out_release_uncharge_unlock: | |
2379 | pte_unmap_unlock(dst_pte, ptl); | |
b616e2d4 | 2380 | ClearPageDirty(page); |
fdd65e88 | 2381 | delete_from_page_cache(page); |
4c27fe4c MR |
2382 | out_release_uncharge: |
2383 | mem_cgroup_cancel_charge(page, memcg, false); | |
2384 | out_release: | |
9cc90c66 | 2385 | unlock_page(page); |
4c27fe4c | 2386 | put_page(page); |
4c27fe4c | 2387 | out_unacct_blocks: |
0f079694 | 2388 | shmem_inode_unacct_blocks(inode, 1); |
4c27fe4c MR |
2389 | goto out; |
2390 | } | |
2391 | ||
8d103963 MR |
2392 | int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm, |
2393 | pmd_t *dst_pmd, | |
2394 | struct vm_area_struct *dst_vma, | |
2395 | unsigned long dst_addr, | |
2396 | unsigned long src_addr, | |
2397 | struct page **pagep) | |
2398 | { | |
2399 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2400 | dst_addr, src_addr, false, pagep); | |
2401 | } | |
2402 | ||
2403 | int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm, | |
2404 | pmd_t *dst_pmd, | |
2405 | struct vm_area_struct *dst_vma, | |
2406 | unsigned long dst_addr) | |
2407 | { | |
2408 | struct page *page = NULL; | |
2409 | ||
2410 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2411 | dst_addr, 0, true, &page); | |
2412 | } | |
2413 | ||
1da177e4 | 2414 | #ifdef CONFIG_TMPFS |
92e1d5be | 2415 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2416 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2417 | |
6d9d88d0 JS |
2418 | #ifdef CONFIG_TMPFS_XATTR |
2419 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2420 | #else | |
2421 | #define shmem_initxattrs NULL | |
2422 | #endif | |
2423 | ||
1da177e4 | 2424 | static int |
800d15a5 NP |
2425 | shmem_write_begin(struct file *file, struct address_space *mapping, |
2426 | loff_t pos, unsigned len, unsigned flags, | |
2427 | struct page **pagep, void **fsdata) | |
1da177e4 | 2428 | { |
800d15a5 | 2429 | struct inode *inode = mapping->host; |
40e041a2 | 2430 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2431 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
2432 | |
2433 | /* i_mutex is held by caller */ | |
3f472cc9 | 2434 | if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) { |
40e041a2 DH |
2435 | if (info->seals & F_SEAL_WRITE) |
2436 | return -EPERM; | |
2437 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2438 | return -EPERM; | |
2439 | } | |
2440 | ||
9e18eb29 | 2441 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
2442 | } |
2443 | ||
2444 | static int | |
2445 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2446 | loff_t pos, unsigned len, unsigned copied, | |
2447 | struct page *page, void *fsdata) | |
2448 | { | |
2449 | struct inode *inode = mapping->host; | |
2450 | ||
d3602444 HD |
2451 | if (pos + copied > inode->i_size) |
2452 | i_size_write(inode, pos + copied); | |
2453 | ||
ec9516fb | 2454 | if (!PageUptodate(page)) { |
800d8c63 KS |
2455 | struct page *head = compound_head(page); |
2456 | if (PageTransCompound(page)) { | |
2457 | int i; | |
2458 | ||
2459 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2460 | if (head + i == page) | |
2461 | continue; | |
2462 | clear_highpage(head + i); | |
2463 | flush_dcache_page(head + i); | |
2464 | } | |
2465 | } | |
09cbfeaf KS |
2466 | if (copied < PAGE_SIZE) { |
2467 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2468 | zero_user_segments(page, 0, from, |
09cbfeaf | 2469 | from + copied, PAGE_SIZE); |
ec9516fb | 2470 | } |
800d8c63 | 2471 | SetPageUptodate(head); |
ec9516fb | 2472 | } |
800d15a5 | 2473 | set_page_dirty(page); |
6746aff7 | 2474 | unlock_page(page); |
09cbfeaf | 2475 | put_page(page); |
800d15a5 | 2476 | |
800d15a5 | 2477 | return copied; |
1da177e4 LT |
2478 | } |
2479 | ||
2ba5bbed | 2480 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2481 | { |
6e58e79d AV |
2482 | struct file *file = iocb->ki_filp; |
2483 | struct inode *inode = file_inode(file); | |
1da177e4 | 2484 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2485 | pgoff_t index; |
2486 | unsigned long offset; | |
a0ee5ec5 | 2487 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2488 | int error = 0; |
cb66a7a1 | 2489 | ssize_t retval = 0; |
6e58e79d | 2490 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2491 | |
2492 | /* | |
2493 | * Might this read be for a stacking filesystem? Then when reading | |
2494 | * holes of a sparse file, we actually need to allocate those pages, | |
2495 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2496 | */ | |
777eda2c | 2497 | if (!iter_is_iovec(to)) |
75edd345 | 2498 | sgp = SGP_CACHE; |
1da177e4 | 2499 | |
09cbfeaf KS |
2500 | index = *ppos >> PAGE_SHIFT; |
2501 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2502 | |
2503 | for (;;) { | |
2504 | struct page *page = NULL; | |
41ffe5d5 HD |
2505 | pgoff_t end_index; |
2506 | unsigned long nr, ret; | |
1da177e4 LT |
2507 | loff_t i_size = i_size_read(inode); |
2508 | ||
09cbfeaf | 2509 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2510 | if (index > end_index) |
2511 | break; | |
2512 | if (index == end_index) { | |
09cbfeaf | 2513 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2514 | if (nr <= offset) |
2515 | break; | |
2516 | } | |
2517 | ||
9e18eb29 | 2518 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2519 | if (error) { |
2520 | if (error == -EINVAL) | |
2521 | error = 0; | |
1da177e4 LT |
2522 | break; |
2523 | } | |
75edd345 HD |
2524 | if (page) { |
2525 | if (sgp == SGP_CACHE) | |
2526 | set_page_dirty(page); | |
d3602444 | 2527 | unlock_page(page); |
75edd345 | 2528 | } |
1da177e4 LT |
2529 | |
2530 | /* | |
2531 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2532 | * are called without i_mutex protection against truncate |
1da177e4 | 2533 | */ |
09cbfeaf | 2534 | nr = PAGE_SIZE; |
1da177e4 | 2535 | i_size = i_size_read(inode); |
09cbfeaf | 2536 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2537 | if (index == end_index) { |
09cbfeaf | 2538 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2539 | if (nr <= offset) { |
2540 | if (page) | |
09cbfeaf | 2541 | put_page(page); |
1da177e4 LT |
2542 | break; |
2543 | } | |
2544 | } | |
2545 | nr -= offset; | |
2546 | ||
2547 | if (page) { | |
2548 | /* | |
2549 | * If users can be writing to this page using arbitrary | |
2550 | * virtual addresses, take care about potential aliasing | |
2551 | * before reading the page on the kernel side. | |
2552 | */ | |
2553 | if (mapping_writably_mapped(mapping)) | |
2554 | flush_dcache_page(page); | |
2555 | /* | |
2556 | * Mark the page accessed if we read the beginning. | |
2557 | */ | |
2558 | if (!offset) | |
2559 | mark_page_accessed(page); | |
b5810039 | 2560 | } else { |
1da177e4 | 2561 | page = ZERO_PAGE(0); |
09cbfeaf | 2562 | get_page(page); |
b5810039 | 2563 | } |
1da177e4 LT |
2564 | |
2565 | /* | |
2566 | * Ok, we have the page, and it's up-to-date, so | |
2567 | * now we can copy it to user space... | |
1da177e4 | 2568 | */ |
2ba5bbed | 2569 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2570 | retval += ret; |
1da177e4 | 2571 | offset += ret; |
09cbfeaf KS |
2572 | index += offset >> PAGE_SHIFT; |
2573 | offset &= ~PAGE_MASK; | |
1da177e4 | 2574 | |
09cbfeaf | 2575 | put_page(page); |
2ba5bbed | 2576 | if (!iov_iter_count(to)) |
1da177e4 | 2577 | break; |
6e58e79d AV |
2578 | if (ret < nr) { |
2579 | error = -EFAULT; | |
2580 | break; | |
2581 | } | |
1da177e4 LT |
2582 | cond_resched(); |
2583 | } | |
2584 | ||
09cbfeaf | 2585 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2586 | file_accessed(file); |
2587 | return retval ? retval : error; | |
1da177e4 LT |
2588 | } |
2589 | ||
220f2ac9 HD |
2590 | /* |
2591 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
2592 | */ | |
2593 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2594 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2595 | { |
2596 | struct page *page; | |
2597 | struct pagevec pvec; | |
2598 | pgoff_t indices[PAGEVEC_SIZE]; | |
2599 | bool done = false; | |
2600 | int i; | |
2601 | ||
86679820 | 2602 | pagevec_init(&pvec); |
220f2ac9 HD |
2603 | pvec.nr = 1; /* start small: we may be there already */ |
2604 | while (!done) { | |
0cd6144a | 2605 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2606 | pvec.nr, pvec.pages, indices); |
2607 | if (!pvec.nr) { | |
965c8e59 | 2608 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2609 | index = end; |
2610 | break; | |
2611 | } | |
2612 | for (i = 0; i < pvec.nr; i++, index++) { | |
2613 | if (index < indices[i]) { | |
965c8e59 | 2614 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2615 | done = true; |
2616 | break; | |
2617 | } | |
2618 | index = indices[i]; | |
2619 | } | |
2620 | page = pvec.pages[i]; | |
2621 | if (page && !radix_tree_exceptional_entry(page)) { | |
2622 | if (!PageUptodate(page)) | |
2623 | page = NULL; | |
2624 | } | |
2625 | if (index >= end || | |
965c8e59 AM |
2626 | (page && whence == SEEK_DATA) || |
2627 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2628 | done = true; |
2629 | break; | |
2630 | } | |
2631 | } | |
0cd6144a | 2632 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2633 | pagevec_release(&pvec); |
2634 | pvec.nr = PAGEVEC_SIZE; | |
2635 | cond_resched(); | |
2636 | } | |
2637 | return index; | |
2638 | } | |
2639 | ||
965c8e59 | 2640 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2641 | { |
2642 | struct address_space *mapping = file->f_mapping; | |
2643 | struct inode *inode = mapping->host; | |
2644 | pgoff_t start, end; | |
2645 | loff_t new_offset; | |
2646 | ||
965c8e59 AM |
2647 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2648 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2649 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2650 | inode_lock(inode); |
220f2ac9 HD |
2651 | /* We're holding i_mutex so we can access i_size directly */ |
2652 | ||
442703d2 | 2653 | if (offset < 0 || offset >= inode->i_size) |
220f2ac9 HD |
2654 | offset = -ENXIO; |
2655 | else { | |
09cbfeaf KS |
2656 | start = offset >> PAGE_SHIFT; |
2657 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2658 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2659 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2660 | if (new_offset > offset) { |
2661 | if (new_offset < inode->i_size) | |
2662 | offset = new_offset; | |
965c8e59 | 2663 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2664 | offset = -ENXIO; |
2665 | else | |
2666 | offset = inode->i_size; | |
2667 | } | |
2668 | } | |
2669 | ||
387aae6f HD |
2670 | if (offset >= 0) |
2671 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2672 | inode_unlock(inode); |
220f2ac9 HD |
2673 | return offset; |
2674 | } | |
2675 | ||
05f65b5c DH |
2676 | /* |
2677 | * We need a tag: a new tag would expand every radix_tree_node by 8 bytes, | |
2678 | * so reuse a tag which we firmly believe is never set or cleared on shmem. | |
2679 | */ | |
2680 | #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE | |
2681 | #define LAST_SCAN 4 /* about 150ms max */ | |
2682 | ||
2683 | static void shmem_tag_pins(struct address_space *mapping) | |
2684 | { | |
2685 | struct radix_tree_iter iter; | |
2686 | void **slot; | |
2687 | pgoff_t start; | |
2688 | struct page *page; | |
eb8e49f7 | 2689 | unsigned int tagged = 0; |
05f65b5c DH |
2690 | |
2691 | lru_add_drain(); | |
2692 | start = 0; | |
05f65b5c | 2693 | |
eb8e49f7 | 2694 | spin_lock_irq(&mapping->tree_lock); |
05f65b5c | 2695 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
8cf36be7 | 2696 | page = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); |
05f65b5c | 2697 | if (!page || radix_tree_exception(page)) { |
2cf938aa MW |
2698 | if (radix_tree_deref_retry(page)) { |
2699 | slot = radix_tree_iter_retry(&iter); | |
2700 | continue; | |
2701 | } | |
05f65b5c | 2702 | } else if (page_count(page) - page_mapcount(page) > 1) { |
05f65b5c DH |
2703 | radix_tree_tag_set(&mapping->page_tree, iter.index, |
2704 | SHMEM_TAG_PINNED); | |
05f65b5c DH |
2705 | } |
2706 | ||
eb8e49f7 MWO |
2707 | if (++tagged % 1024) |
2708 | continue; | |
2709 | ||
2710 | slot = radix_tree_iter_resume(slot, &iter); | |
2711 | spin_unlock_irq(&mapping->tree_lock); | |
2712 | cond_resched(); | |
2713 | spin_lock_irq(&mapping->tree_lock); | |
05f65b5c | 2714 | } |
eb8e49f7 | 2715 | spin_unlock_irq(&mapping->tree_lock); |
05f65b5c DH |
2716 | } |
2717 | ||
2718 | /* | |
2719 | * Setting SEAL_WRITE requires us to verify there's no pending writer. However, | |
2720 | * via get_user_pages(), drivers might have some pending I/O without any active | |
2721 | * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages | |
2722 | * and see whether it has an elevated ref-count. If so, we tag them and wait for | |
2723 | * them to be dropped. | |
2724 | * The caller must guarantee that no new user will acquire writable references | |
2725 | * to those pages to avoid races. | |
2726 | */ | |
40e041a2 DH |
2727 | static int shmem_wait_for_pins(struct address_space *mapping) |
2728 | { | |
05f65b5c DH |
2729 | struct radix_tree_iter iter; |
2730 | void **slot; | |
2731 | pgoff_t start; | |
2732 | struct page *page; | |
2733 | int error, scan; | |
2734 | ||
2735 | shmem_tag_pins(mapping); | |
2736 | ||
2737 | error = 0; | |
2738 | for (scan = 0; scan <= LAST_SCAN; scan++) { | |
2739 | if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED)) | |
2740 | break; | |
2741 | ||
2742 | if (!scan) | |
2743 | lru_add_drain_all(); | |
2744 | else if (schedule_timeout_killable((HZ << scan) / 200)) | |
2745 | scan = LAST_SCAN; | |
2746 | ||
2747 | start = 0; | |
2748 | rcu_read_lock(); | |
05f65b5c DH |
2749 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, |
2750 | start, SHMEM_TAG_PINNED) { | |
2751 | ||
2752 | page = radix_tree_deref_slot(slot); | |
2753 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2754 | if (radix_tree_deref_retry(page)) { |
2755 | slot = radix_tree_iter_retry(&iter); | |
2756 | continue; | |
2757 | } | |
05f65b5c DH |
2758 | |
2759 | page = NULL; | |
2760 | } | |
2761 | ||
2762 | if (page && | |
2763 | page_count(page) - page_mapcount(page) != 1) { | |
2764 | if (scan < LAST_SCAN) | |
2765 | goto continue_resched; | |
2766 | ||
2767 | /* | |
2768 | * On the last scan, we clean up all those tags | |
2769 | * we inserted; but make a note that we still | |
2770 | * found pages pinned. | |
2771 | */ | |
2772 | error = -EBUSY; | |
2773 | } | |
2774 | ||
2775 | spin_lock_irq(&mapping->tree_lock); | |
2776 | radix_tree_tag_clear(&mapping->page_tree, | |
2777 | iter.index, SHMEM_TAG_PINNED); | |
2778 | spin_unlock_irq(&mapping->tree_lock); | |
2779 | continue_resched: | |
2780 | if (need_resched()) { | |
148deab2 | 2781 | slot = radix_tree_iter_resume(slot, &iter); |
05f65b5c | 2782 | cond_resched_rcu(); |
05f65b5c DH |
2783 | } |
2784 | } | |
2785 | rcu_read_unlock(); | |
2786 | } | |
2787 | ||
2788 | return error; | |
40e041a2 DH |
2789 | } |
2790 | ||
2791 | #define F_ALL_SEALS (F_SEAL_SEAL | \ | |
2792 | F_SEAL_SHRINK | \ | |
2793 | F_SEAL_GROW | \ | |
2794 | F_SEAL_WRITE) | |
2795 | ||
2796 | int shmem_add_seals(struct file *file, unsigned int seals) | |
2797 | { | |
2798 | struct inode *inode = file_inode(file); | |
2799 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2800 | int error; | |
2801 | ||
2802 | /* | |
2803 | * SEALING | |
2804 | * Sealing allows multiple parties to share a shmem-file but restrict | |
2805 | * access to a specific subset of file operations. Seals can only be | |
2806 | * added, but never removed. This way, mutually untrusted parties can | |
2807 | * share common memory regions with a well-defined policy. A malicious | |
2808 | * peer can thus never perform unwanted operations on a shared object. | |
2809 | * | |
2810 | * Seals are only supported on special shmem-files and always affect | |
2811 | * the whole underlying inode. Once a seal is set, it may prevent some | |
2812 | * kinds of access to the file. Currently, the following seals are | |
2813 | * defined: | |
2814 | * SEAL_SEAL: Prevent further seals from being set on this file | |
2815 | * SEAL_SHRINK: Prevent the file from shrinking | |
2816 | * SEAL_GROW: Prevent the file from growing | |
2817 | * SEAL_WRITE: Prevent write access to the file | |
2818 | * | |
2819 | * As we don't require any trust relationship between two parties, we | |
2820 | * must prevent seals from being removed. Therefore, sealing a file | |
2821 | * only adds a given set of seals to the file, it never touches | |
2822 | * existing seals. Furthermore, the "setting seals"-operation can be | |
2823 | * sealed itself, which basically prevents any further seal from being | |
2824 | * added. | |
2825 | * | |
2826 | * Semantics of sealing are only defined on volatile files. Only | |
2827 | * anonymous shmem files support sealing. More importantly, seals are | |
2828 | * never written to disk. Therefore, there's no plan to support it on | |
2829 | * other file types. | |
2830 | */ | |
2831 | ||
2832 | if (file->f_op != &shmem_file_operations) | |
2833 | return -EINVAL; | |
2834 | if (!(file->f_mode & FMODE_WRITE)) | |
2835 | return -EPERM; | |
2836 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
2837 | return -EINVAL; | |
2838 | ||
5955102c | 2839 | inode_lock(inode); |
40e041a2 DH |
2840 | |
2841 | if (info->seals & F_SEAL_SEAL) { | |
2842 | error = -EPERM; | |
2843 | goto unlock; | |
2844 | } | |
2845 | ||
2846 | if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) { | |
2847 | error = mapping_deny_writable(file->f_mapping); | |
2848 | if (error) | |
2849 | goto unlock; | |
2850 | ||
2851 | error = shmem_wait_for_pins(file->f_mapping); | |
2852 | if (error) { | |
2853 | mapping_allow_writable(file->f_mapping); | |
2854 | goto unlock; | |
2855 | } | |
2856 | } | |
2857 | ||
2858 | info->seals |= seals; | |
2859 | error = 0; | |
2860 | ||
2861 | unlock: | |
5955102c | 2862 | inode_unlock(inode); |
40e041a2 DH |
2863 | return error; |
2864 | } | |
2865 | EXPORT_SYMBOL_GPL(shmem_add_seals); | |
2866 | ||
2867 | int shmem_get_seals(struct file *file) | |
2868 | { | |
2869 | if (file->f_op != &shmem_file_operations) | |
2870 | return -EINVAL; | |
2871 | ||
2872 | return SHMEM_I(file_inode(file))->seals; | |
2873 | } | |
2874 | EXPORT_SYMBOL_GPL(shmem_get_seals); | |
2875 | ||
2876 | long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg) | |
2877 | { | |
2878 | long error; | |
2879 | ||
2880 | switch (cmd) { | |
2881 | case F_ADD_SEALS: | |
2882 | /* disallow upper 32bit */ | |
2883 | if (arg > UINT_MAX) | |
2884 | return -EINVAL; | |
2885 | ||
2886 | error = shmem_add_seals(file, arg); | |
2887 | break; | |
2888 | case F_GET_SEALS: | |
2889 | error = shmem_get_seals(file); | |
2890 | break; | |
2891 | default: | |
2892 | error = -EINVAL; | |
2893 | break; | |
2894 | } | |
2895 | ||
2896 | return error; | |
2897 | } | |
2898 | ||
83e4fa9c HD |
2899 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2900 | loff_t len) | |
2901 | { | |
496ad9aa | 2902 | struct inode *inode = file_inode(file); |
e2d12e22 | 2903 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2904 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2905 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2906 | pgoff_t start, index, end; |
2907 | int error; | |
83e4fa9c | 2908 | |
13ace4d0 HD |
2909 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2910 | return -EOPNOTSUPP; | |
2911 | ||
5955102c | 2912 | inode_lock(inode); |
83e4fa9c HD |
2913 | |
2914 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2915 | struct address_space *mapping = file->f_mapping; | |
2916 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2917 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2918 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2919 | |
40e041a2 DH |
2920 | /* protected by i_mutex */ |
2921 | if (info->seals & F_SEAL_WRITE) { | |
2922 | error = -EPERM; | |
2923 | goto out; | |
2924 | } | |
2925 | ||
8e205f77 | 2926 | shmem_falloc.waitq = &shmem_falloc_waitq; |
e572fc28 | 2927 | shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT; |
f00cdc6d HD |
2928 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; |
2929 | spin_lock(&inode->i_lock); | |
2930 | inode->i_private = &shmem_falloc; | |
2931 | spin_unlock(&inode->i_lock); | |
2932 | ||
83e4fa9c HD |
2933 | if ((u64)unmap_end > (u64)unmap_start) |
2934 | unmap_mapping_range(mapping, unmap_start, | |
2935 | 1 + unmap_end - unmap_start, 0); | |
2936 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2937 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2938 | |
2939 | spin_lock(&inode->i_lock); | |
2940 | inode->i_private = NULL; | |
2941 | wake_up_all(&shmem_falloc_waitq); | |
2055da97 | 2942 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head)); |
8e205f77 | 2943 | spin_unlock(&inode->i_lock); |
83e4fa9c | 2944 | error = 0; |
8e205f77 | 2945 | goto out; |
e2d12e22 HD |
2946 | } |
2947 | ||
2948 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2949 | error = inode_newsize_ok(inode, offset + len); | |
2950 | if (error) | |
2951 | goto out; | |
2952 | ||
40e041a2 DH |
2953 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2954 | error = -EPERM; | |
2955 | goto out; | |
2956 | } | |
2957 | ||
09cbfeaf KS |
2958 | start = offset >> PAGE_SHIFT; |
2959 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2960 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2961 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2962 | error = -ENOSPC; | |
2963 | goto out; | |
83e4fa9c HD |
2964 | } |
2965 | ||
8e205f77 | 2966 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2967 | shmem_falloc.start = start; |
2968 | shmem_falloc.next = start; | |
2969 | shmem_falloc.nr_falloced = 0; | |
2970 | shmem_falloc.nr_unswapped = 0; | |
2971 | spin_lock(&inode->i_lock); | |
2972 | inode->i_private = &shmem_falloc; | |
2973 | spin_unlock(&inode->i_lock); | |
2974 | ||
e2d12e22 HD |
2975 | for (index = start; index < end; index++) { |
2976 | struct page *page; | |
2977 | ||
2978 | /* | |
2979 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2980 | * been interrupted because we are using up too much memory. | |
2981 | */ | |
2982 | if (signal_pending(current)) | |
2983 | error = -EINTR; | |
1aac1400 HD |
2984 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2985 | error = -ENOMEM; | |
e2d12e22 | 2986 | else |
9e18eb29 | 2987 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2988 | if (error) { |
1635f6a7 | 2989 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2990 | if (index > start) { |
2991 | shmem_undo_range(inode, | |
2992 | (loff_t)start << PAGE_SHIFT, | |
2993 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2994 | } | |
1aac1400 | 2995 | goto undone; |
e2d12e22 HD |
2996 | } |
2997 | ||
1aac1400 HD |
2998 | /* |
2999 | * Inform shmem_writepage() how far we have reached. | |
3000 | * No need for lock or barrier: we have the page lock. | |
3001 | */ | |
3002 | shmem_falloc.next++; | |
3003 | if (!PageUptodate(page)) | |
3004 | shmem_falloc.nr_falloced++; | |
3005 | ||
e2d12e22 | 3006 | /* |
1635f6a7 HD |
3007 | * If !PageUptodate, leave it that way so that freeable pages |
3008 | * can be recognized if we need to rollback on error later. | |
3009 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
3010 | * than free the pages we are allocating (and SGP_CACHE pages |
3011 | * might still be clean: we now need to mark those dirty too). | |
3012 | */ | |
3013 | set_page_dirty(page); | |
3014 | unlock_page(page); | |
09cbfeaf | 3015 | put_page(page); |
e2d12e22 HD |
3016 | cond_resched(); |
3017 | } | |
3018 | ||
3019 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
3020 | i_size_write(inode, offset + len); | |
078cd827 | 3021 | inode->i_ctime = current_time(inode); |
1aac1400 HD |
3022 | undone: |
3023 | spin_lock(&inode->i_lock); | |
3024 | inode->i_private = NULL; | |
3025 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 3026 | out: |
5955102c | 3027 | inode_unlock(inode); |
83e4fa9c HD |
3028 | return error; |
3029 | } | |
3030 | ||
726c3342 | 3031 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 3032 | { |
726c3342 | 3033 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
3034 | |
3035 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 3036 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 3037 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 3038 | if (sbinfo->max_blocks) { |
1da177e4 | 3039 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
3040 | buf->f_bavail = |
3041 | buf->f_bfree = sbinfo->max_blocks - | |
3042 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
3043 | } |
3044 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
3045 | buf->f_files = sbinfo->max_inodes; |
3046 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
3047 | } |
3048 | /* else leave those fields 0 like simple_statfs */ | |
3049 | return 0; | |
3050 | } | |
3051 | ||
3052 | /* | |
3053 | * File creation. Allocate an inode, and we're done.. | |
3054 | */ | |
3055 | static int | |
1a67aafb | 3056 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3057 | { |
0b0a0806 | 3058 | struct inode *inode; |
1da177e4 LT |
3059 | int error = -ENOSPC; |
3060 | ||
454abafe | 3061 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 3062 | if (inode) { |
feda821e CH |
3063 | error = simple_acl_create(dir, inode); |
3064 | if (error) | |
3065 | goto out_iput; | |
2a7dba39 | 3066 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 3067 | &dentry->d_name, |
6d9d88d0 | 3068 | shmem_initxattrs, NULL); |
feda821e CH |
3069 | if (error && error != -EOPNOTSUPP) |
3070 | goto out_iput; | |
37ec43cd | 3071 | |
718deb6b | 3072 | error = 0; |
1da177e4 | 3073 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3074 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3075 | d_instantiate(dentry, inode); |
3076 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
3077 | } |
3078 | return error; | |
feda821e CH |
3079 | out_iput: |
3080 | iput(inode); | |
3081 | return error; | |
1da177e4 LT |
3082 | } |
3083 | ||
60545d0d AV |
3084 | static int |
3085 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
3086 | { | |
3087 | struct inode *inode; | |
3088 | int error = -ENOSPC; | |
3089 | ||
3090 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
3091 | if (inode) { | |
3092 | error = security_inode_init_security(inode, dir, | |
3093 | NULL, | |
3094 | shmem_initxattrs, NULL); | |
feda821e CH |
3095 | if (error && error != -EOPNOTSUPP) |
3096 | goto out_iput; | |
3097 | error = simple_acl_create(dir, inode); | |
3098 | if (error) | |
3099 | goto out_iput; | |
60545d0d AV |
3100 | d_tmpfile(dentry, inode); |
3101 | } | |
3102 | return error; | |
feda821e CH |
3103 | out_iput: |
3104 | iput(inode); | |
3105 | return error; | |
60545d0d AV |
3106 | } |
3107 | ||
18bb1db3 | 3108 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
3109 | { |
3110 | int error; | |
3111 | ||
3112 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
3113 | return error; | |
d8c76e6f | 3114 | inc_nlink(dir); |
1da177e4 LT |
3115 | return 0; |
3116 | } | |
3117 | ||
4acdaf27 | 3118 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 3119 | bool excl) |
1da177e4 LT |
3120 | { |
3121 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
3122 | } | |
3123 | ||
3124 | /* | |
3125 | * Link a file.. | |
3126 | */ | |
3127 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
3128 | { | |
75c3cfa8 | 3129 | struct inode *inode = d_inode(old_dentry); |
0d3ec953 | 3130 | int ret = 0; |
1da177e4 LT |
3131 | |
3132 | /* | |
3133 | * No ordinary (disk based) filesystem counts links as inodes; | |
3134 | * but each new link needs a new dentry, pinning lowmem, and | |
3135 | * tmpfs dentries cannot be pruned until they are unlinked. | |
be0c5542 DW |
3136 | * But if an O_TMPFILE file is linked into the tmpfs, the |
3137 | * first link must skip that, to get the accounting right. | |
1da177e4 | 3138 | */ |
be0c5542 DW |
3139 | if (inode->i_nlink) { |
3140 | ret = shmem_reserve_inode(inode->i_sb); | |
3141 | if (ret) | |
3142 | goto out; | |
3143 | } | |
1da177e4 LT |
3144 | |
3145 | dir->i_size += BOGO_DIRENT_SIZE; | |
078cd827 | 3146 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
d8c76e6f | 3147 | inc_nlink(inode); |
7de9c6ee | 3148 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
3149 | dget(dentry); /* Extra pinning count for the created dentry */ |
3150 | d_instantiate(dentry, inode); | |
5b04c689 PE |
3151 | out: |
3152 | return ret; | |
1da177e4 LT |
3153 | } |
3154 | ||
3155 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
3156 | { | |
75c3cfa8 | 3157 | struct inode *inode = d_inode(dentry); |
1da177e4 | 3158 | |
5b04c689 PE |
3159 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
3160 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
3161 | |
3162 | dir->i_size -= BOGO_DIRENT_SIZE; | |
078cd827 | 3163 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
9a53c3a7 | 3164 | drop_nlink(inode); |
1da177e4 LT |
3165 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
3166 | return 0; | |
3167 | } | |
3168 | ||
3169 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
3170 | { | |
3171 | if (!simple_empty(dentry)) | |
3172 | return -ENOTEMPTY; | |
3173 | ||
75c3cfa8 | 3174 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 3175 | drop_nlink(dir); |
1da177e4 LT |
3176 | return shmem_unlink(dir, dentry); |
3177 | } | |
3178 | ||
37456771 MS |
3179 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
3180 | { | |
e36cb0b8 DH |
3181 | bool old_is_dir = d_is_dir(old_dentry); |
3182 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
3183 | |
3184 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
3185 | if (old_is_dir) { | |
3186 | drop_nlink(old_dir); | |
3187 | inc_nlink(new_dir); | |
3188 | } else { | |
3189 | drop_nlink(new_dir); | |
3190 | inc_nlink(old_dir); | |
3191 | } | |
3192 | } | |
3193 | old_dir->i_ctime = old_dir->i_mtime = | |
3194 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 | 3195 | d_inode(old_dentry)->i_ctime = |
078cd827 | 3196 | d_inode(new_dentry)->i_ctime = current_time(old_dir); |
37456771 MS |
3197 | |
3198 | return 0; | |
3199 | } | |
3200 | ||
46fdb794 MS |
3201 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
3202 | { | |
3203 | struct dentry *whiteout; | |
3204 | int error; | |
3205 | ||
3206 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
3207 | if (!whiteout) | |
3208 | return -ENOMEM; | |
3209 | ||
3210 | error = shmem_mknod(old_dir, whiteout, | |
3211 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
3212 | dput(whiteout); | |
3213 | if (error) | |
3214 | return error; | |
3215 | ||
3216 | /* | |
3217 | * Cheat and hash the whiteout while the old dentry is still in | |
3218 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
3219 | * | |
3220 | * d_lookup() will consistently find one of them at this point, | |
3221 | * not sure which one, but that isn't even important. | |
3222 | */ | |
3223 | d_rehash(whiteout); | |
3224 | return 0; | |
3225 | } | |
3226 | ||
1da177e4 LT |
3227 | /* |
3228 | * The VFS layer already does all the dentry stuff for rename, | |
3229 | * we just have to decrement the usage count for the target if | |
3230 | * it exists so that the VFS layer correctly free's it when it | |
3231 | * gets overwritten. | |
3232 | */ | |
3b69ff51 | 3233 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 3234 | { |
75c3cfa8 | 3235 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
3236 | int they_are_dirs = S_ISDIR(inode->i_mode); |
3237 | ||
46fdb794 | 3238 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
3239 | return -EINVAL; |
3240 | ||
37456771 MS |
3241 | if (flags & RENAME_EXCHANGE) |
3242 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
3243 | ||
1da177e4 LT |
3244 | if (!simple_empty(new_dentry)) |
3245 | return -ENOTEMPTY; | |
3246 | ||
46fdb794 MS |
3247 | if (flags & RENAME_WHITEOUT) { |
3248 | int error; | |
3249 | ||
3250 | error = shmem_whiteout(old_dir, old_dentry); | |
3251 | if (error) | |
3252 | return error; | |
3253 | } | |
3254 | ||
75c3cfa8 | 3255 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 3256 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 3257 | if (they_are_dirs) { |
75c3cfa8 | 3258 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 3259 | drop_nlink(old_dir); |
b928095b | 3260 | } |
1da177e4 | 3261 | } else if (they_are_dirs) { |
9a53c3a7 | 3262 | drop_nlink(old_dir); |
d8c76e6f | 3263 | inc_nlink(new_dir); |
1da177e4 LT |
3264 | } |
3265 | ||
3266 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
3267 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
3268 | old_dir->i_ctime = old_dir->i_mtime = | |
3269 | new_dir->i_ctime = new_dir->i_mtime = | |
078cd827 | 3270 | inode->i_ctime = current_time(old_dir); |
1da177e4 LT |
3271 | return 0; |
3272 | } | |
3273 | ||
3274 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
3275 | { | |
3276 | int error; | |
3277 | int len; | |
3278 | struct inode *inode; | |
9276aad6 | 3279 | struct page *page; |
1da177e4 LT |
3280 | |
3281 | len = strlen(symname) + 1; | |
09cbfeaf | 3282 | if (len > PAGE_SIZE) |
1da177e4 LT |
3283 | return -ENAMETOOLONG; |
3284 | ||
454abafe | 3285 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
3286 | if (!inode) |
3287 | return -ENOSPC; | |
3288 | ||
9d8f13ba | 3289 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3290 | shmem_initxattrs, NULL); |
570bc1c2 SS |
3291 | if (error) { |
3292 | if (error != -EOPNOTSUPP) { | |
3293 | iput(inode); | |
3294 | return error; | |
3295 | } | |
3296 | error = 0; | |
3297 | } | |
3298 | ||
1da177e4 | 3299 | inode->i_size = len-1; |
69f07ec9 | 3300 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3301 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3302 | if (!inode->i_link) { | |
69f07ec9 HD |
3303 | iput(inode); |
3304 | return -ENOMEM; | |
3305 | } | |
3306 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3307 | } else { |
e8ecde25 | 3308 | inode_nohighmem(inode); |
9e18eb29 | 3309 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
3310 | if (error) { |
3311 | iput(inode); | |
3312 | return error; | |
3313 | } | |
14fcc23f | 3314 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 3315 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 3316 | memcpy(page_address(page), symname, len); |
ec9516fb | 3317 | SetPageUptodate(page); |
1da177e4 | 3318 | set_page_dirty(page); |
6746aff7 | 3319 | unlock_page(page); |
09cbfeaf | 3320 | put_page(page); |
1da177e4 | 3321 | } |
1da177e4 | 3322 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3323 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3324 | d_instantiate(dentry, inode); |
3325 | dget(dentry); | |
3326 | return 0; | |
3327 | } | |
3328 | ||
fceef393 | 3329 | static void shmem_put_link(void *arg) |
1da177e4 | 3330 | { |
fceef393 AV |
3331 | mark_page_accessed(arg); |
3332 | put_page(arg); | |
1da177e4 LT |
3333 | } |
3334 | ||
6b255391 | 3335 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
3336 | struct inode *inode, |
3337 | struct delayed_call *done) | |
1da177e4 | 3338 | { |
1da177e4 | 3339 | struct page *page = NULL; |
6b255391 | 3340 | int error; |
6a6c9904 AV |
3341 | if (!dentry) { |
3342 | page = find_get_page(inode->i_mapping, 0); | |
3343 | if (!page) | |
3344 | return ERR_PTR(-ECHILD); | |
3345 | if (!PageUptodate(page)) { | |
3346 | put_page(page); | |
3347 | return ERR_PTR(-ECHILD); | |
3348 | } | |
3349 | } else { | |
9e18eb29 | 3350 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3351 | if (error) |
3352 | return ERR_PTR(error); | |
3353 | unlock_page(page); | |
3354 | } | |
fceef393 | 3355 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3356 | return page_address(page); |
1da177e4 LT |
3357 | } |
3358 | ||
b09e0fa4 | 3359 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3360 | /* |
b09e0fa4 EP |
3361 | * Superblocks without xattr inode operations may get some security.* xattr |
3362 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3363 | * like ACLs, we also need to implement the security.* handlers at |
3364 | * filesystem level, though. | |
3365 | */ | |
3366 | ||
6d9d88d0 JS |
3367 | /* |
3368 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3369 | */ | |
3370 | static int shmem_initxattrs(struct inode *inode, | |
3371 | const struct xattr *xattr_array, | |
3372 | void *fs_info) | |
3373 | { | |
3374 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3375 | const struct xattr *xattr; | |
38f38657 | 3376 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3377 | size_t len; |
3378 | ||
3379 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3380 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3381 | if (!new_xattr) |
3382 | return -ENOMEM; | |
3383 | ||
3384 | len = strlen(xattr->name) + 1; | |
3385 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3386 | GFP_KERNEL); | |
3387 | if (!new_xattr->name) { | |
3388 | kfree(new_xattr); | |
3389 | return -ENOMEM; | |
3390 | } | |
3391 | ||
3392 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3393 | XATTR_SECURITY_PREFIX_LEN); | |
3394 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3395 | xattr->name, len); | |
3396 | ||
38f38657 | 3397 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3398 | } |
3399 | ||
3400 | return 0; | |
3401 | } | |
3402 | ||
aa7c5241 | 3403 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3404 | struct dentry *unused, struct inode *inode, |
3405 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3406 | { |
b296821a | 3407 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3408 | |
aa7c5241 | 3409 | name = xattr_full_name(handler, name); |
38f38657 | 3410 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3411 | } |
3412 | ||
aa7c5241 | 3413 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
59301226 AV |
3414 | struct dentry *unused, struct inode *inode, |
3415 | const char *name, const void *value, | |
3416 | size_t size, int flags) | |
b09e0fa4 | 3417 | { |
59301226 | 3418 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3419 | |
aa7c5241 | 3420 | name = xattr_full_name(handler, name); |
38f38657 | 3421 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
3422 | } |
3423 | ||
aa7c5241 AG |
3424 | static const struct xattr_handler shmem_security_xattr_handler = { |
3425 | .prefix = XATTR_SECURITY_PREFIX, | |
3426 | .get = shmem_xattr_handler_get, | |
3427 | .set = shmem_xattr_handler_set, | |
3428 | }; | |
b09e0fa4 | 3429 | |
aa7c5241 AG |
3430 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3431 | .prefix = XATTR_TRUSTED_PREFIX, | |
3432 | .get = shmem_xattr_handler_get, | |
3433 | .set = shmem_xattr_handler_set, | |
3434 | }; | |
b09e0fa4 | 3435 | |
aa7c5241 AG |
3436 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3437 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3438 | &posix_acl_access_xattr_handler, | |
3439 | &posix_acl_default_xattr_handler, | |
3440 | #endif | |
3441 | &shmem_security_xattr_handler, | |
3442 | &shmem_trusted_xattr_handler, | |
3443 | NULL | |
3444 | }; | |
b09e0fa4 EP |
3445 | |
3446 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3447 | { | |
75c3cfa8 | 3448 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3449 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3450 | } |
3451 | #endif /* CONFIG_TMPFS_XATTR */ | |
3452 | ||
69f07ec9 | 3453 | static const struct inode_operations shmem_short_symlink_operations = { |
6b255391 | 3454 | .get_link = simple_get_link, |
b09e0fa4 | 3455 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3456 | .listxattr = shmem_listxattr, |
b09e0fa4 EP |
3457 | #endif |
3458 | }; | |
3459 | ||
3460 | static const struct inode_operations shmem_symlink_inode_operations = { | |
6b255391 | 3461 | .get_link = shmem_get_link, |
b09e0fa4 | 3462 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3463 | .listxattr = shmem_listxattr, |
39f0247d | 3464 | #endif |
b09e0fa4 | 3465 | }; |
39f0247d | 3466 | |
91828a40 DG |
3467 | static struct dentry *shmem_get_parent(struct dentry *child) |
3468 | { | |
3469 | return ERR_PTR(-ESTALE); | |
3470 | } | |
3471 | ||
3472 | static int shmem_match(struct inode *ino, void *vfh) | |
3473 | { | |
3474 | __u32 *fh = vfh; | |
c088e31d | 3475 | __u64 inum = fh[1]; |
91828a40 DG |
3476 | return ino->i_ino == inum && fh[0] == ino->i_generation; |
3477 | } | |
3478 | ||
480b116c CH |
3479 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3480 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3481 | { |
91828a40 | 3482 | struct inode *inode; |
480b116c | 3483 | struct dentry *dentry = NULL; |
35c2a7f4 | 3484 | u64 inum; |
480b116c | 3485 | |
c088e31d | 3486 | if (fh_len < 2) |
480b116c | 3487 | return NULL; |
91828a40 | 3488 | |
c088e31d SF |
3489 | inum = fid->raw[1]; |
3490 | inode = ilookup5(sb, inum, shmem_match, fid->raw); | |
91828a40 | 3491 | if (inode) { |
480b116c | 3492 | dentry = d_find_alias(inode); |
91828a40 DG |
3493 | iput(inode); |
3494 | } | |
3495 | ||
480b116c | 3496 | return dentry; |
91828a40 DG |
3497 | } |
3498 | ||
b0b0382b AV |
3499 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3500 | struct inode *parent) | |
91828a40 | 3501 | { |
c088e31d SF |
3502 | if (*len < 2) { |
3503 | *len = 2; | |
94e07a75 | 3504 | return FILEID_INVALID; |
5fe0c237 | 3505 | } |
91828a40 | 3506 | |
91828a40 DG |
3507 | fh[0] = inode->i_generation; |
3508 | fh[1] = inode->i_ino; | |
91828a40 | 3509 | |
c088e31d | 3510 | *len = 2; |
91828a40 DG |
3511 | return 1; |
3512 | } | |
3513 | ||
39655164 | 3514 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3515 | .get_parent = shmem_get_parent, |
91828a40 | 3516 | .encode_fh = shmem_encode_fh, |
480b116c | 3517 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3518 | }; |
3519 | ||
680d794b AM |
3520 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
3521 | bool remount) | |
1da177e4 LT |
3522 | { |
3523 | char *this_char, *value, *rest; | |
49cd0a5c | 3524 | struct mempolicy *mpol = NULL; |
8751e039 EB |
3525 | uid_t uid; |
3526 | gid_t gid; | |
1da177e4 | 3527 | |
b00dc3ad HD |
3528 | while (options != NULL) { |
3529 | this_char = options; | |
3530 | for (;;) { | |
3531 | /* | |
3532 | * NUL-terminate this option: unfortunately, | |
3533 | * mount options form a comma-separated list, | |
3534 | * but mpol's nodelist may also contain commas. | |
3535 | */ | |
3536 | options = strchr(options, ','); | |
3537 | if (options == NULL) | |
3538 | break; | |
3539 | options++; | |
3540 | if (!isdigit(*options)) { | |
3541 | options[-1] = '\0'; | |
3542 | break; | |
3543 | } | |
3544 | } | |
1da177e4 LT |
3545 | if (!*this_char) |
3546 | continue; | |
3547 | if ((value = strchr(this_char,'=')) != NULL) { | |
3548 | *value++ = 0; | |
3549 | } else { | |
1170532b JP |
3550 | pr_err("tmpfs: No value for mount option '%s'\n", |
3551 | this_char); | |
49cd0a5c | 3552 | goto error; |
1da177e4 LT |
3553 | } |
3554 | ||
3555 | if (!strcmp(this_char,"size")) { | |
3556 | unsigned long long size; | |
3557 | size = memparse(value,&rest); | |
3558 | if (*rest == '%') { | |
3559 | size <<= PAGE_SHIFT; | |
3560 | size *= totalram_pages; | |
3561 | do_div(size, 100); | |
3562 | rest++; | |
3563 | } | |
3564 | if (*rest) | |
3565 | goto bad_val; | |
680d794b | 3566 | sbinfo->max_blocks = |
09cbfeaf | 3567 | DIV_ROUND_UP(size, PAGE_SIZE); |
1da177e4 | 3568 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 3569 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
3570 | if (*rest) |
3571 | goto bad_val; | |
3572 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 3573 | sbinfo->max_inodes = memparse(value, &rest); |
c088e31d | 3574 | if (*rest || sbinfo->max_inodes < 2) |
1da177e4 LT |
3575 | goto bad_val; |
3576 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 3577 | if (remount) |
1da177e4 | 3578 | continue; |
680d794b | 3579 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
3580 | if (*rest) |
3581 | goto bad_val; | |
3582 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 3583 | if (remount) |
1da177e4 | 3584 | continue; |
8751e039 | 3585 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3586 | if (*rest) |
3587 | goto bad_val; | |
8751e039 EB |
3588 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
3589 | if (!uid_valid(sbinfo->uid)) | |
3590 | goto bad_val; | |
1da177e4 | 3591 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 3592 | if (remount) |
1da177e4 | 3593 | continue; |
8751e039 | 3594 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3595 | if (*rest) |
3596 | goto bad_val; | |
8751e039 EB |
3597 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
3598 | if (!gid_valid(sbinfo->gid)) | |
3599 | goto bad_val; | |
e496cf3d | 3600 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3601 | } else if (!strcmp(this_char, "huge")) { |
3602 | int huge; | |
3603 | huge = shmem_parse_huge(value); | |
3604 | if (huge < 0) | |
3605 | goto bad_val; | |
3606 | if (!has_transparent_hugepage() && | |
3607 | huge != SHMEM_HUGE_NEVER) | |
3608 | goto bad_val; | |
3609 | sbinfo->huge = huge; | |
3610 | #endif | |
3611 | #ifdef CONFIG_NUMA | |
7339ff83 | 3612 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
3613 | mpol_put(mpol); |
3614 | mpol = NULL; | |
3615 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 3616 | goto bad_val; |
5a6e75f8 | 3617 | #endif |
1da177e4 | 3618 | } else { |
1170532b | 3619 | pr_err("tmpfs: Bad mount option %s\n", this_char); |
49cd0a5c | 3620 | goto error; |
1da177e4 LT |
3621 | } |
3622 | } | |
49cd0a5c | 3623 | sbinfo->mpol = mpol; |
1da177e4 LT |
3624 | return 0; |
3625 | ||
3626 | bad_val: | |
1170532b | 3627 | pr_err("tmpfs: Bad value '%s' for mount option '%s'\n", |
1da177e4 | 3628 | value, this_char); |
49cd0a5c GT |
3629 | error: |
3630 | mpol_put(mpol); | |
1da177e4 LT |
3631 | return 1; |
3632 | ||
3633 | } | |
3634 | ||
3635 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
3636 | { | |
3637 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 3638 | struct shmem_sb_info config = *sbinfo; |
c088e31d | 3639 | int inodes; |
0edd73b3 HD |
3640 | int error = -EINVAL; |
3641 | ||
5f00110f | 3642 | config.mpol = NULL; |
680d794b | 3643 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 3644 | return error; |
1da177e4 | 3645 | |
0edd73b3 | 3646 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3647 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 3648 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 3649 | goto out; |
680d794b | 3650 | if (config.max_inodes < inodes) |
0edd73b3 HD |
3651 | goto out; |
3652 | /* | |
54af6042 | 3653 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
3654 | * but we must separately disallow unlimited->limited, because |
3655 | * in that case we have no record of how much is already in use. | |
3656 | */ | |
680d794b | 3657 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 3658 | goto out; |
680d794b | 3659 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
3660 | goto out; |
3661 | ||
3662 | error = 0; | |
5a6e75f8 | 3663 | sbinfo->huge = config.huge; |
680d794b | 3664 | sbinfo->max_blocks = config.max_blocks; |
680d794b AM |
3665 | sbinfo->max_inodes = config.max_inodes; |
3666 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 3667 | |
5f00110f GT |
3668 | /* |
3669 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3670 | */ | |
3671 | if (config.mpol) { | |
3672 | mpol_put(sbinfo->mpol); | |
3673 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
3674 | } | |
0edd73b3 HD |
3675 | out: |
3676 | spin_unlock(&sbinfo->stat_lock); | |
3677 | return error; | |
1da177e4 | 3678 | } |
680d794b | 3679 | |
34c80b1d | 3680 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3681 | { |
34c80b1d | 3682 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b AM |
3683 | |
3684 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3685 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3686 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 3687 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
c088e31d | 3688 | seq_printf(seq, ",nr_inodes=%d", sbinfo->max_inodes); |
680d794b | 3689 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) |
09208d15 | 3690 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3691 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3692 | seq_printf(seq, ",uid=%u", | |
3693 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3694 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3695 | seq_printf(seq, ",gid=%u", | |
3696 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
e496cf3d | 3697 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3698 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
3699 | if (sbinfo->huge) | |
3700 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3701 | #endif | |
71fe804b | 3702 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b AM |
3703 | return 0; |
3704 | } | |
9183df25 DH |
3705 | |
3706 | #define MFD_NAME_PREFIX "memfd:" | |
3707 | #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1) | |
3708 | #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN) | |
3709 | ||
749df87b | 3710 | #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB) |
9183df25 DH |
3711 | |
3712 | SYSCALL_DEFINE2(memfd_create, | |
3713 | const char __user *, uname, | |
3714 | unsigned int, flags) | |
3715 | { | |
3716 | struct shmem_inode_info *info; | |
3717 | struct file *file; | |
3718 | int fd, error; | |
3719 | char *name; | |
3720 | long len; | |
3721 | ||
749df87b MK |
3722 | if (!(flags & MFD_HUGETLB)) { |
3723 | if (flags & ~(unsigned int)MFD_ALL_FLAGS) | |
3724 | return -EINVAL; | |
3725 | } else { | |
3726 | /* Sealing not supported in hugetlbfs (MFD_HUGETLB) */ | |
3727 | if (flags & MFD_ALLOW_SEALING) | |
3728 | return -EINVAL; | |
3729 | /* Allow huge page size encoding in flags. */ | |
3730 | if (flags & ~(unsigned int)(MFD_ALL_FLAGS | | |
3731 | (MFD_HUGE_MASK << MFD_HUGE_SHIFT))) | |
3732 | return -EINVAL; | |
3733 | } | |
9183df25 DH |
3734 | |
3735 | /* length includes terminating zero */ | |
3736 | len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1); | |
3737 | if (len <= 0) | |
3738 | return -EFAULT; | |
3739 | if (len > MFD_NAME_MAX_LEN + 1) | |
3740 | return -EINVAL; | |
3741 | ||
0ee931c4 | 3742 | name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_KERNEL); |
9183df25 DH |
3743 | if (!name) |
3744 | return -ENOMEM; | |
3745 | ||
3746 | strcpy(name, MFD_NAME_PREFIX); | |
3747 | if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) { | |
3748 | error = -EFAULT; | |
3749 | goto err_name; | |
3750 | } | |
3751 | ||
3752 | /* terminating-zero may have changed after strnlen_user() returned */ | |
3753 | if (name[len + MFD_NAME_PREFIX_LEN - 1]) { | |
3754 | error = -EFAULT; | |
3755 | goto err_name; | |
3756 | } | |
3757 | ||
3758 | fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0); | |
3759 | if (fd < 0) { | |
3760 | error = fd; | |
3761 | goto err_name; | |
3762 | } | |
3763 | ||
749df87b MK |
3764 | if (flags & MFD_HUGETLB) { |
3765 | struct user_struct *user = NULL; | |
3766 | ||
3767 | file = hugetlb_file_setup(name, 0, VM_NORESERVE, &user, | |
3768 | HUGETLB_ANONHUGE_INODE, | |
3769 | (flags >> MFD_HUGE_SHIFT) & | |
3770 | MFD_HUGE_MASK); | |
3771 | } else | |
3772 | file = shmem_file_setup(name, 0, VM_NORESERVE); | |
9183df25 DH |
3773 | if (IS_ERR(file)) { |
3774 | error = PTR_ERR(file); | |
3775 | goto err_fd; | |
3776 | } | |
9183df25 DH |
3777 | file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; |
3778 | file->f_flags |= O_RDWR | O_LARGEFILE; | |
749df87b MK |
3779 | |
3780 | if (flags & MFD_ALLOW_SEALING) { | |
3781 | /* | |
3782 | * flags check at beginning of function ensures | |
3783 | * this is not a hugetlbfs (MFD_HUGETLB) file. | |
3784 | */ | |
3785 | info = SHMEM_I(file_inode(file)); | |
9183df25 | 3786 | info->seals &= ~F_SEAL_SEAL; |
749df87b | 3787 | } |
9183df25 DH |
3788 | |
3789 | fd_install(fd, file); | |
3790 | kfree(name); | |
3791 | return fd; | |
3792 | ||
3793 | err_fd: | |
3794 | put_unused_fd(fd); | |
3795 | err_name: | |
3796 | kfree(name); | |
3797 | return error; | |
3798 | } | |
3799 | ||
680d794b | 3800 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3801 | |
3802 | static void shmem_put_super(struct super_block *sb) | |
3803 | { | |
602586a8 HD |
3804 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3805 | ||
c088e31d SF |
3806 | if (!sbinfo->idr_nouse) |
3807 | idr_destroy(&sbinfo->idr); | |
602586a8 | 3808 | percpu_counter_destroy(&sbinfo->used_blocks); |
49cd0a5c | 3809 | mpol_put(sbinfo->mpol); |
602586a8 | 3810 | kfree(sbinfo); |
1da177e4 LT |
3811 | sb->s_fs_info = NULL; |
3812 | } | |
3813 | ||
2b2af54a | 3814 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3815 | { |
3816 | struct inode *inode; | |
0edd73b3 | 3817 | struct shmem_sb_info *sbinfo; |
680d794b AM |
3818 | int err = -ENOMEM; |
3819 | ||
3820 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3821 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b AM |
3822 | L1_CACHE_BYTES), GFP_KERNEL); |
3823 | if (!sbinfo) | |
3824 | return -ENOMEM; | |
3825 | ||
c088e31d SF |
3826 | mutex_init(&sbinfo->idr_lock); |
3827 | idr_init(&sbinfo->idr); | |
680d794b | 3828 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
3829 | sbinfo->uid = current_fsuid(); |
3830 | sbinfo->gid = current_fsgid(); | |
680d794b | 3831 | sb->s_fs_info = sbinfo; |
1da177e4 | 3832 | |
0edd73b3 | 3833 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3834 | /* |
3835 | * Per default we only allow half of the physical ram per | |
3836 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3837 | * but the internal instance is left unlimited. | |
3838 | */ | |
1751e8a6 | 3839 | if (!(sb->s_flags & SB_KERNMOUNT)) { |
680d794b AM |
3840 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3841 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3842 | if (shmem_parse_options(data, sbinfo, false)) { | |
3843 | err = -EINVAL; | |
3844 | goto failed; | |
3845 | } | |
ca4e0519 | 3846 | } else { |
1751e8a6 | 3847 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 3848 | } |
91828a40 | 3849 | sb->s_export_op = &shmem_export_ops; |
1751e8a6 | 3850 | sb->s_flags |= SB_NOSEC; |
1da177e4 | 3851 | #else |
1751e8a6 | 3852 | sb->s_flags |= SB_NOUSER; |
1da177e4 LT |
3853 | #endif |
3854 | ||
0edd73b3 | 3855 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3856 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3857 | goto failed; |
680d794b | 3858 | sbinfo->free_inodes = sbinfo->max_inodes; |
779750d2 KS |
3859 | spin_lock_init(&sbinfo->shrinklist_lock); |
3860 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 3861 | |
285b2c4f | 3862 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3863 | sb->s_blocksize = PAGE_SIZE; |
3864 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3865 | sb->s_magic = TMPFS_MAGIC; |
3866 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3867 | sb->s_time_gran = 1; |
b09e0fa4 | 3868 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3869 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3870 | #endif |
3871 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
1751e8a6 | 3872 | sb->s_flags |= SB_POSIXACL; |
39f0247d | 3873 | #endif |
2b4db796 | 3874 | uuid_gen(&sb->s_uuid); |
0edd73b3 | 3875 | |
454abafe | 3876 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3877 | if (!inode) |
3878 | goto failed; | |
680d794b AM |
3879 | inode->i_uid = sbinfo->uid; |
3880 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3881 | sb->s_root = d_make_root(inode); |
3882 | if (!sb->s_root) | |
48fde701 | 3883 | goto failed; |
1da177e4 LT |
3884 | return 0; |
3885 | ||
1da177e4 LT |
3886 | failed: |
3887 | shmem_put_super(sb); | |
3888 | return err; | |
3889 | } | |
3890 | ||
fcc234f8 | 3891 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3892 | |
3893 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3894 | { | |
41ffe5d5 HD |
3895 | struct shmem_inode_info *info; |
3896 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3897 | if (!info) | |
1da177e4 | 3898 | return NULL; |
41ffe5d5 | 3899 | return &info->vfs_inode; |
1da177e4 LT |
3900 | } |
3901 | ||
41ffe5d5 | 3902 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3903 | { |
3904 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84e710da AV |
3905 | if (S_ISLNK(inode->i_mode)) |
3906 | kfree(inode->i_link); | |
fa0d7e3d NP |
3907 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3908 | } | |
3909 | ||
1da177e4 LT |
3910 | static void shmem_destroy_inode(struct inode *inode) |
3911 | { | |
09208d15 | 3912 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3913 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3914 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3915 | } |
3916 | ||
41ffe5d5 | 3917 | static void shmem_init_inode(void *foo) |
1da177e4 | 3918 | { |
41ffe5d5 HD |
3919 | struct shmem_inode_info *info = foo; |
3920 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3921 | } |
3922 | ||
9a8ec03e | 3923 | static void shmem_init_inodecache(void) |
1da177e4 LT |
3924 | { |
3925 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3926 | sizeof(struct shmem_inode_info), | |
5d097056 | 3927 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3928 | } |
3929 | ||
41ffe5d5 | 3930 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3931 | { |
1a1d92c1 | 3932 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3933 | } |
3934 | ||
c088e31d SF |
3935 | static __init void shmem_no_idr(struct super_block *sb) |
3936 | { | |
3937 | struct shmem_sb_info *sbinfo; | |
3938 | ||
3939 | sbinfo = SHMEM_SB(sb); | |
3940 | sbinfo->idr_nouse = true; | |
3941 | idr_destroy(&sbinfo->idr); | |
3942 | } | |
3943 | ||
f5e54d6e | 3944 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3945 | .writepage = shmem_writepage, |
76719325 | 3946 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3947 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3948 | .write_begin = shmem_write_begin, |
3949 | .write_end = shmem_write_end, | |
1da177e4 | 3950 | #endif |
1c93923c | 3951 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3952 | .migratepage = migrate_page, |
1c93923c | 3953 | #endif |
aa261f54 | 3954 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3955 | }; |
3956 | ||
15ad7cdc | 3957 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3958 | .mmap = shmem_mmap, |
c01d5b30 | 3959 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3960 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3961 | .llseek = shmem_file_llseek, |
2ba5bbed | 3962 | .read_iter = shmem_file_read_iter, |
8174202b | 3963 | .write_iter = generic_file_write_iter, |
1b061d92 | 3964 | .fsync = noop_fsync, |
82c156f8 | 3965 | .splice_read = generic_file_splice_read, |
f6cb85d0 | 3966 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3967 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3968 | #endif |
3969 | }; | |
3970 | ||
92e1d5be | 3971 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3972 | .getattr = shmem_getattr, |
94c1e62d | 3973 | .setattr = shmem_setattr, |
b09e0fa4 | 3974 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3975 | .listxattr = shmem_listxattr, |
feda821e | 3976 | .set_acl = simple_set_acl, |
b09e0fa4 | 3977 | #endif |
1da177e4 LT |
3978 | }; |
3979 | ||
92e1d5be | 3980 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3981 | #ifdef CONFIG_TMPFS |
3982 | .create = shmem_create, | |
3983 | .lookup = simple_lookup, | |
3984 | .link = shmem_link, | |
3985 | .unlink = shmem_unlink, | |
3986 | .symlink = shmem_symlink, | |
3987 | .mkdir = shmem_mkdir, | |
3988 | .rmdir = shmem_rmdir, | |
3989 | .mknod = shmem_mknod, | |
2773bf00 | 3990 | .rename = shmem_rename2, |
60545d0d | 3991 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3992 | #endif |
b09e0fa4 | 3993 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3994 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3995 | #endif |
39f0247d | 3996 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3997 | .setattr = shmem_setattr, |
feda821e | 3998 | .set_acl = simple_set_acl, |
39f0247d AG |
3999 | #endif |
4000 | }; | |
4001 | ||
92e1d5be | 4002 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 4003 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 4004 | .listxattr = shmem_listxattr, |
b09e0fa4 | 4005 | #endif |
39f0247d | 4006 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 4007 | .setattr = shmem_setattr, |
feda821e | 4008 | .set_acl = simple_set_acl, |
39f0247d | 4009 | #endif |
1da177e4 LT |
4010 | }; |
4011 | ||
759b9775 | 4012 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
4013 | .alloc_inode = shmem_alloc_inode, |
4014 | .destroy_inode = shmem_destroy_inode, | |
4015 | #ifdef CONFIG_TMPFS | |
4016 | .statfs = shmem_statfs, | |
4017 | .remount_fs = shmem_remount_fs, | |
680d794b | 4018 | .show_options = shmem_show_options, |
1da177e4 | 4019 | #endif |
1f895f75 | 4020 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
4021 | .drop_inode = generic_delete_inode, |
4022 | .put_super = shmem_put_super, | |
779750d2 KS |
4023 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
4024 | .nr_cached_objects = shmem_unused_huge_count, | |
4025 | .free_cached_objects = shmem_unused_huge_scan, | |
4026 | #endif | |
1da177e4 LT |
4027 | }; |
4028 | ||
f0f37e2f | 4029 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 4030 | .fault = shmem_fault, |
d7c17551 | 4031 | .map_pages = filemap_map_pages, |
1da177e4 LT |
4032 | #ifdef CONFIG_NUMA |
4033 | .set_policy = shmem_set_policy, | |
4034 | .get_policy = shmem_get_policy, | |
4035 | #endif | |
4036 | }; | |
4037 | ||
3c26ff6e AV |
4038 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
4039 | int flags, const char *dev_name, void *data) | |
1da177e4 | 4040 | { |
3c26ff6e | 4041 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
4042 | } |
4043 | ||
41ffe5d5 | 4044 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
4045 | .owner = THIS_MODULE, |
4046 | .name = "tmpfs", | |
3c26ff6e | 4047 | .mount = shmem_mount, |
1da177e4 | 4048 | .kill_sb = kill_litter_super, |
2b8576cb | 4049 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 4050 | }; |
1da177e4 | 4051 | |
41ffe5d5 | 4052 | int __init shmem_init(void) |
1da177e4 LT |
4053 | { |
4054 | int error; | |
4055 | ||
16203a7a RL |
4056 | /* If rootfs called this, don't re-init */ |
4057 | if (shmem_inode_cachep) | |
4058 | return 0; | |
4059 | ||
9a8ec03e | 4060 | shmem_init_inodecache(); |
1da177e4 | 4061 | |
41ffe5d5 | 4062 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 4063 | if (error) { |
1170532b | 4064 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
4065 | goto out2; |
4066 | } | |
95dc112a | 4067 | |
ca4e0519 | 4068 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
4069 | if (IS_ERR(shm_mnt)) { |
4070 | error = PTR_ERR(shm_mnt); | |
1170532b | 4071 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
4072 | goto out1; |
4073 | } | |
c088e31d | 4074 | shmem_no_idr(shm_mnt->mnt_sb); |
5a6e75f8 | 4075 | |
e496cf3d | 4076 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
435c0b87 | 4077 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4078 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4079 | else | |
4080 | shmem_huge = 0; /* just in case it was patched */ | |
4081 | #endif | |
1da177e4 LT |
4082 | return 0; |
4083 | ||
4084 | out1: | |
41ffe5d5 | 4085 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 4086 | out2: |
41ffe5d5 | 4087 | shmem_destroy_inodecache(); |
1da177e4 LT |
4088 | shm_mnt = ERR_PTR(error); |
4089 | return error; | |
4090 | } | |
853ac43a | 4091 | |
e496cf3d | 4092 | #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS) |
5a6e75f8 KS |
4093 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
4094 | struct kobj_attribute *attr, char *buf) | |
4095 | { | |
4096 | int values[] = { | |
4097 | SHMEM_HUGE_ALWAYS, | |
4098 | SHMEM_HUGE_WITHIN_SIZE, | |
4099 | SHMEM_HUGE_ADVISE, | |
4100 | SHMEM_HUGE_NEVER, | |
4101 | SHMEM_HUGE_DENY, | |
4102 | SHMEM_HUGE_FORCE, | |
4103 | }; | |
4104 | int i, count; | |
4105 | ||
4106 | for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) { | |
4107 | const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s "; | |
4108 | ||
4109 | count += sprintf(buf + count, fmt, | |
4110 | shmem_format_huge(values[i])); | |
4111 | } | |
4112 | buf[count - 1] = '\n'; | |
4113 | return count; | |
4114 | } | |
4115 | ||
4116 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
4117 | struct kobj_attribute *attr, const char *buf, size_t count) | |
4118 | { | |
4119 | char tmp[16]; | |
4120 | int huge; | |
4121 | ||
4122 | if (count + 1 > sizeof(tmp)) | |
4123 | return -EINVAL; | |
4124 | memcpy(tmp, buf, count); | |
4125 | tmp[count] = '\0'; | |
4126 | if (count && tmp[count - 1] == '\n') | |
4127 | tmp[count - 1] = '\0'; | |
4128 | ||
4129 | huge = shmem_parse_huge(tmp); | |
4130 | if (huge == -EINVAL) | |
4131 | return -EINVAL; | |
4132 | if (!has_transparent_hugepage() && | |
4133 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
4134 | return -EINVAL; | |
4135 | ||
4136 | shmem_huge = huge; | |
435c0b87 | 4137 | if (shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4138 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4139 | return count; | |
4140 | } | |
4141 | ||
4142 | struct kobj_attribute shmem_enabled_attr = | |
4143 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
3b33719c | 4144 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */ |
f3f0e1d2 | 4145 | |
3b33719c | 4146 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
f3f0e1d2 KS |
4147 | bool shmem_huge_enabled(struct vm_area_struct *vma) |
4148 | { | |
4149 | struct inode *inode = file_inode(vma->vm_file); | |
4150 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4151 | loff_t i_size; | |
4152 | pgoff_t off; | |
4153 | ||
4154 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
4155 | return true; | |
4156 | if (shmem_huge == SHMEM_HUGE_DENY) | |
4157 | return false; | |
4158 | switch (sbinfo->huge) { | |
4159 | case SHMEM_HUGE_NEVER: | |
4160 | return false; | |
4161 | case SHMEM_HUGE_ALWAYS: | |
4162 | return true; | |
4163 | case SHMEM_HUGE_WITHIN_SIZE: | |
4164 | off = round_up(vma->vm_pgoff, HPAGE_PMD_NR); | |
4165 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
4166 | if (i_size >= HPAGE_PMD_SIZE && | |
4167 | i_size >> PAGE_SHIFT >= off) | |
4168 | return true; | |
c8402871 | 4169 | /* fall through */ |
f3f0e1d2 KS |
4170 | case SHMEM_HUGE_ADVISE: |
4171 | /* TODO: implement fadvise() hints */ | |
4172 | return (vma->vm_flags & VM_HUGEPAGE); | |
4173 | default: | |
4174 | VM_BUG_ON(1); | |
4175 | return false; | |
4176 | } | |
4177 | } | |
3b33719c | 4178 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 4179 | |
853ac43a MM |
4180 | #else /* !CONFIG_SHMEM */ |
4181 | ||
4182 | /* | |
4183 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
4184 | * | |
4185 | * This is intended for small system where the benefits of the full | |
4186 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4187 | * their complexity. On systems without swap this code should be | |
4188 | * effectively equivalent, but much lighter weight. | |
4189 | */ | |
4190 | ||
41ffe5d5 | 4191 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4192 | .name = "tmpfs", |
3c26ff6e | 4193 | .mount = ramfs_mount, |
853ac43a | 4194 | .kill_sb = kill_litter_super, |
2b8576cb | 4195 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4196 | }; |
4197 | ||
41ffe5d5 | 4198 | int __init shmem_init(void) |
853ac43a | 4199 | { |
41ffe5d5 | 4200 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4201 | |
41ffe5d5 | 4202 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
4203 | BUG_ON(IS_ERR(shm_mnt)); |
4204 | ||
4205 | return 0; | |
4206 | } | |
4207 | ||
41ffe5d5 | 4208 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
4209 | { |
4210 | return 0; | |
4211 | } | |
4212 | ||
3f96b79a HD |
4213 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
4214 | { | |
4215 | return 0; | |
4216 | } | |
4217 | ||
24513264 HD |
4218 | void shmem_unlock_mapping(struct address_space *mapping) |
4219 | { | |
4220 | } | |
4221 | ||
c01d5b30 HD |
4222 | #ifdef CONFIG_MMU |
4223 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4224 | unsigned long addr, unsigned long len, | |
4225 | unsigned long pgoff, unsigned long flags) | |
4226 | { | |
4227 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4228 | } | |
4229 | #endif | |
4230 | ||
41ffe5d5 | 4231 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4232 | { |
41ffe5d5 | 4233 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4234 | } |
4235 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4236 | ||
0b0a0806 HD |
4237 | #define shmem_vm_ops generic_file_vm_ops |
4238 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 4239 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
4240 | #define shmem_acct_size(flags, size) 0 |
4241 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
4242 | |
4243 | #endif /* CONFIG_SHMEM */ | |
4244 | ||
4245 | /* common code */ | |
1da177e4 | 4246 | |
19938e35 | 4247 | static const struct dentry_operations anon_ops = { |
118b2302 | 4248 | .d_dname = simple_dname |
3451538a AV |
4249 | }; |
4250 | ||
703321b6 | 4251 | static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size, |
c7277090 | 4252 | unsigned long flags, unsigned int i_flags) |
1da177e4 | 4253 | { |
6b4d0b27 | 4254 | struct file *res; |
1da177e4 | 4255 | struct inode *inode; |
2c48b9c4 | 4256 | struct path path; |
3451538a | 4257 | struct super_block *sb; |
1da177e4 LT |
4258 | struct qstr this; |
4259 | ||
703321b6 MA |
4260 | if (IS_ERR(mnt)) |
4261 | return ERR_CAST(mnt); | |
1da177e4 | 4262 | |
285b2c4f | 4263 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4264 | return ERR_PTR(-EINVAL); |
4265 | ||
4266 | if (shmem_acct_size(flags, size)) | |
4267 | return ERR_PTR(-ENOMEM); | |
4268 | ||
6b4d0b27 | 4269 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
4270 | this.name = name; |
4271 | this.len = strlen(name); | |
4272 | this.hash = 0; /* will go */ | |
703321b6 MA |
4273 | sb = mnt->mnt_sb; |
4274 | path.mnt = mntget(mnt); | |
3451538a | 4275 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 4276 | if (!path.dentry) |
1da177e4 | 4277 | goto put_memory; |
3451538a | 4278 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 4279 | |
6b4d0b27 | 4280 | res = ERR_PTR(-ENOSPC); |
3451538a | 4281 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 4282 | if (!inode) |
66ee4b88 | 4283 | goto put_memory; |
1da177e4 | 4284 | |
c7277090 | 4285 | inode->i_flags |= i_flags; |
2c48b9c4 | 4286 | d_instantiate(path.dentry, inode); |
1da177e4 | 4287 | inode->i_size = size; |
6d6b77f1 | 4288 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
4289 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
4290 | if (IS_ERR(res)) | |
66ee4b88 | 4291 | goto put_path; |
4b42af81 | 4292 | |
6b4d0b27 | 4293 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 4294 | &shmem_file_operations); |
6b4d0b27 | 4295 | if (IS_ERR(res)) |
66ee4b88 | 4296 | goto put_path; |
4b42af81 | 4297 | |
6b4d0b27 | 4298 | return res; |
1da177e4 | 4299 | |
1da177e4 LT |
4300 | put_memory: |
4301 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
4302 | put_path: |
4303 | path_put(&path); | |
6b4d0b27 | 4304 | return res; |
1da177e4 | 4305 | } |
c7277090 EP |
4306 | |
4307 | /** | |
4308 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4309 | * kernel internal. There will be NO LSM permission checks against the | |
4310 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4311 | * higher layer. The users are the big_key and shm implementations. LSM |
4312 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4313 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4314 | * @size: size to be set for the file | |
4315 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4316 | */ | |
4317 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4318 | { | |
703321b6 | 4319 | return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE); |
c7277090 EP |
4320 | } |
4321 | ||
4322 | /** | |
4323 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4324 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4325 | * @size: size to be set for the file | |
4326 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4327 | */ | |
4328 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4329 | { | |
703321b6 | 4330 | return __shmem_file_setup(shm_mnt, name, size, flags, 0); |
c7277090 | 4331 | } |
395e0ddc | 4332 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4333 | |
703321b6 MA |
4334 | /** |
4335 | * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs | |
4336 | * @mnt: the tmpfs mount where the file will be created | |
4337 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4338 | * @size: size to be set for the file | |
4339 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4340 | */ | |
4341 | struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name, | |
4342 | loff_t size, unsigned long flags) | |
4343 | { | |
4344 | return __shmem_file_setup(mnt, name, size, flags, 0); | |
4345 | } | |
4346 | EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt); | |
4347 | ||
46711810 | 4348 | /** |
1da177e4 | 4349 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
4350 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
4351 | */ | |
4352 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4353 | { | |
4354 | struct file *file; | |
4355 | loff_t size = vma->vm_end - vma->vm_start; | |
4356 | ||
66fc1303 HD |
4357 | /* |
4358 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
4359 | * between XFS directory reading and selinux: since this file is only | |
4360 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4361 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4362 | */ | |
703321b6 | 4363 | file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags); |
1da177e4 LT |
4364 | if (IS_ERR(file)) |
4365 | return PTR_ERR(file); | |
4366 | ||
4367 | if (vma->vm_file) | |
4368 | fput(vma->vm_file); | |
4369 | vma->vm_file = file; | |
4370 | vma->vm_ops = &shmem_vm_ops; | |
f3f0e1d2 | 4371 | |
e496cf3d | 4372 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
4373 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
4374 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
4375 | khugepaged_enter(vma, vma->vm_flags); | |
4376 | } | |
4377 | ||
1da177e4 LT |
4378 | return 0; |
4379 | } | |
d9d90e5e HD |
4380 | |
4381 | /** | |
4382 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
4383 | * @mapping: the page's address_space | |
4384 | * @index: the page index | |
4385 | * @gfp: the page allocator flags to use if allocating | |
4386 | * | |
4387 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4388 | * with any new page allocations done using the specified allocation flags. | |
4389 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
4390 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
4391 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4392 | * | |
68da9f05 HD |
4393 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4394 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4395 | */ |
4396 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4397 | pgoff_t index, gfp_t gfp) | |
4398 | { | |
68da9f05 HD |
4399 | #ifdef CONFIG_SHMEM |
4400 | struct inode *inode = mapping->host; | |
9276aad6 | 4401 | struct page *page; |
68da9f05 HD |
4402 | int error; |
4403 | ||
4404 | BUG_ON(mapping->a_ops != &shmem_aops); | |
9e18eb29 | 4405 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
cfda0526 | 4406 | gfp, NULL, NULL, NULL); |
68da9f05 HD |
4407 | if (error) |
4408 | page = ERR_PTR(error); | |
4409 | else | |
4410 | unlock_page(page); | |
4411 | return page; | |
4412 | #else | |
4413 | /* | |
4414 | * The tiny !SHMEM case uses ramfs without swap | |
4415 | */ | |
d9d90e5e | 4416 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4417 | #endif |
d9d90e5e HD |
4418 | } |
4419 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |