]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Resizable virtual memory filesystem for Linux. | |
3 | * | |
4 | * Copyright (C) 2000 Linus Torvalds. | |
5 | * 2000 Transmeta Corp. | |
6 | * 2000-2001 Christoph Rohland | |
7 | * 2000-2001 SAP AG | |
8 | * 2002 Red Hat Inc. | |
6922c0c7 HD |
9 | * Copyright (C) 2002-2011 Hugh Dickins. |
10 | * Copyright (C) 2011 Google Inc. | |
0edd73b3 | 11 | * Copyright (C) 2002-2005 VERITAS Software Corporation. |
1da177e4 LT |
12 | * Copyright (C) 2004 Andi Kleen, SuSE Labs |
13 | * | |
14 | * Extended attribute support for tmpfs: | |
15 | * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net> | |
16 | * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com> | |
17 | * | |
853ac43a MM |
18 | * tiny-shmem: |
19 | * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com> | |
20 | * | |
1da177e4 LT |
21 | * This file is released under the GPL. |
22 | */ | |
23 | ||
853ac43a MM |
24 | #include <linux/fs.h> |
25 | #include <linux/init.h> | |
26 | #include <linux/vfs.h> | |
27 | #include <linux/mount.h> | |
250297ed | 28 | #include <linux/ramfs.h> |
caefba17 | 29 | #include <linux/pagemap.h> |
853ac43a MM |
30 | #include <linux/file.h> |
31 | #include <linux/mm.h> | |
174cd4b1 | 32 | #include <linux/sched/signal.h> |
b95f1b31 | 33 | #include <linux/export.h> |
853ac43a | 34 | #include <linux/swap.h> |
e2e40f2c | 35 | #include <linux/uio.h> |
f3f0e1d2 | 36 | #include <linux/khugepaged.h> |
853ac43a | 37 | |
95cc09d6 AA |
38 | #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */ |
39 | ||
853ac43a MM |
40 | static struct vfsmount *shm_mnt; |
41 | ||
42 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
43 | /* |
44 | * This virtual memory filesystem is heavily based on the ramfs. It | |
45 | * extends ramfs by the ability to use swap and honor resource limits | |
46 | * which makes it a completely usable filesystem. | |
47 | */ | |
48 | ||
39f0247d | 49 | #include <linux/xattr.h> |
a5694255 | 50 | #include <linux/exportfs.h> |
1c7c474c | 51 | #include <linux/posix_acl.h> |
feda821e | 52 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 53 | #include <linux/mman.h> |
1da177e4 LT |
54 | #include <linux/string.h> |
55 | #include <linux/slab.h> | |
56 | #include <linux/backing-dev.h> | |
57 | #include <linux/shmem_fs.h> | |
1da177e4 | 58 | #include <linux/writeback.h> |
1da177e4 | 59 | #include <linux/blkdev.h> |
bda97eab | 60 | #include <linux/pagevec.h> |
41ffe5d5 | 61 | #include <linux/percpu_counter.h> |
83e4fa9c | 62 | #include <linux/falloc.h> |
708e3508 | 63 | #include <linux/splice.h> |
1da177e4 LT |
64 | #include <linux/security.h> |
65 | #include <linux/swapops.h> | |
66 | #include <linux/mempolicy.h> | |
67 | #include <linux/namei.h> | |
b00dc3ad | 68 | #include <linux/ctype.h> |
304dbdb7 | 69 | #include <linux/migrate.h> |
c1f60a5a | 70 | #include <linux/highmem.h> |
680d794b | 71 | #include <linux/seq_file.h> |
92562927 | 72 | #include <linux/magic.h> |
9183df25 | 73 | #include <linux/syscalls.h> |
40e041a2 | 74 | #include <linux/fcntl.h> |
9183df25 | 75 | #include <uapi/linux/memfd.h> |
cfda0526 | 76 | #include <linux/userfaultfd_k.h> |
4c27fe4c | 77 | #include <linux/rmap.h> |
2b4db796 | 78 | #include <linux/uuid.h> |
304dbdb7 | 79 | |
7c0f6ba6 | 80 | #include <linux/uaccess.h> |
1da177e4 LT |
81 | #include <asm/pgtable.h> |
82 | ||
dd56b046 MG |
83 | #include "internal.h" |
84 | ||
09cbfeaf KS |
85 | #define BLOCKS_PER_PAGE (PAGE_SIZE/512) |
86 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT) | |
1da177e4 | 87 | |
1da177e4 LT |
88 | /* Pretend that each entry is of this size in directory's i_size */ |
89 | #define BOGO_DIRENT_SIZE 20 | |
90 | ||
69f07ec9 HD |
91 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
92 | #define SHORT_SYMLINK_LEN 128 | |
93 | ||
1aac1400 | 94 | /* |
f00cdc6d HD |
95 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
96 | * inode->i_private (with i_mutex making sure that it has only one user at | |
97 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
98 | */ |
99 | struct shmem_falloc { | |
8e205f77 | 100 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
101 | pgoff_t start; /* start of range currently being fallocated */ |
102 | pgoff_t next; /* the next page offset to be fallocated */ | |
103 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
104 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
105 | }; | |
106 | ||
b76db735 | 107 | #ifdef CONFIG_TMPFS |
680d794b AM |
108 | static unsigned long shmem_default_max_blocks(void) |
109 | { | |
110 | return totalram_pages / 2; | |
111 | } | |
112 | ||
113 | static unsigned long shmem_default_max_inodes(void) | |
114 | { | |
115 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
116 | } | |
b76db735 | 117 | #endif |
680d794b | 118 | |
bde05d1c HD |
119 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
120 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
121 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 | 122 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 123 | struct page **pagep, enum sgp_type sgp, |
cfda0526 MR |
124 | gfp_t gfp, struct vm_area_struct *vma, |
125 | struct vm_fault *vmf, int *fault_type); | |
68da9f05 | 126 | |
f3f0e1d2 | 127 | int shmem_getpage(struct inode *inode, pgoff_t index, |
9e18eb29 | 128 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
129 | { |
130 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
cfda0526 | 131 | mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL); |
68da9f05 | 132 | } |
1da177e4 | 133 | |
1da177e4 LT |
134 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
135 | { | |
136 | return sb->s_fs_info; | |
137 | } | |
138 | ||
139 | /* | |
140 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
141 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
142 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
143 | * consistent with the pre-accounting of private mappings ... | |
144 | */ | |
145 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
146 | { | |
0b0a0806 | 147 | return (flags & VM_NORESERVE) ? |
191c5424 | 148 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
149 | } |
150 | ||
151 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
152 | { | |
0b0a0806 | 153 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
154 | vm_unacct_memory(VM_ACCT(size)); |
155 | } | |
156 | ||
77142517 KK |
157 | static inline int shmem_reacct_size(unsigned long flags, |
158 | loff_t oldsize, loff_t newsize) | |
159 | { | |
160 | if (!(flags & VM_NORESERVE)) { | |
161 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
162 | return security_vm_enough_memory_mm(current->mm, | |
163 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
164 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
165 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
166 | } | |
167 | return 0; | |
168 | } | |
169 | ||
1da177e4 LT |
170 | /* |
171 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 172 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
173 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
174 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
175 | */ | |
800d8c63 | 176 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 177 | { |
800d8c63 KS |
178 | if (!(flags & VM_NORESERVE)) |
179 | return 0; | |
180 | ||
181 | return security_vm_enough_memory_mm(current->mm, | |
182 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
183 | } |
184 | ||
185 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
186 | { | |
0b0a0806 | 187 | if (flags & VM_NORESERVE) |
09cbfeaf | 188 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
189 | } |
190 | ||
759b9775 | 191 | static const struct super_operations shmem_ops; |
f5e54d6e | 192 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 193 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
194 | static const struct inode_operations shmem_inode_operations; |
195 | static const struct inode_operations shmem_dir_inode_operations; | |
196 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 197 | static const struct vm_operations_struct shmem_vm_ops; |
779750d2 | 198 | static struct file_system_type shmem_fs_type; |
1da177e4 | 199 | |
b0506e48 MR |
200 | bool vma_is_shmem(struct vm_area_struct *vma) |
201 | { | |
202 | return vma->vm_ops == &shmem_vm_ops; | |
203 | } | |
204 | ||
1da177e4 | 205 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 206 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 207 | |
5b04c689 PE |
208 | static int shmem_reserve_inode(struct super_block *sb) |
209 | { | |
210 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
211 | if (sbinfo->max_inodes) { | |
212 | spin_lock(&sbinfo->stat_lock); | |
213 | if (!sbinfo->free_inodes) { | |
214 | spin_unlock(&sbinfo->stat_lock); | |
215 | return -ENOSPC; | |
216 | } | |
217 | sbinfo->free_inodes--; | |
218 | spin_unlock(&sbinfo->stat_lock); | |
219 | } | |
220 | return 0; | |
221 | } | |
222 | ||
223 | static void shmem_free_inode(struct super_block *sb) | |
224 | { | |
225 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
226 | if (sbinfo->max_inodes) { | |
227 | spin_lock(&sbinfo->stat_lock); | |
228 | sbinfo->free_inodes++; | |
229 | spin_unlock(&sbinfo->stat_lock); | |
230 | } | |
231 | } | |
232 | ||
46711810 | 233 | /** |
41ffe5d5 | 234 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
235 | * @inode: inode to recalc |
236 | * | |
237 | * We have to calculate the free blocks since the mm can drop | |
238 | * undirtied hole pages behind our back. | |
239 | * | |
240 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
241 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
242 | * | |
243 | * It has to be called with the spinlock held. | |
244 | */ | |
245 | static void shmem_recalc_inode(struct inode *inode) | |
246 | { | |
247 | struct shmem_inode_info *info = SHMEM_I(inode); | |
248 | long freed; | |
249 | ||
250 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
251 | if (freed > 0) { | |
54af6042 HD |
252 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
253 | if (sbinfo->max_blocks) | |
254 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 255 | info->alloced -= freed; |
54af6042 | 256 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 257 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
258 | } |
259 | } | |
260 | ||
800d8c63 KS |
261 | bool shmem_charge(struct inode *inode, long pages) |
262 | { | |
263 | struct shmem_inode_info *info = SHMEM_I(inode); | |
264 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4595ef88 | 265 | unsigned long flags; |
800d8c63 KS |
266 | |
267 | if (shmem_acct_block(info->flags, pages)) | |
268 | return false; | |
4595ef88 | 269 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
270 | info->alloced += pages; |
271 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
272 | shmem_recalc_inode(inode); | |
4595ef88 | 273 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
274 | inode->i_mapping->nrpages += pages; |
275 | ||
276 | if (!sbinfo->max_blocks) | |
277 | return true; | |
278 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
279 | sbinfo->max_blocks - pages) > 0) { | |
280 | inode->i_mapping->nrpages -= pages; | |
4595ef88 | 281 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
282 | info->alloced -= pages; |
283 | shmem_recalc_inode(inode); | |
4595ef88 | 284 | spin_unlock_irqrestore(&info->lock, flags); |
71664665 | 285 | shmem_unacct_blocks(info->flags, pages); |
800d8c63 KS |
286 | return false; |
287 | } | |
288 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
289 | return true; | |
290 | } | |
291 | ||
292 | void shmem_uncharge(struct inode *inode, long pages) | |
293 | { | |
294 | struct shmem_inode_info *info = SHMEM_I(inode); | |
295 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4595ef88 | 296 | unsigned long flags; |
800d8c63 | 297 | |
4595ef88 | 298 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
299 | info->alloced -= pages; |
300 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
301 | shmem_recalc_inode(inode); | |
4595ef88 | 302 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
303 | |
304 | if (sbinfo->max_blocks) | |
305 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
71664665 | 306 | shmem_unacct_blocks(info->flags, pages); |
800d8c63 KS |
307 | } |
308 | ||
7a5d0fbb HD |
309 | /* |
310 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
311 | */ | |
312 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
313 | pgoff_t index, void *expected, void *replacement) | |
314 | { | |
f7942430 | 315 | struct radix_tree_node *node; |
7a5d0fbb | 316 | void **pslot; |
6dbaf22c | 317 | void *item; |
7a5d0fbb HD |
318 | |
319 | VM_BUG_ON(!expected); | |
6dbaf22c | 320 | VM_BUG_ON(!replacement); |
f7942430 JW |
321 | item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot); |
322 | if (!item) | |
6dbaf22c | 323 | return -ENOENT; |
7a5d0fbb HD |
324 | if (item != expected) |
325 | return -ENOENT; | |
4d693d08 JW |
326 | __radix_tree_replace(&mapping->page_tree, node, pslot, |
327 | replacement, NULL, NULL); | |
7a5d0fbb HD |
328 | return 0; |
329 | } | |
330 | ||
d1899228 HD |
331 | /* |
332 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
333 | * that an entry was not already brought back from swap by a racing thread. | |
334 | * | |
335 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
336 | * might be reused, and again be SwapCache, using the same swap as before. | |
337 | */ | |
338 | static bool shmem_confirm_swap(struct address_space *mapping, | |
339 | pgoff_t index, swp_entry_t swap) | |
340 | { | |
341 | void *item; | |
342 | ||
343 | rcu_read_lock(); | |
344 | item = radix_tree_lookup(&mapping->page_tree, index); | |
345 | rcu_read_unlock(); | |
346 | return item == swp_to_radix_entry(swap); | |
347 | } | |
348 | ||
5a6e75f8 KS |
349 | /* |
350 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
351 | * | |
352 | * SHMEM_HUGE_NEVER: | |
353 | * disables huge pages for the mount; | |
354 | * SHMEM_HUGE_ALWAYS: | |
355 | * enables huge pages for the mount; | |
356 | * SHMEM_HUGE_WITHIN_SIZE: | |
357 | * only allocate huge pages if the page will be fully within i_size, | |
358 | * also respect fadvise()/madvise() hints; | |
359 | * SHMEM_HUGE_ADVISE: | |
360 | * only allocate huge pages if requested with fadvise()/madvise(); | |
361 | */ | |
362 | ||
363 | #define SHMEM_HUGE_NEVER 0 | |
364 | #define SHMEM_HUGE_ALWAYS 1 | |
365 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
366 | #define SHMEM_HUGE_ADVISE 3 | |
367 | ||
368 | /* | |
369 | * Special values. | |
370 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
371 | * | |
372 | * SHMEM_HUGE_DENY: | |
373 | * disables huge on shm_mnt and all mounts, for emergency use; | |
374 | * SHMEM_HUGE_FORCE: | |
375 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
376 | * | |
377 | */ | |
378 | #define SHMEM_HUGE_DENY (-1) | |
379 | #define SHMEM_HUGE_FORCE (-2) | |
380 | ||
e496cf3d | 381 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
382 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
383 | ||
384 | int shmem_huge __read_mostly; | |
385 | ||
f1f5929c | 386 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
387 | static int shmem_parse_huge(const char *str) |
388 | { | |
389 | if (!strcmp(str, "never")) | |
390 | return SHMEM_HUGE_NEVER; | |
391 | if (!strcmp(str, "always")) | |
392 | return SHMEM_HUGE_ALWAYS; | |
393 | if (!strcmp(str, "within_size")) | |
394 | return SHMEM_HUGE_WITHIN_SIZE; | |
395 | if (!strcmp(str, "advise")) | |
396 | return SHMEM_HUGE_ADVISE; | |
397 | if (!strcmp(str, "deny")) | |
398 | return SHMEM_HUGE_DENY; | |
399 | if (!strcmp(str, "force")) | |
400 | return SHMEM_HUGE_FORCE; | |
401 | return -EINVAL; | |
402 | } | |
403 | ||
404 | static const char *shmem_format_huge(int huge) | |
405 | { | |
406 | switch (huge) { | |
407 | case SHMEM_HUGE_NEVER: | |
408 | return "never"; | |
409 | case SHMEM_HUGE_ALWAYS: | |
410 | return "always"; | |
411 | case SHMEM_HUGE_WITHIN_SIZE: | |
412 | return "within_size"; | |
413 | case SHMEM_HUGE_ADVISE: | |
414 | return "advise"; | |
415 | case SHMEM_HUGE_DENY: | |
416 | return "deny"; | |
417 | case SHMEM_HUGE_FORCE: | |
418 | return "force"; | |
419 | default: | |
420 | VM_BUG_ON(1); | |
421 | return "bad_val"; | |
422 | } | |
423 | } | |
f1f5929c | 424 | #endif |
5a6e75f8 | 425 | |
779750d2 KS |
426 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
427 | struct shrink_control *sc, unsigned long nr_to_split) | |
428 | { | |
429 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 430 | LIST_HEAD(to_remove); |
779750d2 KS |
431 | struct inode *inode; |
432 | struct shmem_inode_info *info; | |
433 | struct page *page; | |
434 | unsigned long batch = sc ? sc->nr_to_scan : 128; | |
435 | int removed = 0, split = 0; | |
436 | ||
437 | if (list_empty(&sbinfo->shrinklist)) | |
438 | return SHRINK_STOP; | |
439 | ||
440 | spin_lock(&sbinfo->shrinklist_lock); | |
441 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
442 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
443 | ||
444 | /* pin the inode */ | |
445 | inode = igrab(&info->vfs_inode); | |
446 | ||
447 | /* inode is about to be evicted */ | |
448 | if (!inode) { | |
449 | list_del_init(&info->shrinklist); | |
450 | removed++; | |
451 | goto next; | |
452 | } | |
453 | ||
454 | /* Check if there's anything to gain */ | |
455 | if (round_up(inode->i_size, PAGE_SIZE) == | |
456 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 457 | list_move(&info->shrinklist, &to_remove); |
779750d2 | 458 | removed++; |
779750d2 KS |
459 | goto next; |
460 | } | |
461 | ||
462 | list_move(&info->shrinklist, &list); | |
463 | next: | |
464 | if (!--batch) | |
465 | break; | |
466 | } | |
467 | spin_unlock(&sbinfo->shrinklist_lock); | |
468 | ||
253fd0f0 KS |
469 | list_for_each_safe(pos, next, &to_remove) { |
470 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
471 | inode = &info->vfs_inode; | |
472 | list_del_init(&info->shrinklist); | |
473 | iput(inode); | |
474 | } | |
475 | ||
779750d2 KS |
476 | list_for_each_safe(pos, next, &list) { |
477 | int ret; | |
478 | ||
479 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
480 | inode = &info->vfs_inode; | |
481 | ||
482 | if (nr_to_split && split >= nr_to_split) { | |
483 | iput(inode); | |
484 | continue; | |
485 | } | |
486 | ||
487 | page = find_lock_page(inode->i_mapping, | |
488 | (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT); | |
489 | if (!page) | |
490 | goto drop; | |
491 | ||
492 | if (!PageTransHuge(page)) { | |
493 | unlock_page(page); | |
494 | put_page(page); | |
495 | goto drop; | |
496 | } | |
497 | ||
498 | ret = split_huge_page(page); | |
499 | unlock_page(page); | |
500 | put_page(page); | |
501 | ||
502 | if (ret) { | |
503 | /* split failed: leave it on the list */ | |
504 | iput(inode); | |
505 | continue; | |
506 | } | |
507 | ||
508 | split++; | |
509 | drop: | |
510 | list_del_init(&info->shrinklist); | |
511 | removed++; | |
512 | iput(inode); | |
513 | } | |
514 | ||
515 | spin_lock(&sbinfo->shrinklist_lock); | |
516 | list_splice_tail(&list, &sbinfo->shrinklist); | |
517 | sbinfo->shrinklist_len -= removed; | |
518 | spin_unlock(&sbinfo->shrinklist_lock); | |
519 | ||
520 | return split; | |
521 | } | |
522 | ||
523 | static long shmem_unused_huge_scan(struct super_block *sb, | |
524 | struct shrink_control *sc) | |
525 | { | |
526 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
527 | ||
528 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
529 | return SHRINK_STOP; | |
530 | ||
531 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
532 | } | |
533 | ||
534 | static long shmem_unused_huge_count(struct super_block *sb, | |
535 | struct shrink_control *sc) | |
536 | { | |
537 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
538 | return READ_ONCE(sbinfo->shrinklist_len); | |
539 | } | |
e496cf3d | 540 | #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 KS |
541 | |
542 | #define shmem_huge SHMEM_HUGE_DENY | |
543 | ||
779750d2 KS |
544 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
545 | struct shrink_control *sc, unsigned long nr_to_split) | |
546 | { | |
547 | return 0; | |
548 | } | |
e496cf3d | 549 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 550 | |
46f65ec1 HD |
551 | /* |
552 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
553 | */ | |
554 | static int shmem_add_to_page_cache(struct page *page, | |
555 | struct address_space *mapping, | |
fed400a1 | 556 | pgoff_t index, void *expected) |
46f65ec1 | 557 | { |
800d8c63 | 558 | int error, nr = hpage_nr_pages(page); |
46f65ec1 | 559 | |
800d8c63 KS |
560 | VM_BUG_ON_PAGE(PageTail(page), page); |
561 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
562 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
563 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 564 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 565 | |
800d8c63 | 566 | page_ref_add(page, nr); |
b065b432 HD |
567 | page->mapping = mapping; |
568 | page->index = index; | |
569 | ||
570 | spin_lock_irq(&mapping->tree_lock); | |
800d8c63 KS |
571 | if (PageTransHuge(page)) { |
572 | void __rcu **results; | |
573 | pgoff_t idx; | |
574 | int i; | |
575 | ||
576 | error = 0; | |
577 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, | |
578 | &results, &idx, index, 1) && | |
579 | idx < index + HPAGE_PMD_NR) { | |
580 | error = -EEXIST; | |
581 | } | |
582 | ||
583 | if (!error) { | |
584 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
585 | error = radix_tree_insert(&mapping->page_tree, | |
586 | index + i, page + i); | |
587 | VM_BUG_ON(error); | |
588 | } | |
589 | count_vm_event(THP_FILE_ALLOC); | |
590 | } | |
591 | } else if (!expected) { | |
b065b432 | 592 | error = radix_tree_insert(&mapping->page_tree, index, page); |
800d8c63 | 593 | } else { |
b065b432 HD |
594 | error = shmem_radix_tree_replace(mapping, index, expected, |
595 | page); | |
800d8c63 KS |
596 | } |
597 | ||
46f65ec1 | 598 | if (!error) { |
800d8c63 KS |
599 | mapping->nrpages += nr; |
600 | if (PageTransHuge(page)) | |
11fb9989 MG |
601 | __inc_node_page_state(page, NR_SHMEM_THPS); |
602 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); | |
603 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr); | |
b065b432 HD |
604 | spin_unlock_irq(&mapping->tree_lock); |
605 | } else { | |
606 | page->mapping = NULL; | |
607 | spin_unlock_irq(&mapping->tree_lock); | |
800d8c63 | 608 | page_ref_sub(page, nr); |
46f65ec1 | 609 | } |
46f65ec1 HD |
610 | return error; |
611 | } | |
612 | ||
6922c0c7 HD |
613 | /* |
614 | * Like delete_from_page_cache, but substitutes swap for page. | |
615 | */ | |
616 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
617 | { | |
618 | struct address_space *mapping = page->mapping; | |
619 | int error; | |
620 | ||
800d8c63 KS |
621 | VM_BUG_ON_PAGE(PageCompound(page), page); |
622 | ||
6922c0c7 HD |
623 | spin_lock_irq(&mapping->tree_lock); |
624 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
625 | page->mapping = NULL; | |
626 | mapping->nrpages--; | |
11fb9989 MG |
627 | __dec_node_page_state(page, NR_FILE_PAGES); |
628 | __dec_node_page_state(page, NR_SHMEM); | |
6922c0c7 | 629 | spin_unlock_irq(&mapping->tree_lock); |
09cbfeaf | 630 | put_page(page); |
6922c0c7 HD |
631 | BUG_ON(error); |
632 | } | |
633 | ||
7a5d0fbb HD |
634 | /* |
635 | * Remove swap entry from radix tree, free the swap and its page cache. | |
636 | */ | |
637 | static int shmem_free_swap(struct address_space *mapping, | |
638 | pgoff_t index, void *radswap) | |
639 | { | |
6dbaf22c | 640 | void *old; |
7a5d0fbb HD |
641 | |
642 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 643 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 644 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
645 | if (old != radswap) |
646 | return -ENOENT; | |
647 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
648 | return 0; | |
7a5d0fbb HD |
649 | } |
650 | ||
6a15a370 VB |
651 | /* |
652 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 653 | * given offsets are swapped out. |
6a15a370 VB |
654 | * |
655 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
656 | * as long as the inode doesn't go away and racy results are not a problem. | |
657 | */ | |
48131e03 VB |
658 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
659 | pgoff_t start, pgoff_t end) | |
6a15a370 | 660 | { |
6a15a370 VB |
661 | struct radix_tree_iter iter; |
662 | void **slot; | |
663 | struct page *page; | |
48131e03 | 664 | unsigned long swapped = 0; |
6a15a370 VB |
665 | |
666 | rcu_read_lock(); | |
667 | ||
6a15a370 VB |
668 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
669 | if (iter.index >= end) | |
670 | break; | |
671 | ||
672 | page = radix_tree_deref_slot(slot); | |
673 | ||
2cf938aa MW |
674 | if (radix_tree_deref_retry(page)) { |
675 | slot = radix_tree_iter_retry(&iter); | |
676 | continue; | |
677 | } | |
6a15a370 VB |
678 | |
679 | if (radix_tree_exceptional_entry(page)) | |
680 | swapped++; | |
681 | ||
682 | if (need_resched()) { | |
148deab2 | 683 | slot = radix_tree_iter_resume(slot, &iter); |
6a15a370 | 684 | cond_resched_rcu(); |
6a15a370 VB |
685 | } |
686 | } | |
687 | ||
688 | rcu_read_unlock(); | |
689 | ||
690 | return swapped << PAGE_SHIFT; | |
691 | } | |
692 | ||
48131e03 VB |
693 | /* |
694 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
695 | * given vma is swapped out. | |
696 | * | |
697 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
698 | * as long as the inode doesn't go away and racy results are not a problem. | |
699 | */ | |
700 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
701 | { | |
702 | struct inode *inode = file_inode(vma->vm_file); | |
703 | struct shmem_inode_info *info = SHMEM_I(inode); | |
704 | struct address_space *mapping = inode->i_mapping; | |
705 | unsigned long swapped; | |
706 | ||
707 | /* Be careful as we don't hold info->lock */ | |
708 | swapped = READ_ONCE(info->swapped); | |
709 | ||
710 | /* | |
711 | * The easier cases are when the shmem object has nothing in swap, or | |
712 | * the vma maps it whole. Then we can simply use the stats that we | |
713 | * already track. | |
714 | */ | |
715 | if (!swapped) | |
716 | return 0; | |
717 | ||
718 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
719 | return swapped << PAGE_SHIFT; | |
720 | ||
721 | /* Here comes the more involved part */ | |
722 | return shmem_partial_swap_usage(mapping, | |
723 | linear_page_index(vma, vma->vm_start), | |
724 | linear_page_index(vma, vma->vm_end)); | |
725 | } | |
726 | ||
24513264 HD |
727 | /* |
728 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
729 | */ | |
730 | void shmem_unlock_mapping(struct address_space *mapping) | |
731 | { | |
732 | struct pagevec pvec; | |
733 | pgoff_t indices[PAGEVEC_SIZE]; | |
734 | pgoff_t index = 0; | |
735 | ||
736 | pagevec_init(&pvec, 0); | |
737 | /* | |
738 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
739 | */ | |
740 | while (!mapping_unevictable(mapping)) { | |
741 | /* | |
742 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
743 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
744 | */ | |
0cd6144a JW |
745 | pvec.nr = find_get_entries(mapping, index, |
746 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
747 | if (!pvec.nr) |
748 | break; | |
749 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 750 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
751 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
752 | pagevec_release(&pvec); | |
753 | cond_resched(); | |
754 | } | |
7a5d0fbb HD |
755 | } |
756 | ||
757 | /* | |
758 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 759 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 760 | */ |
1635f6a7 HD |
761 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
762 | bool unfalloc) | |
1da177e4 | 763 | { |
285b2c4f | 764 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 765 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
766 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
767 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
768 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
769 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 770 | struct pagevec pvec; |
7a5d0fbb HD |
771 | pgoff_t indices[PAGEVEC_SIZE]; |
772 | long nr_swaps_freed = 0; | |
285b2c4f | 773 | pgoff_t index; |
bda97eab HD |
774 | int i; |
775 | ||
83e4fa9c HD |
776 | if (lend == -1) |
777 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
778 | |
779 | pagevec_init(&pvec, 0); | |
780 | index = start; | |
83e4fa9c | 781 | while (index < end) { |
0cd6144a JW |
782 | pvec.nr = find_get_entries(mapping, index, |
783 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
784 | pvec.pages, indices); | |
7a5d0fbb HD |
785 | if (!pvec.nr) |
786 | break; | |
bda97eab HD |
787 | for (i = 0; i < pagevec_count(&pvec); i++) { |
788 | struct page *page = pvec.pages[i]; | |
789 | ||
7a5d0fbb | 790 | index = indices[i]; |
83e4fa9c | 791 | if (index >= end) |
bda97eab HD |
792 | break; |
793 | ||
7a5d0fbb | 794 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
795 | if (unfalloc) |
796 | continue; | |
7a5d0fbb HD |
797 | nr_swaps_freed += !shmem_free_swap(mapping, |
798 | index, page); | |
bda97eab | 799 | continue; |
7a5d0fbb HD |
800 | } |
801 | ||
800d8c63 KS |
802 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
803 | ||
7a5d0fbb | 804 | if (!trylock_page(page)) |
bda97eab | 805 | continue; |
800d8c63 KS |
806 | |
807 | if (PageTransTail(page)) { | |
808 | /* Middle of THP: zero out the page */ | |
809 | clear_highpage(page); | |
810 | unlock_page(page); | |
811 | continue; | |
812 | } else if (PageTransHuge(page)) { | |
813 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
814 | /* | |
815 | * Range ends in the middle of THP: | |
816 | * zero out the page | |
817 | */ | |
818 | clear_highpage(page); | |
819 | unlock_page(page); | |
820 | continue; | |
821 | } | |
822 | index += HPAGE_PMD_NR - 1; | |
823 | i += HPAGE_PMD_NR - 1; | |
824 | } | |
825 | ||
1635f6a7 | 826 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
827 | VM_BUG_ON_PAGE(PageTail(page), page); |
828 | if (page_mapping(page) == mapping) { | |
309381fe | 829 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
830 | truncate_inode_page(mapping, page); |
831 | } | |
bda97eab | 832 | } |
bda97eab HD |
833 | unlock_page(page); |
834 | } | |
0cd6144a | 835 | pagevec_remove_exceptionals(&pvec); |
24513264 | 836 | pagevec_release(&pvec); |
bda97eab HD |
837 | cond_resched(); |
838 | index++; | |
839 | } | |
1da177e4 | 840 | |
83e4fa9c | 841 | if (partial_start) { |
bda97eab | 842 | struct page *page = NULL; |
9e18eb29 | 843 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 844 | if (page) { |
09cbfeaf | 845 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
846 | if (start > end) { |
847 | top = partial_end; | |
848 | partial_end = 0; | |
849 | } | |
850 | zero_user_segment(page, partial_start, top); | |
851 | set_page_dirty(page); | |
852 | unlock_page(page); | |
09cbfeaf | 853 | put_page(page); |
83e4fa9c HD |
854 | } |
855 | } | |
856 | if (partial_end) { | |
857 | struct page *page = NULL; | |
9e18eb29 | 858 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
859 | if (page) { |
860 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
861 | set_page_dirty(page); |
862 | unlock_page(page); | |
09cbfeaf | 863 | put_page(page); |
bda97eab HD |
864 | } |
865 | } | |
83e4fa9c HD |
866 | if (start >= end) |
867 | return; | |
bda97eab HD |
868 | |
869 | index = start; | |
b1a36650 | 870 | while (index < end) { |
bda97eab | 871 | cond_resched(); |
0cd6144a JW |
872 | |
873 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 874 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 875 | pvec.pages, indices); |
7a5d0fbb | 876 | if (!pvec.nr) { |
b1a36650 HD |
877 | /* If all gone or hole-punch or unfalloc, we're done */ |
878 | if (index == start || end != -1) | |
bda97eab | 879 | break; |
b1a36650 | 880 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
881 | index = start; |
882 | continue; | |
883 | } | |
bda97eab HD |
884 | for (i = 0; i < pagevec_count(&pvec); i++) { |
885 | struct page *page = pvec.pages[i]; | |
886 | ||
7a5d0fbb | 887 | index = indices[i]; |
83e4fa9c | 888 | if (index >= end) |
bda97eab HD |
889 | break; |
890 | ||
7a5d0fbb | 891 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
892 | if (unfalloc) |
893 | continue; | |
b1a36650 HD |
894 | if (shmem_free_swap(mapping, index, page)) { |
895 | /* Swap was replaced by page: retry */ | |
896 | index--; | |
897 | break; | |
898 | } | |
899 | nr_swaps_freed++; | |
7a5d0fbb HD |
900 | continue; |
901 | } | |
902 | ||
bda97eab | 903 | lock_page(page); |
800d8c63 KS |
904 | |
905 | if (PageTransTail(page)) { | |
906 | /* Middle of THP: zero out the page */ | |
907 | clear_highpage(page); | |
908 | unlock_page(page); | |
909 | /* | |
910 | * Partial thp truncate due 'start' in middle | |
911 | * of THP: don't need to look on these pages | |
912 | * again on !pvec.nr restart. | |
913 | */ | |
914 | if (index != round_down(end, HPAGE_PMD_NR)) | |
915 | start++; | |
916 | continue; | |
917 | } else if (PageTransHuge(page)) { | |
918 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
919 | /* | |
920 | * Range ends in the middle of THP: | |
921 | * zero out the page | |
922 | */ | |
923 | clear_highpage(page); | |
924 | unlock_page(page); | |
925 | continue; | |
926 | } | |
927 | index += HPAGE_PMD_NR - 1; | |
928 | i += HPAGE_PMD_NR - 1; | |
929 | } | |
930 | ||
1635f6a7 | 931 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
932 | VM_BUG_ON_PAGE(PageTail(page), page); |
933 | if (page_mapping(page) == mapping) { | |
309381fe | 934 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 935 | truncate_inode_page(mapping, page); |
b1a36650 HD |
936 | } else { |
937 | /* Page was replaced by swap: retry */ | |
938 | unlock_page(page); | |
939 | index--; | |
940 | break; | |
1635f6a7 | 941 | } |
7a5d0fbb | 942 | } |
bda97eab HD |
943 | unlock_page(page); |
944 | } | |
0cd6144a | 945 | pagevec_remove_exceptionals(&pvec); |
24513264 | 946 | pagevec_release(&pvec); |
bda97eab HD |
947 | index++; |
948 | } | |
94c1e62d | 949 | |
4595ef88 | 950 | spin_lock_irq(&info->lock); |
7a5d0fbb | 951 | info->swapped -= nr_swaps_freed; |
1da177e4 | 952 | shmem_recalc_inode(inode); |
4595ef88 | 953 | spin_unlock_irq(&info->lock); |
1635f6a7 | 954 | } |
1da177e4 | 955 | |
1635f6a7 HD |
956 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
957 | { | |
958 | shmem_undo_range(inode, lstart, lend, false); | |
078cd827 | 959 | inode->i_ctime = inode->i_mtime = current_time(inode); |
1da177e4 | 960 | } |
94c1e62d | 961 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 962 | |
a528d35e DH |
963 | static int shmem_getattr(const struct path *path, struct kstat *stat, |
964 | u32 request_mask, unsigned int query_flags) | |
44a30220 | 965 | { |
a528d35e | 966 | struct inode *inode = path->dentry->d_inode; |
44a30220 YZ |
967 | struct shmem_inode_info *info = SHMEM_I(inode); |
968 | ||
d0424c42 | 969 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
4595ef88 | 970 | spin_lock_irq(&info->lock); |
d0424c42 | 971 | shmem_recalc_inode(inode); |
4595ef88 | 972 | spin_unlock_irq(&info->lock); |
d0424c42 | 973 | } |
44a30220 | 974 | generic_fillattr(inode, stat); |
44a30220 YZ |
975 | return 0; |
976 | } | |
977 | ||
94c1e62d | 978 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 979 | { |
75c3cfa8 | 980 | struct inode *inode = d_inode(dentry); |
40e041a2 | 981 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 982 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 LT |
983 | int error; |
984 | ||
31051c85 | 985 | error = setattr_prepare(dentry, attr); |
db78b877 CH |
986 | if (error) |
987 | return error; | |
988 | ||
94c1e62d HD |
989 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
990 | loff_t oldsize = inode->i_size; | |
991 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 992 | |
40e041a2 DH |
993 | /* protected by i_mutex */ |
994 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
995 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
996 | return -EPERM; | |
997 | ||
94c1e62d | 998 | if (newsize != oldsize) { |
77142517 KK |
999 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
1000 | oldsize, newsize); | |
1001 | if (error) | |
1002 | return error; | |
94c1e62d | 1003 | i_size_write(inode, newsize); |
078cd827 | 1004 | inode->i_ctime = inode->i_mtime = current_time(inode); |
94c1e62d | 1005 | } |
afa2db2f | 1006 | if (newsize <= oldsize) { |
94c1e62d | 1007 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
1008 | if (oldsize > holebegin) |
1009 | unmap_mapping_range(inode->i_mapping, | |
1010 | holebegin, 0, 1); | |
1011 | if (info->alloced) | |
1012 | shmem_truncate_range(inode, | |
1013 | newsize, (loff_t)-1); | |
94c1e62d | 1014 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
1015 | if (oldsize > holebegin) |
1016 | unmap_mapping_range(inode->i_mapping, | |
1017 | holebegin, 0, 1); | |
779750d2 KS |
1018 | |
1019 | /* | |
1020 | * Part of the huge page can be beyond i_size: subject | |
1021 | * to shrink under memory pressure. | |
1022 | */ | |
1023 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { | |
1024 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1025 | /* |
1026 | * _careful to defend against unlocked access to | |
1027 | * ->shrink_list in shmem_unused_huge_shrink() | |
1028 | */ | |
1029 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1030 | list_add_tail(&info->shrinklist, |
1031 | &sbinfo->shrinklist); | |
1032 | sbinfo->shrinklist_len++; | |
1033 | } | |
1034 | spin_unlock(&sbinfo->shrinklist_lock); | |
1035 | } | |
94c1e62d | 1036 | } |
1da177e4 LT |
1037 | } |
1038 | ||
db78b877 | 1039 | setattr_copy(inode, attr); |
db78b877 | 1040 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 1041 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
1042 | return error; |
1043 | } | |
1044 | ||
1f895f75 | 1045 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1046 | { |
1da177e4 | 1047 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1048 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 | 1049 | |
3889e6e7 | 1050 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
1051 | shmem_unacct_size(info->flags, inode->i_size); |
1052 | inode->i_size = 0; | |
3889e6e7 | 1053 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1054 | if (!list_empty(&info->shrinklist)) { |
1055 | spin_lock(&sbinfo->shrinklist_lock); | |
1056 | if (!list_empty(&info->shrinklist)) { | |
1057 | list_del_init(&info->shrinklist); | |
1058 | sbinfo->shrinklist_len--; | |
1059 | } | |
1060 | spin_unlock(&sbinfo->shrinklist_lock); | |
1061 | } | |
1da177e4 | 1062 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 1063 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 1064 | list_del_init(&info->swaplist); |
cb5f7b9a | 1065 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1066 | } |
3ed47db3 | 1067 | } |
b09e0fa4 | 1068 | |
38f38657 | 1069 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 1070 | WARN_ON(inode->i_blocks); |
5b04c689 | 1071 | shmem_free_inode(inode->i_sb); |
dbd5768f | 1072 | clear_inode(inode); |
1da177e4 LT |
1073 | } |
1074 | ||
478922e2 MW |
1075 | static unsigned long find_swap_entry(struct radix_tree_root *root, void *item) |
1076 | { | |
1077 | struct radix_tree_iter iter; | |
1078 | void **slot; | |
1079 | unsigned long found = -1; | |
1080 | unsigned int checked = 0; | |
1081 | ||
1082 | rcu_read_lock(); | |
1083 | radix_tree_for_each_slot(slot, root, &iter, 0) { | |
1084 | if (*slot == item) { | |
1085 | found = iter.index; | |
1086 | break; | |
1087 | } | |
1088 | checked++; | |
1089 | if ((checked % 4096) != 0) | |
1090 | continue; | |
1091 | slot = radix_tree_iter_resume(slot, &iter); | |
1092 | cond_resched_rcu(); | |
1093 | } | |
1094 | ||
1095 | rcu_read_unlock(); | |
1096 | return found; | |
1097 | } | |
1098 | ||
46f65ec1 HD |
1099 | /* |
1100 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1101 | */ | |
41ffe5d5 | 1102 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 1103 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 1104 | { |
285b2c4f | 1105 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 1106 | void *radswap; |
41ffe5d5 | 1107 | pgoff_t index; |
bde05d1c HD |
1108 | gfp_t gfp; |
1109 | int error = 0; | |
1da177e4 | 1110 | |
46f65ec1 | 1111 | radswap = swp_to_radix_entry(swap); |
478922e2 | 1112 | index = find_swap_entry(&mapping->page_tree, radswap); |
46f65ec1 | 1113 | if (index == -1) |
00501b53 | 1114 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 1115 | |
1b1b32f2 HD |
1116 | /* |
1117 | * Move _head_ to start search for next from here. | |
1f895f75 | 1118 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 1119 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 1120 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
1121 | */ |
1122 | if (shmem_swaplist.next != &info->swaplist) | |
1123 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 1124 | |
bde05d1c HD |
1125 | gfp = mapping_gfp_mask(mapping); |
1126 | if (shmem_should_replace_page(*pagep, gfp)) { | |
1127 | mutex_unlock(&shmem_swaplist_mutex); | |
1128 | error = shmem_replace_page(pagep, gfp, info, index); | |
1129 | mutex_lock(&shmem_swaplist_mutex); | |
1130 | /* | |
1131 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
1132 | * allocation, but the inode might have been freed while we |
1133 | * dropped it: although a racing shmem_evict_inode() cannot | |
1134 | * complete without emptying the radix_tree, our page lock | |
1135 | * on this swapcache page is not enough to prevent that - | |
1136 | * free_swap_and_cache() of our swap entry will only | |
1137 | * trylock_page(), removing swap from radix_tree whatever. | |
1138 | * | |
1139 | * We must not proceed to shmem_add_to_page_cache() if the | |
1140 | * inode has been freed, but of course we cannot rely on | |
1141 | * inode or mapping or info to check that. However, we can | |
1142 | * safely check if our swap entry is still in use (and here | |
1143 | * it can't have got reused for another page): if it's still | |
1144 | * in use, then the inode cannot have been freed yet, and we | |
1145 | * can safely proceed (if it's no longer in use, that tells | |
1146 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
1147 | */ |
1148 | if (!page_swapcount(*pagep)) | |
1149 | error = -ENOENT; | |
1150 | } | |
1151 | ||
d13d1443 | 1152 | /* |
778dd893 HD |
1153 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
1154 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
1155 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 1156 | */ |
bde05d1c HD |
1157 | if (!error) |
1158 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 1159 | radswap); |
48f170fb | 1160 | if (error != -ENOMEM) { |
46f65ec1 HD |
1161 | /* |
1162 | * Truncation and eviction use free_swap_and_cache(), which | |
1163 | * only does trylock page: if we raced, best clean up here. | |
1164 | */ | |
bde05d1c HD |
1165 | delete_from_swap_cache(*pagep); |
1166 | set_page_dirty(*pagep); | |
46f65ec1 | 1167 | if (!error) { |
4595ef88 | 1168 | spin_lock_irq(&info->lock); |
46f65ec1 | 1169 | info->swapped--; |
4595ef88 | 1170 | spin_unlock_irq(&info->lock); |
46f65ec1 HD |
1171 | swap_free(swap); |
1172 | } | |
1da177e4 | 1173 | } |
2e0e26c7 | 1174 | return error; |
1da177e4 LT |
1175 | } |
1176 | ||
1177 | /* | |
46f65ec1 | 1178 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 1179 | */ |
41ffe5d5 | 1180 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 1181 | { |
41ffe5d5 | 1182 | struct list_head *this, *next; |
1da177e4 | 1183 | struct shmem_inode_info *info; |
00501b53 | 1184 | struct mem_cgroup *memcg; |
bde05d1c HD |
1185 | int error = 0; |
1186 | ||
1187 | /* | |
1188 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 1189 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 1190 | */ |
0142ef6c | 1191 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 1192 | goto out; |
778dd893 HD |
1193 | |
1194 | /* | |
1195 | * Charge page using GFP_KERNEL while we can wait, before taking | |
1196 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
1197 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 1198 | */ |
f627c2f5 KS |
1199 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg, |
1200 | false); | |
778dd893 HD |
1201 | if (error) |
1202 | goto out; | |
46f65ec1 | 1203 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 1204 | error = -EAGAIN; |
1da177e4 | 1205 | |
cb5f7b9a | 1206 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
1207 | list_for_each_safe(this, next, &shmem_swaplist) { |
1208 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 1209 | if (info->swapped) |
00501b53 | 1210 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
1211 | else |
1212 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1213 | cond_resched(); |
00501b53 | 1214 | if (error != -EAGAIN) |
778dd893 | 1215 | break; |
00501b53 | 1216 | /* found nothing in this: move on to search the next */ |
1da177e4 | 1217 | } |
cb5f7b9a | 1218 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1219 | |
00501b53 JW |
1220 | if (error) { |
1221 | if (error != -ENOMEM) | |
1222 | error = 0; | |
f627c2f5 | 1223 | mem_cgroup_cancel_charge(page, memcg, false); |
00501b53 | 1224 | } else |
f627c2f5 | 1225 | mem_cgroup_commit_charge(page, memcg, true, false); |
778dd893 | 1226 | out: |
aaa46865 | 1227 | unlock_page(page); |
09cbfeaf | 1228 | put_page(page); |
778dd893 | 1229 | return error; |
1da177e4 LT |
1230 | } |
1231 | ||
1232 | /* | |
1233 | * Move the page from the page cache to the swap cache. | |
1234 | */ | |
1235 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1236 | { | |
1237 | struct shmem_inode_info *info; | |
1da177e4 | 1238 | struct address_space *mapping; |
1da177e4 | 1239 | struct inode *inode; |
6922c0c7 HD |
1240 | swp_entry_t swap; |
1241 | pgoff_t index; | |
1da177e4 | 1242 | |
800d8c63 | 1243 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1244 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1245 | mapping = page->mapping; |
1246 | index = page->index; | |
1247 | inode = mapping->host; | |
1248 | info = SHMEM_I(inode); | |
1249 | if (info->flags & VM_LOCKED) | |
1250 | goto redirty; | |
d9fe526a | 1251 | if (!total_swap_pages) |
1da177e4 LT |
1252 | goto redirty; |
1253 | ||
d9fe526a | 1254 | /* |
97b713ba CH |
1255 | * Our capabilities prevent regular writeback or sync from ever calling |
1256 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1257 | * its underlying filesystem, in which case tmpfs should write out to | |
1258 | * swap only in response to memory pressure, and not for the writeback | |
1259 | * threads or sync. | |
d9fe526a | 1260 | */ |
48f170fb HD |
1261 | if (!wbc->for_reclaim) { |
1262 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1263 | goto redirty; | |
1264 | } | |
1635f6a7 HD |
1265 | |
1266 | /* | |
1267 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1268 | * value into swapfile.c, the only way we can correctly account for a | |
1269 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1270 | * |
1271 | * That's okay for a page already fallocated earlier, but if we have | |
1272 | * not yet completed the fallocation, then (a) we want to keep track | |
1273 | * of this page in case we have to undo it, and (b) it may not be a | |
1274 | * good idea to continue anyway, once we're pushing into swap. So | |
1275 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1276 | */ |
1277 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1278 | if (inode->i_private) { |
1279 | struct shmem_falloc *shmem_falloc; | |
1280 | spin_lock(&inode->i_lock); | |
1281 | shmem_falloc = inode->i_private; | |
1282 | if (shmem_falloc && | |
8e205f77 | 1283 | !shmem_falloc->waitq && |
1aac1400 HD |
1284 | index >= shmem_falloc->start && |
1285 | index < shmem_falloc->next) | |
1286 | shmem_falloc->nr_unswapped++; | |
1287 | else | |
1288 | shmem_falloc = NULL; | |
1289 | spin_unlock(&inode->i_lock); | |
1290 | if (shmem_falloc) | |
1291 | goto redirty; | |
1292 | } | |
1635f6a7 HD |
1293 | clear_highpage(page); |
1294 | flush_dcache_page(page); | |
1295 | SetPageUptodate(page); | |
1296 | } | |
1297 | ||
38d8b4e6 | 1298 | swap = get_swap_page(page); |
48f170fb HD |
1299 | if (!swap.val) |
1300 | goto redirty; | |
d9fe526a | 1301 | |
37e84351 VD |
1302 | if (mem_cgroup_try_charge_swap(page, swap)) |
1303 | goto free_swap; | |
1304 | ||
b1dea800 HD |
1305 | /* |
1306 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1307 | * if it's not already there. Do it now before the page is |
1308 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1309 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1310 | * we've incremented swapped, because shmem_unuse_inode() will |
1311 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1312 | */ |
48f170fb HD |
1313 | mutex_lock(&shmem_swaplist_mutex); |
1314 | if (list_empty(&info->swaplist)) | |
1315 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 1316 | |
48f170fb | 1317 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
4595ef88 | 1318 | spin_lock_irq(&info->lock); |
6922c0c7 | 1319 | shmem_recalc_inode(inode); |
267a4c76 | 1320 | info->swapped++; |
4595ef88 | 1321 | spin_unlock_irq(&info->lock); |
6922c0c7 | 1322 | |
267a4c76 HD |
1323 | swap_shmem_alloc(swap); |
1324 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1325 | ||
6922c0c7 | 1326 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1327 | BUG_ON(page_mapped(page)); |
9fab5619 | 1328 | swap_writepage(page, wbc); |
1da177e4 LT |
1329 | return 0; |
1330 | } | |
1331 | ||
6922c0c7 | 1332 | mutex_unlock(&shmem_swaplist_mutex); |
37e84351 | 1333 | free_swap: |
75f6d6d2 | 1334 | put_swap_page(page, swap); |
1da177e4 LT |
1335 | redirty: |
1336 | set_page_dirty(page); | |
d9fe526a HD |
1337 | if (wbc->for_reclaim) |
1338 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1339 | unlock_page(page); | |
1340 | return 0; | |
1da177e4 LT |
1341 | } |
1342 | ||
75edd345 | 1343 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1344 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1345 | { |
095f1fc4 | 1346 | char buffer[64]; |
680d794b | 1347 | |
71fe804b | 1348 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1349 | return; /* show nothing */ |
680d794b | 1350 | |
a7a88b23 | 1351 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1352 | |
1353 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1354 | } |
71fe804b LS |
1355 | |
1356 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1357 | { | |
1358 | struct mempolicy *mpol = NULL; | |
1359 | if (sbinfo->mpol) { | |
1360 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1361 | mpol = sbinfo->mpol; | |
1362 | mpol_get(mpol); | |
1363 | spin_unlock(&sbinfo->stat_lock); | |
1364 | } | |
1365 | return mpol; | |
1366 | } | |
75edd345 HD |
1367 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1368 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1369 | { | |
1370 | } | |
1371 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1372 | { | |
1373 | return NULL; | |
1374 | } | |
1375 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1376 | #ifndef CONFIG_NUMA | |
1377 | #define vm_policy vm_private_data | |
1378 | #endif | |
680d794b | 1379 | |
800d8c63 KS |
1380 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1381 | struct shmem_inode_info *info, pgoff_t index) | |
1382 | { | |
1383 | /* Create a pseudo vma that just contains the policy */ | |
1384 | vma->vm_start = 0; | |
1385 | /* Bias interleave by inode number to distribute better across nodes */ | |
1386 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
1387 | vma->vm_ops = NULL; | |
1388 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); | |
1389 | } | |
1390 | ||
1391 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1392 | { | |
1393 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1394 | mpol_cond_put(vma->vm_policy); | |
1395 | } | |
1396 | ||
41ffe5d5 HD |
1397 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1398 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1399 | { |
1da177e4 | 1400 | struct vm_area_struct pvma; |
18a2f371 | 1401 | struct page *page; |
52cd3b07 | 1402 | |
800d8c63 | 1403 | shmem_pseudo_vma_init(&pvma, info, index); |
18a2f371 | 1404 | page = swapin_readahead(swap, gfp, &pvma, 0); |
800d8c63 | 1405 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1406 | |
800d8c63 KS |
1407 | return page; |
1408 | } | |
1409 | ||
1410 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1411 | struct shmem_inode_info *info, pgoff_t index) | |
1412 | { | |
1413 | struct vm_area_struct pvma; | |
1414 | struct inode *inode = &info->vfs_inode; | |
1415 | struct address_space *mapping = inode->i_mapping; | |
4620a06e | 1416 | pgoff_t idx, hindex; |
800d8c63 KS |
1417 | void __rcu **results; |
1418 | struct page *page; | |
1419 | ||
e496cf3d | 1420 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1421 | return NULL; |
1422 | ||
4620a06e | 1423 | hindex = round_down(index, HPAGE_PMD_NR); |
800d8c63 KS |
1424 | rcu_read_lock(); |
1425 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx, | |
1426 | hindex, 1) && idx < hindex + HPAGE_PMD_NR) { | |
1427 | rcu_read_unlock(); | |
1428 | return NULL; | |
1429 | } | |
1430 | rcu_read_unlock(); | |
18a2f371 | 1431 | |
800d8c63 KS |
1432 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1433 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
1434 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); | |
1435 | shmem_pseudo_vma_destroy(&pvma); | |
1436 | if (page) | |
1437 | prep_transhuge_page(page); | |
18a2f371 | 1438 | return page; |
1da177e4 LT |
1439 | } |
1440 | ||
02098fea | 1441 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1442 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1443 | { |
1444 | struct vm_area_struct pvma; | |
18a2f371 | 1445 | struct page *page; |
1da177e4 | 1446 | |
800d8c63 KS |
1447 | shmem_pseudo_vma_init(&pvma, info, index); |
1448 | page = alloc_page_vma(gfp, &pvma, 0); | |
1449 | shmem_pseudo_vma_destroy(&pvma); | |
1450 | ||
1451 | return page; | |
1452 | } | |
1453 | ||
1454 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
1455 | struct shmem_inode_info *info, struct shmem_sb_info *sbinfo, | |
1456 | pgoff_t index, bool huge) | |
1457 | { | |
1458 | struct page *page; | |
1459 | int nr; | |
1460 | int err = -ENOSPC; | |
52cd3b07 | 1461 | |
e496cf3d | 1462 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1463 | huge = false; |
1464 | nr = huge ? HPAGE_PMD_NR : 1; | |
1465 | ||
1466 | if (shmem_acct_block(info->flags, nr)) | |
1467 | goto failed; | |
1468 | if (sbinfo->max_blocks) { | |
1469 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
1470 | sbinfo->max_blocks - nr) > 0) | |
1471 | goto unacct; | |
1472 | percpu_counter_add(&sbinfo->used_blocks, nr); | |
1473 | } | |
1474 | ||
1475 | if (huge) | |
1476 | page = shmem_alloc_hugepage(gfp, info, index); | |
1477 | else | |
1478 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1479 | if (page) { |
1480 | __SetPageLocked(page); | |
1481 | __SetPageSwapBacked(page); | |
800d8c63 | 1482 | return page; |
75edd345 | 1483 | } |
18a2f371 | 1484 | |
800d8c63 KS |
1485 | err = -ENOMEM; |
1486 | if (sbinfo->max_blocks) | |
1487 | percpu_counter_add(&sbinfo->used_blocks, -nr); | |
1488 | unacct: | |
1489 | shmem_unacct_blocks(info->flags, nr); | |
1490 | failed: | |
1491 | return ERR_PTR(err); | |
1da177e4 | 1492 | } |
71fe804b | 1493 | |
bde05d1c HD |
1494 | /* |
1495 | * When a page is moved from swapcache to shmem filecache (either by the | |
1496 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1497 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1498 | * ignorance of the mapping it belongs to. If that mapping has special | |
1499 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1500 | * we may need to copy to a suitable page before moving to filecache. | |
1501 | * | |
1502 | * In a future release, this may well be extended to respect cpuset and | |
1503 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1504 | * but for now it is a simple matter of zone. | |
1505 | */ | |
1506 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1507 | { | |
1508 | return page_zonenum(page) > gfp_zone(gfp); | |
1509 | } | |
1510 | ||
1511 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1512 | struct shmem_inode_info *info, pgoff_t index) | |
1513 | { | |
1514 | struct page *oldpage, *newpage; | |
1515 | struct address_space *swap_mapping; | |
1516 | pgoff_t swap_index; | |
1517 | int error; | |
1518 | ||
1519 | oldpage = *pagep; | |
1520 | swap_index = page_private(oldpage); | |
1521 | swap_mapping = page_mapping(oldpage); | |
1522 | ||
1523 | /* | |
1524 | * We have arrived here because our zones are constrained, so don't | |
1525 | * limit chance of success by further cpuset and node constraints. | |
1526 | */ | |
1527 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1528 | newpage = shmem_alloc_page(gfp, info, index); | |
1529 | if (!newpage) | |
1530 | return -ENOMEM; | |
bde05d1c | 1531 | |
09cbfeaf | 1532 | get_page(newpage); |
bde05d1c | 1533 | copy_highpage(newpage, oldpage); |
0142ef6c | 1534 | flush_dcache_page(newpage); |
bde05d1c | 1535 | |
9956edf3 HD |
1536 | __SetPageLocked(newpage); |
1537 | __SetPageSwapBacked(newpage); | |
bde05d1c | 1538 | SetPageUptodate(newpage); |
bde05d1c | 1539 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1540 | SetPageSwapCache(newpage); |
1541 | ||
1542 | /* | |
1543 | * Our caller will very soon move newpage out of swapcache, but it's | |
1544 | * a nice clean interface for us to replace oldpage by newpage there. | |
1545 | */ | |
1546 | spin_lock_irq(&swap_mapping->tree_lock); | |
1547 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1548 | newpage); | |
0142ef6c | 1549 | if (!error) { |
11fb9989 MG |
1550 | __inc_node_page_state(newpage, NR_FILE_PAGES); |
1551 | __dec_node_page_state(oldpage, NR_FILE_PAGES); | |
0142ef6c | 1552 | } |
bde05d1c | 1553 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1554 | |
0142ef6c HD |
1555 | if (unlikely(error)) { |
1556 | /* | |
1557 | * Is this possible? I think not, now that our callers check | |
1558 | * both PageSwapCache and page_private after getting page lock; | |
1559 | * but be defensive. Reverse old to newpage for clear and free. | |
1560 | */ | |
1561 | oldpage = newpage; | |
1562 | } else { | |
6a93ca8f | 1563 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1564 | lru_cache_add_anon(newpage); |
1565 | *pagep = newpage; | |
1566 | } | |
bde05d1c HD |
1567 | |
1568 | ClearPageSwapCache(oldpage); | |
1569 | set_page_private(oldpage, 0); | |
1570 | ||
1571 | unlock_page(oldpage); | |
09cbfeaf KS |
1572 | put_page(oldpage); |
1573 | put_page(oldpage); | |
0142ef6c | 1574 | return error; |
bde05d1c HD |
1575 | } |
1576 | ||
1da177e4 | 1577 | /* |
68da9f05 | 1578 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1579 | * |
1580 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1581 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1582 | * entry since a page cannot live in both the swap and page cache. |
1583 | * | |
1584 | * fault_mm and fault_type are only supplied by shmem_fault: | |
1585 | * otherwise they are NULL. | |
1da177e4 | 1586 | */ |
41ffe5d5 | 1587 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 1588 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
cfda0526 | 1589 | struct vm_area_struct *vma, struct vm_fault *vmf, int *fault_type) |
1da177e4 LT |
1590 | { |
1591 | struct address_space *mapping = inode->i_mapping; | |
23f919d4 | 1592 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 | 1593 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1594 | struct mm_struct *charge_mm; |
00501b53 | 1595 | struct mem_cgroup *memcg; |
27ab7006 | 1596 | struct page *page; |
1da177e4 | 1597 | swp_entry_t swap; |
657e3038 | 1598 | enum sgp_type sgp_huge = sgp; |
800d8c63 | 1599 | pgoff_t hindex = index; |
1da177e4 | 1600 | int error; |
54af6042 | 1601 | int once = 0; |
1635f6a7 | 1602 | int alloced = 0; |
1da177e4 | 1603 | |
09cbfeaf | 1604 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1605 | return -EFBIG; |
657e3038 KS |
1606 | if (sgp == SGP_NOHUGE || sgp == SGP_HUGE) |
1607 | sgp = SGP_CACHE; | |
1da177e4 | 1608 | repeat: |
54af6042 | 1609 | swap.val = 0; |
0cd6144a | 1610 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1611 | if (radix_tree_exceptional_entry(page)) { |
1612 | swap = radix_to_swp_entry(page); | |
1613 | page = NULL; | |
1614 | } | |
1615 | ||
75edd345 | 1616 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1617 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 1618 | error = -EINVAL; |
267a4c76 | 1619 | goto unlock; |
54af6042 HD |
1620 | } |
1621 | ||
66d2f4d2 HD |
1622 | if (page && sgp == SGP_WRITE) |
1623 | mark_page_accessed(page); | |
1624 | ||
1635f6a7 HD |
1625 | /* fallocated page? */ |
1626 | if (page && !PageUptodate(page)) { | |
1627 | if (sgp != SGP_READ) | |
1628 | goto clear; | |
1629 | unlock_page(page); | |
09cbfeaf | 1630 | put_page(page); |
1635f6a7 HD |
1631 | page = NULL; |
1632 | } | |
54af6042 | 1633 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1634 | *pagep = page; |
1635 | return 0; | |
27ab7006 HD |
1636 | } |
1637 | ||
1638 | /* | |
54af6042 HD |
1639 | * Fast cache lookup did not find it: |
1640 | * bring it back from swap or allocate. | |
27ab7006 | 1641 | */ |
54af6042 | 1642 | sbinfo = SHMEM_SB(inode->i_sb); |
cfda0526 | 1643 | charge_mm = vma ? vma->vm_mm : current->mm; |
1da177e4 | 1644 | |
1da177e4 LT |
1645 | if (swap.val) { |
1646 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1647 | page = lookup_swap_cache(swap); |
1648 | if (!page) { | |
9e18eb29 ALC |
1649 | /* Or update major stats only when swapin succeeds?? */ |
1650 | if (fault_type) { | |
68da9f05 | 1651 | *fault_type |= VM_FAULT_MAJOR; |
9e18eb29 | 1652 | count_vm_event(PGMAJFAULT); |
2262185c | 1653 | count_memcg_event_mm(charge_mm, PGMAJFAULT); |
9e18eb29 ALC |
1654 | } |
1655 | /* Here we actually start the io */ | |
41ffe5d5 | 1656 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1657 | if (!page) { |
54af6042 HD |
1658 | error = -ENOMEM; |
1659 | goto failed; | |
1da177e4 | 1660 | } |
1da177e4 LT |
1661 | } |
1662 | ||
1663 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1664 | lock_page(page); |
0142ef6c | 1665 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1666 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1667 | error = -EEXIST; /* try again */ |
d1899228 | 1668 | goto unlock; |
bde05d1c | 1669 | } |
27ab7006 | 1670 | if (!PageUptodate(page)) { |
1da177e4 | 1671 | error = -EIO; |
54af6042 | 1672 | goto failed; |
1da177e4 | 1673 | } |
54af6042 HD |
1674 | wait_on_page_writeback(page); |
1675 | ||
bde05d1c HD |
1676 | if (shmem_should_replace_page(page, gfp)) { |
1677 | error = shmem_replace_page(&page, gfp, info, index); | |
1678 | if (error) | |
1679 | goto failed; | |
1da177e4 | 1680 | } |
27ab7006 | 1681 | |
9e18eb29 | 1682 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
f627c2f5 | 1683 | false); |
d1899228 | 1684 | if (!error) { |
aa3b1895 | 1685 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1686 | swp_to_radix_entry(swap)); |
215c02bc HD |
1687 | /* |
1688 | * We already confirmed swap under page lock, and make | |
1689 | * no memory allocation here, so usually no possibility | |
1690 | * of error; but free_swap_and_cache() only trylocks a | |
1691 | * page, so it is just possible that the entry has been | |
1692 | * truncated or holepunched since swap was confirmed. | |
1693 | * shmem_undo_range() will have done some of the | |
1694 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1695 | * the rest. |
215c02bc HD |
1696 | * Reset swap.val? No, leave it so "failed" goes back to |
1697 | * "repeat": reading a hole and writing should succeed. | |
1698 | */ | |
00501b53 | 1699 | if (error) { |
f627c2f5 | 1700 | mem_cgroup_cancel_charge(page, memcg, false); |
215c02bc | 1701 | delete_from_swap_cache(page); |
00501b53 | 1702 | } |
d1899228 | 1703 | } |
54af6042 HD |
1704 | if (error) |
1705 | goto failed; | |
1706 | ||
f627c2f5 | 1707 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1708 | |
4595ef88 | 1709 | spin_lock_irq(&info->lock); |
285b2c4f | 1710 | info->swapped--; |
54af6042 | 1711 | shmem_recalc_inode(inode); |
4595ef88 | 1712 | spin_unlock_irq(&info->lock); |
54af6042 | 1713 | |
66d2f4d2 HD |
1714 | if (sgp == SGP_WRITE) |
1715 | mark_page_accessed(page); | |
1716 | ||
54af6042 | 1717 | delete_from_swap_cache(page); |
27ab7006 HD |
1718 | set_page_dirty(page); |
1719 | swap_free(swap); | |
1720 | ||
54af6042 | 1721 | } else { |
cfda0526 MR |
1722 | if (vma && userfaultfd_missing(vma)) { |
1723 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
1724 | return 0; | |
1725 | } | |
1726 | ||
800d8c63 KS |
1727 | /* shmem_symlink() */ |
1728 | if (mapping->a_ops != &shmem_aops) | |
1729 | goto alloc_nohuge; | |
657e3038 | 1730 | if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE) |
800d8c63 KS |
1731 | goto alloc_nohuge; |
1732 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1733 | goto alloc_huge; | |
1734 | switch (sbinfo->huge) { | |
1735 | loff_t i_size; | |
1736 | pgoff_t off; | |
1737 | case SHMEM_HUGE_NEVER: | |
1738 | goto alloc_nohuge; | |
1739 | case SHMEM_HUGE_WITHIN_SIZE: | |
1740 | off = round_up(index, HPAGE_PMD_NR); | |
1741 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1742 | if (i_size >= HPAGE_PMD_SIZE && | |
1743 | i_size >> PAGE_SHIFT >= off) | |
1744 | goto alloc_huge; | |
1745 | /* fallthrough */ | |
1746 | case SHMEM_HUGE_ADVISE: | |
657e3038 KS |
1747 | if (sgp_huge == SGP_HUGE) |
1748 | goto alloc_huge; | |
1749 | /* TODO: implement fadvise() hints */ | |
800d8c63 | 1750 | goto alloc_nohuge; |
54af6042 | 1751 | } |
1da177e4 | 1752 | |
800d8c63 KS |
1753 | alloc_huge: |
1754 | page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1755 | index, true); | |
1756 | if (IS_ERR(page)) { | |
1757 | alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1758 | index, false); | |
1da177e4 | 1759 | } |
800d8c63 | 1760 | if (IS_ERR(page)) { |
779750d2 | 1761 | int retry = 5; |
800d8c63 KS |
1762 | error = PTR_ERR(page); |
1763 | page = NULL; | |
779750d2 KS |
1764 | if (error != -ENOSPC) |
1765 | goto failed; | |
1766 | /* | |
1767 | * Try to reclaim some spece by splitting a huge page | |
1768 | * beyond i_size on the filesystem. | |
1769 | */ | |
1770 | while (retry--) { | |
1771 | int ret; | |
1772 | ret = shmem_unused_huge_shrink(sbinfo, NULL, 1); | |
1773 | if (ret == SHRINK_STOP) | |
1774 | break; | |
1775 | if (ret) | |
1776 | goto alloc_nohuge; | |
1777 | } | |
800d8c63 KS |
1778 | goto failed; |
1779 | } | |
1780 | ||
1781 | if (PageTransHuge(page)) | |
1782 | hindex = round_down(index, HPAGE_PMD_NR); | |
1783 | else | |
1784 | hindex = index; | |
1785 | ||
66d2f4d2 | 1786 | if (sgp == SGP_WRITE) |
eb39d618 | 1787 | __SetPageReferenced(page); |
66d2f4d2 | 1788 | |
9e18eb29 | 1789 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
800d8c63 | 1790 | PageTransHuge(page)); |
54af6042 | 1791 | if (error) |
800d8c63 KS |
1792 | goto unacct; |
1793 | error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK, | |
1794 | compound_order(page)); | |
b065b432 | 1795 | if (!error) { |
800d8c63 | 1796 | error = shmem_add_to_page_cache(page, mapping, hindex, |
fed400a1 | 1797 | NULL); |
b065b432 HD |
1798 | radix_tree_preload_end(); |
1799 | } | |
1800 | if (error) { | |
800d8c63 KS |
1801 | mem_cgroup_cancel_charge(page, memcg, |
1802 | PageTransHuge(page)); | |
1803 | goto unacct; | |
b065b432 | 1804 | } |
800d8c63 KS |
1805 | mem_cgroup_commit_charge(page, memcg, false, |
1806 | PageTransHuge(page)); | |
54af6042 HD |
1807 | lru_cache_add_anon(page); |
1808 | ||
4595ef88 | 1809 | spin_lock_irq(&info->lock); |
800d8c63 KS |
1810 | info->alloced += 1 << compound_order(page); |
1811 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); | |
54af6042 | 1812 | shmem_recalc_inode(inode); |
4595ef88 | 1813 | spin_unlock_irq(&info->lock); |
1635f6a7 | 1814 | alloced = true; |
54af6042 | 1815 | |
779750d2 KS |
1816 | if (PageTransHuge(page) && |
1817 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < | |
1818 | hindex + HPAGE_PMD_NR - 1) { | |
1819 | /* | |
1820 | * Part of the huge page is beyond i_size: subject | |
1821 | * to shrink under memory pressure. | |
1822 | */ | |
1823 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1824 | /* |
1825 | * _careful to defend against unlocked access to | |
1826 | * ->shrink_list in shmem_unused_huge_shrink() | |
1827 | */ | |
1828 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1829 | list_add_tail(&info->shrinklist, |
1830 | &sbinfo->shrinklist); | |
1831 | sbinfo->shrinklist_len++; | |
1832 | } | |
1833 | spin_unlock(&sbinfo->shrinklist_lock); | |
1834 | } | |
1835 | ||
ec9516fb | 1836 | /* |
1635f6a7 HD |
1837 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1838 | */ | |
1839 | if (sgp == SGP_FALLOC) | |
1840 | sgp = SGP_WRITE; | |
1841 | clear: | |
1842 | /* | |
1843 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1844 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1845 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb | 1846 | */ |
800d8c63 KS |
1847 | if (sgp != SGP_WRITE && !PageUptodate(page)) { |
1848 | struct page *head = compound_head(page); | |
1849 | int i; | |
1850 | ||
1851 | for (i = 0; i < (1 << compound_order(head)); i++) { | |
1852 | clear_highpage(head + i); | |
1853 | flush_dcache_page(head + i); | |
1854 | } | |
1855 | SetPageUptodate(head); | |
ec9516fb | 1856 | } |
1da177e4 | 1857 | } |
bde05d1c | 1858 | |
54af6042 | 1859 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1860 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1861 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1862 | if (alloced) { |
1863 | ClearPageDirty(page); | |
1864 | delete_from_page_cache(page); | |
4595ef88 | 1865 | spin_lock_irq(&info->lock); |
267a4c76 | 1866 | shmem_recalc_inode(inode); |
4595ef88 | 1867 | spin_unlock_irq(&info->lock); |
267a4c76 | 1868 | } |
54af6042 | 1869 | error = -EINVAL; |
267a4c76 | 1870 | goto unlock; |
e83c32e8 | 1871 | } |
800d8c63 | 1872 | *pagep = page + index - hindex; |
54af6042 | 1873 | return 0; |
1da177e4 | 1874 | |
59a16ead | 1875 | /* |
54af6042 | 1876 | * Error recovery. |
59a16ead | 1877 | */ |
54af6042 | 1878 | unacct: |
800d8c63 KS |
1879 | if (sbinfo->max_blocks) |
1880 | percpu_counter_sub(&sbinfo->used_blocks, | |
1881 | 1 << compound_order(page)); | |
1882 | shmem_unacct_blocks(info->flags, 1 << compound_order(page)); | |
1883 | ||
1884 | if (PageTransHuge(page)) { | |
1885 | unlock_page(page); | |
1886 | put_page(page); | |
1887 | goto alloc_nohuge; | |
1888 | } | |
54af6042 | 1889 | failed: |
267a4c76 | 1890 | if (swap.val && !shmem_confirm_swap(mapping, index, swap)) |
d1899228 HD |
1891 | error = -EEXIST; |
1892 | unlock: | |
27ab7006 | 1893 | if (page) { |
54af6042 | 1894 | unlock_page(page); |
09cbfeaf | 1895 | put_page(page); |
54af6042 HD |
1896 | } |
1897 | if (error == -ENOSPC && !once++) { | |
4595ef88 | 1898 | spin_lock_irq(&info->lock); |
54af6042 | 1899 | shmem_recalc_inode(inode); |
4595ef88 | 1900 | spin_unlock_irq(&info->lock); |
27ab7006 | 1901 | goto repeat; |
ff36b801 | 1902 | } |
d1899228 | 1903 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1904 | goto repeat; |
1905 | return error; | |
1da177e4 LT |
1906 | } |
1907 | ||
10d20bd2 LT |
1908 | /* |
1909 | * This is like autoremove_wake_function, but it removes the wait queue | |
1910 | * entry unconditionally - even if something else had already woken the | |
1911 | * target. | |
1912 | */ | |
ac6424b9 | 1913 | static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
10d20bd2 LT |
1914 | { |
1915 | int ret = default_wake_function(wait, mode, sync, key); | |
2055da97 | 1916 | list_del_init(&wait->entry); |
10d20bd2 LT |
1917 | return ret; |
1918 | } | |
1919 | ||
11bac800 | 1920 | static int shmem_fault(struct vm_fault *vmf) |
1da177e4 | 1921 | { |
11bac800 | 1922 | struct vm_area_struct *vma = vmf->vma; |
496ad9aa | 1923 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 1924 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
657e3038 | 1925 | enum sgp_type sgp; |
1da177e4 | 1926 | int error; |
68da9f05 | 1927 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1928 | |
f00cdc6d HD |
1929 | /* |
1930 | * Trinity finds that probing a hole which tmpfs is punching can | |
1931 | * prevent the hole-punch from ever completing: which in turn | |
1932 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1933 | * faulting pages into the hole while it's being punched. Although |
1934 | * shmem_undo_range() does remove the additions, it may be unable to | |
1935 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1936 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1937 | * | |
1938 | * It does not matter if we sometimes reach this check just before the | |
1939 | * hole-punch begins, so that one fault then races with the punch: | |
1940 | * we just need to make racing faults a rare case. | |
1941 | * | |
1942 | * The implementation below would be much simpler if we just used a | |
1943 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1944 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1945 | */ |
1946 | if (unlikely(inode->i_private)) { | |
1947 | struct shmem_falloc *shmem_falloc; | |
1948 | ||
1949 | spin_lock(&inode->i_lock); | |
1950 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1951 | if (shmem_falloc && |
1952 | shmem_falloc->waitq && | |
1953 | vmf->pgoff >= shmem_falloc->start && | |
1954 | vmf->pgoff < shmem_falloc->next) { | |
1955 | wait_queue_head_t *shmem_falloc_waitq; | |
10d20bd2 | 1956 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); |
8e205f77 HD |
1957 | |
1958 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1959 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1960 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1961 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1962 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1963 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1964 | } |
8e205f77 HD |
1965 | |
1966 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1967 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1968 | TASK_UNINTERRUPTIBLE); | |
1969 | spin_unlock(&inode->i_lock); | |
1970 | schedule(); | |
1971 | ||
1972 | /* | |
1973 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1974 | * stack of the hole-punching task: shmem_falloc_waitq | |
1975 | * is usually invalid by the time we reach here, but | |
1976 | * finish_wait() does not dereference it in that case; | |
1977 | * though i_lock needed lest racing with wake_up_all(). | |
1978 | */ | |
1979 | spin_lock(&inode->i_lock); | |
1980 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1981 | spin_unlock(&inode->i_lock); | |
1982 | return ret; | |
f00cdc6d | 1983 | } |
8e205f77 | 1984 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1985 | } |
1986 | ||
657e3038 | 1987 | sgp = SGP_CACHE; |
18600332 MH |
1988 | |
1989 | if ((vma->vm_flags & VM_NOHUGEPAGE) || | |
1990 | test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) | |
657e3038 | 1991 | sgp = SGP_NOHUGE; |
18600332 MH |
1992 | else if (vma->vm_flags & VM_HUGEPAGE) |
1993 | sgp = SGP_HUGE; | |
657e3038 KS |
1994 | |
1995 | error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp, | |
cfda0526 | 1996 | gfp, vma, vmf, &ret); |
d0217ac0 NP |
1997 | if (error) |
1998 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1999 | return ret; |
1da177e4 LT |
2000 | } |
2001 | ||
c01d5b30 HD |
2002 | unsigned long shmem_get_unmapped_area(struct file *file, |
2003 | unsigned long uaddr, unsigned long len, | |
2004 | unsigned long pgoff, unsigned long flags) | |
2005 | { | |
2006 | unsigned long (*get_area)(struct file *, | |
2007 | unsigned long, unsigned long, unsigned long, unsigned long); | |
2008 | unsigned long addr; | |
2009 | unsigned long offset; | |
2010 | unsigned long inflated_len; | |
2011 | unsigned long inflated_addr; | |
2012 | unsigned long inflated_offset; | |
2013 | ||
2014 | if (len > TASK_SIZE) | |
2015 | return -ENOMEM; | |
2016 | ||
2017 | get_area = current->mm->get_unmapped_area; | |
2018 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2019 | ||
e496cf3d | 2020 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
c01d5b30 HD |
2021 | return addr; |
2022 | if (IS_ERR_VALUE(addr)) | |
2023 | return addr; | |
2024 | if (addr & ~PAGE_MASK) | |
2025 | return addr; | |
2026 | if (addr > TASK_SIZE - len) | |
2027 | return addr; | |
2028 | ||
2029 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2030 | return addr; | |
2031 | if (len < HPAGE_PMD_SIZE) | |
2032 | return addr; | |
2033 | if (flags & MAP_FIXED) | |
2034 | return addr; | |
2035 | /* | |
2036 | * Our priority is to support MAP_SHARED mapped hugely; | |
2037 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
2038 | * But if caller specified an address hint, respect that as before. | |
2039 | */ | |
2040 | if (uaddr) | |
2041 | return addr; | |
2042 | ||
2043 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2044 | struct super_block *sb; | |
2045 | ||
2046 | if (file) { | |
2047 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2048 | sb = file_inode(file)->i_sb; | |
2049 | } else { | |
2050 | /* | |
2051 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2052 | * for "/dev/zero", to create a shared anonymous object. | |
2053 | */ | |
2054 | if (IS_ERR(shm_mnt)) | |
2055 | return addr; | |
2056 | sb = shm_mnt->mnt_sb; | |
2057 | } | |
3089bf61 | 2058 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2059 | return addr; |
2060 | } | |
2061 | ||
2062 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2063 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2064 | return addr; | |
2065 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2066 | return addr; | |
2067 | ||
2068 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2069 | if (inflated_len > TASK_SIZE) | |
2070 | return addr; | |
2071 | if (inflated_len < len) | |
2072 | return addr; | |
2073 | ||
2074 | inflated_addr = get_area(NULL, 0, inflated_len, 0, flags); | |
2075 | if (IS_ERR_VALUE(inflated_addr)) | |
2076 | return addr; | |
2077 | if (inflated_addr & ~PAGE_MASK) | |
2078 | return addr; | |
2079 | ||
2080 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2081 | inflated_addr += offset - inflated_offset; | |
2082 | if (inflated_offset > offset) | |
2083 | inflated_addr += HPAGE_PMD_SIZE; | |
2084 | ||
2085 | if (inflated_addr > TASK_SIZE - len) | |
2086 | return addr; | |
2087 | return inflated_addr; | |
2088 | } | |
2089 | ||
1da177e4 | 2090 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2091 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2092 | { |
496ad9aa | 2093 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2094 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2095 | } |
2096 | ||
d8dc74f2 AB |
2097 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
2098 | unsigned long addr) | |
1da177e4 | 2099 | { |
496ad9aa | 2100 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2101 | pgoff_t index; |
1da177e4 | 2102 | |
41ffe5d5 HD |
2103 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2104 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
2105 | } |
2106 | #endif | |
2107 | ||
2108 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
2109 | { | |
496ad9aa | 2110 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2111 | struct shmem_inode_info *info = SHMEM_I(inode); |
2112 | int retval = -ENOMEM; | |
2113 | ||
4595ef88 | 2114 | spin_lock_irq(&info->lock); |
1da177e4 LT |
2115 | if (lock && !(info->flags & VM_LOCKED)) { |
2116 | if (!user_shm_lock(inode->i_size, user)) | |
2117 | goto out_nomem; | |
2118 | info->flags |= VM_LOCKED; | |
89e004ea | 2119 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
2120 | } |
2121 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
2122 | user_shm_unlock(inode->i_size, user); | |
2123 | info->flags &= ~VM_LOCKED; | |
89e004ea | 2124 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2125 | } |
2126 | retval = 0; | |
89e004ea | 2127 | |
1da177e4 | 2128 | out_nomem: |
4595ef88 | 2129 | spin_unlock_irq(&info->lock); |
1da177e4 LT |
2130 | return retval; |
2131 | } | |
2132 | ||
9b83a6a8 | 2133 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
2134 | { |
2135 | file_accessed(file); | |
2136 | vma->vm_ops = &shmem_vm_ops; | |
e496cf3d | 2137 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
2138 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
2139 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
2140 | khugepaged_enter(vma, vma->vm_flags); | |
2141 | } | |
1da177e4 LT |
2142 | return 0; |
2143 | } | |
2144 | ||
454abafe | 2145 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 2146 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
2147 | { |
2148 | struct inode *inode; | |
2149 | struct shmem_inode_info *info; | |
2150 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
2151 | ||
5b04c689 PE |
2152 | if (shmem_reserve_inode(sb)) |
2153 | return NULL; | |
1da177e4 LT |
2154 | |
2155 | inode = new_inode(sb); | |
2156 | if (inode) { | |
85fe4025 | 2157 | inode->i_ino = get_next_ino(); |
454abafe | 2158 | inode_init_owner(inode, dir, mode); |
1da177e4 | 2159 | inode->i_blocks = 0; |
078cd827 | 2160 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
91828a40 | 2161 | inode->i_generation = get_seconds(); |
1da177e4 LT |
2162 | info = SHMEM_I(inode); |
2163 | memset(info, 0, (char *)inode - (char *)info); | |
2164 | spin_lock_init(&info->lock); | |
40e041a2 | 2165 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 2166 | info->flags = flags & VM_NORESERVE; |
779750d2 | 2167 | INIT_LIST_HEAD(&info->shrinklist); |
1da177e4 | 2168 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 2169 | simple_xattrs_init(&info->xattrs); |
72c04902 | 2170 | cache_no_acl(inode); |
1da177e4 LT |
2171 | |
2172 | switch (mode & S_IFMT) { | |
2173 | default: | |
39f0247d | 2174 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
2175 | init_special_inode(inode, mode, dev); |
2176 | break; | |
2177 | case S_IFREG: | |
14fcc23f | 2178 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
2179 | inode->i_op = &shmem_inode_operations; |
2180 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
2181 | mpol_shared_policy_init(&info->policy, |
2182 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
2183 | break; |
2184 | case S_IFDIR: | |
d8c76e6f | 2185 | inc_nlink(inode); |
1da177e4 LT |
2186 | /* Some things misbehave if size == 0 on a directory */ |
2187 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2188 | inode->i_op = &shmem_dir_inode_operations; | |
2189 | inode->i_fop = &simple_dir_operations; | |
2190 | break; | |
2191 | case S_IFLNK: | |
2192 | /* | |
2193 | * Must not load anything in the rbtree, | |
2194 | * mpol_free_shared_policy will not be called. | |
2195 | */ | |
71fe804b | 2196 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
2197 | break; |
2198 | } | |
5b04c689 PE |
2199 | } else |
2200 | shmem_free_inode(sb); | |
1da177e4 LT |
2201 | return inode; |
2202 | } | |
2203 | ||
0cd6144a JW |
2204 | bool shmem_mapping(struct address_space *mapping) |
2205 | { | |
f8005451 | 2206 | return mapping->a_ops == &shmem_aops; |
0cd6144a JW |
2207 | } |
2208 | ||
4c27fe4c MR |
2209 | int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm, |
2210 | pmd_t *dst_pmd, | |
2211 | struct vm_area_struct *dst_vma, | |
2212 | unsigned long dst_addr, | |
2213 | unsigned long src_addr, | |
2214 | struct page **pagep) | |
2215 | { | |
2216 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2217 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2218 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
2219 | struct address_space *mapping = inode->i_mapping; | |
2220 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2221 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
2222 | struct mem_cgroup *memcg; | |
2223 | spinlock_t *ptl; | |
2224 | void *page_kaddr; | |
2225 | struct page *page; | |
2226 | pte_t _dst_pte, *dst_pte; | |
2227 | int ret; | |
2228 | ||
cb658a45 AA |
2229 | ret = -ENOMEM; |
2230 | if (shmem_acct_block(info->flags, 1)) | |
2231 | goto out; | |
2232 | if (sbinfo->max_blocks) { | |
2233 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
2234 | sbinfo->max_blocks) >= 0) | |
2235 | goto out_unacct_blocks; | |
2236 | percpu_counter_inc(&sbinfo->used_blocks); | |
2237 | } | |
4c27fe4c | 2238 | |
cb658a45 | 2239 | if (!*pagep) { |
4c27fe4c MR |
2240 | page = shmem_alloc_page(gfp, info, pgoff); |
2241 | if (!page) | |
2242 | goto out_dec_used_blocks; | |
2243 | ||
2244 | page_kaddr = kmap_atomic(page); | |
2245 | ret = copy_from_user(page_kaddr, (const void __user *)src_addr, | |
2246 | PAGE_SIZE); | |
2247 | kunmap_atomic(page_kaddr); | |
2248 | ||
2249 | /* fallback to copy_from_user outside mmap_sem */ | |
2250 | if (unlikely(ret)) { | |
2251 | *pagep = page; | |
cb658a45 AA |
2252 | if (sbinfo->max_blocks) |
2253 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
2254 | shmem_unacct_blocks(info->flags, 1); | |
4c27fe4c MR |
2255 | /* don't free the page */ |
2256 | return -EFAULT; | |
2257 | } | |
2258 | } else { | |
2259 | page = *pagep; | |
2260 | *pagep = NULL; | |
2261 | } | |
2262 | ||
9cc90c66 AA |
2263 | VM_BUG_ON(PageLocked(page) || PageSwapBacked(page)); |
2264 | __SetPageLocked(page); | |
2265 | __SetPageSwapBacked(page); | |
a425d358 | 2266 | __SetPageUptodate(page); |
9cc90c66 | 2267 | |
4c27fe4c MR |
2268 | ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false); |
2269 | if (ret) | |
2270 | goto out_release; | |
2271 | ||
2272 | ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK); | |
2273 | if (!ret) { | |
2274 | ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL); | |
2275 | radix_tree_preload_end(); | |
2276 | } | |
2277 | if (ret) | |
2278 | goto out_release_uncharge; | |
2279 | ||
2280 | mem_cgroup_commit_charge(page, memcg, false, false); | |
2281 | ||
2282 | _dst_pte = mk_pte(page, dst_vma->vm_page_prot); | |
2283 | if (dst_vma->vm_flags & VM_WRITE) | |
2284 | _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte)); | |
2285 | ||
2286 | ret = -EEXIST; | |
2287 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); | |
2288 | if (!pte_none(*dst_pte)) | |
2289 | goto out_release_uncharge_unlock; | |
2290 | ||
4c27fe4c MR |
2291 | lru_cache_add_anon(page); |
2292 | ||
2293 | spin_lock(&info->lock); | |
2294 | info->alloced++; | |
2295 | inode->i_blocks += BLOCKS_PER_PAGE; | |
2296 | shmem_recalc_inode(inode); | |
2297 | spin_unlock(&info->lock); | |
2298 | ||
2299 | inc_mm_counter(dst_mm, mm_counter_file(page)); | |
2300 | page_add_file_rmap(page, false); | |
2301 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
2302 | ||
2303 | /* No need to invalidate - it was non-present before */ | |
2304 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
2305 | unlock_page(page); | |
2306 | pte_unmap_unlock(dst_pte, ptl); | |
2307 | ret = 0; | |
2308 | out: | |
2309 | return ret; | |
2310 | out_release_uncharge_unlock: | |
2311 | pte_unmap_unlock(dst_pte, ptl); | |
2312 | out_release_uncharge: | |
2313 | mem_cgroup_cancel_charge(page, memcg, false); | |
2314 | out_release: | |
9cc90c66 | 2315 | unlock_page(page); |
4c27fe4c MR |
2316 | put_page(page); |
2317 | out_dec_used_blocks: | |
2318 | if (sbinfo->max_blocks) | |
2319 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
2320 | out_unacct_blocks: | |
2321 | shmem_unacct_blocks(info->flags, 1); | |
2322 | goto out; | |
2323 | } | |
2324 | ||
1da177e4 | 2325 | #ifdef CONFIG_TMPFS |
92e1d5be | 2326 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2327 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2328 | |
6d9d88d0 JS |
2329 | #ifdef CONFIG_TMPFS_XATTR |
2330 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2331 | #else | |
2332 | #define shmem_initxattrs NULL | |
2333 | #endif | |
2334 | ||
1da177e4 | 2335 | static int |
800d15a5 NP |
2336 | shmem_write_begin(struct file *file, struct address_space *mapping, |
2337 | loff_t pos, unsigned len, unsigned flags, | |
2338 | struct page **pagep, void **fsdata) | |
1da177e4 | 2339 | { |
800d15a5 | 2340 | struct inode *inode = mapping->host; |
40e041a2 | 2341 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2342 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
2343 | |
2344 | /* i_mutex is held by caller */ | |
3f472cc9 | 2345 | if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) { |
40e041a2 DH |
2346 | if (info->seals & F_SEAL_WRITE) |
2347 | return -EPERM; | |
2348 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2349 | return -EPERM; | |
2350 | } | |
2351 | ||
9e18eb29 | 2352 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
2353 | } |
2354 | ||
2355 | static int | |
2356 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2357 | loff_t pos, unsigned len, unsigned copied, | |
2358 | struct page *page, void *fsdata) | |
2359 | { | |
2360 | struct inode *inode = mapping->host; | |
2361 | ||
d3602444 HD |
2362 | if (pos + copied > inode->i_size) |
2363 | i_size_write(inode, pos + copied); | |
2364 | ||
ec9516fb | 2365 | if (!PageUptodate(page)) { |
800d8c63 KS |
2366 | struct page *head = compound_head(page); |
2367 | if (PageTransCompound(page)) { | |
2368 | int i; | |
2369 | ||
2370 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2371 | if (head + i == page) | |
2372 | continue; | |
2373 | clear_highpage(head + i); | |
2374 | flush_dcache_page(head + i); | |
2375 | } | |
2376 | } | |
09cbfeaf KS |
2377 | if (copied < PAGE_SIZE) { |
2378 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2379 | zero_user_segments(page, 0, from, |
09cbfeaf | 2380 | from + copied, PAGE_SIZE); |
ec9516fb | 2381 | } |
800d8c63 | 2382 | SetPageUptodate(head); |
ec9516fb | 2383 | } |
800d15a5 | 2384 | set_page_dirty(page); |
6746aff7 | 2385 | unlock_page(page); |
09cbfeaf | 2386 | put_page(page); |
800d15a5 | 2387 | |
800d15a5 | 2388 | return copied; |
1da177e4 LT |
2389 | } |
2390 | ||
2ba5bbed | 2391 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2392 | { |
6e58e79d AV |
2393 | struct file *file = iocb->ki_filp; |
2394 | struct inode *inode = file_inode(file); | |
1da177e4 | 2395 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2396 | pgoff_t index; |
2397 | unsigned long offset; | |
a0ee5ec5 | 2398 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2399 | int error = 0; |
cb66a7a1 | 2400 | ssize_t retval = 0; |
6e58e79d | 2401 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2402 | |
2403 | /* | |
2404 | * Might this read be for a stacking filesystem? Then when reading | |
2405 | * holes of a sparse file, we actually need to allocate those pages, | |
2406 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2407 | */ | |
777eda2c | 2408 | if (!iter_is_iovec(to)) |
75edd345 | 2409 | sgp = SGP_CACHE; |
1da177e4 | 2410 | |
09cbfeaf KS |
2411 | index = *ppos >> PAGE_SHIFT; |
2412 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2413 | |
2414 | for (;;) { | |
2415 | struct page *page = NULL; | |
41ffe5d5 HD |
2416 | pgoff_t end_index; |
2417 | unsigned long nr, ret; | |
1da177e4 LT |
2418 | loff_t i_size = i_size_read(inode); |
2419 | ||
09cbfeaf | 2420 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2421 | if (index > end_index) |
2422 | break; | |
2423 | if (index == end_index) { | |
09cbfeaf | 2424 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2425 | if (nr <= offset) |
2426 | break; | |
2427 | } | |
2428 | ||
9e18eb29 | 2429 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2430 | if (error) { |
2431 | if (error == -EINVAL) | |
2432 | error = 0; | |
1da177e4 LT |
2433 | break; |
2434 | } | |
75edd345 HD |
2435 | if (page) { |
2436 | if (sgp == SGP_CACHE) | |
2437 | set_page_dirty(page); | |
d3602444 | 2438 | unlock_page(page); |
75edd345 | 2439 | } |
1da177e4 LT |
2440 | |
2441 | /* | |
2442 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2443 | * are called without i_mutex protection against truncate |
1da177e4 | 2444 | */ |
09cbfeaf | 2445 | nr = PAGE_SIZE; |
1da177e4 | 2446 | i_size = i_size_read(inode); |
09cbfeaf | 2447 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2448 | if (index == end_index) { |
09cbfeaf | 2449 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2450 | if (nr <= offset) { |
2451 | if (page) | |
09cbfeaf | 2452 | put_page(page); |
1da177e4 LT |
2453 | break; |
2454 | } | |
2455 | } | |
2456 | nr -= offset; | |
2457 | ||
2458 | if (page) { | |
2459 | /* | |
2460 | * If users can be writing to this page using arbitrary | |
2461 | * virtual addresses, take care about potential aliasing | |
2462 | * before reading the page on the kernel side. | |
2463 | */ | |
2464 | if (mapping_writably_mapped(mapping)) | |
2465 | flush_dcache_page(page); | |
2466 | /* | |
2467 | * Mark the page accessed if we read the beginning. | |
2468 | */ | |
2469 | if (!offset) | |
2470 | mark_page_accessed(page); | |
b5810039 | 2471 | } else { |
1da177e4 | 2472 | page = ZERO_PAGE(0); |
09cbfeaf | 2473 | get_page(page); |
b5810039 | 2474 | } |
1da177e4 LT |
2475 | |
2476 | /* | |
2477 | * Ok, we have the page, and it's up-to-date, so | |
2478 | * now we can copy it to user space... | |
1da177e4 | 2479 | */ |
2ba5bbed | 2480 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2481 | retval += ret; |
1da177e4 | 2482 | offset += ret; |
09cbfeaf KS |
2483 | index += offset >> PAGE_SHIFT; |
2484 | offset &= ~PAGE_MASK; | |
1da177e4 | 2485 | |
09cbfeaf | 2486 | put_page(page); |
2ba5bbed | 2487 | if (!iov_iter_count(to)) |
1da177e4 | 2488 | break; |
6e58e79d AV |
2489 | if (ret < nr) { |
2490 | error = -EFAULT; | |
2491 | break; | |
2492 | } | |
1da177e4 LT |
2493 | cond_resched(); |
2494 | } | |
2495 | ||
09cbfeaf | 2496 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2497 | file_accessed(file); |
2498 | return retval ? retval : error; | |
1da177e4 LT |
2499 | } |
2500 | ||
220f2ac9 HD |
2501 | /* |
2502 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
2503 | */ | |
2504 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2505 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2506 | { |
2507 | struct page *page; | |
2508 | struct pagevec pvec; | |
2509 | pgoff_t indices[PAGEVEC_SIZE]; | |
2510 | bool done = false; | |
2511 | int i; | |
2512 | ||
2513 | pagevec_init(&pvec, 0); | |
2514 | pvec.nr = 1; /* start small: we may be there already */ | |
2515 | while (!done) { | |
0cd6144a | 2516 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2517 | pvec.nr, pvec.pages, indices); |
2518 | if (!pvec.nr) { | |
965c8e59 | 2519 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2520 | index = end; |
2521 | break; | |
2522 | } | |
2523 | for (i = 0; i < pvec.nr; i++, index++) { | |
2524 | if (index < indices[i]) { | |
965c8e59 | 2525 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2526 | done = true; |
2527 | break; | |
2528 | } | |
2529 | index = indices[i]; | |
2530 | } | |
2531 | page = pvec.pages[i]; | |
2532 | if (page && !radix_tree_exceptional_entry(page)) { | |
2533 | if (!PageUptodate(page)) | |
2534 | page = NULL; | |
2535 | } | |
2536 | if (index >= end || | |
965c8e59 AM |
2537 | (page && whence == SEEK_DATA) || |
2538 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2539 | done = true; |
2540 | break; | |
2541 | } | |
2542 | } | |
0cd6144a | 2543 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2544 | pagevec_release(&pvec); |
2545 | pvec.nr = PAGEVEC_SIZE; | |
2546 | cond_resched(); | |
2547 | } | |
2548 | return index; | |
2549 | } | |
2550 | ||
965c8e59 | 2551 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2552 | { |
2553 | struct address_space *mapping = file->f_mapping; | |
2554 | struct inode *inode = mapping->host; | |
2555 | pgoff_t start, end; | |
2556 | loff_t new_offset; | |
2557 | ||
965c8e59 AM |
2558 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2559 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2560 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2561 | inode_lock(inode); |
220f2ac9 HD |
2562 | /* We're holding i_mutex so we can access i_size directly */ |
2563 | ||
2564 | if (offset < 0) | |
2565 | offset = -EINVAL; | |
2566 | else if (offset >= inode->i_size) | |
2567 | offset = -ENXIO; | |
2568 | else { | |
09cbfeaf KS |
2569 | start = offset >> PAGE_SHIFT; |
2570 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2571 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2572 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2573 | if (new_offset > offset) { |
2574 | if (new_offset < inode->i_size) | |
2575 | offset = new_offset; | |
965c8e59 | 2576 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2577 | offset = -ENXIO; |
2578 | else | |
2579 | offset = inode->i_size; | |
2580 | } | |
2581 | } | |
2582 | ||
387aae6f HD |
2583 | if (offset >= 0) |
2584 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2585 | inode_unlock(inode); |
220f2ac9 HD |
2586 | return offset; |
2587 | } | |
2588 | ||
05f65b5c DH |
2589 | /* |
2590 | * We need a tag: a new tag would expand every radix_tree_node by 8 bytes, | |
2591 | * so reuse a tag which we firmly believe is never set or cleared on shmem. | |
2592 | */ | |
2593 | #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE | |
2594 | #define LAST_SCAN 4 /* about 150ms max */ | |
2595 | ||
2596 | static void shmem_tag_pins(struct address_space *mapping) | |
2597 | { | |
2598 | struct radix_tree_iter iter; | |
2599 | void **slot; | |
2600 | pgoff_t start; | |
2601 | struct page *page; | |
2602 | ||
2603 | lru_add_drain(); | |
2604 | start = 0; | |
2605 | rcu_read_lock(); | |
2606 | ||
05f65b5c DH |
2607 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
2608 | page = radix_tree_deref_slot(slot); | |
2609 | if (!page || radix_tree_exception(page)) { | |
2cf938aa MW |
2610 | if (radix_tree_deref_retry(page)) { |
2611 | slot = radix_tree_iter_retry(&iter); | |
2612 | continue; | |
2613 | } | |
05f65b5c DH |
2614 | } else if (page_count(page) - page_mapcount(page) > 1) { |
2615 | spin_lock_irq(&mapping->tree_lock); | |
2616 | radix_tree_tag_set(&mapping->page_tree, iter.index, | |
2617 | SHMEM_TAG_PINNED); | |
2618 | spin_unlock_irq(&mapping->tree_lock); | |
2619 | } | |
2620 | ||
2621 | if (need_resched()) { | |
148deab2 | 2622 | slot = radix_tree_iter_resume(slot, &iter); |
05f65b5c | 2623 | cond_resched_rcu(); |
05f65b5c DH |
2624 | } |
2625 | } | |
2626 | rcu_read_unlock(); | |
2627 | } | |
2628 | ||
2629 | /* | |
2630 | * Setting SEAL_WRITE requires us to verify there's no pending writer. However, | |
2631 | * via get_user_pages(), drivers might have some pending I/O without any active | |
2632 | * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages | |
2633 | * and see whether it has an elevated ref-count. If so, we tag them and wait for | |
2634 | * them to be dropped. | |
2635 | * The caller must guarantee that no new user will acquire writable references | |
2636 | * to those pages to avoid races. | |
2637 | */ | |
40e041a2 DH |
2638 | static int shmem_wait_for_pins(struct address_space *mapping) |
2639 | { | |
05f65b5c DH |
2640 | struct radix_tree_iter iter; |
2641 | void **slot; | |
2642 | pgoff_t start; | |
2643 | struct page *page; | |
2644 | int error, scan; | |
2645 | ||
2646 | shmem_tag_pins(mapping); | |
2647 | ||
2648 | error = 0; | |
2649 | for (scan = 0; scan <= LAST_SCAN; scan++) { | |
2650 | if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED)) | |
2651 | break; | |
2652 | ||
2653 | if (!scan) | |
2654 | lru_add_drain_all(); | |
2655 | else if (schedule_timeout_killable((HZ << scan) / 200)) | |
2656 | scan = LAST_SCAN; | |
2657 | ||
2658 | start = 0; | |
2659 | rcu_read_lock(); | |
05f65b5c DH |
2660 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, |
2661 | start, SHMEM_TAG_PINNED) { | |
2662 | ||
2663 | page = radix_tree_deref_slot(slot); | |
2664 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2665 | if (radix_tree_deref_retry(page)) { |
2666 | slot = radix_tree_iter_retry(&iter); | |
2667 | continue; | |
2668 | } | |
05f65b5c DH |
2669 | |
2670 | page = NULL; | |
2671 | } | |
2672 | ||
2673 | if (page && | |
2674 | page_count(page) - page_mapcount(page) != 1) { | |
2675 | if (scan < LAST_SCAN) | |
2676 | goto continue_resched; | |
2677 | ||
2678 | /* | |
2679 | * On the last scan, we clean up all those tags | |
2680 | * we inserted; but make a note that we still | |
2681 | * found pages pinned. | |
2682 | */ | |
2683 | error = -EBUSY; | |
2684 | } | |
2685 | ||
2686 | spin_lock_irq(&mapping->tree_lock); | |
2687 | radix_tree_tag_clear(&mapping->page_tree, | |
2688 | iter.index, SHMEM_TAG_PINNED); | |
2689 | spin_unlock_irq(&mapping->tree_lock); | |
2690 | continue_resched: | |
2691 | if (need_resched()) { | |
148deab2 | 2692 | slot = radix_tree_iter_resume(slot, &iter); |
05f65b5c | 2693 | cond_resched_rcu(); |
05f65b5c DH |
2694 | } |
2695 | } | |
2696 | rcu_read_unlock(); | |
2697 | } | |
2698 | ||
2699 | return error; | |
40e041a2 DH |
2700 | } |
2701 | ||
2702 | #define F_ALL_SEALS (F_SEAL_SEAL | \ | |
2703 | F_SEAL_SHRINK | \ | |
2704 | F_SEAL_GROW | \ | |
2705 | F_SEAL_WRITE) | |
2706 | ||
2707 | int shmem_add_seals(struct file *file, unsigned int seals) | |
2708 | { | |
2709 | struct inode *inode = file_inode(file); | |
2710 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2711 | int error; | |
2712 | ||
2713 | /* | |
2714 | * SEALING | |
2715 | * Sealing allows multiple parties to share a shmem-file but restrict | |
2716 | * access to a specific subset of file operations. Seals can only be | |
2717 | * added, but never removed. This way, mutually untrusted parties can | |
2718 | * share common memory regions with a well-defined policy. A malicious | |
2719 | * peer can thus never perform unwanted operations on a shared object. | |
2720 | * | |
2721 | * Seals are only supported on special shmem-files and always affect | |
2722 | * the whole underlying inode. Once a seal is set, it may prevent some | |
2723 | * kinds of access to the file. Currently, the following seals are | |
2724 | * defined: | |
2725 | * SEAL_SEAL: Prevent further seals from being set on this file | |
2726 | * SEAL_SHRINK: Prevent the file from shrinking | |
2727 | * SEAL_GROW: Prevent the file from growing | |
2728 | * SEAL_WRITE: Prevent write access to the file | |
2729 | * | |
2730 | * As we don't require any trust relationship between two parties, we | |
2731 | * must prevent seals from being removed. Therefore, sealing a file | |
2732 | * only adds a given set of seals to the file, it never touches | |
2733 | * existing seals. Furthermore, the "setting seals"-operation can be | |
2734 | * sealed itself, which basically prevents any further seal from being | |
2735 | * added. | |
2736 | * | |
2737 | * Semantics of sealing are only defined on volatile files. Only | |
2738 | * anonymous shmem files support sealing. More importantly, seals are | |
2739 | * never written to disk. Therefore, there's no plan to support it on | |
2740 | * other file types. | |
2741 | */ | |
2742 | ||
2743 | if (file->f_op != &shmem_file_operations) | |
2744 | return -EINVAL; | |
2745 | if (!(file->f_mode & FMODE_WRITE)) | |
2746 | return -EPERM; | |
2747 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
2748 | return -EINVAL; | |
2749 | ||
5955102c | 2750 | inode_lock(inode); |
40e041a2 DH |
2751 | |
2752 | if (info->seals & F_SEAL_SEAL) { | |
2753 | error = -EPERM; | |
2754 | goto unlock; | |
2755 | } | |
2756 | ||
2757 | if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) { | |
2758 | error = mapping_deny_writable(file->f_mapping); | |
2759 | if (error) | |
2760 | goto unlock; | |
2761 | ||
2762 | error = shmem_wait_for_pins(file->f_mapping); | |
2763 | if (error) { | |
2764 | mapping_allow_writable(file->f_mapping); | |
2765 | goto unlock; | |
2766 | } | |
2767 | } | |
2768 | ||
2769 | info->seals |= seals; | |
2770 | error = 0; | |
2771 | ||
2772 | unlock: | |
5955102c | 2773 | inode_unlock(inode); |
40e041a2 DH |
2774 | return error; |
2775 | } | |
2776 | EXPORT_SYMBOL_GPL(shmem_add_seals); | |
2777 | ||
2778 | int shmem_get_seals(struct file *file) | |
2779 | { | |
2780 | if (file->f_op != &shmem_file_operations) | |
2781 | return -EINVAL; | |
2782 | ||
2783 | return SHMEM_I(file_inode(file))->seals; | |
2784 | } | |
2785 | EXPORT_SYMBOL_GPL(shmem_get_seals); | |
2786 | ||
2787 | long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg) | |
2788 | { | |
2789 | long error; | |
2790 | ||
2791 | switch (cmd) { | |
2792 | case F_ADD_SEALS: | |
2793 | /* disallow upper 32bit */ | |
2794 | if (arg > UINT_MAX) | |
2795 | return -EINVAL; | |
2796 | ||
2797 | error = shmem_add_seals(file, arg); | |
2798 | break; | |
2799 | case F_GET_SEALS: | |
2800 | error = shmem_get_seals(file); | |
2801 | break; | |
2802 | default: | |
2803 | error = -EINVAL; | |
2804 | break; | |
2805 | } | |
2806 | ||
2807 | return error; | |
2808 | } | |
2809 | ||
83e4fa9c HD |
2810 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2811 | loff_t len) | |
2812 | { | |
496ad9aa | 2813 | struct inode *inode = file_inode(file); |
e2d12e22 | 2814 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2815 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2816 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2817 | pgoff_t start, index, end; |
2818 | int error; | |
83e4fa9c | 2819 | |
13ace4d0 HD |
2820 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2821 | return -EOPNOTSUPP; | |
2822 | ||
5955102c | 2823 | inode_lock(inode); |
83e4fa9c HD |
2824 | |
2825 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2826 | struct address_space *mapping = file->f_mapping; | |
2827 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2828 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2829 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2830 | |
40e041a2 DH |
2831 | /* protected by i_mutex */ |
2832 | if (info->seals & F_SEAL_WRITE) { | |
2833 | error = -EPERM; | |
2834 | goto out; | |
2835 | } | |
2836 | ||
8e205f77 | 2837 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
2838 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
2839 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
2840 | spin_lock(&inode->i_lock); | |
2841 | inode->i_private = &shmem_falloc; | |
2842 | spin_unlock(&inode->i_lock); | |
2843 | ||
83e4fa9c HD |
2844 | if ((u64)unmap_end > (u64)unmap_start) |
2845 | unmap_mapping_range(mapping, unmap_start, | |
2846 | 1 + unmap_end - unmap_start, 0); | |
2847 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2848 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2849 | |
2850 | spin_lock(&inode->i_lock); | |
2851 | inode->i_private = NULL; | |
2852 | wake_up_all(&shmem_falloc_waitq); | |
2055da97 | 2853 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head)); |
8e205f77 | 2854 | spin_unlock(&inode->i_lock); |
83e4fa9c | 2855 | error = 0; |
8e205f77 | 2856 | goto out; |
e2d12e22 HD |
2857 | } |
2858 | ||
2859 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2860 | error = inode_newsize_ok(inode, offset + len); | |
2861 | if (error) | |
2862 | goto out; | |
2863 | ||
40e041a2 DH |
2864 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2865 | error = -EPERM; | |
2866 | goto out; | |
2867 | } | |
2868 | ||
09cbfeaf KS |
2869 | start = offset >> PAGE_SHIFT; |
2870 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2871 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2872 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2873 | error = -ENOSPC; | |
2874 | goto out; | |
83e4fa9c HD |
2875 | } |
2876 | ||
8e205f77 | 2877 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2878 | shmem_falloc.start = start; |
2879 | shmem_falloc.next = start; | |
2880 | shmem_falloc.nr_falloced = 0; | |
2881 | shmem_falloc.nr_unswapped = 0; | |
2882 | spin_lock(&inode->i_lock); | |
2883 | inode->i_private = &shmem_falloc; | |
2884 | spin_unlock(&inode->i_lock); | |
2885 | ||
e2d12e22 HD |
2886 | for (index = start; index < end; index++) { |
2887 | struct page *page; | |
2888 | ||
2889 | /* | |
2890 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2891 | * been interrupted because we are using up too much memory. | |
2892 | */ | |
2893 | if (signal_pending(current)) | |
2894 | error = -EINTR; | |
1aac1400 HD |
2895 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2896 | error = -ENOMEM; | |
e2d12e22 | 2897 | else |
9e18eb29 | 2898 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2899 | if (error) { |
1635f6a7 | 2900 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2901 | if (index > start) { |
2902 | shmem_undo_range(inode, | |
2903 | (loff_t)start << PAGE_SHIFT, | |
2904 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2905 | } | |
1aac1400 | 2906 | goto undone; |
e2d12e22 HD |
2907 | } |
2908 | ||
1aac1400 HD |
2909 | /* |
2910 | * Inform shmem_writepage() how far we have reached. | |
2911 | * No need for lock or barrier: we have the page lock. | |
2912 | */ | |
2913 | shmem_falloc.next++; | |
2914 | if (!PageUptodate(page)) | |
2915 | shmem_falloc.nr_falloced++; | |
2916 | ||
e2d12e22 | 2917 | /* |
1635f6a7 HD |
2918 | * If !PageUptodate, leave it that way so that freeable pages |
2919 | * can be recognized if we need to rollback on error later. | |
2920 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2921 | * than free the pages we are allocating (and SGP_CACHE pages |
2922 | * might still be clean: we now need to mark those dirty too). | |
2923 | */ | |
2924 | set_page_dirty(page); | |
2925 | unlock_page(page); | |
09cbfeaf | 2926 | put_page(page); |
e2d12e22 HD |
2927 | cond_resched(); |
2928 | } | |
2929 | ||
2930 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2931 | i_size_write(inode, offset + len); | |
078cd827 | 2932 | inode->i_ctime = current_time(inode); |
1aac1400 HD |
2933 | undone: |
2934 | spin_lock(&inode->i_lock); | |
2935 | inode->i_private = NULL; | |
2936 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2937 | out: |
5955102c | 2938 | inode_unlock(inode); |
83e4fa9c HD |
2939 | return error; |
2940 | } | |
2941 | ||
726c3342 | 2942 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2943 | { |
726c3342 | 2944 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2945 | |
2946 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2947 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2948 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2949 | if (sbinfo->max_blocks) { |
1da177e4 | 2950 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2951 | buf->f_bavail = |
2952 | buf->f_bfree = sbinfo->max_blocks - | |
2953 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2954 | } |
2955 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2956 | buf->f_files = sbinfo->max_inodes; |
2957 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2958 | } |
2959 | /* else leave those fields 0 like simple_statfs */ | |
2960 | return 0; | |
2961 | } | |
2962 | ||
2963 | /* | |
2964 | * File creation. Allocate an inode, and we're done.. | |
2965 | */ | |
2966 | static int | |
1a67aafb | 2967 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2968 | { |
0b0a0806 | 2969 | struct inode *inode; |
1da177e4 LT |
2970 | int error = -ENOSPC; |
2971 | ||
454abafe | 2972 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2973 | if (inode) { |
feda821e CH |
2974 | error = simple_acl_create(dir, inode); |
2975 | if (error) | |
2976 | goto out_iput; | |
2a7dba39 | 2977 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2978 | &dentry->d_name, |
6d9d88d0 | 2979 | shmem_initxattrs, NULL); |
feda821e CH |
2980 | if (error && error != -EOPNOTSUPP) |
2981 | goto out_iput; | |
37ec43cd | 2982 | |
718deb6b | 2983 | error = 0; |
1da177e4 | 2984 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 2985 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
2986 | d_instantiate(dentry, inode); |
2987 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2988 | } |
2989 | return error; | |
feda821e CH |
2990 | out_iput: |
2991 | iput(inode); | |
2992 | return error; | |
1da177e4 LT |
2993 | } |
2994 | ||
60545d0d AV |
2995 | static int |
2996 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2997 | { | |
2998 | struct inode *inode; | |
2999 | int error = -ENOSPC; | |
3000 | ||
3001 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
3002 | if (inode) { | |
3003 | error = security_inode_init_security(inode, dir, | |
3004 | NULL, | |
3005 | shmem_initxattrs, NULL); | |
feda821e CH |
3006 | if (error && error != -EOPNOTSUPP) |
3007 | goto out_iput; | |
3008 | error = simple_acl_create(dir, inode); | |
3009 | if (error) | |
3010 | goto out_iput; | |
60545d0d AV |
3011 | d_tmpfile(dentry, inode); |
3012 | } | |
3013 | return error; | |
feda821e CH |
3014 | out_iput: |
3015 | iput(inode); | |
3016 | return error; | |
60545d0d AV |
3017 | } |
3018 | ||
18bb1db3 | 3019 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
3020 | { |
3021 | int error; | |
3022 | ||
3023 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
3024 | return error; | |
d8c76e6f | 3025 | inc_nlink(dir); |
1da177e4 LT |
3026 | return 0; |
3027 | } | |
3028 | ||
4acdaf27 | 3029 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 3030 | bool excl) |
1da177e4 LT |
3031 | { |
3032 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
3033 | } | |
3034 | ||
3035 | /* | |
3036 | * Link a file.. | |
3037 | */ | |
3038 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
3039 | { | |
75c3cfa8 | 3040 | struct inode *inode = d_inode(old_dentry); |
5b04c689 | 3041 | int ret; |
1da177e4 LT |
3042 | |
3043 | /* | |
3044 | * No ordinary (disk based) filesystem counts links as inodes; | |
3045 | * but each new link needs a new dentry, pinning lowmem, and | |
3046 | * tmpfs dentries cannot be pruned until they are unlinked. | |
3047 | */ | |
5b04c689 PE |
3048 | ret = shmem_reserve_inode(inode->i_sb); |
3049 | if (ret) | |
3050 | goto out; | |
1da177e4 LT |
3051 | |
3052 | dir->i_size += BOGO_DIRENT_SIZE; | |
078cd827 | 3053 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
d8c76e6f | 3054 | inc_nlink(inode); |
7de9c6ee | 3055 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
3056 | dget(dentry); /* Extra pinning count for the created dentry */ |
3057 | d_instantiate(dentry, inode); | |
5b04c689 PE |
3058 | out: |
3059 | return ret; | |
1da177e4 LT |
3060 | } |
3061 | ||
3062 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
3063 | { | |
75c3cfa8 | 3064 | struct inode *inode = d_inode(dentry); |
1da177e4 | 3065 | |
5b04c689 PE |
3066 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
3067 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
3068 | |
3069 | dir->i_size -= BOGO_DIRENT_SIZE; | |
078cd827 | 3070 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
9a53c3a7 | 3071 | drop_nlink(inode); |
1da177e4 LT |
3072 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
3073 | return 0; | |
3074 | } | |
3075 | ||
3076 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
3077 | { | |
3078 | if (!simple_empty(dentry)) | |
3079 | return -ENOTEMPTY; | |
3080 | ||
75c3cfa8 | 3081 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 3082 | drop_nlink(dir); |
1da177e4 LT |
3083 | return shmem_unlink(dir, dentry); |
3084 | } | |
3085 | ||
37456771 MS |
3086 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
3087 | { | |
e36cb0b8 DH |
3088 | bool old_is_dir = d_is_dir(old_dentry); |
3089 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
3090 | |
3091 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
3092 | if (old_is_dir) { | |
3093 | drop_nlink(old_dir); | |
3094 | inc_nlink(new_dir); | |
3095 | } else { | |
3096 | drop_nlink(new_dir); | |
3097 | inc_nlink(old_dir); | |
3098 | } | |
3099 | } | |
3100 | old_dir->i_ctime = old_dir->i_mtime = | |
3101 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 | 3102 | d_inode(old_dentry)->i_ctime = |
078cd827 | 3103 | d_inode(new_dentry)->i_ctime = current_time(old_dir); |
37456771 MS |
3104 | |
3105 | return 0; | |
3106 | } | |
3107 | ||
46fdb794 MS |
3108 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
3109 | { | |
3110 | struct dentry *whiteout; | |
3111 | int error; | |
3112 | ||
3113 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
3114 | if (!whiteout) | |
3115 | return -ENOMEM; | |
3116 | ||
3117 | error = shmem_mknod(old_dir, whiteout, | |
3118 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
3119 | dput(whiteout); | |
3120 | if (error) | |
3121 | return error; | |
3122 | ||
3123 | /* | |
3124 | * Cheat and hash the whiteout while the old dentry is still in | |
3125 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
3126 | * | |
3127 | * d_lookup() will consistently find one of them at this point, | |
3128 | * not sure which one, but that isn't even important. | |
3129 | */ | |
3130 | d_rehash(whiteout); | |
3131 | return 0; | |
3132 | } | |
3133 | ||
1da177e4 LT |
3134 | /* |
3135 | * The VFS layer already does all the dentry stuff for rename, | |
3136 | * we just have to decrement the usage count for the target if | |
3137 | * it exists so that the VFS layer correctly free's it when it | |
3138 | * gets overwritten. | |
3139 | */ | |
3b69ff51 | 3140 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 3141 | { |
75c3cfa8 | 3142 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
3143 | int they_are_dirs = S_ISDIR(inode->i_mode); |
3144 | ||
46fdb794 | 3145 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
3146 | return -EINVAL; |
3147 | ||
37456771 MS |
3148 | if (flags & RENAME_EXCHANGE) |
3149 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
3150 | ||
1da177e4 LT |
3151 | if (!simple_empty(new_dentry)) |
3152 | return -ENOTEMPTY; | |
3153 | ||
46fdb794 MS |
3154 | if (flags & RENAME_WHITEOUT) { |
3155 | int error; | |
3156 | ||
3157 | error = shmem_whiteout(old_dir, old_dentry); | |
3158 | if (error) | |
3159 | return error; | |
3160 | } | |
3161 | ||
75c3cfa8 | 3162 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 3163 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 3164 | if (they_are_dirs) { |
75c3cfa8 | 3165 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 3166 | drop_nlink(old_dir); |
b928095b | 3167 | } |
1da177e4 | 3168 | } else if (they_are_dirs) { |
9a53c3a7 | 3169 | drop_nlink(old_dir); |
d8c76e6f | 3170 | inc_nlink(new_dir); |
1da177e4 LT |
3171 | } |
3172 | ||
3173 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
3174 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
3175 | old_dir->i_ctime = old_dir->i_mtime = | |
3176 | new_dir->i_ctime = new_dir->i_mtime = | |
078cd827 | 3177 | inode->i_ctime = current_time(old_dir); |
1da177e4 LT |
3178 | return 0; |
3179 | } | |
3180 | ||
3181 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
3182 | { | |
3183 | int error; | |
3184 | int len; | |
3185 | struct inode *inode; | |
9276aad6 | 3186 | struct page *page; |
1da177e4 LT |
3187 | struct shmem_inode_info *info; |
3188 | ||
3189 | len = strlen(symname) + 1; | |
09cbfeaf | 3190 | if (len > PAGE_SIZE) |
1da177e4 LT |
3191 | return -ENAMETOOLONG; |
3192 | ||
454abafe | 3193 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
3194 | if (!inode) |
3195 | return -ENOSPC; | |
3196 | ||
9d8f13ba | 3197 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3198 | shmem_initxattrs, NULL); |
570bc1c2 SS |
3199 | if (error) { |
3200 | if (error != -EOPNOTSUPP) { | |
3201 | iput(inode); | |
3202 | return error; | |
3203 | } | |
3204 | error = 0; | |
3205 | } | |
3206 | ||
1da177e4 LT |
3207 | info = SHMEM_I(inode); |
3208 | inode->i_size = len-1; | |
69f07ec9 | 3209 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3210 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3211 | if (!inode->i_link) { | |
69f07ec9 HD |
3212 | iput(inode); |
3213 | return -ENOMEM; | |
3214 | } | |
3215 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3216 | } else { |
e8ecde25 | 3217 | inode_nohighmem(inode); |
9e18eb29 | 3218 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
3219 | if (error) { |
3220 | iput(inode); | |
3221 | return error; | |
3222 | } | |
14fcc23f | 3223 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 3224 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 3225 | memcpy(page_address(page), symname, len); |
ec9516fb | 3226 | SetPageUptodate(page); |
1da177e4 | 3227 | set_page_dirty(page); |
6746aff7 | 3228 | unlock_page(page); |
09cbfeaf | 3229 | put_page(page); |
1da177e4 | 3230 | } |
1da177e4 | 3231 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3232 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3233 | d_instantiate(dentry, inode); |
3234 | dget(dentry); | |
3235 | return 0; | |
3236 | } | |
3237 | ||
fceef393 | 3238 | static void shmem_put_link(void *arg) |
1da177e4 | 3239 | { |
fceef393 AV |
3240 | mark_page_accessed(arg); |
3241 | put_page(arg); | |
1da177e4 LT |
3242 | } |
3243 | ||
6b255391 | 3244 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
3245 | struct inode *inode, |
3246 | struct delayed_call *done) | |
1da177e4 | 3247 | { |
1da177e4 | 3248 | struct page *page = NULL; |
6b255391 | 3249 | int error; |
6a6c9904 AV |
3250 | if (!dentry) { |
3251 | page = find_get_page(inode->i_mapping, 0); | |
3252 | if (!page) | |
3253 | return ERR_PTR(-ECHILD); | |
3254 | if (!PageUptodate(page)) { | |
3255 | put_page(page); | |
3256 | return ERR_PTR(-ECHILD); | |
3257 | } | |
3258 | } else { | |
9e18eb29 | 3259 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3260 | if (error) |
3261 | return ERR_PTR(error); | |
3262 | unlock_page(page); | |
3263 | } | |
fceef393 | 3264 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3265 | return page_address(page); |
1da177e4 LT |
3266 | } |
3267 | ||
b09e0fa4 | 3268 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3269 | /* |
b09e0fa4 EP |
3270 | * Superblocks without xattr inode operations may get some security.* xattr |
3271 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3272 | * like ACLs, we also need to implement the security.* handlers at |
3273 | * filesystem level, though. | |
3274 | */ | |
3275 | ||
6d9d88d0 JS |
3276 | /* |
3277 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3278 | */ | |
3279 | static int shmem_initxattrs(struct inode *inode, | |
3280 | const struct xattr *xattr_array, | |
3281 | void *fs_info) | |
3282 | { | |
3283 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3284 | const struct xattr *xattr; | |
38f38657 | 3285 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3286 | size_t len; |
3287 | ||
3288 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3289 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3290 | if (!new_xattr) |
3291 | return -ENOMEM; | |
3292 | ||
3293 | len = strlen(xattr->name) + 1; | |
3294 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3295 | GFP_KERNEL); | |
3296 | if (!new_xattr->name) { | |
3297 | kfree(new_xattr); | |
3298 | return -ENOMEM; | |
3299 | } | |
3300 | ||
3301 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3302 | XATTR_SECURITY_PREFIX_LEN); | |
3303 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3304 | xattr->name, len); | |
3305 | ||
38f38657 | 3306 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3307 | } |
3308 | ||
3309 | return 0; | |
3310 | } | |
3311 | ||
aa7c5241 | 3312 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3313 | struct dentry *unused, struct inode *inode, |
3314 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3315 | { |
b296821a | 3316 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3317 | |
aa7c5241 | 3318 | name = xattr_full_name(handler, name); |
38f38657 | 3319 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3320 | } |
3321 | ||
aa7c5241 | 3322 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
59301226 AV |
3323 | struct dentry *unused, struct inode *inode, |
3324 | const char *name, const void *value, | |
3325 | size_t size, int flags) | |
b09e0fa4 | 3326 | { |
59301226 | 3327 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3328 | |
aa7c5241 | 3329 | name = xattr_full_name(handler, name); |
38f38657 | 3330 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
3331 | } |
3332 | ||
aa7c5241 AG |
3333 | static const struct xattr_handler shmem_security_xattr_handler = { |
3334 | .prefix = XATTR_SECURITY_PREFIX, | |
3335 | .get = shmem_xattr_handler_get, | |
3336 | .set = shmem_xattr_handler_set, | |
3337 | }; | |
b09e0fa4 | 3338 | |
aa7c5241 AG |
3339 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3340 | .prefix = XATTR_TRUSTED_PREFIX, | |
3341 | .get = shmem_xattr_handler_get, | |
3342 | .set = shmem_xattr_handler_set, | |
3343 | }; | |
b09e0fa4 | 3344 | |
aa7c5241 AG |
3345 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3346 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3347 | &posix_acl_access_xattr_handler, | |
3348 | &posix_acl_default_xattr_handler, | |
3349 | #endif | |
3350 | &shmem_security_xattr_handler, | |
3351 | &shmem_trusted_xattr_handler, | |
3352 | NULL | |
3353 | }; | |
b09e0fa4 EP |
3354 | |
3355 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3356 | { | |
75c3cfa8 | 3357 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3358 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3359 | } |
3360 | #endif /* CONFIG_TMPFS_XATTR */ | |
3361 | ||
69f07ec9 | 3362 | static const struct inode_operations shmem_short_symlink_operations = { |
6b255391 | 3363 | .get_link = simple_get_link, |
b09e0fa4 | 3364 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3365 | .listxattr = shmem_listxattr, |
b09e0fa4 EP |
3366 | #endif |
3367 | }; | |
3368 | ||
3369 | static const struct inode_operations shmem_symlink_inode_operations = { | |
6b255391 | 3370 | .get_link = shmem_get_link, |
b09e0fa4 | 3371 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3372 | .listxattr = shmem_listxattr, |
39f0247d | 3373 | #endif |
b09e0fa4 | 3374 | }; |
39f0247d | 3375 | |
91828a40 DG |
3376 | static struct dentry *shmem_get_parent(struct dentry *child) |
3377 | { | |
3378 | return ERR_PTR(-ESTALE); | |
3379 | } | |
3380 | ||
3381 | static int shmem_match(struct inode *ino, void *vfh) | |
3382 | { | |
3383 | __u32 *fh = vfh; | |
3384 | __u64 inum = fh[2]; | |
3385 | inum = (inum << 32) | fh[1]; | |
3386 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
3387 | } | |
3388 | ||
480b116c CH |
3389 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3390 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3391 | { |
91828a40 | 3392 | struct inode *inode; |
480b116c | 3393 | struct dentry *dentry = NULL; |
35c2a7f4 | 3394 | u64 inum; |
480b116c CH |
3395 | |
3396 | if (fh_len < 3) | |
3397 | return NULL; | |
91828a40 | 3398 | |
35c2a7f4 HD |
3399 | inum = fid->raw[2]; |
3400 | inum = (inum << 32) | fid->raw[1]; | |
3401 | ||
480b116c CH |
3402 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
3403 | shmem_match, fid->raw); | |
91828a40 | 3404 | if (inode) { |
480b116c | 3405 | dentry = d_find_alias(inode); |
91828a40 DG |
3406 | iput(inode); |
3407 | } | |
3408 | ||
480b116c | 3409 | return dentry; |
91828a40 DG |
3410 | } |
3411 | ||
b0b0382b AV |
3412 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3413 | struct inode *parent) | |
91828a40 | 3414 | { |
5fe0c237 AK |
3415 | if (*len < 3) { |
3416 | *len = 3; | |
94e07a75 | 3417 | return FILEID_INVALID; |
5fe0c237 | 3418 | } |
91828a40 | 3419 | |
1d3382cb | 3420 | if (inode_unhashed(inode)) { |
91828a40 DG |
3421 | /* Unfortunately insert_inode_hash is not idempotent, |
3422 | * so as we hash inodes here rather than at creation | |
3423 | * time, we need a lock to ensure we only try | |
3424 | * to do it once | |
3425 | */ | |
3426 | static DEFINE_SPINLOCK(lock); | |
3427 | spin_lock(&lock); | |
1d3382cb | 3428 | if (inode_unhashed(inode)) |
91828a40 DG |
3429 | __insert_inode_hash(inode, |
3430 | inode->i_ino + inode->i_generation); | |
3431 | spin_unlock(&lock); | |
3432 | } | |
3433 | ||
3434 | fh[0] = inode->i_generation; | |
3435 | fh[1] = inode->i_ino; | |
3436 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
3437 | ||
3438 | *len = 3; | |
3439 | return 1; | |
3440 | } | |
3441 | ||
39655164 | 3442 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3443 | .get_parent = shmem_get_parent, |
91828a40 | 3444 | .encode_fh = shmem_encode_fh, |
480b116c | 3445 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3446 | }; |
3447 | ||
680d794b AM |
3448 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
3449 | bool remount) | |
1da177e4 LT |
3450 | { |
3451 | char *this_char, *value, *rest; | |
49cd0a5c | 3452 | struct mempolicy *mpol = NULL; |
8751e039 EB |
3453 | uid_t uid; |
3454 | gid_t gid; | |
1da177e4 | 3455 | |
b00dc3ad HD |
3456 | while (options != NULL) { |
3457 | this_char = options; | |
3458 | for (;;) { | |
3459 | /* | |
3460 | * NUL-terminate this option: unfortunately, | |
3461 | * mount options form a comma-separated list, | |
3462 | * but mpol's nodelist may also contain commas. | |
3463 | */ | |
3464 | options = strchr(options, ','); | |
3465 | if (options == NULL) | |
3466 | break; | |
3467 | options++; | |
3468 | if (!isdigit(*options)) { | |
3469 | options[-1] = '\0'; | |
3470 | break; | |
3471 | } | |
3472 | } | |
1da177e4 LT |
3473 | if (!*this_char) |
3474 | continue; | |
3475 | if ((value = strchr(this_char,'=')) != NULL) { | |
3476 | *value++ = 0; | |
3477 | } else { | |
1170532b JP |
3478 | pr_err("tmpfs: No value for mount option '%s'\n", |
3479 | this_char); | |
49cd0a5c | 3480 | goto error; |
1da177e4 LT |
3481 | } |
3482 | ||
3483 | if (!strcmp(this_char,"size")) { | |
3484 | unsigned long long size; | |
3485 | size = memparse(value,&rest); | |
3486 | if (*rest == '%') { | |
3487 | size <<= PAGE_SHIFT; | |
3488 | size *= totalram_pages; | |
3489 | do_div(size, 100); | |
3490 | rest++; | |
3491 | } | |
3492 | if (*rest) | |
3493 | goto bad_val; | |
680d794b | 3494 | sbinfo->max_blocks = |
09cbfeaf | 3495 | DIV_ROUND_UP(size, PAGE_SIZE); |
1da177e4 | 3496 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 3497 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
3498 | if (*rest) |
3499 | goto bad_val; | |
3500 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 3501 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
3502 | if (*rest) |
3503 | goto bad_val; | |
3504 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 3505 | if (remount) |
1da177e4 | 3506 | continue; |
680d794b | 3507 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
3508 | if (*rest) |
3509 | goto bad_val; | |
3510 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 3511 | if (remount) |
1da177e4 | 3512 | continue; |
8751e039 | 3513 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3514 | if (*rest) |
3515 | goto bad_val; | |
8751e039 EB |
3516 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
3517 | if (!uid_valid(sbinfo->uid)) | |
3518 | goto bad_val; | |
1da177e4 | 3519 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 3520 | if (remount) |
1da177e4 | 3521 | continue; |
8751e039 | 3522 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3523 | if (*rest) |
3524 | goto bad_val; | |
8751e039 EB |
3525 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
3526 | if (!gid_valid(sbinfo->gid)) | |
3527 | goto bad_val; | |
e496cf3d | 3528 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3529 | } else if (!strcmp(this_char, "huge")) { |
3530 | int huge; | |
3531 | huge = shmem_parse_huge(value); | |
3532 | if (huge < 0) | |
3533 | goto bad_val; | |
3534 | if (!has_transparent_hugepage() && | |
3535 | huge != SHMEM_HUGE_NEVER) | |
3536 | goto bad_val; | |
3537 | sbinfo->huge = huge; | |
3538 | #endif | |
3539 | #ifdef CONFIG_NUMA | |
7339ff83 | 3540 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
3541 | mpol_put(mpol); |
3542 | mpol = NULL; | |
3543 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 3544 | goto bad_val; |
5a6e75f8 | 3545 | #endif |
1da177e4 | 3546 | } else { |
1170532b | 3547 | pr_err("tmpfs: Bad mount option %s\n", this_char); |
49cd0a5c | 3548 | goto error; |
1da177e4 LT |
3549 | } |
3550 | } | |
49cd0a5c | 3551 | sbinfo->mpol = mpol; |
1da177e4 LT |
3552 | return 0; |
3553 | ||
3554 | bad_val: | |
1170532b | 3555 | pr_err("tmpfs: Bad value '%s' for mount option '%s'\n", |
1da177e4 | 3556 | value, this_char); |
49cd0a5c GT |
3557 | error: |
3558 | mpol_put(mpol); | |
1da177e4 LT |
3559 | return 1; |
3560 | ||
3561 | } | |
3562 | ||
3563 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
3564 | { | |
3565 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 3566 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
3567 | unsigned long inodes; |
3568 | int error = -EINVAL; | |
3569 | ||
5f00110f | 3570 | config.mpol = NULL; |
680d794b | 3571 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 3572 | return error; |
1da177e4 | 3573 | |
0edd73b3 | 3574 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3575 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 3576 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 3577 | goto out; |
680d794b | 3578 | if (config.max_inodes < inodes) |
0edd73b3 HD |
3579 | goto out; |
3580 | /* | |
54af6042 | 3581 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
3582 | * but we must separately disallow unlimited->limited, because |
3583 | * in that case we have no record of how much is already in use. | |
3584 | */ | |
680d794b | 3585 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 3586 | goto out; |
680d794b | 3587 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
3588 | goto out; |
3589 | ||
3590 | error = 0; | |
5a6e75f8 | 3591 | sbinfo->huge = config.huge; |
680d794b | 3592 | sbinfo->max_blocks = config.max_blocks; |
680d794b AM |
3593 | sbinfo->max_inodes = config.max_inodes; |
3594 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 3595 | |
5f00110f GT |
3596 | /* |
3597 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3598 | */ | |
3599 | if (config.mpol) { | |
3600 | mpol_put(sbinfo->mpol); | |
3601 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
3602 | } | |
0edd73b3 HD |
3603 | out: |
3604 | spin_unlock(&sbinfo->stat_lock); | |
3605 | return error; | |
1da177e4 | 3606 | } |
680d794b | 3607 | |
34c80b1d | 3608 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3609 | { |
34c80b1d | 3610 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b AM |
3611 | |
3612 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3613 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3614 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b AM |
3615 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
3616 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
3617 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 3618 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3619 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3620 | seq_printf(seq, ",uid=%u", | |
3621 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3622 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3623 | seq_printf(seq, ",gid=%u", | |
3624 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
e496cf3d | 3625 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3626 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
3627 | if (sbinfo->huge) | |
3628 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3629 | #endif | |
71fe804b | 3630 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b AM |
3631 | return 0; |
3632 | } | |
9183df25 DH |
3633 | |
3634 | #define MFD_NAME_PREFIX "memfd:" | |
3635 | #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1) | |
3636 | #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN) | |
3637 | ||
3638 | #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING) | |
3639 | ||
3640 | SYSCALL_DEFINE2(memfd_create, | |
3641 | const char __user *, uname, | |
3642 | unsigned int, flags) | |
3643 | { | |
3644 | struct shmem_inode_info *info; | |
3645 | struct file *file; | |
3646 | int fd, error; | |
3647 | char *name; | |
3648 | long len; | |
3649 | ||
3650 | if (flags & ~(unsigned int)MFD_ALL_FLAGS) | |
3651 | return -EINVAL; | |
3652 | ||
3653 | /* length includes terminating zero */ | |
3654 | len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1); | |
3655 | if (len <= 0) | |
3656 | return -EFAULT; | |
3657 | if (len > MFD_NAME_MAX_LEN + 1) | |
3658 | return -EINVAL; | |
3659 | ||
3660 | name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY); | |
3661 | if (!name) | |
3662 | return -ENOMEM; | |
3663 | ||
3664 | strcpy(name, MFD_NAME_PREFIX); | |
3665 | if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) { | |
3666 | error = -EFAULT; | |
3667 | goto err_name; | |
3668 | } | |
3669 | ||
3670 | /* terminating-zero may have changed after strnlen_user() returned */ | |
3671 | if (name[len + MFD_NAME_PREFIX_LEN - 1]) { | |
3672 | error = -EFAULT; | |
3673 | goto err_name; | |
3674 | } | |
3675 | ||
3676 | fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0); | |
3677 | if (fd < 0) { | |
3678 | error = fd; | |
3679 | goto err_name; | |
3680 | } | |
3681 | ||
3682 | file = shmem_file_setup(name, 0, VM_NORESERVE); | |
3683 | if (IS_ERR(file)) { | |
3684 | error = PTR_ERR(file); | |
3685 | goto err_fd; | |
3686 | } | |
3687 | info = SHMEM_I(file_inode(file)); | |
3688 | file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; | |
3689 | file->f_flags |= O_RDWR | O_LARGEFILE; | |
3690 | if (flags & MFD_ALLOW_SEALING) | |
3691 | info->seals &= ~F_SEAL_SEAL; | |
3692 | ||
3693 | fd_install(fd, file); | |
3694 | kfree(name); | |
3695 | return fd; | |
3696 | ||
3697 | err_fd: | |
3698 | put_unused_fd(fd); | |
3699 | err_name: | |
3700 | kfree(name); | |
3701 | return error; | |
3702 | } | |
3703 | ||
680d794b | 3704 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3705 | |
3706 | static void shmem_put_super(struct super_block *sb) | |
3707 | { | |
602586a8 HD |
3708 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3709 | ||
3710 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 3711 | mpol_put(sbinfo->mpol); |
602586a8 | 3712 | kfree(sbinfo); |
1da177e4 LT |
3713 | sb->s_fs_info = NULL; |
3714 | } | |
3715 | ||
2b2af54a | 3716 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3717 | { |
3718 | struct inode *inode; | |
0edd73b3 | 3719 | struct shmem_sb_info *sbinfo; |
680d794b AM |
3720 | int err = -ENOMEM; |
3721 | ||
3722 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3723 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b AM |
3724 | L1_CACHE_BYTES), GFP_KERNEL); |
3725 | if (!sbinfo) | |
3726 | return -ENOMEM; | |
3727 | ||
680d794b | 3728 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
3729 | sbinfo->uid = current_fsuid(); |
3730 | sbinfo->gid = current_fsgid(); | |
680d794b | 3731 | sb->s_fs_info = sbinfo; |
1da177e4 | 3732 | |
0edd73b3 | 3733 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3734 | /* |
3735 | * Per default we only allow half of the physical ram per | |
3736 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3737 | * but the internal instance is left unlimited. | |
3738 | */ | |
ca4e0519 | 3739 | if (!(sb->s_flags & MS_KERNMOUNT)) { |
680d794b AM |
3740 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3741 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3742 | if (shmem_parse_options(data, sbinfo, false)) { | |
3743 | err = -EINVAL; | |
3744 | goto failed; | |
3745 | } | |
ca4e0519 AV |
3746 | } else { |
3747 | sb->s_flags |= MS_NOUSER; | |
1da177e4 | 3748 | } |
91828a40 | 3749 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 3750 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
3751 | #else |
3752 | sb->s_flags |= MS_NOUSER; | |
3753 | #endif | |
3754 | ||
0edd73b3 | 3755 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3756 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3757 | goto failed; |
680d794b | 3758 | sbinfo->free_inodes = sbinfo->max_inodes; |
779750d2 KS |
3759 | spin_lock_init(&sbinfo->shrinklist_lock); |
3760 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 3761 | |
285b2c4f | 3762 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3763 | sb->s_blocksize = PAGE_SIZE; |
3764 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3765 | sb->s_magic = TMPFS_MAGIC; |
3766 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3767 | sb->s_time_gran = 1; |
b09e0fa4 | 3768 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3769 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3770 | #endif |
3771 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
3772 | sb->s_flags |= MS_POSIXACL; |
3773 | #endif | |
2b4db796 | 3774 | uuid_gen(&sb->s_uuid); |
0edd73b3 | 3775 | |
454abafe | 3776 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3777 | if (!inode) |
3778 | goto failed; | |
680d794b AM |
3779 | inode->i_uid = sbinfo->uid; |
3780 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3781 | sb->s_root = d_make_root(inode); |
3782 | if (!sb->s_root) | |
48fde701 | 3783 | goto failed; |
1da177e4 LT |
3784 | return 0; |
3785 | ||
1da177e4 LT |
3786 | failed: |
3787 | shmem_put_super(sb); | |
3788 | return err; | |
3789 | } | |
3790 | ||
fcc234f8 | 3791 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3792 | |
3793 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3794 | { | |
41ffe5d5 HD |
3795 | struct shmem_inode_info *info; |
3796 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3797 | if (!info) | |
1da177e4 | 3798 | return NULL; |
41ffe5d5 | 3799 | return &info->vfs_inode; |
1da177e4 LT |
3800 | } |
3801 | ||
41ffe5d5 | 3802 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3803 | { |
3804 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84e710da AV |
3805 | if (S_ISLNK(inode->i_mode)) |
3806 | kfree(inode->i_link); | |
fa0d7e3d NP |
3807 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3808 | } | |
3809 | ||
1da177e4 LT |
3810 | static void shmem_destroy_inode(struct inode *inode) |
3811 | { | |
09208d15 | 3812 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3813 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3814 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3815 | } |
3816 | ||
41ffe5d5 | 3817 | static void shmem_init_inode(void *foo) |
1da177e4 | 3818 | { |
41ffe5d5 HD |
3819 | struct shmem_inode_info *info = foo; |
3820 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3821 | } |
3822 | ||
41ffe5d5 | 3823 | static int shmem_init_inodecache(void) |
1da177e4 LT |
3824 | { |
3825 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3826 | sizeof(struct shmem_inode_info), | |
5d097056 | 3827 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3828 | return 0; |
3829 | } | |
3830 | ||
41ffe5d5 | 3831 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3832 | { |
1a1d92c1 | 3833 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3834 | } |
3835 | ||
f5e54d6e | 3836 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3837 | .writepage = shmem_writepage, |
76719325 | 3838 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3839 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3840 | .write_begin = shmem_write_begin, |
3841 | .write_end = shmem_write_end, | |
1da177e4 | 3842 | #endif |
1c93923c | 3843 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3844 | .migratepage = migrate_page, |
1c93923c | 3845 | #endif |
aa261f54 | 3846 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3847 | }; |
3848 | ||
15ad7cdc | 3849 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3850 | .mmap = shmem_mmap, |
c01d5b30 | 3851 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3852 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3853 | .llseek = shmem_file_llseek, |
2ba5bbed | 3854 | .read_iter = shmem_file_read_iter, |
8174202b | 3855 | .write_iter = generic_file_write_iter, |
1b061d92 | 3856 | .fsync = noop_fsync, |
82c156f8 | 3857 | .splice_read = generic_file_splice_read, |
f6cb85d0 | 3858 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3859 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3860 | #endif |
3861 | }; | |
3862 | ||
92e1d5be | 3863 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3864 | .getattr = shmem_getattr, |
94c1e62d | 3865 | .setattr = shmem_setattr, |
b09e0fa4 | 3866 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3867 | .listxattr = shmem_listxattr, |
feda821e | 3868 | .set_acl = simple_set_acl, |
b09e0fa4 | 3869 | #endif |
1da177e4 LT |
3870 | }; |
3871 | ||
92e1d5be | 3872 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3873 | #ifdef CONFIG_TMPFS |
3874 | .create = shmem_create, | |
3875 | .lookup = simple_lookup, | |
3876 | .link = shmem_link, | |
3877 | .unlink = shmem_unlink, | |
3878 | .symlink = shmem_symlink, | |
3879 | .mkdir = shmem_mkdir, | |
3880 | .rmdir = shmem_rmdir, | |
3881 | .mknod = shmem_mknod, | |
2773bf00 | 3882 | .rename = shmem_rename2, |
60545d0d | 3883 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3884 | #endif |
b09e0fa4 | 3885 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3886 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3887 | #endif |
39f0247d | 3888 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3889 | .setattr = shmem_setattr, |
feda821e | 3890 | .set_acl = simple_set_acl, |
39f0247d AG |
3891 | #endif |
3892 | }; | |
3893 | ||
92e1d5be | 3894 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3895 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3896 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3897 | #endif |
39f0247d | 3898 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3899 | .setattr = shmem_setattr, |
feda821e | 3900 | .set_acl = simple_set_acl, |
39f0247d | 3901 | #endif |
1da177e4 LT |
3902 | }; |
3903 | ||
759b9775 | 3904 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
3905 | .alloc_inode = shmem_alloc_inode, |
3906 | .destroy_inode = shmem_destroy_inode, | |
3907 | #ifdef CONFIG_TMPFS | |
3908 | .statfs = shmem_statfs, | |
3909 | .remount_fs = shmem_remount_fs, | |
680d794b | 3910 | .show_options = shmem_show_options, |
1da177e4 | 3911 | #endif |
1f895f75 | 3912 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3913 | .drop_inode = generic_delete_inode, |
3914 | .put_super = shmem_put_super, | |
779750d2 KS |
3915 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
3916 | .nr_cached_objects = shmem_unused_huge_count, | |
3917 | .free_cached_objects = shmem_unused_huge_scan, | |
3918 | #endif | |
1da177e4 LT |
3919 | }; |
3920 | ||
f0f37e2f | 3921 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3922 | .fault = shmem_fault, |
d7c17551 | 3923 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3924 | #ifdef CONFIG_NUMA |
3925 | .set_policy = shmem_set_policy, | |
3926 | .get_policy = shmem_get_policy, | |
3927 | #endif | |
3928 | }; | |
3929 | ||
3c26ff6e AV |
3930 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
3931 | int flags, const char *dev_name, void *data) | |
1da177e4 | 3932 | { |
3c26ff6e | 3933 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
3934 | } |
3935 | ||
41ffe5d5 | 3936 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3937 | .owner = THIS_MODULE, |
3938 | .name = "tmpfs", | |
3c26ff6e | 3939 | .mount = shmem_mount, |
1da177e4 | 3940 | .kill_sb = kill_litter_super, |
2b8576cb | 3941 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 3942 | }; |
1da177e4 | 3943 | |
41ffe5d5 | 3944 | int __init shmem_init(void) |
1da177e4 LT |
3945 | { |
3946 | int error; | |
3947 | ||
16203a7a RL |
3948 | /* If rootfs called this, don't re-init */ |
3949 | if (shmem_inode_cachep) | |
3950 | return 0; | |
3951 | ||
41ffe5d5 | 3952 | error = shmem_init_inodecache(); |
1da177e4 LT |
3953 | if (error) |
3954 | goto out3; | |
3955 | ||
41ffe5d5 | 3956 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 3957 | if (error) { |
1170532b | 3958 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
3959 | goto out2; |
3960 | } | |
95dc112a | 3961 | |
ca4e0519 | 3962 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
3963 | if (IS_ERR(shm_mnt)) { |
3964 | error = PTR_ERR(shm_mnt); | |
1170532b | 3965 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
3966 | goto out1; |
3967 | } | |
5a6e75f8 | 3968 | |
e496cf3d | 3969 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
435c0b87 | 3970 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
3971 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
3972 | else | |
3973 | shmem_huge = 0; /* just in case it was patched */ | |
3974 | #endif | |
1da177e4 LT |
3975 | return 0; |
3976 | ||
3977 | out1: | |
41ffe5d5 | 3978 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 3979 | out2: |
41ffe5d5 | 3980 | shmem_destroy_inodecache(); |
1da177e4 LT |
3981 | out3: |
3982 | shm_mnt = ERR_PTR(error); | |
3983 | return error; | |
3984 | } | |
853ac43a | 3985 | |
e496cf3d | 3986 | #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS) |
5a6e75f8 KS |
3987 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
3988 | struct kobj_attribute *attr, char *buf) | |
3989 | { | |
3990 | int values[] = { | |
3991 | SHMEM_HUGE_ALWAYS, | |
3992 | SHMEM_HUGE_WITHIN_SIZE, | |
3993 | SHMEM_HUGE_ADVISE, | |
3994 | SHMEM_HUGE_NEVER, | |
3995 | SHMEM_HUGE_DENY, | |
3996 | SHMEM_HUGE_FORCE, | |
3997 | }; | |
3998 | int i, count; | |
3999 | ||
4000 | for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) { | |
4001 | const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s "; | |
4002 | ||
4003 | count += sprintf(buf + count, fmt, | |
4004 | shmem_format_huge(values[i])); | |
4005 | } | |
4006 | buf[count - 1] = '\n'; | |
4007 | return count; | |
4008 | } | |
4009 | ||
4010 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
4011 | struct kobj_attribute *attr, const char *buf, size_t count) | |
4012 | { | |
4013 | char tmp[16]; | |
4014 | int huge; | |
4015 | ||
4016 | if (count + 1 > sizeof(tmp)) | |
4017 | return -EINVAL; | |
4018 | memcpy(tmp, buf, count); | |
4019 | tmp[count] = '\0'; | |
4020 | if (count && tmp[count - 1] == '\n') | |
4021 | tmp[count - 1] = '\0'; | |
4022 | ||
4023 | huge = shmem_parse_huge(tmp); | |
4024 | if (huge == -EINVAL) | |
4025 | return -EINVAL; | |
4026 | if (!has_transparent_hugepage() && | |
4027 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
4028 | return -EINVAL; | |
4029 | ||
4030 | shmem_huge = huge; | |
435c0b87 | 4031 | if (shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4032 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4033 | return count; | |
4034 | } | |
4035 | ||
4036 | struct kobj_attribute shmem_enabled_attr = | |
4037 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
3b33719c | 4038 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */ |
f3f0e1d2 | 4039 | |
3b33719c | 4040 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
f3f0e1d2 KS |
4041 | bool shmem_huge_enabled(struct vm_area_struct *vma) |
4042 | { | |
4043 | struct inode *inode = file_inode(vma->vm_file); | |
4044 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4045 | loff_t i_size; | |
4046 | pgoff_t off; | |
4047 | ||
4048 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
4049 | return true; | |
4050 | if (shmem_huge == SHMEM_HUGE_DENY) | |
4051 | return false; | |
4052 | switch (sbinfo->huge) { | |
4053 | case SHMEM_HUGE_NEVER: | |
4054 | return false; | |
4055 | case SHMEM_HUGE_ALWAYS: | |
4056 | return true; | |
4057 | case SHMEM_HUGE_WITHIN_SIZE: | |
4058 | off = round_up(vma->vm_pgoff, HPAGE_PMD_NR); | |
4059 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
4060 | if (i_size >= HPAGE_PMD_SIZE && | |
4061 | i_size >> PAGE_SHIFT >= off) | |
4062 | return true; | |
4063 | case SHMEM_HUGE_ADVISE: | |
4064 | /* TODO: implement fadvise() hints */ | |
4065 | return (vma->vm_flags & VM_HUGEPAGE); | |
4066 | default: | |
4067 | VM_BUG_ON(1); | |
4068 | return false; | |
4069 | } | |
4070 | } | |
3b33719c | 4071 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 4072 | |
853ac43a MM |
4073 | #else /* !CONFIG_SHMEM */ |
4074 | ||
4075 | /* | |
4076 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
4077 | * | |
4078 | * This is intended for small system where the benefits of the full | |
4079 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4080 | * their complexity. On systems without swap this code should be | |
4081 | * effectively equivalent, but much lighter weight. | |
4082 | */ | |
4083 | ||
41ffe5d5 | 4084 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4085 | .name = "tmpfs", |
3c26ff6e | 4086 | .mount = ramfs_mount, |
853ac43a | 4087 | .kill_sb = kill_litter_super, |
2b8576cb | 4088 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4089 | }; |
4090 | ||
41ffe5d5 | 4091 | int __init shmem_init(void) |
853ac43a | 4092 | { |
41ffe5d5 | 4093 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4094 | |
41ffe5d5 | 4095 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
4096 | BUG_ON(IS_ERR(shm_mnt)); |
4097 | ||
4098 | return 0; | |
4099 | } | |
4100 | ||
41ffe5d5 | 4101 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
4102 | { |
4103 | return 0; | |
4104 | } | |
4105 | ||
3f96b79a HD |
4106 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
4107 | { | |
4108 | return 0; | |
4109 | } | |
4110 | ||
24513264 HD |
4111 | void shmem_unlock_mapping(struct address_space *mapping) |
4112 | { | |
4113 | } | |
4114 | ||
c01d5b30 HD |
4115 | #ifdef CONFIG_MMU |
4116 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4117 | unsigned long addr, unsigned long len, | |
4118 | unsigned long pgoff, unsigned long flags) | |
4119 | { | |
4120 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4121 | } | |
4122 | #endif | |
4123 | ||
41ffe5d5 | 4124 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4125 | { |
41ffe5d5 | 4126 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4127 | } |
4128 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4129 | ||
0b0a0806 HD |
4130 | #define shmem_vm_ops generic_file_vm_ops |
4131 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 4132 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
4133 | #define shmem_acct_size(flags, size) 0 |
4134 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
4135 | |
4136 | #endif /* CONFIG_SHMEM */ | |
4137 | ||
4138 | /* common code */ | |
1da177e4 | 4139 | |
19938e35 | 4140 | static const struct dentry_operations anon_ops = { |
118b2302 | 4141 | .d_dname = simple_dname |
3451538a AV |
4142 | }; |
4143 | ||
c7277090 EP |
4144 | static struct file *__shmem_file_setup(const char *name, loff_t size, |
4145 | unsigned long flags, unsigned int i_flags) | |
1da177e4 | 4146 | { |
6b4d0b27 | 4147 | struct file *res; |
1da177e4 | 4148 | struct inode *inode; |
2c48b9c4 | 4149 | struct path path; |
3451538a | 4150 | struct super_block *sb; |
1da177e4 LT |
4151 | struct qstr this; |
4152 | ||
4153 | if (IS_ERR(shm_mnt)) | |
6b4d0b27 | 4154 | return ERR_CAST(shm_mnt); |
1da177e4 | 4155 | |
285b2c4f | 4156 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4157 | return ERR_PTR(-EINVAL); |
4158 | ||
4159 | if (shmem_acct_size(flags, size)) | |
4160 | return ERR_PTR(-ENOMEM); | |
4161 | ||
6b4d0b27 | 4162 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
4163 | this.name = name; |
4164 | this.len = strlen(name); | |
4165 | this.hash = 0; /* will go */ | |
3451538a | 4166 | sb = shm_mnt->mnt_sb; |
66ee4b88 | 4167 | path.mnt = mntget(shm_mnt); |
3451538a | 4168 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 4169 | if (!path.dentry) |
1da177e4 | 4170 | goto put_memory; |
3451538a | 4171 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 4172 | |
6b4d0b27 | 4173 | res = ERR_PTR(-ENOSPC); |
3451538a | 4174 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 4175 | if (!inode) |
66ee4b88 | 4176 | goto put_memory; |
1da177e4 | 4177 | |
c7277090 | 4178 | inode->i_flags |= i_flags; |
2c48b9c4 | 4179 | d_instantiate(path.dentry, inode); |
1da177e4 | 4180 | inode->i_size = size; |
6d6b77f1 | 4181 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
4182 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
4183 | if (IS_ERR(res)) | |
66ee4b88 | 4184 | goto put_path; |
4b42af81 | 4185 | |
6b4d0b27 | 4186 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 4187 | &shmem_file_operations); |
6b4d0b27 | 4188 | if (IS_ERR(res)) |
66ee4b88 | 4189 | goto put_path; |
4b42af81 | 4190 | |
6b4d0b27 | 4191 | return res; |
1da177e4 | 4192 | |
1da177e4 LT |
4193 | put_memory: |
4194 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
4195 | put_path: |
4196 | path_put(&path); | |
6b4d0b27 | 4197 | return res; |
1da177e4 | 4198 | } |
c7277090 EP |
4199 | |
4200 | /** | |
4201 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4202 | * kernel internal. There will be NO LSM permission checks against the | |
4203 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4204 | * higher layer. The users are the big_key and shm implementations. LSM |
4205 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4206 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4207 | * @size: size to be set for the file | |
4208 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4209 | */ | |
4210 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4211 | { | |
4212 | return __shmem_file_setup(name, size, flags, S_PRIVATE); | |
4213 | } | |
4214 | ||
4215 | /** | |
4216 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4217 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4218 | * @size: size to be set for the file | |
4219 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4220 | */ | |
4221 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4222 | { | |
4223 | return __shmem_file_setup(name, size, flags, 0); | |
4224 | } | |
395e0ddc | 4225 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4226 | |
46711810 | 4227 | /** |
1da177e4 | 4228 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
4229 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
4230 | */ | |
4231 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4232 | { | |
4233 | struct file *file; | |
4234 | loff_t size = vma->vm_end - vma->vm_start; | |
4235 | ||
66fc1303 HD |
4236 | /* |
4237 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
4238 | * between XFS directory reading and selinux: since this file is only | |
4239 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4240 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4241 | */ | |
4242 | file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE); | |
1da177e4 LT |
4243 | if (IS_ERR(file)) |
4244 | return PTR_ERR(file); | |
4245 | ||
4246 | if (vma->vm_file) | |
4247 | fput(vma->vm_file); | |
4248 | vma->vm_file = file; | |
4249 | vma->vm_ops = &shmem_vm_ops; | |
f3f0e1d2 | 4250 | |
e496cf3d | 4251 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
4252 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
4253 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
4254 | khugepaged_enter(vma, vma->vm_flags); | |
4255 | } | |
4256 | ||
1da177e4 LT |
4257 | return 0; |
4258 | } | |
d9d90e5e HD |
4259 | |
4260 | /** | |
4261 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
4262 | * @mapping: the page's address_space | |
4263 | * @index: the page index | |
4264 | * @gfp: the page allocator flags to use if allocating | |
4265 | * | |
4266 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4267 | * with any new page allocations done using the specified allocation flags. | |
4268 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
4269 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
4270 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4271 | * | |
68da9f05 HD |
4272 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4273 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4274 | */ |
4275 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4276 | pgoff_t index, gfp_t gfp) | |
4277 | { | |
68da9f05 HD |
4278 | #ifdef CONFIG_SHMEM |
4279 | struct inode *inode = mapping->host; | |
9276aad6 | 4280 | struct page *page; |
68da9f05 HD |
4281 | int error; |
4282 | ||
4283 | BUG_ON(mapping->a_ops != &shmem_aops); | |
9e18eb29 | 4284 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
cfda0526 | 4285 | gfp, NULL, NULL, NULL); |
68da9f05 HD |
4286 | if (error) |
4287 | page = ERR_PTR(error); | |
4288 | else | |
4289 | unlock_page(page); | |
4290 | return page; | |
4291 | #else | |
4292 | /* | |
4293 | * The tiny !SHMEM case uses ramfs without swap | |
4294 | */ | |
d9d90e5e | 4295 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4296 | #endif |
d9d90e5e HD |
4297 | } |
4298 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |