]>
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> |
853ac43a | 37 | |
95cc09d6 AA |
38 | #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */ |
39 | ||
853ac43a MM |
40 | static struct vfsmount *shm_mnt; |
41 | ||
42 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
43 | /* |
44 | * This virtual memory filesystem is heavily based on the ramfs. It | |
45 | * extends ramfs by the ability to use swap and honor resource limits | |
46 | * which makes it a completely usable filesystem. | |
47 | */ | |
48 | ||
39f0247d | 49 | #include <linux/xattr.h> |
a5694255 | 50 | #include <linux/exportfs.h> |
1c7c474c | 51 | #include <linux/posix_acl.h> |
feda821e | 52 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 53 | #include <linux/mman.h> |
1da177e4 LT |
54 | #include <linux/string.h> |
55 | #include <linux/slab.h> | |
56 | #include <linux/backing-dev.h> | |
57 | #include <linux/shmem_fs.h> | |
1da177e4 | 58 | #include <linux/writeback.h> |
1da177e4 | 59 | #include <linux/blkdev.h> |
bda97eab | 60 | #include <linux/pagevec.h> |
41ffe5d5 | 61 | #include <linux/percpu_counter.h> |
83e4fa9c | 62 | #include <linux/falloc.h> |
708e3508 | 63 | #include <linux/splice.h> |
1da177e4 LT |
64 | #include <linux/security.h> |
65 | #include <linux/swapops.h> | |
66 | #include <linux/mempolicy.h> | |
67 | #include <linux/namei.h> | |
b00dc3ad | 68 | #include <linux/ctype.h> |
304dbdb7 | 69 | #include <linux/migrate.h> |
c1f60a5a | 70 | #include <linux/highmem.h> |
680d794b | 71 | #include <linux/seq_file.h> |
92562927 | 72 | #include <linux/magic.h> |
9183df25 | 73 | #include <linux/syscalls.h> |
40e041a2 | 74 | #include <linux/fcntl.h> |
9183df25 | 75 | #include <uapi/linux/memfd.h> |
cfda0526 | 76 | #include <linux/userfaultfd_k.h> |
4c27fe4c | 77 | #include <linux/rmap.h> |
2b4db796 | 78 | #include <linux/uuid.h> |
304dbdb7 | 79 | |
7c0f6ba6 | 80 | #include <linux/uaccess.h> |
1da177e4 LT |
81 | #include <asm/pgtable.h> |
82 | ||
dd56b046 MG |
83 | #include "internal.h" |
84 | ||
09cbfeaf KS |
85 | #define BLOCKS_PER_PAGE (PAGE_SIZE/512) |
86 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT) | |
1da177e4 | 87 | |
1da177e4 LT |
88 | /* Pretend that each entry is of this size in directory's i_size */ |
89 | #define BOGO_DIRENT_SIZE 20 | |
90 | ||
69f07ec9 HD |
91 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
92 | #define SHORT_SYMLINK_LEN 128 | |
93 | ||
1aac1400 | 94 | /* |
f00cdc6d HD |
95 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
96 | * inode->i_private (with i_mutex making sure that it has only one user at | |
97 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
98 | */ |
99 | struct shmem_falloc { | |
8e205f77 | 100 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
101 | pgoff_t start; /* start of range currently being fallocated */ |
102 | pgoff_t next; /* the next page offset to be fallocated */ | |
103 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
104 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
105 | }; | |
106 | ||
b76db735 | 107 | #ifdef CONFIG_TMPFS |
680d794b AM |
108 | static unsigned long shmem_default_max_blocks(void) |
109 | { | |
110 | return totalram_pages / 2; | |
111 | } | |
112 | ||
b6450630 | 113 | static int shmem_default_max_inodes(void) |
680d794b | 114 | { |
b6450630 SF |
115 | unsigned long ul; |
116 | ||
117 | ul = INT_MAX; | |
118 | ul = min3(ul, totalram_pages - totalhigh_pages, totalram_pages / 2); | |
119 | return ul; | |
680d794b | 120 | } |
b76db735 | 121 | #endif |
680d794b | 122 | |
bde05d1c HD |
123 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
124 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
125 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 | 126 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 127 | struct page **pagep, enum sgp_type sgp, |
cfda0526 MR |
128 | gfp_t gfp, struct vm_area_struct *vma, |
129 | struct vm_fault *vmf, int *fault_type); | |
68da9f05 | 130 | |
f3f0e1d2 | 131 | int shmem_getpage(struct inode *inode, pgoff_t index, |
9e18eb29 | 132 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
133 | { |
134 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
cfda0526 | 135 | mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL); |
68da9f05 | 136 | } |
1da177e4 | 137 | |
1da177e4 LT |
138 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
139 | { | |
140 | return sb->s_fs_info; | |
141 | } | |
142 | ||
143 | /* | |
144 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
145 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
146 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
147 | * consistent with the pre-accounting of private mappings ... | |
148 | */ | |
149 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
150 | { | |
0b0a0806 | 151 | return (flags & VM_NORESERVE) ? |
191c5424 | 152 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
153 | } |
154 | ||
155 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
156 | { | |
0b0a0806 | 157 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
158 | vm_unacct_memory(VM_ACCT(size)); |
159 | } | |
160 | ||
77142517 KK |
161 | static inline int shmem_reacct_size(unsigned long flags, |
162 | loff_t oldsize, loff_t newsize) | |
163 | { | |
164 | if (!(flags & VM_NORESERVE)) { | |
165 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
166 | return security_vm_enough_memory_mm(current->mm, | |
167 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
168 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
169 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
170 | } | |
171 | return 0; | |
172 | } | |
173 | ||
1da177e4 LT |
174 | /* |
175 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 176 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
177 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
178 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
179 | */ | |
800d8c63 | 180 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 181 | { |
800d8c63 KS |
182 | if (!(flags & VM_NORESERVE)) |
183 | return 0; | |
184 | ||
185 | return security_vm_enough_memory_mm(current->mm, | |
186 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
187 | } |
188 | ||
189 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
190 | { | |
0b0a0806 | 191 | if (flags & VM_NORESERVE) |
09cbfeaf | 192 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
193 | } |
194 | ||
759b9775 | 195 | static const struct super_operations shmem_ops; |
f5e54d6e | 196 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 197 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
198 | static const struct inode_operations shmem_inode_operations; |
199 | static const struct inode_operations shmem_dir_inode_operations; | |
200 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 201 | static const struct vm_operations_struct shmem_vm_ops; |
779750d2 | 202 | static struct file_system_type shmem_fs_type; |
1da177e4 | 203 | |
b0506e48 MR |
204 | bool vma_is_shmem(struct vm_area_struct *vma) |
205 | { | |
206 | return vma->vm_ops == &shmem_vm_ops; | |
207 | } | |
208 | ||
1da177e4 | 209 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 210 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 211 | |
5b04c689 PE |
212 | static int shmem_reserve_inode(struct super_block *sb) |
213 | { | |
214 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
215 | if (sbinfo->max_inodes) { | |
216 | spin_lock(&sbinfo->stat_lock); | |
217 | if (!sbinfo->free_inodes) { | |
218 | spin_unlock(&sbinfo->stat_lock); | |
219 | return -ENOSPC; | |
220 | } | |
221 | sbinfo->free_inodes--; | |
222 | spin_unlock(&sbinfo->stat_lock); | |
223 | } | |
224 | return 0; | |
225 | } | |
226 | ||
227 | static void shmem_free_inode(struct super_block *sb) | |
228 | { | |
229 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
230 | if (sbinfo->max_inodes) { | |
231 | spin_lock(&sbinfo->stat_lock); | |
232 | sbinfo->free_inodes++; | |
233 | spin_unlock(&sbinfo->stat_lock); | |
234 | } | |
235 | } | |
236 | ||
46711810 | 237 | /** |
41ffe5d5 | 238 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
239 | * @inode: inode to recalc |
240 | * | |
241 | * We have to calculate the free blocks since the mm can drop | |
242 | * undirtied hole pages behind our back. | |
243 | * | |
244 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
245 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
246 | * | |
247 | * It has to be called with the spinlock held. | |
248 | */ | |
249 | static void shmem_recalc_inode(struct inode *inode) | |
250 | { | |
251 | struct shmem_inode_info *info = SHMEM_I(inode); | |
252 | long freed; | |
253 | ||
254 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
255 | if (freed > 0) { | |
54af6042 HD |
256 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
257 | if (sbinfo->max_blocks) | |
258 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 259 | info->alloced -= freed; |
54af6042 | 260 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 261 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
262 | } |
263 | } | |
264 | ||
800d8c63 KS |
265 | bool shmem_charge(struct inode *inode, long pages) |
266 | { | |
267 | struct shmem_inode_info *info = SHMEM_I(inode); | |
268 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4595ef88 | 269 | unsigned long flags; |
800d8c63 KS |
270 | |
271 | if (shmem_acct_block(info->flags, pages)) | |
272 | return false; | |
4595ef88 | 273 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
274 | info->alloced += pages; |
275 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
276 | shmem_recalc_inode(inode); | |
4595ef88 | 277 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
278 | inode->i_mapping->nrpages += pages; |
279 | ||
280 | if (!sbinfo->max_blocks) | |
281 | return true; | |
282 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
283 | sbinfo->max_blocks - pages) > 0) { | |
284 | inode->i_mapping->nrpages -= pages; | |
4595ef88 | 285 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
286 | info->alloced -= pages; |
287 | shmem_recalc_inode(inode); | |
4595ef88 | 288 | spin_unlock_irqrestore(&info->lock, flags); |
71664665 | 289 | shmem_unacct_blocks(info->flags, pages); |
800d8c63 KS |
290 | return false; |
291 | } | |
292 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
293 | return true; | |
294 | } | |
295 | ||
296 | void shmem_uncharge(struct inode *inode, long pages) | |
297 | { | |
298 | struct shmem_inode_info *info = SHMEM_I(inode); | |
299 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4595ef88 | 300 | unsigned long flags; |
800d8c63 | 301 | |
4595ef88 | 302 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
303 | info->alloced -= pages; |
304 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
305 | shmem_recalc_inode(inode); | |
4595ef88 | 306 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
307 | |
308 | if (sbinfo->max_blocks) | |
309 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
71664665 | 310 | shmem_unacct_blocks(info->flags, pages); |
800d8c63 KS |
311 | } |
312 | ||
7a5d0fbb HD |
313 | /* |
314 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
315 | */ | |
316 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
317 | pgoff_t index, void *expected, void *replacement) | |
318 | { | |
f7942430 | 319 | struct radix_tree_node *node; |
7a5d0fbb | 320 | void **pslot; |
6dbaf22c | 321 | void *item; |
7a5d0fbb HD |
322 | |
323 | VM_BUG_ON(!expected); | |
6dbaf22c | 324 | VM_BUG_ON(!replacement); |
f7942430 JW |
325 | item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot); |
326 | if (!item) | |
6dbaf22c | 327 | return -ENOENT; |
7a5d0fbb HD |
328 | if (item != expected) |
329 | return -ENOENT; | |
4d693d08 JW |
330 | __radix_tree_replace(&mapping->page_tree, node, pslot, |
331 | replacement, NULL, NULL); | |
7a5d0fbb HD |
332 | return 0; |
333 | } | |
334 | ||
d1899228 HD |
335 | /* |
336 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
337 | * that an entry was not already brought back from swap by a racing thread. | |
338 | * | |
339 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
340 | * might be reused, and again be SwapCache, using the same swap as before. | |
341 | */ | |
342 | static bool shmem_confirm_swap(struct address_space *mapping, | |
343 | pgoff_t index, swp_entry_t swap) | |
344 | { | |
345 | void *item; | |
346 | ||
347 | rcu_read_lock(); | |
348 | item = radix_tree_lookup(&mapping->page_tree, index); | |
349 | rcu_read_unlock(); | |
350 | return item == swp_to_radix_entry(swap); | |
351 | } | |
352 | ||
5a6e75f8 KS |
353 | /* |
354 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
355 | * | |
356 | * SHMEM_HUGE_NEVER: | |
357 | * disables huge pages for the mount; | |
358 | * SHMEM_HUGE_ALWAYS: | |
359 | * enables huge pages for the mount; | |
360 | * SHMEM_HUGE_WITHIN_SIZE: | |
361 | * only allocate huge pages if the page will be fully within i_size, | |
362 | * also respect fadvise()/madvise() hints; | |
363 | * SHMEM_HUGE_ADVISE: | |
364 | * only allocate huge pages if requested with fadvise()/madvise(); | |
365 | */ | |
366 | ||
367 | #define SHMEM_HUGE_NEVER 0 | |
368 | #define SHMEM_HUGE_ALWAYS 1 | |
369 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
370 | #define SHMEM_HUGE_ADVISE 3 | |
371 | ||
372 | /* | |
373 | * Special values. | |
374 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
375 | * | |
376 | * SHMEM_HUGE_DENY: | |
377 | * disables huge on shm_mnt and all mounts, for emergency use; | |
378 | * SHMEM_HUGE_FORCE: | |
379 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
380 | * | |
381 | */ | |
382 | #define SHMEM_HUGE_DENY (-1) | |
383 | #define SHMEM_HUGE_FORCE (-2) | |
384 | ||
e496cf3d | 385 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
386 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
387 | ||
388 | int shmem_huge __read_mostly; | |
389 | ||
f1f5929c | 390 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
391 | static int shmem_parse_huge(const char *str) |
392 | { | |
393 | if (!strcmp(str, "never")) | |
394 | return SHMEM_HUGE_NEVER; | |
395 | if (!strcmp(str, "always")) | |
396 | return SHMEM_HUGE_ALWAYS; | |
397 | if (!strcmp(str, "within_size")) | |
398 | return SHMEM_HUGE_WITHIN_SIZE; | |
399 | if (!strcmp(str, "advise")) | |
400 | return SHMEM_HUGE_ADVISE; | |
401 | if (!strcmp(str, "deny")) | |
402 | return SHMEM_HUGE_DENY; | |
403 | if (!strcmp(str, "force")) | |
404 | return SHMEM_HUGE_FORCE; | |
405 | return -EINVAL; | |
406 | } | |
407 | ||
408 | static const char *shmem_format_huge(int huge) | |
409 | { | |
410 | switch (huge) { | |
411 | case SHMEM_HUGE_NEVER: | |
412 | return "never"; | |
413 | case SHMEM_HUGE_ALWAYS: | |
414 | return "always"; | |
415 | case SHMEM_HUGE_WITHIN_SIZE: | |
416 | return "within_size"; | |
417 | case SHMEM_HUGE_ADVISE: | |
418 | return "advise"; | |
419 | case SHMEM_HUGE_DENY: | |
420 | return "deny"; | |
421 | case SHMEM_HUGE_FORCE: | |
422 | return "force"; | |
423 | default: | |
424 | VM_BUG_ON(1); | |
425 | return "bad_val"; | |
426 | } | |
427 | } | |
f1f5929c | 428 | #endif |
5a6e75f8 | 429 | |
779750d2 KS |
430 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
431 | struct shrink_control *sc, unsigned long nr_to_split) | |
432 | { | |
433 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 434 | LIST_HEAD(to_remove); |
779750d2 KS |
435 | struct inode *inode; |
436 | struct shmem_inode_info *info; | |
437 | struct page *page; | |
438 | unsigned long batch = sc ? sc->nr_to_scan : 128; | |
439 | int removed = 0, split = 0; | |
440 | ||
441 | if (list_empty(&sbinfo->shrinklist)) | |
442 | return SHRINK_STOP; | |
443 | ||
444 | spin_lock(&sbinfo->shrinklist_lock); | |
445 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
446 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
447 | ||
448 | /* pin the inode */ | |
449 | inode = igrab(&info->vfs_inode); | |
450 | ||
451 | /* inode is about to be evicted */ | |
452 | if (!inode) { | |
453 | list_del_init(&info->shrinklist); | |
454 | removed++; | |
455 | goto next; | |
456 | } | |
457 | ||
458 | /* Check if there's anything to gain */ | |
459 | if (round_up(inode->i_size, PAGE_SIZE) == | |
460 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 461 | list_move(&info->shrinklist, &to_remove); |
779750d2 | 462 | removed++; |
779750d2 KS |
463 | goto next; |
464 | } | |
465 | ||
466 | list_move(&info->shrinklist, &list); | |
467 | next: | |
468 | if (!--batch) | |
469 | break; | |
470 | } | |
471 | spin_unlock(&sbinfo->shrinklist_lock); | |
472 | ||
253fd0f0 KS |
473 | list_for_each_safe(pos, next, &to_remove) { |
474 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
475 | inode = &info->vfs_inode; | |
476 | list_del_init(&info->shrinklist); | |
477 | iput(inode); | |
478 | } | |
479 | ||
779750d2 KS |
480 | list_for_each_safe(pos, next, &list) { |
481 | int ret; | |
482 | ||
483 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
484 | inode = &info->vfs_inode; | |
485 | ||
486 | if (nr_to_split && split >= nr_to_split) { | |
487 | iput(inode); | |
488 | continue; | |
489 | } | |
490 | ||
491 | page = find_lock_page(inode->i_mapping, | |
492 | (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT); | |
493 | if (!page) | |
494 | goto drop; | |
495 | ||
496 | if (!PageTransHuge(page)) { | |
497 | unlock_page(page); | |
498 | put_page(page); | |
499 | goto drop; | |
500 | } | |
501 | ||
502 | ret = split_huge_page(page); | |
503 | unlock_page(page); | |
504 | put_page(page); | |
505 | ||
506 | if (ret) { | |
507 | /* split failed: leave it on the list */ | |
508 | iput(inode); | |
509 | continue; | |
510 | } | |
511 | ||
512 | split++; | |
513 | drop: | |
514 | list_del_init(&info->shrinklist); | |
515 | removed++; | |
516 | iput(inode); | |
517 | } | |
518 | ||
519 | spin_lock(&sbinfo->shrinklist_lock); | |
520 | list_splice_tail(&list, &sbinfo->shrinklist); | |
521 | sbinfo->shrinklist_len -= removed; | |
522 | spin_unlock(&sbinfo->shrinklist_lock); | |
523 | ||
524 | return split; | |
525 | } | |
526 | ||
527 | static long shmem_unused_huge_scan(struct super_block *sb, | |
528 | struct shrink_control *sc) | |
529 | { | |
530 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
531 | ||
532 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
533 | return SHRINK_STOP; | |
534 | ||
535 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
536 | } | |
537 | ||
538 | static long shmem_unused_huge_count(struct super_block *sb, | |
539 | struct shrink_control *sc) | |
540 | { | |
541 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
542 | return READ_ONCE(sbinfo->shrinklist_len); | |
543 | } | |
e496cf3d | 544 | #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 KS |
545 | |
546 | #define shmem_huge SHMEM_HUGE_DENY | |
547 | ||
779750d2 KS |
548 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
549 | struct shrink_control *sc, unsigned long nr_to_split) | |
550 | { | |
551 | return 0; | |
552 | } | |
e496cf3d | 553 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 554 | |
46f65ec1 HD |
555 | /* |
556 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
557 | */ | |
558 | static int shmem_add_to_page_cache(struct page *page, | |
559 | struct address_space *mapping, | |
fed400a1 | 560 | pgoff_t index, void *expected) |
46f65ec1 | 561 | { |
800d8c63 | 562 | int error, nr = hpage_nr_pages(page); |
46f65ec1 | 563 | |
800d8c63 KS |
564 | VM_BUG_ON_PAGE(PageTail(page), page); |
565 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
566 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
567 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 568 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 569 | |
800d8c63 | 570 | page_ref_add(page, nr); |
b065b432 HD |
571 | page->mapping = mapping; |
572 | page->index = index; | |
573 | ||
574 | spin_lock_irq(&mapping->tree_lock); | |
800d8c63 KS |
575 | if (PageTransHuge(page)) { |
576 | void __rcu **results; | |
577 | pgoff_t idx; | |
578 | int i; | |
579 | ||
580 | error = 0; | |
581 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, | |
582 | &results, &idx, index, 1) && | |
583 | idx < index + HPAGE_PMD_NR) { | |
584 | error = -EEXIST; | |
585 | } | |
586 | ||
587 | if (!error) { | |
588 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
589 | error = radix_tree_insert(&mapping->page_tree, | |
590 | index + i, page + i); | |
591 | VM_BUG_ON(error); | |
592 | } | |
593 | count_vm_event(THP_FILE_ALLOC); | |
594 | } | |
595 | } else if (!expected) { | |
b065b432 | 596 | error = radix_tree_insert(&mapping->page_tree, index, page); |
800d8c63 | 597 | } else { |
b065b432 HD |
598 | error = shmem_radix_tree_replace(mapping, index, expected, |
599 | page); | |
800d8c63 KS |
600 | } |
601 | ||
46f65ec1 | 602 | if (!error) { |
800d8c63 KS |
603 | mapping->nrpages += nr; |
604 | if (PageTransHuge(page)) | |
11fb9989 MG |
605 | __inc_node_page_state(page, NR_SHMEM_THPS); |
606 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); | |
607 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr); | |
b065b432 HD |
608 | spin_unlock_irq(&mapping->tree_lock); |
609 | } else { | |
610 | page->mapping = NULL; | |
611 | spin_unlock_irq(&mapping->tree_lock); | |
800d8c63 | 612 | page_ref_sub(page, nr); |
46f65ec1 | 613 | } |
46f65ec1 HD |
614 | return error; |
615 | } | |
616 | ||
6922c0c7 HD |
617 | /* |
618 | * Like delete_from_page_cache, but substitutes swap for page. | |
619 | */ | |
620 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
621 | { | |
622 | struct address_space *mapping = page->mapping; | |
623 | int error; | |
624 | ||
800d8c63 KS |
625 | VM_BUG_ON_PAGE(PageCompound(page), page); |
626 | ||
6922c0c7 HD |
627 | spin_lock_irq(&mapping->tree_lock); |
628 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
629 | page->mapping = NULL; | |
630 | mapping->nrpages--; | |
11fb9989 MG |
631 | __dec_node_page_state(page, NR_FILE_PAGES); |
632 | __dec_node_page_state(page, NR_SHMEM); | |
6922c0c7 | 633 | spin_unlock_irq(&mapping->tree_lock); |
09cbfeaf | 634 | put_page(page); |
6922c0c7 HD |
635 | BUG_ON(error); |
636 | } | |
637 | ||
7a5d0fbb HD |
638 | /* |
639 | * Remove swap entry from radix tree, free the swap and its page cache. | |
640 | */ | |
641 | static int shmem_free_swap(struct address_space *mapping, | |
642 | pgoff_t index, void *radswap) | |
643 | { | |
6dbaf22c | 644 | void *old; |
7a5d0fbb HD |
645 | |
646 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 647 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 648 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
649 | if (old != radswap) |
650 | return -ENOENT; | |
651 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
652 | return 0; | |
7a5d0fbb HD |
653 | } |
654 | ||
6a15a370 VB |
655 | /* |
656 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 657 | * given offsets are swapped out. |
6a15a370 VB |
658 | * |
659 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
660 | * as long as the inode doesn't go away and racy results are not a problem. | |
661 | */ | |
48131e03 VB |
662 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
663 | pgoff_t start, pgoff_t end) | |
6a15a370 | 664 | { |
6a15a370 VB |
665 | struct radix_tree_iter iter; |
666 | void **slot; | |
667 | struct page *page; | |
48131e03 | 668 | unsigned long swapped = 0; |
6a15a370 VB |
669 | |
670 | rcu_read_lock(); | |
671 | ||
6a15a370 VB |
672 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
673 | if (iter.index >= end) | |
674 | break; | |
675 | ||
676 | page = radix_tree_deref_slot(slot); | |
677 | ||
2cf938aa MW |
678 | if (radix_tree_deref_retry(page)) { |
679 | slot = radix_tree_iter_retry(&iter); | |
680 | continue; | |
681 | } | |
6a15a370 VB |
682 | |
683 | if (radix_tree_exceptional_entry(page)) | |
684 | swapped++; | |
685 | ||
686 | if (need_resched()) { | |
148deab2 | 687 | slot = radix_tree_iter_resume(slot, &iter); |
6a15a370 | 688 | cond_resched_rcu(); |
6a15a370 VB |
689 | } |
690 | } | |
691 | ||
692 | rcu_read_unlock(); | |
693 | ||
694 | return swapped << PAGE_SHIFT; | |
695 | } | |
696 | ||
48131e03 VB |
697 | /* |
698 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
699 | * given vma is swapped out. | |
700 | * | |
701 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
702 | * as long as the inode doesn't go away and racy results are not a problem. | |
703 | */ | |
704 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
705 | { | |
706 | struct inode *inode = file_inode(vma->vm_file); | |
707 | struct shmem_inode_info *info = SHMEM_I(inode); | |
708 | struct address_space *mapping = inode->i_mapping; | |
709 | unsigned long swapped; | |
710 | ||
711 | /* Be careful as we don't hold info->lock */ | |
712 | swapped = READ_ONCE(info->swapped); | |
713 | ||
714 | /* | |
715 | * The easier cases are when the shmem object has nothing in swap, or | |
716 | * the vma maps it whole. Then we can simply use the stats that we | |
717 | * already track. | |
718 | */ | |
719 | if (!swapped) | |
720 | return 0; | |
721 | ||
722 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
723 | return swapped << PAGE_SHIFT; | |
724 | ||
725 | /* Here comes the more involved part */ | |
726 | return shmem_partial_swap_usage(mapping, | |
727 | linear_page_index(vma, vma->vm_start), | |
728 | linear_page_index(vma, vma->vm_end)); | |
729 | } | |
730 | ||
24513264 HD |
731 | /* |
732 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
733 | */ | |
734 | void shmem_unlock_mapping(struct address_space *mapping) | |
735 | { | |
736 | struct pagevec pvec; | |
737 | pgoff_t indices[PAGEVEC_SIZE]; | |
738 | pgoff_t index = 0; | |
739 | ||
740 | pagevec_init(&pvec, 0); | |
741 | /* | |
742 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
743 | */ | |
744 | while (!mapping_unevictable(mapping)) { | |
745 | /* | |
746 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
747 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
748 | */ | |
0cd6144a JW |
749 | pvec.nr = find_get_entries(mapping, index, |
750 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
751 | if (!pvec.nr) |
752 | break; | |
753 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 754 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
755 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
756 | pagevec_release(&pvec); | |
757 | cond_resched(); | |
758 | } | |
7a5d0fbb HD |
759 | } |
760 | ||
761 | /* | |
762 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 763 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 764 | */ |
1635f6a7 HD |
765 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
766 | bool unfalloc) | |
1da177e4 | 767 | { |
285b2c4f | 768 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 769 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
770 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
771 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
772 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
773 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 774 | struct pagevec pvec; |
7a5d0fbb HD |
775 | pgoff_t indices[PAGEVEC_SIZE]; |
776 | long nr_swaps_freed = 0; | |
285b2c4f | 777 | pgoff_t index; |
bda97eab HD |
778 | int i; |
779 | ||
83e4fa9c HD |
780 | if (lend == -1) |
781 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
782 | |
783 | pagevec_init(&pvec, 0); | |
784 | index = start; | |
83e4fa9c | 785 | while (index < end) { |
0cd6144a JW |
786 | pvec.nr = find_get_entries(mapping, index, |
787 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
788 | pvec.pages, indices); | |
7a5d0fbb HD |
789 | if (!pvec.nr) |
790 | break; | |
bda97eab HD |
791 | for (i = 0; i < pagevec_count(&pvec); i++) { |
792 | struct page *page = pvec.pages[i]; | |
793 | ||
7a5d0fbb | 794 | index = indices[i]; |
83e4fa9c | 795 | if (index >= end) |
bda97eab HD |
796 | break; |
797 | ||
7a5d0fbb | 798 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
799 | if (unfalloc) |
800 | continue; | |
7a5d0fbb HD |
801 | nr_swaps_freed += !shmem_free_swap(mapping, |
802 | index, page); | |
bda97eab | 803 | continue; |
7a5d0fbb HD |
804 | } |
805 | ||
800d8c63 KS |
806 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
807 | ||
7a5d0fbb | 808 | if (!trylock_page(page)) |
bda97eab | 809 | continue; |
800d8c63 KS |
810 | |
811 | if (PageTransTail(page)) { | |
812 | /* Middle of THP: zero out the page */ | |
813 | clear_highpage(page); | |
814 | unlock_page(page); | |
815 | continue; | |
816 | } else if (PageTransHuge(page)) { | |
817 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
818 | /* | |
819 | * Range ends in the middle of THP: | |
820 | * zero out the page | |
821 | */ | |
822 | clear_highpage(page); | |
823 | unlock_page(page); | |
824 | continue; | |
825 | } | |
826 | index += HPAGE_PMD_NR - 1; | |
827 | i += HPAGE_PMD_NR - 1; | |
828 | } | |
829 | ||
1635f6a7 | 830 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
831 | VM_BUG_ON_PAGE(PageTail(page), page); |
832 | if (page_mapping(page) == mapping) { | |
309381fe | 833 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
834 | truncate_inode_page(mapping, page); |
835 | } | |
bda97eab | 836 | } |
bda97eab HD |
837 | unlock_page(page); |
838 | } | |
0cd6144a | 839 | pagevec_remove_exceptionals(&pvec); |
24513264 | 840 | pagevec_release(&pvec); |
bda97eab HD |
841 | cond_resched(); |
842 | index++; | |
843 | } | |
1da177e4 | 844 | |
83e4fa9c | 845 | if (partial_start) { |
bda97eab | 846 | struct page *page = NULL; |
9e18eb29 | 847 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 848 | if (page) { |
09cbfeaf | 849 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
850 | if (start > end) { |
851 | top = partial_end; | |
852 | partial_end = 0; | |
853 | } | |
854 | zero_user_segment(page, partial_start, top); | |
855 | set_page_dirty(page); | |
856 | unlock_page(page); | |
09cbfeaf | 857 | put_page(page); |
83e4fa9c HD |
858 | } |
859 | } | |
860 | if (partial_end) { | |
861 | struct page *page = NULL; | |
9e18eb29 | 862 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
863 | if (page) { |
864 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
865 | set_page_dirty(page); |
866 | unlock_page(page); | |
09cbfeaf | 867 | put_page(page); |
bda97eab HD |
868 | } |
869 | } | |
83e4fa9c HD |
870 | if (start >= end) |
871 | return; | |
bda97eab HD |
872 | |
873 | index = start; | |
b1a36650 | 874 | while (index < end) { |
bda97eab | 875 | cond_resched(); |
0cd6144a JW |
876 | |
877 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 878 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 879 | pvec.pages, indices); |
7a5d0fbb | 880 | if (!pvec.nr) { |
b1a36650 HD |
881 | /* If all gone or hole-punch or unfalloc, we're done */ |
882 | if (index == start || end != -1) | |
bda97eab | 883 | break; |
b1a36650 | 884 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
885 | index = start; |
886 | continue; | |
887 | } | |
bda97eab HD |
888 | for (i = 0; i < pagevec_count(&pvec); i++) { |
889 | struct page *page = pvec.pages[i]; | |
890 | ||
7a5d0fbb | 891 | index = indices[i]; |
83e4fa9c | 892 | if (index >= end) |
bda97eab HD |
893 | break; |
894 | ||
7a5d0fbb | 895 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
896 | if (unfalloc) |
897 | continue; | |
b1a36650 HD |
898 | if (shmem_free_swap(mapping, index, page)) { |
899 | /* Swap was replaced by page: retry */ | |
900 | index--; | |
901 | break; | |
902 | } | |
903 | nr_swaps_freed++; | |
7a5d0fbb HD |
904 | continue; |
905 | } | |
906 | ||
bda97eab | 907 | lock_page(page); |
800d8c63 KS |
908 | |
909 | if (PageTransTail(page)) { | |
910 | /* Middle of THP: zero out the page */ | |
911 | clear_highpage(page); | |
912 | unlock_page(page); | |
913 | /* | |
914 | * Partial thp truncate due 'start' in middle | |
915 | * of THP: don't need to look on these pages | |
916 | * again on !pvec.nr restart. | |
917 | */ | |
918 | if (index != round_down(end, HPAGE_PMD_NR)) | |
919 | start++; | |
920 | continue; | |
921 | } else if (PageTransHuge(page)) { | |
922 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
923 | /* | |
924 | * Range ends in the middle of THP: | |
925 | * zero out the page | |
926 | */ | |
927 | clear_highpage(page); | |
928 | unlock_page(page); | |
929 | continue; | |
930 | } | |
931 | index += HPAGE_PMD_NR - 1; | |
932 | i += HPAGE_PMD_NR - 1; | |
933 | } | |
934 | ||
1635f6a7 | 935 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
936 | VM_BUG_ON_PAGE(PageTail(page), page); |
937 | if (page_mapping(page) == mapping) { | |
309381fe | 938 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 939 | truncate_inode_page(mapping, page); |
b1a36650 HD |
940 | } else { |
941 | /* Page was replaced by swap: retry */ | |
942 | unlock_page(page); | |
943 | index--; | |
944 | break; | |
1635f6a7 | 945 | } |
7a5d0fbb | 946 | } |
bda97eab HD |
947 | unlock_page(page); |
948 | } | |
0cd6144a | 949 | pagevec_remove_exceptionals(&pvec); |
24513264 | 950 | pagevec_release(&pvec); |
bda97eab HD |
951 | index++; |
952 | } | |
94c1e62d | 953 | |
4595ef88 | 954 | spin_lock_irq(&info->lock); |
7a5d0fbb | 955 | info->swapped -= nr_swaps_freed; |
1da177e4 | 956 | shmem_recalc_inode(inode); |
4595ef88 | 957 | spin_unlock_irq(&info->lock); |
1635f6a7 | 958 | } |
1da177e4 | 959 | |
1635f6a7 HD |
960 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
961 | { | |
962 | shmem_undo_range(inode, lstart, lend, false); | |
078cd827 | 963 | inode->i_ctime = inode->i_mtime = current_time(inode); |
1da177e4 | 964 | } |
94c1e62d | 965 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 966 | |
a528d35e DH |
967 | static int shmem_getattr(const struct path *path, struct kstat *stat, |
968 | u32 request_mask, unsigned int query_flags) | |
44a30220 | 969 | { |
a528d35e | 970 | struct inode *inode = path->dentry->d_inode; |
44a30220 YZ |
971 | struct shmem_inode_info *info = SHMEM_I(inode); |
972 | ||
d0424c42 | 973 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
4595ef88 | 974 | spin_lock_irq(&info->lock); |
d0424c42 | 975 | shmem_recalc_inode(inode); |
4595ef88 | 976 | spin_unlock_irq(&info->lock); |
d0424c42 | 977 | } |
44a30220 | 978 | generic_fillattr(inode, stat); |
44a30220 YZ |
979 | return 0; |
980 | } | |
981 | ||
94c1e62d | 982 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 983 | { |
75c3cfa8 | 984 | struct inode *inode = d_inode(dentry); |
40e041a2 | 985 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 986 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 LT |
987 | int error; |
988 | ||
31051c85 | 989 | error = setattr_prepare(dentry, attr); |
db78b877 CH |
990 | if (error) |
991 | return error; | |
992 | ||
94c1e62d HD |
993 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
994 | loff_t oldsize = inode->i_size; | |
995 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 996 | |
40e041a2 DH |
997 | /* protected by i_mutex */ |
998 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
999 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
1000 | return -EPERM; | |
1001 | ||
94c1e62d | 1002 | if (newsize != oldsize) { |
77142517 KK |
1003 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
1004 | oldsize, newsize); | |
1005 | if (error) | |
1006 | return error; | |
94c1e62d | 1007 | i_size_write(inode, newsize); |
078cd827 | 1008 | inode->i_ctime = inode->i_mtime = current_time(inode); |
94c1e62d | 1009 | } |
afa2db2f | 1010 | if (newsize <= oldsize) { |
94c1e62d | 1011 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
1012 | if (oldsize > holebegin) |
1013 | unmap_mapping_range(inode->i_mapping, | |
1014 | holebegin, 0, 1); | |
1015 | if (info->alloced) | |
1016 | shmem_truncate_range(inode, | |
1017 | newsize, (loff_t)-1); | |
94c1e62d | 1018 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
1019 | if (oldsize > holebegin) |
1020 | unmap_mapping_range(inode->i_mapping, | |
1021 | holebegin, 0, 1); | |
779750d2 KS |
1022 | |
1023 | /* | |
1024 | * Part of the huge page can be beyond i_size: subject | |
1025 | * to shrink under memory pressure. | |
1026 | */ | |
1027 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { | |
1028 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1029 | /* |
1030 | * _careful to defend against unlocked access to | |
1031 | * ->shrink_list in shmem_unused_huge_shrink() | |
1032 | */ | |
1033 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1034 | list_add_tail(&info->shrinklist, |
1035 | &sbinfo->shrinklist); | |
1036 | sbinfo->shrinklist_len++; | |
1037 | } | |
1038 | spin_unlock(&sbinfo->shrinklist_lock); | |
1039 | } | |
94c1e62d | 1040 | } |
1da177e4 LT |
1041 | } |
1042 | ||
db78b877 | 1043 | setattr_copy(inode, attr); |
db78b877 | 1044 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 1045 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
1046 | return error; |
1047 | } | |
1048 | ||
1f895f75 | 1049 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1050 | { |
1da177e4 | 1051 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1052 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 | 1053 | |
3889e6e7 | 1054 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
1055 | shmem_unacct_size(info->flags, inode->i_size); |
1056 | inode->i_size = 0; | |
3889e6e7 | 1057 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1058 | if (!list_empty(&info->shrinklist)) { |
1059 | spin_lock(&sbinfo->shrinklist_lock); | |
1060 | if (!list_empty(&info->shrinklist)) { | |
1061 | list_del_init(&info->shrinklist); | |
1062 | sbinfo->shrinklist_len--; | |
1063 | } | |
1064 | spin_unlock(&sbinfo->shrinklist_lock); | |
1065 | } | |
1da177e4 | 1066 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 1067 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 1068 | list_del_init(&info->swaplist); |
cb5f7b9a | 1069 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1070 | } |
3ed47db3 | 1071 | } |
b09e0fa4 | 1072 | |
38f38657 | 1073 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 1074 | WARN_ON(inode->i_blocks); |
b6450630 SF |
1075 | if (!sbinfo->idr_nouse && inode->i_ino) { |
1076 | mutex_lock(&sbinfo->idr_lock); | |
1077 | idr_remove(&sbinfo->idr, inode->i_ino); | |
1078 | mutex_unlock(&sbinfo->idr_lock); | |
1079 | } | |
5b04c689 | 1080 | shmem_free_inode(inode->i_sb); |
dbd5768f | 1081 | clear_inode(inode); |
1da177e4 LT |
1082 | } |
1083 | ||
478922e2 MW |
1084 | static unsigned long find_swap_entry(struct radix_tree_root *root, void *item) |
1085 | { | |
1086 | struct radix_tree_iter iter; | |
1087 | void **slot; | |
1088 | unsigned long found = -1; | |
1089 | unsigned int checked = 0; | |
1090 | ||
1091 | rcu_read_lock(); | |
1092 | radix_tree_for_each_slot(slot, root, &iter, 0) { | |
1093 | if (*slot == item) { | |
1094 | found = iter.index; | |
1095 | break; | |
1096 | } | |
1097 | checked++; | |
1098 | if ((checked % 4096) != 0) | |
1099 | continue; | |
1100 | slot = radix_tree_iter_resume(slot, &iter); | |
1101 | cond_resched_rcu(); | |
1102 | } | |
1103 | ||
1104 | rcu_read_unlock(); | |
1105 | return found; | |
1106 | } | |
1107 | ||
46f65ec1 HD |
1108 | /* |
1109 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1110 | */ | |
41ffe5d5 | 1111 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 1112 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 1113 | { |
285b2c4f | 1114 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 1115 | void *radswap; |
41ffe5d5 | 1116 | pgoff_t index; |
bde05d1c HD |
1117 | gfp_t gfp; |
1118 | int error = 0; | |
1da177e4 | 1119 | |
46f65ec1 | 1120 | radswap = swp_to_radix_entry(swap); |
478922e2 | 1121 | index = find_swap_entry(&mapping->page_tree, radswap); |
46f65ec1 | 1122 | if (index == -1) |
00501b53 | 1123 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 1124 | |
1b1b32f2 HD |
1125 | /* |
1126 | * Move _head_ to start search for next from here. | |
1f895f75 | 1127 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 1128 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 1129 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
1130 | */ |
1131 | if (shmem_swaplist.next != &info->swaplist) | |
1132 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 1133 | |
bde05d1c HD |
1134 | gfp = mapping_gfp_mask(mapping); |
1135 | if (shmem_should_replace_page(*pagep, gfp)) { | |
1136 | mutex_unlock(&shmem_swaplist_mutex); | |
1137 | error = shmem_replace_page(pagep, gfp, info, index); | |
1138 | mutex_lock(&shmem_swaplist_mutex); | |
1139 | /* | |
1140 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
1141 | * allocation, but the inode might have been freed while we |
1142 | * dropped it: although a racing shmem_evict_inode() cannot | |
1143 | * complete without emptying the radix_tree, our page lock | |
1144 | * on this swapcache page is not enough to prevent that - | |
1145 | * free_swap_and_cache() of our swap entry will only | |
1146 | * trylock_page(), removing swap from radix_tree whatever. | |
1147 | * | |
1148 | * We must not proceed to shmem_add_to_page_cache() if the | |
1149 | * inode has been freed, but of course we cannot rely on | |
1150 | * inode or mapping or info to check that. However, we can | |
1151 | * safely check if our swap entry is still in use (and here | |
1152 | * it can't have got reused for another page): if it's still | |
1153 | * in use, then the inode cannot have been freed yet, and we | |
1154 | * can safely proceed (if it's no longer in use, that tells | |
1155 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
1156 | */ |
1157 | if (!page_swapcount(*pagep)) | |
1158 | error = -ENOENT; | |
1159 | } | |
1160 | ||
d13d1443 | 1161 | /* |
778dd893 HD |
1162 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
1163 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
1164 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 1165 | */ |
bde05d1c HD |
1166 | if (!error) |
1167 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 1168 | radswap); |
48f170fb | 1169 | if (error != -ENOMEM) { |
46f65ec1 HD |
1170 | /* |
1171 | * Truncation and eviction use free_swap_and_cache(), which | |
1172 | * only does trylock page: if we raced, best clean up here. | |
1173 | */ | |
bde05d1c HD |
1174 | delete_from_swap_cache(*pagep); |
1175 | set_page_dirty(*pagep); | |
46f65ec1 | 1176 | if (!error) { |
4595ef88 | 1177 | spin_lock_irq(&info->lock); |
46f65ec1 | 1178 | info->swapped--; |
4595ef88 | 1179 | spin_unlock_irq(&info->lock); |
46f65ec1 HD |
1180 | swap_free(swap); |
1181 | } | |
1da177e4 | 1182 | } |
2e0e26c7 | 1183 | return error; |
1da177e4 LT |
1184 | } |
1185 | ||
1186 | /* | |
46f65ec1 | 1187 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 1188 | */ |
41ffe5d5 | 1189 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 1190 | { |
41ffe5d5 | 1191 | struct list_head *this, *next; |
1da177e4 | 1192 | struct shmem_inode_info *info; |
00501b53 | 1193 | struct mem_cgroup *memcg; |
bde05d1c HD |
1194 | int error = 0; |
1195 | ||
1196 | /* | |
1197 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 1198 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 1199 | */ |
0142ef6c | 1200 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 1201 | goto out; |
778dd893 HD |
1202 | |
1203 | /* | |
1204 | * Charge page using GFP_KERNEL while we can wait, before taking | |
1205 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
1206 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 1207 | */ |
f627c2f5 KS |
1208 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg, |
1209 | false); | |
778dd893 HD |
1210 | if (error) |
1211 | goto out; | |
46f65ec1 | 1212 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 1213 | error = -EAGAIN; |
1da177e4 | 1214 | |
cb5f7b9a | 1215 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
1216 | list_for_each_safe(this, next, &shmem_swaplist) { |
1217 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 1218 | if (info->swapped) |
00501b53 | 1219 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
1220 | else |
1221 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1222 | cond_resched(); |
00501b53 | 1223 | if (error != -EAGAIN) |
778dd893 | 1224 | break; |
00501b53 | 1225 | /* found nothing in this: move on to search the next */ |
1da177e4 | 1226 | } |
cb5f7b9a | 1227 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1228 | |
00501b53 JW |
1229 | if (error) { |
1230 | if (error != -ENOMEM) | |
1231 | error = 0; | |
f627c2f5 | 1232 | mem_cgroup_cancel_charge(page, memcg, false); |
00501b53 | 1233 | } else |
f627c2f5 | 1234 | mem_cgroup_commit_charge(page, memcg, true, false); |
778dd893 | 1235 | out: |
aaa46865 | 1236 | unlock_page(page); |
09cbfeaf | 1237 | put_page(page); |
778dd893 | 1238 | return error; |
1da177e4 LT |
1239 | } |
1240 | ||
1241 | /* | |
1242 | * Move the page from the page cache to the swap cache. | |
1243 | */ | |
1244 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1245 | { | |
1246 | struct shmem_inode_info *info; | |
1da177e4 | 1247 | struct address_space *mapping; |
1da177e4 | 1248 | struct inode *inode; |
6922c0c7 HD |
1249 | swp_entry_t swap; |
1250 | pgoff_t index; | |
1da177e4 | 1251 | |
800d8c63 | 1252 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1253 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1254 | mapping = page->mapping; |
1255 | index = page->index; | |
1256 | inode = mapping->host; | |
1257 | info = SHMEM_I(inode); | |
1258 | if (info->flags & VM_LOCKED) | |
1259 | goto redirty; | |
d9fe526a | 1260 | if (!total_swap_pages) |
1da177e4 LT |
1261 | goto redirty; |
1262 | ||
d9fe526a | 1263 | /* |
97b713ba CH |
1264 | * Our capabilities prevent regular writeback or sync from ever calling |
1265 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1266 | * its underlying filesystem, in which case tmpfs should write out to | |
1267 | * swap only in response to memory pressure, and not for the writeback | |
1268 | * threads or sync. | |
d9fe526a | 1269 | */ |
48f170fb HD |
1270 | if (!wbc->for_reclaim) { |
1271 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1272 | goto redirty; | |
1273 | } | |
1635f6a7 HD |
1274 | |
1275 | /* | |
1276 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1277 | * value into swapfile.c, the only way we can correctly account for a | |
1278 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1279 | * |
1280 | * That's okay for a page already fallocated earlier, but if we have | |
1281 | * not yet completed the fallocation, then (a) we want to keep track | |
1282 | * of this page in case we have to undo it, and (b) it may not be a | |
1283 | * good idea to continue anyway, once we're pushing into swap. So | |
1284 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1285 | */ |
1286 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1287 | if (inode->i_private) { |
1288 | struct shmem_falloc *shmem_falloc; | |
1289 | spin_lock(&inode->i_lock); | |
1290 | shmem_falloc = inode->i_private; | |
1291 | if (shmem_falloc && | |
8e205f77 | 1292 | !shmem_falloc->waitq && |
1aac1400 HD |
1293 | index >= shmem_falloc->start && |
1294 | index < shmem_falloc->next) | |
1295 | shmem_falloc->nr_unswapped++; | |
1296 | else | |
1297 | shmem_falloc = NULL; | |
1298 | spin_unlock(&inode->i_lock); | |
1299 | if (shmem_falloc) | |
1300 | goto redirty; | |
1301 | } | |
1635f6a7 HD |
1302 | clear_highpage(page); |
1303 | flush_dcache_page(page); | |
1304 | SetPageUptodate(page); | |
1305 | } | |
1306 | ||
38d8b4e6 | 1307 | swap = get_swap_page(page); |
48f170fb HD |
1308 | if (!swap.val) |
1309 | goto redirty; | |
d9fe526a | 1310 | |
37e84351 VD |
1311 | if (mem_cgroup_try_charge_swap(page, swap)) |
1312 | goto free_swap; | |
1313 | ||
b1dea800 HD |
1314 | /* |
1315 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1316 | * if it's not already there. Do it now before the page is |
1317 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1318 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1319 | * we've incremented swapped, because shmem_unuse_inode() will |
1320 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1321 | */ |
48f170fb HD |
1322 | mutex_lock(&shmem_swaplist_mutex); |
1323 | if (list_empty(&info->swaplist)) | |
1324 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 1325 | |
48f170fb | 1326 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
4595ef88 | 1327 | spin_lock_irq(&info->lock); |
6922c0c7 | 1328 | shmem_recalc_inode(inode); |
267a4c76 | 1329 | info->swapped++; |
4595ef88 | 1330 | spin_unlock_irq(&info->lock); |
6922c0c7 | 1331 | |
267a4c76 HD |
1332 | swap_shmem_alloc(swap); |
1333 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1334 | ||
6922c0c7 | 1335 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1336 | BUG_ON(page_mapped(page)); |
9fab5619 | 1337 | swap_writepage(page, wbc); |
1da177e4 LT |
1338 | return 0; |
1339 | } | |
1340 | ||
6922c0c7 | 1341 | mutex_unlock(&shmem_swaplist_mutex); |
37e84351 | 1342 | free_swap: |
75f6d6d2 | 1343 | put_swap_page(page, swap); |
1da177e4 LT |
1344 | redirty: |
1345 | set_page_dirty(page); | |
d9fe526a HD |
1346 | if (wbc->for_reclaim) |
1347 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1348 | unlock_page(page); | |
1349 | return 0; | |
1da177e4 LT |
1350 | } |
1351 | ||
75edd345 | 1352 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1353 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1354 | { |
095f1fc4 | 1355 | char buffer[64]; |
680d794b | 1356 | |
71fe804b | 1357 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1358 | return; /* show nothing */ |
680d794b | 1359 | |
a7a88b23 | 1360 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1361 | |
1362 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1363 | } |
71fe804b LS |
1364 | |
1365 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1366 | { | |
1367 | struct mempolicy *mpol = NULL; | |
1368 | if (sbinfo->mpol) { | |
1369 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1370 | mpol = sbinfo->mpol; | |
1371 | mpol_get(mpol); | |
1372 | spin_unlock(&sbinfo->stat_lock); | |
1373 | } | |
1374 | return mpol; | |
1375 | } | |
75edd345 HD |
1376 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1377 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1378 | { | |
1379 | } | |
1380 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1381 | { | |
1382 | return NULL; | |
1383 | } | |
1384 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1385 | #ifndef CONFIG_NUMA | |
1386 | #define vm_policy vm_private_data | |
1387 | #endif | |
680d794b | 1388 | |
800d8c63 KS |
1389 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1390 | struct shmem_inode_info *info, pgoff_t index) | |
1391 | { | |
1392 | /* Create a pseudo vma that just contains the policy */ | |
1393 | vma->vm_start = 0; | |
1394 | /* Bias interleave by inode number to distribute better across nodes */ | |
1395 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
1396 | vma->vm_ops = NULL; | |
1397 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); | |
1398 | } | |
1399 | ||
1400 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1401 | { | |
1402 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1403 | mpol_cond_put(vma->vm_policy); | |
1404 | } | |
1405 | ||
41ffe5d5 HD |
1406 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1407 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1408 | { |
1da177e4 | 1409 | struct vm_area_struct pvma; |
18a2f371 | 1410 | struct page *page; |
52cd3b07 | 1411 | |
800d8c63 | 1412 | shmem_pseudo_vma_init(&pvma, info, index); |
18a2f371 | 1413 | page = swapin_readahead(swap, gfp, &pvma, 0); |
800d8c63 | 1414 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1415 | |
800d8c63 KS |
1416 | return page; |
1417 | } | |
1418 | ||
1419 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1420 | struct shmem_inode_info *info, pgoff_t index) | |
1421 | { | |
1422 | struct vm_area_struct pvma; | |
1423 | struct inode *inode = &info->vfs_inode; | |
1424 | struct address_space *mapping = inode->i_mapping; | |
4620a06e | 1425 | pgoff_t idx, hindex; |
800d8c63 KS |
1426 | void __rcu **results; |
1427 | struct page *page; | |
1428 | ||
e496cf3d | 1429 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1430 | return NULL; |
1431 | ||
4620a06e | 1432 | hindex = round_down(index, HPAGE_PMD_NR); |
800d8c63 KS |
1433 | rcu_read_lock(); |
1434 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx, | |
1435 | hindex, 1) && idx < hindex + HPAGE_PMD_NR) { | |
1436 | rcu_read_unlock(); | |
1437 | return NULL; | |
1438 | } | |
1439 | rcu_read_unlock(); | |
18a2f371 | 1440 | |
800d8c63 KS |
1441 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1442 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
1443 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); | |
1444 | shmem_pseudo_vma_destroy(&pvma); | |
1445 | if (page) | |
1446 | prep_transhuge_page(page); | |
18a2f371 | 1447 | return page; |
1da177e4 LT |
1448 | } |
1449 | ||
02098fea | 1450 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1451 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1452 | { |
1453 | struct vm_area_struct pvma; | |
18a2f371 | 1454 | struct page *page; |
1da177e4 | 1455 | |
800d8c63 KS |
1456 | shmem_pseudo_vma_init(&pvma, info, index); |
1457 | page = alloc_page_vma(gfp, &pvma, 0); | |
1458 | shmem_pseudo_vma_destroy(&pvma); | |
1459 | ||
1460 | return page; | |
1461 | } | |
1462 | ||
1463 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
1464 | struct shmem_inode_info *info, struct shmem_sb_info *sbinfo, | |
1465 | pgoff_t index, bool huge) | |
1466 | { | |
1467 | struct page *page; | |
1468 | int nr; | |
1469 | int err = -ENOSPC; | |
52cd3b07 | 1470 | |
e496cf3d | 1471 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1472 | huge = false; |
1473 | nr = huge ? HPAGE_PMD_NR : 1; | |
1474 | ||
1475 | if (shmem_acct_block(info->flags, nr)) | |
1476 | goto failed; | |
1477 | if (sbinfo->max_blocks) { | |
1478 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
1479 | sbinfo->max_blocks - nr) > 0) | |
1480 | goto unacct; | |
1481 | percpu_counter_add(&sbinfo->used_blocks, nr); | |
1482 | } | |
1483 | ||
1484 | if (huge) | |
1485 | page = shmem_alloc_hugepage(gfp, info, index); | |
1486 | else | |
1487 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1488 | if (page) { |
1489 | __SetPageLocked(page); | |
1490 | __SetPageSwapBacked(page); | |
800d8c63 | 1491 | return page; |
75edd345 | 1492 | } |
18a2f371 | 1493 | |
800d8c63 KS |
1494 | err = -ENOMEM; |
1495 | if (sbinfo->max_blocks) | |
1496 | percpu_counter_add(&sbinfo->used_blocks, -nr); | |
1497 | unacct: | |
1498 | shmem_unacct_blocks(info->flags, nr); | |
1499 | failed: | |
1500 | return ERR_PTR(err); | |
1da177e4 | 1501 | } |
71fe804b | 1502 | |
bde05d1c HD |
1503 | /* |
1504 | * When a page is moved from swapcache to shmem filecache (either by the | |
1505 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1506 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1507 | * ignorance of the mapping it belongs to. If that mapping has special | |
1508 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1509 | * we may need to copy to a suitable page before moving to filecache. | |
1510 | * | |
1511 | * In a future release, this may well be extended to respect cpuset and | |
1512 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1513 | * but for now it is a simple matter of zone. | |
1514 | */ | |
1515 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1516 | { | |
1517 | return page_zonenum(page) > gfp_zone(gfp); | |
1518 | } | |
1519 | ||
1520 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1521 | struct shmem_inode_info *info, pgoff_t index) | |
1522 | { | |
1523 | struct page *oldpage, *newpage; | |
1524 | struct address_space *swap_mapping; | |
1525 | pgoff_t swap_index; | |
1526 | int error; | |
1527 | ||
1528 | oldpage = *pagep; | |
1529 | swap_index = page_private(oldpage); | |
1530 | swap_mapping = page_mapping(oldpage); | |
1531 | ||
1532 | /* | |
1533 | * We have arrived here because our zones are constrained, so don't | |
1534 | * limit chance of success by further cpuset and node constraints. | |
1535 | */ | |
1536 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1537 | newpage = shmem_alloc_page(gfp, info, index); | |
1538 | if (!newpage) | |
1539 | return -ENOMEM; | |
bde05d1c | 1540 | |
09cbfeaf | 1541 | get_page(newpage); |
bde05d1c | 1542 | copy_highpage(newpage, oldpage); |
0142ef6c | 1543 | flush_dcache_page(newpage); |
bde05d1c | 1544 | |
9956edf3 HD |
1545 | __SetPageLocked(newpage); |
1546 | __SetPageSwapBacked(newpage); | |
bde05d1c | 1547 | SetPageUptodate(newpage); |
bde05d1c | 1548 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1549 | SetPageSwapCache(newpage); |
1550 | ||
1551 | /* | |
1552 | * Our caller will very soon move newpage out of swapcache, but it's | |
1553 | * a nice clean interface for us to replace oldpage by newpage there. | |
1554 | */ | |
1555 | spin_lock_irq(&swap_mapping->tree_lock); | |
1556 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1557 | newpage); | |
0142ef6c | 1558 | if (!error) { |
11fb9989 MG |
1559 | __inc_node_page_state(newpage, NR_FILE_PAGES); |
1560 | __dec_node_page_state(oldpage, NR_FILE_PAGES); | |
0142ef6c | 1561 | } |
bde05d1c | 1562 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1563 | |
0142ef6c HD |
1564 | if (unlikely(error)) { |
1565 | /* | |
1566 | * Is this possible? I think not, now that our callers check | |
1567 | * both PageSwapCache and page_private after getting page lock; | |
1568 | * but be defensive. Reverse old to newpage for clear and free. | |
1569 | */ | |
1570 | oldpage = newpage; | |
1571 | } else { | |
6a93ca8f | 1572 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1573 | lru_cache_add_anon(newpage); |
1574 | *pagep = newpage; | |
1575 | } | |
bde05d1c HD |
1576 | |
1577 | ClearPageSwapCache(oldpage); | |
1578 | set_page_private(oldpage, 0); | |
1579 | ||
1580 | unlock_page(oldpage); | |
09cbfeaf KS |
1581 | put_page(oldpage); |
1582 | put_page(oldpage); | |
0142ef6c | 1583 | return error; |
bde05d1c HD |
1584 | } |
1585 | ||
1da177e4 | 1586 | /* |
68da9f05 | 1587 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1588 | * |
1589 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1590 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1591 | * entry since a page cannot live in both the swap and page cache. |
1592 | * | |
1593 | * fault_mm and fault_type are only supplied by shmem_fault: | |
1594 | * otherwise they are NULL. | |
1da177e4 | 1595 | */ |
41ffe5d5 | 1596 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 1597 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
cfda0526 | 1598 | struct vm_area_struct *vma, struct vm_fault *vmf, int *fault_type) |
1da177e4 LT |
1599 | { |
1600 | struct address_space *mapping = inode->i_mapping; | |
23f919d4 | 1601 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 | 1602 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1603 | struct mm_struct *charge_mm; |
00501b53 | 1604 | struct mem_cgroup *memcg; |
27ab7006 | 1605 | struct page *page; |
1da177e4 | 1606 | swp_entry_t swap; |
657e3038 | 1607 | enum sgp_type sgp_huge = sgp; |
800d8c63 | 1608 | pgoff_t hindex = index; |
1da177e4 | 1609 | int error; |
54af6042 | 1610 | int once = 0; |
1635f6a7 | 1611 | int alloced = 0; |
1da177e4 | 1612 | |
09cbfeaf | 1613 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1614 | return -EFBIG; |
657e3038 KS |
1615 | if (sgp == SGP_NOHUGE || sgp == SGP_HUGE) |
1616 | sgp = SGP_CACHE; | |
1da177e4 | 1617 | repeat: |
54af6042 | 1618 | swap.val = 0; |
0cd6144a | 1619 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1620 | if (radix_tree_exceptional_entry(page)) { |
1621 | swap = radix_to_swp_entry(page); | |
1622 | page = NULL; | |
1623 | } | |
1624 | ||
75edd345 | 1625 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1626 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 1627 | error = -EINVAL; |
267a4c76 | 1628 | goto unlock; |
54af6042 HD |
1629 | } |
1630 | ||
66d2f4d2 HD |
1631 | if (page && sgp == SGP_WRITE) |
1632 | mark_page_accessed(page); | |
1633 | ||
1635f6a7 HD |
1634 | /* fallocated page? */ |
1635 | if (page && !PageUptodate(page)) { | |
1636 | if (sgp != SGP_READ) | |
1637 | goto clear; | |
1638 | unlock_page(page); | |
09cbfeaf | 1639 | put_page(page); |
1635f6a7 HD |
1640 | page = NULL; |
1641 | } | |
54af6042 | 1642 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1643 | *pagep = page; |
1644 | return 0; | |
27ab7006 HD |
1645 | } |
1646 | ||
1647 | /* | |
54af6042 HD |
1648 | * Fast cache lookup did not find it: |
1649 | * bring it back from swap or allocate. | |
27ab7006 | 1650 | */ |
54af6042 | 1651 | sbinfo = SHMEM_SB(inode->i_sb); |
cfda0526 | 1652 | charge_mm = vma ? vma->vm_mm : current->mm; |
1da177e4 | 1653 | |
1da177e4 LT |
1654 | if (swap.val) { |
1655 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1656 | page = lookup_swap_cache(swap); |
1657 | if (!page) { | |
9e18eb29 ALC |
1658 | /* Or update major stats only when swapin succeeds?? */ |
1659 | if (fault_type) { | |
68da9f05 | 1660 | *fault_type |= VM_FAULT_MAJOR; |
9e18eb29 | 1661 | count_vm_event(PGMAJFAULT); |
2262185c | 1662 | count_memcg_event_mm(charge_mm, PGMAJFAULT); |
9e18eb29 ALC |
1663 | } |
1664 | /* Here we actually start the io */ | |
41ffe5d5 | 1665 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1666 | if (!page) { |
54af6042 HD |
1667 | error = -ENOMEM; |
1668 | goto failed; | |
1da177e4 | 1669 | } |
1da177e4 LT |
1670 | } |
1671 | ||
1672 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1673 | lock_page(page); |
0142ef6c | 1674 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1675 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1676 | error = -EEXIST; /* try again */ |
d1899228 | 1677 | goto unlock; |
bde05d1c | 1678 | } |
27ab7006 | 1679 | if (!PageUptodate(page)) { |
1da177e4 | 1680 | error = -EIO; |
54af6042 | 1681 | goto failed; |
1da177e4 | 1682 | } |
54af6042 HD |
1683 | wait_on_page_writeback(page); |
1684 | ||
bde05d1c HD |
1685 | if (shmem_should_replace_page(page, gfp)) { |
1686 | error = shmem_replace_page(&page, gfp, info, index); | |
1687 | if (error) | |
1688 | goto failed; | |
1da177e4 | 1689 | } |
27ab7006 | 1690 | |
9e18eb29 | 1691 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
f627c2f5 | 1692 | false); |
d1899228 | 1693 | if (!error) { |
aa3b1895 | 1694 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1695 | swp_to_radix_entry(swap)); |
215c02bc HD |
1696 | /* |
1697 | * We already confirmed swap under page lock, and make | |
1698 | * no memory allocation here, so usually no possibility | |
1699 | * of error; but free_swap_and_cache() only trylocks a | |
1700 | * page, so it is just possible that the entry has been | |
1701 | * truncated or holepunched since swap was confirmed. | |
1702 | * shmem_undo_range() will have done some of the | |
1703 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1704 | * the rest. |
215c02bc HD |
1705 | * Reset swap.val? No, leave it so "failed" goes back to |
1706 | * "repeat": reading a hole and writing should succeed. | |
1707 | */ | |
00501b53 | 1708 | if (error) { |
f627c2f5 | 1709 | mem_cgroup_cancel_charge(page, memcg, false); |
215c02bc | 1710 | delete_from_swap_cache(page); |
00501b53 | 1711 | } |
d1899228 | 1712 | } |
54af6042 HD |
1713 | if (error) |
1714 | goto failed; | |
1715 | ||
f627c2f5 | 1716 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1717 | |
4595ef88 | 1718 | spin_lock_irq(&info->lock); |
285b2c4f | 1719 | info->swapped--; |
54af6042 | 1720 | shmem_recalc_inode(inode); |
4595ef88 | 1721 | spin_unlock_irq(&info->lock); |
54af6042 | 1722 | |
66d2f4d2 HD |
1723 | if (sgp == SGP_WRITE) |
1724 | mark_page_accessed(page); | |
1725 | ||
54af6042 | 1726 | delete_from_swap_cache(page); |
27ab7006 HD |
1727 | set_page_dirty(page); |
1728 | swap_free(swap); | |
1729 | ||
54af6042 | 1730 | } else { |
cfda0526 MR |
1731 | if (vma && userfaultfd_missing(vma)) { |
1732 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
1733 | return 0; | |
1734 | } | |
1735 | ||
800d8c63 KS |
1736 | /* shmem_symlink() */ |
1737 | if (mapping->a_ops != &shmem_aops) | |
1738 | goto alloc_nohuge; | |
657e3038 | 1739 | if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE) |
800d8c63 KS |
1740 | goto alloc_nohuge; |
1741 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1742 | goto alloc_huge; | |
1743 | switch (sbinfo->huge) { | |
1744 | loff_t i_size; | |
1745 | pgoff_t off; | |
1746 | case SHMEM_HUGE_NEVER: | |
1747 | goto alloc_nohuge; | |
1748 | case SHMEM_HUGE_WITHIN_SIZE: | |
1749 | off = round_up(index, HPAGE_PMD_NR); | |
1750 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1751 | if (i_size >= HPAGE_PMD_SIZE && | |
1752 | i_size >> PAGE_SHIFT >= off) | |
1753 | goto alloc_huge; | |
1754 | /* fallthrough */ | |
1755 | case SHMEM_HUGE_ADVISE: | |
657e3038 KS |
1756 | if (sgp_huge == SGP_HUGE) |
1757 | goto alloc_huge; | |
1758 | /* TODO: implement fadvise() hints */ | |
800d8c63 | 1759 | goto alloc_nohuge; |
54af6042 | 1760 | } |
1da177e4 | 1761 | |
800d8c63 KS |
1762 | alloc_huge: |
1763 | page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1764 | index, true); | |
1765 | if (IS_ERR(page)) { | |
1766 | alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1767 | index, false); | |
1da177e4 | 1768 | } |
800d8c63 | 1769 | if (IS_ERR(page)) { |
779750d2 | 1770 | int retry = 5; |
800d8c63 KS |
1771 | error = PTR_ERR(page); |
1772 | page = NULL; | |
779750d2 KS |
1773 | if (error != -ENOSPC) |
1774 | goto failed; | |
1775 | /* | |
1776 | * Try to reclaim some spece by splitting a huge page | |
1777 | * beyond i_size on the filesystem. | |
1778 | */ | |
1779 | while (retry--) { | |
1780 | int ret; | |
1781 | ret = shmem_unused_huge_shrink(sbinfo, NULL, 1); | |
1782 | if (ret == SHRINK_STOP) | |
1783 | break; | |
1784 | if (ret) | |
1785 | goto alloc_nohuge; | |
1786 | } | |
800d8c63 KS |
1787 | goto failed; |
1788 | } | |
1789 | ||
1790 | if (PageTransHuge(page)) | |
1791 | hindex = round_down(index, HPAGE_PMD_NR); | |
1792 | else | |
1793 | hindex = index; | |
1794 | ||
66d2f4d2 | 1795 | if (sgp == SGP_WRITE) |
eb39d618 | 1796 | __SetPageReferenced(page); |
66d2f4d2 | 1797 | |
9e18eb29 | 1798 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
800d8c63 | 1799 | PageTransHuge(page)); |
54af6042 | 1800 | if (error) |
800d8c63 KS |
1801 | goto unacct; |
1802 | error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK, | |
1803 | compound_order(page)); | |
b065b432 | 1804 | if (!error) { |
800d8c63 | 1805 | error = shmem_add_to_page_cache(page, mapping, hindex, |
fed400a1 | 1806 | NULL); |
b065b432 HD |
1807 | radix_tree_preload_end(); |
1808 | } | |
1809 | if (error) { | |
800d8c63 KS |
1810 | mem_cgroup_cancel_charge(page, memcg, |
1811 | PageTransHuge(page)); | |
1812 | goto unacct; | |
b065b432 | 1813 | } |
800d8c63 KS |
1814 | mem_cgroup_commit_charge(page, memcg, false, |
1815 | PageTransHuge(page)); | |
54af6042 HD |
1816 | lru_cache_add_anon(page); |
1817 | ||
4595ef88 | 1818 | spin_lock_irq(&info->lock); |
800d8c63 KS |
1819 | info->alloced += 1 << compound_order(page); |
1820 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); | |
54af6042 | 1821 | shmem_recalc_inode(inode); |
4595ef88 | 1822 | spin_unlock_irq(&info->lock); |
1635f6a7 | 1823 | alloced = true; |
54af6042 | 1824 | |
779750d2 KS |
1825 | if (PageTransHuge(page) && |
1826 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < | |
1827 | hindex + HPAGE_PMD_NR - 1) { | |
1828 | /* | |
1829 | * Part of the huge page is beyond i_size: subject | |
1830 | * to shrink under memory pressure. | |
1831 | */ | |
1832 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1833 | /* |
1834 | * _careful to defend against unlocked access to | |
1835 | * ->shrink_list in shmem_unused_huge_shrink() | |
1836 | */ | |
1837 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1838 | list_add_tail(&info->shrinklist, |
1839 | &sbinfo->shrinklist); | |
1840 | sbinfo->shrinklist_len++; | |
1841 | } | |
1842 | spin_unlock(&sbinfo->shrinklist_lock); | |
1843 | } | |
1844 | ||
ec9516fb | 1845 | /* |
1635f6a7 HD |
1846 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1847 | */ | |
1848 | if (sgp == SGP_FALLOC) | |
1849 | sgp = SGP_WRITE; | |
1850 | clear: | |
1851 | /* | |
1852 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1853 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1854 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb | 1855 | */ |
800d8c63 KS |
1856 | if (sgp != SGP_WRITE && !PageUptodate(page)) { |
1857 | struct page *head = compound_head(page); | |
1858 | int i; | |
1859 | ||
1860 | for (i = 0; i < (1 << compound_order(head)); i++) { | |
1861 | clear_highpage(head + i); | |
1862 | flush_dcache_page(head + i); | |
1863 | } | |
1864 | SetPageUptodate(head); | |
ec9516fb | 1865 | } |
1da177e4 | 1866 | } |
bde05d1c | 1867 | |
54af6042 | 1868 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1869 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1870 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1871 | if (alloced) { |
1872 | ClearPageDirty(page); | |
1873 | delete_from_page_cache(page); | |
4595ef88 | 1874 | spin_lock_irq(&info->lock); |
267a4c76 | 1875 | shmem_recalc_inode(inode); |
4595ef88 | 1876 | spin_unlock_irq(&info->lock); |
267a4c76 | 1877 | } |
54af6042 | 1878 | error = -EINVAL; |
267a4c76 | 1879 | goto unlock; |
e83c32e8 | 1880 | } |
800d8c63 | 1881 | *pagep = page + index - hindex; |
54af6042 | 1882 | return 0; |
1da177e4 | 1883 | |
59a16ead | 1884 | /* |
54af6042 | 1885 | * Error recovery. |
59a16ead | 1886 | */ |
54af6042 | 1887 | unacct: |
800d8c63 KS |
1888 | if (sbinfo->max_blocks) |
1889 | percpu_counter_sub(&sbinfo->used_blocks, | |
1890 | 1 << compound_order(page)); | |
1891 | shmem_unacct_blocks(info->flags, 1 << compound_order(page)); | |
1892 | ||
1893 | if (PageTransHuge(page)) { | |
1894 | unlock_page(page); | |
1895 | put_page(page); | |
1896 | goto alloc_nohuge; | |
1897 | } | |
54af6042 | 1898 | failed: |
267a4c76 | 1899 | if (swap.val && !shmem_confirm_swap(mapping, index, swap)) |
d1899228 HD |
1900 | error = -EEXIST; |
1901 | unlock: | |
27ab7006 | 1902 | if (page) { |
54af6042 | 1903 | unlock_page(page); |
09cbfeaf | 1904 | put_page(page); |
54af6042 HD |
1905 | } |
1906 | if (error == -ENOSPC && !once++) { | |
4595ef88 | 1907 | spin_lock_irq(&info->lock); |
54af6042 | 1908 | shmem_recalc_inode(inode); |
4595ef88 | 1909 | spin_unlock_irq(&info->lock); |
27ab7006 | 1910 | goto repeat; |
ff36b801 | 1911 | } |
d1899228 | 1912 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1913 | goto repeat; |
1914 | return error; | |
1da177e4 LT |
1915 | } |
1916 | ||
10d20bd2 LT |
1917 | /* |
1918 | * This is like autoremove_wake_function, but it removes the wait queue | |
1919 | * entry unconditionally - even if something else had already woken the | |
1920 | * target. | |
1921 | */ | |
ac6424b9 | 1922 | static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
10d20bd2 LT |
1923 | { |
1924 | int ret = default_wake_function(wait, mode, sync, key); | |
2055da97 | 1925 | list_del_init(&wait->entry); |
10d20bd2 LT |
1926 | return ret; |
1927 | } | |
1928 | ||
11bac800 | 1929 | static int shmem_fault(struct vm_fault *vmf) |
1da177e4 | 1930 | { |
11bac800 | 1931 | struct vm_area_struct *vma = vmf->vma; |
496ad9aa | 1932 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 1933 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
657e3038 | 1934 | enum sgp_type sgp; |
1da177e4 | 1935 | int error; |
68da9f05 | 1936 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1937 | |
f00cdc6d HD |
1938 | /* |
1939 | * Trinity finds that probing a hole which tmpfs is punching can | |
1940 | * prevent the hole-punch from ever completing: which in turn | |
1941 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1942 | * faulting pages into the hole while it's being punched. Although |
1943 | * shmem_undo_range() does remove the additions, it may be unable to | |
1944 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1945 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1946 | * | |
1947 | * It does not matter if we sometimes reach this check just before the | |
1948 | * hole-punch begins, so that one fault then races with the punch: | |
1949 | * we just need to make racing faults a rare case. | |
1950 | * | |
1951 | * The implementation below would be much simpler if we just used a | |
1952 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1953 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1954 | */ |
1955 | if (unlikely(inode->i_private)) { | |
1956 | struct shmem_falloc *shmem_falloc; | |
1957 | ||
1958 | spin_lock(&inode->i_lock); | |
1959 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1960 | if (shmem_falloc && |
1961 | shmem_falloc->waitq && | |
1962 | vmf->pgoff >= shmem_falloc->start && | |
1963 | vmf->pgoff < shmem_falloc->next) { | |
1964 | wait_queue_head_t *shmem_falloc_waitq; | |
10d20bd2 | 1965 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); |
8e205f77 HD |
1966 | |
1967 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1968 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1969 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1970 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1971 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1972 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1973 | } |
8e205f77 HD |
1974 | |
1975 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1976 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1977 | TASK_UNINTERRUPTIBLE); | |
1978 | spin_unlock(&inode->i_lock); | |
1979 | schedule(); | |
1980 | ||
1981 | /* | |
1982 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1983 | * stack of the hole-punching task: shmem_falloc_waitq | |
1984 | * is usually invalid by the time we reach here, but | |
1985 | * finish_wait() does not dereference it in that case; | |
1986 | * though i_lock needed lest racing with wake_up_all(). | |
1987 | */ | |
1988 | spin_lock(&inode->i_lock); | |
1989 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1990 | spin_unlock(&inode->i_lock); | |
1991 | return ret; | |
f00cdc6d | 1992 | } |
8e205f77 | 1993 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1994 | } |
1995 | ||
657e3038 | 1996 | sgp = SGP_CACHE; |
18600332 MH |
1997 | |
1998 | if ((vma->vm_flags & VM_NOHUGEPAGE) || | |
1999 | test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) | |
657e3038 | 2000 | sgp = SGP_NOHUGE; |
18600332 MH |
2001 | else if (vma->vm_flags & VM_HUGEPAGE) |
2002 | sgp = SGP_HUGE; | |
657e3038 KS |
2003 | |
2004 | error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp, | |
cfda0526 | 2005 | gfp, vma, vmf, &ret); |
d0217ac0 NP |
2006 | if (error) |
2007 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 2008 | return ret; |
1da177e4 LT |
2009 | } |
2010 | ||
c01d5b30 HD |
2011 | unsigned long shmem_get_unmapped_area(struct file *file, |
2012 | unsigned long uaddr, unsigned long len, | |
2013 | unsigned long pgoff, unsigned long flags) | |
2014 | { | |
2015 | unsigned long (*get_area)(struct file *, | |
2016 | unsigned long, unsigned long, unsigned long, unsigned long); | |
2017 | unsigned long addr; | |
2018 | unsigned long offset; | |
2019 | unsigned long inflated_len; | |
2020 | unsigned long inflated_addr; | |
2021 | unsigned long inflated_offset; | |
2022 | ||
2023 | if (len > TASK_SIZE) | |
2024 | return -ENOMEM; | |
2025 | ||
2026 | get_area = current->mm->get_unmapped_area; | |
2027 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2028 | ||
e496cf3d | 2029 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
c01d5b30 HD |
2030 | return addr; |
2031 | if (IS_ERR_VALUE(addr)) | |
2032 | return addr; | |
2033 | if (addr & ~PAGE_MASK) | |
2034 | return addr; | |
2035 | if (addr > TASK_SIZE - len) | |
2036 | return addr; | |
2037 | ||
2038 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2039 | return addr; | |
2040 | if (len < HPAGE_PMD_SIZE) | |
2041 | return addr; | |
2042 | if (flags & MAP_FIXED) | |
2043 | return addr; | |
2044 | /* | |
2045 | * Our priority is to support MAP_SHARED mapped hugely; | |
2046 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
2047 | * But if caller specified an address hint, respect that as before. | |
2048 | */ | |
2049 | if (uaddr) | |
2050 | return addr; | |
2051 | ||
2052 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2053 | struct super_block *sb; | |
2054 | ||
2055 | if (file) { | |
2056 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2057 | sb = file_inode(file)->i_sb; | |
2058 | } else { | |
2059 | /* | |
2060 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2061 | * for "/dev/zero", to create a shared anonymous object. | |
2062 | */ | |
2063 | if (IS_ERR(shm_mnt)) | |
2064 | return addr; | |
2065 | sb = shm_mnt->mnt_sb; | |
2066 | } | |
3089bf61 | 2067 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2068 | return addr; |
2069 | } | |
2070 | ||
2071 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2072 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2073 | return addr; | |
2074 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2075 | return addr; | |
2076 | ||
2077 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2078 | if (inflated_len > TASK_SIZE) | |
2079 | return addr; | |
2080 | if (inflated_len < len) | |
2081 | return addr; | |
2082 | ||
2083 | inflated_addr = get_area(NULL, 0, inflated_len, 0, flags); | |
2084 | if (IS_ERR_VALUE(inflated_addr)) | |
2085 | return addr; | |
2086 | if (inflated_addr & ~PAGE_MASK) | |
2087 | return addr; | |
2088 | ||
2089 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2090 | inflated_addr += offset - inflated_offset; | |
2091 | if (inflated_offset > offset) | |
2092 | inflated_addr += HPAGE_PMD_SIZE; | |
2093 | ||
2094 | if (inflated_addr > TASK_SIZE - len) | |
2095 | return addr; | |
2096 | return inflated_addr; | |
2097 | } | |
2098 | ||
1da177e4 | 2099 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2100 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2101 | { |
496ad9aa | 2102 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2103 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2104 | } |
2105 | ||
d8dc74f2 AB |
2106 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
2107 | unsigned long addr) | |
1da177e4 | 2108 | { |
496ad9aa | 2109 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2110 | pgoff_t index; |
1da177e4 | 2111 | |
41ffe5d5 HD |
2112 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2113 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
2114 | } |
2115 | #endif | |
2116 | ||
2117 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
2118 | { | |
496ad9aa | 2119 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2120 | struct shmem_inode_info *info = SHMEM_I(inode); |
2121 | int retval = -ENOMEM; | |
2122 | ||
4595ef88 | 2123 | spin_lock_irq(&info->lock); |
1da177e4 LT |
2124 | if (lock && !(info->flags & VM_LOCKED)) { |
2125 | if (!user_shm_lock(inode->i_size, user)) | |
2126 | goto out_nomem; | |
2127 | info->flags |= VM_LOCKED; | |
89e004ea | 2128 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
2129 | } |
2130 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
2131 | user_shm_unlock(inode->i_size, user); | |
2132 | info->flags &= ~VM_LOCKED; | |
89e004ea | 2133 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2134 | } |
2135 | retval = 0; | |
89e004ea | 2136 | |
1da177e4 | 2137 | out_nomem: |
4595ef88 | 2138 | spin_unlock_irq(&info->lock); |
1da177e4 LT |
2139 | return retval; |
2140 | } | |
2141 | ||
9b83a6a8 | 2142 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
2143 | { |
2144 | file_accessed(file); | |
2145 | vma->vm_ops = &shmem_vm_ops; | |
e496cf3d | 2146 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
2147 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
2148 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
2149 | khugepaged_enter(vma, vma->vm_flags); | |
2150 | } | |
1da177e4 LT |
2151 | return 0; |
2152 | } | |
2153 | ||
454abafe | 2154 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 2155 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
2156 | { |
2157 | struct inode *inode; | |
2158 | struct shmem_inode_info *info; | |
2159 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
b6450630 | 2160 | int ino; |
1da177e4 | 2161 | |
5b04c689 PE |
2162 | if (shmem_reserve_inode(sb)) |
2163 | return NULL; | |
1da177e4 LT |
2164 | |
2165 | inode = new_inode(sb); | |
2166 | if (inode) { | |
454abafe | 2167 | inode_init_owner(inode, dir, mode); |
1da177e4 | 2168 | inode->i_blocks = 0; |
078cd827 | 2169 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
91828a40 | 2170 | inode->i_generation = get_seconds(); |
1da177e4 LT |
2171 | info = SHMEM_I(inode); |
2172 | memset(info, 0, (char *)inode - (char *)info); | |
2173 | spin_lock_init(&info->lock); | |
40e041a2 | 2174 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 2175 | info->flags = flags & VM_NORESERVE; |
779750d2 | 2176 | INIT_LIST_HEAD(&info->shrinklist); |
1da177e4 | 2177 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 2178 | simple_xattrs_init(&info->xattrs); |
72c04902 | 2179 | cache_no_acl(inode); |
1da177e4 LT |
2180 | |
2181 | switch (mode & S_IFMT) { | |
2182 | default: | |
39f0247d | 2183 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
2184 | init_special_inode(inode, mode, dev); |
2185 | break; | |
2186 | case S_IFREG: | |
14fcc23f | 2187 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
2188 | inode->i_op = &shmem_inode_operations; |
2189 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
2190 | mpol_shared_policy_init(&info->policy, |
2191 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
2192 | break; |
2193 | case S_IFDIR: | |
d8c76e6f | 2194 | inc_nlink(inode); |
1da177e4 LT |
2195 | /* Some things misbehave if size == 0 on a directory */ |
2196 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2197 | inode->i_op = &shmem_dir_inode_operations; | |
2198 | inode->i_fop = &simple_dir_operations; | |
2199 | break; | |
2200 | case S_IFLNK: | |
2201 | /* | |
2202 | * Must not load anything in the rbtree, | |
2203 | * mpol_free_shared_policy will not be called. | |
2204 | */ | |
71fe804b | 2205 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
2206 | break; |
2207 | } | |
b6450630 SF |
2208 | |
2209 | if (!sbinfo->idr_nouse) { | |
2210 | /* inum 0 and 1 are unused */ | |
2211 | mutex_lock(&sbinfo->idr_lock); | |
2212 | ino = idr_alloc(&sbinfo->idr, inode, 2, INT_MAX, | |
2213 | GFP_NOFS); | |
2214 | if (ino > 0) { | |
2215 | inode->i_ino = ino; | |
2216 | mutex_unlock(&sbinfo->idr_lock); | |
2217 | __insert_inode_hash(inode, inode->i_ino); | |
2218 | } else { | |
2219 | inode->i_ino = 0; | |
2220 | mutex_unlock(&sbinfo->idr_lock); | |
2221 | iput(inode); | |
2222 | /* shmem_free_inode() will be called */ | |
2223 | inode = NULL; | |
2224 | } | |
2225 | } else | |
2226 | inode->i_ino = get_next_ino(); | |
5b04c689 PE |
2227 | } else |
2228 | shmem_free_inode(sb); | |
1da177e4 LT |
2229 | return inode; |
2230 | } | |
2231 | ||
0cd6144a JW |
2232 | bool shmem_mapping(struct address_space *mapping) |
2233 | { | |
f8005451 | 2234 | return mapping->a_ops == &shmem_aops; |
0cd6144a JW |
2235 | } |
2236 | ||
4c27fe4c MR |
2237 | int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm, |
2238 | pmd_t *dst_pmd, | |
2239 | struct vm_area_struct *dst_vma, | |
2240 | unsigned long dst_addr, | |
2241 | unsigned long src_addr, | |
2242 | struct page **pagep) | |
2243 | { | |
2244 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2245 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2246 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
2247 | struct address_space *mapping = inode->i_mapping; | |
2248 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2249 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
2250 | struct mem_cgroup *memcg; | |
2251 | spinlock_t *ptl; | |
2252 | void *page_kaddr; | |
2253 | struct page *page; | |
2254 | pte_t _dst_pte, *dst_pte; | |
2255 | int ret; | |
2256 | ||
cb658a45 AA |
2257 | ret = -ENOMEM; |
2258 | if (shmem_acct_block(info->flags, 1)) | |
2259 | goto out; | |
2260 | if (sbinfo->max_blocks) { | |
2261 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
2262 | sbinfo->max_blocks) >= 0) | |
2263 | goto out_unacct_blocks; | |
2264 | percpu_counter_inc(&sbinfo->used_blocks); | |
2265 | } | |
4c27fe4c | 2266 | |
cb658a45 | 2267 | if (!*pagep) { |
4c27fe4c MR |
2268 | page = shmem_alloc_page(gfp, info, pgoff); |
2269 | if (!page) | |
2270 | goto out_dec_used_blocks; | |
2271 | ||
2272 | page_kaddr = kmap_atomic(page); | |
2273 | ret = copy_from_user(page_kaddr, (const void __user *)src_addr, | |
2274 | PAGE_SIZE); | |
2275 | kunmap_atomic(page_kaddr); | |
2276 | ||
2277 | /* fallback to copy_from_user outside mmap_sem */ | |
2278 | if (unlikely(ret)) { | |
2279 | *pagep = page; | |
cb658a45 AA |
2280 | if (sbinfo->max_blocks) |
2281 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
2282 | shmem_unacct_blocks(info->flags, 1); | |
4c27fe4c MR |
2283 | /* don't free the page */ |
2284 | return -EFAULT; | |
2285 | } | |
2286 | } else { | |
2287 | page = *pagep; | |
2288 | *pagep = NULL; | |
2289 | } | |
2290 | ||
9cc90c66 AA |
2291 | VM_BUG_ON(PageLocked(page) || PageSwapBacked(page)); |
2292 | __SetPageLocked(page); | |
2293 | __SetPageSwapBacked(page); | |
a425d358 | 2294 | __SetPageUptodate(page); |
9cc90c66 | 2295 | |
4c27fe4c MR |
2296 | ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false); |
2297 | if (ret) | |
2298 | goto out_release; | |
2299 | ||
2300 | ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK); | |
2301 | if (!ret) { | |
2302 | ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL); | |
2303 | radix_tree_preload_end(); | |
2304 | } | |
2305 | if (ret) | |
2306 | goto out_release_uncharge; | |
2307 | ||
2308 | mem_cgroup_commit_charge(page, memcg, false, false); | |
2309 | ||
2310 | _dst_pte = mk_pte(page, dst_vma->vm_page_prot); | |
2311 | if (dst_vma->vm_flags & VM_WRITE) | |
2312 | _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte)); | |
2313 | ||
2314 | ret = -EEXIST; | |
2315 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); | |
2316 | if (!pte_none(*dst_pte)) | |
2317 | goto out_release_uncharge_unlock; | |
2318 | ||
4c27fe4c MR |
2319 | lru_cache_add_anon(page); |
2320 | ||
2321 | spin_lock(&info->lock); | |
2322 | info->alloced++; | |
2323 | inode->i_blocks += BLOCKS_PER_PAGE; | |
2324 | shmem_recalc_inode(inode); | |
2325 | spin_unlock(&info->lock); | |
2326 | ||
2327 | inc_mm_counter(dst_mm, mm_counter_file(page)); | |
2328 | page_add_file_rmap(page, false); | |
2329 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
2330 | ||
2331 | /* No need to invalidate - it was non-present before */ | |
2332 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
2333 | unlock_page(page); | |
2334 | pte_unmap_unlock(dst_pte, ptl); | |
2335 | ret = 0; | |
2336 | out: | |
2337 | return ret; | |
2338 | out_release_uncharge_unlock: | |
2339 | pte_unmap_unlock(dst_pte, ptl); | |
2340 | out_release_uncharge: | |
2341 | mem_cgroup_cancel_charge(page, memcg, false); | |
2342 | out_release: | |
9cc90c66 | 2343 | unlock_page(page); |
4c27fe4c MR |
2344 | put_page(page); |
2345 | out_dec_used_blocks: | |
2346 | if (sbinfo->max_blocks) | |
2347 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
2348 | out_unacct_blocks: | |
2349 | shmem_unacct_blocks(info->flags, 1); | |
2350 | goto out; | |
2351 | } | |
2352 | ||
1da177e4 | 2353 | #ifdef CONFIG_TMPFS |
92e1d5be | 2354 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2355 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2356 | |
6d9d88d0 JS |
2357 | #ifdef CONFIG_TMPFS_XATTR |
2358 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2359 | #else | |
2360 | #define shmem_initxattrs NULL | |
2361 | #endif | |
2362 | ||
1da177e4 | 2363 | static int |
800d15a5 NP |
2364 | shmem_write_begin(struct file *file, struct address_space *mapping, |
2365 | loff_t pos, unsigned len, unsigned flags, | |
2366 | struct page **pagep, void **fsdata) | |
1da177e4 | 2367 | { |
800d15a5 | 2368 | struct inode *inode = mapping->host; |
40e041a2 | 2369 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2370 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
2371 | |
2372 | /* i_mutex is held by caller */ | |
3f472cc9 | 2373 | if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) { |
40e041a2 DH |
2374 | if (info->seals & F_SEAL_WRITE) |
2375 | return -EPERM; | |
2376 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2377 | return -EPERM; | |
2378 | } | |
2379 | ||
9e18eb29 | 2380 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
2381 | } |
2382 | ||
2383 | static int | |
2384 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2385 | loff_t pos, unsigned len, unsigned copied, | |
2386 | struct page *page, void *fsdata) | |
2387 | { | |
2388 | struct inode *inode = mapping->host; | |
2389 | ||
d3602444 HD |
2390 | if (pos + copied > inode->i_size) |
2391 | i_size_write(inode, pos + copied); | |
2392 | ||
ec9516fb | 2393 | if (!PageUptodate(page)) { |
800d8c63 KS |
2394 | struct page *head = compound_head(page); |
2395 | if (PageTransCompound(page)) { | |
2396 | int i; | |
2397 | ||
2398 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2399 | if (head + i == page) | |
2400 | continue; | |
2401 | clear_highpage(head + i); | |
2402 | flush_dcache_page(head + i); | |
2403 | } | |
2404 | } | |
09cbfeaf KS |
2405 | if (copied < PAGE_SIZE) { |
2406 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2407 | zero_user_segments(page, 0, from, |
09cbfeaf | 2408 | from + copied, PAGE_SIZE); |
ec9516fb | 2409 | } |
800d8c63 | 2410 | SetPageUptodate(head); |
ec9516fb | 2411 | } |
800d15a5 | 2412 | set_page_dirty(page); |
6746aff7 | 2413 | unlock_page(page); |
09cbfeaf | 2414 | put_page(page); |
800d15a5 | 2415 | |
800d15a5 | 2416 | return copied; |
1da177e4 LT |
2417 | } |
2418 | ||
2ba5bbed | 2419 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2420 | { |
6e58e79d AV |
2421 | struct file *file = iocb->ki_filp; |
2422 | struct inode *inode = file_inode(file); | |
1da177e4 | 2423 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2424 | pgoff_t index; |
2425 | unsigned long offset; | |
a0ee5ec5 | 2426 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2427 | int error = 0; |
cb66a7a1 | 2428 | ssize_t retval = 0; |
6e58e79d | 2429 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2430 | |
2431 | /* | |
2432 | * Might this read be for a stacking filesystem? Then when reading | |
2433 | * holes of a sparse file, we actually need to allocate those pages, | |
2434 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2435 | */ | |
777eda2c | 2436 | if (!iter_is_iovec(to)) |
75edd345 | 2437 | sgp = SGP_CACHE; |
1da177e4 | 2438 | |
09cbfeaf KS |
2439 | index = *ppos >> PAGE_SHIFT; |
2440 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2441 | |
2442 | for (;;) { | |
2443 | struct page *page = NULL; | |
41ffe5d5 HD |
2444 | pgoff_t end_index; |
2445 | unsigned long nr, ret; | |
1da177e4 LT |
2446 | loff_t i_size = i_size_read(inode); |
2447 | ||
09cbfeaf | 2448 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2449 | if (index > end_index) |
2450 | break; | |
2451 | if (index == end_index) { | |
09cbfeaf | 2452 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2453 | if (nr <= offset) |
2454 | break; | |
2455 | } | |
2456 | ||
9e18eb29 | 2457 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2458 | if (error) { |
2459 | if (error == -EINVAL) | |
2460 | error = 0; | |
1da177e4 LT |
2461 | break; |
2462 | } | |
75edd345 HD |
2463 | if (page) { |
2464 | if (sgp == SGP_CACHE) | |
2465 | set_page_dirty(page); | |
d3602444 | 2466 | unlock_page(page); |
75edd345 | 2467 | } |
1da177e4 LT |
2468 | |
2469 | /* | |
2470 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2471 | * are called without i_mutex protection against truncate |
1da177e4 | 2472 | */ |
09cbfeaf | 2473 | nr = PAGE_SIZE; |
1da177e4 | 2474 | i_size = i_size_read(inode); |
09cbfeaf | 2475 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2476 | if (index == end_index) { |
09cbfeaf | 2477 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2478 | if (nr <= offset) { |
2479 | if (page) | |
09cbfeaf | 2480 | put_page(page); |
1da177e4 LT |
2481 | break; |
2482 | } | |
2483 | } | |
2484 | nr -= offset; | |
2485 | ||
2486 | if (page) { | |
2487 | /* | |
2488 | * If users can be writing to this page using arbitrary | |
2489 | * virtual addresses, take care about potential aliasing | |
2490 | * before reading the page on the kernel side. | |
2491 | */ | |
2492 | if (mapping_writably_mapped(mapping)) | |
2493 | flush_dcache_page(page); | |
2494 | /* | |
2495 | * Mark the page accessed if we read the beginning. | |
2496 | */ | |
2497 | if (!offset) | |
2498 | mark_page_accessed(page); | |
b5810039 | 2499 | } else { |
1da177e4 | 2500 | page = ZERO_PAGE(0); |
09cbfeaf | 2501 | get_page(page); |
b5810039 | 2502 | } |
1da177e4 LT |
2503 | |
2504 | /* | |
2505 | * Ok, we have the page, and it's up-to-date, so | |
2506 | * now we can copy it to user space... | |
1da177e4 | 2507 | */ |
2ba5bbed | 2508 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2509 | retval += ret; |
1da177e4 | 2510 | offset += ret; |
09cbfeaf KS |
2511 | index += offset >> PAGE_SHIFT; |
2512 | offset &= ~PAGE_MASK; | |
1da177e4 | 2513 | |
09cbfeaf | 2514 | put_page(page); |
2ba5bbed | 2515 | if (!iov_iter_count(to)) |
1da177e4 | 2516 | break; |
6e58e79d AV |
2517 | if (ret < nr) { |
2518 | error = -EFAULT; | |
2519 | break; | |
2520 | } | |
1da177e4 LT |
2521 | cond_resched(); |
2522 | } | |
2523 | ||
09cbfeaf | 2524 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2525 | file_accessed(file); |
2526 | return retval ? retval : error; | |
1da177e4 LT |
2527 | } |
2528 | ||
220f2ac9 HD |
2529 | /* |
2530 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
2531 | */ | |
2532 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2533 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2534 | { |
2535 | struct page *page; | |
2536 | struct pagevec pvec; | |
2537 | pgoff_t indices[PAGEVEC_SIZE]; | |
2538 | bool done = false; | |
2539 | int i; | |
2540 | ||
2541 | pagevec_init(&pvec, 0); | |
2542 | pvec.nr = 1; /* start small: we may be there already */ | |
2543 | while (!done) { | |
0cd6144a | 2544 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2545 | pvec.nr, pvec.pages, indices); |
2546 | if (!pvec.nr) { | |
965c8e59 | 2547 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2548 | index = end; |
2549 | break; | |
2550 | } | |
2551 | for (i = 0; i < pvec.nr; i++, index++) { | |
2552 | if (index < indices[i]) { | |
965c8e59 | 2553 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2554 | done = true; |
2555 | break; | |
2556 | } | |
2557 | index = indices[i]; | |
2558 | } | |
2559 | page = pvec.pages[i]; | |
2560 | if (page && !radix_tree_exceptional_entry(page)) { | |
2561 | if (!PageUptodate(page)) | |
2562 | page = NULL; | |
2563 | } | |
2564 | if (index >= end || | |
965c8e59 AM |
2565 | (page && whence == SEEK_DATA) || |
2566 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2567 | done = true; |
2568 | break; | |
2569 | } | |
2570 | } | |
0cd6144a | 2571 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2572 | pagevec_release(&pvec); |
2573 | pvec.nr = PAGEVEC_SIZE; | |
2574 | cond_resched(); | |
2575 | } | |
2576 | return index; | |
2577 | } | |
2578 | ||
965c8e59 | 2579 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2580 | { |
2581 | struct address_space *mapping = file->f_mapping; | |
2582 | struct inode *inode = mapping->host; | |
2583 | pgoff_t start, end; | |
2584 | loff_t new_offset; | |
2585 | ||
965c8e59 AM |
2586 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2587 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2588 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2589 | inode_lock(inode); |
220f2ac9 HD |
2590 | /* We're holding i_mutex so we can access i_size directly */ |
2591 | ||
2592 | if (offset < 0) | |
2593 | offset = -EINVAL; | |
2594 | else if (offset >= inode->i_size) | |
2595 | offset = -ENXIO; | |
2596 | else { | |
09cbfeaf KS |
2597 | start = offset >> PAGE_SHIFT; |
2598 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2599 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2600 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2601 | if (new_offset > offset) { |
2602 | if (new_offset < inode->i_size) | |
2603 | offset = new_offset; | |
965c8e59 | 2604 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2605 | offset = -ENXIO; |
2606 | else | |
2607 | offset = inode->i_size; | |
2608 | } | |
2609 | } | |
2610 | ||
387aae6f HD |
2611 | if (offset >= 0) |
2612 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2613 | inode_unlock(inode); |
220f2ac9 HD |
2614 | return offset; |
2615 | } | |
2616 | ||
05f65b5c DH |
2617 | /* |
2618 | * We need a tag: a new tag would expand every radix_tree_node by 8 bytes, | |
2619 | * so reuse a tag which we firmly believe is never set or cleared on shmem. | |
2620 | */ | |
2621 | #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE | |
2622 | #define LAST_SCAN 4 /* about 150ms max */ | |
2623 | ||
2624 | static void shmem_tag_pins(struct address_space *mapping) | |
2625 | { | |
2626 | struct radix_tree_iter iter; | |
2627 | void **slot; | |
2628 | pgoff_t start; | |
2629 | struct page *page; | |
2630 | ||
2631 | lru_add_drain(); | |
2632 | start = 0; | |
2633 | rcu_read_lock(); | |
2634 | ||
05f65b5c DH |
2635 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
2636 | page = radix_tree_deref_slot(slot); | |
2637 | if (!page || radix_tree_exception(page)) { | |
2cf938aa MW |
2638 | if (radix_tree_deref_retry(page)) { |
2639 | slot = radix_tree_iter_retry(&iter); | |
2640 | continue; | |
2641 | } | |
05f65b5c DH |
2642 | } else if (page_count(page) - page_mapcount(page) > 1) { |
2643 | spin_lock_irq(&mapping->tree_lock); | |
2644 | radix_tree_tag_set(&mapping->page_tree, iter.index, | |
2645 | SHMEM_TAG_PINNED); | |
2646 | spin_unlock_irq(&mapping->tree_lock); | |
2647 | } | |
2648 | ||
2649 | if (need_resched()) { | |
148deab2 | 2650 | slot = radix_tree_iter_resume(slot, &iter); |
05f65b5c | 2651 | cond_resched_rcu(); |
05f65b5c DH |
2652 | } |
2653 | } | |
2654 | rcu_read_unlock(); | |
2655 | } | |
2656 | ||
2657 | /* | |
2658 | * Setting SEAL_WRITE requires us to verify there's no pending writer. However, | |
2659 | * via get_user_pages(), drivers might have some pending I/O without any active | |
2660 | * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages | |
2661 | * and see whether it has an elevated ref-count. If so, we tag them and wait for | |
2662 | * them to be dropped. | |
2663 | * The caller must guarantee that no new user will acquire writable references | |
2664 | * to those pages to avoid races. | |
2665 | */ | |
40e041a2 DH |
2666 | static int shmem_wait_for_pins(struct address_space *mapping) |
2667 | { | |
05f65b5c DH |
2668 | struct radix_tree_iter iter; |
2669 | void **slot; | |
2670 | pgoff_t start; | |
2671 | struct page *page; | |
2672 | int error, scan; | |
2673 | ||
2674 | shmem_tag_pins(mapping); | |
2675 | ||
2676 | error = 0; | |
2677 | for (scan = 0; scan <= LAST_SCAN; scan++) { | |
2678 | if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED)) | |
2679 | break; | |
2680 | ||
2681 | if (!scan) | |
2682 | lru_add_drain_all(); | |
2683 | else if (schedule_timeout_killable((HZ << scan) / 200)) | |
2684 | scan = LAST_SCAN; | |
2685 | ||
2686 | start = 0; | |
2687 | rcu_read_lock(); | |
05f65b5c DH |
2688 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, |
2689 | start, SHMEM_TAG_PINNED) { | |
2690 | ||
2691 | page = radix_tree_deref_slot(slot); | |
2692 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2693 | if (radix_tree_deref_retry(page)) { |
2694 | slot = radix_tree_iter_retry(&iter); | |
2695 | continue; | |
2696 | } | |
05f65b5c DH |
2697 | |
2698 | page = NULL; | |
2699 | } | |
2700 | ||
2701 | if (page && | |
2702 | page_count(page) - page_mapcount(page) != 1) { | |
2703 | if (scan < LAST_SCAN) | |
2704 | goto continue_resched; | |
2705 | ||
2706 | /* | |
2707 | * On the last scan, we clean up all those tags | |
2708 | * we inserted; but make a note that we still | |
2709 | * found pages pinned. | |
2710 | */ | |
2711 | error = -EBUSY; | |
2712 | } | |
2713 | ||
2714 | spin_lock_irq(&mapping->tree_lock); | |
2715 | radix_tree_tag_clear(&mapping->page_tree, | |
2716 | iter.index, SHMEM_TAG_PINNED); | |
2717 | spin_unlock_irq(&mapping->tree_lock); | |
2718 | continue_resched: | |
2719 | if (need_resched()) { | |
148deab2 | 2720 | slot = radix_tree_iter_resume(slot, &iter); |
05f65b5c | 2721 | cond_resched_rcu(); |
05f65b5c DH |
2722 | } |
2723 | } | |
2724 | rcu_read_unlock(); | |
2725 | } | |
2726 | ||
2727 | return error; | |
40e041a2 DH |
2728 | } |
2729 | ||
2730 | #define F_ALL_SEALS (F_SEAL_SEAL | \ | |
2731 | F_SEAL_SHRINK | \ | |
2732 | F_SEAL_GROW | \ | |
2733 | F_SEAL_WRITE) | |
2734 | ||
2735 | int shmem_add_seals(struct file *file, unsigned int seals) | |
2736 | { | |
2737 | struct inode *inode = file_inode(file); | |
2738 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2739 | int error; | |
2740 | ||
2741 | /* | |
2742 | * SEALING | |
2743 | * Sealing allows multiple parties to share a shmem-file but restrict | |
2744 | * access to a specific subset of file operations. Seals can only be | |
2745 | * added, but never removed. This way, mutually untrusted parties can | |
2746 | * share common memory regions with a well-defined policy. A malicious | |
2747 | * peer can thus never perform unwanted operations on a shared object. | |
2748 | * | |
2749 | * Seals are only supported on special shmem-files and always affect | |
2750 | * the whole underlying inode. Once a seal is set, it may prevent some | |
2751 | * kinds of access to the file. Currently, the following seals are | |
2752 | * defined: | |
2753 | * SEAL_SEAL: Prevent further seals from being set on this file | |
2754 | * SEAL_SHRINK: Prevent the file from shrinking | |
2755 | * SEAL_GROW: Prevent the file from growing | |
2756 | * SEAL_WRITE: Prevent write access to the file | |
2757 | * | |
2758 | * As we don't require any trust relationship between two parties, we | |
2759 | * must prevent seals from being removed. Therefore, sealing a file | |
2760 | * only adds a given set of seals to the file, it never touches | |
2761 | * existing seals. Furthermore, the "setting seals"-operation can be | |
2762 | * sealed itself, which basically prevents any further seal from being | |
2763 | * added. | |
2764 | * | |
2765 | * Semantics of sealing are only defined on volatile files. Only | |
2766 | * anonymous shmem files support sealing. More importantly, seals are | |
2767 | * never written to disk. Therefore, there's no plan to support it on | |
2768 | * other file types. | |
2769 | */ | |
2770 | ||
2771 | if (file->f_op != &shmem_file_operations) | |
2772 | return -EINVAL; | |
2773 | if (!(file->f_mode & FMODE_WRITE)) | |
2774 | return -EPERM; | |
2775 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
2776 | return -EINVAL; | |
2777 | ||
5955102c | 2778 | inode_lock(inode); |
40e041a2 DH |
2779 | |
2780 | if (info->seals & F_SEAL_SEAL) { | |
2781 | error = -EPERM; | |
2782 | goto unlock; | |
2783 | } | |
2784 | ||
2785 | if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) { | |
2786 | error = mapping_deny_writable(file->f_mapping); | |
2787 | if (error) | |
2788 | goto unlock; | |
2789 | ||
2790 | error = shmem_wait_for_pins(file->f_mapping); | |
2791 | if (error) { | |
2792 | mapping_allow_writable(file->f_mapping); | |
2793 | goto unlock; | |
2794 | } | |
2795 | } | |
2796 | ||
2797 | info->seals |= seals; | |
2798 | error = 0; | |
2799 | ||
2800 | unlock: | |
5955102c | 2801 | inode_unlock(inode); |
40e041a2 DH |
2802 | return error; |
2803 | } | |
2804 | EXPORT_SYMBOL_GPL(shmem_add_seals); | |
2805 | ||
2806 | int shmem_get_seals(struct file *file) | |
2807 | { | |
2808 | if (file->f_op != &shmem_file_operations) | |
2809 | return -EINVAL; | |
2810 | ||
2811 | return SHMEM_I(file_inode(file))->seals; | |
2812 | } | |
2813 | EXPORT_SYMBOL_GPL(shmem_get_seals); | |
2814 | ||
2815 | long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg) | |
2816 | { | |
2817 | long error; | |
2818 | ||
2819 | switch (cmd) { | |
2820 | case F_ADD_SEALS: | |
2821 | /* disallow upper 32bit */ | |
2822 | if (arg > UINT_MAX) | |
2823 | return -EINVAL; | |
2824 | ||
2825 | error = shmem_add_seals(file, arg); | |
2826 | break; | |
2827 | case F_GET_SEALS: | |
2828 | error = shmem_get_seals(file); | |
2829 | break; | |
2830 | default: | |
2831 | error = -EINVAL; | |
2832 | break; | |
2833 | } | |
2834 | ||
2835 | return error; | |
2836 | } | |
2837 | ||
83e4fa9c HD |
2838 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2839 | loff_t len) | |
2840 | { | |
496ad9aa | 2841 | struct inode *inode = file_inode(file); |
e2d12e22 | 2842 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2843 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2844 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2845 | pgoff_t start, index, end; |
2846 | int error; | |
83e4fa9c | 2847 | |
13ace4d0 HD |
2848 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2849 | return -EOPNOTSUPP; | |
2850 | ||
5955102c | 2851 | inode_lock(inode); |
83e4fa9c HD |
2852 | |
2853 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2854 | struct address_space *mapping = file->f_mapping; | |
2855 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2856 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2857 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2858 | |
40e041a2 DH |
2859 | /* protected by i_mutex */ |
2860 | if (info->seals & F_SEAL_WRITE) { | |
2861 | error = -EPERM; | |
2862 | goto out; | |
2863 | } | |
2864 | ||
8e205f77 | 2865 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
2866 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
2867 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
2868 | spin_lock(&inode->i_lock); | |
2869 | inode->i_private = &shmem_falloc; | |
2870 | spin_unlock(&inode->i_lock); | |
2871 | ||
83e4fa9c HD |
2872 | if ((u64)unmap_end > (u64)unmap_start) |
2873 | unmap_mapping_range(mapping, unmap_start, | |
2874 | 1 + unmap_end - unmap_start, 0); | |
2875 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2876 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2877 | |
2878 | spin_lock(&inode->i_lock); | |
2879 | inode->i_private = NULL; | |
2880 | wake_up_all(&shmem_falloc_waitq); | |
2055da97 | 2881 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head)); |
8e205f77 | 2882 | spin_unlock(&inode->i_lock); |
83e4fa9c | 2883 | error = 0; |
8e205f77 | 2884 | goto out; |
e2d12e22 HD |
2885 | } |
2886 | ||
2887 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2888 | error = inode_newsize_ok(inode, offset + len); | |
2889 | if (error) | |
2890 | goto out; | |
2891 | ||
40e041a2 DH |
2892 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2893 | error = -EPERM; | |
2894 | goto out; | |
2895 | } | |
2896 | ||
09cbfeaf KS |
2897 | start = offset >> PAGE_SHIFT; |
2898 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2899 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2900 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2901 | error = -ENOSPC; | |
2902 | goto out; | |
83e4fa9c HD |
2903 | } |
2904 | ||
8e205f77 | 2905 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2906 | shmem_falloc.start = start; |
2907 | shmem_falloc.next = start; | |
2908 | shmem_falloc.nr_falloced = 0; | |
2909 | shmem_falloc.nr_unswapped = 0; | |
2910 | spin_lock(&inode->i_lock); | |
2911 | inode->i_private = &shmem_falloc; | |
2912 | spin_unlock(&inode->i_lock); | |
2913 | ||
e2d12e22 HD |
2914 | for (index = start; index < end; index++) { |
2915 | struct page *page; | |
2916 | ||
2917 | /* | |
2918 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2919 | * been interrupted because we are using up too much memory. | |
2920 | */ | |
2921 | if (signal_pending(current)) | |
2922 | error = -EINTR; | |
1aac1400 HD |
2923 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2924 | error = -ENOMEM; | |
e2d12e22 | 2925 | else |
9e18eb29 | 2926 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2927 | if (error) { |
1635f6a7 | 2928 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2929 | if (index > start) { |
2930 | shmem_undo_range(inode, | |
2931 | (loff_t)start << PAGE_SHIFT, | |
2932 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2933 | } | |
1aac1400 | 2934 | goto undone; |
e2d12e22 HD |
2935 | } |
2936 | ||
1aac1400 HD |
2937 | /* |
2938 | * Inform shmem_writepage() how far we have reached. | |
2939 | * No need for lock or barrier: we have the page lock. | |
2940 | */ | |
2941 | shmem_falloc.next++; | |
2942 | if (!PageUptodate(page)) | |
2943 | shmem_falloc.nr_falloced++; | |
2944 | ||
e2d12e22 | 2945 | /* |
1635f6a7 HD |
2946 | * If !PageUptodate, leave it that way so that freeable pages |
2947 | * can be recognized if we need to rollback on error later. | |
2948 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2949 | * than free the pages we are allocating (and SGP_CACHE pages |
2950 | * might still be clean: we now need to mark those dirty too). | |
2951 | */ | |
2952 | set_page_dirty(page); | |
2953 | unlock_page(page); | |
09cbfeaf | 2954 | put_page(page); |
e2d12e22 HD |
2955 | cond_resched(); |
2956 | } | |
2957 | ||
2958 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2959 | i_size_write(inode, offset + len); | |
078cd827 | 2960 | inode->i_ctime = current_time(inode); |
1aac1400 HD |
2961 | undone: |
2962 | spin_lock(&inode->i_lock); | |
2963 | inode->i_private = NULL; | |
2964 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2965 | out: |
5955102c | 2966 | inode_unlock(inode); |
83e4fa9c HD |
2967 | return error; |
2968 | } | |
2969 | ||
726c3342 | 2970 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2971 | { |
726c3342 | 2972 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2973 | |
2974 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2975 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2976 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2977 | if (sbinfo->max_blocks) { |
1da177e4 | 2978 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2979 | buf->f_bavail = |
2980 | buf->f_bfree = sbinfo->max_blocks - | |
2981 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2982 | } |
2983 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2984 | buf->f_files = sbinfo->max_inodes; |
2985 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2986 | } |
2987 | /* else leave those fields 0 like simple_statfs */ | |
2988 | return 0; | |
2989 | } | |
2990 | ||
2991 | /* | |
2992 | * File creation. Allocate an inode, and we're done.. | |
2993 | */ | |
2994 | static int | |
1a67aafb | 2995 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2996 | { |
0b0a0806 | 2997 | struct inode *inode; |
1da177e4 LT |
2998 | int error = -ENOSPC; |
2999 | ||
454abafe | 3000 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 3001 | if (inode) { |
feda821e CH |
3002 | error = simple_acl_create(dir, inode); |
3003 | if (error) | |
3004 | goto out_iput; | |
2a7dba39 | 3005 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 3006 | &dentry->d_name, |
6d9d88d0 | 3007 | shmem_initxattrs, NULL); |
feda821e CH |
3008 | if (error && error != -EOPNOTSUPP) |
3009 | goto out_iput; | |
37ec43cd | 3010 | |
718deb6b | 3011 | error = 0; |
1da177e4 | 3012 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3013 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3014 | d_instantiate(dentry, inode); |
3015 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
3016 | } |
3017 | return error; | |
feda821e CH |
3018 | out_iput: |
3019 | iput(inode); | |
3020 | return error; | |
1da177e4 LT |
3021 | } |
3022 | ||
60545d0d AV |
3023 | static int |
3024 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
3025 | { | |
3026 | struct inode *inode; | |
3027 | int error = -ENOSPC; | |
3028 | ||
3029 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
3030 | if (inode) { | |
3031 | error = security_inode_init_security(inode, dir, | |
3032 | NULL, | |
3033 | shmem_initxattrs, NULL); | |
feda821e CH |
3034 | if (error && error != -EOPNOTSUPP) |
3035 | goto out_iput; | |
3036 | error = simple_acl_create(dir, inode); | |
3037 | if (error) | |
3038 | goto out_iput; | |
60545d0d AV |
3039 | d_tmpfile(dentry, inode); |
3040 | } | |
3041 | return error; | |
feda821e CH |
3042 | out_iput: |
3043 | iput(inode); | |
3044 | return error; | |
60545d0d AV |
3045 | } |
3046 | ||
18bb1db3 | 3047 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
3048 | { |
3049 | int error; | |
3050 | ||
3051 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
3052 | return error; | |
d8c76e6f | 3053 | inc_nlink(dir); |
1da177e4 LT |
3054 | return 0; |
3055 | } | |
3056 | ||
4acdaf27 | 3057 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 3058 | bool excl) |
1da177e4 LT |
3059 | { |
3060 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
3061 | } | |
3062 | ||
3063 | /* | |
3064 | * Link a file.. | |
3065 | */ | |
3066 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
3067 | { | |
75c3cfa8 | 3068 | struct inode *inode = d_inode(old_dentry); |
5b04c689 | 3069 | int ret; |
1da177e4 LT |
3070 | |
3071 | /* | |
3072 | * No ordinary (disk based) filesystem counts links as inodes; | |
3073 | * but each new link needs a new dentry, pinning lowmem, and | |
3074 | * tmpfs dentries cannot be pruned until they are unlinked. | |
3075 | */ | |
5b04c689 PE |
3076 | ret = shmem_reserve_inode(inode->i_sb); |
3077 | if (ret) | |
3078 | goto out; | |
1da177e4 LT |
3079 | |
3080 | dir->i_size += BOGO_DIRENT_SIZE; | |
078cd827 | 3081 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
d8c76e6f | 3082 | inc_nlink(inode); |
7de9c6ee | 3083 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
3084 | dget(dentry); /* Extra pinning count for the created dentry */ |
3085 | d_instantiate(dentry, inode); | |
5b04c689 PE |
3086 | out: |
3087 | return ret; | |
1da177e4 LT |
3088 | } |
3089 | ||
3090 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
3091 | { | |
75c3cfa8 | 3092 | struct inode *inode = d_inode(dentry); |
1da177e4 | 3093 | |
5b04c689 PE |
3094 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
3095 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
3096 | |
3097 | dir->i_size -= BOGO_DIRENT_SIZE; | |
078cd827 | 3098 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
9a53c3a7 | 3099 | drop_nlink(inode); |
1da177e4 LT |
3100 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
3101 | return 0; | |
3102 | } | |
3103 | ||
3104 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
3105 | { | |
3106 | if (!simple_empty(dentry)) | |
3107 | return -ENOTEMPTY; | |
3108 | ||
75c3cfa8 | 3109 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 3110 | drop_nlink(dir); |
1da177e4 LT |
3111 | return shmem_unlink(dir, dentry); |
3112 | } | |
3113 | ||
37456771 MS |
3114 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
3115 | { | |
e36cb0b8 DH |
3116 | bool old_is_dir = d_is_dir(old_dentry); |
3117 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
3118 | |
3119 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
3120 | if (old_is_dir) { | |
3121 | drop_nlink(old_dir); | |
3122 | inc_nlink(new_dir); | |
3123 | } else { | |
3124 | drop_nlink(new_dir); | |
3125 | inc_nlink(old_dir); | |
3126 | } | |
3127 | } | |
3128 | old_dir->i_ctime = old_dir->i_mtime = | |
3129 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 | 3130 | d_inode(old_dentry)->i_ctime = |
078cd827 | 3131 | d_inode(new_dentry)->i_ctime = current_time(old_dir); |
37456771 MS |
3132 | |
3133 | return 0; | |
3134 | } | |
3135 | ||
46fdb794 MS |
3136 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
3137 | { | |
3138 | struct dentry *whiteout; | |
3139 | int error; | |
3140 | ||
3141 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
3142 | if (!whiteout) | |
3143 | return -ENOMEM; | |
3144 | ||
3145 | error = shmem_mknod(old_dir, whiteout, | |
3146 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
3147 | dput(whiteout); | |
3148 | if (error) | |
3149 | return error; | |
3150 | ||
3151 | /* | |
3152 | * Cheat and hash the whiteout while the old dentry is still in | |
3153 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
3154 | * | |
3155 | * d_lookup() will consistently find one of them at this point, | |
3156 | * not sure which one, but that isn't even important. | |
3157 | */ | |
3158 | d_rehash(whiteout); | |
3159 | return 0; | |
3160 | } | |
3161 | ||
1da177e4 LT |
3162 | /* |
3163 | * The VFS layer already does all the dentry stuff for rename, | |
3164 | * we just have to decrement the usage count for the target if | |
3165 | * it exists so that the VFS layer correctly free's it when it | |
3166 | * gets overwritten. | |
3167 | */ | |
3b69ff51 | 3168 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 3169 | { |
75c3cfa8 | 3170 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
3171 | int they_are_dirs = S_ISDIR(inode->i_mode); |
3172 | ||
46fdb794 | 3173 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
3174 | return -EINVAL; |
3175 | ||
37456771 MS |
3176 | if (flags & RENAME_EXCHANGE) |
3177 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
3178 | ||
1da177e4 LT |
3179 | if (!simple_empty(new_dentry)) |
3180 | return -ENOTEMPTY; | |
3181 | ||
46fdb794 MS |
3182 | if (flags & RENAME_WHITEOUT) { |
3183 | int error; | |
3184 | ||
3185 | error = shmem_whiteout(old_dir, old_dentry); | |
3186 | if (error) | |
3187 | return error; | |
3188 | } | |
3189 | ||
75c3cfa8 | 3190 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 3191 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 3192 | if (they_are_dirs) { |
75c3cfa8 | 3193 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 3194 | drop_nlink(old_dir); |
b928095b | 3195 | } |
1da177e4 | 3196 | } else if (they_are_dirs) { |
9a53c3a7 | 3197 | drop_nlink(old_dir); |
d8c76e6f | 3198 | inc_nlink(new_dir); |
1da177e4 LT |
3199 | } |
3200 | ||
3201 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
3202 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
3203 | old_dir->i_ctime = old_dir->i_mtime = | |
3204 | new_dir->i_ctime = new_dir->i_mtime = | |
078cd827 | 3205 | inode->i_ctime = current_time(old_dir); |
1da177e4 LT |
3206 | return 0; |
3207 | } | |
3208 | ||
3209 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
3210 | { | |
3211 | int error; | |
3212 | int len; | |
3213 | struct inode *inode; | |
9276aad6 | 3214 | struct page *page; |
1da177e4 LT |
3215 | struct shmem_inode_info *info; |
3216 | ||
3217 | len = strlen(symname) + 1; | |
09cbfeaf | 3218 | if (len > PAGE_SIZE) |
1da177e4 LT |
3219 | return -ENAMETOOLONG; |
3220 | ||
454abafe | 3221 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
3222 | if (!inode) |
3223 | return -ENOSPC; | |
3224 | ||
9d8f13ba | 3225 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3226 | shmem_initxattrs, NULL); |
570bc1c2 SS |
3227 | if (error) { |
3228 | if (error != -EOPNOTSUPP) { | |
3229 | iput(inode); | |
3230 | return error; | |
3231 | } | |
3232 | error = 0; | |
3233 | } | |
3234 | ||
1da177e4 LT |
3235 | info = SHMEM_I(inode); |
3236 | inode->i_size = len-1; | |
69f07ec9 | 3237 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3238 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3239 | if (!inode->i_link) { | |
69f07ec9 HD |
3240 | iput(inode); |
3241 | return -ENOMEM; | |
3242 | } | |
3243 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3244 | } else { |
e8ecde25 | 3245 | inode_nohighmem(inode); |
9e18eb29 | 3246 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
3247 | if (error) { |
3248 | iput(inode); | |
3249 | return error; | |
3250 | } | |
14fcc23f | 3251 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 3252 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 3253 | memcpy(page_address(page), symname, len); |
ec9516fb | 3254 | SetPageUptodate(page); |
1da177e4 | 3255 | set_page_dirty(page); |
6746aff7 | 3256 | unlock_page(page); |
09cbfeaf | 3257 | put_page(page); |
1da177e4 | 3258 | } |
1da177e4 | 3259 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3260 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3261 | d_instantiate(dentry, inode); |
3262 | dget(dentry); | |
3263 | return 0; | |
3264 | } | |
3265 | ||
fceef393 | 3266 | static void shmem_put_link(void *arg) |
1da177e4 | 3267 | { |
fceef393 AV |
3268 | mark_page_accessed(arg); |
3269 | put_page(arg); | |
1da177e4 LT |
3270 | } |
3271 | ||
6b255391 | 3272 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
3273 | struct inode *inode, |
3274 | struct delayed_call *done) | |
1da177e4 | 3275 | { |
1da177e4 | 3276 | struct page *page = NULL; |
6b255391 | 3277 | int error; |
6a6c9904 AV |
3278 | if (!dentry) { |
3279 | page = find_get_page(inode->i_mapping, 0); | |
3280 | if (!page) | |
3281 | return ERR_PTR(-ECHILD); | |
3282 | if (!PageUptodate(page)) { | |
3283 | put_page(page); | |
3284 | return ERR_PTR(-ECHILD); | |
3285 | } | |
3286 | } else { | |
9e18eb29 | 3287 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3288 | if (error) |
3289 | return ERR_PTR(error); | |
3290 | unlock_page(page); | |
3291 | } | |
fceef393 | 3292 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3293 | return page_address(page); |
1da177e4 LT |
3294 | } |
3295 | ||
b09e0fa4 | 3296 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3297 | /* |
b09e0fa4 EP |
3298 | * Superblocks without xattr inode operations may get some security.* xattr |
3299 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3300 | * like ACLs, we also need to implement the security.* handlers at |
3301 | * filesystem level, though. | |
3302 | */ | |
3303 | ||
6d9d88d0 JS |
3304 | /* |
3305 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3306 | */ | |
3307 | static int shmem_initxattrs(struct inode *inode, | |
3308 | const struct xattr *xattr_array, | |
3309 | void *fs_info) | |
3310 | { | |
3311 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3312 | const struct xattr *xattr; | |
38f38657 | 3313 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3314 | size_t len; |
3315 | ||
3316 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3317 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3318 | if (!new_xattr) |
3319 | return -ENOMEM; | |
3320 | ||
3321 | len = strlen(xattr->name) + 1; | |
3322 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3323 | GFP_KERNEL); | |
3324 | if (!new_xattr->name) { | |
3325 | kfree(new_xattr); | |
3326 | return -ENOMEM; | |
3327 | } | |
3328 | ||
3329 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3330 | XATTR_SECURITY_PREFIX_LEN); | |
3331 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3332 | xattr->name, len); | |
3333 | ||
38f38657 | 3334 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3335 | } |
3336 | ||
3337 | return 0; | |
3338 | } | |
3339 | ||
aa7c5241 | 3340 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3341 | struct dentry *unused, struct inode *inode, |
3342 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3343 | { |
b296821a | 3344 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3345 | |
aa7c5241 | 3346 | name = xattr_full_name(handler, name); |
38f38657 | 3347 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3348 | } |
3349 | ||
aa7c5241 | 3350 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
59301226 AV |
3351 | struct dentry *unused, struct inode *inode, |
3352 | const char *name, const void *value, | |
3353 | size_t size, int flags) | |
b09e0fa4 | 3354 | { |
59301226 | 3355 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3356 | |
aa7c5241 | 3357 | name = xattr_full_name(handler, name); |
38f38657 | 3358 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
3359 | } |
3360 | ||
aa7c5241 AG |
3361 | static const struct xattr_handler shmem_security_xattr_handler = { |
3362 | .prefix = XATTR_SECURITY_PREFIX, | |
3363 | .get = shmem_xattr_handler_get, | |
3364 | .set = shmem_xattr_handler_set, | |
3365 | }; | |
b09e0fa4 | 3366 | |
aa7c5241 AG |
3367 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3368 | .prefix = XATTR_TRUSTED_PREFIX, | |
3369 | .get = shmem_xattr_handler_get, | |
3370 | .set = shmem_xattr_handler_set, | |
3371 | }; | |
b09e0fa4 | 3372 | |
aa7c5241 AG |
3373 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3374 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3375 | &posix_acl_access_xattr_handler, | |
3376 | &posix_acl_default_xattr_handler, | |
3377 | #endif | |
3378 | &shmem_security_xattr_handler, | |
3379 | &shmem_trusted_xattr_handler, | |
3380 | NULL | |
3381 | }; | |
b09e0fa4 EP |
3382 | |
3383 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3384 | { | |
75c3cfa8 | 3385 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3386 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3387 | } |
3388 | #endif /* CONFIG_TMPFS_XATTR */ | |
3389 | ||
69f07ec9 | 3390 | static const struct inode_operations shmem_short_symlink_operations = { |
6b255391 | 3391 | .get_link = simple_get_link, |
b09e0fa4 | 3392 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3393 | .listxattr = shmem_listxattr, |
b09e0fa4 EP |
3394 | #endif |
3395 | }; | |
3396 | ||
3397 | static const struct inode_operations shmem_symlink_inode_operations = { | |
6b255391 | 3398 | .get_link = shmem_get_link, |
b09e0fa4 | 3399 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3400 | .listxattr = shmem_listxattr, |
39f0247d | 3401 | #endif |
b09e0fa4 | 3402 | }; |
39f0247d | 3403 | |
91828a40 DG |
3404 | static struct dentry *shmem_get_parent(struct dentry *child) |
3405 | { | |
3406 | return ERR_PTR(-ESTALE); | |
3407 | } | |
3408 | ||
3409 | static int shmem_match(struct inode *ino, void *vfh) | |
3410 | { | |
3411 | __u32 *fh = vfh; | |
b6450630 | 3412 | __u64 inum = fh[1]; |
91828a40 DG |
3413 | return ino->i_ino == inum && fh[0] == ino->i_generation; |
3414 | } | |
3415 | ||
480b116c CH |
3416 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3417 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3418 | { |
91828a40 | 3419 | struct inode *inode; |
480b116c | 3420 | struct dentry *dentry = NULL; |
35c2a7f4 | 3421 | u64 inum; |
480b116c | 3422 | |
b6450630 | 3423 | if (fh_len < 2) |
480b116c | 3424 | return NULL; |
91828a40 | 3425 | |
b6450630 SF |
3426 | inum = fid->raw[1]; |
3427 | inode = ilookup5(sb, inum, shmem_match, fid->raw); | |
91828a40 | 3428 | if (inode) { |
480b116c | 3429 | dentry = d_find_alias(inode); |
91828a40 DG |
3430 | iput(inode); |
3431 | } | |
3432 | ||
480b116c | 3433 | return dentry; |
91828a40 DG |
3434 | } |
3435 | ||
b0b0382b AV |
3436 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3437 | struct inode *parent) | |
91828a40 | 3438 | { |
b6450630 SF |
3439 | if (*len < 2) { |
3440 | *len = 2; | |
94e07a75 | 3441 | return FILEID_INVALID; |
5fe0c237 | 3442 | } |
91828a40 | 3443 | |
91828a40 DG |
3444 | fh[0] = inode->i_generation; |
3445 | fh[1] = inode->i_ino; | |
91828a40 | 3446 | |
b6450630 | 3447 | *len = 2; |
91828a40 DG |
3448 | return 1; |
3449 | } | |
3450 | ||
39655164 | 3451 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3452 | .get_parent = shmem_get_parent, |
91828a40 | 3453 | .encode_fh = shmem_encode_fh, |
480b116c | 3454 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3455 | }; |
3456 | ||
680d794b AM |
3457 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
3458 | bool remount) | |
1da177e4 LT |
3459 | { |
3460 | char *this_char, *value, *rest; | |
49cd0a5c | 3461 | struct mempolicy *mpol = NULL; |
8751e039 EB |
3462 | uid_t uid; |
3463 | gid_t gid; | |
1da177e4 | 3464 | |
b00dc3ad HD |
3465 | while (options != NULL) { |
3466 | this_char = options; | |
3467 | for (;;) { | |
3468 | /* | |
3469 | * NUL-terminate this option: unfortunately, | |
3470 | * mount options form a comma-separated list, | |
3471 | * but mpol's nodelist may also contain commas. | |
3472 | */ | |
3473 | options = strchr(options, ','); | |
3474 | if (options == NULL) | |
3475 | break; | |
3476 | options++; | |
3477 | if (!isdigit(*options)) { | |
3478 | options[-1] = '\0'; | |
3479 | break; | |
3480 | } | |
3481 | } | |
1da177e4 LT |
3482 | if (!*this_char) |
3483 | continue; | |
3484 | if ((value = strchr(this_char,'=')) != NULL) { | |
3485 | *value++ = 0; | |
3486 | } else { | |
1170532b JP |
3487 | pr_err("tmpfs: No value for mount option '%s'\n", |
3488 | this_char); | |
49cd0a5c | 3489 | goto error; |
1da177e4 LT |
3490 | } |
3491 | ||
3492 | if (!strcmp(this_char,"size")) { | |
3493 | unsigned long long size; | |
3494 | size = memparse(value,&rest); | |
3495 | if (*rest == '%') { | |
3496 | size <<= PAGE_SHIFT; | |
3497 | size *= totalram_pages; | |
3498 | do_div(size, 100); | |
3499 | rest++; | |
3500 | } | |
3501 | if (*rest) | |
3502 | goto bad_val; | |
680d794b | 3503 | sbinfo->max_blocks = |
09cbfeaf | 3504 | DIV_ROUND_UP(size, PAGE_SIZE); |
1da177e4 | 3505 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 3506 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
3507 | if (*rest) |
3508 | goto bad_val; | |
3509 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 3510 | sbinfo->max_inodes = memparse(value, &rest); |
b6450630 | 3511 | if (*rest || sbinfo->max_inodes < 2) |
1da177e4 LT |
3512 | goto bad_val; |
3513 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 3514 | if (remount) |
1da177e4 | 3515 | continue; |
680d794b | 3516 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
3517 | if (*rest) |
3518 | goto bad_val; | |
3519 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 3520 | if (remount) |
1da177e4 | 3521 | continue; |
8751e039 | 3522 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3523 | if (*rest) |
3524 | goto bad_val; | |
8751e039 EB |
3525 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
3526 | if (!uid_valid(sbinfo->uid)) | |
3527 | goto bad_val; | |
1da177e4 | 3528 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 3529 | if (remount) |
1da177e4 | 3530 | continue; |
8751e039 | 3531 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3532 | if (*rest) |
3533 | goto bad_val; | |
8751e039 EB |
3534 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
3535 | if (!gid_valid(sbinfo->gid)) | |
3536 | goto bad_val; | |
e496cf3d | 3537 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3538 | } else if (!strcmp(this_char, "huge")) { |
3539 | int huge; | |
3540 | huge = shmem_parse_huge(value); | |
3541 | if (huge < 0) | |
3542 | goto bad_val; | |
3543 | if (!has_transparent_hugepage() && | |
3544 | huge != SHMEM_HUGE_NEVER) | |
3545 | goto bad_val; | |
3546 | sbinfo->huge = huge; | |
3547 | #endif | |
3548 | #ifdef CONFIG_NUMA | |
7339ff83 | 3549 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
3550 | mpol_put(mpol); |
3551 | mpol = NULL; | |
3552 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 3553 | goto bad_val; |
5a6e75f8 | 3554 | #endif |
1da177e4 | 3555 | } else { |
1170532b | 3556 | pr_err("tmpfs: Bad mount option %s\n", this_char); |
49cd0a5c | 3557 | goto error; |
1da177e4 LT |
3558 | } |
3559 | } | |
49cd0a5c | 3560 | sbinfo->mpol = mpol; |
1da177e4 LT |
3561 | return 0; |
3562 | ||
3563 | bad_val: | |
1170532b | 3564 | pr_err("tmpfs: Bad value '%s' for mount option '%s'\n", |
1da177e4 | 3565 | value, this_char); |
49cd0a5c GT |
3566 | error: |
3567 | mpol_put(mpol); | |
1da177e4 LT |
3568 | return 1; |
3569 | ||
3570 | } | |
3571 | ||
3572 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
3573 | { | |
3574 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 3575 | struct shmem_sb_info config = *sbinfo; |
b6450630 | 3576 | int inodes; |
0edd73b3 HD |
3577 | int error = -EINVAL; |
3578 | ||
5f00110f | 3579 | config.mpol = NULL; |
680d794b | 3580 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 3581 | return error; |
1da177e4 | 3582 | |
0edd73b3 | 3583 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3584 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 3585 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 3586 | goto out; |
680d794b | 3587 | if (config.max_inodes < inodes) |
0edd73b3 HD |
3588 | goto out; |
3589 | /* | |
54af6042 | 3590 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
3591 | * but we must separately disallow unlimited->limited, because |
3592 | * in that case we have no record of how much is already in use. | |
3593 | */ | |
680d794b | 3594 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 3595 | goto out; |
680d794b | 3596 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
3597 | goto out; |
3598 | ||
3599 | error = 0; | |
5a6e75f8 | 3600 | sbinfo->huge = config.huge; |
680d794b | 3601 | sbinfo->max_blocks = config.max_blocks; |
680d794b AM |
3602 | sbinfo->max_inodes = config.max_inodes; |
3603 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 3604 | |
5f00110f GT |
3605 | /* |
3606 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3607 | */ | |
3608 | if (config.mpol) { | |
3609 | mpol_put(sbinfo->mpol); | |
3610 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
3611 | } | |
0edd73b3 HD |
3612 | out: |
3613 | spin_unlock(&sbinfo->stat_lock); | |
3614 | return error; | |
1da177e4 | 3615 | } |
680d794b | 3616 | |
34c80b1d | 3617 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3618 | { |
34c80b1d | 3619 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b AM |
3620 | |
3621 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3622 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3623 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 3624 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
b6450630 | 3625 | seq_printf(seq, ",nr_inodes=%d", sbinfo->max_inodes); |
680d794b | 3626 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) |
09208d15 | 3627 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3628 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3629 | seq_printf(seq, ",uid=%u", | |
3630 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3631 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3632 | seq_printf(seq, ",gid=%u", | |
3633 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
e496cf3d | 3634 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3635 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
3636 | if (sbinfo->huge) | |
3637 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3638 | #endif | |
71fe804b | 3639 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b AM |
3640 | return 0; |
3641 | } | |
9183df25 DH |
3642 | |
3643 | #define MFD_NAME_PREFIX "memfd:" | |
3644 | #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1) | |
3645 | #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN) | |
3646 | ||
3647 | #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING) | |
3648 | ||
3649 | SYSCALL_DEFINE2(memfd_create, | |
3650 | const char __user *, uname, | |
3651 | unsigned int, flags) | |
3652 | { | |
3653 | struct shmem_inode_info *info; | |
3654 | struct file *file; | |
3655 | int fd, error; | |
3656 | char *name; | |
3657 | long len; | |
3658 | ||
3659 | if (flags & ~(unsigned int)MFD_ALL_FLAGS) | |
3660 | return -EINVAL; | |
3661 | ||
3662 | /* length includes terminating zero */ | |
3663 | len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1); | |
3664 | if (len <= 0) | |
3665 | return -EFAULT; | |
3666 | if (len > MFD_NAME_MAX_LEN + 1) | |
3667 | return -EINVAL; | |
3668 | ||
3669 | name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY); | |
3670 | if (!name) | |
3671 | return -ENOMEM; | |
3672 | ||
3673 | strcpy(name, MFD_NAME_PREFIX); | |
3674 | if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) { | |
3675 | error = -EFAULT; | |
3676 | goto err_name; | |
3677 | } | |
3678 | ||
3679 | /* terminating-zero may have changed after strnlen_user() returned */ | |
3680 | if (name[len + MFD_NAME_PREFIX_LEN - 1]) { | |
3681 | error = -EFAULT; | |
3682 | goto err_name; | |
3683 | } | |
3684 | ||
3685 | fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0); | |
3686 | if (fd < 0) { | |
3687 | error = fd; | |
3688 | goto err_name; | |
3689 | } | |
3690 | ||
3691 | file = shmem_file_setup(name, 0, VM_NORESERVE); | |
3692 | if (IS_ERR(file)) { | |
3693 | error = PTR_ERR(file); | |
3694 | goto err_fd; | |
3695 | } | |
3696 | info = SHMEM_I(file_inode(file)); | |
3697 | file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; | |
3698 | file->f_flags |= O_RDWR | O_LARGEFILE; | |
3699 | if (flags & MFD_ALLOW_SEALING) | |
3700 | info->seals &= ~F_SEAL_SEAL; | |
3701 | ||
3702 | fd_install(fd, file); | |
3703 | kfree(name); | |
3704 | return fd; | |
3705 | ||
3706 | err_fd: | |
3707 | put_unused_fd(fd); | |
3708 | err_name: | |
3709 | kfree(name); | |
3710 | return error; | |
3711 | } | |
3712 | ||
680d794b | 3713 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3714 | |
3715 | static void shmem_put_super(struct super_block *sb) | |
3716 | { | |
602586a8 HD |
3717 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3718 | ||
b6450630 SF |
3719 | if (!sbinfo->idr_nouse) |
3720 | idr_destroy(&sbinfo->idr); | |
602586a8 | 3721 | percpu_counter_destroy(&sbinfo->used_blocks); |
49cd0a5c | 3722 | mpol_put(sbinfo->mpol); |
602586a8 | 3723 | kfree(sbinfo); |
1da177e4 LT |
3724 | sb->s_fs_info = NULL; |
3725 | } | |
3726 | ||
2b2af54a | 3727 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3728 | { |
3729 | struct inode *inode; | |
0edd73b3 | 3730 | struct shmem_sb_info *sbinfo; |
680d794b AM |
3731 | int err = -ENOMEM; |
3732 | ||
3733 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3734 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b AM |
3735 | L1_CACHE_BYTES), GFP_KERNEL); |
3736 | if (!sbinfo) | |
3737 | return -ENOMEM; | |
3738 | ||
b6450630 SF |
3739 | mutex_init(&sbinfo->idr_lock); |
3740 | idr_init(&sbinfo->idr); | |
680d794b | 3741 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
3742 | sbinfo->uid = current_fsuid(); |
3743 | sbinfo->gid = current_fsgid(); | |
680d794b | 3744 | sb->s_fs_info = sbinfo; |
1da177e4 | 3745 | |
0edd73b3 | 3746 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3747 | /* |
3748 | * Per default we only allow half of the physical ram per | |
3749 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3750 | * but the internal instance is left unlimited. | |
3751 | */ | |
ca4e0519 | 3752 | if (!(sb->s_flags & MS_KERNMOUNT)) { |
680d794b AM |
3753 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3754 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3755 | if (shmem_parse_options(data, sbinfo, false)) { | |
3756 | err = -EINVAL; | |
3757 | goto failed; | |
3758 | } | |
ca4e0519 AV |
3759 | } else { |
3760 | sb->s_flags |= MS_NOUSER; | |
1da177e4 | 3761 | } |
91828a40 | 3762 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 3763 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
3764 | #else |
3765 | sb->s_flags |= MS_NOUSER; | |
3766 | #endif | |
3767 | ||
0edd73b3 | 3768 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3769 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3770 | goto failed; |
680d794b | 3771 | sbinfo->free_inodes = sbinfo->max_inodes; |
779750d2 KS |
3772 | spin_lock_init(&sbinfo->shrinklist_lock); |
3773 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 3774 | |
285b2c4f | 3775 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3776 | sb->s_blocksize = PAGE_SIZE; |
3777 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3778 | sb->s_magic = TMPFS_MAGIC; |
3779 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3780 | sb->s_time_gran = 1; |
b09e0fa4 | 3781 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3782 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3783 | #endif |
3784 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
3785 | sb->s_flags |= MS_POSIXACL; |
3786 | #endif | |
2b4db796 | 3787 | uuid_gen(&sb->s_uuid); |
0edd73b3 | 3788 | |
454abafe | 3789 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3790 | if (!inode) |
3791 | goto failed; | |
680d794b AM |
3792 | inode->i_uid = sbinfo->uid; |
3793 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3794 | sb->s_root = d_make_root(inode); |
3795 | if (!sb->s_root) | |
48fde701 | 3796 | goto failed; |
1da177e4 LT |
3797 | return 0; |
3798 | ||
1da177e4 LT |
3799 | failed: |
3800 | shmem_put_super(sb); | |
3801 | return err; | |
3802 | } | |
3803 | ||
fcc234f8 | 3804 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3805 | |
3806 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3807 | { | |
41ffe5d5 HD |
3808 | struct shmem_inode_info *info; |
3809 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3810 | if (!info) | |
1da177e4 | 3811 | return NULL; |
41ffe5d5 | 3812 | return &info->vfs_inode; |
1da177e4 LT |
3813 | } |
3814 | ||
41ffe5d5 | 3815 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3816 | { |
3817 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84e710da AV |
3818 | if (S_ISLNK(inode->i_mode)) |
3819 | kfree(inode->i_link); | |
fa0d7e3d NP |
3820 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3821 | } | |
3822 | ||
1da177e4 LT |
3823 | static void shmem_destroy_inode(struct inode *inode) |
3824 | { | |
09208d15 | 3825 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3826 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3827 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3828 | } |
3829 | ||
41ffe5d5 | 3830 | static void shmem_init_inode(void *foo) |
1da177e4 | 3831 | { |
41ffe5d5 HD |
3832 | struct shmem_inode_info *info = foo; |
3833 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3834 | } |
3835 | ||
41ffe5d5 | 3836 | static int shmem_init_inodecache(void) |
1da177e4 LT |
3837 | { |
3838 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3839 | sizeof(struct shmem_inode_info), | |
5d097056 | 3840 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3841 | return 0; |
3842 | } | |
3843 | ||
41ffe5d5 | 3844 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3845 | { |
1a1d92c1 | 3846 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3847 | } |
3848 | ||
b6450630 SF |
3849 | static __init void shmem_no_idr(struct super_block *sb) |
3850 | { | |
3851 | struct shmem_sb_info *sbinfo; | |
3852 | ||
3853 | sbinfo = SHMEM_SB(sb); | |
3854 | sbinfo->idr_nouse = true; | |
3855 | idr_destroy(&sbinfo->idr); | |
3856 | } | |
3857 | ||
f5e54d6e | 3858 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3859 | .writepage = shmem_writepage, |
76719325 | 3860 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3861 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3862 | .write_begin = shmem_write_begin, |
3863 | .write_end = shmem_write_end, | |
1da177e4 | 3864 | #endif |
1c93923c | 3865 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3866 | .migratepage = migrate_page, |
1c93923c | 3867 | #endif |
aa261f54 | 3868 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3869 | }; |
3870 | ||
15ad7cdc | 3871 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3872 | .mmap = shmem_mmap, |
c01d5b30 | 3873 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3874 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3875 | .llseek = shmem_file_llseek, |
2ba5bbed | 3876 | .read_iter = shmem_file_read_iter, |
8174202b | 3877 | .write_iter = generic_file_write_iter, |
1b061d92 | 3878 | .fsync = noop_fsync, |
82c156f8 | 3879 | .splice_read = generic_file_splice_read, |
f6cb85d0 | 3880 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3881 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3882 | #endif |
3883 | }; | |
3884 | ||
92e1d5be | 3885 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3886 | .getattr = shmem_getattr, |
94c1e62d | 3887 | .setattr = shmem_setattr, |
b09e0fa4 | 3888 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3889 | .listxattr = shmem_listxattr, |
feda821e | 3890 | .set_acl = simple_set_acl, |
b09e0fa4 | 3891 | #endif |
1da177e4 LT |
3892 | }; |
3893 | ||
92e1d5be | 3894 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3895 | #ifdef CONFIG_TMPFS |
3896 | .create = shmem_create, | |
3897 | .lookup = simple_lookup, | |
3898 | .link = shmem_link, | |
3899 | .unlink = shmem_unlink, | |
3900 | .symlink = shmem_symlink, | |
3901 | .mkdir = shmem_mkdir, | |
3902 | .rmdir = shmem_rmdir, | |
3903 | .mknod = shmem_mknod, | |
2773bf00 | 3904 | .rename = shmem_rename2, |
60545d0d | 3905 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3906 | #endif |
b09e0fa4 | 3907 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3908 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3909 | #endif |
39f0247d | 3910 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3911 | .setattr = shmem_setattr, |
feda821e | 3912 | .set_acl = simple_set_acl, |
39f0247d AG |
3913 | #endif |
3914 | }; | |
3915 | ||
92e1d5be | 3916 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3917 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3918 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3919 | #endif |
39f0247d | 3920 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3921 | .setattr = shmem_setattr, |
feda821e | 3922 | .set_acl = simple_set_acl, |
39f0247d | 3923 | #endif |
1da177e4 LT |
3924 | }; |
3925 | ||
759b9775 | 3926 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
3927 | .alloc_inode = shmem_alloc_inode, |
3928 | .destroy_inode = shmem_destroy_inode, | |
3929 | #ifdef CONFIG_TMPFS | |
3930 | .statfs = shmem_statfs, | |
3931 | .remount_fs = shmem_remount_fs, | |
680d794b | 3932 | .show_options = shmem_show_options, |
1da177e4 | 3933 | #endif |
1f895f75 | 3934 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3935 | .drop_inode = generic_delete_inode, |
3936 | .put_super = shmem_put_super, | |
779750d2 KS |
3937 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
3938 | .nr_cached_objects = shmem_unused_huge_count, | |
3939 | .free_cached_objects = shmem_unused_huge_scan, | |
3940 | #endif | |
1da177e4 LT |
3941 | }; |
3942 | ||
f0f37e2f | 3943 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3944 | .fault = shmem_fault, |
d7c17551 | 3945 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3946 | #ifdef CONFIG_NUMA |
3947 | .set_policy = shmem_set_policy, | |
3948 | .get_policy = shmem_get_policy, | |
3949 | #endif | |
3950 | }; | |
3951 | ||
3c26ff6e AV |
3952 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
3953 | int flags, const char *dev_name, void *data) | |
1da177e4 | 3954 | { |
3c26ff6e | 3955 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
3956 | } |
3957 | ||
41ffe5d5 | 3958 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3959 | .owner = THIS_MODULE, |
3960 | .name = "tmpfs", | |
3c26ff6e | 3961 | .mount = shmem_mount, |
1da177e4 | 3962 | .kill_sb = kill_litter_super, |
2b8576cb | 3963 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 3964 | }; |
1da177e4 | 3965 | |
41ffe5d5 | 3966 | int __init shmem_init(void) |
1da177e4 LT |
3967 | { |
3968 | int error; | |
3969 | ||
16203a7a RL |
3970 | /* If rootfs called this, don't re-init */ |
3971 | if (shmem_inode_cachep) | |
3972 | return 0; | |
3973 | ||
41ffe5d5 | 3974 | error = shmem_init_inodecache(); |
1da177e4 LT |
3975 | if (error) |
3976 | goto out3; | |
3977 | ||
41ffe5d5 | 3978 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 3979 | if (error) { |
1170532b | 3980 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
3981 | goto out2; |
3982 | } | |
95dc112a | 3983 | |
ca4e0519 | 3984 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
3985 | if (IS_ERR(shm_mnt)) { |
3986 | error = PTR_ERR(shm_mnt); | |
1170532b | 3987 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
3988 | goto out1; |
3989 | } | |
b6450630 | 3990 | shmem_no_idr(shm_mnt->mnt_sb); |
5a6e75f8 | 3991 | |
e496cf3d | 3992 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
435c0b87 | 3993 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
3994 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
3995 | else | |
3996 | shmem_huge = 0; /* just in case it was patched */ | |
3997 | #endif | |
1da177e4 LT |
3998 | return 0; |
3999 | ||
4000 | out1: | |
41ffe5d5 | 4001 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 4002 | out2: |
41ffe5d5 | 4003 | shmem_destroy_inodecache(); |
1da177e4 LT |
4004 | out3: |
4005 | shm_mnt = ERR_PTR(error); | |
4006 | return error; | |
4007 | } | |
853ac43a | 4008 | |
e496cf3d | 4009 | #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS) |
5a6e75f8 KS |
4010 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
4011 | struct kobj_attribute *attr, char *buf) | |
4012 | { | |
4013 | int values[] = { | |
4014 | SHMEM_HUGE_ALWAYS, | |
4015 | SHMEM_HUGE_WITHIN_SIZE, | |
4016 | SHMEM_HUGE_ADVISE, | |
4017 | SHMEM_HUGE_NEVER, | |
4018 | SHMEM_HUGE_DENY, | |
4019 | SHMEM_HUGE_FORCE, | |
4020 | }; | |
4021 | int i, count; | |
4022 | ||
4023 | for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) { | |
4024 | const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s "; | |
4025 | ||
4026 | count += sprintf(buf + count, fmt, | |
4027 | shmem_format_huge(values[i])); | |
4028 | } | |
4029 | buf[count - 1] = '\n'; | |
4030 | return count; | |
4031 | } | |
4032 | ||
4033 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
4034 | struct kobj_attribute *attr, const char *buf, size_t count) | |
4035 | { | |
4036 | char tmp[16]; | |
4037 | int huge; | |
4038 | ||
4039 | if (count + 1 > sizeof(tmp)) | |
4040 | return -EINVAL; | |
4041 | memcpy(tmp, buf, count); | |
4042 | tmp[count] = '\0'; | |
4043 | if (count && tmp[count - 1] == '\n') | |
4044 | tmp[count - 1] = '\0'; | |
4045 | ||
4046 | huge = shmem_parse_huge(tmp); | |
4047 | if (huge == -EINVAL) | |
4048 | return -EINVAL; | |
4049 | if (!has_transparent_hugepage() && | |
4050 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
4051 | return -EINVAL; | |
4052 | ||
4053 | shmem_huge = huge; | |
435c0b87 | 4054 | if (shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4055 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4056 | return count; | |
4057 | } | |
4058 | ||
4059 | struct kobj_attribute shmem_enabled_attr = | |
4060 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
3b33719c | 4061 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */ |
f3f0e1d2 | 4062 | |
3b33719c | 4063 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
f3f0e1d2 KS |
4064 | bool shmem_huge_enabled(struct vm_area_struct *vma) |
4065 | { | |
4066 | struct inode *inode = file_inode(vma->vm_file); | |
4067 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4068 | loff_t i_size; | |
4069 | pgoff_t off; | |
4070 | ||
4071 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
4072 | return true; | |
4073 | if (shmem_huge == SHMEM_HUGE_DENY) | |
4074 | return false; | |
4075 | switch (sbinfo->huge) { | |
4076 | case SHMEM_HUGE_NEVER: | |
4077 | return false; | |
4078 | case SHMEM_HUGE_ALWAYS: | |
4079 | return true; | |
4080 | case SHMEM_HUGE_WITHIN_SIZE: | |
4081 | off = round_up(vma->vm_pgoff, HPAGE_PMD_NR); | |
4082 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
4083 | if (i_size >= HPAGE_PMD_SIZE && | |
4084 | i_size >> PAGE_SHIFT >= off) | |
4085 | return true; | |
4086 | case SHMEM_HUGE_ADVISE: | |
4087 | /* TODO: implement fadvise() hints */ | |
4088 | return (vma->vm_flags & VM_HUGEPAGE); | |
4089 | default: | |
4090 | VM_BUG_ON(1); | |
4091 | return false; | |
4092 | } | |
4093 | } | |
3b33719c | 4094 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 4095 | |
853ac43a MM |
4096 | #else /* !CONFIG_SHMEM */ |
4097 | ||
4098 | /* | |
4099 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
4100 | * | |
4101 | * This is intended for small system where the benefits of the full | |
4102 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4103 | * their complexity. On systems without swap this code should be | |
4104 | * effectively equivalent, but much lighter weight. | |
4105 | */ | |
4106 | ||
41ffe5d5 | 4107 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4108 | .name = "tmpfs", |
3c26ff6e | 4109 | .mount = ramfs_mount, |
853ac43a | 4110 | .kill_sb = kill_litter_super, |
2b8576cb | 4111 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4112 | }; |
4113 | ||
41ffe5d5 | 4114 | int __init shmem_init(void) |
853ac43a | 4115 | { |
41ffe5d5 | 4116 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4117 | |
41ffe5d5 | 4118 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
4119 | BUG_ON(IS_ERR(shm_mnt)); |
4120 | ||
4121 | return 0; | |
4122 | } | |
4123 | ||
41ffe5d5 | 4124 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
4125 | { |
4126 | return 0; | |
4127 | } | |
4128 | ||
3f96b79a HD |
4129 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
4130 | { | |
4131 | return 0; | |
4132 | } | |
4133 | ||
24513264 HD |
4134 | void shmem_unlock_mapping(struct address_space *mapping) |
4135 | { | |
4136 | } | |
4137 | ||
c01d5b30 HD |
4138 | #ifdef CONFIG_MMU |
4139 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4140 | unsigned long addr, unsigned long len, | |
4141 | unsigned long pgoff, unsigned long flags) | |
4142 | { | |
4143 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4144 | } | |
4145 | #endif | |
4146 | ||
41ffe5d5 | 4147 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4148 | { |
41ffe5d5 | 4149 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4150 | } |
4151 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4152 | ||
0b0a0806 HD |
4153 | #define shmem_vm_ops generic_file_vm_ops |
4154 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 4155 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
4156 | #define shmem_acct_size(flags, size) 0 |
4157 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
4158 | |
4159 | #endif /* CONFIG_SHMEM */ | |
4160 | ||
4161 | /* common code */ | |
1da177e4 | 4162 | |
19938e35 | 4163 | static const struct dentry_operations anon_ops = { |
118b2302 | 4164 | .d_dname = simple_dname |
3451538a AV |
4165 | }; |
4166 | ||
c7277090 EP |
4167 | static struct file *__shmem_file_setup(const char *name, loff_t size, |
4168 | unsigned long flags, unsigned int i_flags) | |
1da177e4 | 4169 | { |
6b4d0b27 | 4170 | struct file *res; |
1da177e4 | 4171 | struct inode *inode; |
2c48b9c4 | 4172 | struct path path; |
3451538a | 4173 | struct super_block *sb; |
1da177e4 LT |
4174 | struct qstr this; |
4175 | ||
4176 | if (IS_ERR(shm_mnt)) | |
6b4d0b27 | 4177 | return ERR_CAST(shm_mnt); |
1da177e4 | 4178 | |
285b2c4f | 4179 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4180 | return ERR_PTR(-EINVAL); |
4181 | ||
4182 | if (shmem_acct_size(flags, size)) | |
4183 | return ERR_PTR(-ENOMEM); | |
4184 | ||
6b4d0b27 | 4185 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
4186 | this.name = name; |
4187 | this.len = strlen(name); | |
4188 | this.hash = 0; /* will go */ | |
3451538a | 4189 | sb = shm_mnt->mnt_sb; |
66ee4b88 | 4190 | path.mnt = mntget(shm_mnt); |
3451538a | 4191 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 4192 | if (!path.dentry) |
1da177e4 | 4193 | goto put_memory; |
3451538a | 4194 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 4195 | |
6b4d0b27 | 4196 | res = ERR_PTR(-ENOSPC); |
3451538a | 4197 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 4198 | if (!inode) |
66ee4b88 | 4199 | goto put_memory; |
1da177e4 | 4200 | |
c7277090 | 4201 | inode->i_flags |= i_flags; |
2c48b9c4 | 4202 | d_instantiate(path.dentry, inode); |
1da177e4 | 4203 | inode->i_size = size; |
6d6b77f1 | 4204 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
4205 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
4206 | if (IS_ERR(res)) | |
66ee4b88 | 4207 | goto put_path; |
4b42af81 | 4208 | |
6b4d0b27 | 4209 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 4210 | &shmem_file_operations); |
6b4d0b27 | 4211 | if (IS_ERR(res)) |
66ee4b88 | 4212 | goto put_path; |
4b42af81 | 4213 | |
6b4d0b27 | 4214 | return res; |
1da177e4 | 4215 | |
1da177e4 LT |
4216 | put_memory: |
4217 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
4218 | put_path: |
4219 | path_put(&path); | |
6b4d0b27 | 4220 | return res; |
1da177e4 | 4221 | } |
c7277090 EP |
4222 | |
4223 | /** | |
4224 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4225 | * kernel internal. There will be NO LSM permission checks against the | |
4226 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4227 | * higher layer. The users are the big_key and shm implementations. LSM |
4228 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4229 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4230 | * @size: size to be set for the file | |
4231 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4232 | */ | |
4233 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4234 | { | |
4235 | return __shmem_file_setup(name, size, flags, S_PRIVATE); | |
4236 | } | |
4237 | ||
4238 | /** | |
4239 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4240 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4241 | * @size: size to be set for the file | |
4242 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4243 | */ | |
4244 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4245 | { | |
4246 | return __shmem_file_setup(name, size, flags, 0); | |
4247 | } | |
395e0ddc | 4248 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4249 | |
46711810 | 4250 | /** |
1da177e4 | 4251 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
4252 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
4253 | */ | |
4254 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4255 | { | |
4256 | struct file *file; | |
4257 | loff_t size = vma->vm_end - vma->vm_start; | |
4258 | ||
66fc1303 HD |
4259 | /* |
4260 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
4261 | * between XFS directory reading and selinux: since this file is only | |
4262 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4263 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4264 | */ | |
4265 | file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE); | |
1da177e4 LT |
4266 | if (IS_ERR(file)) |
4267 | return PTR_ERR(file); | |
4268 | ||
4269 | if (vma->vm_file) | |
4270 | fput(vma->vm_file); | |
4271 | vma->vm_file = file; | |
4272 | vma->vm_ops = &shmem_vm_ops; | |
f3f0e1d2 | 4273 | |
e496cf3d | 4274 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
4275 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
4276 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
4277 | khugepaged_enter(vma, vma->vm_flags); | |
4278 | } | |
4279 | ||
1da177e4 LT |
4280 | return 0; |
4281 | } | |
d9d90e5e HD |
4282 | |
4283 | /** | |
4284 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
4285 | * @mapping: the page's address_space | |
4286 | * @index: the page index | |
4287 | * @gfp: the page allocator flags to use if allocating | |
4288 | * | |
4289 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4290 | * with any new page allocations done using the specified allocation flags. | |
4291 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
4292 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
4293 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4294 | * | |
68da9f05 HD |
4295 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4296 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4297 | */ |
4298 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4299 | pgoff_t index, gfp_t gfp) | |
4300 | { | |
68da9f05 HD |
4301 | #ifdef CONFIG_SHMEM |
4302 | struct inode *inode = mapping->host; | |
9276aad6 | 4303 | struct page *page; |
68da9f05 HD |
4304 | int error; |
4305 | ||
4306 | BUG_ON(mapping->a_ops != &shmem_aops); | |
9e18eb29 | 4307 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
cfda0526 | 4308 | gfp, NULL, NULL, NULL); |
68da9f05 HD |
4309 | if (error) |
4310 | page = ERR_PTR(error); | |
4311 | else | |
4312 | unlock_page(page); | |
4313 | return page; | |
4314 | #else | |
4315 | /* | |
4316 | * The tiny !SHMEM case uses ramfs without swap | |
4317 | */ | |
d9d90e5e | 4318 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4319 | #endif |
d9d90e5e HD |
4320 | } |
4321 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |