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