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> | |
b95f1b31 | 32 | #include <linux/export.h> |
853ac43a | 33 | #include <linux/swap.h> |
e2e40f2c | 34 | #include <linux/uio.h> |
f3f0e1d2 | 35 | #include <linux/khugepaged.h> |
853ac43a | 36 | |
95cc09d6 AA |
37 | #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */ |
38 | ||
853ac43a MM |
39 | static struct vfsmount *shm_mnt; |
40 | ||
41 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
42 | /* |
43 | * This virtual memory filesystem is heavily based on the ramfs. It | |
44 | * extends ramfs by the ability to use swap and honor resource limits | |
45 | * which makes it a completely usable filesystem. | |
46 | */ | |
47 | ||
39f0247d | 48 | #include <linux/xattr.h> |
a5694255 | 49 | #include <linux/exportfs.h> |
1c7c474c | 50 | #include <linux/posix_acl.h> |
feda821e | 51 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 52 | #include <linux/mman.h> |
1da177e4 LT |
53 | #include <linux/string.h> |
54 | #include <linux/slab.h> | |
55 | #include <linux/backing-dev.h> | |
56 | #include <linux/shmem_fs.h> | |
1da177e4 | 57 | #include <linux/writeback.h> |
1da177e4 | 58 | #include <linux/blkdev.h> |
bda97eab | 59 | #include <linux/pagevec.h> |
41ffe5d5 | 60 | #include <linux/percpu_counter.h> |
83e4fa9c | 61 | #include <linux/falloc.h> |
708e3508 | 62 | #include <linux/splice.h> |
1da177e4 LT |
63 | #include <linux/security.h> |
64 | #include <linux/swapops.h> | |
65 | #include <linux/mempolicy.h> | |
66 | #include <linux/namei.h> | |
b00dc3ad | 67 | #include <linux/ctype.h> |
304dbdb7 | 68 | #include <linux/migrate.h> |
c1f60a5a | 69 | #include <linux/highmem.h> |
680d794b | 70 | #include <linux/seq_file.h> |
92562927 | 71 | #include <linux/magic.h> |
9183df25 | 72 | #include <linux/syscalls.h> |
40e041a2 | 73 | #include <linux/fcntl.h> |
9183df25 | 74 | #include <uapi/linux/memfd.h> |
cfda0526 | 75 | #include <linux/userfaultfd_k.h> |
4c27fe4c | 76 | #include <linux/rmap.h> |
304dbdb7 | 77 | |
7c0f6ba6 | 78 | #include <linux/uaccess.h> |
1da177e4 LT |
79 | #include <asm/pgtable.h> |
80 | ||
dd56b046 MG |
81 | #include "internal.h" |
82 | ||
09cbfeaf KS |
83 | #define BLOCKS_PER_PAGE (PAGE_SIZE/512) |
84 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT) | |
1da177e4 | 85 | |
1da177e4 LT |
86 | /* Pretend that each entry is of this size in directory's i_size */ |
87 | #define BOGO_DIRENT_SIZE 20 | |
88 | ||
69f07ec9 HD |
89 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
90 | #define SHORT_SYMLINK_LEN 128 | |
91 | ||
1aac1400 | 92 | /* |
f00cdc6d HD |
93 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
94 | * inode->i_private (with i_mutex making sure that it has only one user at | |
95 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
96 | */ |
97 | struct shmem_falloc { | |
8e205f77 | 98 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
99 | pgoff_t start; /* start of range currently being fallocated */ |
100 | pgoff_t next; /* the next page offset to be fallocated */ | |
101 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
102 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
103 | }; | |
104 | ||
b76db735 | 105 | #ifdef CONFIG_TMPFS |
680d794b | 106 | static unsigned long shmem_default_max_blocks(void) |
107 | { | |
108 | return totalram_pages / 2; | |
109 | } | |
110 | ||
111 | static unsigned long shmem_default_max_inodes(void) | |
112 | { | |
113 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
114 | } | |
b76db735 | 115 | #endif |
680d794b | 116 | |
bde05d1c HD |
117 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
118 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
119 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 | 120 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 121 | struct page **pagep, enum sgp_type sgp, |
cfda0526 MR |
122 | gfp_t gfp, struct vm_area_struct *vma, |
123 | struct vm_fault *vmf, int *fault_type); | |
68da9f05 | 124 | |
f3f0e1d2 | 125 | int shmem_getpage(struct inode *inode, pgoff_t index, |
9e18eb29 | 126 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
127 | { |
128 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
cfda0526 | 129 | mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL); |
68da9f05 | 130 | } |
1da177e4 | 131 | |
1da177e4 LT |
132 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
133 | { | |
134 | return sb->s_fs_info; | |
135 | } | |
136 | ||
137 | /* | |
138 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
139 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
140 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
141 | * consistent with the pre-accounting of private mappings ... | |
142 | */ | |
143 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
144 | { | |
0b0a0806 | 145 | return (flags & VM_NORESERVE) ? |
191c5424 | 146 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
147 | } |
148 | ||
149 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
150 | { | |
0b0a0806 | 151 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
152 | vm_unacct_memory(VM_ACCT(size)); |
153 | } | |
154 | ||
77142517 KK |
155 | static inline int shmem_reacct_size(unsigned long flags, |
156 | loff_t oldsize, loff_t newsize) | |
157 | { | |
158 | if (!(flags & VM_NORESERVE)) { | |
159 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
160 | return security_vm_enough_memory_mm(current->mm, | |
161 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
162 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
163 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
164 | } | |
165 | return 0; | |
166 | } | |
167 | ||
1da177e4 LT |
168 | /* |
169 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 170 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
171 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
172 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
173 | */ | |
800d8c63 | 174 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 175 | { |
800d8c63 KS |
176 | if (!(flags & VM_NORESERVE)) |
177 | return 0; | |
178 | ||
179 | return security_vm_enough_memory_mm(current->mm, | |
180 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
181 | } |
182 | ||
183 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
184 | { | |
0b0a0806 | 185 | if (flags & VM_NORESERVE) |
09cbfeaf | 186 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
187 | } |
188 | ||
759b9775 | 189 | static const struct super_operations shmem_ops; |
f5e54d6e | 190 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 191 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
192 | static const struct inode_operations shmem_inode_operations; |
193 | static const struct inode_operations shmem_dir_inode_operations; | |
194 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 195 | static const struct vm_operations_struct shmem_vm_ops; |
779750d2 | 196 | static struct file_system_type shmem_fs_type; |
1da177e4 | 197 | |
b0506e48 MR |
198 | bool vma_is_shmem(struct vm_area_struct *vma) |
199 | { | |
200 | return vma->vm_ops == &shmem_vm_ops; | |
201 | } | |
202 | ||
1da177e4 | 203 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 204 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 205 | |
5b04c689 PE |
206 | static int shmem_reserve_inode(struct super_block *sb) |
207 | { | |
208 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
209 | if (sbinfo->max_inodes) { | |
210 | spin_lock(&sbinfo->stat_lock); | |
211 | if (!sbinfo->free_inodes) { | |
212 | spin_unlock(&sbinfo->stat_lock); | |
213 | return -ENOSPC; | |
214 | } | |
215 | sbinfo->free_inodes--; | |
216 | spin_unlock(&sbinfo->stat_lock); | |
217 | } | |
218 | return 0; | |
219 | } | |
220 | ||
221 | static void shmem_free_inode(struct super_block *sb) | |
222 | { | |
223 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
224 | if (sbinfo->max_inodes) { | |
225 | spin_lock(&sbinfo->stat_lock); | |
226 | sbinfo->free_inodes++; | |
227 | spin_unlock(&sbinfo->stat_lock); | |
228 | } | |
229 | } | |
230 | ||
46711810 | 231 | /** |
41ffe5d5 | 232 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
233 | * @inode: inode to recalc |
234 | * | |
235 | * We have to calculate the free blocks since the mm can drop | |
236 | * undirtied hole pages behind our back. | |
237 | * | |
238 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
239 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
240 | * | |
241 | * It has to be called with the spinlock held. | |
242 | */ | |
243 | static void shmem_recalc_inode(struct inode *inode) | |
244 | { | |
245 | struct shmem_inode_info *info = SHMEM_I(inode); | |
246 | long freed; | |
247 | ||
248 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
249 | if (freed > 0) { | |
54af6042 HD |
250 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
251 | if (sbinfo->max_blocks) | |
252 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 253 | info->alloced -= freed; |
54af6042 | 254 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 255 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
256 | } |
257 | } | |
258 | ||
800d8c63 KS |
259 | bool shmem_charge(struct inode *inode, long pages) |
260 | { | |
261 | struct shmem_inode_info *info = SHMEM_I(inode); | |
262 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4595ef88 | 263 | unsigned long flags; |
800d8c63 KS |
264 | |
265 | if (shmem_acct_block(info->flags, pages)) | |
266 | return false; | |
4595ef88 | 267 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
268 | info->alloced += pages; |
269 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
270 | shmem_recalc_inode(inode); | |
4595ef88 | 271 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
272 | inode->i_mapping->nrpages += pages; |
273 | ||
274 | if (!sbinfo->max_blocks) | |
275 | return true; | |
276 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
277 | sbinfo->max_blocks - pages) > 0) { | |
278 | inode->i_mapping->nrpages -= pages; | |
4595ef88 | 279 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
280 | info->alloced -= pages; |
281 | shmem_recalc_inode(inode); | |
4595ef88 | 282 | spin_unlock_irqrestore(&info->lock, flags); |
71664665 | 283 | shmem_unacct_blocks(info->flags, pages); |
800d8c63 KS |
284 | return false; |
285 | } | |
286 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
287 | return true; | |
288 | } | |
289 | ||
290 | void shmem_uncharge(struct inode *inode, long pages) | |
291 | { | |
292 | struct shmem_inode_info *info = SHMEM_I(inode); | |
293 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4595ef88 | 294 | unsigned long flags; |
800d8c63 | 295 | |
4595ef88 | 296 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
297 | info->alloced -= pages; |
298 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
299 | shmem_recalc_inode(inode); | |
4595ef88 | 300 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
301 | |
302 | if (sbinfo->max_blocks) | |
303 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
71664665 | 304 | shmem_unacct_blocks(info->flags, pages); |
800d8c63 KS |
305 | } |
306 | ||
7a5d0fbb HD |
307 | /* |
308 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
309 | */ | |
310 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
311 | pgoff_t index, void *expected, void *replacement) | |
312 | { | |
f7942430 | 313 | struct radix_tree_node *node; |
7a5d0fbb | 314 | void **pslot; |
6dbaf22c | 315 | void *item; |
7a5d0fbb HD |
316 | |
317 | VM_BUG_ON(!expected); | |
6dbaf22c | 318 | VM_BUG_ON(!replacement); |
f7942430 JW |
319 | item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot); |
320 | if (!item) | |
6dbaf22c | 321 | return -ENOENT; |
7a5d0fbb HD |
322 | if (item != expected) |
323 | return -ENOENT; | |
4d693d08 JW |
324 | __radix_tree_replace(&mapping->page_tree, node, pslot, |
325 | replacement, NULL, NULL); | |
7a5d0fbb HD |
326 | return 0; |
327 | } | |
328 | ||
d1899228 HD |
329 | /* |
330 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
331 | * that an entry was not already brought back from swap by a racing thread. | |
332 | * | |
333 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
334 | * might be reused, and again be SwapCache, using the same swap as before. | |
335 | */ | |
336 | static bool shmem_confirm_swap(struct address_space *mapping, | |
337 | pgoff_t index, swp_entry_t swap) | |
338 | { | |
339 | void *item; | |
340 | ||
341 | rcu_read_lock(); | |
342 | item = radix_tree_lookup(&mapping->page_tree, index); | |
343 | rcu_read_unlock(); | |
344 | return item == swp_to_radix_entry(swap); | |
345 | } | |
346 | ||
5a6e75f8 KS |
347 | /* |
348 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
349 | * | |
350 | * SHMEM_HUGE_NEVER: | |
351 | * disables huge pages for the mount; | |
352 | * SHMEM_HUGE_ALWAYS: | |
353 | * enables huge pages for the mount; | |
354 | * SHMEM_HUGE_WITHIN_SIZE: | |
355 | * only allocate huge pages if the page will be fully within i_size, | |
356 | * also respect fadvise()/madvise() hints; | |
357 | * SHMEM_HUGE_ADVISE: | |
358 | * only allocate huge pages if requested with fadvise()/madvise(); | |
359 | */ | |
360 | ||
361 | #define SHMEM_HUGE_NEVER 0 | |
362 | #define SHMEM_HUGE_ALWAYS 1 | |
363 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
364 | #define SHMEM_HUGE_ADVISE 3 | |
365 | ||
366 | /* | |
367 | * Special values. | |
368 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
369 | * | |
370 | * SHMEM_HUGE_DENY: | |
371 | * disables huge on shm_mnt and all mounts, for emergency use; | |
372 | * SHMEM_HUGE_FORCE: | |
373 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
374 | * | |
375 | */ | |
376 | #define SHMEM_HUGE_DENY (-1) | |
377 | #define SHMEM_HUGE_FORCE (-2) | |
378 | ||
e496cf3d | 379 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
380 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
381 | ||
382 | int shmem_huge __read_mostly; | |
383 | ||
f1f5929c | 384 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
385 | static int shmem_parse_huge(const char *str) |
386 | { | |
387 | if (!strcmp(str, "never")) | |
388 | return SHMEM_HUGE_NEVER; | |
389 | if (!strcmp(str, "always")) | |
390 | return SHMEM_HUGE_ALWAYS; | |
391 | if (!strcmp(str, "within_size")) | |
392 | return SHMEM_HUGE_WITHIN_SIZE; | |
393 | if (!strcmp(str, "advise")) | |
394 | return SHMEM_HUGE_ADVISE; | |
395 | if (!strcmp(str, "deny")) | |
396 | return SHMEM_HUGE_DENY; | |
397 | if (!strcmp(str, "force")) | |
398 | return SHMEM_HUGE_FORCE; | |
399 | return -EINVAL; | |
400 | } | |
401 | ||
402 | static const char *shmem_format_huge(int huge) | |
403 | { | |
404 | switch (huge) { | |
405 | case SHMEM_HUGE_NEVER: | |
406 | return "never"; | |
407 | case SHMEM_HUGE_ALWAYS: | |
408 | return "always"; | |
409 | case SHMEM_HUGE_WITHIN_SIZE: | |
410 | return "within_size"; | |
411 | case SHMEM_HUGE_ADVISE: | |
412 | return "advise"; | |
413 | case SHMEM_HUGE_DENY: | |
414 | return "deny"; | |
415 | case SHMEM_HUGE_FORCE: | |
416 | return "force"; | |
417 | default: | |
418 | VM_BUG_ON(1); | |
419 | return "bad_val"; | |
420 | } | |
421 | } | |
f1f5929c | 422 | #endif |
5a6e75f8 | 423 | |
779750d2 KS |
424 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
425 | struct shrink_control *sc, unsigned long nr_to_split) | |
426 | { | |
427 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 428 | LIST_HEAD(to_remove); |
779750d2 KS |
429 | struct inode *inode; |
430 | struct shmem_inode_info *info; | |
431 | struct page *page; | |
432 | unsigned long batch = sc ? sc->nr_to_scan : 128; | |
433 | int removed = 0, split = 0; | |
434 | ||
435 | if (list_empty(&sbinfo->shrinklist)) | |
436 | return SHRINK_STOP; | |
437 | ||
438 | spin_lock(&sbinfo->shrinklist_lock); | |
439 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
440 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
441 | ||
442 | /* pin the inode */ | |
443 | inode = igrab(&info->vfs_inode); | |
444 | ||
445 | /* inode is about to be evicted */ | |
446 | if (!inode) { | |
447 | list_del_init(&info->shrinklist); | |
448 | removed++; | |
449 | goto next; | |
450 | } | |
451 | ||
452 | /* Check if there's anything to gain */ | |
453 | if (round_up(inode->i_size, PAGE_SIZE) == | |
454 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 455 | list_move(&info->shrinklist, &to_remove); |
779750d2 | 456 | removed++; |
779750d2 KS |
457 | goto next; |
458 | } | |
459 | ||
460 | list_move(&info->shrinklist, &list); | |
461 | next: | |
462 | if (!--batch) | |
463 | break; | |
464 | } | |
465 | spin_unlock(&sbinfo->shrinklist_lock); | |
466 | ||
253fd0f0 KS |
467 | list_for_each_safe(pos, next, &to_remove) { |
468 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
469 | inode = &info->vfs_inode; | |
470 | list_del_init(&info->shrinklist); | |
471 | iput(inode); | |
472 | } | |
473 | ||
779750d2 KS |
474 | list_for_each_safe(pos, next, &list) { |
475 | int ret; | |
476 | ||
477 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
478 | inode = &info->vfs_inode; | |
479 | ||
480 | if (nr_to_split && split >= nr_to_split) { | |
481 | iput(inode); | |
482 | continue; | |
483 | } | |
484 | ||
485 | page = find_lock_page(inode->i_mapping, | |
486 | (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT); | |
487 | if (!page) | |
488 | goto drop; | |
489 | ||
490 | if (!PageTransHuge(page)) { | |
491 | unlock_page(page); | |
492 | put_page(page); | |
493 | goto drop; | |
494 | } | |
495 | ||
496 | ret = split_huge_page(page); | |
497 | unlock_page(page); | |
498 | put_page(page); | |
499 | ||
500 | if (ret) { | |
501 | /* split failed: leave it on the list */ | |
502 | iput(inode); | |
503 | continue; | |
504 | } | |
505 | ||
506 | split++; | |
507 | drop: | |
508 | list_del_init(&info->shrinklist); | |
509 | removed++; | |
510 | iput(inode); | |
511 | } | |
512 | ||
513 | spin_lock(&sbinfo->shrinklist_lock); | |
514 | list_splice_tail(&list, &sbinfo->shrinklist); | |
515 | sbinfo->shrinklist_len -= removed; | |
516 | spin_unlock(&sbinfo->shrinklist_lock); | |
517 | ||
518 | return split; | |
519 | } | |
520 | ||
521 | static long shmem_unused_huge_scan(struct super_block *sb, | |
522 | struct shrink_control *sc) | |
523 | { | |
524 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
525 | ||
526 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
527 | return SHRINK_STOP; | |
528 | ||
529 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
530 | } | |
531 | ||
532 | static long shmem_unused_huge_count(struct super_block *sb, | |
533 | struct shrink_control *sc) | |
534 | { | |
535 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
536 | return READ_ONCE(sbinfo->shrinklist_len); | |
537 | } | |
e496cf3d | 538 | #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 KS |
539 | |
540 | #define shmem_huge SHMEM_HUGE_DENY | |
541 | ||
779750d2 KS |
542 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
543 | struct shrink_control *sc, unsigned long nr_to_split) | |
544 | { | |
545 | return 0; | |
546 | } | |
e496cf3d | 547 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 548 | |
46f65ec1 HD |
549 | /* |
550 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
551 | */ | |
552 | static int shmem_add_to_page_cache(struct page *page, | |
553 | struct address_space *mapping, | |
fed400a1 | 554 | pgoff_t index, void *expected) |
46f65ec1 | 555 | { |
800d8c63 | 556 | int error, nr = hpage_nr_pages(page); |
46f65ec1 | 557 | |
800d8c63 KS |
558 | VM_BUG_ON_PAGE(PageTail(page), page); |
559 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
560 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
561 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 562 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 563 | |
800d8c63 | 564 | page_ref_add(page, nr); |
b065b432 HD |
565 | page->mapping = mapping; |
566 | page->index = index; | |
567 | ||
568 | spin_lock_irq(&mapping->tree_lock); | |
800d8c63 KS |
569 | if (PageTransHuge(page)) { |
570 | void __rcu **results; | |
571 | pgoff_t idx; | |
572 | int i; | |
573 | ||
574 | error = 0; | |
575 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, | |
576 | &results, &idx, index, 1) && | |
577 | idx < index + HPAGE_PMD_NR) { | |
578 | error = -EEXIST; | |
579 | } | |
580 | ||
581 | if (!error) { | |
582 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
583 | error = radix_tree_insert(&mapping->page_tree, | |
584 | index + i, page + i); | |
585 | VM_BUG_ON(error); | |
586 | } | |
587 | count_vm_event(THP_FILE_ALLOC); | |
588 | } | |
589 | } else if (!expected) { | |
b065b432 | 590 | error = radix_tree_insert(&mapping->page_tree, index, page); |
800d8c63 | 591 | } else { |
b065b432 HD |
592 | error = shmem_radix_tree_replace(mapping, index, expected, |
593 | page); | |
800d8c63 KS |
594 | } |
595 | ||
46f65ec1 | 596 | if (!error) { |
800d8c63 KS |
597 | mapping->nrpages += nr; |
598 | if (PageTransHuge(page)) | |
11fb9989 MG |
599 | __inc_node_page_state(page, NR_SHMEM_THPS); |
600 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); | |
601 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr); | |
b065b432 HD |
602 | spin_unlock_irq(&mapping->tree_lock); |
603 | } else { | |
604 | page->mapping = NULL; | |
605 | spin_unlock_irq(&mapping->tree_lock); | |
800d8c63 | 606 | page_ref_sub(page, nr); |
46f65ec1 | 607 | } |
46f65ec1 HD |
608 | return error; |
609 | } | |
610 | ||
6922c0c7 HD |
611 | /* |
612 | * Like delete_from_page_cache, but substitutes swap for page. | |
613 | */ | |
614 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
615 | { | |
616 | struct address_space *mapping = page->mapping; | |
617 | int error; | |
618 | ||
800d8c63 KS |
619 | VM_BUG_ON_PAGE(PageCompound(page), page); |
620 | ||
6922c0c7 HD |
621 | spin_lock_irq(&mapping->tree_lock); |
622 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
623 | page->mapping = NULL; | |
624 | mapping->nrpages--; | |
11fb9989 MG |
625 | __dec_node_page_state(page, NR_FILE_PAGES); |
626 | __dec_node_page_state(page, NR_SHMEM); | |
6922c0c7 | 627 | spin_unlock_irq(&mapping->tree_lock); |
09cbfeaf | 628 | put_page(page); |
6922c0c7 HD |
629 | BUG_ON(error); |
630 | } | |
631 | ||
7a5d0fbb HD |
632 | /* |
633 | * Remove swap entry from radix tree, free the swap and its page cache. | |
634 | */ | |
635 | static int shmem_free_swap(struct address_space *mapping, | |
636 | pgoff_t index, void *radswap) | |
637 | { | |
6dbaf22c | 638 | void *old; |
7a5d0fbb HD |
639 | |
640 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 641 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 642 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
643 | if (old != radswap) |
644 | return -ENOENT; | |
645 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
646 | return 0; | |
7a5d0fbb HD |
647 | } |
648 | ||
6a15a370 VB |
649 | /* |
650 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 651 | * given offsets are swapped out. |
6a15a370 VB |
652 | * |
653 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
654 | * as long as the inode doesn't go away and racy results are not a problem. | |
655 | */ | |
48131e03 VB |
656 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
657 | pgoff_t start, pgoff_t end) | |
6a15a370 | 658 | { |
6a15a370 VB |
659 | struct radix_tree_iter iter; |
660 | void **slot; | |
661 | struct page *page; | |
48131e03 | 662 | unsigned long swapped = 0; |
6a15a370 VB |
663 | |
664 | rcu_read_lock(); | |
665 | ||
6a15a370 VB |
666 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
667 | if (iter.index >= end) | |
668 | break; | |
669 | ||
670 | page = radix_tree_deref_slot(slot); | |
671 | ||
2cf938aa MW |
672 | if (radix_tree_deref_retry(page)) { |
673 | slot = radix_tree_iter_retry(&iter); | |
674 | continue; | |
675 | } | |
6a15a370 VB |
676 | |
677 | if (radix_tree_exceptional_entry(page)) | |
678 | swapped++; | |
679 | ||
680 | if (need_resched()) { | |
148deab2 | 681 | slot = radix_tree_iter_resume(slot, &iter); |
6a15a370 | 682 | cond_resched_rcu(); |
6a15a370 VB |
683 | } |
684 | } | |
685 | ||
686 | rcu_read_unlock(); | |
687 | ||
688 | return swapped << PAGE_SHIFT; | |
689 | } | |
690 | ||
48131e03 VB |
691 | /* |
692 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
693 | * given vma is swapped out. | |
694 | * | |
695 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
696 | * as long as the inode doesn't go away and racy results are not a problem. | |
697 | */ | |
698 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
699 | { | |
700 | struct inode *inode = file_inode(vma->vm_file); | |
701 | struct shmem_inode_info *info = SHMEM_I(inode); | |
702 | struct address_space *mapping = inode->i_mapping; | |
703 | unsigned long swapped; | |
704 | ||
705 | /* Be careful as we don't hold info->lock */ | |
706 | swapped = READ_ONCE(info->swapped); | |
707 | ||
708 | /* | |
709 | * The easier cases are when the shmem object has nothing in swap, or | |
710 | * the vma maps it whole. Then we can simply use the stats that we | |
711 | * already track. | |
712 | */ | |
713 | if (!swapped) | |
714 | return 0; | |
715 | ||
716 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
717 | return swapped << PAGE_SHIFT; | |
718 | ||
719 | /* Here comes the more involved part */ | |
720 | return shmem_partial_swap_usage(mapping, | |
721 | linear_page_index(vma, vma->vm_start), | |
722 | linear_page_index(vma, vma->vm_end)); | |
723 | } | |
724 | ||
24513264 HD |
725 | /* |
726 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
727 | */ | |
728 | void shmem_unlock_mapping(struct address_space *mapping) | |
729 | { | |
730 | struct pagevec pvec; | |
731 | pgoff_t indices[PAGEVEC_SIZE]; | |
732 | pgoff_t index = 0; | |
733 | ||
734 | pagevec_init(&pvec, 0); | |
735 | /* | |
736 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
737 | */ | |
738 | while (!mapping_unevictable(mapping)) { | |
739 | /* | |
740 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
741 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
742 | */ | |
0cd6144a JW |
743 | pvec.nr = find_get_entries(mapping, index, |
744 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
745 | if (!pvec.nr) |
746 | break; | |
747 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 748 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
749 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
750 | pagevec_release(&pvec); | |
751 | cond_resched(); | |
752 | } | |
7a5d0fbb HD |
753 | } |
754 | ||
755 | /* | |
756 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 757 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 758 | */ |
1635f6a7 HD |
759 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
760 | bool unfalloc) | |
1da177e4 | 761 | { |
285b2c4f | 762 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 763 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
764 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
765 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
766 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
767 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 768 | struct pagevec pvec; |
7a5d0fbb HD |
769 | pgoff_t indices[PAGEVEC_SIZE]; |
770 | long nr_swaps_freed = 0; | |
285b2c4f | 771 | pgoff_t index; |
bda97eab HD |
772 | int i; |
773 | ||
83e4fa9c HD |
774 | if (lend == -1) |
775 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
776 | |
777 | pagevec_init(&pvec, 0); | |
778 | index = start; | |
83e4fa9c | 779 | while (index < end) { |
0cd6144a JW |
780 | pvec.nr = find_get_entries(mapping, index, |
781 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
782 | pvec.pages, indices); | |
7a5d0fbb HD |
783 | if (!pvec.nr) |
784 | break; | |
bda97eab HD |
785 | for (i = 0; i < pagevec_count(&pvec); i++) { |
786 | struct page *page = pvec.pages[i]; | |
787 | ||
7a5d0fbb | 788 | index = indices[i]; |
83e4fa9c | 789 | if (index >= end) |
bda97eab HD |
790 | break; |
791 | ||
7a5d0fbb | 792 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
793 | if (unfalloc) |
794 | continue; | |
7a5d0fbb HD |
795 | nr_swaps_freed += !shmem_free_swap(mapping, |
796 | index, page); | |
bda97eab | 797 | continue; |
7a5d0fbb HD |
798 | } |
799 | ||
800d8c63 KS |
800 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
801 | ||
7a5d0fbb | 802 | if (!trylock_page(page)) |
bda97eab | 803 | continue; |
800d8c63 KS |
804 | |
805 | if (PageTransTail(page)) { | |
806 | /* Middle of THP: zero out the page */ | |
807 | clear_highpage(page); | |
808 | unlock_page(page); | |
809 | continue; | |
810 | } else if (PageTransHuge(page)) { | |
811 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
812 | /* | |
813 | * Range ends in the middle of THP: | |
814 | * zero out the page | |
815 | */ | |
816 | clear_highpage(page); | |
817 | unlock_page(page); | |
818 | continue; | |
819 | } | |
820 | index += HPAGE_PMD_NR - 1; | |
821 | i += HPAGE_PMD_NR - 1; | |
822 | } | |
823 | ||
1635f6a7 | 824 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
825 | VM_BUG_ON_PAGE(PageTail(page), page); |
826 | if (page_mapping(page) == mapping) { | |
309381fe | 827 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
828 | truncate_inode_page(mapping, page); |
829 | } | |
bda97eab | 830 | } |
bda97eab HD |
831 | unlock_page(page); |
832 | } | |
0cd6144a | 833 | pagevec_remove_exceptionals(&pvec); |
24513264 | 834 | pagevec_release(&pvec); |
bda97eab HD |
835 | cond_resched(); |
836 | index++; | |
837 | } | |
1da177e4 | 838 | |
83e4fa9c | 839 | if (partial_start) { |
bda97eab | 840 | struct page *page = NULL; |
9e18eb29 | 841 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 842 | if (page) { |
09cbfeaf | 843 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
844 | if (start > end) { |
845 | top = partial_end; | |
846 | partial_end = 0; | |
847 | } | |
848 | zero_user_segment(page, partial_start, top); | |
849 | set_page_dirty(page); | |
850 | unlock_page(page); | |
09cbfeaf | 851 | put_page(page); |
83e4fa9c HD |
852 | } |
853 | } | |
854 | if (partial_end) { | |
855 | struct page *page = NULL; | |
9e18eb29 | 856 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
857 | if (page) { |
858 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
859 | set_page_dirty(page); |
860 | unlock_page(page); | |
09cbfeaf | 861 | put_page(page); |
bda97eab HD |
862 | } |
863 | } | |
83e4fa9c HD |
864 | if (start >= end) |
865 | return; | |
bda97eab HD |
866 | |
867 | index = start; | |
b1a36650 | 868 | while (index < end) { |
bda97eab | 869 | cond_resched(); |
0cd6144a JW |
870 | |
871 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 872 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 873 | pvec.pages, indices); |
7a5d0fbb | 874 | if (!pvec.nr) { |
b1a36650 HD |
875 | /* If all gone or hole-punch or unfalloc, we're done */ |
876 | if (index == start || end != -1) | |
bda97eab | 877 | break; |
b1a36650 | 878 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
879 | index = start; |
880 | continue; | |
881 | } | |
bda97eab HD |
882 | for (i = 0; i < pagevec_count(&pvec); i++) { |
883 | struct page *page = pvec.pages[i]; | |
884 | ||
7a5d0fbb | 885 | index = indices[i]; |
83e4fa9c | 886 | if (index >= end) |
bda97eab HD |
887 | break; |
888 | ||
7a5d0fbb | 889 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
890 | if (unfalloc) |
891 | continue; | |
b1a36650 HD |
892 | if (shmem_free_swap(mapping, index, page)) { |
893 | /* Swap was replaced by page: retry */ | |
894 | index--; | |
895 | break; | |
896 | } | |
897 | nr_swaps_freed++; | |
7a5d0fbb HD |
898 | continue; |
899 | } | |
900 | ||
bda97eab | 901 | lock_page(page); |
800d8c63 KS |
902 | |
903 | if (PageTransTail(page)) { | |
904 | /* Middle of THP: zero out the page */ | |
905 | clear_highpage(page); | |
906 | unlock_page(page); | |
907 | /* | |
908 | * Partial thp truncate due 'start' in middle | |
909 | * of THP: don't need to look on these pages | |
910 | * again on !pvec.nr restart. | |
911 | */ | |
912 | if (index != round_down(end, HPAGE_PMD_NR)) | |
913 | start++; | |
914 | continue; | |
915 | } else if (PageTransHuge(page)) { | |
916 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
917 | /* | |
918 | * Range ends in the middle of THP: | |
919 | * zero out the page | |
920 | */ | |
921 | clear_highpage(page); | |
922 | unlock_page(page); | |
923 | continue; | |
924 | } | |
925 | index += HPAGE_PMD_NR - 1; | |
926 | i += HPAGE_PMD_NR - 1; | |
927 | } | |
928 | ||
1635f6a7 | 929 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
930 | VM_BUG_ON_PAGE(PageTail(page), page); |
931 | if (page_mapping(page) == mapping) { | |
309381fe | 932 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 933 | truncate_inode_page(mapping, page); |
b1a36650 HD |
934 | } else { |
935 | /* Page was replaced by swap: retry */ | |
936 | unlock_page(page); | |
937 | index--; | |
938 | break; | |
1635f6a7 | 939 | } |
7a5d0fbb | 940 | } |
bda97eab HD |
941 | unlock_page(page); |
942 | } | |
0cd6144a | 943 | pagevec_remove_exceptionals(&pvec); |
24513264 | 944 | pagevec_release(&pvec); |
bda97eab HD |
945 | index++; |
946 | } | |
94c1e62d | 947 | |
4595ef88 | 948 | spin_lock_irq(&info->lock); |
7a5d0fbb | 949 | info->swapped -= nr_swaps_freed; |
1da177e4 | 950 | shmem_recalc_inode(inode); |
4595ef88 | 951 | spin_unlock_irq(&info->lock); |
1635f6a7 | 952 | } |
1da177e4 | 953 | |
1635f6a7 HD |
954 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
955 | { | |
956 | shmem_undo_range(inode, lstart, lend, false); | |
078cd827 | 957 | inode->i_ctime = inode->i_mtime = current_time(inode); |
1da177e4 | 958 | } |
94c1e62d | 959 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 960 | |
44a30220 YZ |
961 | static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry, |
962 | struct kstat *stat) | |
963 | { | |
964 | struct inode *inode = dentry->d_inode; | |
965 | struct shmem_inode_info *info = SHMEM_I(inode); | |
966 | ||
d0424c42 | 967 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
4595ef88 | 968 | spin_lock_irq(&info->lock); |
d0424c42 | 969 | shmem_recalc_inode(inode); |
4595ef88 | 970 | spin_unlock_irq(&info->lock); |
d0424c42 | 971 | } |
44a30220 | 972 | generic_fillattr(inode, stat); |
44a30220 YZ |
973 | return 0; |
974 | } | |
975 | ||
94c1e62d | 976 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 977 | { |
75c3cfa8 | 978 | struct inode *inode = d_inode(dentry); |
40e041a2 | 979 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 980 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 LT |
981 | int error; |
982 | ||
31051c85 | 983 | error = setattr_prepare(dentry, attr); |
db78b877 CH |
984 | if (error) |
985 | return error; | |
986 | ||
94c1e62d HD |
987 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
988 | loff_t oldsize = inode->i_size; | |
989 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 990 | |
40e041a2 DH |
991 | /* protected by i_mutex */ |
992 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
993 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
994 | return -EPERM; | |
995 | ||
94c1e62d | 996 | if (newsize != oldsize) { |
77142517 KK |
997 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
998 | oldsize, newsize); | |
999 | if (error) | |
1000 | return error; | |
94c1e62d | 1001 | i_size_write(inode, newsize); |
078cd827 | 1002 | inode->i_ctime = inode->i_mtime = current_time(inode); |
94c1e62d | 1003 | } |
afa2db2f | 1004 | if (newsize <= oldsize) { |
94c1e62d | 1005 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
1006 | if (oldsize > holebegin) |
1007 | unmap_mapping_range(inode->i_mapping, | |
1008 | holebegin, 0, 1); | |
1009 | if (info->alloced) | |
1010 | shmem_truncate_range(inode, | |
1011 | newsize, (loff_t)-1); | |
94c1e62d | 1012 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
1013 | if (oldsize > holebegin) |
1014 | unmap_mapping_range(inode->i_mapping, | |
1015 | holebegin, 0, 1); | |
779750d2 KS |
1016 | |
1017 | /* | |
1018 | * Part of the huge page can be beyond i_size: subject | |
1019 | * to shrink under memory pressure. | |
1020 | */ | |
1021 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { | |
1022 | spin_lock(&sbinfo->shrinklist_lock); | |
1023 | if (list_empty(&info->shrinklist)) { | |
1024 | list_add_tail(&info->shrinklist, | |
1025 | &sbinfo->shrinklist); | |
1026 | sbinfo->shrinklist_len++; | |
1027 | } | |
1028 | spin_unlock(&sbinfo->shrinklist_lock); | |
1029 | } | |
94c1e62d | 1030 | } |
1da177e4 LT |
1031 | } |
1032 | ||
db78b877 | 1033 | setattr_copy(inode, attr); |
db78b877 | 1034 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 1035 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
1036 | return error; |
1037 | } | |
1038 | ||
1f895f75 | 1039 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1040 | { |
1da177e4 | 1041 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1042 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 | 1043 | |
3889e6e7 | 1044 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
1045 | shmem_unacct_size(info->flags, inode->i_size); |
1046 | inode->i_size = 0; | |
3889e6e7 | 1047 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1048 | if (!list_empty(&info->shrinklist)) { |
1049 | spin_lock(&sbinfo->shrinklist_lock); | |
1050 | if (!list_empty(&info->shrinklist)) { | |
1051 | list_del_init(&info->shrinklist); | |
1052 | sbinfo->shrinklist_len--; | |
1053 | } | |
1054 | spin_unlock(&sbinfo->shrinklist_lock); | |
1055 | } | |
1da177e4 | 1056 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 1057 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 1058 | list_del_init(&info->swaplist); |
cb5f7b9a | 1059 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1060 | } |
3ed47db3 | 1061 | } |
b09e0fa4 | 1062 | |
38f38657 | 1063 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 1064 | WARN_ON(inode->i_blocks); |
5b04c689 | 1065 | shmem_free_inode(inode->i_sb); |
dbd5768f | 1066 | clear_inode(inode); |
1da177e4 LT |
1067 | } |
1068 | ||
478922e2 MW |
1069 | static unsigned long find_swap_entry(struct radix_tree_root *root, void *item) |
1070 | { | |
1071 | struct radix_tree_iter iter; | |
1072 | void **slot; | |
1073 | unsigned long found = -1; | |
1074 | unsigned int checked = 0; | |
1075 | ||
1076 | rcu_read_lock(); | |
1077 | radix_tree_for_each_slot(slot, root, &iter, 0) { | |
1078 | if (*slot == item) { | |
1079 | found = iter.index; | |
1080 | break; | |
1081 | } | |
1082 | checked++; | |
1083 | if ((checked % 4096) != 0) | |
1084 | continue; | |
1085 | slot = radix_tree_iter_resume(slot, &iter); | |
1086 | cond_resched_rcu(); | |
1087 | } | |
1088 | ||
1089 | rcu_read_unlock(); | |
1090 | return found; | |
1091 | } | |
1092 | ||
46f65ec1 HD |
1093 | /* |
1094 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1095 | */ | |
41ffe5d5 | 1096 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 1097 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 1098 | { |
285b2c4f | 1099 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 1100 | void *radswap; |
41ffe5d5 | 1101 | pgoff_t index; |
bde05d1c HD |
1102 | gfp_t gfp; |
1103 | int error = 0; | |
1da177e4 | 1104 | |
46f65ec1 | 1105 | radswap = swp_to_radix_entry(swap); |
478922e2 | 1106 | index = find_swap_entry(&mapping->page_tree, radswap); |
46f65ec1 | 1107 | if (index == -1) |
00501b53 | 1108 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 1109 | |
1b1b32f2 HD |
1110 | /* |
1111 | * Move _head_ to start search for next from here. | |
1f895f75 | 1112 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 1113 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 1114 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
1115 | */ |
1116 | if (shmem_swaplist.next != &info->swaplist) | |
1117 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 1118 | |
bde05d1c HD |
1119 | gfp = mapping_gfp_mask(mapping); |
1120 | if (shmem_should_replace_page(*pagep, gfp)) { | |
1121 | mutex_unlock(&shmem_swaplist_mutex); | |
1122 | error = shmem_replace_page(pagep, gfp, info, index); | |
1123 | mutex_lock(&shmem_swaplist_mutex); | |
1124 | /* | |
1125 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
1126 | * allocation, but the inode might have been freed while we |
1127 | * dropped it: although a racing shmem_evict_inode() cannot | |
1128 | * complete without emptying the radix_tree, our page lock | |
1129 | * on this swapcache page is not enough to prevent that - | |
1130 | * free_swap_and_cache() of our swap entry will only | |
1131 | * trylock_page(), removing swap from radix_tree whatever. | |
1132 | * | |
1133 | * We must not proceed to shmem_add_to_page_cache() if the | |
1134 | * inode has been freed, but of course we cannot rely on | |
1135 | * inode or mapping or info to check that. However, we can | |
1136 | * safely check if our swap entry is still in use (and here | |
1137 | * it can't have got reused for another page): if it's still | |
1138 | * in use, then the inode cannot have been freed yet, and we | |
1139 | * can safely proceed (if it's no longer in use, that tells | |
1140 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
1141 | */ |
1142 | if (!page_swapcount(*pagep)) | |
1143 | error = -ENOENT; | |
1144 | } | |
1145 | ||
d13d1443 | 1146 | /* |
778dd893 HD |
1147 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
1148 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
1149 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 1150 | */ |
bde05d1c HD |
1151 | if (!error) |
1152 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 1153 | radswap); |
48f170fb | 1154 | if (error != -ENOMEM) { |
46f65ec1 HD |
1155 | /* |
1156 | * Truncation and eviction use free_swap_and_cache(), which | |
1157 | * only does trylock page: if we raced, best clean up here. | |
1158 | */ | |
bde05d1c HD |
1159 | delete_from_swap_cache(*pagep); |
1160 | set_page_dirty(*pagep); | |
46f65ec1 | 1161 | if (!error) { |
4595ef88 | 1162 | spin_lock_irq(&info->lock); |
46f65ec1 | 1163 | info->swapped--; |
4595ef88 | 1164 | spin_unlock_irq(&info->lock); |
46f65ec1 HD |
1165 | swap_free(swap); |
1166 | } | |
1da177e4 | 1167 | } |
2e0e26c7 | 1168 | return error; |
1da177e4 LT |
1169 | } |
1170 | ||
1171 | /* | |
46f65ec1 | 1172 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 1173 | */ |
41ffe5d5 | 1174 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 1175 | { |
41ffe5d5 | 1176 | struct list_head *this, *next; |
1da177e4 | 1177 | struct shmem_inode_info *info; |
00501b53 | 1178 | struct mem_cgroup *memcg; |
bde05d1c HD |
1179 | int error = 0; |
1180 | ||
1181 | /* | |
1182 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 1183 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 1184 | */ |
0142ef6c | 1185 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 1186 | goto out; |
778dd893 HD |
1187 | |
1188 | /* | |
1189 | * Charge page using GFP_KERNEL while we can wait, before taking | |
1190 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
1191 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 1192 | */ |
f627c2f5 KS |
1193 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg, |
1194 | false); | |
778dd893 HD |
1195 | if (error) |
1196 | goto out; | |
46f65ec1 | 1197 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 1198 | error = -EAGAIN; |
1da177e4 | 1199 | |
cb5f7b9a | 1200 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
1201 | list_for_each_safe(this, next, &shmem_swaplist) { |
1202 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 1203 | if (info->swapped) |
00501b53 | 1204 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
1205 | else |
1206 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1207 | cond_resched(); |
00501b53 | 1208 | if (error != -EAGAIN) |
778dd893 | 1209 | break; |
00501b53 | 1210 | /* found nothing in this: move on to search the next */ |
1da177e4 | 1211 | } |
cb5f7b9a | 1212 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1213 | |
00501b53 JW |
1214 | if (error) { |
1215 | if (error != -ENOMEM) | |
1216 | error = 0; | |
f627c2f5 | 1217 | mem_cgroup_cancel_charge(page, memcg, false); |
00501b53 | 1218 | } else |
f627c2f5 | 1219 | mem_cgroup_commit_charge(page, memcg, true, false); |
778dd893 | 1220 | out: |
aaa46865 | 1221 | unlock_page(page); |
09cbfeaf | 1222 | put_page(page); |
778dd893 | 1223 | return error; |
1da177e4 LT |
1224 | } |
1225 | ||
1226 | /* | |
1227 | * Move the page from the page cache to the swap cache. | |
1228 | */ | |
1229 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1230 | { | |
1231 | struct shmem_inode_info *info; | |
1da177e4 | 1232 | struct address_space *mapping; |
1da177e4 | 1233 | struct inode *inode; |
6922c0c7 HD |
1234 | swp_entry_t swap; |
1235 | pgoff_t index; | |
1da177e4 | 1236 | |
800d8c63 | 1237 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1238 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1239 | mapping = page->mapping; |
1240 | index = page->index; | |
1241 | inode = mapping->host; | |
1242 | info = SHMEM_I(inode); | |
1243 | if (info->flags & VM_LOCKED) | |
1244 | goto redirty; | |
d9fe526a | 1245 | if (!total_swap_pages) |
1da177e4 LT |
1246 | goto redirty; |
1247 | ||
d9fe526a | 1248 | /* |
97b713ba CH |
1249 | * Our capabilities prevent regular writeback or sync from ever calling |
1250 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1251 | * its underlying filesystem, in which case tmpfs should write out to | |
1252 | * swap only in response to memory pressure, and not for the writeback | |
1253 | * threads or sync. | |
d9fe526a | 1254 | */ |
48f170fb HD |
1255 | if (!wbc->for_reclaim) { |
1256 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1257 | goto redirty; | |
1258 | } | |
1635f6a7 HD |
1259 | |
1260 | /* | |
1261 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1262 | * value into swapfile.c, the only way we can correctly account for a | |
1263 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1264 | * |
1265 | * That's okay for a page already fallocated earlier, but if we have | |
1266 | * not yet completed the fallocation, then (a) we want to keep track | |
1267 | * of this page in case we have to undo it, and (b) it may not be a | |
1268 | * good idea to continue anyway, once we're pushing into swap. So | |
1269 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1270 | */ |
1271 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1272 | if (inode->i_private) { |
1273 | struct shmem_falloc *shmem_falloc; | |
1274 | spin_lock(&inode->i_lock); | |
1275 | shmem_falloc = inode->i_private; | |
1276 | if (shmem_falloc && | |
8e205f77 | 1277 | !shmem_falloc->waitq && |
1aac1400 HD |
1278 | index >= shmem_falloc->start && |
1279 | index < shmem_falloc->next) | |
1280 | shmem_falloc->nr_unswapped++; | |
1281 | else | |
1282 | shmem_falloc = NULL; | |
1283 | spin_unlock(&inode->i_lock); | |
1284 | if (shmem_falloc) | |
1285 | goto redirty; | |
1286 | } | |
1635f6a7 HD |
1287 | clear_highpage(page); |
1288 | flush_dcache_page(page); | |
1289 | SetPageUptodate(page); | |
1290 | } | |
1291 | ||
48f170fb HD |
1292 | swap = get_swap_page(); |
1293 | if (!swap.val) | |
1294 | goto redirty; | |
d9fe526a | 1295 | |
37e84351 VD |
1296 | if (mem_cgroup_try_charge_swap(page, swap)) |
1297 | goto free_swap; | |
1298 | ||
b1dea800 HD |
1299 | /* |
1300 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1301 | * if it's not already there. Do it now before the page is |
1302 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1303 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1304 | * we've incremented swapped, because shmem_unuse_inode() will |
1305 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1306 | */ |
48f170fb HD |
1307 | mutex_lock(&shmem_swaplist_mutex); |
1308 | if (list_empty(&info->swaplist)) | |
1309 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 1310 | |
48f170fb | 1311 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
4595ef88 | 1312 | spin_lock_irq(&info->lock); |
6922c0c7 | 1313 | shmem_recalc_inode(inode); |
267a4c76 | 1314 | info->swapped++; |
4595ef88 | 1315 | spin_unlock_irq(&info->lock); |
6922c0c7 | 1316 | |
267a4c76 HD |
1317 | swap_shmem_alloc(swap); |
1318 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1319 | ||
6922c0c7 | 1320 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1321 | BUG_ON(page_mapped(page)); |
9fab5619 | 1322 | swap_writepage(page, wbc); |
1da177e4 LT |
1323 | return 0; |
1324 | } | |
1325 | ||
6922c0c7 | 1326 | mutex_unlock(&shmem_swaplist_mutex); |
37e84351 | 1327 | free_swap: |
0a31bc97 | 1328 | swapcache_free(swap); |
1da177e4 LT |
1329 | redirty: |
1330 | set_page_dirty(page); | |
d9fe526a HD |
1331 | if (wbc->for_reclaim) |
1332 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1333 | unlock_page(page); | |
1334 | return 0; | |
1da177e4 LT |
1335 | } |
1336 | ||
75edd345 | 1337 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1338 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1339 | { |
095f1fc4 | 1340 | char buffer[64]; |
680d794b | 1341 | |
71fe804b | 1342 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1343 | return; /* show nothing */ |
680d794b | 1344 | |
a7a88b23 | 1345 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1346 | |
1347 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1348 | } |
71fe804b LS |
1349 | |
1350 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1351 | { | |
1352 | struct mempolicy *mpol = NULL; | |
1353 | if (sbinfo->mpol) { | |
1354 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1355 | mpol = sbinfo->mpol; | |
1356 | mpol_get(mpol); | |
1357 | spin_unlock(&sbinfo->stat_lock); | |
1358 | } | |
1359 | return mpol; | |
1360 | } | |
75edd345 HD |
1361 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1362 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1363 | { | |
1364 | } | |
1365 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1366 | { | |
1367 | return NULL; | |
1368 | } | |
1369 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1370 | #ifndef CONFIG_NUMA | |
1371 | #define vm_policy vm_private_data | |
1372 | #endif | |
680d794b | 1373 | |
800d8c63 KS |
1374 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1375 | struct shmem_inode_info *info, pgoff_t index) | |
1376 | { | |
1377 | /* Create a pseudo vma that just contains the policy */ | |
1378 | vma->vm_start = 0; | |
1379 | /* Bias interleave by inode number to distribute better across nodes */ | |
1380 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
1381 | vma->vm_ops = NULL; | |
1382 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); | |
1383 | } | |
1384 | ||
1385 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1386 | { | |
1387 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1388 | mpol_cond_put(vma->vm_policy); | |
1389 | } | |
1390 | ||
41ffe5d5 HD |
1391 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1392 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1393 | { |
1da177e4 | 1394 | struct vm_area_struct pvma; |
18a2f371 | 1395 | struct page *page; |
52cd3b07 | 1396 | |
800d8c63 | 1397 | shmem_pseudo_vma_init(&pvma, info, index); |
18a2f371 | 1398 | page = swapin_readahead(swap, gfp, &pvma, 0); |
800d8c63 | 1399 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1400 | |
800d8c63 KS |
1401 | return page; |
1402 | } | |
1403 | ||
1404 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1405 | struct shmem_inode_info *info, pgoff_t index) | |
1406 | { | |
1407 | struct vm_area_struct pvma; | |
1408 | struct inode *inode = &info->vfs_inode; | |
1409 | struct address_space *mapping = inode->i_mapping; | |
4620a06e | 1410 | pgoff_t idx, hindex; |
800d8c63 KS |
1411 | void __rcu **results; |
1412 | struct page *page; | |
1413 | ||
e496cf3d | 1414 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1415 | return NULL; |
1416 | ||
4620a06e | 1417 | hindex = round_down(index, HPAGE_PMD_NR); |
800d8c63 KS |
1418 | rcu_read_lock(); |
1419 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx, | |
1420 | hindex, 1) && idx < hindex + HPAGE_PMD_NR) { | |
1421 | rcu_read_unlock(); | |
1422 | return NULL; | |
1423 | } | |
1424 | rcu_read_unlock(); | |
18a2f371 | 1425 | |
800d8c63 KS |
1426 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1427 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
1428 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); | |
1429 | shmem_pseudo_vma_destroy(&pvma); | |
1430 | if (page) | |
1431 | prep_transhuge_page(page); | |
18a2f371 | 1432 | return page; |
1da177e4 LT |
1433 | } |
1434 | ||
02098fea | 1435 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1436 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1437 | { |
1438 | struct vm_area_struct pvma; | |
18a2f371 | 1439 | struct page *page; |
1da177e4 | 1440 | |
800d8c63 KS |
1441 | shmem_pseudo_vma_init(&pvma, info, index); |
1442 | page = alloc_page_vma(gfp, &pvma, 0); | |
1443 | shmem_pseudo_vma_destroy(&pvma); | |
1444 | ||
1445 | return page; | |
1446 | } | |
1447 | ||
1448 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
1449 | struct shmem_inode_info *info, struct shmem_sb_info *sbinfo, | |
1450 | pgoff_t index, bool huge) | |
1451 | { | |
1452 | struct page *page; | |
1453 | int nr; | |
1454 | int err = -ENOSPC; | |
52cd3b07 | 1455 | |
e496cf3d | 1456 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1457 | huge = false; |
1458 | nr = huge ? HPAGE_PMD_NR : 1; | |
1459 | ||
1460 | if (shmem_acct_block(info->flags, nr)) | |
1461 | goto failed; | |
1462 | if (sbinfo->max_blocks) { | |
1463 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
1464 | sbinfo->max_blocks - nr) > 0) | |
1465 | goto unacct; | |
1466 | percpu_counter_add(&sbinfo->used_blocks, nr); | |
1467 | } | |
1468 | ||
1469 | if (huge) | |
1470 | page = shmem_alloc_hugepage(gfp, info, index); | |
1471 | else | |
1472 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1473 | if (page) { |
1474 | __SetPageLocked(page); | |
1475 | __SetPageSwapBacked(page); | |
800d8c63 | 1476 | return page; |
75edd345 | 1477 | } |
18a2f371 | 1478 | |
800d8c63 KS |
1479 | err = -ENOMEM; |
1480 | if (sbinfo->max_blocks) | |
1481 | percpu_counter_add(&sbinfo->used_blocks, -nr); | |
1482 | unacct: | |
1483 | shmem_unacct_blocks(info->flags, nr); | |
1484 | failed: | |
1485 | return ERR_PTR(err); | |
1da177e4 | 1486 | } |
71fe804b | 1487 | |
bde05d1c HD |
1488 | /* |
1489 | * When a page is moved from swapcache to shmem filecache (either by the | |
1490 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1491 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1492 | * ignorance of the mapping it belongs to. If that mapping has special | |
1493 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1494 | * we may need to copy to a suitable page before moving to filecache. | |
1495 | * | |
1496 | * In a future release, this may well be extended to respect cpuset and | |
1497 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1498 | * but for now it is a simple matter of zone. | |
1499 | */ | |
1500 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1501 | { | |
1502 | return page_zonenum(page) > gfp_zone(gfp); | |
1503 | } | |
1504 | ||
1505 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1506 | struct shmem_inode_info *info, pgoff_t index) | |
1507 | { | |
1508 | struct page *oldpage, *newpage; | |
1509 | struct address_space *swap_mapping; | |
1510 | pgoff_t swap_index; | |
1511 | int error; | |
1512 | ||
1513 | oldpage = *pagep; | |
1514 | swap_index = page_private(oldpage); | |
1515 | swap_mapping = page_mapping(oldpage); | |
1516 | ||
1517 | /* | |
1518 | * We have arrived here because our zones are constrained, so don't | |
1519 | * limit chance of success by further cpuset and node constraints. | |
1520 | */ | |
1521 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1522 | newpage = shmem_alloc_page(gfp, info, index); | |
1523 | if (!newpage) | |
1524 | return -ENOMEM; | |
bde05d1c | 1525 | |
09cbfeaf | 1526 | get_page(newpage); |
bde05d1c | 1527 | copy_highpage(newpage, oldpage); |
0142ef6c | 1528 | flush_dcache_page(newpage); |
bde05d1c | 1529 | |
9956edf3 HD |
1530 | __SetPageLocked(newpage); |
1531 | __SetPageSwapBacked(newpage); | |
bde05d1c | 1532 | SetPageUptodate(newpage); |
bde05d1c | 1533 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1534 | SetPageSwapCache(newpage); |
1535 | ||
1536 | /* | |
1537 | * Our caller will very soon move newpage out of swapcache, but it's | |
1538 | * a nice clean interface for us to replace oldpage by newpage there. | |
1539 | */ | |
1540 | spin_lock_irq(&swap_mapping->tree_lock); | |
1541 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1542 | newpage); | |
0142ef6c | 1543 | if (!error) { |
11fb9989 MG |
1544 | __inc_node_page_state(newpage, NR_FILE_PAGES); |
1545 | __dec_node_page_state(oldpage, NR_FILE_PAGES); | |
0142ef6c | 1546 | } |
bde05d1c | 1547 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1548 | |
0142ef6c HD |
1549 | if (unlikely(error)) { |
1550 | /* | |
1551 | * Is this possible? I think not, now that our callers check | |
1552 | * both PageSwapCache and page_private after getting page lock; | |
1553 | * but be defensive. Reverse old to newpage for clear and free. | |
1554 | */ | |
1555 | oldpage = newpage; | |
1556 | } else { | |
6a93ca8f | 1557 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1558 | lru_cache_add_anon(newpage); |
1559 | *pagep = newpage; | |
1560 | } | |
bde05d1c HD |
1561 | |
1562 | ClearPageSwapCache(oldpage); | |
1563 | set_page_private(oldpage, 0); | |
1564 | ||
1565 | unlock_page(oldpage); | |
09cbfeaf KS |
1566 | put_page(oldpage); |
1567 | put_page(oldpage); | |
0142ef6c | 1568 | return error; |
bde05d1c HD |
1569 | } |
1570 | ||
1da177e4 | 1571 | /* |
68da9f05 | 1572 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1573 | * |
1574 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1575 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1576 | * entry since a page cannot live in both the swap and page cache. |
1577 | * | |
1578 | * fault_mm and fault_type are only supplied by shmem_fault: | |
1579 | * otherwise they are NULL. | |
1da177e4 | 1580 | */ |
41ffe5d5 | 1581 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 1582 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
cfda0526 | 1583 | struct vm_area_struct *vma, struct vm_fault *vmf, int *fault_type) |
1da177e4 LT |
1584 | { |
1585 | struct address_space *mapping = inode->i_mapping; | |
23f919d4 | 1586 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 | 1587 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1588 | struct mm_struct *charge_mm; |
00501b53 | 1589 | struct mem_cgroup *memcg; |
27ab7006 | 1590 | struct page *page; |
1da177e4 | 1591 | swp_entry_t swap; |
657e3038 | 1592 | enum sgp_type sgp_huge = sgp; |
800d8c63 | 1593 | pgoff_t hindex = index; |
1da177e4 | 1594 | int error; |
54af6042 | 1595 | int once = 0; |
1635f6a7 | 1596 | int alloced = 0; |
1da177e4 | 1597 | |
09cbfeaf | 1598 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1599 | return -EFBIG; |
657e3038 KS |
1600 | if (sgp == SGP_NOHUGE || sgp == SGP_HUGE) |
1601 | sgp = SGP_CACHE; | |
1da177e4 | 1602 | repeat: |
54af6042 | 1603 | swap.val = 0; |
0cd6144a | 1604 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1605 | if (radix_tree_exceptional_entry(page)) { |
1606 | swap = radix_to_swp_entry(page); | |
1607 | page = NULL; | |
1608 | } | |
1609 | ||
75edd345 | 1610 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1611 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 1612 | error = -EINVAL; |
267a4c76 | 1613 | goto unlock; |
54af6042 HD |
1614 | } |
1615 | ||
66d2f4d2 HD |
1616 | if (page && sgp == SGP_WRITE) |
1617 | mark_page_accessed(page); | |
1618 | ||
1635f6a7 HD |
1619 | /* fallocated page? */ |
1620 | if (page && !PageUptodate(page)) { | |
1621 | if (sgp != SGP_READ) | |
1622 | goto clear; | |
1623 | unlock_page(page); | |
09cbfeaf | 1624 | put_page(page); |
1635f6a7 HD |
1625 | page = NULL; |
1626 | } | |
54af6042 | 1627 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1628 | *pagep = page; |
1629 | return 0; | |
27ab7006 HD |
1630 | } |
1631 | ||
1632 | /* | |
54af6042 HD |
1633 | * Fast cache lookup did not find it: |
1634 | * bring it back from swap or allocate. | |
27ab7006 | 1635 | */ |
54af6042 | 1636 | sbinfo = SHMEM_SB(inode->i_sb); |
cfda0526 | 1637 | charge_mm = vma ? vma->vm_mm : current->mm; |
1da177e4 | 1638 | |
1da177e4 LT |
1639 | if (swap.val) { |
1640 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1641 | page = lookup_swap_cache(swap); |
1642 | if (!page) { | |
9e18eb29 ALC |
1643 | /* Or update major stats only when swapin succeeds?? */ |
1644 | if (fault_type) { | |
68da9f05 | 1645 | *fault_type |= VM_FAULT_MAJOR; |
9e18eb29 | 1646 | count_vm_event(PGMAJFAULT); |
cfda0526 MR |
1647 | mem_cgroup_count_vm_event(charge_mm, |
1648 | PGMAJFAULT); | |
9e18eb29 ALC |
1649 | } |
1650 | /* Here we actually start the io */ | |
41ffe5d5 | 1651 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1652 | if (!page) { |
54af6042 HD |
1653 | error = -ENOMEM; |
1654 | goto failed; | |
1da177e4 | 1655 | } |
1da177e4 LT |
1656 | } |
1657 | ||
1658 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1659 | lock_page(page); |
0142ef6c | 1660 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1661 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1662 | error = -EEXIST; /* try again */ |
d1899228 | 1663 | goto unlock; |
bde05d1c | 1664 | } |
27ab7006 | 1665 | if (!PageUptodate(page)) { |
1da177e4 | 1666 | error = -EIO; |
54af6042 | 1667 | goto failed; |
1da177e4 | 1668 | } |
54af6042 HD |
1669 | wait_on_page_writeback(page); |
1670 | ||
bde05d1c HD |
1671 | if (shmem_should_replace_page(page, gfp)) { |
1672 | error = shmem_replace_page(&page, gfp, info, index); | |
1673 | if (error) | |
1674 | goto failed; | |
1da177e4 | 1675 | } |
27ab7006 | 1676 | |
9e18eb29 | 1677 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
f627c2f5 | 1678 | false); |
d1899228 | 1679 | if (!error) { |
aa3b1895 | 1680 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1681 | swp_to_radix_entry(swap)); |
215c02bc HD |
1682 | /* |
1683 | * We already confirmed swap under page lock, and make | |
1684 | * no memory allocation here, so usually no possibility | |
1685 | * of error; but free_swap_and_cache() only trylocks a | |
1686 | * page, so it is just possible that the entry has been | |
1687 | * truncated or holepunched since swap was confirmed. | |
1688 | * shmem_undo_range() will have done some of the | |
1689 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1690 | * the rest. |
215c02bc HD |
1691 | * Reset swap.val? No, leave it so "failed" goes back to |
1692 | * "repeat": reading a hole and writing should succeed. | |
1693 | */ | |
00501b53 | 1694 | if (error) { |
f627c2f5 | 1695 | mem_cgroup_cancel_charge(page, memcg, false); |
215c02bc | 1696 | delete_from_swap_cache(page); |
00501b53 | 1697 | } |
d1899228 | 1698 | } |
54af6042 HD |
1699 | if (error) |
1700 | goto failed; | |
1701 | ||
f627c2f5 | 1702 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1703 | |
4595ef88 | 1704 | spin_lock_irq(&info->lock); |
285b2c4f | 1705 | info->swapped--; |
54af6042 | 1706 | shmem_recalc_inode(inode); |
4595ef88 | 1707 | spin_unlock_irq(&info->lock); |
54af6042 | 1708 | |
66d2f4d2 HD |
1709 | if (sgp == SGP_WRITE) |
1710 | mark_page_accessed(page); | |
1711 | ||
54af6042 | 1712 | delete_from_swap_cache(page); |
27ab7006 HD |
1713 | set_page_dirty(page); |
1714 | swap_free(swap); | |
1715 | ||
54af6042 | 1716 | } else { |
cfda0526 MR |
1717 | if (vma && userfaultfd_missing(vma)) { |
1718 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
1719 | return 0; | |
1720 | } | |
1721 | ||
800d8c63 KS |
1722 | /* shmem_symlink() */ |
1723 | if (mapping->a_ops != &shmem_aops) | |
1724 | goto alloc_nohuge; | |
657e3038 | 1725 | if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE) |
800d8c63 KS |
1726 | goto alloc_nohuge; |
1727 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1728 | goto alloc_huge; | |
1729 | switch (sbinfo->huge) { | |
1730 | loff_t i_size; | |
1731 | pgoff_t off; | |
1732 | case SHMEM_HUGE_NEVER: | |
1733 | goto alloc_nohuge; | |
1734 | case SHMEM_HUGE_WITHIN_SIZE: | |
1735 | off = round_up(index, HPAGE_PMD_NR); | |
1736 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1737 | if (i_size >= HPAGE_PMD_SIZE && | |
1738 | i_size >> PAGE_SHIFT >= off) | |
1739 | goto alloc_huge; | |
1740 | /* fallthrough */ | |
1741 | case SHMEM_HUGE_ADVISE: | |
657e3038 KS |
1742 | if (sgp_huge == SGP_HUGE) |
1743 | goto alloc_huge; | |
1744 | /* TODO: implement fadvise() hints */ | |
800d8c63 | 1745 | goto alloc_nohuge; |
54af6042 | 1746 | } |
1da177e4 | 1747 | |
800d8c63 KS |
1748 | alloc_huge: |
1749 | page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1750 | index, true); | |
1751 | if (IS_ERR(page)) { | |
1752 | alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1753 | index, false); | |
1da177e4 | 1754 | } |
800d8c63 | 1755 | if (IS_ERR(page)) { |
779750d2 | 1756 | int retry = 5; |
800d8c63 KS |
1757 | error = PTR_ERR(page); |
1758 | page = NULL; | |
779750d2 KS |
1759 | if (error != -ENOSPC) |
1760 | goto failed; | |
1761 | /* | |
1762 | * Try to reclaim some spece by splitting a huge page | |
1763 | * beyond i_size on the filesystem. | |
1764 | */ | |
1765 | while (retry--) { | |
1766 | int ret; | |
1767 | ret = shmem_unused_huge_shrink(sbinfo, NULL, 1); | |
1768 | if (ret == SHRINK_STOP) | |
1769 | break; | |
1770 | if (ret) | |
1771 | goto alloc_nohuge; | |
1772 | } | |
800d8c63 KS |
1773 | goto failed; |
1774 | } | |
1775 | ||
1776 | if (PageTransHuge(page)) | |
1777 | hindex = round_down(index, HPAGE_PMD_NR); | |
1778 | else | |
1779 | hindex = index; | |
1780 | ||
66d2f4d2 | 1781 | if (sgp == SGP_WRITE) |
eb39d618 | 1782 | __SetPageReferenced(page); |
66d2f4d2 | 1783 | |
9e18eb29 | 1784 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
800d8c63 | 1785 | PageTransHuge(page)); |
54af6042 | 1786 | if (error) |
800d8c63 KS |
1787 | goto unacct; |
1788 | error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK, | |
1789 | compound_order(page)); | |
b065b432 | 1790 | if (!error) { |
800d8c63 | 1791 | error = shmem_add_to_page_cache(page, mapping, hindex, |
fed400a1 | 1792 | NULL); |
b065b432 HD |
1793 | radix_tree_preload_end(); |
1794 | } | |
1795 | if (error) { | |
800d8c63 KS |
1796 | mem_cgroup_cancel_charge(page, memcg, |
1797 | PageTransHuge(page)); | |
1798 | goto unacct; | |
b065b432 | 1799 | } |
800d8c63 KS |
1800 | mem_cgroup_commit_charge(page, memcg, false, |
1801 | PageTransHuge(page)); | |
54af6042 HD |
1802 | lru_cache_add_anon(page); |
1803 | ||
4595ef88 | 1804 | spin_lock_irq(&info->lock); |
800d8c63 KS |
1805 | info->alloced += 1 << compound_order(page); |
1806 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); | |
54af6042 | 1807 | shmem_recalc_inode(inode); |
4595ef88 | 1808 | spin_unlock_irq(&info->lock); |
1635f6a7 | 1809 | alloced = true; |
54af6042 | 1810 | |
779750d2 KS |
1811 | if (PageTransHuge(page) && |
1812 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < | |
1813 | hindex + HPAGE_PMD_NR - 1) { | |
1814 | /* | |
1815 | * Part of the huge page is beyond i_size: subject | |
1816 | * to shrink under memory pressure. | |
1817 | */ | |
1818 | spin_lock(&sbinfo->shrinklist_lock); | |
1819 | if (list_empty(&info->shrinklist)) { | |
1820 | list_add_tail(&info->shrinklist, | |
1821 | &sbinfo->shrinklist); | |
1822 | sbinfo->shrinklist_len++; | |
1823 | } | |
1824 | spin_unlock(&sbinfo->shrinklist_lock); | |
1825 | } | |
1826 | ||
ec9516fb | 1827 | /* |
1635f6a7 HD |
1828 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1829 | */ | |
1830 | if (sgp == SGP_FALLOC) | |
1831 | sgp = SGP_WRITE; | |
1832 | clear: | |
1833 | /* | |
1834 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1835 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1836 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb | 1837 | */ |
800d8c63 KS |
1838 | if (sgp != SGP_WRITE && !PageUptodate(page)) { |
1839 | struct page *head = compound_head(page); | |
1840 | int i; | |
1841 | ||
1842 | for (i = 0; i < (1 << compound_order(head)); i++) { | |
1843 | clear_highpage(head + i); | |
1844 | flush_dcache_page(head + i); | |
1845 | } | |
1846 | SetPageUptodate(head); | |
ec9516fb | 1847 | } |
1da177e4 | 1848 | } |
bde05d1c | 1849 | |
54af6042 | 1850 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1851 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1852 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1853 | if (alloced) { |
1854 | ClearPageDirty(page); | |
1855 | delete_from_page_cache(page); | |
4595ef88 | 1856 | spin_lock_irq(&info->lock); |
267a4c76 | 1857 | shmem_recalc_inode(inode); |
4595ef88 | 1858 | spin_unlock_irq(&info->lock); |
267a4c76 | 1859 | } |
54af6042 | 1860 | error = -EINVAL; |
267a4c76 | 1861 | goto unlock; |
e83c32e8 | 1862 | } |
800d8c63 | 1863 | *pagep = page + index - hindex; |
54af6042 | 1864 | return 0; |
1da177e4 | 1865 | |
59a16ead | 1866 | /* |
54af6042 | 1867 | * Error recovery. |
59a16ead | 1868 | */ |
54af6042 | 1869 | unacct: |
800d8c63 KS |
1870 | if (sbinfo->max_blocks) |
1871 | percpu_counter_sub(&sbinfo->used_blocks, | |
1872 | 1 << compound_order(page)); | |
1873 | shmem_unacct_blocks(info->flags, 1 << compound_order(page)); | |
1874 | ||
1875 | if (PageTransHuge(page)) { | |
1876 | unlock_page(page); | |
1877 | put_page(page); | |
1878 | goto alloc_nohuge; | |
1879 | } | |
54af6042 | 1880 | failed: |
267a4c76 | 1881 | if (swap.val && !shmem_confirm_swap(mapping, index, swap)) |
d1899228 HD |
1882 | error = -EEXIST; |
1883 | unlock: | |
27ab7006 | 1884 | if (page) { |
54af6042 | 1885 | unlock_page(page); |
09cbfeaf | 1886 | put_page(page); |
54af6042 HD |
1887 | } |
1888 | if (error == -ENOSPC && !once++) { | |
4595ef88 | 1889 | spin_lock_irq(&info->lock); |
54af6042 | 1890 | shmem_recalc_inode(inode); |
4595ef88 | 1891 | spin_unlock_irq(&info->lock); |
27ab7006 | 1892 | goto repeat; |
ff36b801 | 1893 | } |
d1899228 | 1894 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1895 | goto repeat; |
1896 | return error; | |
1da177e4 LT |
1897 | } |
1898 | ||
10d20bd2 LT |
1899 | /* |
1900 | * This is like autoremove_wake_function, but it removes the wait queue | |
1901 | * entry unconditionally - even if something else had already woken the | |
1902 | * target. | |
1903 | */ | |
1904 | static int synchronous_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key) | |
1905 | { | |
1906 | int ret = default_wake_function(wait, mode, sync, key); | |
1907 | list_del_init(&wait->task_list); | |
1908 | return ret; | |
1909 | } | |
1910 | ||
d0217ac0 | 1911 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1912 | { |
496ad9aa | 1913 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 1914 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
657e3038 | 1915 | enum sgp_type sgp; |
1da177e4 | 1916 | int error; |
68da9f05 | 1917 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1918 | |
f00cdc6d HD |
1919 | /* |
1920 | * Trinity finds that probing a hole which tmpfs is punching can | |
1921 | * prevent the hole-punch from ever completing: which in turn | |
1922 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1923 | * faulting pages into the hole while it's being punched. Although |
1924 | * shmem_undo_range() does remove the additions, it may be unable to | |
1925 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1926 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1927 | * | |
1928 | * It does not matter if we sometimes reach this check just before the | |
1929 | * hole-punch begins, so that one fault then races with the punch: | |
1930 | * we just need to make racing faults a rare case. | |
1931 | * | |
1932 | * The implementation below would be much simpler if we just used a | |
1933 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1934 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1935 | */ |
1936 | if (unlikely(inode->i_private)) { | |
1937 | struct shmem_falloc *shmem_falloc; | |
1938 | ||
1939 | spin_lock(&inode->i_lock); | |
1940 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1941 | if (shmem_falloc && |
1942 | shmem_falloc->waitq && | |
1943 | vmf->pgoff >= shmem_falloc->start && | |
1944 | vmf->pgoff < shmem_falloc->next) { | |
1945 | wait_queue_head_t *shmem_falloc_waitq; | |
10d20bd2 | 1946 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); |
8e205f77 HD |
1947 | |
1948 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1949 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1950 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1951 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1952 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1953 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1954 | } |
8e205f77 HD |
1955 | |
1956 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1957 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1958 | TASK_UNINTERRUPTIBLE); | |
1959 | spin_unlock(&inode->i_lock); | |
1960 | schedule(); | |
1961 | ||
1962 | /* | |
1963 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1964 | * stack of the hole-punching task: shmem_falloc_waitq | |
1965 | * is usually invalid by the time we reach here, but | |
1966 | * finish_wait() does not dereference it in that case; | |
1967 | * though i_lock needed lest racing with wake_up_all(). | |
1968 | */ | |
1969 | spin_lock(&inode->i_lock); | |
1970 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1971 | spin_unlock(&inode->i_lock); | |
1972 | return ret; | |
f00cdc6d | 1973 | } |
8e205f77 | 1974 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1975 | } |
1976 | ||
657e3038 KS |
1977 | sgp = SGP_CACHE; |
1978 | if (vma->vm_flags & VM_HUGEPAGE) | |
1979 | sgp = SGP_HUGE; | |
1980 | else if (vma->vm_flags & VM_NOHUGEPAGE) | |
1981 | sgp = SGP_NOHUGE; | |
1982 | ||
1983 | error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp, | |
cfda0526 | 1984 | gfp, vma, vmf, &ret); |
d0217ac0 NP |
1985 | if (error) |
1986 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1987 | return ret; |
1da177e4 LT |
1988 | } |
1989 | ||
c01d5b30 HD |
1990 | unsigned long shmem_get_unmapped_area(struct file *file, |
1991 | unsigned long uaddr, unsigned long len, | |
1992 | unsigned long pgoff, unsigned long flags) | |
1993 | { | |
1994 | unsigned long (*get_area)(struct file *, | |
1995 | unsigned long, unsigned long, unsigned long, unsigned long); | |
1996 | unsigned long addr; | |
1997 | unsigned long offset; | |
1998 | unsigned long inflated_len; | |
1999 | unsigned long inflated_addr; | |
2000 | unsigned long inflated_offset; | |
2001 | ||
2002 | if (len > TASK_SIZE) | |
2003 | return -ENOMEM; | |
2004 | ||
2005 | get_area = current->mm->get_unmapped_area; | |
2006 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2007 | ||
e496cf3d | 2008 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
c01d5b30 HD |
2009 | return addr; |
2010 | if (IS_ERR_VALUE(addr)) | |
2011 | return addr; | |
2012 | if (addr & ~PAGE_MASK) | |
2013 | return addr; | |
2014 | if (addr > TASK_SIZE - len) | |
2015 | return addr; | |
2016 | ||
2017 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2018 | return addr; | |
2019 | if (len < HPAGE_PMD_SIZE) | |
2020 | return addr; | |
2021 | if (flags & MAP_FIXED) | |
2022 | return addr; | |
2023 | /* | |
2024 | * Our priority is to support MAP_SHARED mapped hugely; | |
2025 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
2026 | * But if caller specified an address hint, respect that as before. | |
2027 | */ | |
2028 | if (uaddr) | |
2029 | return addr; | |
2030 | ||
2031 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2032 | struct super_block *sb; | |
2033 | ||
2034 | if (file) { | |
2035 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2036 | sb = file_inode(file)->i_sb; | |
2037 | } else { | |
2038 | /* | |
2039 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2040 | * for "/dev/zero", to create a shared anonymous object. | |
2041 | */ | |
2042 | if (IS_ERR(shm_mnt)) | |
2043 | return addr; | |
2044 | sb = shm_mnt->mnt_sb; | |
2045 | } | |
3089bf61 | 2046 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2047 | return addr; |
2048 | } | |
2049 | ||
2050 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2051 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2052 | return addr; | |
2053 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2054 | return addr; | |
2055 | ||
2056 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2057 | if (inflated_len > TASK_SIZE) | |
2058 | return addr; | |
2059 | if (inflated_len < len) | |
2060 | return addr; | |
2061 | ||
2062 | inflated_addr = get_area(NULL, 0, inflated_len, 0, flags); | |
2063 | if (IS_ERR_VALUE(inflated_addr)) | |
2064 | return addr; | |
2065 | if (inflated_addr & ~PAGE_MASK) | |
2066 | return addr; | |
2067 | ||
2068 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2069 | inflated_addr += offset - inflated_offset; | |
2070 | if (inflated_offset > offset) | |
2071 | inflated_addr += HPAGE_PMD_SIZE; | |
2072 | ||
2073 | if (inflated_addr > TASK_SIZE - len) | |
2074 | return addr; | |
2075 | return inflated_addr; | |
2076 | } | |
2077 | ||
1da177e4 | 2078 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2079 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2080 | { |
496ad9aa | 2081 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2082 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2083 | } |
2084 | ||
d8dc74f2 AB |
2085 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
2086 | unsigned long addr) | |
1da177e4 | 2087 | { |
496ad9aa | 2088 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2089 | pgoff_t index; |
1da177e4 | 2090 | |
41ffe5d5 HD |
2091 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2092 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
2093 | } |
2094 | #endif | |
2095 | ||
2096 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
2097 | { | |
496ad9aa | 2098 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2099 | struct shmem_inode_info *info = SHMEM_I(inode); |
2100 | int retval = -ENOMEM; | |
2101 | ||
4595ef88 | 2102 | spin_lock_irq(&info->lock); |
1da177e4 LT |
2103 | if (lock && !(info->flags & VM_LOCKED)) { |
2104 | if (!user_shm_lock(inode->i_size, user)) | |
2105 | goto out_nomem; | |
2106 | info->flags |= VM_LOCKED; | |
89e004ea | 2107 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
2108 | } |
2109 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
2110 | user_shm_unlock(inode->i_size, user); | |
2111 | info->flags &= ~VM_LOCKED; | |
89e004ea | 2112 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2113 | } |
2114 | retval = 0; | |
89e004ea | 2115 | |
1da177e4 | 2116 | out_nomem: |
4595ef88 | 2117 | spin_unlock_irq(&info->lock); |
1da177e4 LT |
2118 | return retval; |
2119 | } | |
2120 | ||
9b83a6a8 | 2121 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
2122 | { |
2123 | file_accessed(file); | |
2124 | vma->vm_ops = &shmem_vm_ops; | |
e496cf3d | 2125 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
2126 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
2127 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
2128 | khugepaged_enter(vma, vma->vm_flags); | |
2129 | } | |
1da177e4 LT |
2130 | return 0; |
2131 | } | |
2132 | ||
454abafe | 2133 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 2134 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
2135 | { |
2136 | struct inode *inode; | |
2137 | struct shmem_inode_info *info; | |
2138 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
2139 | ||
5b04c689 PE |
2140 | if (shmem_reserve_inode(sb)) |
2141 | return NULL; | |
1da177e4 LT |
2142 | |
2143 | inode = new_inode(sb); | |
2144 | if (inode) { | |
85fe4025 | 2145 | inode->i_ino = get_next_ino(); |
454abafe | 2146 | inode_init_owner(inode, dir, mode); |
1da177e4 | 2147 | inode->i_blocks = 0; |
078cd827 | 2148 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
91828a40 | 2149 | inode->i_generation = get_seconds(); |
1da177e4 LT |
2150 | info = SHMEM_I(inode); |
2151 | memset(info, 0, (char *)inode - (char *)info); | |
2152 | spin_lock_init(&info->lock); | |
40e041a2 | 2153 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 2154 | info->flags = flags & VM_NORESERVE; |
779750d2 | 2155 | INIT_LIST_HEAD(&info->shrinklist); |
1da177e4 | 2156 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 2157 | simple_xattrs_init(&info->xattrs); |
72c04902 | 2158 | cache_no_acl(inode); |
1da177e4 LT |
2159 | |
2160 | switch (mode & S_IFMT) { | |
2161 | default: | |
39f0247d | 2162 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
2163 | init_special_inode(inode, mode, dev); |
2164 | break; | |
2165 | case S_IFREG: | |
14fcc23f | 2166 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
2167 | inode->i_op = &shmem_inode_operations; |
2168 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
2169 | mpol_shared_policy_init(&info->policy, |
2170 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
2171 | break; |
2172 | case S_IFDIR: | |
d8c76e6f | 2173 | inc_nlink(inode); |
1da177e4 LT |
2174 | /* Some things misbehave if size == 0 on a directory */ |
2175 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2176 | inode->i_op = &shmem_dir_inode_operations; | |
2177 | inode->i_fop = &simple_dir_operations; | |
2178 | break; | |
2179 | case S_IFLNK: | |
2180 | /* | |
2181 | * Must not load anything in the rbtree, | |
2182 | * mpol_free_shared_policy will not be called. | |
2183 | */ | |
71fe804b | 2184 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
2185 | break; |
2186 | } | |
5b04c689 PE |
2187 | } else |
2188 | shmem_free_inode(sb); | |
1da177e4 LT |
2189 | return inode; |
2190 | } | |
2191 | ||
0cd6144a JW |
2192 | bool shmem_mapping(struct address_space *mapping) |
2193 | { | |
f8005451 | 2194 | return mapping->a_ops == &shmem_aops; |
0cd6144a JW |
2195 | } |
2196 | ||
4c27fe4c MR |
2197 | int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm, |
2198 | pmd_t *dst_pmd, | |
2199 | struct vm_area_struct *dst_vma, | |
2200 | unsigned long dst_addr, | |
2201 | unsigned long src_addr, | |
2202 | struct page **pagep) | |
2203 | { | |
2204 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2205 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2206 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
2207 | struct address_space *mapping = inode->i_mapping; | |
2208 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2209 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
2210 | struct mem_cgroup *memcg; | |
2211 | spinlock_t *ptl; | |
2212 | void *page_kaddr; | |
2213 | struct page *page; | |
2214 | pte_t _dst_pte, *dst_pte; | |
2215 | int ret; | |
2216 | ||
cb658a45 AA |
2217 | ret = -ENOMEM; |
2218 | if (shmem_acct_block(info->flags, 1)) | |
2219 | goto out; | |
2220 | if (sbinfo->max_blocks) { | |
2221 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
2222 | sbinfo->max_blocks) >= 0) | |
2223 | goto out_unacct_blocks; | |
2224 | percpu_counter_inc(&sbinfo->used_blocks); | |
2225 | } | |
4c27fe4c | 2226 | |
cb658a45 | 2227 | if (!*pagep) { |
4c27fe4c MR |
2228 | page = shmem_alloc_page(gfp, info, pgoff); |
2229 | if (!page) | |
2230 | goto out_dec_used_blocks; | |
2231 | ||
2232 | page_kaddr = kmap_atomic(page); | |
2233 | ret = copy_from_user(page_kaddr, (const void __user *)src_addr, | |
2234 | PAGE_SIZE); | |
2235 | kunmap_atomic(page_kaddr); | |
2236 | ||
2237 | /* fallback to copy_from_user outside mmap_sem */ | |
2238 | if (unlikely(ret)) { | |
2239 | *pagep = page; | |
cb658a45 AA |
2240 | if (sbinfo->max_blocks) |
2241 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
2242 | shmem_unacct_blocks(info->flags, 1); | |
4c27fe4c MR |
2243 | /* don't free the page */ |
2244 | return -EFAULT; | |
2245 | } | |
2246 | } else { | |
2247 | page = *pagep; | |
2248 | *pagep = NULL; | |
2249 | } | |
2250 | ||
9cc90c66 AA |
2251 | VM_BUG_ON(PageLocked(page) || PageSwapBacked(page)); |
2252 | __SetPageLocked(page); | |
2253 | __SetPageSwapBacked(page); | |
a425d358 | 2254 | __SetPageUptodate(page); |
9cc90c66 | 2255 | |
4c27fe4c MR |
2256 | ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false); |
2257 | if (ret) | |
2258 | goto out_release; | |
2259 | ||
2260 | ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK); | |
2261 | if (!ret) { | |
2262 | ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL); | |
2263 | radix_tree_preload_end(); | |
2264 | } | |
2265 | if (ret) | |
2266 | goto out_release_uncharge; | |
2267 | ||
2268 | mem_cgroup_commit_charge(page, memcg, false, false); | |
2269 | ||
2270 | _dst_pte = mk_pte(page, dst_vma->vm_page_prot); | |
2271 | if (dst_vma->vm_flags & VM_WRITE) | |
2272 | _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte)); | |
2273 | ||
2274 | ret = -EEXIST; | |
2275 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); | |
2276 | if (!pte_none(*dst_pte)) | |
2277 | goto out_release_uncharge_unlock; | |
2278 | ||
4c27fe4c MR |
2279 | lru_cache_add_anon(page); |
2280 | ||
2281 | spin_lock(&info->lock); | |
2282 | info->alloced++; | |
2283 | inode->i_blocks += BLOCKS_PER_PAGE; | |
2284 | shmem_recalc_inode(inode); | |
2285 | spin_unlock(&info->lock); | |
2286 | ||
2287 | inc_mm_counter(dst_mm, mm_counter_file(page)); | |
2288 | page_add_file_rmap(page, false); | |
2289 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
2290 | ||
2291 | /* No need to invalidate - it was non-present before */ | |
2292 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
2293 | unlock_page(page); | |
2294 | pte_unmap_unlock(dst_pte, ptl); | |
2295 | ret = 0; | |
2296 | out: | |
2297 | return ret; | |
2298 | out_release_uncharge_unlock: | |
2299 | pte_unmap_unlock(dst_pte, ptl); | |
2300 | out_release_uncharge: | |
2301 | mem_cgroup_cancel_charge(page, memcg, false); | |
2302 | out_release: | |
9cc90c66 | 2303 | unlock_page(page); |
4c27fe4c MR |
2304 | put_page(page); |
2305 | out_dec_used_blocks: | |
2306 | if (sbinfo->max_blocks) | |
2307 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
2308 | out_unacct_blocks: | |
2309 | shmem_unacct_blocks(info->flags, 1); | |
2310 | goto out; | |
2311 | } | |
2312 | ||
1da177e4 | 2313 | #ifdef CONFIG_TMPFS |
92e1d5be | 2314 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2315 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2316 | |
6d9d88d0 JS |
2317 | #ifdef CONFIG_TMPFS_XATTR |
2318 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2319 | #else | |
2320 | #define shmem_initxattrs NULL | |
2321 | #endif | |
2322 | ||
1da177e4 | 2323 | static int |
800d15a5 NP |
2324 | shmem_write_begin(struct file *file, struct address_space *mapping, |
2325 | loff_t pos, unsigned len, unsigned flags, | |
2326 | struct page **pagep, void **fsdata) | |
1da177e4 | 2327 | { |
800d15a5 | 2328 | struct inode *inode = mapping->host; |
40e041a2 | 2329 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2330 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
2331 | |
2332 | /* i_mutex is held by caller */ | |
2333 | if (unlikely(info->seals)) { | |
2334 | if (info->seals & F_SEAL_WRITE) | |
2335 | return -EPERM; | |
2336 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2337 | return -EPERM; | |
2338 | } | |
2339 | ||
9e18eb29 | 2340 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
2341 | } |
2342 | ||
2343 | static int | |
2344 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2345 | loff_t pos, unsigned len, unsigned copied, | |
2346 | struct page *page, void *fsdata) | |
2347 | { | |
2348 | struct inode *inode = mapping->host; | |
2349 | ||
d3602444 HD |
2350 | if (pos + copied > inode->i_size) |
2351 | i_size_write(inode, pos + copied); | |
2352 | ||
ec9516fb | 2353 | if (!PageUptodate(page)) { |
800d8c63 KS |
2354 | struct page *head = compound_head(page); |
2355 | if (PageTransCompound(page)) { | |
2356 | int i; | |
2357 | ||
2358 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2359 | if (head + i == page) | |
2360 | continue; | |
2361 | clear_highpage(head + i); | |
2362 | flush_dcache_page(head + i); | |
2363 | } | |
2364 | } | |
09cbfeaf KS |
2365 | if (copied < PAGE_SIZE) { |
2366 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2367 | zero_user_segments(page, 0, from, |
09cbfeaf | 2368 | from + copied, PAGE_SIZE); |
ec9516fb | 2369 | } |
800d8c63 | 2370 | SetPageUptodate(head); |
ec9516fb | 2371 | } |
800d15a5 | 2372 | set_page_dirty(page); |
6746aff7 | 2373 | unlock_page(page); |
09cbfeaf | 2374 | put_page(page); |
800d15a5 | 2375 | |
800d15a5 | 2376 | return copied; |
1da177e4 LT |
2377 | } |
2378 | ||
2ba5bbed | 2379 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2380 | { |
6e58e79d AV |
2381 | struct file *file = iocb->ki_filp; |
2382 | struct inode *inode = file_inode(file); | |
1da177e4 | 2383 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2384 | pgoff_t index; |
2385 | unsigned long offset; | |
a0ee5ec5 | 2386 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2387 | int error = 0; |
cb66a7a1 | 2388 | ssize_t retval = 0; |
6e58e79d | 2389 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2390 | |
2391 | /* | |
2392 | * Might this read be for a stacking filesystem? Then when reading | |
2393 | * holes of a sparse file, we actually need to allocate those pages, | |
2394 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2395 | */ | |
777eda2c | 2396 | if (!iter_is_iovec(to)) |
75edd345 | 2397 | sgp = SGP_CACHE; |
1da177e4 | 2398 | |
09cbfeaf KS |
2399 | index = *ppos >> PAGE_SHIFT; |
2400 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2401 | |
2402 | for (;;) { | |
2403 | struct page *page = NULL; | |
41ffe5d5 HD |
2404 | pgoff_t end_index; |
2405 | unsigned long nr, ret; | |
1da177e4 LT |
2406 | loff_t i_size = i_size_read(inode); |
2407 | ||
09cbfeaf | 2408 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2409 | if (index > end_index) |
2410 | break; | |
2411 | if (index == end_index) { | |
09cbfeaf | 2412 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2413 | if (nr <= offset) |
2414 | break; | |
2415 | } | |
2416 | ||
9e18eb29 | 2417 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2418 | if (error) { |
2419 | if (error == -EINVAL) | |
2420 | error = 0; | |
1da177e4 LT |
2421 | break; |
2422 | } | |
75edd345 HD |
2423 | if (page) { |
2424 | if (sgp == SGP_CACHE) | |
2425 | set_page_dirty(page); | |
d3602444 | 2426 | unlock_page(page); |
75edd345 | 2427 | } |
1da177e4 LT |
2428 | |
2429 | /* | |
2430 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2431 | * are called without i_mutex protection against truncate |
1da177e4 | 2432 | */ |
09cbfeaf | 2433 | nr = PAGE_SIZE; |
1da177e4 | 2434 | i_size = i_size_read(inode); |
09cbfeaf | 2435 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2436 | if (index == end_index) { |
09cbfeaf | 2437 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2438 | if (nr <= offset) { |
2439 | if (page) | |
09cbfeaf | 2440 | put_page(page); |
1da177e4 LT |
2441 | break; |
2442 | } | |
2443 | } | |
2444 | nr -= offset; | |
2445 | ||
2446 | if (page) { | |
2447 | /* | |
2448 | * If users can be writing to this page using arbitrary | |
2449 | * virtual addresses, take care about potential aliasing | |
2450 | * before reading the page on the kernel side. | |
2451 | */ | |
2452 | if (mapping_writably_mapped(mapping)) | |
2453 | flush_dcache_page(page); | |
2454 | /* | |
2455 | * Mark the page accessed if we read the beginning. | |
2456 | */ | |
2457 | if (!offset) | |
2458 | mark_page_accessed(page); | |
b5810039 | 2459 | } else { |
1da177e4 | 2460 | page = ZERO_PAGE(0); |
09cbfeaf | 2461 | get_page(page); |
b5810039 | 2462 | } |
1da177e4 LT |
2463 | |
2464 | /* | |
2465 | * Ok, we have the page, and it's up-to-date, so | |
2466 | * now we can copy it to user space... | |
1da177e4 | 2467 | */ |
2ba5bbed | 2468 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2469 | retval += ret; |
1da177e4 | 2470 | offset += ret; |
09cbfeaf KS |
2471 | index += offset >> PAGE_SHIFT; |
2472 | offset &= ~PAGE_MASK; | |
1da177e4 | 2473 | |
09cbfeaf | 2474 | put_page(page); |
2ba5bbed | 2475 | if (!iov_iter_count(to)) |
1da177e4 | 2476 | break; |
6e58e79d AV |
2477 | if (ret < nr) { |
2478 | error = -EFAULT; | |
2479 | break; | |
2480 | } | |
1da177e4 LT |
2481 | cond_resched(); |
2482 | } | |
2483 | ||
09cbfeaf | 2484 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2485 | file_accessed(file); |
2486 | return retval ? retval : error; | |
1da177e4 LT |
2487 | } |
2488 | ||
220f2ac9 HD |
2489 | /* |
2490 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
2491 | */ | |
2492 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2493 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2494 | { |
2495 | struct page *page; | |
2496 | struct pagevec pvec; | |
2497 | pgoff_t indices[PAGEVEC_SIZE]; | |
2498 | bool done = false; | |
2499 | int i; | |
2500 | ||
2501 | pagevec_init(&pvec, 0); | |
2502 | pvec.nr = 1; /* start small: we may be there already */ | |
2503 | while (!done) { | |
0cd6144a | 2504 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2505 | pvec.nr, pvec.pages, indices); |
2506 | if (!pvec.nr) { | |
965c8e59 | 2507 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2508 | index = end; |
2509 | break; | |
2510 | } | |
2511 | for (i = 0; i < pvec.nr; i++, index++) { | |
2512 | if (index < indices[i]) { | |
965c8e59 | 2513 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2514 | done = true; |
2515 | break; | |
2516 | } | |
2517 | index = indices[i]; | |
2518 | } | |
2519 | page = pvec.pages[i]; | |
2520 | if (page && !radix_tree_exceptional_entry(page)) { | |
2521 | if (!PageUptodate(page)) | |
2522 | page = NULL; | |
2523 | } | |
2524 | if (index >= end || | |
965c8e59 AM |
2525 | (page && whence == SEEK_DATA) || |
2526 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2527 | done = true; |
2528 | break; | |
2529 | } | |
2530 | } | |
0cd6144a | 2531 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2532 | pagevec_release(&pvec); |
2533 | pvec.nr = PAGEVEC_SIZE; | |
2534 | cond_resched(); | |
2535 | } | |
2536 | return index; | |
2537 | } | |
2538 | ||
965c8e59 | 2539 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2540 | { |
2541 | struct address_space *mapping = file->f_mapping; | |
2542 | struct inode *inode = mapping->host; | |
2543 | pgoff_t start, end; | |
2544 | loff_t new_offset; | |
2545 | ||
965c8e59 AM |
2546 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2547 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2548 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2549 | inode_lock(inode); |
220f2ac9 HD |
2550 | /* We're holding i_mutex so we can access i_size directly */ |
2551 | ||
2552 | if (offset < 0) | |
2553 | offset = -EINVAL; | |
2554 | else if (offset >= inode->i_size) | |
2555 | offset = -ENXIO; | |
2556 | else { | |
09cbfeaf KS |
2557 | start = offset >> PAGE_SHIFT; |
2558 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2559 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2560 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2561 | if (new_offset > offset) { |
2562 | if (new_offset < inode->i_size) | |
2563 | offset = new_offset; | |
965c8e59 | 2564 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2565 | offset = -ENXIO; |
2566 | else | |
2567 | offset = inode->i_size; | |
2568 | } | |
2569 | } | |
2570 | ||
387aae6f HD |
2571 | if (offset >= 0) |
2572 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2573 | inode_unlock(inode); |
220f2ac9 HD |
2574 | return offset; |
2575 | } | |
2576 | ||
05f65b5c DH |
2577 | /* |
2578 | * We need a tag: a new tag would expand every radix_tree_node by 8 bytes, | |
2579 | * so reuse a tag which we firmly believe is never set or cleared on shmem. | |
2580 | */ | |
2581 | #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE | |
2582 | #define LAST_SCAN 4 /* about 150ms max */ | |
2583 | ||
2584 | static void shmem_tag_pins(struct address_space *mapping) | |
2585 | { | |
2586 | struct radix_tree_iter iter; | |
2587 | void **slot; | |
2588 | pgoff_t start; | |
2589 | struct page *page; | |
2590 | ||
2591 | lru_add_drain(); | |
2592 | start = 0; | |
2593 | rcu_read_lock(); | |
2594 | ||
05f65b5c DH |
2595 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
2596 | page = radix_tree_deref_slot(slot); | |
2597 | if (!page || radix_tree_exception(page)) { | |
2cf938aa MW |
2598 | if (radix_tree_deref_retry(page)) { |
2599 | slot = radix_tree_iter_retry(&iter); | |
2600 | continue; | |
2601 | } | |
05f65b5c DH |
2602 | } else if (page_count(page) - page_mapcount(page) > 1) { |
2603 | spin_lock_irq(&mapping->tree_lock); | |
2604 | radix_tree_tag_set(&mapping->page_tree, iter.index, | |
2605 | SHMEM_TAG_PINNED); | |
2606 | spin_unlock_irq(&mapping->tree_lock); | |
2607 | } | |
2608 | ||
2609 | if (need_resched()) { | |
148deab2 | 2610 | slot = radix_tree_iter_resume(slot, &iter); |
05f65b5c | 2611 | cond_resched_rcu(); |
05f65b5c DH |
2612 | } |
2613 | } | |
2614 | rcu_read_unlock(); | |
2615 | } | |
2616 | ||
2617 | /* | |
2618 | * Setting SEAL_WRITE requires us to verify there's no pending writer. However, | |
2619 | * via get_user_pages(), drivers might have some pending I/O without any active | |
2620 | * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages | |
2621 | * and see whether it has an elevated ref-count. If so, we tag them and wait for | |
2622 | * them to be dropped. | |
2623 | * The caller must guarantee that no new user will acquire writable references | |
2624 | * to those pages to avoid races. | |
2625 | */ | |
40e041a2 DH |
2626 | static int shmem_wait_for_pins(struct address_space *mapping) |
2627 | { | |
05f65b5c DH |
2628 | struct radix_tree_iter iter; |
2629 | void **slot; | |
2630 | pgoff_t start; | |
2631 | struct page *page; | |
2632 | int error, scan; | |
2633 | ||
2634 | shmem_tag_pins(mapping); | |
2635 | ||
2636 | error = 0; | |
2637 | for (scan = 0; scan <= LAST_SCAN; scan++) { | |
2638 | if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED)) | |
2639 | break; | |
2640 | ||
2641 | if (!scan) | |
2642 | lru_add_drain_all(); | |
2643 | else if (schedule_timeout_killable((HZ << scan) / 200)) | |
2644 | scan = LAST_SCAN; | |
2645 | ||
2646 | start = 0; | |
2647 | rcu_read_lock(); | |
05f65b5c DH |
2648 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, |
2649 | start, SHMEM_TAG_PINNED) { | |
2650 | ||
2651 | page = radix_tree_deref_slot(slot); | |
2652 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2653 | if (radix_tree_deref_retry(page)) { |
2654 | slot = radix_tree_iter_retry(&iter); | |
2655 | continue; | |
2656 | } | |
05f65b5c DH |
2657 | |
2658 | page = NULL; | |
2659 | } | |
2660 | ||
2661 | if (page && | |
2662 | page_count(page) - page_mapcount(page) != 1) { | |
2663 | if (scan < LAST_SCAN) | |
2664 | goto continue_resched; | |
2665 | ||
2666 | /* | |
2667 | * On the last scan, we clean up all those tags | |
2668 | * we inserted; but make a note that we still | |
2669 | * found pages pinned. | |
2670 | */ | |
2671 | error = -EBUSY; | |
2672 | } | |
2673 | ||
2674 | spin_lock_irq(&mapping->tree_lock); | |
2675 | radix_tree_tag_clear(&mapping->page_tree, | |
2676 | iter.index, SHMEM_TAG_PINNED); | |
2677 | spin_unlock_irq(&mapping->tree_lock); | |
2678 | continue_resched: | |
2679 | if (need_resched()) { | |
148deab2 | 2680 | slot = radix_tree_iter_resume(slot, &iter); |
05f65b5c | 2681 | cond_resched_rcu(); |
05f65b5c DH |
2682 | } |
2683 | } | |
2684 | rcu_read_unlock(); | |
2685 | } | |
2686 | ||
2687 | return error; | |
40e041a2 DH |
2688 | } |
2689 | ||
2690 | #define F_ALL_SEALS (F_SEAL_SEAL | \ | |
2691 | F_SEAL_SHRINK | \ | |
2692 | F_SEAL_GROW | \ | |
2693 | F_SEAL_WRITE) | |
2694 | ||
2695 | int shmem_add_seals(struct file *file, unsigned int seals) | |
2696 | { | |
2697 | struct inode *inode = file_inode(file); | |
2698 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2699 | int error; | |
2700 | ||
2701 | /* | |
2702 | * SEALING | |
2703 | * Sealing allows multiple parties to share a shmem-file but restrict | |
2704 | * access to a specific subset of file operations. Seals can only be | |
2705 | * added, but never removed. This way, mutually untrusted parties can | |
2706 | * share common memory regions with a well-defined policy. A malicious | |
2707 | * peer can thus never perform unwanted operations on a shared object. | |
2708 | * | |
2709 | * Seals are only supported on special shmem-files and always affect | |
2710 | * the whole underlying inode. Once a seal is set, it may prevent some | |
2711 | * kinds of access to the file. Currently, the following seals are | |
2712 | * defined: | |
2713 | * SEAL_SEAL: Prevent further seals from being set on this file | |
2714 | * SEAL_SHRINK: Prevent the file from shrinking | |
2715 | * SEAL_GROW: Prevent the file from growing | |
2716 | * SEAL_WRITE: Prevent write access to the file | |
2717 | * | |
2718 | * As we don't require any trust relationship between two parties, we | |
2719 | * must prevent seals from being removed. Therefore, sealing a file | |
2720 | * only adds a given set of seals to the file, it never touches | |
2721 | * existing seals. Furthermore, the "setting seals"-operation can be | |
2722 | * sealed itself, which basically prevents any further seal from being | |
2723 | * added. | |
2724 | * | |
2725 | * Semantics of sealing are only defined on volatile files. Only | |
2726 | * anonymous shmem files support sealing. More importantly, seals are | |
2727 | * never written to disk. Therefore, there's no plan to support it on | |
2728 | * other file types. | |
2729 | */ | |
2730 | ||
2731 | if (file->f_op != &shmem_file_operations) | |
2732 | return -EINVAL; | |
2733 | if (!(file->f_mode & FMODE_WRITE)) | |
2734 | return -EPERM; | |
2735 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
2736 | return -EINVAL; | |
2737 | ||
5955102c | 2738 | inode_lock(inode); |
40e041a2 DH |
2739 | |
2740 | if (info->seals & F_SEAL_SEAL) { | |
2741 | error = -EPERM; | |
2742 | goto unlock; | |
2743 | } | |
2744 | ||
2745 | if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) { | |
2746 | error = mapping_deny_writable(file->f_mapping); | |
2747 | if (error) | |
2748 | goto unlock; | |
2749 | ||
2750 | error = shmem_wait_for_pins(file->f_mapping); | |
2751 | if (error) { | |
2752 | mapping_allow_writable(file->f_mapping); | |
2753 | goto unlock; | |
2754 | } | |
2755 | } | |
2756 | ||
2757 | info->seals |= seals; | |
2758 | error = 0; | |
2759 | ||
2760 | unlock: | |
5955102c | 2761 | inode_unlock(inode); |
40e041a2 DH |
2762 | return error; |
2763 | } | |
2764 | EXPORT_SYMBOL_GPL(shmem_add_seals); | |
2765 | ||
2766 | int shmem_get_seals(struct file *file) | |
2767 | { | |
2768 | if (file->f_op != &shmem_file_operations) | |
2769 | return -EINVAL; | |
2770 | ||
2771 | return SHMEM_I(file_inode(file))->seals; | |
2772 | } | |
2773 | EXPORT_SYMBOL_GPL(shmem_get_seals); | |
2774 | ||
2775 | long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg) | |
2776 | { | |
2777 | long error; | |
2778 | ||
2779 | switch (cmd) { | |
2780 | case F_ADD_SEALS: | |
2781 | /* disallow upper 32bit */ | |
2782 | if (arg > UINT_MAX) | |
2783 | return -EINVAL; | |
2784 | ||
2785 | error = shmem_add_seals(file, arg); | |
2786 | break; | |
2787 | case F_GET_SEALS: | |
2788 | error = shmem_get_seals(file); | |
2789 | break; | |
2790 | default: | |
2791 | error = -EINVAL; | |
2792 | break; | |
2793 | } | |
2794 | ||
2795 | return error; | |
2796 | } | |
2797 | ||
83e4fa9c HD |
2798 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2799 | loff_t len) | |
2800 | { | |
496ad9aa | 2801 | struct inode *inode = file_inode(file); |
e2d12e22 | 2802 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2803 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2804 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2805 | pgoff_t start, index, end; |
2806 | int error; | |
83e4fa9c | 2807 | |
13ace4d0 HD |
2808 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2809 | return -EOPNOTSUPP; | |
2810 | ||
5955102c | 2811 | inode_lock(inode); |
83e4fa9c HD |
2812 | |
2813 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2814 | struct address_space *mapping = file->f_mapping; | |
2815 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2816 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2817 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2818 | |
40e041a2 DH |
2819 | /* protected by i_mutex */ |
2820 | if (info->seals & F_SEAL_WRITE) { | |
2821 | error = -EPERM; | |
2822 | goto out; | |
2823 | } | |
2824 | ||
8e205f77 | 2825 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
2826 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
2827 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
2828 | spin_lock(&inode->i_lock); | |
2829 | inode->i_private = &shmem_falloc; | |
2830 | spin_unlock(&inode->i_lock); | |
2831 | ||
83e4fa9c HD |
2832 | if ((u64)unmap_end > (u64)unmap_start) |
2833 | unmap_mapping_range(mapping, unmap_start, | |
2834 | 1 + unmap_end - unmap_start, 0); | |
2835 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2836 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2837 | |
2838 | spin_lock(&inode->i_lock); | |
2839 | inode->i_private = NULL; | |
2840 | wake_up_all(&shmem_falloc_waitq); | |
10d20bd2 | 2841 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.task_list)); |
8e205f77 | 2842 | spin_unlock(&inode->i_lock); |
83e4fa9c | 2843 | error = 0; |
8e205f77 | 2844 | goto out; |
e2d12e22 HD |
2845 | } |
2846 | ||
2847 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2848 | error = inode_newsize_ok(inode, offset + len); | |
2849 | if (error) | |
2850 | goto out; | |
2851 | ||
40e041a2 DH |
2852 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2853 | error = -EPERM; | |
2854 | goto out; | |
2855 | } | |
2856 | ||
09cbfeaf KS |
2857 | start = offset >> PAGE_SHIFT; |
2858 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2859 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2860 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2861 | error = -ENOSPC; | |
2862 | goto out; | |
83e4fa9c HD |
2863 | } |
2864 | ||
8e205f77 | 2865 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2866 | shmem_falloc.start = start; |
2867 | shmem_falloc.next = start; | |
2868 | shmem_falloc.nr_falloced = 0; | |
2869 | shmem_falloc.nr_unswapped = 0; | |
2870 | spin_lock(&inode->i_lock); | |
2871 | inode->i_private = &shmem_falloc; | |
2872 | spin_unlock(&inode->i_lock); | |
2873 | ||
e2d12e22 HD |
2874 | for (index = start; index < end; index++) { |
2875 | struct page *page; | |
2876 | ||
2877 | /* | |
2878 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2879 | * been interrupted because we are using up too much memory. | |
2880 | */ | |
2881 | if (signal_pending(current)) | |
2882 | error = -EINTR; | |
1aac1400 HD |
2883 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2884 | error = -ENOMEM; | |
e2d12e22 | 2885 | else |
9e18eb29 | 2886 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2887 | if (error) { |
1635f6a7 | 2888 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2889 | if (index > start) { |
2890 | shmem_undo_range(inode, | |
2891 | (loff_t)start << PAGE_SHIFT, | |
2892 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2893 | } | |
1aac1400 | 2894 | goto undone; |
e2d12e22 HD |
2895 | } |
2896 | ||
1aac1400 HD |
2897 | /* |
2898 | * Inform shmem_writepage() how far we have reached. | |
2899 | * No need for lock or barrier: we have the page lock. | |
2900 | */ | |
2901 | shmem_falloc.next++; | |
2902 | if (!PageUptodate(page)) | |
2903 | shmem_falloc.nr_falloced++; | |
2904 | ||
e2d12e22 | 2905 | /* |
1635f6a7 HD |
2906 | * If !PageUptodate, leave it that way so that freeable pages |
2907 | * can be recognized if we need to rollback on error later. | |
2908 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2909 | * than free the pages we are allocating (and SGP_CACHE pages |
2910 | * might still be clean: we now need to mark those dirty too). | |
2911 | */ | |
2912 | set_page_dirty(page); | |
2913 | unlock_page(page); | |
09cbfeaf | 2914 | put_page(page); |
e2d12e22 HD |
2915 | cond_resched(); |
2916 | } | |
2917 | ||
2918 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2919 | i_size_write(inode, offset + len); | |
078cd827 | 2920 | inode->i_ctime = current_time(inode); |
1aac1400 HD |
2921 | undone: |
2922 | spin_lock(&inode->i_lock); | |
2923 | inode->i_private = NULL; | |
2924 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2925 | out: |
5955102c | 2926 | inode_unlock(inode); |
83e4fa9c HD |
2927 | return error; |
2928 | } | |
2929 | ||
726c3342 | 2930 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2931 | { |
726c3342 | 2932 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2933 | |
2934 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2935 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2936 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2937 | if (sbinfo->max_blocks) { |
1da177e4 | 2938 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2939 | buf->f_bavail = |
2940 | buf->f_bfree = sbinfo->max_blocks - | |
2941 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2942 | } |
2943 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2944 | buf->f_files = sbinfo->max_inodes; |
2945 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2946 | } |
2947 | /* else leave those fields 0 like simple_statfs */ | |
2948 | return 0; | |
2949 | } | |
2950 | ||
2951 | /* | |
2952 | * File creation. Allocate an inode, and we're done.. | |
2953 | */ | |
2954 | static int | |
1a67aafb | 2955 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2956 | { |
0b0a0806 | 2957 | struct inode *inode; |
1da177e4 LT |
2958 | int error = -ENOSPC; |
2959 | ||
454abafe | 2960 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2961 | if (inode) { |
feda821e CH |
2962 | error = simple_acl_create(dir, inode); |
2963 | if (error) | |
2964 | goto out_iput; | |
2a7dba39 | 2965 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2966 | &dentry->d_name, |
6d9d88d0 | 2967 | shmem_initxattrs, NULL); |
feda821e CH |
2968 | if (error && error != -EOPNOTSUPP) |
2969 | goto out_iput; | |
37ec43cd | 2970 | |
718deb6b | 2971 | error = 0; |
1da177e4 | 2972 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 2973 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
2974 | d_instantiate(dentry, inode); |
2975 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2976 | } |
2977 | return error; | |
feda821e CH |
2978 | out_iput: |
2979 | iput(inode); | |
2980 | return error; | |
1da177e4 LT |
2981 | } |
2982 | ||
60545d0d AV |
2983 | static int |
2984 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2985 | { | |
2986 | struct inode *inode; | |
2987 | int error = -ENOSPC; | |
2988 | ||
2989 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
2990 | if (inode) { | |
2991 | error = security_inode_init_security(inode, dir, | |
2992 | NULL, | |
2993 | shmem_initxattrs, NULL); | |
feda821e CH |
2994 | if (error && error != -EOPNOTSUPP) |
2995 | goto out_iput; | |
2996 | error = simple_acl_create(dir, inode); | |
2997 | if (error) | |
2998 | goto out_iput; | |
60545d0d AV |
2999 | d_tmpfile(dentry, inode); |
3000 | } | |
3001 | return error; | |
feda821e CH |
3002 | out_iput: |
3003 | iput(inode); | |
3004 | return error; | |
60545d0d AV |
3005 | } |
3006 | ||
18bb1db3 | 3007 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
3008 | { |
3009 | int error; | |
3010 | ||
3011 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
3012 | return error; | |
d8c76e6f | 3013 | inc_nlink(dir); |
1da177e4 LT |
3014 | return 0; |
3015 | } | |
3016 | ||
4acdaf27 | 3017 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 3018 | bool excl) |
1da177e4 LT |
3019 | { |
3020 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
3021 | } | |
3022 | ||
3023 | /* | |
3024 | * Link a file.. | |
3025 | */ | |
3026 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
3027 | { | |
75c3cfa8 | 3028 | struct inode *inode = d_inode(old_dentry); |
5b04c689 | 3029 | int ret; |
1da177e4 LT |
3030 | |
3031 | /* | |
3032 | * No ordinary (disk based) filesystem counts links as inodes; | |
3033 | * but each new link needs a new dentry, pinning lowmem, and | |
3034 | * tmpfs dentries cannot be pruned until they are unlinked. | |
3035 | */ | |
5b04c689 PE |
3036 | ret = shmem_reserve_inode(inode->i_sb); |
3037 | if (ret) | |
3038 | goto out; | |
1da177e4 LT |
3039 | |
3040 | dir->i_size += BOGO_DIRENT_SIZE; | |
078cd827 | 3041 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
d8c76e6f | 3042 | inc_nlink(inode); |
7de9c6ee | 3043 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
3044 | dget(dentry); /* Extra pinning count for the created dentry */ |
3045 | d_instantiate(dentry, inode); | |
5b04c689 PE |
3046 | out: |
3047 | return ret; | |
1da177e4 LT |
3048 | } |
3049 | ||
3050 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
3051 | { | |
75c3cfa8 | 3052 | struct inode *inode = d_inode(dentry); |
1da177e4 | 3053 | |
5b04c689 PE |
3054 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
3055 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
3056 | |
3057 | dir->i_size -= BOGO_DIRENT_SIZE; | |
078cd827 | 3058 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
9a53c3a7 | 3059 | drop_nlink(inode); |
1da177e4 LT |
3060 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
3061 | return 0; | |
3062 | } | |
3063 | ||
3064 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
3065 | { | |
3066 | if (!simple_empty(dentry)) | |
3067 | return -ENOTEMPTY; | |
3068 | ||
75c3cfa8 | 3069 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 3070 | drop_nlink(dir); |
1da177e4 LT |
3071 | return shmem_unlink(dir, dentry); |
3072 | } | |
3073 | ||
37456771 MS |
3074 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
3075 | { | |
e36cb0b8 DH |
3076 | bool old_is_dir = d_is_dir(old_dentry); |
3077 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
3078 | |
3079 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
3080 | if (old_is_dir) { | |
3081 | drop_nlink(old_dir); | |
3082 | inc_nlink(new_dir); | |
3083 | } else { | |
3084 | drop_nlink(new_dir); | |
3085 | inc_nlink(old_dir); | |
3086 | } | |
3087 | } | |
3088 | old_dir->i_ctime = old_dir->i_mtime = | |
3089 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 | 3090 | d_inode(old_dentry)->i_ctime = |
078cd827 | 3091 | d_inode(new_dentry)->i_ctime = current_time(old_dir); |
37456771 MS |
3092 | |
3093 | return 0; | |
3094 | } | |
3095 | ||
46fdb794 MS |
3096 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
3097 | { | |
3098 | struct dentry *whiteout; | |
3099 | int error; | |
3100 | ||
3101 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
3102 | if (!whiteout) | |
3103 | return -ENOMEM; | |
3104 | ||
3105 | error = shmem_mknod(old_dir, whiteout, | |
3106 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
3107 | dput(whiteout); | |
3108 | if (error) | |
3109 | return error; | |
3110 | ||
3111 | /* | |
3112 | * Cheat and hash the whiteout while the old dentry is still in | |
3113 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
3114 | * | |
3115 | * d_lookup() will consistently find one of them at this point, | |
3116 | * not sure which one, but that isn't even important. | |
3117 | */ | |
3118 | d_rehash(whiteout); | |
3119 | return 0; | |
3120 | } | |
3121 | ||
1da177e4 LT |
3122 | /* |
3123 | * The VFS layer already does all the dentry stuff for rename, | |
3124 | * we just have to decrement the usage count for the target if | |
3125 | * it exists so that the VFS layer correctly free's it when it | |
3126 | * gets overwritten. | |
3127 | */ | |
3b69ff51 | 3128 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 3129 | { |
75c3cfa8 | 3130 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
3131 | int they_are_dirs = S_ISDIR(inode->i_mode); |
3132 | ||
46fdb794 | 3133 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
3134 | return -EINVAL; |
3135 | ||
37456771 MS |
3136 | if (flags & RENAME_EXCHANGE) |
3137 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
3138 | ||
1da177e4 LT |
3139 | if (!simple_empty(new_dentry)) |
3140 | return -ENOTEMPTY; | |
3141 | ||
46fdb794 MS |
3142 | if (flags & RENAME_WHITEOUT) { |
3143 | int error; | |
3144 | ||
3145 | error = shmem_whiteout(old_dir, old_dentry); | |
3146 | if (error) | |
3147 | return error; | |
3148 | } | |
3149 | ||
75c3cfa8 | 3150 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 3151 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 3152 | if (they_are_dirs) { |
75c3cfa8 | 3153 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 3154 | drop_nlink(old_dir); |
b928095b | 3155 | } |
1da177e4 | 3156 | } else if (they_are_dirs) { |
9a53c3a7 | 3157 | drop_nlink(old_dir); |
d8c76e6f | 3158 | inc_nlink(new_dir); |
1da177e4 LT |
3159 | } |
3160 | ||
3161 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
3162 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
3163 | old_dir->i_ctime = old_dir->i_mtime = | |
3164 | new_dir->i_ctime = new_dir->i_mtime = | |
078cd827 | 3165 | inode->i_ctime = current_time(old_dir); |
1da177e4 LT |
3166 | return 0; |
3167 | } | |
3168 | ||
3169 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
3170 | { | |
3171 | int error; | |
3172 | int len; | |
3173 | struct inode *inode; | |
9276aad6 | 3174 | struct page *page; |
1da177e4 LT |
3175 | struct shmem_inode_info *info; |
3176 | ||
3177 | len = strlen(symname) + 1; | |
09cbfeaf | 3178 | if (len > PAGE_SIZE) |
1da177e4 LT |
3179 | return -ENAMETOOLONG; |
3180 | ||
454abafe | 3181 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
3182 | if (!inode) |
3183 | return -ENOSPC; | |
3184 | ||
9d8f13ba | 3185 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3186 | shmem_initxattrs, NULL); |
570bc1c2 SS |
3187 | if (error) { |
3188 | if (error != -EOPNOTSUPP) { | |
3189 | iput(inode); | |
3190 | return error; | |
3191 | } | |
3192 | error = 0; | |
3193 | } | |
3194 | ||
1da177e4 LT |
3195 | info = SHMEM_I(inode); |
3196 | inode->i_size = len-1; | |
69f07ec9 | 3197 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3198 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3199 | if (!inode->i_link) { | |
69f07ec9 HD |
3200 | iput(inode); |
3201 | return -ENOMEM; | |
3202 | } | |
3203 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3204 | } else { |
e8ecde25 | 3205 | inode_nohighmem(inode); |
9e18eb29 | 3206 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
3207 | if (error) { |
3208 | iput(inode); | |
3209 | return error; | |
3210 | } | |
14fcc23f | 3211 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 3212 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 3213 | memcpy(page_address(page), symname, len); |
ec9516fb | 3214 | SetPageUptodate(page); |
1da177e4 | 3215 | set_page_dirty(page); |
6746aff7 | 3216 | unlock_page(page); |
09cbfeaf | 3217 | put_page(page); |
1da177e4 | 3218 | } |
1da177e4 | 3219 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3220 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3221 | d_instantiate(dentry, inode); |
3222 | dget(dentry); | |
3223 | return 0; | |
3224 | } | |
3225 | ||
fceef393 | 3226 | static void shmem_put_link(void *arg) |
1da177e4 | 3227 | { |
fceef393 AV |
3228 | mark_page_accessed(arg); |
3229 | put_page(arg); | |
1da177e4 LT |
3230 | } |
3231 | ||
6b255391 | 3232 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
3233 | struct inode *inode, |
3234 | struct delayed_call *done) | |
1da177e4 | 3235 | { |
1da177e4 | 3236 | struct page *page = NULL; |
6b255391 | 3237 | int error; |
6a6c9904 AV |
3238 | if (!dentry) { |
3239 | page = find_get_page(inode->i_mapping, 0); | |
3240 | if (!page) | |
3241 | return ERR_PTR(-ECHILD); | |
3242 | if (!PageUptodate(page)) { | |
3243 | put_page(page); | |
3244 | return ERR_PTR(-ECHILD); | |
3245 | } | |
3246 | } else { | |
9e18eb29 | 3247 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3248 | if (error) |
3249 | return ERR_PTR(error); | |
3250 | unlock_page(page); | |
3251 | } | |
fceef393 | 3252 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3253 | return page_address(page); |
1da177e4 LT |
3254 | } |
3255 | ||
b09e0fa4 | 3256 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3257 | /* |
b09e0fa4 EP |
3258 | * Superblocks without xattr inode operations may get some security.* xattr |
3259 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3260 | * like ACLs, we also need to implement the security.* handlers at |
3261 | * filesystem level, though. | |
3262 | */ | |
3263 | ||
6d9d88d0 JS |
3264 | /* |
3265 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3266 | */ | |
3267 | static int shmem_initxattrs(struct inode *inode, | |
3268 | const struct xattr *xattr_array, | |
3269 | void *fs_info) | |
3270 | { | |
3271 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3272 | const struct xattr *xattr; | |
38f38657 | 3273 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3274 | size_t len; |
3275 | ||
3276 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3277 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3278 | if (!new_xattr) |
3279 | return -ENOMEM; | |
3280 | ||
3281 | len = strlen(xattr->name) + 1; | |
3282 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3283 | GFP_KERNEL); | |
3284 | if (!new_xattr->name) { | |
3285 | kfree(new_xattr); | |
3286 | return -ENOMEM; | |
3287 | } | |
3288 | ||
3289 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3290 | XATTR_SECURITY_PREFIX_LEN); | |
3291 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3292 | xattr->name, len); | |
3293 | ||
38f38657 | 3294 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3295 | } |
3296 | ||
3297 | return 0; | |
3298 | } | |
3299 | ||
aa7c5241 | 3300 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3301 | struct dentry *unused, struct inode *inode, |
3302 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3303 | { |
b296821a | 3304 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3305 | |
aa7c5241 | 3306 | name = xattr_full_name(handler, name); |
38f38657 | 3307 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3308 | } |
3309 | ||
aa7c5241 | 3310 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
59301226 AV |
3311 | struct dentry *unused, struct inode *inode, |
3312 | const char *name, const void *value, | |
3313 | size_t size, int flags) | |
b09e0fa4 | 3314 | { |
59301226 | 3315 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3316 | |
aa7c5241 | 3317 | name = xattr_full_name(handler, name); |
38f38657 | 3318 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
3319 | } |
3320 | ||
aa7c5241 AG |
3321 | static const struct xattr_handler shmem_security_xattr_handler = { |
3322 | .prefix = XATTR_SECURITY_PREFIX, | |
3323 | .get = shmem_xattr_handler_get, | |
3324 | .set = shmem_xattr_handler_set, | |
3325 | }; | |
b09e0fa4 | 3326 | |
aa7c5241 AG |
3327 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3328 | .prefix = XATTR_TRUSTED_PREFIX, | |
3329 | .get = shmem_xattr_handler_get, | |
3330 | .set = shmem_xattr_handler_set, | |
3331 | }; | |
b09e0fa4 | 3332 | |
aa7c5241 AG |
3333 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3334 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3335 | &posix_acl_access_xattr_handler, | |
3336 | &posix_acl_default_xattr_handler, | |
3337 | #endif | |
3338 | &shmem_security_xattr_handler, | |
3339 | &shmem_trusted_xattr_handler, | |
3340 | NULL | |
3341 | }; | |
b09e0fa4 EP |
3342 | |
3343 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3344 | { | |
75c3cfa8 | 3345 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3346 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3347 | } |
3348 | #endif /* CONFIG_TMPFS_XATTR */ | |
3349 | ||
69f07ec9 | 3350 | static const struct inode_operations shmem_short_symlink_operations = { |
6b255391 | 3351 | .get_link = simple_get_link, |
b09e0fa4 | 3352 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3353 | .listxattr = shmem_listxattr, |
b09e0fa4 EP |
3354 | #endif |
3355 | }; | |
3356 | ||
3357 | static const struct inode_operations shmem_symlink_inode_operations = { | |
6b255391 | 3358 | .get_link = shmem_get_link, |
b09e0fa4 | 3359 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3360 | .listxattr = shmem_listxattr, |
39f0247d | 3361 | #endif |
b09e0fa4 | 3362 | }; |
39f0247d | 3363 | |
91828a40 DG |
3364 | static struct dentry *shmem_get_parent(struct dentry *child) |
3365 | { | |
3366 | return ERR_PTR(-ESTALE); | |
3367 | } | |
3368 | ||
3369 | static int shmem_match(struct inode *ino, void *vfh) | |
3370 | { | |
3371 | __u32 *fh = vfh; | |
3372 | __u64 inum = fh[2]; | |
3373 | inum = (inum << 32) | fh[1]; | |
3374 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
3375 | } | |
3376 | ||
480b116c CH |
3377 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3378 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3379 | { |
91828a40 | 3380 | struct inode *inode; |
480b116c | 3381 | struct dentry *dentry = NULL; |
35c2a7f4 | 3382 | u64 inum; |
480b116c CH |
3383 | |
3384 | if (fh_len < 3) | |
3385 | return NULL; | |
91828a40 | 3386 | |
35c2a7f4 HD |
3387 | inum = fid->raw[2]; |
3388 | inum = (inum << 32) | fid->raw[1]; | |
3389 | ||
480b116c CH |
3390 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
3391 | shmem_match, fid->raw); | |
91828a40 | 3392 | if (inode) { |
480b116c | 3393 | dentry = d_find_alias(inode); |
91828a40 DG |
3394 | iput(inode); |
3395 | } | |
3396 | ||
480b116c | 3397 | return dentry; |
91828a40 DG |
3398 | } |
3399 | ||
b0b0382b AV |
3400 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3401 | struct inode *parent) | |
91828a40 | 3402 | { |
5fe0c237 AK |
3403 | if (*len < 3) { |
3404 | *len = 3; | |
94e07a75 | 3405 | return FILEID_INVALID; |
5fe0c237 | 3406 | } |
91828a40 | 3407 | |
1d3382cb | 3408 | if (inode_unhashed(inode)) { |
91828a40 DG |
3409 | /* Unfortunately insert_inode_hash is not idempotent, |
3410 | * so as we hash inodes here rather than at creation | |
3411 | * time, we need a lock to ensure we only try | |
3412 | * to do it once | |
3413 | */ | |
3414 | static DEFINE_SPINLOCK(lock); | |
3415 | spin_lock(&lock); | |
1d3382cb | 3416 | if (inode_unhashed(inode)) |
91828a40 DG |
3417 | __insert_inode_hash(inode, |
3418 | inode->i_ino + inode->i_generation); | |
3419 | spin_unlock(&lock); | |
3420 | } | |
3421 | ||
3422 | fh[0] = inode->i_generation; | |
3423 | fh[1] = inode->i_ino; | |
3424 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
3425 | ||
3426 | *len = 3; | |
3427 | return 1; | |
3428 | } | |
3429 | ||
39655164 | 3430 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3431 | .get_parent = shmem_get_parent, |
91828a40 | 3432 | .encode_fh = shmem_encode_fh, |
480b116c | 3433 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3434 | }; |
3435 | ||
680d794b | 3436 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
3437 | bool remount) | |
1da177e4 LT |
3438 | { |
3439 | char *this_char, *value, *rest; | |
49cd0a5c | 3440 | struct mempolicy *mpol = NULL; |
8751e039 EB |
3441 | uid_t uid; |
3442 | gid_t gid; | |
1da177e4 | 3443 | |
b00dc3ad HD |
3444 | while (options != NULL) { |
3445 | this_char = options; | |
3446 | for (;;) { | |
3447 | /* | |
3448 | * NUL-terminate this option: unfortunately, | |
3449 | * mount options form a comma-separated list, | |
3450 | * but mpol's nodelist may also contain commas. | |
3451 | */ | |
3452 | options = strchr(options, ','); | |
3453 | if (options == NULL) | |
3454 | break; | |
3455 | options++; | |
3456 | if (!isdigit(*options)) { | |
3457 | options[-1] = '\0'; | |
3458 | break; | |
3459 | } | |
3460 | } | |
1da177e4 LT |
3461 | if (!*this_char) |
3462 | continue; | |
3463 | if ((value = strchr(this_char,'=')) != NULL) { | |
3464 | *value++ = 0; | |
3465 | } else { | |
1170532b JP |
3466 | pr_err("tmpfs: No value for mount option '%s'\n", |
3467 | this_char); | |
49cd0a5c | 3468 | goto error; |
1da177e4 LT |
3469 | } |
3470 | ||
3471 | if (!strcmp(this_char,"size")) { | |
3472 | unsigned long long size; | |
3473 | size = memparse(value,&rest); | |
3474 | if (*rest == '%') { | |
3475 | size <<= PAGE_SHIFT; | |
3476 | size *= totalram_pages; | |
3477 | do_div(size, 100); | |
3478 | rest++; | |
3479 | } | |
3480 | if (*rest) | |
3481 | goto bad_val; | |
680d794b | 3482 | sbinfo->max_blocks = |
09cbfeaf | 3483 | DIV_ROUND_UP(size, PAGE_SIZE); |
1da177e4 | 3484 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 3485 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
3486 | if (*rest) |
3487 | goto bad_val; | |
3488 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 3489 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
3490 | if (*rest) |
3491 | goto bad_val; | |
3492 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 3493 | if (remount) |
1da177e4 | 3494 | continue; |
680d794b | 3495 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
3496 | if (*rest) |
3497 | goto bad_val; | |
3498 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 3499 | if (remount) |
1da177e4 | 3500 | continue; |
8751e039 | 3501 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3502 | if (*rest) |
3503 | goto bad_val; | |
8751e039 EB |
3504 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
3505 | if (!uid_valid(sbinfo->uid)) | |
3506 | goto bad_val; | |
1da177e4 | 3507 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 3508 | if (remount) |
1da177e4 | 3509 | continue; |
8751e039 | 3510 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3511 | if (*rest) |
3512 | goto bad_val; | |
8751e039 EB |
3513 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
3514 | if (!gid_valid(sbinfo->gid)) | |
3515 | goto bad_val; | |
e496cf3d | 3516 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3517 | } else if (!strcmp(this_char, "huge")) { |
3518 | int huge; | |
3519 | huge = shmem_parse_huge(value); | |
3520 | if (huge < 0) | |
3521 | goto bad_val; | |
3522 | if (!has_transparent_hugepage() && | |
3523 | huge != SHMEM_HUGE_NEVER) | |
3524 | goto bad_val; | |
3525 | sbinfo->huge = huge; | |
3526 | #endif | |
3527 | #ifdef CONFIG_NUMA | |
7339ff83 | 3528 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
3529 | mpol_put(mpol); |
3530 | mpol = NULL; | |
3531 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 3532 | goto bad_val; |
5a6e75f8 | 3533 | #endif |
1da177e4 | 3534 | } else { |
1170532b | 3535 | pr_err("tmpfs: Bad mount option %s\n", this_char); |
49cd0a5c | 3536 | goto error; |
1da177e4 LT |
3537 | } |
3538 | } | |
49cd0a5c | 3539 | sbinfo->mpol = mpol; |
1da177e4 LT |
3540 | return 0; |
3541 | ||
3542 | bad_val: | |
1170532b | 3543 | pr_err("tmpfs: Bad value '%s' for mount option '%s'\n", |
1da177e4 | 3544 | value, this_char); |
49cd0a5c GT |
3545 | error: |
3546 | mpol_put(mpol); | |
1da177e4 LT |
3547 | return 1; |
3548 | ||
3549 | } | |
3550 | ||
3551 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
3552 | { | |
3553 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 3554 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
3555 | unsigned long inodes; |
3556 | int error = -EINVAL; | |
3557 | ||
5f00110f | 3558 | config.mpol = NULL; |
680d794b | 3559 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 3560 | return error; |
1da177e4 | 3561 | |
0edd73b3 | 3562 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3563 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 3564 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 3565 | goto out; |
680d794b | 3566 | if (config.max_inodes < inodes) |
0edd73b3 HD |
3567 | goto out; |
3568 | /* | |
54af6042 | 3569 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
3570 | * but we must separately disallow unlimited->limited, because |
3571 | * in that case we have no record of how much is already in use. | |
3572 | */ | |
680d794b | 3573 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 3574 | goto out; |
680d794b | 3575 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
3576 | goto out; |
3577 | ||
3578 | error = 0; | |
5a6e75f8 | 3579 | sbinfo->huge = config.huge; |
680d794b | 3580 | sbinfo->max_blocks = config.max_blocks; |
680d794b | 3581 | sbinfo->max_inodes = config.max_inodes; |
3582 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 3583 | |
5f00110f GT |
3584 | /* |
3585 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3586 | */ | |
3587 | if (config.mpol) { | |
3588 | mpol_put(sbinfo->mpol); | |
3589 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
3590 | } | |
0edd73b3 HD |
3591 | out: |
3592 | spin_unlock(&sbinfo->stat_lock); | |
3593 | return error; | |
1da177e4 | 3594 | } |
680d794b | 3595 | |
34c80b1d | 3596 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3597 | { |
34c80b1d | 3598 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b | 3599 | |
3600 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3601 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3602 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 3603 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
3604 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
3605 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 3606 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3607 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3608 | seq_printf(seq, ",uid=%u", | |
3609 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3610 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3611 | seq_printf(seq, ",gid=%u", | |
3612 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
e496cf3d | 3613 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3614 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
3615 | if (sbinfo->huge) | |
3616 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3617 | #endif | |
71fe804b | 3618 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 3619 | return 0; |
3620 | } | |
9183df25 DH |
3621 | |
3622 | #define MFD_NAME_PREFIX "memfd:" | |
3623 | #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1) | |
3624 | #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN) | |
3625 | ||
3626 | #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING) | |
3627 | ||
3628 | SYSCALL_DEFINE2(memfd_create, | |
3629 | const char __user *, uname, | |
3630 | unsigned int, flags) | |
3631 | { | |
3632 | struct shmem_inode_info *info; | |
3633 | struct file *file; | |
3634 | int fd, error; | |
3635 | char *name; | |
3636 | long len; | |
3637 | ||
3638 | if (flags & ~(unsigned int)MFD_ALL_FLAGS) | |
3639 | return -EINVAL; | |
3640 | ||
3641 | /* length includes terminating zero */ | |
3642 | len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1); | |
3643 | if (len <= 0) | |
3644 | return -EFAULT; | |
3645 | if (len > MFD_NAME_MAX_LEN + 1) | |
3646 | return -EINVAL; | |
3647 | ||
3648 | name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY); | |
3649 | if (!name) | |
3650 | return -ENOMEM; | |
3651 | ||
3652 | strcpy(name, MFD_NAME_PREFIX); | |
3653 | if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) { | |
3654 | error = -EFAULT; | |
3655 | goto err_name; | |
3656 | } | |
3657 | ||
3658 | /* terminating-zero may have changed after strnlen_user() returned */ | |
3659 | if (name[len + MFD_NAME_PREFIX_LEN - 1]) { | |
3660 | error = -EFAULT; | |
3661 | goto err_name; | |
3662 | } | |
3663 | ||
3664 | fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0); | |
3665 | if (fd < 0) { | |
3666 | error = fd; | |
3667 | goto err_name; | |
3668 | } | |
3669 | ||
3670 | file = shmem_file_setup(name, 0, VM_NORESERVE); | |
3671 | if (IS_ERR(file)) { | |
3672 | error = PTR_ERR(file); | |
3673 | goto err_fd; | |
3674 | } | |
3675 | info = SHMEM_I(file_inode(file)); | |
3676 | file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; | |
3677 | file->f_flags |= O_RDWR | O_LARGEFILE; | |
3678 | if (flags & MFD_ALLOW_SEALING) | |
3679 | info->seals &= ~F_SEAL_SEAL; | |
3680 | ||
3681 | fd_install(fd, file); | |
3682 | kfree(name); | |
3683 | return fd; | |
3684 | ||
3685 | err_fd: | |
3686 | put_unused_fd(fd); | |
3687 | err_name: | |
3688 | kfree(name); | |
3689 | return error; | |
3690 | } | |
3691 | ||
680d794b | 3692 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3693 | |
3694 | static void shmem_put_super(struct super_block *sb) | |
3695 | { | |
602586a8 HD |
3696 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3697 | ||
3698 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 3699 | mpol_put(sbinfo->mpol); |
602586a8 | 3700 | kfree(sbinfo); |
1da177e4 LT |
3701 | sb->s_fs_info = NULL; |
3702 | } | |
3703 | ||
2b2af54a | 3704 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3705 | { |
3706 | struct inode *inode; | |
0edd73b3 | 3707 | struct shmem_sb_info *sbinfo; |
680d794b | 3708 | int err = -ENOMEM; |
3709 | ||
3710 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3711 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 3712 | L1_CACHE_BYTES), GFP_KERNEL); |
3713 | if (!sbinfo) | |
3714 | return -ENOMEM; | |
3715 | ||
680d794b | 3716 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
3717 | sbinfo->uid = current_fsuid(); |
3718 | sbinfo->gid = current_fsgid(); | |
680d794b | 3719 | sb->s_fs_info = sbinfo; |
1da177e4 | 3720 | |
0edd73b3 | 3721 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3722 | /* |
3723 | * Per default we only allow half of the physical ram per | |
3724 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3725 | * but the internal instance is left unlimited. | |
3726 | */ | |
ca4e0519 | 3727 | if (!(sb->s_flags & MS_KERNMOUNT)) { |
680d794b | 3728 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3729 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3730 | if (shmem_parse_options(data, sbinfo, false)) { | |
3731 | err = -EINVAL; | |
3732 | goto failed; | |
3733 | } | |
ca4e0519 AV |
3734 | } else { |
3735 | sb->s_flags |= MS_NOUSER; | |
1da177e4 | 3736 | } |
91828a40 | 3737 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 3738 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
3739 | #else |
3740 | sb->s_flags |= MS_NOUSER; | |
3741 | #endif | |
3742 | ||
0edd73b3 | 3743 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3744 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3745 | goto failed; |
680d794b | 3746 | sbinfo->free_inodes = sbinfo->max_inodes; |
779750d2 KS |
3747 | spin_lock_init(&sbinfo->shrinklist_lock); |
3748 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 3749 | |
285b2c4f | 3750 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3751 | sb->s_blocksize = PAGE_SIZE; |
3752 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3753 | sb->s_magic = TMPFS_MAGIC; |
3754 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3755 | sb->s_time_gran = 1; |
b09e0fa4 | 3756 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3757 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3758 | #endif |
3759 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
3760 | sb->s_flags |= MS_POSIXACL; |
3761 | #endif | |
0edd73b3 | 3762 | |
454abafe | 3763 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3764 | if (!inode) |
3765 | goto failed; | |
680d794b | 3766 | inode->i_uid = sbinfo->uid; |
3767 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3768 | sb->s_root = d_make_root(inode); |
3769 | if (!sb->s_root) | |
48fde701 | 3770 | goto failed; |
1da177e4 LT |
3771 | return 0; |
3772 | ||
1da177e4 LT |
3773 | failed: |
3774 | shmem_put_super(sb); | |
3775 | return err; | |
3776 | } | |
3777 | ||
fcc234f8 | 3778 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3779 | |
3780 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3781 | { | |
41ffe5d5 HD |
3782 | struct shmem_inode_info *info; |
3783 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3784 | if (!info) | |
1da177e4 | 3785 | return NULL; |
41ffe5d5 | 3786 | return &info->vfs_inode; |
1da177e4 LT |
3787 | } |
3788 | ||
41ffe5d5 | 3789 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3790 | { |
3791 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84e710da AV |
3792 | if (S_ISLNK(inode->i_mode)) |
3793 | kfree(inode->i_link); | |
fa0d7e3d NP |
3794 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3795 | } | |
3796 | ||
1da177e4 LT |
3797 | static void shmem_destroy_inode(struct inode *inode) |
3798 | { | |
09208d15 | 3799 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3800 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3801 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3802 | } |
3803 | ||
41ffe5d5 | 3804 | static void shmem_init_inode(void *foo) |
1da177e4 | 3805 | { |
41ffe5d5 HD |
3806 | struct shmem_inode_info *info = foo; |
3807 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3808 | } |
3809 | ||
41ffe5d5 | 3810 | static int shmem_init_inodecache(void) |
1da177e4 LT |
3811 | { |
3812 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3813 | sizeof(struct shmem_inode_info), | |
5d097056 | 3814 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3815 | return 0; |
3816 | } | |
3817 | ||
41ffe5d5 | 3818 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3819 | { |
1a1d92c1 | 3820 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3821 | } |
3822 | ||
f5e54d6e | 3823 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3824 | .writepage = shmem_writepage, |
76719325 | 3825 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3826 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3827 | .write_begin = shmem_write_begin, |
3828 | .write_end = shmem_write_end, | |
1da177e4 | 3829 | #endif |
1c93923c | 3830 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3831 | .migratepage = migrate_page, |
1c93923c | 3832 | #endif |
aa261f54 | 3833 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3834 | }; |
3835 | ||
15ad7cdc | 3836 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3837 | .mmap = shmem_mmap, |
c01d5b30 | 3838 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3839 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3840 | .llseek = shmem_file_llseek, |
2ba5bbed | 3841 | .read_iter = shmem_file_read_iter, |
8174202b | 3842 | .write_iter = generic_file_write_iter, |
1b061d92 | 3843 | .fsync = noop_fsync, |
82c156f8 | 3844 | .splice_read = generic_file_splice_read, |
f6cb85d0 | 3845 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3846 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3847 | #endif |
3848 | }; | |
3849 | ||
92e1d5be | 3850 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3851 | .getattr = shmem_getattr, |
94c1e62d | 3852 | .setattr = shmem_setattr, |
b09e0fa4 | 3853 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3854 | .listxattr = shmem_listxattr, |
feda821e | 3855 | .set_acl = simple_set_acl, |
b09e0fa4 | 3856 | #endif |
1da177e4 LT |
3857 | }; |
3858 | ||
92e1d5be | 3859 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3860 | #ifdef CONFIG_TMPFS |
3861 | .create = shmem_create, | |
3862 | .lookup = simple_lookup, | |
3863 | .link = shmem_link, | |
3864 | .unlink = shmem_unlink, | |
3865 | .symlink = shmem_symlink, | |
3866 | .mkdir = shmem_mkdir, | |
3867 | .rmdir = shmem_rmdir, | |
3868 | .mknod = shmem_mknod, | |
2773bf00 | 3869 | .rename = shmem_rename2, |
60545d0d | 3870 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3871 | #endif |
b09e0fa4 | 3872 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3873 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3874 | #endif |
39f0247d | 3875 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3876 | .setattr = shmem_setattr, |
feda821e | 3877 | .set_acl = simple_set_acl, |
39f0247d AG |
3878 | #endif |
3879 | }; | |
3880 | ||
92e1d5be | 3881 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3882 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3883 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3884 | #endif |
39f0247d | 3885 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3886 | .setattr = shmem_setattr, |
feda821e | 3887 | .set_acl = simple_set_acl, |
39f0247d | 3888 | #endif |
1da177e4 LT |
3889 | }; |
3890 | ||
759b9775 | 3891 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
3892 | .alloc_inode = shmem_alloc_inode, |
3893 | .destroy_inode = shmem_destroy_inode, | |
3894 | #ifdef CONFIG_TMPFS | |
3895 | .statfs = shmem_statfs, | |
3896 | .remount_fs = shmem_remount_fs, | |
680d794b | 3897 | .show_options = shmem_show_options, |
1da177e4 | 3898 | #endif |
1f895f75 | 3899 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3900 | .drop_inode = generic_delete_inode, |
3901 | .put_super = shmem_put_super, | |
779750d2 KS |
3902 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
3903 | .nr_cached_objects = shmem_unused_huge_count, | |
3904 | .free_cached_objects = shmem_unused_huge_scan, | |
3905 | #endif | |
1da177e4 LT |
3906 | }; |
3907 | ||
f0f37e2f | 3908 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3909 | .fault = shmem_fault, |
d7c17551 | 3910 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3911 | #ifdef CONFIG_NUMA |
3912 | .set_policy = shmem_set_policy, | |
3913 | .get_policy = shmem_get_policy, | |
3914 | #endif | |
3915 | }; | |
3916 | ||
3c26ff6e AV |
3917 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
3918 | int flags, const char *dev_name, void *data) | |
1da177e4 | 3919 | { |
3c26ff6e | 3920 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
3921 | } |
3922 | ||
41ffe5d5 | 3923 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3924 | .owner = THIS_MODULE, |
3925 | .name = "tmpfs", | |
3c26ff6e | 3926 | .mount = shmem_mount, |
1da177e4 | 3927 | .kill_sb = kill_litter_super, |
2b8576cb | 3928 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 3929 | }; |
1da177e4 | 3930 | |
41ffe5d5 | 3931 | int __init shmem_init(void) |
1da177e4 LT |
3932 | { |
3933 | int error; | |
3934 | ||
16203a7a RL |
3935 | /* If rootfs called this, don't re-init */ |
3936 | if (shmem_inode_cachep) | |
3937 | return 0; | |
3938 | ||
41ffe5d5 | 3939 | error = shmem_init_inodecache(); |
1da177e4 LT |
3940 | if (error) |
3941 | goto out3; | |
3942 | ||
41ffe5d5 | 3943 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 3944 | if (error) { |
1170532b | 3945 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
3946 | goto out2; |
3947 | } | |
95dc112a | 3948 | |
ca4e0519 | 3949 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
3950 | if (IS_ERR(shm_mnt)) { |
3951 | error = PTR_ERR(shm_mnt); | |
1170532b | 3952 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
3953 | goto out1; |
3954 | } | |
5a6e75f8 | 3955 | |
e496cf3d | 3956 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3957 | if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY) |
3958 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; | |
3959 | else | |
3960 | shmem_huge = 0; /* just in case it was patched */ | |
3961 | #endif | |
1da177e4 LT |
3962 | return 0; |
3963 | ||
3964 | out1: | |
41ffe5d5 | 3965 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 3966 | out2: |
41ffe5d5 | 3967 | shmem_destroy_inodecache(); |
1da177e4 LT |
3968 | out3: |
3969 | shm_mnt = ERR_PTR(error); | |
3970 | return error; | |
3971 | } | |
853ac43a | 3972 | |
e496cf3d | 3973 | #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS) |
5a6e75f8 KS |
3974 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
3975 | struct kobj_attribute *attr, char *buf) | |
3976 | { | |
3977 | int values[] = { | |
3978 | SHMEM_HUGE_ALWAYS, | |
3979 | SHMEM_HUGE_WITHIN_SIZE, | |
3980 | SHMEM_HUGE_ADVISE, | |
3981 | SHMEM_HUGE_NEVER, | |
3982 | SHMEM_HUGE_DENY, | |
3983 | SHMEM_HUGE_FORCE, | |
3984 | }; | |
3985 | int i, count; | |
3986 | ||
3987 | for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) { | |
3988 | const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s "; | |
3989 | ||
3990 | count += sprintf(buf + count, fmt, | |
3991 | shmem_format_huge(values[i])); | |
3992 | } | |
3993 | buf[count - 1] = '\n'; | |
3994 | return count; | |
3995 | } | |
3996 | ||
3997 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
3998 | struct kobj_attribute *attr, const char *buf, size_t count) | |
3999 | { | |
4000 | char tmp[16]; | |
4001 | int huge; | |
4002 | ||
4003 | if (count + 1 > sizeof(tmp)) | |
4004 | return -EINVAL; | |
4005 | memcpy(tmp, buf, count); | |
4006 | tmp[count] = '\0'; | |
4007 | if (count && tmp[count - 1] == '\n') | |
4008 | tmp[count - 1] = '\0'; | |
4009 | ||
4010 | huge = shmem_parse_huge(tmp); | |
4011 | if (huge == -EINVAL) | |
4012 | return -EINVAL; | |
4013 | if (!has_transparent_hugepage() && | |
4014 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
4015 | return -EINVAL; | |
4016 | ||
4017 | shmem_huge = huge; | |
4018 | if (shmem_huge < SHMEM_HUGE_DENY) | |
4019 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; | |
4020 | return count; | |
4021 | } | |
4022 | ||
4023 | struct kobj_attribute shmem_enabled_attr = | |
4024 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
3b33719c | 4025 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */ |
f3f0e1d2 | 4026 | |
3b33719c | 4027 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
f3f0e1d2 KS |
4028 | bool shmem_huge_enabled(struct vm_area_struct *vma) |
4029 | { | |
4030 | struct inode *inode = file_inode(vma->vm_file); | |
4031 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4032 | loff_t i_size; | |
4033 | pgoff_t off; | |
4034 | ||
4035 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
4036 | return true; | |
4037 | if (shmem_huge == SHMEM_HUGE_DENY) | |
4038 | return false; | |
4039 | switch (sbinfo->huge) { | |
4040 | case SHMEM_HUGE_NEVER: | |
4041 | return false; | |
4042 | case SHMEM_HUGE_ALWAYS: | |
4043 | return true; | |
4044 | case SHMEM_HUGE_WITHIN_SIZE: | |
4045 | off = round_up(vma->vm_pgoff, HPAGE_PMD_NR); | |
4046 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
4047 | if (i_size >= HPAGE_PMD_SIZE && | |
4048 | i_size >> PAGE_SHIFT >= off) | |
4049 | return true; | |
4050 | case SHMEM_HUGE_ADVISE: | |
4051 | /* TODO: implement fadvise() hints */ | |
4052 | return (vma->vm_flags & VM_HUGEPAGE); | |
4053 | default: | |
4054 | VM_BUG_ON(1); | |
4055 | return false; | |
4056 | } | |
4057 | } | |
3b33719c | 4058 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 4059 | |
853ac43a MM |
4060 | #else /* !CONFIG_SHMEM */ |
4061 | ||
4062 | /* | |
4063 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
4064 | * | |
4065 | * This is intended for small system where the benefits of the full | |
4066 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4067 | * their complexity. On systems without swap this code should be | |
4068 | * effectively equivalent, but much lighter weight. | |
4069 | */ | |
4070 | ||
41ffe5d5 | 4071 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4072 | .name = "tmpfs", |
3c26ff6e | 4073 | .mount = ramfs_mount, |
853ac43a | 4074 | .kill_sb = kill_litter_super, |
2b8576cb | 4075 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4076 | }; |
4077 | ||
41ffe5d5 | 4078 | int __init shmem_init(void) |
853ac43a | 4079 | { |
41ffe5d5 | 4080 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4081 | |
41ffe5d5 | 4082 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
4083 | BUG_ON(IS_ERR(shm_mnt)); |
4084 | ||
4085 | return 0; | |
4086 | } | |
4087 | ||
41ffe5d5 | 4088 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
4089 | { |
4090 | return 0; | |
4091 | } | |
4092 | ||
3f96b79a HD |
4093 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
4094 | { | |
4095 | return 0; | |
4096 | } | |
4097 | ||
24513264 HD |
4098 | void shmem_unlock_mapping(struct address_space *mapping) |
4099 | { | |
4100 | } | |
4101 | ||
c01d5b30 HD |
4102 | #ifdef CONFIG_MMU |
4103 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4104 | unsigned long addr, unsigned long len, | |
4105 | unsigned long pgoff, unsigned long flags) | |
4106 | { | |
4107 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4108 | } | |
4109 | #endif | |
4110 | ||
41ffe5d5 | 4111 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4112 | { |
41ffe5d5 | 4113 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4114 | } |
4115 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4116 | ||
0b0a0806 HD |
4117 | #define shmem_vm_ops generic_file_vm_ops |
4118 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 4119 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
4120 | #define shmem_acct_size(flags, size) 0 |
4121 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
4122 | |
4123 | #endif /* CONFIG_SHMEM */ | |
4124 | ||
4125 | /* common code */ | |
1da177e4 | 4126 | |
19938e35 | 4127 | static const struct dentry_operations anon_ops = { |
118b2302 | 4128 | .d_dname = simple_dname |
3451538a AV |
4129 | }; |
4130 | ||
c7277090 EP |
4131 | static struct file *__shmem_file_setup(const char *name, loff_t size, |
4132 | unsigned long flags, unsigned int i_flags) | |
1da177e4 | 4133 | { |
6b4d0b27 | 4134 | struct file *res; |
1da177e4 | 4135 | struct inode *inode; |
2c48b9c4 | 4136 | struct path path; |
3451538a | 4137 | struct super_block *sb; |
1da177e4 LT |
4138 | struct qstr this; |
4139 | ||
4140 | if (IS_ERR(shm_mnt)) | |
6b4d0b27 | 4141 | return ERR_CAST(shm_mnt); |
1da177e4 | 4142 | |
285b2c4f | 4143 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4144 | return ERR_PTR(-EINVAL); |
4145 | ||
4146 | if (shmem_acct_size(flags, size)) | |
4147 | return ERR_PTR(-ENOMEM); | |
4148 | ||
6b4d0b27 | 4149 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
4150 | this.name = name; |
4151 | this.len = strlen(name); | |
4152 | this.hash = 0; /* will go */ | |
3451538a | 4153 | sb = shm_mnt->mnt_sb; |
66ee4b88 | 4154 | path.mnt = mntget(shm_mnt); |
3451538a | 4155 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 4156 | if (!path.dentry) |
1da177e4 | 4157 | goto put_memory; |
3451538a | 4158 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 4159 | |
6b4d0b27 | 4160 | res = ERR_PTR(-ENOSPC); |
3451538a | 4161 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 4162 | if (!inode) |
66ee4b88 | 4163 | goto put_memory; |
1da177e4 | 4164 | |
c7277090 | 4165 | inode->i_flags |= i_flags; |
2c48b9c4 | 4166 | d_instantiate(path.dentry, inode); |
1da177e4 | 4167 | inode->i_size = size; |
6d6b77f1 | 4168 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
4169 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
4170 | if (IS_ERR(res)) | |
66ee4b88 | 4171 | goto put_path; |
4b42af81 | 4172 | |
6b4d0b27 | 4173 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 4174 | &shmem_file_operations); |
6b4d0b27 | 4175 | if (IS_ERR(res)) |
66ee4b88 | 4176 | goto put_path; |
4b42af81 | 4177 | |
6b4d0b27 | 4178 | return res; |
1da177e4 | 4179 | |
1da177e4 LT |
4180 | put_memory: |
4181 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
4182 | put_path: |
4183 | path_put(&path); | |
6b4d0b27 | 4184 | return res; |
1da177e4 | 4185 | } |
c7277090 EP |
4186 | |
4187 | /** | |
4188 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4189 | * kernel internal. There will be NO LSM permission checks against the | |
4190 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4191 | * higher layer. The users are the big_key and shm implementations. LSM |
4192 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4193 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4194 | * @size: size to be set for the file | |
4195 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4196 | */ | |
4197 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4198 | { | |
4199 | return __shmem_file_setup(name, size, flags, S_PRIVATE); | |
4200 | } | |
4201 | ||
4202 | /** | |
4203 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4204 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4205 | * @size: size to be set for the file | |
4206 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4207 | */ | |
4208 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4209 | { | |
4210 | return __shmem_file_setup(name, size, flags, 0); | |
4211 | } | |
395e0ddc | 4212 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4213 | |
46711810 | 4214 | /** |
1da177e4 | 4215 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
4216 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
4217 | */ | |
4218 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4219 | { | |
4220 | struct file *file; | |
4221 | loff_t size = vma->vm_end - vma->vm_start; | |
4222 | ||
66fc1303 HD |
4223 | /* |
4224 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
4225 | * between XFS directory reading and selinux: since this file is only | |
4226 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4227 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4228 | */ | |
4229 | file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE); | |
1da177e4 LT |
4230 | if (IS_ERR(file)) |
4231 | return PTR_ERR(file); | |
4232 | ||
4233 | if (vma->vm_file) | |
4234 | fput(vma->vm_file); | |
4235 | vma->vm_file = file; | |
4236 | vma->vm_ops = &shmem_vm_ops; | |
f3f0e1d2 | 4237 | |
e496cf3d | 4238 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
4239 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
4240 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
4241 | khugepaged_enter(vma, vma->vm_flags); | |
4242 | } | |
4243 | ||
1da177e4 LT |
4244 | return 0; |
4245 | } | |
d9d90e5e HD |
4246 | |
4247 | /** | |
4248 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
4249 | * @mapping: the page's address_space | |
4250 | * @index: the page index | |
4251 | * @gfp: the page allocator flags to use if allocating | |
4252 | * | |
4253 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4254 | * with any new page allocations done using the specified allocation flags. | |
4255 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
4256 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
4257 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4258 | * | |
68da9f05 HD |
4259 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4260 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4261 | */ |
4262 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4263 | pgoff_t index, gfp_t gfp) | |
4264 | { | |
68da9f05 HD |
4265 | #ifdef CONFIG_SHMEM |
4266 | struct inode *inode = mapping->host; | |
9276aad6 | 4267 | struct page *page; |
68da9f05 HD |
4268 | int error; |
4269 | ||
4270 | BUG_ON(mapping->a_ops != &shmem_aops); | |
9e18eb29 | 4271 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
cfda0526 | 4272 | gfp, NULL, NULL, NULL); |
68da9f05 HD |
4273 | if (error) |
4274 | page = ERR_PTR(error); | |
4275 | else | |
4276 | unlock_page(page); | |
4277 | return page; | |
4278 | #else | |
4279 | /* | |
4280 | * The tiny !SHMEM case uses ramfs without swap | |
4281 | */ | |
d9d90e5e | 4282 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4283 | #endif |
d9d90e5e HD |
4284 | } |
4285 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |