]>
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> | |
46c9a946 | 32 | #include <linux/random.h> |
174cd4b1 | 33 | #include <linux/sched/signal.h> |
b95f1b31 | 34 | #include <linux/export.h> |
853ac43a | 35 | #include <linux/swap.h> |
e2e40f2c | 36 | #include <linux/uio.h> |
f3f0e1d2 | 37 | #include <linux/khugepaged.h> |
749df87b | 38 | #include <linux/hugetlb.h> |
b56a2d8a | 39 | #include <linux/frontswap.h> |
626c3920 | 40 | #include <linux/fs_parser.h> |
853ac43a | 41 | |
95cc09d6 AA |
42 | #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */ |
43 | ||
853ac43a MM |
44 | static struct vfsmount *shm_mnt; |
45 | ||
46 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
47 | /* |
48 | * This virtual memory filesystem is heavily based on the ramfs. It | |
49 | * extends ramfs by the ability to use swap and honor resource limits | |
50 | * which makes it a completely usable filesystem. | |
51 | */ | |
52 | ||
39f0247d | 53 | #include <linux/xattr.h> |
a5694255 | 54 | #include <linux/exportfs.h> |
1c7c474c | 55 | #include <linux/posix_acl.h> |
feda821e | 56 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 57 | #include <linux/mman.h> |
1da177e4 LT |
58 | #include <linux/string.h> |
59 | #include <linux/slab.h> | |
60 | #include <linux/backing-dev.h> | |
61 | #include <linux/shmem_fs.h> | |
1da177e4 | 62 | #include <linux/writeback.h> |
1da177e4 | 63 | #include <linux/blkdev.h> |
bda97eab | 64 | #include <linux/pagevec.h> |
41ffe5d5 | 65 | #include <linux/percpu_counter.h> |
83e4fa9c | 66 | #include <linux/falloc.h> |
708e3508 | 67 | #include <linux/splice.h> |
1da177e4 LT |
68 | #include <linux/security.h> |
69 | #include <linux/swapops.h> | |
70 | #include <linux/mempolicy.h> | |
71 | #include <linux/namei.h> | |
b00dc3ad | 72 | #include <linux/ctype.h> |
304dbdb7 | 73 | #include <linux/migrate.h> |
c1f60a5a | 74 | #include <linux/highmem.h> |
680d794b | 75 | #include <linux/seq_file.h> |
92562927 | 76 | #include <linux/magic.h> |
9183df25 | 77 | #include <linux/syscalls.h> |
40e041a2 | 78 | #include <linux/fcntl.h> |
9183df25 | 79 | #include <uapi/linux/memfd.h> |
cfda0526 | 80 | #include <linux/userfaultfd_k.h> |
4c27fe4c | 81 | #include <linux/rmap.h> |
2b4db796 | 82 | #include <linux/uuid.h> |
304dbdb7 | 83 | |
7c0f6ba6 | 84 | #include <linux/uaccess.h> |
1da177e4 LT |
85 | #include <asm/pgtable.h> |
86 | ||
dd56b046 MG |
87 | #include "internal.h" |
88 | ||
09cbfeaf KS |
89 | #define BLOCKS_PER_PAGE (PAGE_SIZE/512) |
90 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT) | |
1da177e4 | 91 | |
1da177e4 LT |
92 | /* Pretend that each entry is of this size in directory's i_size */ |
93 | #define BOGO_DIRENT_SIZE 20 | |
94 | ||
69f07ec9 HD |
95 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
96 | #define SHORT_SYMLINK_LEN 128 | |
97 | ||
1aac1400 | 98 | /* |
f00cdc6d HD |
99 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
100 | * inode->i_private (with i_mutex making sure that it has only one user at | |
101 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
102 | */ |
103 | struct shmem_falloc { | |
8e205f77 | 104 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
105 | pgoff_t start; /* start of range currently being fallocated */ |
106 | pgoff_t next; /* the next page offset to be fallocated */ | |
107 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
108 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
109 | }; | |
110 | ||
0b5071dd AV |
111 | struct shmem_options { |
112 | unsigned long long blocks; | |
e4a3f096 | 113 | int inodes; |
0b5071dd AV |
114 | struct mempolicy *mpol; |
115 | kuid_t uid; | |
116 | kgid_t gid; | |
117 | umode_t mode; | |
118 | int huge; | |
119 | int seen; | |
120 | #define SHMEM_SEEN_BLOCKS 1 | |
121 | #define SHMEM_SEEN_INODES 2 | |
122 | #define SHMEM_SEEN_HUGE 4 | |
123 | }; | |
124 | ||
b76db735 | 125 | #ifdef CONFIG_TMPFS |
680d794b AM |
126 | static unsigned long shmem_default_max_blocks(void) |
127 | { | |
ca79b0c2 | 128 | return totalram_pages() / 2; |
680d794b AM |
129 | } |
130 | ||
a3a49a17 | 131 | static int shmem_default_max_inodes(void) |
680d794b | 132 | { |
ca79b0c2 | 133 | unsigned long nr_pages = totalram_pages(); |
a3a49a17 | 134 | unsigned long ul; |
ca79b0c2 | 135 | |
a3a49a17 SF |
136 | ul = INT_MAX; |
137 | ul = min3(ul, nr_pages - totalhigh_pages(), nr_pages / 2); | |
138 | return ul; | |
680d794b | 139 | } |
b76db735 | 140 | #endif |
680d794b | 141 | |
bde05d1c HD |
142 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
143 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
144 | struct shmem_inode_info *info, pgoff_t index); | |
c5bf121e VRP |
145 | static int shmem_swapin_page(struct inode *inode, pgoff_t index, |
146 | struct page **pagep, enum sgp_type sgp, | |
147 | gfp_t gfp, struct vm_area_struct *vma, | |
148 | vm_fault_t *fault_type); | |
68da9f05 | 149 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 150 | struct page **pagep, enum sgp_type sgp, |
cfda0526 | 151 | gfp_t gfp, struct vm_area_struct *vma, |
2b740303 | 152 | struct vm_fault *vmf, vm_fault_t *fault_type); |
68da9f05 | 153 | |
f3f0e1d2 | 154 | int shmem_getpage(struct inode *inode, pgoff_t index, |
9e18eb29 | 155 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
156 | { |
157 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
cfda0526 | 158 | mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL); |
68da9f05 | 159 | } |
1da177e4 | 160 | |
1da177e4 LT |
161 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
162 | { | |
163 | return sb->s_fs_info; | |
164 | } | |
165 | ||
166 | /* | |
167 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
168 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
169 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
170 | * consistent with the pre-accounting of private mappings ... | |
171 | */ | |
172 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
173 | { | |
0b0a0806 | 174 | return (flags & VM_NORESERVE) ? |
191c5424 | 175 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
176 | } |
177 | ||
178 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
179 | { | |
0b0a0806 | 180 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
181 | vm_unacct_memory(VM_ACCT(size)); |
182 | } | |
183 | ||
77142517 KK |
184 | static inline int shmem_reacct_size(unsigned long flags, |
185 | loff_t oldsize, loff_t newsize) | |
186 | { | |
187 | if (!(flags & VM_NORESERVE)) { | |
188 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
189 | return security_vm_enough_memory_mm(current->mm, | |
190 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
191 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
192 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
193 | } | |
194 | return 0; | |
195 | } | |
196 | ||
1da177e4 LT |
197 | /* |
198 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 199 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
200 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
201 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
202 | */ | |
800d8c63 | 203 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 204 | { |
800d8c63 KS |
205 | if (!(flags & VM_NORESERVE)) |
206 | return 0; | |
207 | ||
208 | return security_vm_enough_memory_mm(current->mm, | |
209 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
210 | } |
211 | ||
212 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
213 | { | |
0b0a0806 | 214 | if (flags & VM_NORESERVE) |
09cbfeaf | 215 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
216 | } |
217 | ||
0f079694 MR |
218 | static inline bool shmem_inode_acct_block(struct inode *inode, long pages) |
219 | { | |
220 | struct shmem_inode_info *info = SHMEM_I(inode); | |
221 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
222 | ||
223 | if (shmem_acct_block(info->flags, pages)) | |
224 | return false; | |
225 | ||
226 | if (sbinfo->max_blocks) { | |
227 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
228 | sbinfo->max_blocks - pages) > 0) | |
229 | goto unacct; | |
230 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
231 | } | |
232 | ||
233 | return true; | |
234 | ||
235 | unacct: | |
236 | shmem_unacct_blocks(info->flags, pages); | |
237 | return false; | |
238 | } | |
239 | ||
240 | static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages) | |
241 | { | |
242 | struct shmem_inode_info *info = SHMEM_I(inode); | |
243 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
244 | ||
245 | if (sbinfo->max_blocks) | |
246 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
247 | shmem_unacct_blocks(info->flags, pages); | |
248 | } | |
249 | ||
759b9775 | 250 | static const struct super_operations shmem_ops; |
f5e54d6e | 251 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 252 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
253 | static const struct inode_operations shmem_inode_operations; |
254 | static const struct inode_operations shmem_dir_inode_operations; | |
255 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 256 | static const struct vm_operations_struct shmem_vm_ops; |
779750d2 | 257 | static struct file_system_type shmem_fs_type; |
1da177e4 | 258 | |
b0506e48 MR |
259 | bool vma_is_shmem(struct vm_area_struct *vma) |
260 | { | |
261 | return vma->vm_ops == &shmem_vm_ops; | |
262 | } | |
263 | ||
1da177e4 | 264 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 265 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 266 | |
5b04c689 PE |
267 | static int shmem_reserve_inode(struct super_block *sb) |
268 | { | |
269 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
270 | if (sbinfo->max_inodes) { | |
271 | spin_lock(&sbinfo->stat_lock); | |
272 | if (!sbinfo->free_inodes) { | |
273 | spin_unlock(&sbinfo->stat_lock); | |
274 | return -ENOSPC; | |
275 | } | |
276 | sbinfo->free_inodes--; | |
277 | spin_unlock(&sbinfo->stat_lock); | |
278 | } | |
279 | return 0; | |
280 | } | |
281 | ||
282 | static void shmem_free_inode(struct super_block *sb) | |
283 | { | |
284 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
285 | if (sbinfo->max_inodes) { | |
286 | spin_lock(&sbinfo->stat_lock); | |
287 | sbinfo->free_inodes++; | |
288 | spin_unlock(&sbinfo->stat_lock); | |
289 | } | |
290 | } | |
291 | ||
46711810 | 292 | /** |
41ffe5d5 | 293 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
294 | * @inode: inode to recalc |
295 | * | |
296 | * We have to calculate the free blocks since the mm can drop | |
297 | * undirtied hole pages behind our back. | |
298 | * | |
299 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
300 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
301 | * | |
302 | * It has to be called with the spinlock held. | |
303 | */ | |
304 | static void shmem_recalc_inode(struct inode *inode) | |
305 | { | |
306 | struct shmem_inode_info *info = SHMEM_I(inode); | |
307 | long freed; | |
308 | ||
309 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
310 | if (freed > 0) { | |
311 | info->alloced -= freed; | |
54af6042 | 312 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
0f079694 | 313 | shmem_inode_unacct_blocks(inode, freed); |
1da177e4 LT |
314 | } |
315 | } | |
316 | ||
800d8c63 KS |
317 | bool shmem_charge(struct inode *inode, long pages) |
318 | { | |
319 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 320 | unsigned long flags; |
800d8c63 | 321 | |
0f079694 | 322 | if (!shmem_inode_acct_block(inode, pages)) |
800d8c63 | 323 | return false; |
b1cc94ab | 324 | |
aaa52e34 HD |
325 | /* nrpages adjustment first, then shmem_recalc_inode() when balanced */ |
326 | inode->i_mapping->nrpages += pages; | |
327 | ||
4595ef88 | 328 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
329 | info->alloced += pages; |
330 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
331 | shmem_recalc_inode(inode); | |
4595ef88 | 332 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 | 333 | |
800d8c63 KS |
334 | return true; |
335 | } | |
336 | ||
337 | void shmem_uncharge(struct inode *inode, long pages) | |
338 | { | |
339 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 340 | unsigned long flags; |
800d8c63 | 341 | |
aaa52e34 HD |
342 | /* nrpages adjustment done by __delete_from_page_cache() or caller */ |
343 | ||
4595ef88 | 344 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
345 | info->alloced -= pages; |
346 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
347 | shmem_recalc_inode(inode); | |
4595ef88 | 348 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 | 349 | |
0f079694 | 350 | shmem_inode_unacct_blocks(inode, pages); |
800d8c63 KS |
351 | } |
352 | ||
7a5d0fbb | 353 | /* |
62f945b6 | 354 | * Replace item expected in xarray by a new item, while holding xa_lock. |
7a5d0fbb | 355 | */ |
62f945b6 | 356 | static int shmem_replace_entry(struct address_space *mapping, |
7a5d0fbb HD |
357 | pgoff_t index, void *expected, void *replacement) |
358 | { | |
62f945b6 | 359 | XA_STATE(xas, &mapping->i_pages, index); |
6dbaf22c | 360 | void *item; |
7a5d0fbb HD |
361 | |
362 | VM_BUG_ON(!expected); | |
6dbaf22c | 363 | VM_BUG_ON(!replacement); |
62f945b6 | 364 | item = xas_load(&xas); |
7a5d0fbb HD |
365 | if (item != expected) |
366 | return -ENOENT; | |
62f945b6 | 367 | xas_store(&xas, replacement); |
7a5d0fbb HD |
368 | return 0; |
369 | } | |
370 | ||
d1899228 HD |
371 | /* |
372 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
373 | * that an entry was not already brought back from swap by a racing thread. | |
374 | * | |
375 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
376 | * might be reused, and again be SwapCache, using the same swap as before. | |
377 | */ | |
378 | static bool shmem_confirm_swap(struct address_space *mapping, | |
379 | pgoff_t index, swp_entry_t swap) | |
380 | { | |
a12831bf | 381 | return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap); |
d1899228 HD |
382 | } |
383 | ||
5a6e75f8 KS |
384 | /* |
385 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
386 | * | |
387 | * SHMEM_HUGE_NEVER: | |
388 | * disables huge pages for the mount; | |
389 | * SHMEM_HUGE_ALWAYS: | |
390 | * enables huge pages for the mount; | |
391 | * SHMEM_HUGE_WITHIN_SIZE: | |
392 | * only allocate huge pages if the page will be fully within i_size, | |
393 | * also respect fadvise()/madvise() hints; | |
394 | * SHMEM_HUGE_ADVISE: | |
395 | * only allocate huge pages if requested with fadvise()/madvise(); | |
396 | */ | |
397 | ||
398 | #define SHMEM_HUGE_NEVER 0 | |
399 | #define SHMEM_HUGE_ALWAYS 1 | |
400 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
401 | #define SHMEM_HUGE_ADVISE 3 | |
402 | ||
403 | /* | |
404 | * Special values. | |
405 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
406 | * | |
407 | * SHMEM_HUGE_DENY: | |
408 | * disables huge on shm_mnt and all mounts, for emergency use; | |
409 | * SHMEM_HUGE_FORCE: | |
410 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
411 | * | |
412 | */ | |
413 | #define SHMEM_HUGE_DENY (-1) | |
414 | #define SHMEM_HUGE_FORCE (-2) | |
415 | ||
e496cf3d | 416 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
417 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
418 | ||
5b9c98f3 | 419 | static int shmem_huge __read_mostly; |
5a6e75f8 | 420 | |
e5f2249a | 421 | #if defined(CONFIG_SYSFS) |
5a6e75f8 KS |
422 | static int shmem_parse_huge(const char *str) |
423 | { | |
424 | if (!strcmp(str, "never")) | |
425 | return SHMEM_HUGE_NEVER; | |
426 | if (!strcmp(str, "always")) | |
427 | return SHMEM_HUGE_ALWAYS; | |
428 | if (!strcmp(str, "within_size")) | |
429 | return SHMEM_HUGE_WITHIN_SIZE; | |
430 | if (!strcmp(str, "advise")) | |
431 | return SHMEM_HUGE_ADVISE; | |
432 | if (!strcmp(str, "deny")) | |
433 | return SHMEM_HUGE_DENY; | |
434 | if (!strcmp(str, "force")) | |
435 | return SHMEM_HUGE_FORCE; | |
436 | return -EINVAL; | |
437 | } | |
e5f2249a | 438 | #endif |
5a6e75f8 | 439 | |
e5f2249a | 440 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
441 | static const char *shmem_format_huge(int huge) |
442 | { | |
443 | switch (huge) { | |
444 | case SHMEM_HUGE_NEVER: | |
445 | return "never"; | |
446 | case SHMEM_HUGE_ALWAYS: | |
447 | return "always"; | |
448 | case SHMEM_HUGE_WITHIN_SIZE: | |
449 | return "within_size"; | |
450 | case SHMEM_HUGE_ADVISE: | |
451 | return "advise"; | |
452 | case SHMEM_HUGE_DENY: | |
453 | return "deny"; | |
454 | case SHMEM_HUGE_FORCE: | |
455 | return "force"; | |
456 | default: | |
457 | VM_BUG_ON(1); | |
458 | return "bad_val"; | |
459 | } | |
460 | } | |
f1f5929c | 461 | #endif |
5a6e75f8 | 462 | |
779750d2 KS |
463 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
464 | struct shrink_control *sc, unsigned long nr_to_split) | |
465 | { | |
466 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 467 | LIST_HEAD(to_remove); |
779750d2 KS |
468 | struct inode *inode; |
469 | struct shmem_inode_info *info; | |
470 | struct page *page; | |
471 | unsigned long batch = sc ? sc->nr_to_scan : 128; | |
472 | int removed = 0, split = 0; | |
473 | ||
474 | if (list_empty(&sbinfo->shrinklist)) | |
475 | return SHRINK_STOP; | |
476 | ||
477 | spin_lock(&sbinfo->shrinklist_lock); | |
478 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
479 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
480 | ||
481 | /* pin the inode */ | |
482 | inode = igrab(&info->vfs_inode); | |
483 | ||
484 | /* inode is about to be evicted */ | |
485 | if (!inode) { | |
486 | list_del_init(&info->shrinklist); | |
487 | removed++; | |
488 | goto next; | |
489 | } | |
490 | ||
491 | /* Check if there's anything to gain */ | |
492 | if (round_up(inode->i_size, PAGE_SIZE) == | |
493 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 494 | list_move(&info->shrinklist, &to_remove); |
779750d2 | 495 | removed++; |
779750d2 KS |
496 | goto next; |
497 | } | |
498 | ||
499 | list_move(&info->shrinklist, &list); | |
500 | next: | |
501 | if (!--batch) | |
502 | break; | |
503 | } | |
504 | spin_unlock(&sbinfo->shrinklist_lock); | |
505 | ||
253fd0f0 KS |
506 | list_for_each_safe(pos, next, &to_remove) { |
507 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
508 | inode = &info->vfs_inode; | |
509 | list_del_init(&info->shrinklist); | |
510 | iput(inode); | |
511 | } | |
512 | ||
779750d2 KS |
513 | list_for_each_safe(pos, next, &list) { |
514 | int ret; | |
515 | ||
516 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
517 | inode = &info->vfs_inode; | |
518 | ||
b3cd54b2 KS |
519 | if (nr_to_split && split >= nr_to_split) |
520 | goto leave; | |
779750d2 | 521 | |
b3cd54b2 | 522 | page = find_get_page(inode->i_mapping, |
779750d2 KS |
523 | (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT); |
524 | if (!page) | |
525 | goto drop; | |
526 | ||
b3cd54b2 | 527 | /* No huge page at the end of the file: nothing to split */ |
779750d2 | 528 | if (!PageTransHuge(page)) { |
779750d2 KS |
529 | put_page(page); |
530 | goto drop; | |
531 | } | |
532 | ||
b3cd54b2 KS |
533 | /* |
534 | * Leave the inode on the list if we failed to lock | |
535 | * the page at this time. | |
536 | * | |
537 | * Waiting for the lock may lead to deadlock in the | |
538 | * reclaim path. | |
539 | */ | |
540 | if (!trylock_page(page)) { | |
541 | put_page(page); | |
542 | goto leave; | |
543 | } | |
544 | ||
779750d2 KS |
545 | ret = split_huge_page(page); |
546 | unlock_page(page); | |
547 | put_page(page); | |
548 | ||
b3cd54b2 KS |
549 | /* If split failed leave the inode on the list */ |
550 | if (ret) | |
551 | goto leave; | |
779750d2 KS |
552 | |
553 | split++; | |
554 | drop: | |
555 | list_del_init(&info->shrinklist); | |
556 | removed++; | |
b3cd54b2 | 557 | leave: |
779750d2 KS |
558 | iput(inode); |
559 | } | |
560 | ||
561 | spin_lock(&sbinfo->shrinklist_lock); | |
562 | list_splice_tail(&list, &sbinfo->shrinklist); | |
563 | sbinfo->shrinklist_len -= removed; | |
564 | spin_unlock(&sbinfo->shrinklist_lock); | |
565 | ||
566 | return split; | |
567 | } | |
568 | ||
569 | static long shmem_unused_huge_scan(struct super_block *sb, | |
570 | struct shrink_control *sc) | |
571 | { | |
572 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
573 | ||
574 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
575 | return SHRINK_STOP; | |
576 | ||
577 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
578 | } | |
579 | ||
580 | static long shmem_unused_huge_count(struct super_block *sb, | |
581 | struct shrink_control *sc) | |
582 | { | |
583 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
584 | return READ_ONCE(sbinfo->shrinklist_len); | |
585 | } | |
e496cf3d | 586 | #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 KS |
587 | |
588 | #define shmem_huge SHMEM_HUGE_DENY | |
589 | ||
779750d2 KS |
590 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
591 | struct shrink_control *sc, unsigned long nr_to_split) | |
592 | { | |
593 | return 0; | |
594 | } | |
e496cf3d | 595 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 596 | |
89fdcd26 YS |
597 | static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo) |
598 | { | |
599 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && | |
600 | (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) && | |
601 | shmem_huge != SHMEM_HUGE_DENY) | |
602 | return true; | |
603 | return false; | |
604 | } | |
605 | ||
46f65ec1 HD |
606 | /* |
607 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
608 | */ | |
609 | static int shmem_add_to_page_cache(struct page *page, | |
610 | struct address_space *mapping, | |
552446a4 | 611 | pgoff_t index, void *expected, gfp_t gfp) |
46f65ec1 | 612 | { |
552446a4 MW |
613 | XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page)); |
614 | unsigned long i = 0; | |
d8c6546b | 615 | unsigned long nr = compound_nr(page); |
46f65ec1 | 616 | |
800d8c63 KS |
617 | VM_BUG_ON_PAGE(PageTail(page), page); |
618 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
619 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
620 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 621 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 622 | |
800d8c63 | 623 | page_ref_add(page, nr); |
b065b432 HD |
624 | page->mapping = mapping; |
625 | page->index = index; | |
626 | ||
552446a4 MW |
627 | do { |
628 | void *entry; | |
629 | xas_lock_irq(&xas); | |
630 | entry = xas_find_conflict(&xas); | |
631 | if (entry != expected) | |
632 | xas_set_err(&xas, -EEXIST); | |
633 | xas_create_range(&xas); | |
634 | if (xas_error(&xas)) | |
635 | goto unlock; | |
636 | next: | |
4101196b | 637 | xas_store(&xas, page); |
552446a4 MW |
638 | if (++i < nr) { |
639 | xas_next(&xas); | |
640 | goto next; | |
800d8c63 | 641 | } |
552446a4 | 642 | if (PageTransHuge(page)) { |
800d8c63 | 643 | count_vm_event(THP_FILE_ALLOC); |
552446a4 | 644 | __inc_node_page_state(page, NR_SHMEM_THPS); |
800d8c63 | 645 | } |
800d8c63 | 646 | mapping->nrpages += nr; |
11fb9989 MG |
647 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); |
648 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr); | |
552446a4 MW |
649 | unlock: |
650 | xas_unlock_irq(&xas); | |
651 | } while (xas_nomem(&xas, gfp)); | |
652 | ||
653 | if (xas_error(&xas)) { | |
b065b432 | 654 | page->mapping = NULL; |
800d8c63 | 655 | page_ref_sub(page, nr); |
552446a4 | 656 | return xas_error(&xas); |
46f65ec1 | 657 | } |
552446a4 MW |
658 | |
659 | return 0; | |
46f65ec1 HD |
660 | } |
661 | ||
6922c0c7 HD |
662 | /* |
663 | * Like delete_from_page_cache, but substitutes swap for page. | |
664 | */ | |
665 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
666 | { | |
667 | struct address_space *mapping = page->mapping; | |
668 | int error; | |
669 | ||
800d8c63 KS |
670 | VM_BUG_ON_PAGE(PageCompound(page), page); |
671 | ||
b93b0163 | 672 | xa_lock_irq(&mapping->i_pages); |
62f945b6 | 673 | error = shmem_replace_entry(mapping, page->index, page, radswap); |
6922c0c7 HD |
674 | page->mapping = NULL; |
675 | mapping->nrpages--; | |
11fb9989 MG |
676 | __dec_node_page_state(page, NR_FILE_PAGES); |
677 | __dec_node_page_state(page, NR_SHMEM); | |
b93b0163 | 678 | xa_unlock_irq(&mapping->i_pages); |
09cbfeaf | 679 | put_page(page); |
6922c0c7 HD |
680 | BUG_ON(error); |
681 | } | |
682 | ||
7a5d0fbb | 683 | /* |
c121d3bb | 684 | * Remove swap entry from page cache, free the swap and its page cache. |
7a5d0fbb HD |
685 | */ |
686 | static int shmem_free_swap(struct address_space *mapping, | |
687 | pgoff_t index, void *radswap) | |
688 | { | |
6dbaf22c | 689 | void *old; |
7a5d0fbb | 690 | |
55f3f7ea | 691 | old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0); |
6dbaf22c JW |
692 | if (old != radswap) |
693 | return -ENOENT; | |
694 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
695 | return 0; | |
7a5d0fbb HD |
696 | } |
697 | ||
6a15a370 VB |
698 | /* |
699 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 700 | * given offsets are swapped out. |
6a15a370 | 701 | * |
b93b0163 | 702 | * This is safe to call without i_mutex or the i_pages lock thanks to RCU, |
6a15a370 VB |
703 | * as long as the inode doesn't go away and racy results are not a problem. |
704 | */ | |
48131e03 VB |
705 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
706 | pgoff_t start, pgoff_t end) | |
6a15a370 | 707 | { |
7ae3424f | 708 | XA_STATE(xas, &mapping->i_pages, start); |
6a15a370 | 709 | struct page *page; |
48131e03 | 710 | unsigned long swapped = 0; |
6a15a370 VB |
711 | |
712 | rcu_read_lock(); | |
7ae3424f MW |
713 | xas_for_each(&xas, page, end - 1) { |
714 | if (xas_retry(&xas, page)) | |
2cf938aa | 715 | continue; |
3159f943 | 716 | if (xa_is_value(page)) |
6a15a370 VB |
717 | swapped++; |
718 | ||
719 | if (need_resched()) { | |
7ae3424f | 720 | xas_pause(&xas); |
6a15a370 | 721 | cond_resched_rcu(); |
6a15a370 VB |
722 | } |
723 | } | |
724 | ||
725 | rcu_read_unlock(); | |
726 | ||
727 | return swapped << PAGE_SHIFT; | |
728 | } | |
729 | ||
48131e03 VB |
730 | /* |
731 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
732 | * given vma is swapped out. | |
733 | * | |
b93b0163 | 734 | * This is safe to call without i_mutex or the i_pages lock thanks to RCU, |
48131e03 VB |
735 | * as long as the inode doesn't go away and racy results are not a problem. |
736 | */ | |
737 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
738 | { | |
739 | struct inode *inode = file_inode(vma->vm_file); | |
740 | struct shmem_inode_info *info = SHMEM_I(inode); | |
741 | struct address_space *mapping = inode->i_mapping; | |
742 | unsigned long swapped; | |
743 | ||
744 | /* Be careful as we don't hold info->lock */ | |
745 | swapped = READ_ONCE(info->swapped); | |
746 | ||
747 | /* | |
748 | * The easier cases are when the shmem object has nothing in swap, or | |
749 | * the vma maps it whole. Then we can simply use the stats that we | |
750 | * already track. | |
751 | */ | |
752 | if (!swapped) | |
753 | return 0; | |
754 | ||
755 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
756 | return swapped << PAGE_SHIFT; | |
757 | ||
758 | /* Here comes the more involved part */ | |
759 | return shmem_partial_swap_usage(mapping, | |
760 | linear_page_index(vma, vma->vm_start), | |
761 | linear_page_index(vma, vma->vm_end)); | |
762 | } | |
763 | ||
24513264 HD |
764 | /* |
765 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
766 | */ | |
767 | void shmem_unlock_mapping(struct address_space *mapping) | |
768 | { | |
769 | struct pagevec pvec; | |
770 | pgoff_t indices[PAGEVEC_SIZE]; | |
771 | pgoff_t index = 0; | |
772 | ||
86679820 | 773 | pagevec_init(&pvec); |
24513264 HD |
774 | /* |
775 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
776 | */ | |
777 | while (!mapping_unevictable(mapping)) { | |
778 | /* | |
779 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
780 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
781 | */ | |
0cd6144a JW |
782 | pvec.nr = find_get_entries(mapping, index, |
783 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
784 | if (!pvec.nr) |
785 | break; | |
786 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 787 | pagevec_remove_exceptionals(&pvec); |
64e3d12f | 788 | check_move_unevictable_pages(&pvec); |
24513264 HD |
789 | pagevec_release(&pvec); |
790 | cond_resched(); | |
791 | } | |
7a5d0fbb HD |
792 | } |
793 | ||
794 | /* | |
7f4446ee | 795 | * Remove range of pages and swap entries from page cache, and free them. |
1635f6a7 | 796 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 797 | */ |
1635f6a7 HD |
798 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
799 | bool unfalloc) | |
1da177e4 | 800 | { |
285b2c4f | 801 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 802 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
803 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
804 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
805 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
806 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 807 | struct pagevec pvec; |
7a5d0fbb HD |
808 | pgoff_t indices[PAGEVEC_SIZE]; |
809 | long nr_swaps_freed = 0; | |
285b2c4f | 810 | pgoff_t index; |
bda97eab HD |
811 | int i; |
812 | ||
83e4fa9c HD |
813 | if (lend == -1) |
814 | end = -1; /* unsigned, so actually very big */ | |
bda97eab | 815 | |
86679820 | 816 | pagevec_init(&pvec); |
bda97eab | 817 | index = start; |
83e4fa9c | 818 | while (index < end) { |
0cd6144a JW |
819 | pvec.nr = find_get_entries(mapping, index, |
820 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
821 | pvec.pages, indices); | |
7a5d0fbb HD |
822 | if (!pvec.nr) |
823 | break; | |
bda97eab HD |
824 | for (i = 0; i < pagevec_count(&pvec); i++) { |
825 | struct page *page = pvec.pages[i]; | |
826 | ||
7a5d0fbb | 827 | index = indices[i]; |
83e4fa9c | 828 | if (index >= end) |
bda97eab HD |
829 | break; |
830 | ||
3159f943 | 831 | if (xa_is_value(page)) { |
1635f6a7 HD |
832 | if (unfalloc) |
833 | continue; | |
7a5d0fbb HD |
834 | nr_swaps_freed += !shmem_free_swap(mapping, |
835 | index, page); | |
bda97eab | 836 | continue; |
7a5d0fbb HD |
837 | } |
838 | ||
800d8c63 KS |
839 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
840 | ||
7a5d0fbb | 841 | if (!trylock_page(page)) |
bda97eab | 842 | continue; |
800d8c63 KS |
843 | |
844 | if (PageTransTail(page)) { | |
845 | /* Middle of THP: zero out the page */ | |
846 | clear_highpage(page); | |
847 | unlock_page(page); | |
848 | continue; | |
849 | } else if (PageTransHuge(page)) { | |
850 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
851 | /* | |
852 | * Range ends in the middle of THP: | |
853 | * zero out the page | |
854 | */ | |
855 | clear_highpage(page); | |
856 | unlock_page(page); | |
857 | continue; | |
858 | } | |
859 | index += HPAGE_PMD_NR - 1; | |
860 | i += HPAGE_PMD_NR - 1; | |
861 | } | |
862 | ||
1635f6a7 | 863 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
864 | VM_BUG_ON_PAGE(PageTail(page), page); |
865 | if (page_mapping(page) == mapping) { | |
309381fe | 866 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
867 | truncate_inode_page(mapping, page); |
868 | } | |
bda97eab | 869 | } |
bda97eab HD |
870 | unlock_page(page); |
871 | } | |
0cd6144a | 872 | pagevec_remove_exceptionals(&pvec); |
24513264 | 873 | pagevec_release(&pvec); |
bda97eab HD |
874 | cond_resched(); |
875 | index++; | |
876 | } | |
1da177e4 | 877 | |
83e4fa9c | 878 | if (partial_start) { |
bda97eab | 879 | struct page *page = NULL; |
9e18eb29 | 880 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 881 | if (page) { |
09cbfeaf | 882 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
883 | if (start > end) { |
884 | top = partial_end; | |
885 | partial_end = 0; | |
886 | } | |
887 | zero_user_segment(page, partial_start, top); | |
888 | set_page_dirty(page); | |
889 | unlock_page(page); | |
09cbfeaf | 890 | put_page(page); |
83e4fa9c HD |
891 | } |
892 | } | |
893 | if (partial_end) { | |
894 | struct page *page = NULL; | |
9e18eb29 | 895 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
896 | if (page) { |
897 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
898 | set_page_dirty(page); |
899 | unlock_page(page); | |
09cbfeaf | 900 | put_page(page); |
bda97eab HD |
901 | } |
902 | } | |
83e4fa9c HD |
903 | if (start >= end) |
904 | return; | |
bda97eab HD |
905 | |
906 | index = start; | |
b1a36650 | 907 | while (index < end) { |
bda97eab | 908 | cond_resched(); |
0cd6144a JW |
909 | |
910 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 911 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 912 | pvec.pages, indices); |
7a5d0fbb | 913 | if (!pvec.nr) { |
b1a36650 HD |
914 | /* If all gone or hole-punch or unfalloc, we're done */ |
915 | if (index == start || end != -1) | |
bda97eab | 916 | break; |
b1a36650 | 917 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
918 | index = start; |
919 | continue; | |
920 | } | |
bda97eab HD |
921 | for (i = 0; i < pagevec_count(&pvec); i++) { |
922 | struct page *page = pvec.pages[i]; | |
923 | ||
7a5d0fbb | 924 | index = indices[i]; |
83e4fa9c | 925 | if (index >= end) |
bda97eab HD |
926 | break; |
927 | ||
3159f943 | 928 | if (xa_is_value(page)) { |
1635f6a7 HD |
929 | if (unfalloc) |
930 | continue; | |
b1a36650 HD |
931 | if (shmem_free_swap(mapping, index, page)) { |
932 | /* Swap was replaced by page: retry */ | |
933 | index--; | |
934 | break; | |
935 | } | |
936 | nr_swaps_freed++; | |
7a5d0fbb HD |
937 | continue; |
938 | } | |
939 | ||
bda97eab | 940 | lock_page(page); |
800d8c63 KS |
941 | |
942 | if (PageTransTail(page)) { | |
943 | /* Middle of THP: zero out the page */ | |
944 | clear_highpage(page); | |
945 | unlock_page(page); | |
946 | /* | |
947 | * Partial thp truncate due 'start' in middle | |
948 | * of THP: don't need to look on these pages | |
949 | * again on !pvec.nr restart. | |
950 | */ | |
951 | if (index != round_down(end, HPAGE_PMD_NR)) | |
952 | start++; | |
953 | continue; | |
954 | } else if (PageTransHuge(page)) { | |
955 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
956 | /* | |
957 | * Range ends in the middle of THP: | |
958 | * zero out the page | |
959 | */ | |
960 | clear_highpage(page); | |
961 | unlock_page(page); | |
962 | continue; | |
963 | } | |
964 | index += HPAGE_PMD_NR - 1; | |
965 | i += HPAGE_PMD_NR - 1; | |
966 | } | |
967 | ||
1635f6a7 | 968 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
969 | VM_BUG_ON_PAGE(PageTail(page), page); |
970 | if (page_mapping(page) == mapping) { | |
309381fe | 971 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 972 | truncate_inode_page(mapping, page); |
b1a36650 HD |
973 | } else { |
974 | /* Page was replaced by swap: retry */ | |
975 | unlock_page(page); | |
976 | index--; | |
977 | break; | |
1635f6a7 | 978 | } |
7a5d0fbb | 979 | } |
bda97eab HD |
980 | unlock_page(page); |
981 | } | |
0cd6144a | 982 | pagevec_remove_exceptionals(&pvec); |
24513264 | 983 | pagevec_release(&pvec); |
bda97eab HD |
984 | index++; |
985 | } | |
94c1e62d | 986 | |
4595ef88 | 987 | spin_lock_irq(&info->lock); |
7a5d0fbb | 988 | info->swapped -= nr_swaps_freed; |
1da177e4 | 989 | shmem_recalc_inode(inode); |
4595ef88 | 990 | spin_unlock_irq(&info->lock); |
1635f6a7 | 991 | } |
1da177e4 | 992 | |
1635f6a7 HD |
993 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
994 | { | |
995 | shmem_undo_range(inode, lstart, lend, false); | |
078cd827 | 996 | inode->i_ctime = inode->i_mtime = current_time(inode); |
1da177e4 | 997 | } |
94c1e62d | 998 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 999 | |
a528d35e DH |
1000 | static int shmem_getattr(const struct path *path, struct kstat *stat, |
1001 | u32 request_mask, unsigned int query_flags) | |
44a30220 | 1002 | { |
a528d35e | 1003 | struct inode *inode = path->dentry->d_inode; |
44a30220 | 1004 | struct shmem_inode_info *info = SHMEM_I(inode); |
89fdcd26 | 1005 | struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb); |
44a30220 | 1006 | |
d0424c42 | 1007 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
4595ef88 | 1008 | spin_lock_irq(&info->lock); |
d0424c42 | 1009 | shmem_recalc_inode(inode); |
4595ef88 | 1010 | spin_unlock_irq(&info->lock); |
d0424c42 | 1011 | } |
44a30220 | 1012 | generic_fillattr(inode, stat); |
89fdcd26 YS |
1013 | |
1014 | if (is_huge_enabled(sb_info)) | |
1015 | stat->blksize = HPAGE_PMD_SIZE; | |
1016 | ||
44a30220 YZ |
1017 | return 0; |
1018 | } | |
1019 | ||
94c1e62d | 1020 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 1021 | { |
75c3cfa8 | 1022 | struct inode *inode = d_inode(dentry); |
40e041a2 | 1023 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1024 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 LT |
1025 | int error; |
1026 | ||
31051c85 | 1027 | error = setattr_prepare(dentry, attr); |
db78b877 CH |
1028 | if (error) |
1029 | return error; | |
1030 | ||
94c1e62d HD |
1031 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
1032 | loff_t oldsize = inode->i_size; | |
1033 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 1034 | |
40e041a2 DH |
1035 | /* protected by i_mutex */ |
1036 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
1037 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
1038 | return -EPERM; | |
1039 | ||
94c1e62d | 1040 | if (newsize != oldsize) { |
77142517 KK |
1041 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
1042 | oldsize, newsize); | |
1043 | if (error) | |
1044 | return error; | |
94c1e62d | 1045 | i_size_write(inode, newsize); |
078cd827 | 1046 | inode->i_ctime = inode->i_mtime = current_time(inode); |
94c1e62d | 1047 | } |
afa2db2f | 1048 | if (newsize <= oldsize) { |
94c1e62d | 1049 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
1050 | if (oldsize > holebegin) |
1051 | unmap_mapping_range(inode->i_mapping, | |
1052 | holebegin, 0, 1); | |
1053 | if (info->alloced) | |
1054 | shmem_truncate_range(inode, | |
1055 | newsize, (loff_t)-1); | |
94c1e62d | 1056 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
1057 | if (oldsize > holebegin) |
1058 | unmap_mapping_range(inode->i_mapping, | |
1059 | holebegin, 0, 1); | |
779750d2 KS |
1060 | |
1061 | /* | |
1062 | * Part of the huge page can be beyond i_size: subject | |
1063 | * to shrink under memory pressure. | |
1064 | */ | |
1065 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { | |
1066 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1067 | /* |
1068 | * _careful to defend against unlocked access to | |
1069 | * ->shrink_list in shmem_unused_huge_shrink() | |
1070 | */ | |
1071 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1072 | list_add_tail(&info->shrinklist, |
1073 | &sbinfo->shrinklist); | |
1074 | sbinfo->shrinklist_len++; | |
1075 | } | |
1076 | spin_unlock(&sbinfo->shrinklist_lock); | |
1077 | } | |
94c1e62d | 1078 | } |
1da177e4 LT |
1079 | } |
1080 | ||
db78b877 | 1081 | setattr_copy(inode, attr); |
db78b877 | 1082 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 1083 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
1084 | return error; |
1085 | } | |
1086 | ||
1f895f75 | 1087 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1088 | { |
1da177e4 | 1089 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1090 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 | 1091 | |
3889e6e7 | 1092 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
1093 | shmem_unacct_size(info->flags, inode->i_size); |
1094 | inode->i_size = 0; | |
3889e6e7 | 1095 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1096 | if (!list_empty(&info->shrinklist)) { |
1097 | spin_lock(&sbinfo->shrinklist_lock); | |
1098 | if (!list_empty(&info->shrinklist)) { | |
1099 | list_del_init(&info->shrinklist); | |
1100 | sbinfo->shrinklist_len--; | |
1101 | } | |
1102 | spin_unlock(&sbinfo->shrinklist_lock); | |
1103 | } | |
af53d3e9 HD |
1104 | while (!list_empty(&info->swaplist)) { |
1105 | /* Wait while shmem_unuse() is scanning this inode... */ | |
1106 | wait_var_event(&info->stop_eviction, | |
1107 | !atomic_read(&info->stop_eviction)); | |
cb5f7b9a | 1108 | mutex_lock(&shmem_swaplist_mutex); |
af53d3e9 HD |
1109 | /* ...but beware of the race if we peeked too early */ |
1110 | if (!atomic_read(&info->stop_eviction)) | |
1111 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1112 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1113 | } |
3ed47db3 | 1114 | } |
b09e0fa4 | 1115 | |
38f38657 | 1116 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 1117 | WARN_ON(inode->i_blocks); |
a3a49a17 SF |
1118 | if (!sbinfo->idr_nouse && inode->i_ino) { |
1119 | mutex_lock(&sbinfo->idr_lock); | |
1120 | idr_remove(&sbinfo->idr, inode->i_ino); | |
1121 | mutex_unlock(&sbinfo->idr_lock); | |
1122 | } | |
5b04c689 | 1123 | shmem_free_inode(inode->i_sb); |
dbd5768f | 1124 | clear_inode(inode); |
1da177e4 LT |
1125 | } |
1126 | ||
b56a2d8a VRP |
1127 | extern struct swap_info_struct *swap_info[]; |
1128 | ||
1129 | static int shmem_find_swap_entries(struct address_space *mapping, | |
1130 | pgoff_t start, unsigned int nr_entries, | |
1131 | struct page **entries, pgoff_t *indices, | |
87039546 | 1132 | unsigned int type, bool frontswap) |
478922e2 | 1133 | { |
b56a2d8a VRP |
1134 | XA_STATE(xas, &mapping->i_pages, start); |
1135 | struct page *page; | |
87039546 | 1136 | swp_entry_t entry; |
b56a2d8a VRP |
1137 | unsigned int ret = 0; |
1138 | ||
1139 | if (!nr_entries) | |
1140 | return 0; | |
478922e2 MW |
1141 | |
1142 | rcu_read_lock(); | |
b56a2d8a VRP |
1143 | xas_for_each(&xas, page, ULONG_MAX) { |
1144 | if (xas_retry(&xas, page)) | |
5b9c98f3 | 1145 | continue; |
b56a2d8a VRP |
1146 | |
1147 | if (!xa_is_value(page)) | |
478922e2 | 1148 | continue; |
b56a2d8a | 1149 | |
87039546 HD |
1150 | entry = radix_to_swp_entry(page); |
1151 | if (swp_type(entry) != type) | |
1152 | continue; | |
1153 | if (frontswap && | |
1154 | !frontswap_test(swap_info[type], swp_offset(entry))) | |
1155 | continue; | |
b56a2d8a VRP |
1156 | |
1157 | indices[ret] = xas.xa_index; | |
1158 | entries[ret] = page; | |
1159 | ||
1160 | if (need_resched()) { | |
1161 | xas_pause(&xas); | |
1162 | cond_resched_rcu(); | |
1163 | } | |
1164 | if (++ret == nr_entries) | |
1165 | break; | |
478922e2 | 1166 | } |
478922e2 | 1167 | rcu_read_unlock(); |
e21a2955 | 1168 | |
b56a2d8a | 1169 | return ret; |
478922e2 MW |
1170 | } |
1171 | ||
46f65ec1 | 1172 | /* |
b56a2d8a VRP |
1173 | * Move the swapped pages for an inode to page cache. Returns the count |
1174 | * of pages swapped in, or the error in case of failure. | |
46f65ec1 | 1175 | */ |
b56a2d8a VRP |
1176 | static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec, |
1177 | pgoff_t *indices) | |
1da177e4 | 1178 | { |
b56a2d8a VRP |
1179 | int i = 0; |
1180 | int ret = 0; | |
bde05d1c | 1181 | int error = 0; |
b56a2d8a | 1182 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 1183 | |
b56a2d8a VRP |
1184 | for (i = 0; i < pvec.nr; i++) { |
1185 | struct page *page = pvec.pages[i]; | |
2e0e26c7 | 1186 | |
b56a2d8a VRP |
1187 | if (!xa_is_value(page)) |
1188 | continue; | |
1189 | error = shmem_swapin_page(inode, indices[i], | |
1190 | &page, SGP_CACHE, | |
1191 | mapping_gfp_mask(mapping), | |
1192 | NULL, NULL); | |
1193 | if (error == 0) { | |
1194 | unlock_page(page); | |
1195 | put_page(page); | |
1196 | ret++; | |
1197 | } | |
1198 | if (error == -ENOMEM) | |
1199 | break; | |
1200 | error = 0; | |
bde05d1c | 1201 | } |
b56a2d8a VRP |
1202 | return error ? error : ret; |
1203 | } | |
bde05d1c | 1204 | |
b56a2d8a VRP |
1205 | /* |
1206 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1207 | */ | |
1208 | static int shmem_unuse_inode(struct inode *inode, unsigned int type, | |
1209 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1210 | { | |
1211 | struct address_space *mapping = inode->i_mapping; | |
1212 | pgoff_t start = 0; | |
1213 | struct pagevec pvec; | |
1214 | pgoff_t indices[PAGEVEC_SIZE]; | |
1215 | bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0); | |
1216 | int ret = 0; | |
1217 | ||
1218 | pagevec_init(&pvec); | |
1219 | do { | |
1220 | unsigned int nr_entries = PAGEVEC_SIZE; | |
1221 | ||
1222 | if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE) | |
1223 | nr_entries = *fs_pages_to_unuse; | |
1224 | ||
1225 | pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries, | |
1226 | pvec.pages, indices, | |
87039546 | 1227 | type, frontswap); |
b56a2d8a VRP |
1228 | if (pvec.nr == 0) { |
1229 | ret = 0; | |
1230 | break; | |
46f65ec1 | 1231 | } |
b56a2d8a VRP |
1232 | |
1233 | ret = shmem_unuse_swap_entries(inode, pvec, indices); | |
1234 | if (ret < 0) | |
1235 | break; | |
1236 | ||
1237 | if (frontswap_partial) { | |
1238 | *fs_pages_to_unuse -= ret; | |
1239 | if (*fs_pages_to_unuse == 0) { | |
1240 | ret = FRONTSWAP_PAGES_UNUSED; | |
1241 | break; | |
1242 | } | |
1243 | } | |
1244 | ||
1245 | start = indices[pvec.nr - 1]; | |
1246 | } while (true); | |
1247 | ||
1248 | return ret; | |
1da177e4 LT |
1249 | } |
1250 | ||
1251 | /* | |
b56a2d8a VRP |
1252 | * Read all the shared memory data that resides in the swap |
1253 | * device 'type' back into memory, so the swap device can be | |
1254 | * unused. | |
1da177e4 | 1255 | */ |
b56a2d8a VRP |
1256 | int shmem_unuse(unsigned int type, bool frontswap, |
1257 | unsigned long *fs_pages_to_unuse) | |
1da177e4 | 1258 | { |
b56a2d8a | 1259 | struct shmem_inode_info *info, *next; |
bde05d1c HD |
1260 | int error = 0; |
1261 | ||
b56a2d8a VRP |
1262 | if (list_empty(&shmem_swaplist)) |
1263 | return 0; | |
1264 | ||
1265 | mutex_lock(&shmem_swaplist_mutex); | |
b56a2d8a VRP |
1266 | list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) { |
1267 | if (!info->swapped) { | |
6922c0c7 | 1268 | list_del_init(&info->swaplist); |
b56a2d8a VRP |
1269 | continue; |
1270 | } | |
af53d3e9 HD |
1271 | /* |
1272 | * Drop the swaplist mutex while searching the inode for swap; | |
1273 | * but before doing so, make sure shmem_evict_inode() will not | |
1274 | * remove placeholder inode from swaplist, nor let it be freed | |
1275 | * (igrab() would protect from unlink, but not from unmount). | |
1276 | */ | |
1277 | atomic_inc(&info->stop_eviction); | |
b56a2d8a | 1278 | mutex_unlock(&shmem_swaplist_mutex); |
b56a2d8a | 1279 | |
af53d3e9 | 1280 | error = shmem_unuse_inode(&info->vfs_inode, type, frontswap, |
b56a2d8a | 1281 | fs_pages_to_unuse); |
cb5f7b9a | 1282 | cond_resched(); |
b56a2d8a VRP |
1283 | |
1284 | mutex_lock(&shmem_swaplist_mutex); | |
1285 | next = list_next_entry(info, swaplist); | |
1286 | if (!info->swapped) | |
1287 | list_del_init(&info->swaplist); | |
af53d3e9 HD |
1288 | if (atomic_dec_and_test(&info->stop_eviction)) |
1289 | wake_up_var(&info->stop_eviction); | |
b56a2d8a | 1290 | if (error) |
778dd893 | 1291 | break; |
1da177e4 | 1292 | } |
cb5f7b9a | 1293 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1294 | |
778dd893 | 1295 | return error; |
1da177e4 LT |
1296 | } |
1297 | ||
1298 | /* | |
1299 | * Move the page from the page cache to the swap cache. | |
1300 | */ | |
1301 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1302 | { | |
1303 | struct shmem_inode_info *info; | |
1da177e4 | 1304 | struct address_space *mapping; |
1da177e4 | 1305 | struct inode *inode; |
6922c0c7 HD |
1306 | swp_entry_t swap; |
1307 | pgoff_t index; | |
1da177e4 | 1308 | |
800d8c63 | 1309 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1310 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1311 | mapping = page->mapping; |
1312 | index = page->index; | |
1313 | inode = mapping->host; | |
1314 | info = SHMEM_I(inode); | |
1315 | if (info->flags & VM_LOCKED) | |
1316 | goto redirty; | |
d9fe526a | 1317 | if (!total_swap_pages) |
1da177e4 LT |
1318 | goto redirty; |
1319 | ||
d9fe526a | 1320 | /* |
97b713ba CH |
1321 | * Our capabilities prevent regular writeback or sync from ever calling |
1322 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1323 | * its underlying filesystem, in which case tmpfs should write out to | |
1324 | * swap only in response to memory pressure, and not for the writeback | |
1325 | * threads or sync. | |
d9fe526a | 1326 | */ |
48f170fb HD |
1327 | if (!wbc->for_reclaim) { |
1328 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1329 | goto redirty; | |
1330 | } | |
1635f6a7 HD |
1331 | |
1332 | /* | |
1333 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1334 | * value into swapfile.c, the only way we can correctly account for a | |
1335 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1336 | * |
1337 | * That's okay for a page already fallocated earlier, but if we have | |
1338 | * not yet completed the fallocation, then (a) we want to keep track | |
1339 | * of this page in case we have to undo it, and (b) it may not be a | |
1340 | * good idea to continue anyway, once we're pushing into swap. So | |
1341 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1342 | */ |
1343 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1344 | if (inode->i_private) { |
1345 | struct shmem_falloc *shmem_falloc; | |
1346 | spin_lock(&inode->i_lock); | |
1347 | shmem_falloc = inode->i_private; | |
1348 | if (shmem_falloc && | |
8e205f77 | 1349 | !shmem_falloc->waitq && |
1aac1400 HD |
1350 | index >= shmem_falloc->start && |
1351 | index < shmem_falloc->next) | |
1352 | shmem_falloc->nr_unswapped++; | |
1353 | else | |
1354 | shmem_falloc = NULL; | |
1355 | spin_unlock(&inode->i_lock); | |
1356 | if (shmem_falloc) | |
1357 | goto redirty; | |
1358 | } | |
1635f6a7 HD |
1359 | clear_highpage(page); |
1360 | flush_dcache_page(page); | |
1361 | SetPageUptodate(page); | |
1362 | } | |
1363 | ||
38d8b4e6 | 1364 | swap = get_swap_page(page); |
48f170fb HD |
1365 | if (!swap.val) |
1366 | goto redirty; | |
d9fe526a | 1367 | |
b1dea800 HD |
1368 | /* |
1369 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1370 | * if it's not already there. Do it now before the page is |
1371 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1372 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1373 | * we've incremented swapped, because shmem_unuse_inode() will |
1374 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1375 | */ |
48f170fb HD |
1376 | mutex_lock(&shmem_swaplist_mutex); |
1377 | if (list_empty(&info->swaplist)) | |
b56a2d8a | 1378 | list_add(&info->swaplist, &shmem_swaplist); |
b1dea800 | 1379 | |
48f170fb | 1380 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
4595ef88 | 1381 | spin_lock_irq(&info->lock); |
6922c0c7 | 1382 | shmem_recalc_inode(inode); |
267a4c76 | 1383 | info->swapped++; |
4595ef88 | 1384 | spin_unlock_irq(&info->lock); |
6922c0c7 | 1385 | |
267a4c76 HD |
1386 | swap_shmem_alloc(swap); |
1387 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1388 | ||
6922c0c7 | 1389 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1390 | BUG_ON(page_mapped(page)); |
9fab5619 | 1391 | swap_writepage(page, wbc); |
1da177e4 LT |
1392 | return 0; |
1393 | } | |
1394 | ||
6922c0c7 | 1395 | mutex_unlock(&shmem_swaplist_mutex); |
75f6d6d2 | 1396 | put_swap_page(page, swap); |
1da177e4 LT |
1397 | redirty: |
1398 | set_page_dirty(page); | |
d9fe526a HD |
1399 | if (wbc->for_reclaim) |
1400 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1401 | unlock_page(page); | |
1402 | return 0; | |
1da177e4 LT |
1403 | } |
1404 | ||
75edd345 | 1405 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1406 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1407 | { |
095f1fc4 | 1408 | char buffer[64]; |
680d794b | 1409 | |
71fe804b | 1410 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1411 | return; /* show nothing */ |
680d794b | 1412 | |
a7a88b23 | 1413 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1414 | |
1415 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1416 | } |
71fe804b LS |
1417 | |
1418 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1419 | { | |
1420 | struct mempolicy *mpol = NULL; | |
1421 | if (sbinfo->mpol) { | |
1422 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1423 | mpol = sbinfo->mpol; | |
1424 | mpol_get(mpol); | |
1425 | spin_unlock(&sbinfo->stat_lock); | |
1426 | } | |
1427 | return mpol; | |
1428 | } | |
75edd345 HD |
1429 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1430 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1431 | { | |
1432 | } | |
1433 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1434 | { | |
1435 | return NULL; | |
1436 | } | |
1437 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1438 | #ifndef CONFIG_NUMA | |
1439 | #define vm_policy vm_private_data | |
1440 | #endif | |
680d794b | 1441 | |
800d8c63 KS |
1442 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1443 | struct shmem_inode_info *info, pgoff_t index) | |
1444 | { | |
1445 | /* Create a pseudo vma that just contains the policy */ | |
2c4541e2 | 1446 | vma_init(vma, NULL); |
800d8c63 KS |
1447 | /* Bias interleave by inode number to distribute better across nodes */ |
1448 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
800d8c63 KS |
1449 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
1450 | } | |
1451 | ||
1452 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1453 | { | |
1454 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1455 | mpol_cond_put(vma->vm_policy); | |
1456 | } | |
1457 | ||
41ffe5d5 HD |
1458 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1459 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1460 | { |
1da177e4 | 1461 | struct vm_area_struct pvma; |
18a2f371 | 1462 | struct page *page; |
e9e9b7ec | 1463 | struct vm_fault vmf; |
52cd3b07 | 1464 | |
800d8c63 | 1465 | shmem_pseudo_vma_init(&pvma, info, index); |
e9e9b7ec MK |
1466 | vmf.vma = &pvma; |
1467 | vmf.address = 0; | |
1468 | page = swap_cluster_readahead(swap, gfp, &vmf); | |
800d8c63 | 1469 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1470 | |
800d8c63 KS |
1471 | return page; |
1472 | } | |
1473 | ||
1474 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1475 | struct shmem_inode_info *info, pgoff_t index) | |
1476 | { | |
1477 | struct vm_area_struct pvma; | |
7b8d046f MW |
1478 | struct address_space *mapping = info->vfs_inode.i_mapping; |
1479 | pgoff_t hindex; | |
800d8c63 KS |
1480 | struct page *page; |
1481 | ||
e496cf3d | 1482 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1483 | return NULL; |
1484 | ||
4620a06e | 1485 | hindex = round_down(index, HPAGE_PMD_NR); |
7b8d046f MW |
1486 | if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1, |
1487 | XA_PRESENT)) | |
800d8c63 | 1488 | return NULL; |
18a2f371 | 1489 | |
800d8c63 KS |
1490 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1491 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
19deb769 | 1492 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); |
800d8c63 KS |
1493 | shmem_pseudo_vma_destroy(&pvma); |
1494 | if (page) | |
1495 | prep_transhuge_page(page); | |
18a2f371 | 1496 | return page; |
1da177e4 LT |
1497 | } |
1498 | ||
02098fea | 1499 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1500 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1501 | { |
1502 | struct vm_area_struct pvma; | |
18a2f371 | 1503 | struct page *page; |
1da177e4 | 1504 | |
800d8c63 KS |
1505 | shmem_pseudo_vma_init(&pvma, info, index); |
1506 | page = alloc_page_vma(gfp, &pvma, 0); | |
1507 | shmem_pseudo_vma_destroy(&pvma); | |
1508 | ||
1509 | return page; | |
1510 | } | |
1511 | ||
1512 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
0f079694 | 1513 | struct inode *inode, |
800d8c63 KS |
1514 | pgoff_t index, bool huge) |
1515 | { | |
0f079694 | 1516 | struct shmem_inode_info *info = SHMEM_I(inode); |
800d8c63 KS |
1517 | struct page *page; |
1518 | int nr; | |
1519 | int err = -ENOSPC; | |
52cd3b07 | 1520 | |
e496cf3d | 1521 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1522 | huge = false; |
1523 | nr = huge ? HPAGE_PMD_NR : 1; | |
1524 | ||
0f079694 | 1525 | if (!shmem_inode_acct_block(inode, nr)) |
800d8c63 | 1526 | goto failed; |
800d8c63 KS |
1527 | |
1528 | if (huge) | |
1529 | page = shmem_alloc_hugepage(gfp, info, index); | |
1530 | else | |
1531 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1532 | if (page) { |
1533 | __SetPageLocked(page); | |
1534 | __SetPageSwapBacked(page); | |
800d8c63 | 1535 | return page; |
75edd345 | 1536 | } |
18a2f371 | 1537 | |
800d8c63 | 1538 | err = -ENOMEM; |
0f079694 | 1539 | shmem_inode_unacct_blocks(inode, nr); |
800d8c63 KS |
1540 | failed: |
1541 | return ERR_PTR(err); | |
1da177e4 | 1542 | } |
71fe804b | 1543 | |
bde05d1c HD |
1544 | /* |
1545 | * When a page is moved from swapcache to shmem filecache (either by the | |
1546 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1547 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1548 | * ignorance of the mapping it belongs to. If that mapping has special | |
1549 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1550 | * we may need to copy to a suitable page before moving to filecache. | |
1551 | * | |
1552 | * In a future release, this may well be extended to respect cpuset and | |
1553 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1554 | * but for now it is a simple matter of zone. | |
1555 | */ | |
1556 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1557 | { | |
1558 | return page_zonenum(page) > gfp_zone(gfp); | |
1559 | } | |
1560 | ||
1561 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1562 | struct shmem_inode_info *info, pgoff_t index) | |
1563 | { | |
1564 | struct page *oldpage, *newpage; | |
1565 | struct address_space *swap_mapping; | |
c1cb20d4 | 1566 | swp_entry_t entry; |
bde05d1c HD |
1567 | pgoff_t swap_index; |
1568 | int error; | |
1569 | ||
1570 | oldpage = *pagep; | |
c1cb20d4 YZ |
1571 | entry.val = page_private(oldpage); |
1572 | swap_index = swp_offset(entry); | |
bde05d1c HD |
1573 | swap_mapping = page_mapping(oldpage); |
1574 | ||
1575 | /* | |
1576 | * We have arrived here because our zones are constrained, so don't | |
1577 | * limit chance of success by further cpuset and node constraints. | |
1578 | */ | |
1579 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1580 | newpage = shmem_alloc_page(gfp, info, index); | |
1581 | if (!newpage) | |
1582 | return -ENOMEM; | |
bde05d1c | 1583 | |
09cbfeaf | 1584 | get_page(newpage); |
bde05d1c | 1585 | copy_highpage(newpage, oldpage); |
0142ef6c | 1586 | flush_dcache_page(newpage); |
bde05d1c | 1587 | |
9956edf3 HD |
1588 | __SetPageLocked(newpage); |
1589 | __SetPageSwapBacked(newpage); | |
bde05d1c | 1590 | SetPageUptodate(newpage); |
c1cb20d4 | 1591 | set_page_private(newpage, entry.val); |
bde05d1c HD |
1592 | SetPageSwapCache(newpage); |
1593 | ||
1594 | /* | |
1595 | * Our caller will very soon move newpage out of swapcache, but it's | |
1596 | * a nice clean interface for us to replace oldpage by newpage there. | |
1597 | */ | |
b93b0163 | 1598 | xa_lock_irq(&swap_mapping->i_pages); |
62f945b6 | 1599 | error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage); |
0142ef6c | 1600 | if (!error) { |
11fb9989 MG |
1601 | __inc_node_page_state(newpage, NR_FILE_PAGES); |
1602 | __dec_node_page_state(oldpage, NR_FILE_PAGES); | |
0142ef6c | 1603 | } |
b93b0163 | 1604 | xa_unlock_irq(&swap_mapping->i_pages); |
bde05d1c | 1605 | |
0142ef6c HD |
1606 | if (unlikely(error)) { |
1607 | /* | |
1608 | * Is this possible? I think not, now that our callers check | |
1609 | * both PageSwapCache and page_private after getting page lock; | |
1610 | * but be defensive. Reverse old to newpage for clear and free. | |
1611 | */ | |
1612 | oldpage = newpage; | |
1613 | } else { | |
6a93ca8f | 1614 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1615 | lru_cache_add_anon(newpage); |
1616 | *pagep = newpage; | |
1617 | } | |
bde05d1c HD |
1618 | |
1619 | ClearPageSwapCache(oldpage); | |
1620 | set_page_private(oldpage, 0); | |
1621 | ||
1622 | unlock_page(oldpage); | |
09cbfeaf KS |
1623 | put_page(oldpage); |
1624 | put_page(oldpage); | |
0142ef6c | 1625 | return error; |
bde05d1c HD |
1626 | } |
1627 | ||
c5bf121e VRP |
1628 | /* |
1629 | * Swap in the page pointed to by *pagep. | |
1630 | * Caller has to make sure that *pagep contains a valid swapped page. | |
1631 | * Returns 0 and the page in pagep if success. On failure, returns the | |
1632 | * the error code and NULL in *pagep. | |
1633 | */ | |
1634 | static int shmem_swapin_page(struct inode *inode, pgoff_t index, | |
1635 | struct page **pagep, enum sgp_type sgp, | |
1636 | gfp_t gfp, struct vm_area_struct *vma, | |
1637 | vm_fault_t *fault_type) | |
1638 | { | |
1639 | struct address_space *mapping = inode->i_mapping; | |
1640 | struct shmem_inode_info *info = SHMEM_I(inode); | |
1641 | struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm; | |
1642 | struct mem_cgroup *memcg; | |
1643 | struct page *page; | |
1644 | swp_entry_t swap; | |
1645 | int error; | |
1646 | ||
1647 | VM_BUG_ON(!*pagep || !xa_is_value(*pagep)); | |
1648 | swap = radix_to_swp_entry(*pagep); | |
1649 | *pagep = NULL; | |
1650 | ||
1651 | /* Look it up and read it in.. */ | |
1652 | page = lookup_swap_cache(swap, NULL, 0); | |
1653 | if (!page) { | |
1654 | /* Or update major stats only when swapin succeeds?? */ | |
1655 | if (fault_type) { | |
1656 | *fault_type |= VM_FAULT_MAJOR; | |
1657 | count_vm_event(PGMAJFAULT); | |
1658 | count_memcg_event_mm(charge_mm, PGMAJFAULT); | |
1659 | } | |
1660 | /* Here we actually start the io */ | |
1661 | page = shmem_swapin(swap, gfp, info, index); | |
1662 | if (!page) { | |
1663 | error = -ENOMEM; | |
1664 | goto failed; | |
1665 | } | |
1666 | } | |
1667 | ||
1668 | /* We have to do this with page locked to prevent races */ | |
1669 | lock_page(page); | |
1670 | if (!PageSwapCache(page) || page_private(page) != swap.val || | |
1671 | !shmem_confirm_swap(mapping, index, swap)) { | |
1672 | error = -EEXIST; | |
1673 | goto unlock; | |
1674 | } | |
1675 | if (!PageUptodate(page)) { | |
1676 | error = -EIO; | |
1677 | goto failed; | |
1678 | } | |
1679 | wait_on_page_writeback(page); | |
1680 | ||
1681 | if (shmem_should_replace_page(page, gfp)) { | |
1682 | error = shmem_replace_page(&page, gfp, info, index); | |
1683 | if (error) | |
1684 | goto failed; | |
1685 | } | |
1686 | ||
1687 | error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg, | |
1688 | false); | |
1689 | if (!error) { | |
1690 | error = shmem_add_to_page_cache(page, mapping, index, | |
1691 | swp_to_radix_entry(swap), gfp); | |
1692 | /* | |
1693 | * We already confirmed swap under page lock, and make | |
1694 | * no memory allocation here, so usually no possibility | |
1695 | * of error; but free_swap_and_cache() only trylocks a | |
1696 | * page, so it is just possible that the entry has been | |
1697 | * truncated or holepunched since swap was confirmed. | |
1698 | * shmem_undo_range() will have done some of the | |
1699 | * unaccounting, now delete_from_swap_cache() will do | |
1700 | * the rest. | |
1701 | */ | |
1702 | if (error) { | |
1703 | mem_cgroup_cancel_charge(page, memcg, false); | |
1704 | delete_from_swap_cache(page); | |
1705 | } | |
1706 | } | |
1707 | if (error) | |
1708 | goto failed; | |
1709 | ||
1710 | mem_cgroup_commit_charge(page, memcg, true, false); | |
1711 | ||
1712 | spin_lock_irq(&info->lock); | |
1713 | info->swapped--; | |
1714 | shmem_recalc_inode(inode); | |
1715 | spin_unlock_irq(&info->lock); | |
1716 | ||
1717 | if (sgp == SGP_WRITE) | |
1718 | mark_page_accessed(page); | |
1719 | ||
1720 | delete_from_swap_cache(page); | |
1721 | set_page_dirty(page); | |
1722 | swap_free(swap); | |
1723 | ||
1724 | *pagep = page; | |
1725 | return 0; | |
1726 | failed: | |
1727 | if (!shmem_confirm_swap(mapping, index, swap)) | |
1728 | error = -EEXIST; | |
1729 | unlock: | |
1730 | if (page) { | |
1731 | unlock_page(page); | |
1732 | put_page(page); | |
1733 | } | |
1734 | ||
1735 | return error; | |
1736 | } | |
1737 | ||
1da177e4 | 1738 | /* |
68da9f05 | 1739 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1740 | * |
1741 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1742 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1743 | * entry since a page cannot live in both the swap and page cache. |
1744 | * | |
28eb3c80 | 1745 | * vmf and fault_type are only supplied by shmem_fault: |
9e18eb29 | 1746 | * otherwise they are NULL. |
1da177e4 | 1747 | */ |
41ffe5d5 | 1748 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 1749 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
2b740303 SJ |
1750 | struct vm_area_struct *vma, struct vm_fault *vmf, |
1751 | vm_fault_t *fault_type) | |
1da177e4 LT |
1752 | { |
1753 | struct address_space *mapping = inode->i_mapping; | |
23f919d4 | 1754 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 | 1755 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1756 | struct mm_struct *charge_mm; |
00501b53 | 1757 | struct mem_cgroup *memcg; |
27ab7006 | 1758 | struct page *page; |
657e3038 | 1759 | enum sgp_type sgp_huge = sgp; |
800d8c63 | 1760 | pgoff_t hindex = index; |
1da177e4 | 1761 | int error; |
54af6042 | 1762 | int once = 0; |
1635f6a7 | 1763 | int alloced = 0; |
1da177e4 | 1764 | |
09cbfeaf | 1765 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1766 | return -EFBIG; |
657e3038 KS |
1767 | if (sgp == SGP_NOHUGE || sgp == SGP_HUGE) |
1768 | sgp = SGP_CACHE; | |
1da177e4 | 1769 | repeat: |
c5bf121e VRP |
1770 | if (sgp <= SGP_CACHE && |
1771 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { | |
1772 | return -EINVAL; | |
1773 | } | |
1774 | ||
1775 | sbinfo = SHMEM_SB(inode->i_sb); | |
1776 | charge_mm = vma ? vma->vm_mm : current->mm; | |
1777 | ||
0cd6144a | 1778 | page = find_lock_entry(mapping, index); |
3159f943 | 1779 | if (xa_is_value(page)) { |
c5bf121e VRP |
1780 | error = shmem_swapin_page(inode, index, &page, |
1781 | sgp, gfp, vma, fault_type); | |
1782 | if (error == -EEXIST) | |
1783 | goto repeat; | |
54af6042 | 1784 | |
c5bf121e VRP |
1785 | *pagep = page; |
1786 | return error; | |
54af6042 HD |
1787 | } |
1788 | ||
66d2f4d2 HD |
1789 | if (page && sgp == SGP_WRITE) |
1790 | mark_page_accessed(page); | |
1791 | ||
1635f6a7 HD |
1792 | /* fallocated page? */ |
1793 | if (page && !PageUptodate(page)) { | |
1794 | if (sgp != SGP_READ) | |
1795 | goto clear; | |
1796 | unlock_page(page); | |
09cbfeaf | 1797 | put_page(page); |
1635f6a7 HD |
1798 | page = NULL; |
1799 | } | |
c5bf121e | 1800 | if (page || sgp == SGP_READ) { |
54af6042 HD |
1801 | *pagep = page; |
1802 | return 0; | |
27ab7006 HD |
1803 | } |
1804 | ||
1805 | /* | |
54af6042 HD |
1806 | * Fast cache lookup did not find it: |
1807 | * bring it back from swap or allocate. | |
27ab7006 | 1808 | */ |
54af6042 | 1809 | |
c5bf121e VRP |
1810 | if (vma && userfaultfd_missing(vma)) { |
1811 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
1812 | return 0; | |
1813 | } | |
cfda0526 | 1814 | |
c5bf121e VRP |
1815 | /* shmem_symlink() */ |
1816 | if (mapping->a_ops != &shmem_aops) | |
1817 | goto alloc_nohuge; | |
1818 | if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE) | |
1819 | goto alloc_nohuge; | |
1820 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1821 | goto alloc_huge; | |
1822 | switch (sbinfo->huge) { | |
1823 | loff_t i_size; | |
1824 | pgoff_t off; | |
1825 | case SHMEM_HUGE_NEVER: | |
1826 | goto alloc_nohuge; | |
1827 | case SHMEM_HUGE_WITHIN_SIZE: | |
1828 | off = round_up(index, HPAGE_PMD_NR); | |
1829 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1830 | if (i_size >= HPAGE_PMD_SIZE && | |
1831 | i_size >> PAGE_SHIFT >= off) | |
800d8c63 | 1832 | goto alloc_huge; |
c5bf121e VRP |
1833 | /* fallthrough */ |
1834 | case SHMEM_HUGE_ADVISE: | |
1835 | if (sgp_huge == SGP_HUGE) | |
1836 | goto alloc_huge; | |
1837 | /* TODO: implement fadvise() hints */ | |
1838 | goto alloc_nohuge; | |
1839 | } | |
1da177e4 | 1840 | |
800d8c63 | 1841 | alloc_huge: |
c5bf121e VRP |
1842 | page = shmem_alloc_and_acct_page(gfp, inode, index, true); |
1843 | if (IS_ERR(page)) { | |
1844 | alloc_nohuge: | |
1845 | page = shmem_alloc_and_acct_page(gfp, inode, | |
1846 | index, false); | |
1847 | } | |
1848 | if (IS_ERR(page)) { | |
1849 | int retry = 5; | |
800d8c63 | 1850 | |
c5bf121e VRP |
1851 | error = PTR_ERR(page); |
1852 | page = NULL; | |
1853 | if (error != -ENOSPC) | |
1854 | goto unlock; | |
1855 | /* | |
1856 | * Try to reclaim some space by splitting a huge page | |
1857 | * beyond i_size on the filesystem. | |
1858 | */ | |
1859 | while (retry--) { | |
1860 | int ret; | |
66d2f4d2 | 1861 | |
c5bf121e VRP |
1862 | ret = shmem_unused_huge_shrink(sbinfo, NULL, 1); |
1863 | if (ret == SHRINK_STOP) | |
1864 | break; | |
1865 | if (ret) | |
1866 | goto alloc_nohuge; | |
b065b432 | 1867 | } |
c5bf121e VRP |
1868 | goto unlock; |
1869 | } | |
54af6042 | 1870 | |
c5bf121e VRP |
1871 | if (PageTransHuge(page)) |
1872 | hindex = round_down(index, HPAGE_PMD_NR); | |
1873 | else | |
1874 | hindex = index; | |
54af6042 | 1875 | |
c5bf121e VRP |
1876 | if (sgp == SGP_WRITE) |
1877 | __SetPageReferenced(page); | |
1878 | ||
1879 | error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg, | |
1880 | PageTransHuge(page)); | |
1881 | if (error) | |
1882 | goto unacct; | |
1883 | error = shmem_add_to_page_cache(page, mapping, hindex, | |
1884 | NULL, gfp & GFP_RECLAIM_MASK); | |
1885 | if (error) { | |
1886 | mem_cgroup_cancel_charge(page, memcg, | |
1887 | PageTransHuge(page)); | |
1888 | goto unacct; | |
1889 | } | |
1890 | mem_cgroup_commit_charge(page, memcg, false, | |
1891 | PageTransHuge(page)); | |
1892 | lru_cache_add_anon(page); | |
779750d2 | 1893 | |
c5bf121e | 1894 | spin_lock_irq(&info->lock); |
d8c6546b | 1895 | info->alloced += compound_nr(page); |
c5bf121e VRP |
1896 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); |
1897 | shmem_recalc_inode(inode); | |
1898 | spin_unlock_irq(&info->lock); | |
1899 | alloced = true; | |
1900 | ||
1901 | if (PageTransHuge(page) && | |
1902 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < | |
1903 | hindex + HPAGE_PMD_NR - 1) { | |
ec9516fb | 1904 | /* |
c5bf121e VRP |
1905 | * Part of the huge page is beyond i_size: subject |
1906 | * to shrink under memory pressure. | |
1635f6a7 | 1907 | */ |
c5bf121e | 1908 | spin_lock(&sbinfo->shrinklist_lock); |
1635f6a7 | 1909 | /* |
c5bf121e VRP |
1910 | * _careful to defend against unlocked access to |
1911 | * ->shrink_list in shmem_unused_huge_shrink() | |
ec9516fb | 1912 | */ |
c5bf121e VRP |
1913 | if (list_empty_careful(&info->shrinklist)) { |
1914 | list_add_tail(&info->shrinklist, | |
1915 | &sbinfo->shrinklist); | |
1916 | sbinfo->shrinklist_len++; | |
1917 | } | |
1918 | spin_unlock(&sbinfo->shrinklist_lock); | |
1919 | } | |
800d8c63 | 1920 | |
c5bf121e VRP |
1921 | /* |
1922 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. | |
1923 | */ | |
1924 | if (sgp == SGP_FALLOC) | |
1925 | sgp = SGP_WRITE; | |
1926 | clear: | |
1927 | /* | |
1928 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1929 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1930 | * it now, lest undo on failure cancel our earlier guarantee. | |
1931 | */ | |
1932 | if (sgp != SGP_WRITE && !PageUptodate(page)) { | |
1933 | struct page *head = compound_head(page); | |
1934 | int i; | |
1935 | ||
d8c6546b | 1936 | for (i = 0; i < compound_nr(head); i++) { |
c5bf121e VRP |
1937 | clear_highpage(head + i); |
1938 | flush_dcache_page(head + i); | |
ec9516fb | 1939 | } |
c5bf121e | 1940 | SetPageUptodate(head); |
1da177e4 | 1941 | } |
bde05d1c | 1942 | |
54af6042 | 1943 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1944 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1945 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1946 | if (alloced) { |
1947 | ClearPageDirty(page); | |
1948 | delete_from_page_cache(page); | |
4595ef88 | 1949 | spin_lock_irq(&info->lock); |
267a4c76 | 1950 | shmem_recalc_inode(inode); |
4595ef88 | 1951 | spin_unlock_irq(&info->lock); |
267a4c76 | 1952 | } |
54af6042 | 1953 | error = -EINVAL; |
267a4c76 | 1954 | goto unlock; |
e83c32e8 | 1955 | } |
800d8c63 | 1956 | *pagep = page + index - hindex; |
54af6042 | 1957 | return 0; |
1da177e4 | 1958 | |
59a16ead | 1959 | /* |
54af6042 | 1960 | * Error recovery. |
59a16ead | 1961 | */ |
54af6042 | 1962 | unacct: |
d8c6546b | 1963 | shmem_inode_unacct_blocks(inode, compound_nr(page)); |
800d8c63 KS |
1964 | |
1965 | if (PageTransHuge(page)) { | |
1966 | unlock_page(page); | |
1967 | put_page(page); | |
1968 | goto alloc_nohuge; | |
1969 | } | |
d1899228 | 1970 | unlock: |
27ab7006 | 1971 | if (page) { |
54af6042 | 1972 | unlock_page(page); |
09cbfeaf | 1973 | put_page(page); |
54af6042 HD |
1974 | } |
1975 | if (error == -ENOSPC && !once++) { | |
4595ef88 | 1976 | spin_lock_irq(&info->lock); |
54af6042 | 1977 | shmem_recalc_inode(inode); |
4595ef88 | 1978 | spin_unlock_irq(&info->lock); |
27ab7006 | 1979 | goto repeat; |
ff36b801 | 1980 | } |
7f4446ee | 1981 | if (error == -EEXIST) |
54af6042 HD |
1982 | goto repeat; |
1983 | return error; | |
1da177e4 LT |
1984 | } |
1985 | ||
10d20bd2 LT |
1986 | /* |
1987 | * This is like autoremove_wake_function, but it removes the wait queue | |
1988 | * entry unconditionally - even if something else had already woken the | |
1989 | * target. | |
1990 | */ | |
ac6424b9 | 1991 | static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
10d20bd2 LT |
1992 | { |
1993 | int ret = default_wake_function(wait, mode, sync, key); | |
2055da97 | 1994 | list_del_init(&wait->entry); |
10d20bd2 LT |
1995 | return ret; |
1996 | } | |
1997 | ||
20acce67 | 1998 | static vm_fault_t shmem_fault(struct vm_fault *vmf) |
1da177e4 | 1999 | { |
11bac800 | 2000 | struct vm_area_struct *vma = vmf->vma; |
496ad9aa | 2001 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 2002 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
657e3038 | 2003 | enum sgp_type sgp; |
20acce67 SJ |
2004 | int err; |
2005 | vm_fault_t ret = VM_FAULT_LOCKED; | |
1da177e4 | 2006 | |
f00cdc6d HD |
2007 | /* |
2008 | * Trinity finds that probing a hole which tmpfs is punching can | |
2009 | * prevent the hole-punch from ever completing: which in turn | |
2010 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
2011 | * faulting pages into the hole while it's being punched. Although |
2012 | * shmem_undo_range() does remove the additions, it may be unable to | |
2013 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
2014 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
2015 | * | |
2016 | * It does not matter if we sometimes reach this check just before the | |
2017 | * hole-punch begins, so that one fault then races with the punch: | |
2018 | * we just need to make racing faults a rare case. | |
2019 | * | |
2020 | * The implementation below would be much simpler if we just used a | |
2021 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
2022 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
2023 | */ |
2024 | if (unlikely(inode->i_private)) { | |
2025 | struct shmem_falloc *shmem_falloc; | |
2026 | ||
2027 | spin_lock(&inode->i_lock); | |
2028 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
2029 | if (shmem_falloc && |
2030 | shmem_falloc->waitq && | |
2031 | vmf->pgoff >= shmem_falloc->start && | |
2032 | vmf->pgoff < shmem_falloc->next) { | |
82e4461f | 2033 | struct file *fpin; |
8e205f77 | 2034 | wait_queue_head_t *shmem_falloc_waitq; |
10d20bd2 | 2035 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); |
8e205f77 HD |
2036 | |
2037 | ret = VM_FAULT_NOPAGE; | |
82e4461f KS |
2038 | fpin = maybe_unlock_mmap_for_io(vmf, NULL); |
2039 | if (fpin) | |
8e205f77 | 2040 | ret = VM_FAULT_RETRY; |
8e205f77 HD |
2041 | |
2042 | shmem_falloc_waitq = shmem_falloc->waitq; | |
2043 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
2044 | TASK_UNINTERRUPTIBLE); | |
2045 | spin_unlock(&inode->i_lock); | |
2046 | schedule(); | |
2047 | ||
2048 | /* | |
2049 | * shmem_falloc_waitq points into the shmem_fallocate() | |
2050 | * stack of the hole-punching task: shmem_falloc_waitq | |
2051 | * is usually invalid by the time we reach here, but | |
2052 | * finish_wait() does not dereference it in that case; | |
2053 | * though i_lock needed lest racing with wake_up_all(). | |
2054 | */ | |
2055 | spin_lock(&inode->i_lock); | |
2056 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
2057 | spin_unlock(&inode->i_lock); | |
82e4461f KS |
2058 | |
2059 | if (fpin) | |
2060 | fput(fpin); | |
8e205f77 | 2061 | return ret; |
f00cdc6d | 2062 | } |
8e205f77 | 2063 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
2064 | } |
2065 | ||
657e3038 | 2066 | sgp = SGP_CACHE; |
18600332 MH |
2067 | |
2068 | if ((vma->vm_flags & VM_NOHUGEPAGE) || | |
2069 | test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) | |
657e3038 | 2070 | sgp = SGP_NOHUGE; |
18600332 MH |
2071 | else if (vma->vm_flags & VM_HUGEPAGE) |
2072 | sgp = SGP_HUGE; | |
657e3038 | 2073 | |
20acce67 | 2074 | err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp, |
cfda0526 | 2075 | gfp, vma, vmf, &ret); |
20acce67 SJ |
2076 | if (err) |
2077 | return vmf_error(err); | |
68da9f05 | 2078 | return ret; |
1da177e4 LT |
2079 | } |
2080 | ||
c01d5b30 HD |
2081 | unsigned long shmem_get_unmapped_area(struct file *file, |
2082 | unsigned long uaddr, unsigned long len, | |
2083 | unsigned long pgoff, unsigned long flags) | |
2084 | { | |
2085 | unsigned long (*get_area)(struct file *, | |
2086 | unsigned long, unsigned long, unsigned long, unsigned long); | |
2087 | unsigned long addr; | |
2088 | unsigned long offset; | |
2089 | unsigned long inflated_len; | |
2090 | unsigned long inflated_addr; | |
2091 | unsigned long inflated_offset; | |
2092 | ||
2093 | if (len > TASK_SIZE) | |
2094 | return -ENOMEM; | |
2095 | ||
2096 | get_area = current->mm->get_unmapped_area; | |
2097 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2098 | ||
e496cf3d | 2099 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
c01d5b30 HD |
2100 | return addr; |
2101 | if (IS_ERR_VALUE(addr)) | |
2102 | return addr; | |
2103 | if (addr & ~PAGE_MASK) | |
2104 | return addr; | |
2105 | if (addr > TASK_SIZE - len) | |
2106 | return addr; | |
2107 | ||
2108 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2109 | return addr; | |
2110 | if (len < HPAGE_PMD_SIZE) | |
2111 | return addr; | |
2112 | if (flags & MAP_FIXED) | |
2113 | return addr; | |
2114 | /* | |
2115 | * Our priority is to support MAP_SHARED mapped hugely; | |
2116 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
3917fed7 KS |
2117 | * But if caller specified an address hint and we allocated area there |
2118 | * successfully, respect that as before. | |
c01d5b30 | 2119 | */ |
3917fed7 | 2120 | if (uaddr == addr) |
c01d5b30 HD |
2121 | return addr; |
2122 | ||
2123 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2124 | struct super_block *sb; | |
2125 | ||
2126 | if (file) { | |
2127 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2128 | sb = file_inode(file)->i_sb; | |
2129 | } else { | |
2130 | /* | |
2131 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2132 | * for "/dev/zero", to create a shared anonymous object. | |
2133 | */ | |
2134 | if (IS_ERR(shm_mnt)) | |
2135 | return addr; | |
2136 | sb = shm_mnt->mnt_sb; | |
2137 | } | |
3089bf61 | 2138 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2139 | return addr; |
2140 | } | |
2141 | ||
2142 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2143 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2144 | return addr; | |
2145 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2146 | return addr; | |
2147 | ||
2148 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2149 | if (inflated_len > TASK_SIZE) | |
2150 | return addr; | |
2151 | if (inflated_len < len) | |
2152 | return addr; | |
2153 | ||
3917fed7 | 2154 | inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags); |
c01d5b30 HD |
2155 | if (IS_ERR_VALUE(inflated_addr)) |
2156 | return addr; | |
2157 | if (inflated_addr & ~PAGE_MASK) | |
2158 | return addr; | |
2159 | ||
2160 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2161 | inflated_addr += offset - inflated_offset; | |
2162 | if (inflated_offset > offset) | |
2163 | inflated_addr += HPAGE_PMD_SIZE; | |
2164 | ||
2165 | if (inflated_addr > TASK_SIZE - len) | |
2166 | return addr; | |
2167 | return inflated_addr; | |
2168 | } | |
2169 | ||
1da177e4 | 2170 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2171 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2172 | { |
496ad9aa | 2173 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2174 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2175 | } |
2176 | ||
d8dc74f2 AB |
2177 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
2178 | unsigned long addr) | |
1da177e4 | 2179 | { |
496ad9aa | 2180 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2181 | pgoff_t index; |
1da177e4 | 2182 | |
41ffe5d5 HD |
2183 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2184 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
2185 | } |
2186 | #endif | |
2187 | ||
2188 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
2189 | { | |
496ad9aa | 2190 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2191 | struct shmem_inode_info *info = SHMEM_I(inode); |
2192 | int retval = -ENOMEM; | |
2193 | ||
7c5c644e HD |
2194 | /* |
2195 | * What serializes the accesses to info->flags? | |
2196 | * ipc_lock_object() when called from shmctl_do_lock(), | |
2197 | * no serialization needed when called from shm_destroy(). | |
2198 | */ | |
1da177e4 LT |
2199 | if (lock && !(info->flags & VM_LOCKED)) { |
2200 | if (!user_shm_lock(inode->i_size, user)) | |
2201 | goto out_nomem; | |
2202 | info->flags |= VM_LOCKED; | |
89e004ea | 2203 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
2204 | } |
2205 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
2206 | user_shm_unlock(inode->i_size, user); | |
2207 | info->flags &= ~VM_LOCKED; | |
89e004ea | 2208 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2209 | } |
2210 | retval = 0; | |
89e004ea | 2211 | |
1da177e4 | 2212 | out_nomem: |
1da177e4 LT |
2213 | return retval; |
2214 | } | |
2215 | ||
9b83a6a8 | 2216 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 | 2217 | { |
ab3948f5 | 2218 | struct shmem_inode_info *info = SHMEM_I(file_inode(file)); |
1331ba55 | 2219 | int ret; |
ab3948f5 | 2220 | |
1331ba55 PX |
2221 | ret = seal_check_future_write(info->seals, vma); |
2222 | if (ret) | |
2223 | return ret; | |
ab3948f5 | 2224 | |
1da177e4 LT |
2225 | file_accessed(file); |
2226 | vma->vm_ops = &shmem_vm_ops; | |
e496cf3d | 2227 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
2228 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
2229 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
2230 | khugepaged_enter(vma, vma->vm_flags); | |
2231 | } | |
1da177e4 LT |
2232 | return 0; |
2233 | } | |
2234 | ||
454abafe | 2235 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 2236 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
2237 | { |
2238 | struct inode *inode; | |
2239 | struct shmem_inode_info *info; | |
2240 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
a3a49a17 | 2241 | int ino; |
1da177e4 | 2242 | |
5b04c689 PE |
2243 | if (shmem_reserve_inode(sb)) |
2244 | return NULL; | |
1da177e4 LT |
2245 | |
2246 | inode = new_inode(sb); | |
2247 | if (inode) { | |
454abafe | 2248 | inode_init_owner(inode, dir, mode); |
1da177e4 | 2249 | inode->i_blocks = 0; |
078cd827 | 2250 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
46c9a946 | 2251 | inode->i_generation = prandom_u32(); |
1da177e4 LT |
2252 | info = SHMEM_I(inode); |
2253 | memset(info, 0, (char *)inode - (char *)info); | |
2254 | spin_lock_init(&info->lock); | |
af53d3e9 | 2255 | atomic_set(&info->stop_eviction, 0); |
40e041a2 | 2256 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 2257 | info->flags = flags & VM_NORESERVE; |
779750d2 | 2258 | INIT_LIST_HEAD(&info->shrinklist); |
1da177e4 | 2259 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 2260 | simple_xattrs_init(&info->xattrs); |
72c04902 | 2261 | cache_no_acl(inode); |
1da177e4 LT |
2262 | |
2263 | switch (mode & S_IFMT) { | |
2264 | default: | |
39f0247d | 2265 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
2266 | init_special_inode(inode, mode, dev); |
2267 | break; | |
2268 | case S_IFREG: | |
14fcc23f | 2269 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
2270 | inode->i_op = &shmem_inode_operations; |
2271 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
2272 | mpol_shared_policy_init(&info->policy, |
2273 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
2274 | break; |
2275 | case S_IFDIR: | |
d8c76e6f | 2276 | inc_nlink(inode); |
1da177e4 LT |
2277 | /* Some things misbehave if size == 0 on a directory */ |
2278 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2279 | inode->i_op = &shmem_dir_inode_operations; | |
2280 | inode->i_fop = &simple_dir_operations; | |
2281 | break; | |
2282 | case S_IFLNK: | |
2283 | /* | |
2284 | * Must not load anything in the rbtree, | |
2285 | * mpol_free_shared_policy will not be called. | |
2286 | */ | |
71fe804b | 2287 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
2288 | break; |
2289 | } | |
b45d71fb | 2290 | |
a3a49a17 SF |
2291 | if (!sbinfo->idr_nouse) { |
2292 | /* inum 0 and 1 are unused */ | |
2293 | mutex_lock(&sbinfo->idr_lock); | |
2294 | ino = idr_alloc(&sbinfo->idr, inode, 2, INT_MAX, | |
2295 | GFP_NOFS); | |
2296 | if (ino > 0) { | |
2297 | inode->i_ino = ino; | |
2298 | mutex_unlock(&sbinfo->idr_lock); | |
2299 | __insert_inode_hash(inode, inode->i_ino); | |
2300 | } else { | |
2301 | inode->i_ino = 0; | |
2302 | mutex_unlock(&sbinfo->idr_lock); | |
2303 | iput(inode); | |
2304 | /* shmem_free_inode() will be called */ | |
2305 | inode = NULL; | |
2306 | } | |
2307 | } else | |
2308 | inode->i_ino = get_next_ino(); | |
2309 | ||
b45d71fb | 2310 | lockdep_annotate_inode_mutex_key(inode); |
5b04c689 PE |
2311 | } else |
2312 | shmem_free_inode(sb); | |
1da177e4 LT |
2313 | return inode; |
2314 | } | |
2315 | ||
0cd6144a JW |
2316 | bool shmem_mapping(struct address_space *mapping) |
2317 | { | |
f8005451 | 2318 | return mapping->a_ops == &shmem_aops; |
0cd6144a JW |
2319 | } |
2320 | ||
8d103963 MR |
2321 | static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm, |
2322 | pmd_t *dst_pmd, | |
2323 | struct vm_area_struct *dst_vma, | |
2324 | unsigned long dst_addr, | |
2325 | unsigned long src_addr, | |
2326 | bool zeropage, | |
2327 | struct page **pagep) | |
4c27fe4c MR |
2328 | { |
2329 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2330 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4c27fe4c MR |
2331 | struct address_space *mapping = inode->i_mapping; |
2332 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2333 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
2334 | struct mem_cgroup *memcg; | |
2335 | spinlock_t *ptl; | |
2336 | void *page_kaddr; | |
2337 | struct page *page; | |
2338 | pte_t _dst_pte, *dst_pte; | |
2339 | int ret; | |
e2a50c1f | 2340 | pgoff_t offset, max_off; |
4c27fe4c | 2341 | |
cb658a45 | 2342 | ret = -ENOMEM; |
3f05932a AR |
2343 | if (!shmem_inode_acct_block(inode, 1)) { |
2344 | /* | |
2345 | * We may have got a page, returned -ENOENT triggering a retry, | |
2346 | * and now we find ourselves with -ENOMEM. Release the page, to | |
2347 | * avoid a BUG_ON in our caller. | |
2348 | */ | |
2349 | if (unlikely(*pagep)) { | |
2350 | put_page(*pagep); | |
2351 | *pagep = NULL; | |
2352 | } | |
cb658a45 | 2353 | goto out; |
3f05932a | 2354 | } |
4c27fe4c | 2355 | |
cb658a45 | 2356 | if (!*pagep) { |
4c27fe4c MR |
2357 | page = shmem_alloc_page(gfp, info, pgoff); |
2358 | if (!page) | |
0f079694 | 2359 | goto out_unacct_blocks; |
4c27fe4c | 2360 | |
8d103963 MR |
2361 | if (!zeropage) { /* mcopy_atomic */ |
2362 | page_kaddr = kmap_atomic(page); | |
2363 | ret = copy_from_user(page_kaddr, | |
2364 | (const void __user *)src_addr, | |
2365 | PAGE_SIZE); | |
2366 | kunmap_atomic(page_kaddr); | |
2367 | ||
2368 | /* fallback to copy_from_user outside mmap_sem */ | |
2369 | if (unlikely(ret)) { | |
2370 | *pagep = page; | |
2371 | shmem_inode_unacct_blocks(inode, 1); | |
2372 | /* don't free the page */ | |
9e368259 | 2373 | return -ENOENT; |
8d103963 MR |
2374 | } |
2375 | } else { /* mfill_zeropage_atomic */ | |
2376 | clear_highpage(page); | |
4c27fe4c MR |
2377 | } |
2378 | } else { | |
2379 | page = *pagep; | |
2380 | *pagep = NULL; | |
2381 | } | |
2382 | ||
9cc90c66 AA |
2383 | VM_BUG_ON(PageLocked(page) || PageSwapBacked(page)); |
2384 | __SetPageLocked(page); | |
2385 | __SetPageSwapBacked(page); | |
a425d358 | 2386 | __SetPageUptodate(page); |
9cc90c66 | 2387 | |
e2a50c1f AA |
2388 | ret = -EFAULT; |
2389 | offset = linear_page_index(dst_vma, dst_addr); | |
2390 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); | |
2391 | if (unlikely(offset >= max_off)) | |
2392 | goto out_release; | |
2393 | ||
2cf85583 | 2394 | ret = mem_cgroup_try_charge_delay(page, dst_mm, gfp, &memcg, false); |
4c27fe4c MR |
2395 | if (ret) |
2396 | goto out_release; | |
2397 | ||
552446a4 MW |
2398 | ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL, |
2399 | gfp & GFP_RECLAIM_MASK); | |
4c27fe4c MR |
2400 | if (ret) |
2401 | goto out_release_uncharge; | |
2402 | ||
2403 | mem_cgroup_commit_charge(page, memcg, false, false); | |
2404 | ||
2405 | _dst_pte = mk_pte(page, dst_vma->vm_page_prot); | |
2406 | if (dst_vma->vm_flags & VM_WRITE) | |
2407 | _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte)); | |
dcf7fe9d AA |
2408 | else { |
2409 | /* | |
2410 | * We don't set the pte dirty if the vma has no | |
2411 | * VM_WRITE permission, so mark the page dirty or it | |
2412 | * could be freed from under us. We could do it | |
2413 | * unconditionally before unlock_page(), but doing it | |
2414 | * only if VM_WRITE is not set is faster. | |
2415 | */ | |
2416 | set_page_dirty(page); | |
2417 | } | |
4c27fe4c | 2418 | |
4c27fe4c | 2419 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); |
e2a50c1f AA |
2420 | |
2421 | ret = -EFAULT; | |
2422 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); | |
2423 | if (unlikely(offset >= max_off)) | |
2424 | goto out_release_uncharge_unlock; | |
2425 | ||
2426 | ret = -EEXIST; | |
4c27fe4c MR |
2427 | if (!pte_none(*dst_pte)) |
2428 | goto out_release_uncharge_unlock; | |
2429 | ||
4c27fe4c MR |
2430 | lru_cache_add_anon(page); |
2431 | ||
0f0ec00e | 2432 | spin_lock_irq(&info->lock); |
4c27fe4c MR |
2433 | info->alloced++; |
2434 | inode->i_blocks += BLOCKS_PER_PAGE; | |
2435 | shmem_recalc_inode(inode); | |
0f0ec00e | 2436 | spin_unlock_irq(&info->lock); |
4c27fe4c MR |
2437 | |
2438 | inc_mm_counter(dst_mm, mm_counter_file(page)); | |
2439 | page_add_file_rmap(page, false); | |
2440 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
2441 | ||
2442 | /* No need to invalidate - it was non-present before */ | |
2443 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
4c27fe4c | 2444 | pte_unmap_unlock(dst_pte, ptl); |
e2a50c1f | 2445 | unlock_page(page); |
4c27fe4c MR |
2446 | ret = 0; |
2447 | out: | |
2448 | return ret; | |
2449 | out_release_uncharge_unlock: | |
2450 | pte_unmap_unlock(dst_pte, ptl); | |
dcf7fe9d | 2451 | ClearPageDirty(page); |
e2a50c1f | 2452 | delete_from_page_cache(page); |
4c27fe4c MR |
2453 | out_release_uncharge: |
2454 | mem_cgroup_cancel_charge(page, memcg, false); | |
2455 | out_release: | |
9cc90c66 | 2456 | unlock_page(page); |
4c27fe4c | 2457 | put_page(page); |
4c27fe4c | 2458 | out_unacct_blocks: |
0f079694 | 2459 | shmem_inode_unacct_blocks(inode, 1); |
4c27fe4c MR |
2460 | goto out; |
2461 | } | |
2462 | ||
8d103963 MR |
2463 | int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm, |
2464 | pmd_t *dst_pmd, | |
2465 | struct vm_area_struct *dst_vma, | |
2466 | unsigned long dst_addr, | |
2467 | unsigned long src_addr, | |
2468 | struct page **pagep) | |
2469 | { | |
2470 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2471 | dst_addr, src_addr, false, pagep); | |
2472 | } | |
2473 | ||
2474 | int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm, | |
2475 | pmd_t *dst_pmd, | |
2476 | struct vm_area_struct *dst_vma, | |
2477 | unsigned long dst_addr) | |
2478 | { | |
2479 | struct page *page = NULL; | |
2480 | ||
2481 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2482 | dst_addr, 0, true, &page); | |
2483 | } | |
2484 | ||
1da177e4 | 2485 | #ifdef CONFIG_TMPFS |
92e1d5be | 2486 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2487 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2488 | |
6d9d88d0 JS |
2489 | #ifdef CONFIG_TMPFS_XATTR |
2490 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2491 | #else | |
2492 | #define shmem_initxattrs NULL | |
2493 | #endif | |
2494 | ||
1da177e4 | 2495 | static int |
800d15a5 NP |
2496 | shmem_write_begin(struct file *file, struct address_space *mapping, |
2497 | loff_t pos, unsigned len, unsigned flags, | |
2498 | struct page **pagep, void **fsdata) | |
1da177e4 | 2499 | { |
800d15a5 | 2500 | struct inode *inode = mapping->host; |
40e041a2 | 2501 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2502 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
2503 | |
2504 | /* i_mutex is held by caller */ | |
ab3948f5 JFG |
2505 | if (unlikely(info->seals & (F_SEAL_GROW | |
2506 | F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) { | |
2507 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) | |
40e041a2 DH |
2508 | return -EPERM; |
2509 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2510 | return -EPERM; | |
2511 | } | |
2512 | ||
9e18eb29 | 2513 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
2514 | } |
2515 | ||
2516 | static int | |
2517 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2518 | loff_t pos, unsigned len, unsigned copied, | |
2519 | struct page *page, void *fsdata) | |
2520 | { | |
2521 | struct inode *inode = mapping->host; | |
2522 | ||
d3602444 HD |
2523 | if (pos + copied > inode->i_size) |
2524 | i_size_write(inode, pos + copied); | |
2525 | ||
ec9516fb | 2526 | if (!PageUptodate(page)) { |
800d8c63 KS |
2527 | struct page *head = compound_head(page); |
2528 | if (PageTransCompound(page)) { | |
2529 | int i; | |
2530 | ||
2531 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2532 | if (head + i == page) | |
2533 | continue; | |
2534 | clear_highpage(head + i); | |
2535 | flush_dcache_page(head + i); | |
2536 | } | |
2537 | } | |
09cbfeaf KS |
2538 | if (copied < PAGE_SIZE) { |
2539 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2540 | zero_user_segments(page, 0, from, |
09cbfeaf | 2541 | from + copied, PAGE_SIZE); |
ec9516fb | 2542 | } |
800d8c63 | 2543 | SetPageUptodate(head); |
ec9516fb | 2544 | } |
800d15a5 | 2545 | set_page_dirty(page); |
6746aff7 | 2546 | unlock_page(page); |
09cbfeaf | 2547 | put_page(page); |
800d15a5 | 2548 | |
800d15a5 | 2549 | return copied; |
1da177e4 LT |
2550 | } |
2551 | ||
2ba5bbed | 2552 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2553 | { |
6e58e79d AV |
2554 | struct file *file = iocb->ki_filp; |
2555 | struct inode *inode = file_inode(file); | |
1da177e4 | 2556 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2557 | pgoff_t index; |
2558 | unsigned long offset; | |
a0ee5ec5 | 2559 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2560 | int error = 0; |
cb66a7a1 | 2561 | ssize_t retval = 0; |
6e58e79d | 2562 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2563 | |
2564 | /* | |
2565 | * Might this read be for a stacking filesystem? Then when reading | |
2566 | * holes of a sparse file, we actually need to allocate those pages, | |
2567 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2568 | */ | |
777eda2c | 2569 | if (!iter_is_iovec(to)) |
75edd345 | 2570 | sgp = SGP_CACHE; |
1da177e4 | 2571 | |
09cbfeaf KS |
2572 | index = *ppos >> PAGE_SHIFT; |
2573 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2574 | |
2575 | for (;;) { | |
2576 | struct page *page = NULL; | |
41ffe5d5 HD |
2577 | pgoff_t end_index; |
2578 | unsigned long nr, ret; | |
1da177e4 LT |
2579 | loff_t i_size = i_size_read(inode); |
2580 | ||
09cbfeaf | 2581 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2582 | if (index > end_index) |
2583 | break; | |
2584 | if (index == end_index) { | |
09cbfeaf | 2585 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2586 | if (nr <= offset) |
2587 | break; | |
2588 | } | |
2589 | ||
9e18eb29 | 2590 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2591 | if (error) { |
2592 | if (error == -EINVAL) | |
2593 | error = 0; | |
1da177e4 LT |
2594 | break; |
2595 | } | |
75edd345 HD |
2596 | if (page) { |
2597 | if (sgp == SGP_CACHE) | |
2598 | set_page_dirty(page); | |
d3602444 | 2599 | unlock_page(page); |
75edd345 | 2600 | } |
1da177e4 LT |
2601 | |
2602 | /* | |
2603 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2604 | * are called without i_mutex protection against truncate |
1da177e4 | 2605 | */ |
09cbfeaf | 2606 | nr = PAGE_SIZE; |
1da177e4 | 2607 | i_size = i_size_read(inode); |
09cbfeaf | 2608 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2609 | if (index == end_index) { |
09cbfeaf | 2610 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2611 | if (nr <= offset) { |
2612 | if (page) | |
09cbfeaf | 2613 | put_page(page); |
1da177e4 LT |
2614 | break; |
2615 | } | |
2616 | } | |
2617 | nr -= offset; | |
2618 | ||
2619 | if (page) { | |
2620 | /* | |
2621 | * If users can be writing to this page using arbitrary | |
2622 | * virtual addresses, take care about potential aliasing | |
2623 | * before reading the page on the kernel side. | |
2624 | */ | |
2625 | if (mapping_writably_mapped(mapping)) | |
2626 | flush_dcache_page(page); | |
2627 | /* | |
2628 | * Mark the page accessed if we read the beginning. | |
2629 | */ | |
2630 | if (!offset) | |
2631 | mark_page_accessed(page); | |
b5810039 | 2632 | } else { |
1da177e4 | 2633 | page = ZERO_PAGE(0); |
09cbfeaf | 2634 | get_page(page); |
b5810039 | 2635 | } |
1da177e4 LT |
2636 | |
2637 | /* | |
2638 | * Ok, we have the page, and it's up-to-date, so | |
2639 | * now we can copy it to user space... | |
1da177e4 | 2640 | */ |
2ba5bbed | 2641 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2642 | retval += ret; |
1da177e4 | 2643 | offset += ret; |
09cbfeaf KS |
2644 | index += offset >> PAGE_SHIFT; |
2645 | offset &= ~PAGE_MASK; | |
1da177e4 | 2646 | |
09cbfeaf | 2647 | put_page(page); |
2ba5bbed | 2648 | if (!iov_iter_count(to)) |
1da177e4 | 2649 | break; |
6e58e79d AV |
2650 | if (ret < nr) { |
2651 | error = -EFAULT; | |
2652 | break; | |
2653 | } | |
1da177e4 LT |
2654 | cond_resched(); |
2655 | } | |
2656 | ||
09cbfeaf | 2657 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2658 | file_accessed(file); |
2659 | return retval ? retval : error; | |
1da177e4 LT |
2660 | } |
2661 | ||
220f2ac9 | 2662 | /* |
7f4446ee | 2663 | * llseek SEEK_DATA or SEEK_HOLE through the page cache. |
220f2ac9 HD |
2664 | */ |
2665 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2666 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2667 | { |
2668 | struct page *page; | |
2669 | struct pagevec pvec; | |
2670 | pgoff_t indices[PAGEVEC_SIZE]; | |
2671 | bool done = false; | |
2672 | int i; | |
2673 | ||
86679820 | 2674 | pagevec_init(&pvec); |
220f2ac9 HD |
2675 | pvec.nr = 1; /* start small: we may be there already */ |
2676 | while (!done) { | |
0cd6144a | 2677 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2678 | pvec.nr, pvec.pages, indices); |
2679 | if (!pvec.nr) { | |
965c8e59 | 2680 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2681 | index = end; |
2682 | break; | |
2683 | } | |
2684 | for (i = 0; i < pvec.nr; i++, index++) { | |
2685 | if (index < indices[i]) { | |
965c8e59 | 2686 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2687 | done = true; |
2688 | break; | |
2689 | } | |
2690 | index = indices[i]; | |
2691 | } | |
2692 | page = pvec.pages[i]; | |
3159f943 | 2693 | if (page && !xa_is_value(page)) { |
220f2ac9 HD |
2694 | if (!PageUptodate(page)) |
2695 | page = NULL; | |
2696 | } | |
2697 | if (index >= end || | |
965c8e59 AM |
2698 | (page && whence == SEEK_DATA) || |
2699 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2700 | done = true; |
2701 | break; | |
2702 | } | |
2703 | } | |
0cd6144a | 2704 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2705 | pagevec_release(&pvec); |
2706 | pvec.nr = PAGEVEC_SIZE; | |
2707 | cond_resched(); | |
2708 | } | |
2709 | return index; | |
2710 | } | |
2711 | ||
965c8e59 | 2712 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2713 | { |
2714 | struct address_space *mapping = file->f_mapping; | |
2715 | struct inode *inode = mapping->host; | |
2716 | pgoff_t start, end; | |
2717 | loff_t new_offset; | |
2718 | ||
965c8e59 AM |
2719 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2720 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2721 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2722 | inode_lock(inode); |
220f2ac9 HD |
2723 | /* We're holding i_mutex so we can access i_size directly */ |
2724 | ||
1a413646 | 2725 | if (offset < 0 || offset >= inode->i_size) |
220f2ac9 HD |
2726 | offset = -ENXIO; |
2727 | else { | |
09cbfeaf KS |
2728 | start = offset >> PAGE_SHIFT; |
2729 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2730 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2731 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2732 | if (new_offset > offset) { |
2733 | if (new_offset < inode->i_size) | |
2734 | offset = new_offset; | |
965c8e59 | 2735 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2736 | offset = -ENXIO; |
2737 | else | |
2738 | offset = inode->i_size; | |
2739 | } | |
2740 | } | |
2741 | ||
387aae6f HD |
2742 | if (offset >= 0) |
2743 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2744 | inode_unlock(inode); |
220f2ac9 HD |
2745 | return offset; |
2746 | } | |
2747 | ||
83e4fa9c HD |
2748 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2749 | loff_t len) | |
2750 | { | |
496ad9aa | 2751 | struct inode *inode = file_inode(file); |
e2d12e22 | 2752 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2753 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2754 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2755 | pgoff_t start, index, end; |
2756 | int error; | |
83e4fa9c | 2757 | |
13ace4d0 HD |
2758 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2759 | return -EOPNOTSUPP; | |
2760 | ||
5955102c | 2761 | inode_lock(inode); |
83e4fa9c HD |
2762 | |
2763 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2764 | struct address_space *mapping = file->f_mapping; | |
2765 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2766 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2767 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2768 | |
40e041a2 | 2769 | /* protected by i_mutex */ |
ab3948f5 | 2770 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { |
40e041a2 DH |
2771 | error = -EPERM; |
2772 | goto out; | |
2773 | } | |
2774 | ||
8e205f77 | 2775 | shmem_falloc.waitq = &shmem_falloc_waitq; |
91da400f | 2776 | shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT; |
f00cdc6d HD |
2777 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; |
2778 | spin_lock(&inode->i_lock); | |
2779 | inode->i_private = &shmem_falloc; | |
2780 | spin_unlock(&inode->i_lock); | |
2781 | ||
83e4fa9c HD |
2782 | if ((u64)unmap_end > (u64)unmap_start) |
2783 | unmap_mapping_range(mapping, unmap_start, | |
2784 | 1 + unmap_end - unmap_start, 0); | |
2785 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2786 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2787 | |
2788 | spin_lock(&inode->i_lock); | |
2789 | inode->i_private = NULL; | |
2790 | wake_up_all(&shmem_falloc_waitq); | |
2055da97 | 2791 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head)); |
8e205f77 | 2792 | spin_unlock(&inode->i_lock); |
83e4fa9c | 2793 | error = 0; |
8e205f77 | 2794 | goto out; |
e2d12e22 HD |
2795 | } |
2796 | ||
2797 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2798 | error = inode_newsize_ok(inode, offset + len); | |
2799 | if (error) | |
2800 | goto out; | |
2801 | ||
40e041a2 DH |
2802 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2803 | error = -EPERM; | |
2804 | goto out; | |
2805 | } | |
2806 | ||
09cbfeaf KS |
2807 | start = offset >> PAGE_SHIFT; |
2808 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2809 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2810 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2811 | error = -ENOSPC; | |
2812 | goto out; | |
83e4fa9c HD |
2813 | } |
2814 | ||
8e205f77 | 2815 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2816 | shmem_falloc.start = start; |
2817 | shmem_falloc.next = start; | |
2818 | shmem_falloc.nr_falloced = 0; | |
2819 | shmem_falloc.nr_unswapped = 0; | |
2820 | spin_lock(&inode->i_lock); | |
2821 | inode->i_private = &shmem_falloc; | |
2822 | spin_unlock(&inode->i_lock); | |
2823 | ||
e2d12e22 HD |
2824 | for (index = start; index < end; index++) { |
2825 | struct page *page; | |
2826 | ||
2827 | /* | |
2828 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2829 | * been interrupted because we are using up too much memory. | |
2830 | */ | |
2831 | if (signal_pending(current)) | |
2832 | error = -EINTR; | |
1aac1400 HD |
2833 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2834 | error = -ENOMEM; | |
e2d12e22 | 2835 | else |
9e18eb29 | 2836 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2837 | if (error) { |
1635f6a7 | 2838 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2839 | if (index > start) { |
2840 | shmem_undo_range(inode, | |
2841 | (loff_t)start << PAGE_SHIFT, | |
2842 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2843 | } | |
1aac1400 | 2844 | goto undone; |
e2d12e22 HD |
2845 | } |
2846 | ||
1aac1400 HD |
2847 | /* |
2848 | * Inform shmem_writepage() how far we have reached. | |
2849 | * No need for lock or barrier: we have the page lock. | |
2850 | */ | |
2851 | shmem_falloc.next++; | |
2852 | if (!PageUptodate(page)) | |
2853 | shmem_falloc.nr_falloced++; | |
2854 | ||
e2d12e22 | 2855 | /* |
1635f6a7 HD |
2856 | * If !PageUptodate, leave it that way so that freeable pages |
2857 | * can be recognized if we need to rollback on error later. | |
2858 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2859 | * than free the pages we are allocating (and SGP_CACHE pages |
2860 | * might still be clean: we now need to mark those dirty too). | |
2861 | */ | |
2862 | set_page_dirty(page); | |
2863 | unlock_page(page); | |
09cbfeaf | 2864 | put_page(page); |
e2d12e22 HD |
2865 | cond_resched(); |
2866 | } | |
2867 | ||
2868 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2869 | i_size_write(inode, offset + len); | |
078cd827 | 2870 | inode->i_ctime = current_time(inode); |
1aac1400 HD |
2871 | undone: |
2872 | spin_lock(&inode->i_lock); | |
2873 | inode->i_private = NULL; | |
2874 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2875 | out: |
5955102c | 2876 | inode_unlock(inode); |
83e4fa9c HD |
2877 | return error; |
2878 | } | |
2879 | ||
726c3342 | 2880 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2881 | { |
726c3342 | 2882 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2883 | |
2884 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2885 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2886 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2887 | if (sbinfo->max_blocks) { |
1da177e4 | 2888 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2889 | buf->f_bavail = |
2890 | buf->f_bfree = sbinfo->max_blocks - | |
2891 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2892 | } |
2893 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2894 | buf->f_files = sbinfo->max_inodes; |
2895 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2896 | } |
2897 | /* else leave those fields 0 like simple_statfs */ | |
2898 | return 0; | |
2899 | } | |
2900 | ||
2901 | /* | |
2902 | * File creation. Allocate an inode, and we're done.. | |
2903 | */ | |
2904 | static int | |
1a67aafb | 2905 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2906 | { |
0b0a0806 | 2907 | struct inode *inode; |
1da177e4 LT |
2908 | int error = -ENOSPC; |
2909 | ||
454abafe | 2910 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2911 | if (inode) { |
feda821e CH |
2912 | error = simple_acl_create(dir, inode); |
2913 | if (error) | |
2914 | goto out_iput; | |
2a7dba39 | 2915 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2916 | &dentry->d_name, |
6d9d88d0 | 2917 | shmem_initxattrs, NULL); |
feda821e CH |
2918 | if (error && error != -EOPNOTSUPP) |
2919 | goto out_iput; | |
37ec43cd | 2920 | |
718deb6b | 2921 | error = 0; |
1da177e4 | 2922 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 2923 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
2924 | d_instantiate(dentry, inode); |
2925 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2926 | } |
2927 | return error; | |
feda821e CH |
2928 | out_iput: |
2929 | iput(inode); | |
2930 | return error; | |
1da177e4 LT |
2931 | } |
2932 | ||
60545d0d AV |
2933 | static int |
2934 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2935 | { | |
2936 | struct inode *inode; | |
2937 | int error = -ENOSPC; | |
2938 | ||
2939 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
2940 | if (inode) { | |
2941 | error = security_inode_init_security(inode, dir, | |
2942 | NULL, | |
2943 | shmem_initxattrs, NULL); | |
feda821e CH |
2944 | if (error && error != -EOPNOTSUPP) |
2945 | goto out_iput; | |
2946 | error = simple_acl_create(dir, inode); | |
2947 | if (error) | |
2948 | goto out_iput; | |
60545d0d AV |
2949 | d_tmpfile(dentry, inode); |
2950 | } | |
2951 | return error; | |
feda821e CH |
2952 | out_iput: |
2953 | iput(inode); | |
2954 | return error; | |
60545d0d AV |
2955 | } |
2956 | ||
18bb1db3 | 2957 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2958 | { |
2959 | int error; | |
2960 | ||
2961 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
2962 | return error; | |
d8c76e6f | 2963 | inc_nlink(dir); |
1da177e4 LT |
2964 | return 0; |
2965 | } | |
2966 | ||
4acdaf27 | 2967 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2968 | bool excl) |
1da177e4 LT |
2969 | { |
2970 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
2971 | } | |
2972 | ||
2973 | /* | |
2974 | * Link a file.. | |
2975 | */ | |
2976 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2977 | { | |
75c3cfa8 | 2978 | struct inode *inode = d_inode(old_dentry); |
29b00e60 | 2979 | int ret = 0; |
1da177e4 LT |
2980 | |
2981 | /* | |
2982 | * No ordinary (disk based) filesystem counts links as inodes; | |
2983 | * but each new link needs a new dentry, pinning lowmem, and | |
2984 | * tmpfs dentries cannot be pruned until they are unlinked. | |
1062af92 DW |
2985 | * But if an O_TMPFILE file is linked into the tmpfs, the |
2986 | * first link must skip that, to get the accounting right. | |
1da177e4 | 2987 | */ |
1062af92 DW |
2988 | if (inode->i_nlink) { |
2989 | ret = shmem_reserve_inode(inode->i_sb); | |
2990 | if (ret) | |
2991 | goto out; | |
2992 | } | |
1da177e4 LT |
2993 | |
2994 | dir->i_size += BOGO_DIRENT_SIZE; | |
078cd827 | 2995 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
d8c76e6f | 2996 | inc_nlink(inode); |
7de9c6ee | 2997 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
2998 | dget(dentry); /* Extra pinning count for the created dentry */ |
2999 | d_instantiate(dentry, inode); | |
5b04c689 PE |
3000 | out: |
3001 | return ret; | |
1da177e4 LT |
3002 | } |
3003 | ||
3004 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
3005 | { | |
75c3cfa8 | 3006 | struct inode *inode = d_inode(dentry); |
1da177e4 | 3007 | |
5b04c689 PE |
3008 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
3009 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
3010 | |
3011 | dir->i_size -= BOGO_DIRENT_SIZE; | |
078cd827 | 3012 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
9a53c3a7 | 3013 | drop_nlink(inode); |
1da177e4 LT |
3014 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
3015 | return 0; | |
3016 | } | |
3017 | ||
3018 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
3019 | { | |
3020 | if (!simple_empty(dentry)) | |
3021 | return -ENOTEMPTY; | |
3022 | ||
75c3cfa8 | 3023 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 3024 | drop_nlink(dir); |
1da177e4 LT |
3025 | return shmem_unlink(dir, dentry); |
3026 | } | |
3027 | ||
37456771 MS |
3028 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
3029 | { | |
e36cb0b8 DH |
3030 | bool old_is_dir = d_is_dir(old_dentry); |
3031 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
3032 | |
3033 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
3034 | if (old_is_dir) { | |
3035 | drop_nlink(old_dir); | |
3036 | inc_nlink(new_dir); | |
3037 | } else { | |
3038 | drop_nlink(new_dir); | |
3039 | inc_nlink(old_dir); | |
3040 | } | |
3041 | } | |
3042 | old_dir->i_ctime = old_dir->i_mtime = | |
3043 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 | 3044 | d_inode(old_dentry)->i_ctime = |
078cd827 | 3045 | d_inode(new_dentry)->i_ctime = current_time(old_dir); |
37456771 MS |
3046 | |
3047 | return 0; | |
3048 | } | |
3049 | ||
46fdb794 MS |
3050 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
3051 | { | |
3052 | struct dentry *whiteout; | |
3053 | int error; | |
3054 | ||
3055 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
3056 | if (!whiteout) | |
3057 | return -ENOMEM; | |
3058 | ||
3059 | error = shmem_mknod(old_dir, whiteout, | |
3060 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
3061 | dput(whiteout); | |
3062 | if (error) | |
3063 | return error; | |
3064 | ||
3065 | /* | |
3066 | * Cheat and hash the whiteout while the old dentry is still in | |
3067 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
3068 | * | |
3069 | * d_lookup() will consistently find one of them at this point, | |
3070 | * not sure which one, but that isn't even important. | |
3071 | */ | |
3072 | d_rehash(whiteout); | |
3073 | return 0; | |
3074 | } | |
3075 | ||
1da177e4 LT |
3076 | /* |
3077 | * The VFS layer already does all the dentry stuff for rename, | |
3078 | * we just have to decrement the usage count for the target if | |
3079 | * it exists so that the VFS layer correctly free's it when it | |
3080 | * gets overwritten. | |
3081 | */ | |
3b69ff51 | 3082 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 3083 | { |
75c3cfa8 | 3084 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
3085 | int they_are_dirs = S_ISDIR(inode->i_mode); |
3086 | ||
46fdb794 | 3087 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
3088 | return -EINVAL; |
3089 | ||
37456771 MS |
3090 | if (flags & RENAME_EXCHANGE) |
3091 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
3092 | ||
1da177e4 LT |
3093 | if (!simple_empty(new_dentry)) |
3094 | return -ENOTEMPTY; | |
3095 | ||
46fdb794 MS |
3096 | if (flags & RENAME_WHITEOUT) { |
3097 | int error; | |
3098 | ||
3099 | error = shmem_whiteout(old_dir, old_dentry); | |
3100 | if (error) | |
3101 | return error; | |
3102 | } | |
3103 | ||
75c3cfa8 | 3104 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 3105 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 3106 | if (they_are_dirs) { |
75c3cfa8 | 3107 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 3108 | drop_nlink(old_dir); |
b928095b | 3109 | } |
1da177e4 | 3110 | } else if (they_are_dirs) { |
9a53c3a7 | 3111 | drop_nlink(old_dir); |
d8c76e6f | 3112 | inc_nlink(new_dir); |
1da177e4 LT |
3113 | } |
3114 | ||
3115 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
3116 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
3117 | old_dir->i_ctime = old_dir->i_mtime = | |
3118 | new_dir->i_ctime = new_dir->i_mtime = | |
078cd827 | 3119 | inode->i_ctime = current_time(old_dir); |
1da177e4 LT |
3120 | return 0; |
3121 | } | |
3122 | ||
3123 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
3124 | { | |
3125 | int error; | |
3126 | int len; | |
3127 | struct inode *inode; | |
9276aad6 | 3128 | struct page *page; |
1da177e4 LT |
3129 | |
3130 | len = strlen(symname) + 1; | |
09cbfeaf | 3131 | if (len > PAGE_SIZE) |
1da177e4 LT |
3132 | return -ENAMETOOLONG; |
3133 | ||
0825a6f9 JP |
3134 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0, |
3135 | VM_NORESERVE); | |
1da177e4 LT |
3136 | if (!inode) |
3137 | return -ENOSPC; | |
3138 | ||
9d8f13ba | 3139 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3140 | shmem_initxattrs, NULL); |
570bc1c2 SS |
3141 | if (error) { |
3142 | if (error != -EOPNOTSUPP) { | |
3143 | iput(inode); | |
3144 | return error; | |
3145 | } | |
3146 | error = 0; | |
3147 | } | |
3148 | ||
1da177e4 | 3149 | inode->i_size = len-1; |
69f07ec9 | 3150 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3151 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3152 | if (!inode->i_link) { | |
69f07ec9 HD |
3153 | iput(inode); |
3154 | return -ENOMEM; | |
3155 | } | |
3156 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3157 | } else { |
e8ecde25 | 3158 | inode_nohighmem(inode); |
9e18eb29 | 3159 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
3160 | if (error) { |
3161 | iput(inode); | |
3162 | return error; | |
3163 | } | |
14fcc23f | 3164 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 3165 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 3166 | memcpy(page_address(page), symname, len); |
ec9516fb | 3167 | SetPageUptodate(page); |
1da177e4 | 3168 | set_page_dirty(page); |
6746aff7 | 3169 | unlock_page(page); |
09cbfeaf | 3170 | put_page(page); |
1da177e4 | 3171 | } |
1da177e4 | 3172 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3173 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3174 | d_instantiate(dentry, inode); |
3175 | dget(dentry); | |
3176 | return 0; | |
3177 | } | |
3178 | ||
fceef393 | 3179 | static void shmem_put_link(void *arg) |
1da177e4 | 3180 | { |
fceef393 AV |
3181 | mark_page_accessed(arg); |
3182 | put_page(arg); | |
1da177e4 LT |
3183 | } |
3184 | ||
6b255391 | 3185 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
3186 | struct inode *inode, |
3187 | struct delayed_call *done) | |
1da177e4 | 3188 | { |
1da177e4 | 3189 | struct page *page = NULL; |
6b255391 | 3190 | int error; |
6a6c9904 AV |
3191 | if (!dentry) { |
3192 | page = find_get_page(inode->i_mapping, 0); | |
3193 | if (!page) | |
3194 | return ERR_PTR(-ECHILD); | |
3195 | if (!PageUptodate(page)) { | |
3196 | put_page(page); | |
3197 | return ERR_PTR(-ECHILD); | |
3198 | } | |
3199 | } else { | |
9e18eb29 | 3200 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3201 | if (error) |
3202 | return ERR_PTR(error); | |
3203 | unlock_page(page); | |
3204 | } | |
fceef393 | 3205 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3206 | return page_address(page); |
1da177e4 LT |
3207 | } |
3208 | ||
b09e0fa4 | 3209 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3210 | /* |
b09e0fa4 EP |
3211 | * Superblocks without xattr inode operations may get some security.* xattr |
3212 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3213 | * like ACLs, we also need to implement the security.* handlers at |
3214 | * filesystem level, though. | |
3215 | */ | |
3216 | ||
6d9d88d0 JS |
3217 | /* |
3218 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3219 | */ | |
3220 | static int shmem_initxattrs(struct inode *inode, | |
3221 | const struct xattr *xattr_array, | |
3222 | void *fs_info) | |
3223 | { | |
3224 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3225 | const struct xattr *xattr; | |
38f38657 | 3226 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3227 | size_t len; |
3228 | ||
3229 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3230 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3231 | if (!new_xattr) |
3232 | return -ENOMEM; | |
3233 | ||
3234 | len = strlen(xattr->name) + 1; | |
3235 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3236 | GFP_KERNEL); | |
3237 | if (!new_xattr->name) { | |
3238 | kfree(new_xattr); | |
3239 | return -ENOMEM; | |
3240 | } | |
3241 | ||
3242 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3243 | XATTR_SECURITY_PREFIX_LEN); | |
3244 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3245 | xattr->name, len); | |
3246 | ||
38f38657 | 3247 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3248 | } |
3249 | ||
3250 | return 0; | |
3251 | } | |
3252 | ||
aa7c5241 | 3253 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3254 | struct dentry *unused, struct inode *inode, |
3255 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3256 | { |
b296821a | 3257 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3258 | |
aa7c5241 | 3259 | name = xattr_full_name(handler, name); |
38f38657 | 3260 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3261 | } |
3262 | ||
aa7c5241 | 3263 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
59301226 AV |
3264 | struct dentry *unused, struct inode *inode, |
3265 | const char *name, const void *value, | |
3266 | size_t size, int flags) | |
b09e0fa4 | 3267 | { |
59301226 | 3268 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3269 | |
aa7c5241 | 3270 | name = xattr_full_name(handler, name); |
38f38657 | 3271 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
3272 | } |
3273 | ||
aa7c5241 AG |
3274 | static const struct xattr_handler shmem_security_xattr_handler = { |
3275 | .prefix = XATTR_SECURITY_PREFIX, | |
3276 | .get = shmem_xattr_handler_get, | |
3277 | .set = shmem_xattr_handler_set, | |
3278 | }; | |
b09e0fa4 | 3279 | |
aa7c5241 AG |
3280 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3281 | .prefix = XATTR_TRUSTED_PREFIX, | |
3282 | .get = shmem_xattr_handler_get, | |
3283 | .set = shmem_xattr_handler_set, | |
3284 | }; | |
b09e0fa4 | 3285 | |
aa7c5241 AG |
3286 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3287 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3288 | &posix_acl_access_xattr_handler, | |
3289 | &posix_acl_default_xattr_handler, | |
3290 | #endif | |
3291 | &shmem_security_xattr_handler, | |
3292 | &shmem_trusted_xattr_handler, | |
3293 | NULL | |
3294 | }; | |
b09e0fa4 EP |
3295 | |
3296 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3297 | { | |
75c3cfa8 | 3298 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3299 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3300 | } |
3301 | #endif /* CONFIG_TMPFS_XATTR */ | |
3302 | ||
69f07ec9 | 3303 | static const struct inode_operations shmem_short_symlink_operations = { |
6b255391 | 3304 | .get_link = simple_get_link, |
b09e0fa4 | 3305 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3306 | .listxattr = shmem_listxattr, |
b09e0fa4 EP |
3307 | #endif |
3308 | }; | |
3309 | ||
3310 | static const struct inode_operations shmem_symlink_inode_operations = { | |
6b255391 | 3311 | .get_link = shmem_get_link, |
b09e0fa4 | 3312 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3313 | .listxattr = shmem_listxattr, |
39f0247d | 3314 | #endif |
b09e0fa4 | 3315 | }; |
39f0247d | 3316 | |
91828a40 DG |
3317 | static struct dentry *shmem_get_parent(struct dentry *child) |
3318 | { | |
3319 | return ERR_PTR(-ESTALE); | |
3320 | } | |
3321 | ||
3322 | static int shmem_match(struct inode *ino, void *vfh) | |
3323 | { | |
3324 | __u32 *fh = vfh; | |
a3a49a17 | 3325 | __u64 inum = fh[1]; |
91828a40 DG |
3326 | return ino->i_ino == inum && fh[0] == ino->i_generation; |
3327 | } | |
3328 | ||
12ba780d AG |
3329 | /* Find any alias of inode, but prefer a hashed alias */ |
3330 | static struct dentry *shmem_find_alias(struct inode *inode) | |
3331 | { | |
3332 | struct dentry *alias = d_find_alias(inode); | |
3333 | ||
3334 | return alias ?: d_find_any_alias(inode); | |
3335 | } | |
3336 | ||
3337 | ||
480b116c CH |
3338 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3339 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3340 | { |
91828a40 | 3341 | struct inode *inode; |
480b116c | 3342 | struct dentry *dentry = NULL; |
35c2a7f4 | 3343 | u64 inum; |
480b116c | 3344 | |
a3a49a17 | 3345 | if (fh_len < 2) |
480b116c | 3346 | return NULL; |
91828a40 | 3347 | |
a3a49a17 SF |
3348 | inum = fid->raw[1]; |
3349 | inode = ilookup5(sb, inum, shmem_match, fid->raw); | |
91828a40 | 3350 | if (inode) { |
12ba780d | 3351 | dentry = shmem_find_alias(inode); |
91828a40 DG |
3352 | iput(inode); |
3353 | } | |
3354 | ||
480b116c | 3355 | return dentry; |
91828a40 DG |
3356 | } |
3357 | ||
b0b0382b AV |
3358 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3359 | struct inode *parent) | |
91828a40 | 3360 | { |
a3a49a17 SF |
3361 | if (*len < 2) { |
3362 | *len = 2; | |
94e07a75 | 3363 | return FILEID_INVALID; |
5fe0c237 | 3364 | } |
91828a40 | 3365 | |
91828a40 DG |
3366 | fh[0] = inode->i_generation; |
3367 | fh[1] = inode->i_ino; | |
91828a40 | 3368 | |
a3a49a17 | 3369 | *len = 2; |
91828a40 DG |
3370 | return 1; |
3371 | } | |
3372 | ||
39655164 | 3373 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3374 | .get_parent = shmem_get_parent, |
91828a40 | 3375 | .encode_fh = shmem_encode_fh, |
480b116c | 3376 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3377 | }; |
3378 | ||
626c3920 AV |
3379 | enum shmem_param { |
3380 | Opt_gid, | |
3381 | Opt_huge, | |
3382 | Opt_mode, | |
3383 | Opt_mpol, | |
3384 | Opt_nr_blocks, | |
3385 | Opt_nr_inodes, | |
3386 | Opt_size, | |
3387 | Opt_uid, | |
3388 | }; | |
3389 | ||
3390 | static const struct fs_parameter_spec shmem_param_specs[] = { | |
3391 | fsparam_u32 ("gid", Opt_gid), | |
3392 | fsparam_enum ("huge", Opt_huge), | |
3393 | fsparam_u32oct("mode", Opt_mode), | |
3394 | fsparam_string("mpol", Opt_mpol), | |
3395 | fsparam_string("nr_blocks", Opt_nr_blocks), | |
3396 | fsparam_string("nr_inodes", Opt_nr_inodes), | |
3397 | fsparam_string("size", Opt_size), | |
3398 | fsparam_u32 ("uid", Opt_uid), | |
3399 | {} | |
3400 | }; | |
3401 | ||
3402 | static const struct fs_parameter_enum shmem_param_enums[] = { | |
3403 | { Opt_huge, "never", SHMEM_HUGE_NEVER }, | |
3404 | { Opt_huge, "always", SHMEM_HUGE_ALWAYS }, | |
3405 | { Opt_huge, "within_size", SHMEM_HUGE_WITHIN_SIZE }, | |
3406 | { Opt_huge, "advise", SHMEM_HUGE_ADVISE }, | |
3407 | {} | |
3408 | }; | |
3409 | ||
3410 | const struct fs_parameter_description shmem_fs_parameters = { | |
3411 | .name = "tmpfs", | |
3412 | .specs = shmem_param_specs, | |
3413 | .enums = shmem_param_enums, | |
3414 | }; | |
3415 | ||
f3235626 | 3416 | static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param) |
1da177e4 | 3417 | { |
f3235626 | 3418 | struct shmem_options *ctx = fc->fs_private; |
626c3920 AV |
3419 | struct fs_parse_result result; |
3420 | unsigned long long size; | |
e04dc423 | 3421 | char *rest; |
626c3920 AV |
3422 | int opt; |
3423 | ||
3424 | opt = fs_parse(fc, &shmem_fs_parameters, param, &result); | |
f3235626 | 3425 | if (opt < 0) |
626c3920 | 3426 | return opt; |
1da177e4 | 3427 | |
626c3920 AV |
3428 | switch (opt) { |
3429 | case Opt_size: | |
3430 | size = memparse(param->string, &rest); | |
e04dc423 AV |
3431 | if (*rest == '%') { |
3432 | size <<= PAGE_SHIFT; | |
3433 | size *= totalram_pages(); | |
3434 | do_div(size, 100); | |
3435 | rest++; | |
3436 | } | |
3437 | if (*rest) | |
626c3920 | 3438 | goto bad_value; |
e04dc423 AV |
3439 | ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE); |
3440 | ctx->seen |= SHMEM_SEEN_BLOCKS; | |
626c3920 AV |
3441 | break; |
3442 | case Opt_nr_blocks: | |
3443 | ctx->blocks = memparse(param->string, &rest); | |
e04dc423 | 3444 | if (*rest) |
626c3920 | 3445 | goto bad_value; |
e04dc423 | 3446 | ctx->seen |= SHMEM_SEEN_BLOCKS; |
626c3920 AV |
3447 | break; |
3448 | case Opt_nr_inodes: | |
3449 | ctx->inodes = memparse(param->string, &rest); | |
a3a49a17 | 3450 | if (*rest || ctx->inodes < 2) |
626c3920 | 3451 | goto bad_value; |
e04dc423 | 3452 | ctx->seen |= SHMEM_SEEN_INODES; |
626c3920 AV |
3453 | break; |
3454 | case Opt_mode: | |
3455 | ctx->mode = result.uint_32 & 07777; | |
3456 | break; | |
3457 | case Opt_uid: | |
3458 | ctx->uid = make_kuid(current_user_ns(), result.uint_32); | |
e04dc423 | 3459 | if (!uid_valid(ctx->uid)) |
626c3920 AV |
3460 | goto bad_value; |
3461 | break; | |
3462 | case Opt_gid: | |
3463 | ctx->gid = make_kgid(current_user_ns(), result.uint_32); | |
e04dc423 | 3464 | if (!gid_valid(ctx->gid)) |
626c3920 AV |
3465 | goto bad_value; |
3466 | break; | |
3467 | case Opt_huge: | |
3468 | ctx->huge = result.uint_32; | |
3469 | if (ctx->huge != SHMEM_HUGE_NEVER && | |
3470 | !(IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && | |
3471 | has_transparent_hugepage())) | |
3472 | goto unsupported_parameter; | |
e04dc423 | 3473 | ctx->seen |= SHMEM_SEEN_HUGE; |
626c3920 AV |
3474 | break; |
3475 | case Opt_mpol: | |
3476 | if (IS_ENABLED(CONFIG_NUMA)) { | |
3477 | mpol_put(ctx->mpol); | |
3478 | ctx->mpol = NULL; | |
3479 | if (mpol_parse_str(param->string, &ctx->mpol)) | |
3480 | goto bad_value; | |
3481 | break; | |
3482 | } | |
3483 | goto unsupported_parameter; | |
e04dc423 AV |
3484 | } |
3485 | return 0; | |
3486 | ||
626c3920 AV |
3487 | unsupported_parameter: |
3488 | return invalf(fc, "tmpfs: Unsupported parameter '%s'", param->key); | |
3489 | bad_value: | |
3490 | return invalf(fc, "tmpfs: Bad value for '%s'", param->key); | |
e04dc423 AV |
3491 | } |
3492 | ||
f3235626 | 3493 | static int shmem_parse_options(struct fs_context *fc, void *data) |
e04dc423 | 3494 | { |
f3235626 DH |
3495 | char *options = data; |
3496 | ||
33f37c64 AV |
3497 | if (options) { |
3498 | int err = security_sb_eat_lsm_opts(options, &fc->security); | |
3499 | if (err) | |
3500 | return err; | |
3501 | } | |
3502 | ||
b00dc3ad | 3503 | while (options != NULL) { |
626c3920 | 3504 | char *this_char = options; |
b00dc3ad HD |
3505 | for (;;) { |
3506 | /* | |
3507 | * NUL-terminate this option: unfortunately, | |
3508 | * mount options form a comma-separated list, | |
3509 | * but mpol's nodelist may also contain commas. | |
3510 | */ | |
3511 | options = strchr(options, ','); | |
3512 | if (options == NULL) | |
3513 | break; | |
3514 | options++; | |
3515 | if (!isdigit(*options)) { | |
3516 | options[-1] = '\0'; | |
3517 | break; | |
3518 | } | |
3519 | } | |
626c3920 AV |
3520 | if (*this_char) { |
3521 | char *value = strchr(this_char,'='); | |
f3235626 | 3522 | size_t len = 0; |
626c3920 AV |
3523 | int err; |
3524 | ||
3525 | if (value) { | |
3526 | *value++ = '\0'; | |
f3235626 | 3527 | len = strlen(value); |
626c3920 | 3528 | } |
f3235626 DH |
3529 | err = vfs_parse_fs_string(fc, this_char, value, len); |
3530 | if (err < 0) | |
3531 | return err; | |
1da177e4 | 3532 | } |
1da177e4 LT |
3533 | } |
3534 | return 0; | |
1da177e4 LT |
3535 | } |
3536 | ||
f3235626 DH |
3537 | /* |
3538 | * Reconfigure a shmem filesystem. | |
3539 | * | |
3540 | * Note that we disallow change from limited->unlimited blocks/inodes while any | |
3541 | * are in use; but we must separately disallow unlimited->limited, because in | |
3542 | * that case we have no record of how much is already in use. | |
3543 | */ | |
3544 | static int shmem_reconfigure(struct fs_context *fc) | |
1da177e4 | 3545 | { |
f3235626 DH |
3546 | struct shmem_options *ctx = fc->fs_private; |
3547 | struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb); | |
a3a49a17 | 3548 | int inodes; |
f3235626 | 3549 | const char *err; |
1da177e4 | 3550 | |
0edd73b3 | 3551 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3552 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
f3235626 DH |
3553 | if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) { |
3554 | if (!sbinfo->max_blocks) { | |
3555 | err = "Cannot retroactively limit size"; | |
0b5071dd | 3556 | goto out; |
f3235626 | 3557 | } |
0b5071dd | 3558 | if (percpu_counter_compare(&sbinfo->used_blocks, |
f3235626 DH |
3559 | ctx->blocks) > 0) { |
3560 | err = "Too small a size for current use"; | |
0b5071dd | 3561 | goto out; |
f3235626 | 3562 | } |
0b5071dd | 3563 | } |
f3235626 DH |
3564 | if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) { |
3565 | if (!sbinfo->max_inodes) { | |
3566 | err = "Cannot retroactively limit inodes"; | |
0b5071dd | 3567 | goto out; |
f3235626 DH |
3568 | } |
3569 | if (ctx->inodes < inodes) { | |
3570 | err = "Too few inodes for current use"; | |
0b5071dd | 3571 | goto out; |
f3235626 | 3572 | } |
0b5071dd | 3573 | } |
0edd73b3 | 3574 | |
f3235626 DH |
3575 | if (ctx->seen & SHMEM_SEEN_HUGE) |
3576 | sbinfo->huge = ctx->huge; | |
3577 | if (ctx->seen & SHMEM_SEEN_BLOCKS) | |
3578 | sbinfo->max_blocks = ctx->blocks; | |
3579 | if (ctx->seen & SHMEM_SEEN_INODES) { | |
3580 | sbinfo->max_inodes = ctx->inodes; | |
3581 | sbinfo->free_inodes = ctx->inodes - inodes; | |
0b5071dd | 3582 | } |
71fe804b | 3583 | |
5f00110f GT |
3584 | /* |
3585 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3586 | */ | |
f3235626 | 3587 | if (ctx->mpol) { |
5f00110f | 3588 | mpol_put(sbinfo->mpol); |
f3235626 DH |
3589 | sbinfo->mpol = ctx->mpol; /* transfers initial ref */ |
3590 | ctx->mpol = NULL; | |
5f00110f | 3591 | } |
f3235626 DH |
3592 | spin_unlock(&sbinfo->stat_lock); |
3593 | return 0; | |
0edd73b3 HD |
3594 | out: |
3595 | spin_unlock(&sbinfo->stat_lock); | |
f3235626 | 3596 | return invalf(fc, "tmpfs: %s", err); |
1da177e4 | 3597 | } |
680d794b | 3598 | |
34c80b1d | 3599 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3600 | { |
34c80b1d | 3601 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b AM |
3602 | |
3603 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3604 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3605 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 3606 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
a3a49a17 | 3607 | seq_printf(seq, ",nr_inodes=%d", sbinfo->max_inodes); |
0825a6f9 | 3608 | if (sbinfo->mode != (0777 | S_ISVTX)) |
09208d15 | 3609 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3610 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3611 | seq_printf(seq, ",uid=%u", | |
3612 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3613 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3614 | seq_printf(seq, ",gid=%u", | |
3615 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
e496cf3d | 3616 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3617 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
3618 | if (sbinfo->huge) | |
3619 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3620 | #endif | |
71fe804b | 3621 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b AM |
3622 | return 0; |
3623 | } | |
9183df25 | 3624 | |
680d794b | 3625 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3626 | |
3627 | static void shmem_put_super(struct super_block *sb) | |
3628 | { | |
602586a8 HD |
3629 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3630 | ||
a3a49a17 SF |
3631 | if (!sbinfo->idr_nouse) |
3632 | idr_destroy(&sbinfo->idr); | |
602586a8 | 3633 | percpu_counter_destroy(&sbinfo->used_blocks); |
49cd0a5c | 3634 | mpol_put(sbinfo->mpol); |
602586a8 | 3635 | kfree(sbinfo); |
1da177e4 LT |
3636 | sb->s_fs_info = NULL; |
3637 | } | |
3638 | ||
f3235626 | 3639 | static int shmem_fill_super(struct super_block *sb, struct fs_context *fc) |
1da177e4 | 3640 | { |
f3235626 | 3641 | struct shmem_options *ctx = fc->fs_private; |
1da177e4 | 3642 | struct inode *inode; |
0edd73b3 | 3643 | struct shmem_sb_info *sbinfo; |
680d794b AM |
3644 | int err = -ENOMEM; |
3645 | ||
3646 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3647 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b AM |
3648 | L1_CACHE_BYTES), GFP_KERNEL); |
3649 | if (!sbinfo) | |
3650 | return -ENOMEM; | |
3651 | ||
680d794b | 3652 | sb->s_fs_info = sbinfo; |
1da177e4 | 3653 | |
0edd73b3 | 3654 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3655 | /* |
3656 | * Per default we only allow half of the physical ram per | |
3657 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3658 | * but the internal instance is left unlimited. | |
3659 | */ | |
1751e8a6 | 3660 | if (!(sb->s_flags & SB_KERNMOUNT)) { |
f3235626 DH |
3661 | if (!(ctx->seen & SHMEM_SEEN_BLOCKS)) |
3662 | ctx->blocks = shmem_default_max_blocks(); | |
3663 | if (!(ctx->seen & SHMEM_SEEN_INODES)) | |
3664 | ctx->inodes = shmem_default_max_inodes(); | |
ca4e0519 | 3665 | } else { |
1751e8a6 | 3666 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 3667 | } |
91828a40 | 3668 | sb->s_export_op = &shmem_export_ops; |
1751e8a6 | 3669 | sb->s_flags |= SB_NOSEC; |
1da177e4 | 3670 | #else |
1751e8a6 | 3671 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 3672 | #endif |
e4a3f096 SF |
3673 | mutex_init(&sbinfo->idr_lock); |
3674 | idr_init(&sbinfo->idr); | |
f3235626 DH |
3675 | sbinfo->max_blocks = ctx->blocks; |
3676 | sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes; | |
3677 | sbinfo->uid = ctx->uid; | |
3678 | sbinfo->gid = ctx->gid; | |
3679 | sbinfo->mode = ctx->mode; | |
3680 | sbinfo->huge = ctx->huge; | |
3681 | sbinfo->mpol = ctx->mpol; | |
3682 | ctx->mpol = NULL; | |
1da177e4 | 3683 | |
0edd73b3 | 3684 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3685 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3686 | goto failed; |
779750d2 KS |
3687 | spin_lock_init(&sbinfo->shrinklist_lock); |
3688 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 3689 | |
285b2c4f | 3690 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3691 | sb->s_blocksize = PAGE_SIZE; |
3692 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3693 | sb->s_magic = TMPFS_MAGIC; |
3694 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3695 | sb->s_time_gran = 1; |
b09e0fa4 | 3696 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3697 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3698 | #endif |
3699 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
1751e8a6 | 3700 | sb->s_flags |= SB_POSIXACL; |
39f0247d | 3701 | #endif |
2b4db796 | 3702 | uuid_gen(&sb->s_uuid); |
0edd73b3 | 3703 | |
454abafe | 3704 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3705 | if (!inode) |
3706 | goto failed; | |
680d794b AM |
3707 | inode->i_uid = sbinfo->uid; |
3708 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3709 | sb->s_root = d_make_root(inode); |
3710 | if (!sb->s_root) | |
48fde701 | 3711 | goto failed; |
1da177e4 LT |
3712 | return 0; |
3713 | ||
1da177e4 LT |
3714 | failed: |
3715 | shmem_put_super(sb); | |
3716 | return err; | |
3717 | } | |
3718 | ||
f3235626 DH |
3719 | static int shmem_get_tree(struct fs_context *fc) |
3720 | { | |
3721 | return get_tree_nodev(fc, shmem_fill_super); | |
3722 | } | |
3723 | ||
3724 | static void shmem_free_fc(struct fs_context *fc) | |
3725 | { | |
3726 | struct shmem_options *ctx = fc->fs_private; | |
3727 | ||
3728 | if (ctx) { | |
3729 | mpol_put(ctx->mpol); | |
3730 | kfree(ctx); | |
3731 | } | |
3732 | } | |
3733 | ||
3734 | static const struct fs_context_operations shmem_fs_context_ops = { | |
3735 | .free = shmem_free_fc, | |
3736 | .get_tree = shmem_get_tree, | |
3737 | #ifdef CONFIG_TMPFS | |
3738 | .parse_monolithic = shmem_parse_options, | |
3739 | .parse_param = shmem_parse_one, | |
3740 | .reconfigure = shmem_reconfigure, | |
3741 | #endif | |
3742 | }; | |
3743 | ||
fcc234f8 | 3744 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3745 | |
3746 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3747 | { | |
41ffe5d5 HD |
3748 | struct shmem_inode_info *info; |
3749 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3750 | if (!info) | |
1da177e4 | 3751 | return NULL; |
41ffe5d5 | 3752 | return &info->vfs_inode; |
1da177e4 LT |
3753 | } |
3754 | ||
74b1da56 | 3755 | static void shmem_free_in_core_inode(struct inode *inode) |
fa0d7e3d | 3756 | { |
84e710da AV |
3757 | if (S_ISLNK(inode->i_mode)) |
3758 | kfree(inode->i_link); | |
fa0d7e3d NP |
3759 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3760 | } | |
3761 | ||
1da177e4 LT |
3762 | static void shmem_destroy_inode(struct inode *inode) |
3763 | { | |
09208d15 | 3764 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3765 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
1da177e4 LT |
3766 | } |
3767 | ||
41ffe5d5 | 3768 | static void shmem_init_inode(void *foo) |
1da177e4 | 3769 | { |
41ffe5d5 HD |
3770 | struct shmem_inode_info *info = foo; |
3771 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3772 | } |
3773 | ||
9a8ec03e | 3774 | static void shmem_init_inodecache(void) |
1da177e4 LT |
3775 | { |
3776 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3777 | sizeof(struct shmem_inode_info), | |
5d097056 | 3778 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3779 | } |
3780 | ||
41ffe5d5 | 3781 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3782 | { |
1a1d92c1 | 3783 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3784 | } |
3785 | ||
a3a49a17 SF |
3786 | static __init void shmem_no_idr(struct super_block *sb) |
3787 | { | |
3788 | struct shmem_sb_info *sbinfo; | |
3789 | ||
3790 | sbinfo = SHMEM_SB(sb); | |
3791 | sbinfo->idr_nouse = true; | |
3792 | idr_destroy(&sbinfo->idr); | |
3793 | } | |
3794 | ||
f5e54d6e | 3795 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3796 | .writepage = shmem_writepage, |
76719325 | 3797 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3798 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3799 | .write_begin = shmem_write_begin, |
3800 | .write_end = shmem_write_end, | |
1da177e4 | 3801 | #endif |
1c93923c | 3802 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3803 | .migratepage = migrate_page, |
1c93923c | 3804 | #endif |
aa261f54 | 3805 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3806 | }; |
3807 | ||
15ad7cdc | 3808 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3809 | .mmap = shmem_mmap, |
c01d5b30 | 3810 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3811 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3812 | .llseek = shmem_file_llseek, |
2ba5bbed | 3813 | .read_iter = shmem_file_read_iter, |
8174202b | 3814 | .write_iter = generic_file_write_iter, |
1b061d92 | 3815 | .fsync = noop_fsync, |
82c156f8 | 3816 | .splice_read = generic_file_splice_read, |
f6cb85d0 | 3817 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3818 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3819 | #endif |
3820 | }; | |
3821 | ||
92e1d5be | 3822 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3823 | .getattr = shmem_getattr, |
94c1e62d | 3824 | .setattr = shmem_setattr, |
b09e0fa4 | 3825 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3826 | .listxattr = shmem_listxattr, |
feda821e | 3827 | .set_acl = simple_set_acl, |
b09e0fa4 | 3828 | #endif |
1da177e4 LT |
3829 | }; |
3830 | ||
92e1d5be | 3831 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3832 | #ifdef CONFIG_TMPFS |
3833 | .create = shmem_create, | |
3834 | .lookup = simple_lookup, | |
3835 | .link = shmem_link, | |
3836 | .unlink = shmem_unlink, | |
3837 | .symlink = shmem_symlink, | |
3838 | .mkdir = shmem_mkdir, | |
3839 | .rmdir = shmem_rmdir, | |
3840 | .mknod = shmem_mknod, | |
2773bf00 | 3841 | .rename = shmem_rename2, |
60545d0d | 3842 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3843 | #endif |
b09e0fa4 | 3844 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3845 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3846 | #endif |
39f0247d | 3847 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3848 | .setattr = shmem_setattr, |
feda821e | 3849 | .set_acl = simple_set_acl, |
39f0247d AG |
3850 | #endif |
3851 | }; | |
3852 | ||
92e1d5be | 3853 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3854 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3855 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3856 | #endif |
39f0247d | 3857 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3858 | .setattr = shmem_setattr, |
feda821e | 3859 | .set_acl = simple_set_acl, |
39f0247d | 3860 | #endif |
1da177e4 LT |
3861 | }; |
3862 | ||
759b9775 | 3863 | static const struct super_operations shmem_ops = { |
1da177e4 | 3864 | .alloc_inode = shmem_alloc_inode, |
74b1da56 | 3865 | .free_inode = shmem_free_in_core_inode, |
1da177e4 LT |
3866 | .destroy_inode = shmem_destroy_inode, |
3867 | #ifdef CONFIG_TMPFS | |
3868 | .statfs = shmem_statfs, | |
680d794b | 3869 | .show_options = shmem_show_options, |
1da177e4 | 3870 | #endif |
1f895f75 | 3871 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3872 | .drop_inode = generic_delete_inode, |
3873 | .put_super = shmem_put_super, | |
779750d2 KS |
3874 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
3875 | .nr_cached_objects = shmem_unused_huge_count, | |
3876 | .free_cached_objects = shmem_unused_huge_scan, | |
3877 | #endif | |
1da177e4 LT |
3878 | }; |
3879 | ||
f0f37e2f | 3880 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3881 | .fault = shmem_fault, |
d7c17551 | 3882 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3883 | #ifdef CONFIG_NUMA |
3884 | .set_policy = shmem_set_policy, | |
3885 | .get_policy = shmem_get_policy, | |
3886 | #endif | |
3887 | }; | |
3888 | ||
f3235626 | 3889 | int shmem_init_fs_context(struct fs_context *fc) |
1da177e4 | 3890 | { |
f3235626 DH |
3891 | struct shmem_options *ctx; |
3892 | ||
3893 | ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL); | |
3894 | if (!ctx) | |
3895 | return -ENOMEM; | |
3896 | ||
3897 | ctx->mode = 0777 | S_ISVTX; | |
3898 | ctx->uid = current_fsuid(); | |
3899 | ctx->gid = current_fsgid(); | |
3900 | ||
3901 | fc->fs_private = ctx; | |
3902 | fc->ops = &shmem_fs_context_ops; | |
3903 | return 0; | |
1da177e4 LT |
3904 | } |
3905 | ||
41ffe5d5 | 3906 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3907 | .owner = THIS_MODULE, |
3908 | .name = "tmpfs", | |
f3235626 DH |
3909 | .init_fs_context = shmem_init_fs_context, |
3910 | #ifdef CONFIG_TMPFS | |
3911 | .parameters = &shmem_fs_parameters, | |
3912 | #endif | |
1da177e4 | 3913 | .kill_sb = kill_litter_super, |
2b8576cb | 3914 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 3915 | }; |
1da177e4 | 3916 | |
41ffe5d5 | 3917 | int __init shmem_init(void) |
1da177e4 LT |
3918 | { |
3919 | int error; | |
3920 | ||
9a8ec03e | 3921 | shmem_init_inodecache(); |
1da177e4 | 3922 | |
41ffe5d5 | 3923 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 3924 | if (error) { |
1170532b | 3925 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
3926 | goto out2; |
3927 | } | |
95dc112a | 3928 | |
ca4e0519 | 3929 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
3930 | if (IS_ERR(shm_mnt)) { |
3931 | error = PTR_ERR(shm_mnt); | |
1170532b | 3932 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
3933 | goto out1; |
3934 | } | |
a3a49a17 | 3935 | shmem_no_idr(shm_mnt->mnt_sb); |
5a6e75f8 | 3936 | |
e496cf3d | 3937 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
435c0b87 | 3938 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
3939 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
3940 | else | |
3941 | shmem_huge = 0; /* just in case it was patched */ | |
3942 | #endif | |
1da177e4 LT |
3943 | return 0; |
3944 | ||
3945 | out1: | |
41ffe5d5 | 3946 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 3947 | out2: |
41ffe5d5 | 3948 | shmem_destroy_inodecache(); |
1da177e4 LT |
3949 | shm_mnt = ERR_PTR(error); |
3950 | return error; | |
3951 | } | |
853ac43a | 3952 | |
e496cf3d | 3953 | #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS) |
5a6e75f8 KS |
3954 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
3955 | struct kobj_attribute *attr, char *buf) | |
3956 | { | |
3957 | int values[] = { | |
3958 | SHMEM_HUGE_ALWAYS, | |
3959 | SHMEM_HUGE_WITHIN_SIZE, | |
3960 | SHMEM_HUGE_ADVISE, | |
3961 | SHMEM_HUGE_NEVER, | |
3962 | SHMEM_HUGE_DENY, | |
3963 | SHMEM_HUGE_FORCE, | |
3964 | }; | |
3965 | int i, count; | |
3966 | ||
3967 | for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) { | |
3968 | const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s "; | |
3969 | ||
3970 | count += sprintf(buf + count, fmt, | |
3971 | shmem_format_huge(values[i])); | |
3972 | } | |
3973 | buf[count - 1] = '\n'; | |
3974 | return count; | |
3975 | } | |
3976 | ||
3977 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
3978 | struct kobj_attribute *attr, const char *buf, size_t count) | |
3979 | { | |
3980 | char tmp[16]; | |
3981 | int huge; | |
3982 | ||
3983 | if (count + 1 > sizeof(tmp)) | |
3984 | return -EINVAL; | |
3985 | memcpy(tmp, buf, count); | |
3986 | tmp[count] = '\0'; | |
3987 | if (count && tmp[count - 1] == '\n') | |
3988 | tmp[count - 1] = '\0'; | |
3989 | ||
3990 | huge = shmem_parse_huge(tmp); | |
3991 | if (huge == -EINVAL) | |
3992 | return -EINVAL; | |
3993 | if (!has_transparent_hugepage() && | |
3994 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
3995 | return -EINVAL; | |
3996 | ||
3997 | shmem_huge = huge; | |
435c0b87 | 3998 | if (shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
3999 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4000 | return count; | |
4001 | } | |
4002 | ||
4003 | struct kobj_attribute shmem_enabled_attr = | |
4004 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
3b33719c | 4005 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */ |
f3f0e1d2 | 4006 | |
3b33719c | 4007 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
f3f0e1d2 KS |
4008 | bool shmem_huge_enabled(struct vm_area_struct *vma) |
4009 | { | |
4010 | struct inode *inode = file_inode(vma->vm_file); | |
4011 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4012 | loff_t i_size; | |
4013 | pgoff_t off; | |
4014 | ||
c0630669 YS |
4015 | if ((vma->vm_flags & VM_NOHUGEPAGE) || |
4016 | test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) | |
4017 | return false; | |
f3f0e1d2 KS |
4018 | if (shmem_huge == SHMEM_HUGE_FORCE) |
4019 | return true; | |
4020 | if (shmem_huge == SHMEM_HUGE_DENY) | |
4021 | return false; | |
4022 | switch (sbinfo->huge) { | |
4023 | case SHMEM_HUGE_NEVER: | |
4024 | return false; | |
4025 | case SHMEM_HUGE_ALWAYS: | |
4026 | return true; | |
4027 | case SHMEM_HUGE_WITHIN_SIZE: | |
4028 | off = round_up(vma->vm_pgoff, HPAGE_PMD_NR); | |
4029 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
4030 | if (i_size >= HPAGE_PMD_SIZE && | |
4031 | i_size >> PAGE_SHIFT >= off) | |
4032 | return true; | |
c8402871 | 4033 | /* fall through */ |
f3f0e1d2 KS |
4034 | case SHMEM_HUGE_ADVISE: |
4035 | /* TODO: implement fadvise() hints */ | |
4036 | return (vma->vm_flags & VM_HUGEPAGE); | |
4037 | default: | |
4038 | VM_BUG_ON(1); | |
4039 | return false; | |
4040 | } | |
4041 | } | |
3b33719c | 4042 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 4043 | |
853ac43a MM |
4044 | #else /* !CONFIG_SHMEM */ |
4045 | ||
4046 | /* | |
4047 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
4048 | * | |
4049 | * This is intended for small system where the benefits of the full | |
4050 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4051 | * their complexity. On systems without swap this code should be | |
4052 | * effectively equivalent, but much lighter weight. | |
4053 | */ | |
4054 | ||
41ffe5d5 | 4055 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4056 | .name = "tmpfs", |
f3235626 DH |
4057 | .init_fs_context = ramfs_init_fs_context, |
4058 | .parameters = &ramfs_fs_parameters, | |
853ac43a | 4059 | .kill_sb = kill_litter_super, |
2b8576cb | 4060 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4061 | }; |
4062 | ||
41ffe5d5 | 4063 | int __init shmem_init(void) |
853ac43a | 4064 | { |
41ffe5d5 | 4065 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4066 | |
41ffe5d5 | 4067 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
4068 | BUG_ON(IS_ERR(shm_mnt)); |
4069 | ||
4070 | return 0; | |
4071 | } | |
4072 | ||
b56a2d8a VRP |
4073 | int shmem_unuse(unsigned int type, bool frontswap, |
4074 | unsigned long *fs_pages_to_unuse) | |
853ac43a MM |
4075 | { |
4076 | return 0; | |
4077 | } | |
4078 | ||
3f96b79a HD |
4079 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
4080 | { | |
4081 | return 0; | |
4082 | } | |
4083 | ||
24513264 HD |
4084 | void shmem_unlock_mapping(struct address_space *mapping) |
4085 | { | |
4086 | } | |
4087 | ||
c01d5b30 HD |
4088 | #ifdef CONFIG_MMU |
4089 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4090 | unsigned long addr, unsigned long len, | |
4091 | unsigned long pgoff, unsigned long flags) | |
4092 | { | |
4093 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4094 | } | |
4095 | #endif | |
4096 | ||
41ffe5d5 | 4097 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4098 | { |
41ffe5d5 | 4099 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4100 | } |
4101 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4102 | ||
0b0a0806 HD |
4103 | #define shmem_vm_ops generic_file_vm_ops |
4104 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 4105 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
4106 | #define shmem_acct_size(flags, size) 0 |
4107 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
4108 | |
4109 | #endif /* CONFIG_SHMEM */ | |
4110 | ||
4111 | /* common code */ | |
1da177e4 | 4112 | |
703321b6 | 4113 | static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size, |
c7277090 | 4114 | unsigned long flags, unsigned int i_flags) |
1da177e4 | 4115 | { |
1da177e4 | 4116 | struct inode *inode; |
93dec2da | 4117 | struct file *res; |
1da177e4 | 4118 | |
703321b6 MA |
4119 | if (IS_ERR(mnt)) |
4120 | return ERR_CAST(mnt); | |
1da177e4 | 4121 | |
285b2c4f | 4122 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4123 | return ERR_PTR(-EINVAL); |
4124 | ||
4125 | if (shmem_acct_size(flags, size)) | |
4126 | return ERR_PTR(-ENOMEM); | |
4127 | ||
93dec2da AV |
4128 | inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0, |
4129 | flags); | |
dac2d1f6 AV |
4130 | if (unlikely(!inode)) { |
4131 | shmem_unacct_size(flags, size); | |
4132 | return ERR_PTR(-ENOSPC); | |
4133 | } | |
c7277090 | 4134 | inode->i_flags |= i_flags; |
1da177e4 | 4135 | inode->i_size = size; |
6d6b77f1 | 4136 | clear_nlink(inode); /* It is unlinked */ |
26567cdb | 4137 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
93dec2da AV |
4138 | if (!IS_ERR(res)) |
4139 | res = alloc_file_pseudo(inode, mnt, name, O_RDWR, | |
4140 | &shmem_file_operations); | |
26567cdb | 4141 | if (IS_ERR(res)) |
93dec2da | 4142 | iput(inode); |
6b4d0b27 | 4143 | return res; |
1da177e4 | 4144 | } |
c7277090 EP |
4145 | |
4146 | /** | |
4147 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4148 | * kernel internal. There will be NO LSM permission checks against the | |
4149 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4150 | * higher layer. The users are the big_key and shm implementations. LSM |
4151 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4152 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4153 | * @size: size to be set for the file | |
4154 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4155 | */ | |
4156 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4157 | { | |
703321b6 | 4158 | return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE); |
c7277090 EP |
4159 | } |
4160 | ||
4161 | /** | |
4162 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4163 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4164 | * @size: size to be set for the file | |
4165 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4166 | */ | |
4167 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4168 | { | |
703321b6 | 4169 | return __shmem_file_setup(shm_mnt, name, size, flags, 0); |
c7277090 | 4170 | } |
395e0ddc | 4171 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4172 | |
703321b6 MA |
4173 | /** |
4174 | * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs | |
4175 | * @mnt: the tmpfs mount where the file will be created | |
4176 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4177 | * @size: size to be set for the file | |
4178 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4179 | */ | |
4180 | struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name, | |
4181 | loff_t size, unsigned long flags) | |
4182 | { | |
4183 | return __shmem_file_setup(mnt, name, size, flags, 0); | |
4184 | } | |
4185 | EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt); | |
4186 | ||
46711810 | 4187 | /** |
1da177e4 | 4188 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
4189 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
4190 | */ | |
4191 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4192 | { | |
4193 | struct file *file; | |
4194 | loff_t size = vma->vm_end - vma->vm_start; | |
4195 | ||
66fc1303 HD |
4196 | /* |
4197 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
4198 | * between XFS directory reading and selinux: since this file is only | |
4199 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4200 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4201 | */ | |
703321b6 | 4202 | file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags); |
1da177e4 LT |
4203 | if (IS_ERR(file)) |
4204 | return PTR_ERR(file); | |
4205 | ||
4206 | if (vma->vm_file) | |
4207 | fput(vma->vm_file); | |
4208 | vma->vm_file = file; | |
4209 | vma->vm_ops = &shmem_vm_ops; | |
f3f0e1d2 | 4210 | |
e496cf3d | 4211 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
4212 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
4213 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
4214 | khugepaged_enter(vma, vma->vm_flags); | |
4215 | } | |
4216 | ||
1da177e4 LT |
4217 | return 0; |
4218 | } | |
7670e053 | 4219 | EXPORT_SYMBOL_GPL(shmem_zero_setup); |
d9d90e5e HD |
4220 | |
4221 | /** | |
4222 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
4223 | * @mapping: the page's address_space | |
4224 | * @index: the page index | |
4225 | * @gfp: the page allocator flags to use if allocating | |
4226 | * | |
4227 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4228 | * with any new page allocations done using the specified allocation flags. | |
4229 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
4230 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
4231 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4232 | * | |
68da9f05 HD |
4233 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4234 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4235 | */ |
4236 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4237 | pgoff_t index, gfp_t gfp) | |
4238 | { | |
68da9f05 HD |
4239 | #ifdef CONFIG_SHMEM |
4240 | struct inode *inode = mapping->host; | |
9276aad6 | 4241 | struct page *page; |
68da9f05 HD |
4242 | int error; |
4243 | ||
4244 | BUG_ON(mapping->a_ops != &shmem_aops); | |
9e18eb29 | 4245 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
cfda0526 | 4246 | gfp, NULL, NULL, NULL); |
68da9f05 HD |
4247 | if (error) |
4248 | page = ERR_PTR(error); | |
4249 | else | |
4250 | unlock_page(page); | |
4251 | return page; | |
4252 | #else | |
4253 | /* | |
4254 | * The tiny !SHMEM case uses ramfs without swap | |
4255 | */ | |
d9d90e5e | 4256 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4257 | #endif |
d9d90e5e HD |
4258 | } |
4259 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |