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> |
853ac43a | 39 | |
95cc09d6 AA |
40 | #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */ |
41 | ||
853ac43a MM |
42 | static struct vfsmount *shm_mnt; |
43 | ||
44 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
45 | /* |
46 | * This virtual memory filesystem is heavily based on the ramfs. It | |
47 | * extends ramfs by the ability to use swap and honor resource limits | |
48 | * which makes it a completely usable filesystem. | |
49 | */ | |
50 | ||
39f0247d | 51 | #include <linux/xattr.h> |
a5694255 | 52 | #include <linux/exportfs.h> |
1c7c474c | 53 | #include <linux/posix_acl.h> |
feda821e | 54 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 55 | #include <linux/mman.h> |
1da177e4 LT |
56 | #include <linux/string.h> |
57 | #include <linux/slab.h> | |
58 | #include <linux/backing-dev.h> | |
59 | #include <linux/shmem_fs.h> | |
1da177e4 | 60 | #include <linux/writeback.h> |
1da177e4 | 61 | #include <linux/blkdev.h> |
bda97eab | 62 | #include <linux/pagevec.h> |
41ffe5d5 | 63 | #include <linux/percpu_counter.h> |
83e4fa9c | 64 | #include <linux/falloc.h> |
708e3508 | 65 | #include <linux/splice.h> |
1da177e4 LT |
66 | #include <linux/security.h> |
67 | #include <linux/swapops.h> | |
68 | #include <linux/mempolicy.h> | |
69 | #include <linux/namei.h> | |
b00dc3ad | 70 | #include <linux/ctype.h> |
304dbdb7 | 71 | #include <linux/migrate.h> |
c1f60a5a | 72 | #include <linux/highmem.h> |
680d794b | 73 | #include <linux/seq_file.h> |
92562927 | 74 | #include <linux/magic.h> |
9183df25 | 75 | #include <linux/syscalls.h> |
40e041a2 | 76 | #include <linux/fcntl.h> |
9183df25 | 77 | #include <uapi/linux/memfd.h> |
cfda0526 | 78 | #include <linux/userfaultfd_k.h> |
4c27fe4c | 79 | #include <linux/rmap.h> |
2b4db796 | 80 | #include <linux/uuid.h> |
304dbdb7 | 81 | |
7c0f6ba6 | 82 | #include <linux/uaccess.h> |
1da177e4 LT |
83 | #include <asm/pgtable.h> |
84 | ||
dd56b046 MG |
85 | #include "internal.h" |
86 | ||
09cbfeaf KS |
87 | #define BLOCKS_PER_PAGE (PAGE_SIZE/512) |
88 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT) | |
1da177e4 | 89 | |
1da177e4 LT |
90 | /* Pretend that each entry is of this size in directory's i_size */ |
91 | #define BOGO_DIRENT_SIZE 20 | |
92 | ||
69f07ec9 HD |
93 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
94 | #define SHORT_SYMLINK_LEN 128 | |
95 | ||
1aac1400 | 96 | /* |
f00cdc6d HD |
97 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
98 | * inode->i_private (with i_mutex making sure that it has only one user at | |
99 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
100 | */ |
101 | struct shmem_falloc { | |
8e205f77 | 102 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
103 | pgoff_t start; /* start of range currently being fallocated */ |
104 | pgoff_t next; /* the next page offset to be fallocated */ | |
105 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
106 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
107 | }; | |
108 | ||
b76db735 | 109 | #ifdef CONFIG_TMPFS |
680d794b | 110 | static unsigned long shmem_default_max_blocks(void) |
111 | { | |
112 | return totalram_pages / 2; | |
113 | } | |
114 | ||
115 | static unsigned long shmem_default_max_inodes(void) | |
116 | { | |
117 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
118 | } | |
b76db735 | 119 | #endif |
680d794b | 120 | |
bde05d1c HD |
121 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
122 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
123 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 | 124 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 125 | struct page **pagep, enum sgp_type sgp, |
cfda0526 | 126 | gfp_t gfp, struct vm_area_struct *vma, |
2b740303 | 127 | struct vm_fault *vmf, vm_fault_t *fault_type); |
68da9f05 | 128 | |
f3f0e1d2 | 129 | int shmem_getpage(struct inode *inode, pgoff_t index, |
9e18eb29 | 130 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
131 | { |
132 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
cfda0526 | 133 | mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL); |
68da9f05 | 134 | } |
1da177e4 | 135 | |
1da177e4 LT |
136 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
137 | { | |
138 | return sb->s_fs_info; | |
139 | } | |
140 | ||
141 | /* | |
142 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
143 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
144 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
145 | * consistent with the pre-accounting of private mappings ... | |
146 | */ | |
147 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
148 | { | |
0b0a0806 | 149 | return (flags & VM_NORESERVE) ? |
191c5424 | 150 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
151 | } |
152 | ||
153 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
154 | { | |
0b0a0806 | 155 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
156 | vm_unacct_memory(VM_ACCT(size)); |
157 | } | |
158 | ||
77142517 KK |
159 | static inline int shmem_reacct_size(unsigned long flags, |
160 | loff_t oldsize, loff_t newsize) | |
161 | { | |
162 | if (!(flags & VM_NORESERVE)) { | |
163 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
164 | return security_vm_enough_memory_mm(current->mm, | |
165 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
166 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
167 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
168 | } | |
169 | return 0; | |
170 | } | |
171 | ||
1da177e4 LT |
172 | /* |
173 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 174 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
175 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
176 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
177 | */ | |
800d8c63 | 178 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 179 | { |
800d8c63 KS |
180 | if (!(flags & VM_NORESERVE)) |
181 | return 0; | |
182 | ||
183 | return security_vm_enough_memory_mm(current->mm, | |
184 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
185 | } |
186 | ||
187 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
188 | { | |
0b0a0806 | 189 | if (flags & VM_NORESERVE) |
09cbfeaf | 190 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
191 | } |
192 | ||
0f079694 MR |
193 | static inline bool shmem_inode_acct_block(struct inode *inode, long pages) |
194 | { | |
195 | struct shmem_inode_info *info = SHMEM_I(inode); | |
196 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
197 | ||
198 | if (shmem_acct_block(info->flags, pages)) | |
199 | return false; | |
200 | ||
201 | if (sbinfo->max_blocks) { | |
202 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
203 | sbinfo->max_blocks - pages) > 0) | |
204 | goto unacct; | |
205 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
206 | } | |
207 | ||
208 | return true; | |
209 | ||
210 | unacct: | |
211 | shmem_unacct_blocks(info->flags, pages); | |
212 | return false; | |
213 | } | |
214 | ||
215 | static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages) | |
216 | { | |
217 | struct shmem_inode_info *info = SHMEM_I(inode); | |
218 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
219 | ||
220 | if (sbinfo->max_blocks) | |
221 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
222 | shmem_unacct_blocks(info->flags, pages); | |
223 | } | |
224 | ||
759b9775 | 225 | static const struct super_operations shmem_ops; |
f5e54d6e | 226 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 227 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
228 | static const struct inode_operations shmem_inode_operations; |
229 | static const struct inode_operations shmem_dir_inode_operations; | |
230 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 231 | static const struct vm_operations_struct shmem_vm_ops; |
779750d2 | 232 | static struct file_system_type shmem_fs_type; |
1da177e4 | 233 | |
b0506e48 MR |
234 | bool vma_is_shmem(struct vm_area_struct *vma) |
235 | { | |
236 | return vma->vm_ops == &shmem_vm_ops; | |
237 | } | |
238 | ||
1da177e4 | 239 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 240 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 241 | |
5b04c689 PE |
242 | static int shmem_reserve_inode(struct super_block *sb) |
243 | { | |
244 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
245 | if (sbinfo->max_inodes) { | |
246 | spin_lock(&sbinfo->stat_lock); | |
247 | if (!sbinfo->free_inodes) { | |
248 | spin_unlock(&sbinfo->stat_lock); | |
249 | return -ENOSPC; | |
250 | } | |
251 | sbinfo->free_inodes--; | |
252 | spin_unlock(&sbinfo->stat_lock); | |
253 | } | |
254 | return 0; | |
255 | } | |
256 | ||
257 | static void shmem_free_inode(struct super_block *sb) | |
258 | { | |
259 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
260 | if (sbinfo->max_inodes) { | |
261 | spin_lock(&sbinfo->stat_lock); | |
262 | sbinfo->free_inodes++; | |
263 | spin_unlock(&sbinfo->stat_lock); | |
264 | } | |
265 | } | |
266 | ||
46711810 | 267 | /** |
41ffe5d5 | 268 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
269 | * @inode: inode to recalc |
270 | * | |
271 | * We have to calculate the free blocks since the mm can drop | |
272 | * undirtied hole pages behind our back. | |
273 | * | |
274 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
275 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
276 | * | |
277 | * It has to be called with the spinlock held. | |
278 | */ | |
279 | static void shmem_recalc_inode(struct inode *inode) | |
280 | { | |
281 | struct shmem_inode_info *info = SHMEM_I(inode); | |
282 | long freed; | |
283 | ||
284 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
285 | if (freed > 0) { | |
286 | info->alloced -= freed; | |
54af6042 | 287 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
0f079694 | 288 | shmem_inode_unacct_blocks(inode, freed); |
1da177e4 LT |
289 | } |
290 | } | |
291 | ||
800d8c63 KS |
292 | bool shmem_charge(struct inode *inode, long pages) |
293 | { | |
294 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 295 | unsigned long flags; |
800d8c63 | 296 | |
0f079694 | 297 | if (!shmem_inode_acct_block(inode, pages)) |
800d8c63 | 298 | return false; |
b1cc94ab | 299 | |
4595ef88 | 300 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
301 | info->alloced += pages; |
302 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
303 | shmem_recalc_inode(inode); | |
4595ef88 | 304 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
305 | inode->i_mapping->nrpages += pages; |
306 | ||
800d8c63 KS |
307 | return true; |
308 | } | |
309 | ||
310 | void shmem_uncharge(struct inode *inode, long pages) | |
311 | { | |
312 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 313 | unsigned long flags; |
800d8c63 | 314 | |
4595ef88 | 315 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
316 | info->alloced -= pages; |
317 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
318 | shmem_recalc_inode(inode); | |
4595ef88 | 319 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 | 320 | |
0f079694 | 321 | shmem_inode_unacct_blocks(inode, pages); |
800d8c63 KS |
322 | } |
323 | ||
7a5d0fbb | 324 | /* |
62f945b6 | 325 | * Replace item expected in xarray by a new item, while holding xa_lock. |
7a5d0fbb | 326 | */ |
62f945b6 | 327 | static int shmem_replace_entry(struct address_space *mapping, |
7a5d0fbb HD |
328 | pgoff_t index, void *expected, void *replacement) |
329 | { | |
62f945b6 | 330 | XA_STATE(xas, &mapping->i_pages, index); |
6dbaf22c | 331 | void *item; |
7a5d0fbb HD |
332 | |
333 | VM_BUG_ON(!expected); | |
6dbaf22c | 334 | VM_BUG_ON(!replacement); |
62f945b6 | 335 | item = xas_load(&xas); |
7a5d0fbb HD |
336 | if (item != expected) |
337 | return -ENOENT; | |
62f945b6 | 338 | xas_store(&xas, replacement); |
7a5d0fbb HD |
339 | return 0; |
340 | } | |
341 | ||
d1899228 HD |
342 | /* |
343 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
344 | * that an entry was not already brought back from swap by a racing thread. | |
345 | * | |
346 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
347 | * might be reused, and again be SwapCache, using the same swap as before. | |
348 | */ | |
349 | static bool shmem_confirm_swap(struct address_space *mapping, | |
350 | pgoff_t index, swp_entry_t swap) | |
351 | { | |
a12831bf | 352 | return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap); |
d1899228 HD |
353 | } |
354 | ||
5a6e75f8 KS |
355 | /* |
356 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
357 | * | |
358 | * SHMEM_HUGE_NEVER: | |
359 | * disables huge pages for the mount; | |
360 | * SHMEM_HUGE_ALWAYS: | |
361 | * enables huge pages for the mount; | |
362 | * SHMEM_HUGE_WITHIN_SIZE: | |
363 | * only allocate huge pages if the page will be fully within i_size, | |
364 | * also respect fadvise()/madvise() hints; | |
365 | * SHMEM_HUGE_ADVISE: | |
366 | * only allocate huge pages if requested with fadvise()/madvise(); | |
367 | */ | |
368 | ||
369 | #define SHMEM_HUGE_NEVER 0 | |
370 | #define SHMEM_HUGE_ALWAYS 1 | |
371 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
372 | #define SHMEM_HUGE_ADVISE 3 | |
373 | ||
374 | /* | |
375 | * Special values. | |
376 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
377 | * | |
378 | * SHMEM_HUGE_DENY: | |
379 | * disables huge on shm_mnt and all mounts, for emergency use; | |
380 | * SHMEM_HUGE_FORCE: | |
381 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
382 | * | |
383 | */ | |
384 | #define SHMEM_HUGE_DENY (-1) | |
385 | #define SHMEM_HUGE_FORCE (-2) | |
386 | ||
e496cf3d | 387 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
388 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
389 | ||
5b9c98f3 | 390 | static int shmem_huge __read_mostly; |
5a6e75f8 | 391 | |
f1f5929c | 392 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
393 | static int shmem_parse_huge(const char *str) |
394 | { | |
395 | if (!strcmp(str, "never")) | |
396 | return SHMEM_HUGE_NEVER; | |
397 | if (!strcmp(str, "always")) | |
398 | return SHMEM_HUGE_ALWAYS; | |
399 | if (!strcmp(str, "within_size")) | |
400 | return SHMEM_HUGE_WITHIN_SIZE; | |
401 | if (!strcmp(str, "advise")) | |
402 | return SHMEM_HUGE_ADVISE; | |
403 | if (!strcmp(str, "deny")) | |
404 | return SHMEM_HUGE_DENY; | |
405 | if (!strcmp(str, "force")) | |
406 | return SHMEM_HUGE_FORCE; | |
407 | return -EINVAL; | |
408 | } | |
409 | ||
410 | static const char *shmem_format_huge(int huge) | |
411 | { | |
412 | switch (huge) { | |
413 | case SHMEM_HUGE_NEVER: | |
414 | return "never"; | |
415 | case SHMEM_HUGE_ALWAYS: | |
416 | return "always"; | |
417 | case SHMEM_HUGE_WITHIN_SIZE: | |
418 | return "within_size"; | |
419 | case SHMEM_HUGE_ADVISE: | |
420 | return "advise"; | |
421 | case SHMEM_HUGE_DENY: | |
422 | return "deny"; | |
423 | case SHMEM_HUGE_FORCE: | |
424 | return "force"; | |
425 | default: | |
426 | VM_BUG_ON(1); | |
427 | return "bad_val"; | |
428 | } | |
429 | } | |
f1f5929c | 430 | #endif |
5a6e75f8 | 431 | |
779750d2 KS |
432 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
433 | struct shrink_control *sc, unsigned long nr_to_split) | |
434 | { | |
435 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 436 | LIST_HEAD(to_remove); |
779750d2 KS |
437 | struct inode *inode; |
438 | struct shmem_inode_info *info; | |
439 | struct page *page; | |
440 | unsigned long batch = sc ? sc->nr_to_scan : 128; | |
441 | int removed = 0, split = 0; | |
442 | ||
443 | if (list_empty(&sbinfo->shrinklist)) | |
444 | return SHRINK_STOP; | |
445 | ||
446 | spin_lock(&sbinfo->shrinklist_lock); | |
447 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
448 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
449 | ||
450 | /* pin the inode */ | |
451 | inode = igrab(&info->vfs_inode); | |
452 | ||
453 | /* inode is about to be evicted */ | |
454 | if (!inode) { | |
455 | list_del_init(&info->shrinklist); | |
456 | removed++; | |
457 | goto next; | |
458 | } | |
459 | ||
460 | /* Check if there's anything to gain */ | |
461 | if (round_up(inode->i_size, PAGE_SIZE) == | |
462 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 463 | list_move(&info->shrinklist, &to_remove); |
779750d2 | 464 | removed++; |
779750d2 KS |
465 | goto next; |
466 | } | |
467 | ||
468 | list_move(&info->shrinklist, &list); | |
469 | next: | |
470 | if (!--batch) | |
471 | break; | |
472 | } | |
473 | spin_unlock(&sbinfo->shrinklist_lock); | |
474 | ||
253fd0f0 KS |
475 | list_for_each_safe(pos, next, &to_remove) { |
476 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
477 | inode = &info->vfs_inode; | |
478 | list_del_init(&info->shrinklist); | |
479 | iput(inode); | |
480 | } | |
481 | ||
779750d2 KS |
482 | list_for_each_safe(pos, next, &list) { |
483 | int ret; | |
484 | ||
485 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
486 | inode = &info->vfs_inode; | |
487 | ||
b3cd54b2 KS |
488 | if (nr_to_split && split >= nr_to_split) |
489 | goto leave; | |
779750d2 | 490 | |
b3cd54b2 | 491 | page = find_get_page(inode->i_mapping, |
779750d2 KS |
492 | (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT); |
493 | if (!page) | |
494 | goto drop; | |
495 | ||
b3cd54b2 | 496 | /* No huge page at the end of the file: nothing to split */ |
779750d2 | 497 | if (!PageTransHuge(page)) { |
779750d2 KS |
498 | put_page(page); |
499 | goto drop; | |
500 | } | |
501 | ||
b3cd54b2 KS |
502 | /* |
503 | * Leave the inode on the list if we failed to lock | |
504 | * the page at this time. | |
505 | * | |
506 | * Waiting for the lock may lead to deadlock in the | |
507 | * reclaim path. | |
508 | */ | |
509 | if (!trylock_page(page)) { | |
510 | put_page(page); | |
511 | goto leave; | |
512 | } | |
513 | ||
779750d2 KS |
514 | ret = split_huge_page(page); |
515 | unlock_page(page); | |
516 | put_page(page); | |
517 | ||
b3cd54b2 KS |
518 | /* If split failed leave the inode on the list */ |
519 | if (ret) | |
520 | goto leave; | |
779750d2 KS |
521 | |
522 | split++; | |
523 | drop: | |
524 | list_del_init(&info->shrinklist); | |
525 | removed++; | |
b3cd54b2 | 526 | leave: |
779750d2 KS |
527 | iput(inode); |
528 | } | |
529 | ||
530 | spin_lock(&sbinfo->shrinklist_lock); | |
531 | list_splice_tail(&list, &sbinfo->shrinklist); | |
532 | sbinfo->shrinklist_len -= removed; | |
533 | spin_unlock(&sbinfo->shrinklist_lock); | |
534 | ||
535 | return split; | |
536 | } | |
537 | ||
538 | static long shmem_unused_huge_scan(struct super_block *sb, | |
539 | struct shrink_control *sc) | |
540 | { | |
541 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
542 | ||
543 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
544 | return SHRINK_STOP; | |
545 | ||
546 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
547 | } | |
548 | ||
549 | static long shmem_unused_huge_count(struct super_block *sb, | |
550 | struct shrink_control *sc) | |
551 | { | |
552 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
553 | return READ_ONCE(sbinfo->shrinklist_len); | |
554 | } | |
e496cf3d | 555 | #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 KS |
556 | |
557 | #define shmem_huge SHMEM_HUGE_DENY | |
558 | ||
779750d2 KS |
559 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
560 | struct shrink_control *sc, unsigned long nr_to_split) | |
561 | { | |
562 | return 0; | |
563 | } | |
e496cf3d | 564 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 565 | |
89fdcd26 YS |
566 | static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo) |
567 | { | |
568 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && | |
569 | (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) && | |
570 | shmem_huge != SHMEM_HUGE_DENY) | |
571 | return true; | |
572 | return false; | |
573 | } | |
574 | ||
46f65ec1 HD |
575 | /* |
576 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
577 | */ | |
578 | static int shmem_add_to_page_cache(struct page *page, | |
579 | struct address_space *mapping, | |
fed400a1 | 580 | pgoff_t index, void *expected) |
46f65ec1 | 581 | { |
800d8c63 | 582 | int error, nr = hpage_nr_pages(page); |
46f65ec1 | 583 | |
800d8c63 KS |
584 | VM_BUG_ON_PAGE(PageTail(page), page); |
585 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
586 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
587 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 588 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 589 | |
800d8c63 | 590 | page_ref_add(page, nr); |
b065b432 HD |
591 | page->mapping = mapping; |
592 | page->index = index; | |
593 | ||
b93b0163 | 594 | xa_lock_irq(&mapping->i_pages); |
800d8c63 KS |
595 | if (PageTransHuge(page)) { |
596 | void __rcu **results; | |
597 | pgoff_t idx; | |
598 | int i; | |
599 | ||
600 | error = 0; | |
b93b0163 | 601 | if (radix_tree_gang_lookup_slot(&mapping->i_pages, |
800d8c63 KS |
602 | &results, &idx, index, 1) && |
603 | idx < index + HPAGE_PMD_NR) { | |
604 | error = -EEXIST; | |
605 | } | |
606 | ||
607 | if (!error) { | |
608 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
b93b0163 | 609 | error = radix_tree_insert(&mapping->i_pages, |
800d8c63 KS |
610 | index + i, page + i); |
611 | VM_BUG_ON(error); | |
612 | } | |
613 | count_vm_event(THP_FILE_ALLOC); | |
614 | } | |
615 | } else if (!expected) { | |
b93b0163 | 616 | error = radix_tree_insert(&mapping->i_pages, index, page); |
800d8c63 | 617 | } else { |
62f945b6 | 618 | error = shmem_replace_entry(mapping, index, expected, page); |
800d8c63 KS |
619 | } |
620 | ||
46f65ec1 | 621 | if (!error) { |
800d8c63 KS |
622 | mapping->nrpages += nr; |
623 | if (PageTransHuge(page)) | |
11fb9989 MG |
624 | __inc_node_page_state(page, NR_SHMEM_THPS); |
625 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); | |
626 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr); | |
b93b0163 | 627 | xa_unlock_irq(&mapping->i_pages); |
b065b432 HD |
628 | } else { |
629 | page->mapping = NULL; | |
b93b0163 | 630 | xa_unlock_irq(&mapping->i_pages); |
800d8c63 | 631 | page_ref_sub(page, nr); |
46f65ec1 | 632 | } |
46f65ec1 HD |
633 | return error; |
634 | } | |
635 | ||
6922c0c7 HD |
636 | /* |
637 | * Like delete_from_page_cache, but substitutes swap for page. | |
638 | */ | |
639 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
640 | { | |
641 | struct address_space *mapping = page->mapping; | |
642 | int error; | |
643 | ||
800d8c63 KS |
644 | VM_BUG_ON_PAGE(PageCompound(page), page); |
645 | ||
b93b0163 | 646 | xa_lock_irq(&mapping->i_pages); |
62f945b6 | 647 | error = shmem_replace_entry(mapping, page->index, page, radswap); |
6922c0c7 HD |
648 | page->mapping = NULL; |
649 | mapping->nrpages--; | |
11fb9989 MG |
650 | __dec_node_page_state(page, NR_FILE_PAGES); |
651 | __dec_node_page_state(page, NR_SHMEM); | |
b93b0163 | 652 | xa_unlock_irq(&mapping->i_pages); |
09cbfeaf | 653 | put_page(page); |
6922c0c7 HD |
654 | BUG_ON(error); |
655 | } | |
656 | ||
7a5d0fbb HD |
657 | /* |
658 | * Remove swap entry from radix tree, free the swap and its page cache. | |
659 | */ | |
660 | static int shmem_free_swap(struct address_space *mapping, | |
661 | pgoff_t index, void *radswap) | |
662 | { | |
6dbaf22c | 663 | void *old; |
7a5d0fbb | 664 | |
b93b0163 MW |
665 | xa_lock_irq(&mapping->i_pages); |
666 | old = radix_tree_delete_item(&mapping->i_pages, index, radswap); | |
667 | xa_unlock_irq(&mapping->i_pages); | |
6dbaf22c JW |
668 | if (old != radswap) |
669 | return -ENOENT; | |
670 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
671 | return 0; | |
7a5d0fbb HD |
672 | } |
673 | ||
6a15a370 VB |
674 | /* |
675 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 676 | * given offsets are swapped out. |
6a15a370 | 677 | * |
b93b0163 | 678 | * This is safe to call without i_mutex or the i_pages lock thanks to RCU, |
6a15a370 VB |
679 | * as long as the inode doesn't go away and racy results are not a problem. |
680 | */ | |
48131e03 VB |
681 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
682 | pgoff_t start, pgoff_t end) | |
6a15a370 | 683 | { |
6a15a370 | 684 | struct radix_tree_iter iter; |
5b9c98f3 | 685 | void __rcu **slot; |
6a15a370 | 686 | struct page *page; |
48131e03 | 687 | unsigned long swapped = 0; |
6a15a370 VB |
688 | |
689 | rcu_read_lock(); | |
690 | ||
b93b0163 | 691 | radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) { |
6a15a370 VB |
692 | if (iter.index >= end) |
693 | break; | |
694 | ||
695 | page = radix_tree_deref_slot(slot); | |
696 | ||
2cf938aa MW |
697 | if (radix_tree_deref_retry(page)) { |
698 | slot = radix_tree_iter_retry(&iter); | |
699 | continue; | |
700 | } | |
6a15a370 | 701 | |
3159f943 | 702 | if (xa_is_value(page)) |
6a15a370 VB |
703 | swapped++; |
704 | ||
705 | if (need_resched()) { | |
148deab2 | 706 | slot = radix_tree_iter_resume(slot, &iter); |
6a15a370 | 707 | cond_resched_rcu(); |
6a15a370 VB |
708 | } |
709 | } | |
710 | ||
711 | rcu_read_unlock(); | |
712 | ||
713 | return swapped << PAGE_SHIFT; | |
714 | } | |
715 | ||
48131e03 VB |
716 | /* |
717 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
718 | * given vma is swapped out. | |
719 | * | |
b93b0163 | 720 | * This is safe to call without i_mutex or the i_pages lock thanks to RCU, |
48131e03 VB |
721 | * as long as the inode doesn't go away and racy results are not a problem. |
722 | */ | |
723 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
724 | { | |
725 | struct inode *inode = file_inode(vma->vm_file); | |
726 | struct shmem_inode_info *info = SHMEM_I(inode); | |
727 | struct address_space *mapping = inode->i_mapping; | |
728 | unsigned long swapped; | |
729 | ||
730 | /* Be careful as we don't hold info->lock */ | |
731 | swapped = READ_ONCE(info->swapped); | |
732 | ||
733 | /* | |
734 | * The easier cases are when the shmem object has nothing in swap, or | |
735 | * the vma maps it whole. Then we can simply use the stats that we | |
736 | * already track. | |
737 | */ | |
738 | if (!swapped) | |
739 | return 0; | |
740 | ||
741 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
742 | return swapped << PAGE_SHIFT; | |
743 | ||
744 | /* Here comes the more involved part */ | |
745 | return shmem_partial_swap_usage(mapping, | |
746 | linear_page_index(vma, vma->vm_start), | |
747 | linear_page_index(vma, vma->vm_end)); | |
748 | } | |
749 | ||
24513264 HD |
750 | /* |
751 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
752 | */ | |
753 | void shmem_unlock_mapping(struct address_space *mapping) | |
754 | { | |
755 | struct pagevec pvec; | |
756 | pgoff_t indices[PAGEVEC_SIZE]; | |
757 | pgoff_t index = 0; | |
758 | ||
86679820 | 759 | pagevec_init(&pvec); |
24513264 HD |
760 | /* |
761 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
762 | */ | |
763 | while (!mapping_unevictable(mapping)) { | |
764 | /* | |
765 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
766 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
767 | */ | |
0cd6144a JW |
768 | pvec.nr = find_get_entries(mapping, index, |
769 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
770 | if (!pvec.nr) |
771 | break; | |
772 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 773 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
774 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
775 | pagevec_release(&pvec); | |
776 | cond_resched(); | |
777 | } | |
7a5d0fbb HD |
778 | } |
779 | ||
780 | /* | |
781 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 782 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 783 | */ |
1635f6a7 HD |
784 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
785 | bool unfalloc) | |
1da177e4 | 786 | { |
285b2c4f | 787 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 788 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
789 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
790 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
791 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
792 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 793 | struct pagevec pvec; |
7a5d0fbb HD |
794 | pgoff_t indices[PAGEVEC_SIZE]; |
795 | long nr_swaps_freed = 0; | |
285b2c4f | 796 | pgoff_t index; |
bda97eab HD |
797 | int i; |
798 | ||
83e4fa9c HD |
799 | if (lend == -1) |
800 | end = -1; /* unsigned, so actually very big */ | |
bda97eab | 801 | |
86679820 | 802 | pagevec_init(&pvec); |
bda97eab | 803 | index = start; |
83e4fa9c | 804 | while (index < end) { |
0cd6144a JW |
805 | pvec.nr = find_get_entries(mapping, index, |
806 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
807 | pvec.pages, indices); | |
7a5d0fbb HD |
808 | if (!pvec.nr) |
809 | break; | |
bda97eab HD |
810 | for (i = 0; i < pagevec_count(&pvec); i++) { |
811 | struct page *page = pvec.pages[i]; | |
812 | ||
7a5d0fbb | 813 | index = indices[i]; |
83e4fa9c | 814 | if (index >= end) |
bda97eab HD |
815 | break; |
816 | ||
3159f943 | 817 | if (xa_is_value(page)) { |
1635f6a7 HD |
818 | if (unfalloc) |
819 | continue; | |
7a5d0fbb HD |
820 | nr_swaps_freed += !shmem_free_swap(mapping, |
821 | index, page); | |
bda97eab | 822 | continue; |
7a5d0fbb HD |
823 | } |
824 | ||
800d8c63 KS |
825 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
826 | ||
7a5d0fbb | 827 | if (!trylock_page(page)) |
bda97eab | 828 | continue; |
800d8c63 KS |
829 | |
830 | if (PageTransTail(page)) { | |
831 | /* Middle of THP: zero out the page */ | |
832 | clear_highpage(page); | |
833 | unlock_page(page); | |
834 | continue; | |
835 | } else if (PageTransHuge(page)) { | |
836 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
837 | /* | |
838 | * Range ends in the middle of THP: | |
839 | * zero out the page | |
840 | */ | |
841 | clear_highpage(page); | |
842 | unlock_page(page); | |
843 | continue; | |
844 | } | |
845 | index += HPAGE_PMD_NR - 1; | |
846 | i += HPAGE_PMD_NR - 1; | |
847 | } | |
848 | ||
1635f6a7 | 849 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
850 | VM_BUG_ON_PAGE(PageTail(page), page); |
851 | if (page_mapping(page) == mapping) { | |
309381fe | 852 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
853 | truncate_inode_page(mapping, page); |
854 | } | |
bda97eab | 855 | } |
bda97eab HD |
856 | unlock_page(page); |
857 | } | |
0cd6144a | 858 | pagevec_remove_exceptionals(&pvec); |
24513264 | 859 | pagevec_release(&pvec); |
bda97eab HD |
860 | cond_resched(); |
861 | index++; | |
862 | } | |
1da177e4 | 863 | |
83e4fa9c | 864 | if (partial_start) { |
bda97eab | 865 | struct page *page = NULL; |
9e18eb29 | 866 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 867 | if (page) { |
09cbfeaf | 868 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
869 | if (start > end) { |
870 | top = partial_end; | |
871 | partial_end = 0; | |
872 | } | |
873 | zero_user_segment(page, partial_start, top); | |
874 | set_page_dirty(page); | |
875 | unlock_page(page); | |
09cbfeaf | 876 | put_page(page); |
83e4fa9c HD |
877 | } |
878 | } | |
879 | if (partial_end) { | |
880 | struct page *page = NULL; | |
9e18eb29 | 881 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
882 | if (page) { |
883 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
884 | set_page_dirty(page); |
885 | unlock_page(page); | |
09cbfeaf | 886 | put_page(page); |
bda97eab HD |
887 | } |
888 | } | |
83e4fa9c HD |
889 | if (start >= end) |
890 | return; | |
bda97eab HD |
891 | |
892 | index = start; | |
b1a36650 | 893 | while (index < end) { |
bda97eab | 894 | cond_resched(); |
0cd6144a JW |
895 | |
896 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 897 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 898 | pvec.pages, indices); |
7a5d0fbb | 899 | if (!pvec.nr) { |
b1a36650 HD |
900 | /* If all gone or hole-punch or unfalloc, we're done */ |
901 | if (index == start || end != -1) | |
bda97eab | 902 | break; |
b1a36650 | 903 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
904 | index = start; |
905 | continue; | |
906 | } | |
bda97eab HD |
907 | for (i = 0; i < pagevec_count(&pvec); i++) { |
908 | struct page *page = pvec.pages[i]; | |
909 | ||
7a5d0fbb | 910 | index = indices[i]; |
83e4fa9c | 911 | if (index >= end) |
bda97eab HD |
912 | break; |
913 | ||
3159f943 | 914 | if (xa_is_value(page)) { |
1635f6a7 HD |
915 | if (unfalloc) |
916 | continue; | |
b1a36650 HD |
917 | if (shmem_free_swap(mapping, index, page)) { |
918 | /* Swap was replaced by page: retry */ | |
919 | index--; | |
920 | break; | |
921 | } | |
922 | nr_swaps_freed++; | |
7a5d0fbb HD |
923 | continue; |
924 | } | |
925 | ||
bda97eab | 926 | lock_page(page); |
800d8c63 KS |
927 | |
928 | if (PageTransTail(page)) { | |
929 | /* Middle of THP: zero out the page */ | |
930 | clear_highpage(page); | |
931 | unlock_page(page); | |
932 | /* | |
933 | * Partial thp truncate due 'start' in middle | |
934 | * of THP: don't need to look on these pages | |
935 | * again on !pvec.nr restart. | |
936 | */ | |
937 | if (index != round_down(end, HPAGE_PMD_NR)) | |
938 | start++; | |
939 | continue; | |
940 | } else if (PageTransHuge(page)) { | |
941 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
942 | /* | |
943 | * Range ends in the middle of THP: | |
944 | * zero out the page | |
945 | */ | |
946 | clear_highpage(page); | |
947 | unlock_page(page); | |
948 | continue; | |
949 | } | |
950 | index += HPAGE_PMD_NR - 1; | |
951 | i += HPAGE_PMD_NR - 1; | |
952 | } | |
953 | ||
1635f6a7 | 954 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
955 | VM_BUG_ON_PAGE(PageTail(page), page); |
956 | if (page_mapping(page) == mapping) { | |
309381fe | 957 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 958 | truncate_inode_page(mapping, page); |
b1a36650 HD |
959 | } else { |
960 | /* Page was replaced by swap: retry */ | |
961 | unlock_page(page); | |
962 | index--; | |
963 | break; | |
1635f6a7 | 964 | } |
7a5d0fbb | 965 | } |
bda97eab HD |
966 | unlock_page(page); |
967 | } | |
0cd6144a | 968 | pagevec_remove_exceptionals(&pvec); |
24513264 | 969 | pagevec_release(&pvec); |
bda97eab HD |
970 | index++; |
971 | } | |
94c1e62d | 972 | |
4595ef88 | 973 | spin_lock_irq(&info->lock); |
7a5d0fbb | 974 | info->swapped -= nr_swaps_freed; |
1da177e4 | 975 | shmem_recalc_inode(inode); |
4595ef88 | 976 | spin_unlock_irq(&info->lock); |
1635f6a7 | 977 | } |
1da177e4 | 978 | |
1635f6a7 HD |
979 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
980 | { | |
981 | shmem_undo_range(inode, lstart, lend, false); | |
078cd827 | 982 | inode->i_ctime = inode->i_mtime = current_time(inode); |
1da177e4 | 983 | } |
94c1e62d | 984 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 985 | |
a528d35e DH |
986 | static int shmem_getattr(const struct path *path, struct kstat *stat, |
987 | u32 request_mask, unsigned int query_flags) | |
44a30220 | 988 | { |
a528d35e | 989 | struct inode *inode = path->dentry->d_inode; |
44a30220 | 990 | struct shmem_inode_info *info = SHMEM_I(inode); |
89fdcd26 | 991 | struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb); |
44a30220 | 992 | |
d0424c42 | 993 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
4595ef88 | 994 | spin_lock_irq(&info->lock); |
d0424c42 | 995 | shmem_recalc_inode(inode); |
4595ef88 | 996 | spin_unlock_irq(&info->lock); |
d0424c42 | 997 | } |
44a30220 | 998 | generic_fillattr(inode, stat); |
89fdcd26 YS |
999 | |
1000 | if (is_huge_enabled(sb_info)) | |
1001 | stat->blksize = HPAGE_PMD_SIZE; | |
1002 | ||
44a30220 YZ |
1003 | return 0; |
1004 | } | |
1005 | ||
94c1e62d | 1006 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 1007 | { |
75c3cfa8 | 1008 | struct inode *inode = d_inode(dentry); |
40e041a2 | 1009 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1010 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 LT |
1011 | int error; |
1012 | ||
31051c85 | 1013 | error = setattr_prepare(dentry, attr); |
db78b877 CH |
1014 | if (error) |
1015 | return error; | |
1016 | ||
94c1e62d HD |
1017 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
1018 | loff_t oldsize = inode->i_size; | |
1019 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 1020 | |
40e041a2 DH |
1021 | /* protected by i_mutex */ |
1022 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
1023 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
1024 | return -EPERM; | |
1025 | ||
94c1e62d | 1026 | if (newsize != oldsize) { |
77142517 KK |
1027 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
1028 | oldsize, newsize); | |
1029 | if (error) | |
1030 | return error; | |
94c1e62d | 1031 | i_size_write(inode, newsize); |
078cd827 | 1032 | inode->i_ctime = inode->i_mtime = current_time(inode); |
94c1e62d | 1033 | } |
afa2db2f | 1034 | if (newsize <= oldsize) { |
94c1e62d | 1035 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
1036 | if (oldsize > holebegin) |
1037 | unmap_mapping_range(inode->i_mapping, | |
1038 | holebegin, 0, 1); | |
1039 | if (info->alloced) | |
1040 | shmem_truncate_range(inode, | |
1041 | newsize, (loff_t)-1); | |
94c1e62d | 1042 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
1043 | if (oldsize > holebegin) |
1044 | unmap_mapping_range(inode->i_mapping, | |
1045 | holebegin, 0, 1); | |
779750d2 KS |
1046 | |
1047 | /* | |
1048 | * Part of the huge page can be beyond i_size: subject | |
1049 | * to shrink under memory pressure. | |
1050 | */ | |
1051 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { | |
1052 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1053 | /* |
1054 | * _careful to defend against unlocked access to | |
1055 | * ->shrink_list in shmem_unused_huge_shrink() | |
1056 | */ | |
1057 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1058 | list_add_tail(&info->shrinklist, |
1059 | &sbinfo->shrinklist); | |
1060 | sbinfo->shrinklist_len++; | |
1061 | } | |
1062 | spin_unlock(&sbinfo->shrinklist_lock); | |
1063 | } | |
94c1e62d | 1064 | } |
1da177e4 LT |
1065 | } |
1066 | ||
db78b877 | 1067 | setattr_copy(inode, attr); |
db78b877 | 1068 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 1069 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
1070 | return error; |
1071 | } | |
1072 | ||
1f895f75 | 1073 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1074 | { |
1da177e4 | 1075 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1076 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 | 1077 | |
3889e6e7 | 1078 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
1079 | shmem_unacct_size(info->flags, inode->i_size); |
1080 | inode->i_size = 0; | |
3889e6e7 | 1081 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1082 | if (!list_empty(&info->shrinklist)) { |
1083 | spin_lock(&sbinfo->shrinklist_lock); | |
1084 | if (!list_empty(&info->shrinklist)) { | |
1085 | list_del_init(&info->shrinklist); | |
1086 | sbinfo->shrinklist_len--; | |
1087 | } | |
1088 | spin_unlock(&sbinfo->shrinklist_lock); | |
1089 | } | |
1da177e4 | 1090 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 1091 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 1092 | list_del_init(&info->swaplist); |
cb5f7b9a | 1093 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1094 | } |
3ed47db3 | 1095 | } |
b09e0fa4 | 1096 | |
38f38657 | 1097 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 1098 | WARN_ON(inode->i_blocks); |
5b04c689 | 1099 | shmem_free_inode(inode->i_sb); |
dbd5768f | 1100 | clear_inode(inode); |
1da177e4 LT |
1101 | } |
1102 | ||
e21a2955 | 1103 | static unsigned long find_swap_entry(struct xarray *xa, void *item) |
478922e2 | 1104 | { |
e21a2955 | 1105 | XA_STATE(xas, xa, 0); |
478922e2 | 1106 | unsigned int checked = 0; |
e21a2955 | 1107 | void *entry; |
478922e2 MW |
1108 | |
1109 | rcu_read_lock(); | |
e21a2955 MW |
1110 | xas_for_each(&xas, entry, ULONG_MAX) { |
1111 | if (xas_retry(&xas, entry)) | |
5b9c98f3 | 1112 | continue; |
e21a2955 | 1113 | if (entry == item) |
478922e2 | 1114 | break; |
478922e2 | 1115 | checked++; |
e21a2955 | 1116 | if ((checked % XA_CHECK_SCHED) != 0) |
478922e2 | 1117 | continue; |
e21a2955 | 1118 | xas_pause(&xas); |
478922e2 MW |
1119 | cond_resched_rcu(); |
1120 | } | |
478922e2 | 1121 | rcu_read_unlock(); |
e21a2955 MW |
1122 | |
1123 | return entry ? xas.xa_index : -1; | |
478922e2 MW |
1124 | } |
1125 | ||
46f65ec1 HD |
1126 | /* |
1127 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1128 | */ | |
41ffe5d5 | 1129 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 1130 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 1131 | { |
285b2c4f | 1132 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 1133 | void *radswap; |
41ffe5d5 | 1134 | pgoff_t index; |
bde05d1c HD |
1135 | gfp_t gfp; |
1136 | int error = 0; | |
1da177e4 | 1137 | |
46f65ec1 | 1138 | radswap = swp_to_radix_entry(swap); |
b93b0163 | 1139 | index = find_swap_entry(&mapping->i_pages, radswap); |
46f65ec1 | 1140 | if (index == -1) |
00501b53 | 1141 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 1142 | |
1b1b32f2 HD |
1143 | /* |
1144 | * Move _head_ to start search for next from here. | |
1f895f75 | 1145 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 1146 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 1147 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
1148 | */ |
1149 | if (shmem_swaplist.next != &info->swaplist) | |
1150 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 1151 | |
bde05d1c HD |
1152 | gfp = mapping_gfp_mask(mapping); |
1153 | if (shmem_should_replace_page(*pagep, gfp)) { | |
1154 | mutex_unlock(&shmem_swaplist_mutex); | |
1155 | error = shmem_replace_page(pagep, gfp, info, index); | |
1156 | mutex_lock(&shmem_swaplist_mutex); | |
1157 | /* | |
1158 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
1159 | * allocation, but the inode might have been freed while we |
1160 | * dropped it: although a racing shmem_evict_inode() cannot | |
1161 | * complete without emptying the radix_tree, our page lock | |
1162 | * on this swapcache page is not enough to prevent that - | |
1163 | * free_swap_and_cache() of our swap entry will only | |
1164 | * trylock_page(), removing swap from radix_tree whatever. | |
1165 | * | |
1166 | * We must not proceed to shmem_add_to_page_cache() if the | |
1167 | * inode has been freed, but of course we cannot rely on | |
1168 | * inode or mapping or info to check that. However, we can | |
1169 | * safely check if our swap entry is still in use (and here | |
1170 | * it can't have got reused for another page): if it's still | |
1171 | * in use, then the inode cannot have been freed yet, and we | |
1172 | * can safely proceed (if it's no longer in use, that tells | |
1173 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
1174 | */ |
1175 | if (!page_swapcount(*pagep)) | |
1176 | error = -ENOENT; | |
1177 | } | |
1178 | ||
d13d1443 | 1179 | /* |
778dd893 HD |
1180 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
1181 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
1182 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 1183 | */ |
bde05d1c HD |
1184 | if (!error) |
1185 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 1186 | radswap); |
48f170fb | 1187 | if (error != -ENOMEM) { |
46f65ec1 HD |
1188 | /* |
1189 | * Truncation and eviction use free_swap_and_cache(), which | |
1190 | * only does trylock page: if we raced, best clean up here. | |
1191 | */ | |
bde05d1c HD |
1192 | delete_from_swap_cache(*pagep); |
1193 | set_page_dirty(*pagep); | |
46f65ec1 | 1194 | if (!error) { |
4595ef88 | 1195 | spin_lock_irq(&info->lock); |
46f65ec1 | 1196 | info->swapped--; |
4595ef88 | 1197 | spin_unlock_irq(&info->lock); |
46f65ec1 HD |
1198 | swap_free(swap); |
1199 | } | |
1da177e4 | 1200 | } |
2e0e26c7 | 1201 | return error; |
1da177e4 LT |
1202 | } |
1203 | ||
1204 | /* | |
46f65ec1 | 1205 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 1206 | */ |
41ffe5d5 | 1207 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 1208 | { |
41ffe5d5 | 1209 | struct list_head *this, *next; |
1da177e4 | 1210 | struct shmem_inode_info *info; |
00501b53 | 1211 | struct mem_cgroup *memcg; |
bde05d1c HD |
1212 | int error = 0; |
1213 | ||
1214 | /* | |
1215 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 1216 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 1217 | */ |
0142ef6c | 1218 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 1219 | goto out; |
778dd893 HD |
1220 | |
1221 | /* | |
1222 | * Charge page using GFP_KERNEL while we can wait, before taking | |
1223 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
1224 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 1225 | */ |
2cf85583 TH |
1226 | error = mem_cgroup_try_charge_delay(page, current->mm, GFP_KERNEL, |
1227 | &memcg, false); | |
778dd893 HD |
1228 | if (error) |
1229 | goto out; | |
46f65ec1 | 1230 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 1231 | error = -EAGAIN; |
1da177e4 | 1232 | |
cb5f7b9a | 1233 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
1234 | list_for_each_safe(this, next, &shmem_swaplist) { |
1235 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 1236 | if (info->swapped) |
00501b53 | 1237 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
1238 | else |
1239 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1240 | cond_resched(); |
00501b53 | 1241 | if (error != -EAGAIN) |
778dd893 | 1242 | break; |
00501b53 | 1243 | /* found nothing in this: move on to search the next */ |
1da177e4 | 1244 | } |
cb5f7b9a | 1245 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1246 | |
00501b53 JW |
1247 | if (error) { |
1248 | if (error != -ENOMEM) | |
1249 | error = 0; | |
f627c2f5 | 1250 | mem_cgroup_cancel_charge(page, memcg, false); |
00501b53 | 1251 | } else |
f627c2f5 | 1252 | mem_cgroup_commit_charge(page, memcg, true, false); |
778dd893 | 1253 | out: |
aaa46865 | 1254 | unlock_page(page); |
09cbfeaf | 1255 | put_page(page); |
778dd893 | 1256 | return error; |
1da177e4 LT |
1257 | } |
1258 | ||
1259 | /* | |
1260 | * Move the page from the page cache to the swap cache. | |
1261 | */ | |
1262 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1263 | { | |
1264 | struct shmem_inode_info *info; | |
1da177e4 | 1265 | struct address_space *mapping; |
1da177e4 | 1266 | struct inode *inode; |
6922c0c7 HD |
1267 | swp_entry_t swap; |
1268 | pgoff_t index; | |
1da177e4 | 1269 | |
800d8c63 | 1270 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1271 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1272 | mapping = page->mapping; |
1273 | index = page->index; | |
1274 | inode = mapping->host; | |
1275 | info = SHMEM_I(inode); | |
1276 | if (info->flags & VM_LOCKED) | |
1277 | goto redirty; | |
d9fe526a | 1278 | if (!total_swap_pages) |
1da177e4 LT |
1279 | goto redirty; |
1280 | ||
d9fe526a | 1281 | /* |
97b713ba CH |
1282 | * Our capabilities prevent regular writeback or sync from ever calling |
1283 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1284 | * its underlying filesystem, in which case tmpfs should write out to | |
1285 | * swap only in response to memory pressure, and not for the writeback | |
1286 | * threads or sync. | |
d9fe526a | 1287 | */ |
48f170fb HD |
1288 | if (!wbc->for_reclaim) { |
1289 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1290 | goto redirty; | |
1291 | } | |
1635f6a7 HD |
1292 | |
1293 | /* | |
1294 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1295 | * value into swapfile.c, the only way we can correctly account for a | |
1296 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1297 | * |
1298 | * That's okay for a page already fallocated earlier, but if we have | |
1299 | * not yet completed the fallocation, then (a) we want to keep track | |
1300 | * of this page in case we have to undo it, and (b) it may not be a | |
1301 | * good idea to continue anyway, once we're pushing into swap. So | |
1302 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1303 | */ |
1304 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1305 | if (inode->i_private) { |
1306 | struct shmem_falloc *shmem_falloc; | |
1307 | spin_lock(&inode->i_lock); | |
1308 | shmem_falloc = inode->i_private; | |
1309 | if (shmem_falloc && | |
8e205f77 | 1310 | !shmem_falloc->waitq && |
1aac1400 HD |
1311 | index >= shmem_falloc->start && |
1312 | index < shmem_falloc->next) | |
1313 | shmem_falloc->nr_unswapped++; | |
1314 | else | |
1315 | shmem_falloc = NULL; | |
1316 | spin_unlock(&inode->i_lock); | |
1317 | if (shmem_falloc) | |
1318 | goto redirty; | |
1319 | } | |
1635f6a7 HD |
1320 | clear_highpage(page); |
1321 | flush_dcache_page(page); | |
1322 | SetPageUptodate(page); | |
1323 | } | |
1324 | ||
38d8b4e6 | 1325 | swap = get_swap_page(page); |
48f170fb HD |
1326 | if (!swap.val) |
1327 | goto redirty; | |
d9fe526a | 1328 | |
b1dea800 HD |
1329 | /* |
1330 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1331 | * if it's not already there. Do it now before the page is |
1332 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1333 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1334 | * we've incremented swapped, because shmem_unuse_inode() will |
1335 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1336 | */ |
48f170fb HD |
1337 | mutex_lock(&shmem_swaplist_mutex); |
1338 | if (list_empty(&info->swaplist)) | |
1339 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 1340 | |
48f170fb | 1341 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
4595ef88 | 1342 | spin_lock_irq(&info->lock); |
6922c0c7 | 1343 | shmem_recalc_inode(inode); |
267a4c76 | 1344 | info->swapped++; |
4595ef88 | 1345 | spin_unlock_irq(&info->lock); |
6922c0c7 | 1346 | |
267a4c76 HD |
1347 | swap_shmem_alloc(swap); |
1348 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1349 | ||
6922c0c7 | 1350 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1351 | BUG_ON(page_mapped(page)); |
9fab5619 | 1352 | swap_writepage(page, wbc); |
1da177e4 LT |
1353 | return 0; |
1354 | } | |
1355 | ||
6922c0c7 | 1356 | mutex_unlock(&shmem_swaplist_mutex); |
75f6d6d2 | 1357 | put_swap_page(page, swap); |
1da177e4 LT |
1358 | redirty: |
1359 | set_page_dirty(page); | |
d9fe526a HD |
1360 | if (wbc->for_reclaim) |
1361 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1362 | unlock_page(page); | |
1363 | return 0; | |
1da177e4 LT |
1364 | } |
1365 | ||
75edd345 | 1366 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1367 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1368 | { |
095f1fc4 | 1369 | char buffer[64]; |
680d794b | 1370 | |
71fe804b | 1371 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1372 | return; /* show nothing */ |
680d794b | 1373 | |
a7a88b23 | 1374 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1375 | |
1376 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1377 | } |
71fe804b LS |
1378 | |
1379 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1380 | { | |
1381 | struct mempolicy *mpol = NULL; | |
1382 | if (sbinfo->mpol) { | |
1383 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1384 | mpol = sbinfo->mpol; | |
1385 | mpol_get(mpol); | |
1386 | spin_unlock(&sbinfo->stat_lock); | |
1387 | } | |
1388 | return mpol; | |
1389 | } | |
75edd345 HD |
1390 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1391 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1392 | { | |
1393 | } | |
1394 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1395 | { | |
1396 | return NULL; | |
1397 | } | |
1398 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1399 | #ifndef CONFIG_NUMA | |
1400 | #define vm_policy vm_private_data | |
1401 | #endif | |
680d794b | 1402 | |
800d8c63 KS |
1403 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1404 | struct shmem_inode_info *info, pgoff_t index) | |
1405 | { | |
1406 | /* Create a pseudo vma that just contains the policy */ | |
2c4541e2 | 1407 | vma_init(vma, NULL); |
800d8c63 KS |
1408 | /* Bias interleave by inode number to distribute better across nodes */ |
1409 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
800d8c63 KS |
1410 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
1411 | } | |
1412 | ||
1413 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1414 | { | |
1415 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1416 | mpol_cond_put(vma->vm_policy); | |
1417 | } | |
1418 | ||
41ffe5d5 HD |
1419 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1420 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1421 | { |
1da177e4 | 1422 | struct vm_area_struct pvma; |
18a2f371 | 1423 | struct page *page; |
e9e9b7ec | 1424 | struct vm_fault vmf; |
52cd3b07 | 1425 | |
800d8c63 | 1426 | shmem_pseudo_vma_init(&pvma, info, index); |
e9e9b7ec MK |
1427 | vmf.vma = &pvma; |
1428 | vmf.address = 0; | |
1429 | page = swap_cluster_readahead(swap, gfp, &vmf); | |
800d8c63 | 1430 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1431 | |
800d8c63 KS |
1432 | return page; |
1433 | } | |
1434 | ||
1435 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1436 | struct shmem_inode_info *info, pgoff_t index) | |
1437 | { | |
1438 | struct vm_area_struct pvma; | |
1439 | struct inode *inode = &info->vfs_inode; | |
1440 | struct address_space *mapping = inode->i_mapping; | |
4620a06e | 1441 | pgoff_t idx, hindex; |
800d8c63 KS |
1442 | void __rcu **results; |
1443 | struct page *page; | |
1444 | ||
e496cf3d | 1445 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1446 | return NULL; |
1447 | ||
4620a06e | 1448 | hindex = round_down(index, HPAGE_PMD_NR); |
800d8c63 | 1449 | rcu_read_lock(); |
b93b0163 | 1450 | if (radix_tree_gang_lookup_slot(&mapping->i_pages, &results, &idx, |
800d8c63 KS |
1451 | hindex, 1) && idx < hindex + HPAGE_PMD_NR) { |
1452 | rcu_read_unlock(); | |
1453 | return NULL; | |
1454 | } | |
1455 | rcu_read_unlock(); | |
18a2f371 | 1456 | |
800d8c63 KS |
1457 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1458 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
1459 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); | |
1460 | shmem_pseudo_vma_destroy(&pvma); | |
1461 | if (page) | |
1462 | prep_transhuge_page(page); | |
18a2f371 | 1463 | return page; |
1da177e4 LT |
1464 | } |
1465 | ||
02098fea | 1466 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1467 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1468 | { |
1469 | struct vm_area_struct pvma; | |
18a2f371 | 1470 | struct page *page; |
1da177e4 | 1471 | |
800d8c63 KS |
1472 | shmem_pseudo_vma_init(&pvma, info, index); |
1473 | page = alloc_page_vma(gfp, &pvma, 0); | |
1474 | shmem_pseudo_vma_destroy(&pvma); | |
1475 | ||
1476 | return page; | |
1477 | } | |
1478 | ||
1479 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
0f079694 | 1480 | struct inode *inode, |
800d8c63 KS |
1481 | pgoff_t index, bool huge) |
1482 | { | |
0f079694 | 1483 | struct shmem_inode_info *info = SHMEM_I(inode); |
800d8c63 KS |
1484 | struct page *page; |
1485 | int nr; | |
1486 | int err = -ENOSPC; | |
52cd3b07 | 1487 | |
e496cf3d | 1488 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1489 | huge = false; |
1490 | nr = huge ? HPAGE_PMD_NR : 1; | |
1491 | ||
0f079694 | 1492 | if (!shmem_inode_acct_block(inode, nr)) |
800d8c63 | 1493 | goto failed; |
800d8c63 KS |
1494 | |
1495 | if (huge) | |
1496 | page = shmem_alloc_hugepage(gfp, info, index); | |
1497 | else | |
1498 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1499 | if (page) { |
1500 | __SetPageLocked(page); | |
1501 | __SetPageSwapBacked(page); | |
800d8c63 | 1502 | return page; |
75edd345 | 1503 | } |
18a2f371 | 1504 | |
800d8c63 | 1505 | err = -ENOMEM; |
0f079694 | 1506 | shmem_inode_unacct_blocks(inode, nr); |
800d8c63 KS |
1507 | failed: |
1508 | return ERR_PTR(err); | |
1da177e4 | 1509 | } |
71fe804b | 1510 | |
bde05d1c HD |
1511 | /* |
1512 | * When a page is moved from swapcache to shmem filecache (either by the | |
1513 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1514 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1515 | * ignorance of the mapping it belongs to. If that mapping has special | |
1516 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1517 | * we may need to copy to a suitable page before moving to filecache. | |
1518 | * | |
1519 | * In a future release, this may well be extended to respect cpuset and | |
1520 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1521 | * but for now it is a simple matter of zone. | |
1522 | */ | |
1523 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1524 | { | |
1525 | return page_zonenum(page) > gfp_zone(gfp); | |
1526 | } | |
1527 | ||
1528 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1529 | struct shmem_inode_info *info, pgoff_t index) | |
1530 | { | |
1531 | struct page *oldpage, *newpage; | |
1532 | struct address_space *swap_mapping; | |
1533 | pgoff_t swap_index; | |
1534 | int error; | |
1535 | ||
1536 | oldpage = *pagep; | |
1537 | swap_index = page_private(oldpage); | |
1538 | swap_mapping = page_mapping(oldpage); | |
1539 | ||
1540 | /* | |
1541 | * We have arrived here because our zones are constrained, so don't | |
1542 | * limit chance of success by further cpuset and node constraints. | |
1543 | */ | |
1544 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1545 | newpage = shmem_alloc_page(gfp, info, index); | |
1546 | if (!newpage) | |
1547 | return -ENOMEM; | |
bde05d1c | 1548 | |
09cbfeaf | 1549 | get_page(newpage); |
bde05d1c | 1550 | copy_highpage(newpage, oldpage); |
0142ef6c | 1551 | flush_dcache_page(newpage); |
bde05d1c | 1552 | |
9956edf3 HD |
1553 | __SetPageLocked(newpage); |
1554 | __SetPageSwapBacked(newpage); | |
bde05d1c | 1555 | SetPageUptodate(newpage); |
bde05d1c | 1556 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1557 | SetPageSwapCache(newpage); |
1558 | ||
1559 | /* | |
1560 | * Our caller will very soon move newpage out of swapcache, but it's | |
1561 | * a nice clean interface for us to replace oldpage by newpage there. | |
1562 | */ | |
b93b0163 | 1563 | xa_lock_irq(&swap_mapping->i_pages); |
62f945b6 | 1564 | error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage); |
0142ef6c | 1565 | if (!error) { |
11fb9989 MG |
1566 | __inc_node_page_state(newpage, NR_FILE_PAGES); |
1567 | __dec_node_page_state(oldpage, NR_FILE_PAGES); | |
0142ef6c | 1568 | } |
b93b0163 | 1569 | xa_unlock_irq(&swap_mapping->i_pages); |
bde05d1c | 1570 | |
0142ef6c HD |
1571 | if (unlikely(error)) { |
1572 | /* | |
1573 | * Is this possible? I think not, now that our callers check | |
1574 | * both PageSwapCache and page_private after getting page lock; | |
1575 | * but be defensive. Reverse old to newpage for clear and free. | |
1576 | */ | |
1577 | oldpage = newpage; | |
1578 | } else { | |
6a93ca8f | 1579 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1580 | lru_cache_add_anon(newpage); |
1581 | *pagep = newpage; | |
1582 | } | |
bde05d1c HD |
1583 | |
1584 | ClearPageSwapCache(oldpage); | |
1585 | set_page_private(oldpage, 0); | |
1586 | ||
1587 | unlock_page(oldpage); | |
09cbfeaf KS |
1588 | put_page(oldpage); |
1589 | put_page(oldpage); | |
0142ef6c | 1590 | return error; |
bde05d1c HD |
1591 | } |
1592 | ||
1da177e4 | 1593 | /* |
68da9f05 | 1594 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1595 | * |
1596 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1597 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1598 | * entry since a page cannot live in both the swap and page cache. |
1599 | * | |
1600 | * fault_mm and fault_type are only supplied by shmem_fault: | |
1601 | * otherwise they are NULL. | |
1da177e4 | 1602 | */ |
41ffe5d5 | 1603 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 1604 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
2b740303 SJ |
1605 | struct vm_area_struct *vma, struct vm_fault *vmf, |
1606 | vm_fault_t *fault_type) | |
1da177e4 LT |
1607 | { |
1608 | struct address_space *mapping = inode->i_mapping; | |
23f919d4 | 1609 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 | 1610 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1611 | struct mm_struct *charge_mm; |
00501b53 | 1612 | struct mem_cgroup *memcg; |
27ab7006 | 1613 | struct page *page; |
1da177e4 | 1614 | swp_entry_t swap; |
657e3038 | 1615 | enum sgp_type sgp_huge = sgp; |
800d8c63 | 1616 | pgoff_t hindex = index; |
1da177e4 | 1617 | int error; |
54af6042 | 1618 | int once = 0; |
1635f6a7 | 1619 | int alloced = 0; |
1da177e4 | 1620 | |
09cbfeaf | 1621 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1622 | return -EFBIG; |
657e3038 KS |
1623 | if (sgp == SGP_NOHUGE || sgp == SGP_HUGE) |
1624 | sgp = SGP_CACHE; | |
1da177e4 | 1625 | repeat: |
54af6042 | 1626 | swap.val = 0; |
0cd6144a | 1627 | page = find_lock_entry(mapping, index); |
3159f943 | 1628 | if (xa_is_value(page)) { |
54af6042 HD |
1629 | swap = radix_to_swp_entry(page); |
1630 | page = NULL; | |
1631 | } | |
1632 | ||
75edd345 | 1633 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1634 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 1635 | error = -EINVAL; |
267a4c76 | 1636 | goto unlock; |
54af6042 HD |
1637 | } |
1638 | ||
66d2f4d2 HD |
1639 | if (page && sgp == SGP_WRITE) |
1640 | mark_page_accessed(page); | |
1641 | ||
1635f6a7 HD |
1642 | /* fallocated page? */ |
1643 | if (page && !PageUptodate(page)) { | |
1644 | if (sgp != SGP_READ) | |
1645 | goto clear; | |
1646 | unlock_page(page); | |
09cbfeaf | 1647 | put_page(page); |
1635f6a7 HD |
1648 | page = NULL; |
1649 | } | |
54af6042 | 1650 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1651 | *pagep = page; |
1652 | return 0; | |
27ab7006 HD |
1653 | } |
1654 | ||
1655 | /* | |
54af6042 HD |
1656 | * Fast cache lookup did not find it: |
1657 | * bring it back from swap or allocate. | |
27ab7006 | 1658 | */ |
54af6042 | 1659 | sbinfo = SHMEM_SB(inode->i_sb); |
cfda0526 | 1660 | charge_mm = vma ? vma->vm_mm : current->mm; |
1da177e4 | 1661 | |
1da177e4 LT |
1662 | if (swap.val) { |
1663 | /* Look it up and read it in.. */ | |
ec560175 | 1664 | page = lookup_swap_cache(swap, NULL, 0); |
27ab7006 | 1665 | if (!page) { |
9e18eb29 ALC |
1666 | /* Or update major stats only when swapin succeeds?? */ |
1667 | if (fault_type) { | |
68da9f05 | 1668 | *fault_type |= VM_FAULT_MAJOR; |
9e18eb29 | 1669 | count_vm_event(PGMAJFAULT); |
2262185c | 1670 | count_memcg_event_mm(charge_mm, PGMAJFAULT); |
9e18eb29 ALC |
1671 | } |
1672 | /* Here we actually start the io */ | |
41ffe5d5 | 1673 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1674 | if (!page) { |
54af6042 HD |
1675 | error = -ENOMEM; |
1676 | goto failed; | |
1da177e4 | 1677 | } |
1da177e4 LT |
1678 | } |
1679 | ||
1680 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1681 | lock_page(page); |
0142ef6c | 1682 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1683 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1684 | error = -EEXIST; /* try again */ |
d1899228 | 1685 | goto unlock; |
bde05d1c | 1686 | } |
27ab7006 | 1687 | if (!PageUptodate(page)) { |
1da177e4 | 1688 | error = -EIO; |
54af6042 | 1689 | goto failed; |
1da177e4 | 1690 | } |
54af6042 HD |
1691 | wait_on_page_writeback(page); |
1692 | ||
bde05d1c HD |
1693 | if (shmem_should_replace_page(page, gfp)) { |
1694 | error = shmem_replace_page(&page, gfp, info, index); | |
1695 | if (error) | |
1696 | goto failed; | |
1da177e4 | 1697 | } |
27ab7006 | 1698 | |
2cf85583 | 1699 | error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg, |
f627c2f5 | 1700 | false); |
d1899228 | 1701 | if (!error) { |
aa3b1895 | 1702 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1703 | swp_to_radix_entry(swap)); |
215c02bc HD |
1704 | /* |
1705 | * We already confirmed swap under page lock, and make | |
1706 | * no memory allocation here, so usually no possibility | |
1707 | * of error; but free_swap_and_cache() only trylocks a | |
1708 | * page, so it is just possible that the entry has been | |
1709 | * truncated or holepunched since swap was confirmed. | |
1710 | * shmem_undo_range() will have done some of the | |
1711 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1712 | * the rest. |
215c02bc HD |
1713 | * Reset swap.val? No, leave it so "failed" goes back to |
1714 | * "repeat": reading a hole and writing should succeed. | |
1715 | */ | |
00501b53 | 1716 | if (error) { |
f627c2f5 | 1717 | mem_cgroup_cancel_charge(page, memcg, false); |
215c02bc | 1718 | delete_from_swap_cache(page); |
00501b53 | 1719 | } |
d1899228 | 1720 | } |
54af6042 HD |
1721 | if (error) |
1722 | goto failed; | |
1723 | ||
f627c2f5 | 1724 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1725 | |
4595ef88 | 1726 | spin_lock_irq(&info->lock); |
285b2c4f | 1727 | info->swapped--; |
54af6042 | 1728 | shmem_recalc_inode(inode); |
4595ef88 | 1729 | spin_unlock_irq(&info->lock); |
54af6042 | 1730 | |
66d2f4d2 HD |
1731 | if (sgp == SGP_WRITE) |
1732 | mark_page_accessed(page); | |
1733 | ||
54af6042 | 1734 | delete_from_swap_cache(page); |
27ab7006 HD |
1735 | set_page_dirty(page); |
1736 | swap_free(swap); | |
1737 | ||
54af6042 | 1738 | } else { |
cfda0526 MR |
1739 | if (vma && userfaultfd_missing(vma)) { |
1740 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
1741 | return 0; | |
1742 | } | |
1743 | ||
800d8c63 KS |
1744 | /* shmem_symlink() */ |
1745 | if (mapping->a_ops != &shmem_aops) | |
1746 | goto alloc_nohuge; | |
657e3038 | 1747 | if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE) |
800d8c63 KS |
1748 | goto alloc_nohuge; |
1749 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1750 | goto alloc_huge; | |
1751 | switch (sbinfo->huge) { | |
1752 | loff_t i_size; | |
1753 | pgoff_t off; | |
1754 | case SHMEM_HUGE_NEVER: | |
1755 | goto alloc_nohuge; | |
1756 | case SHMEM_HUGE_WITHIN_SIZE: | |
1757 | off = round_up(index, HPAGE_PMD_NR); | |
1758 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1759 | if (i_size >= HPAGE_PMD_SIZE && | |
1760 | i_size >> PAGE_SHIFT >= off) | |
1761 | goto alloc_huge; | |
1762 | /* fallthrough */ | |
1763 | case SHMEM_HUGE_ADVISE: | |
657e3038 KS |
1764 | if (sgp_huge == SGP_HUGE) |
1765 | goto alloc_huge; | |
1766 | /* TODO: implement fadvise() hints */ | |
800d8c63 | 1767 | goto alloc_nohuge; |
54af6042 | 1768 | } |
1da177e4 | 1769 | |
800d8c63 | 1770 | alloc_huge: |
0f079694 | 1771 | page = shmem_alloc_and_acct_page(gfp, inode, index, true); |
800d8c63 | 1772 | if (IS_ERR(page)) { |
0f079694 | 1773 | alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, inode, |
800d8c63 | 1774 | index, false); |
1da177e4 | 1775 | } |
800d8c63 | 1776 | if (IS_ERR(page)) { |
779750d2 | 1777 | int retry = 5; |
800d8c63 KS |
1778 | error = PTR_ERR(page); |
1779 | page = NULL; | |
779750d2 KS |
1780 | if (error != -ENOSPC) |
1781 | goto failed; | |
1782 | /* | |
1783 | * Try to reclaim some spece by splitting a huge page | |
1784 | * beyond i_size on the filesystem. | |
1785 | */ | |
1786 | while (retry--) { | |
1787 | int ret; | |
1788 | ret = shmem_unused_huge_shrink(sbinfo, NULL, 1); | |
1789 | if (ret == SHRINK_STOP) | |
1790 | break; | |
1791 | if (ret) | |
1792 | goto alloc_nohuge; | |
1793 | } | |
800d8c63 KS |
1794 | goto failed; |
1795 | } | |
1796 | ||
1797 | if (PageTransHuge(page)) | |
1798 | hindex = round_down(index, HPAGE_PMD_NR); | |
1799 | else | |
1800 | hindex = index; | |
1801 | ||
66d2f4d2 | 1802 | if (sgp == SGP_WRITE) |
eb39d618 | 1803 | __SetPageReferenced(page); |
66d2f4d2 | 1804 | |
2cf85583 | 1805 | error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg, |
800d8c63 | 1806 | PageTransHuge(page)); |
54af6042 | 1807 | if (error) |
800d8c63 KS |
1808 | goto unacct; |
1809 | error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK, | |
1810 | compound_order(page)); | |
b065b432 | 1811 | if (!error) { |
800d8c63 | 1812 | error = shmem_add_to_page_cache(page, mapping, hindex, |
fed400a1 | 1813 | NULL); |
b065b432 HD |
1814 | radix_tree_preload_end(); |
1815 | } | |
1816 | if (error) { | |
800d8c63 KS |
1817 | mem_cgroup_cancel_charge(page, memcg, |
1818 | PageTransHuge(page)); | |
1819 | goto unacct; | |
b065b432 | 1820 | } |
800d8c63 KS |
1821 | mem_cgroup_commit_charge(page, memcg, false, |
1822 | PageTransHuge(page)); | |
54af6042 HD |
1823 | lru_cache_add_anon(page); |
1824 | ||
4595ef88 | 1825 | spin_lock_irq(&info->lock); |
800d8c63 KS |
1826 | info->alloced += 1 << compound_order(page); |
1827 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); | |
54af6042 | 1828 | shmem_recalc_inode(inode); |
4595ef88 | 1829 | spin_unlock_irq(&info->lock); |
1635f6a7 | 1830 | alloced = true; |
54af6042 | 1831 | |
779750d2 KS |
1832 | if (PageTransHuge(page) && |
1833 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < | |
1834 | hindex + HPAGE_PMD_NR - 1) { | |
1835 | /* | |
1836 | * Part of the huge page is beyond i_size: subject | |
1837 | * to shrink under memory pressure. | |
1838 | */ | |
1839 | spin_lock(&sbinfo->shrinklist_lock); | |
d041353d CW |
1840 | /* |
1841 | * _careful to defend against unlocked access to | |
1842 | * ->shrink_list in shmem_unused_huge_shrink() | |
1843 | */ | |
1844 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1845 | list_add_tail(&info->shrinklist, |
1846 | &sbinfo->shrinklist); | |
1847 | sbinfo->shrinklist_len++; | |
1848 | } | |
1849 | spin_unlock(&sbinfo->shrinklist_lock); | |
1850 | } | |
1851 | ||
ec9516fb | 1852 | /* |
1635f6a7 HD |
1853 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1854 | */ | |
1855 | if (sgp == SGP_FALLOC) | |
1856 | sgp = SGP_WRITE; | |
1857 | clear: | |
1858 | /* | |
1859 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1860 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1861 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb | 1862 | */ |
800d8c63 KS |
1863 | if (sgp != SGP_WRITE && !PageUptodate(page)) { |
1864 | struct page *head = compound_head(page); | |
1865 | int i; | |
1866 | ||
1867 | for (i = 0; i < (1 << compound_order(head)); i++) { | |
1868 | clear_highpage(head + i); | |
1869 | flush_dcache_page(head + i); | |
1870 | } | |
1871 | SetPageUptodate(head); | |
ec9516fb | 1872 | } |
1da177e4 | 1873 | } |
bde05d1c | 1874 | |
54af6042 | 1875 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1876 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1877 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1878 | if (alloced) { |
1879 | ClearPageDirty(page); | |
1880 | delete_from_page_cache(page); | |
4595ef88 | 1881 | spin_lock_irq(&info->lock); |
267a4c76 | 1882 | shmem_recalc_inode(inode); |
4595ef88 | 1883 | spin_unlock_irq(&info->lock); |
267a4c76 | 1884 | } |
54af6042 | 1885 | error = -EINVAL; |
267a4c76 | 1886 | goto unlock; |
e83c32e8 | 1887 | } |
800d8c63 | 1888 | *pagep = page + index - hindex; |
54af6042 | 1889 | return 0; |
1da177e4 | 1890 | |
59a16ead | 1891 | /* |
54af6042 | 1892 | * Error recovery. |
59a16ead | 1893 | */ |
54af6042 | 1894 | unacct: |
0f079694 | 1895 | shmem_inode_unacct_blocks(inode, 1 << compound_order(page)); |
800d8c63 KS |
1896 | |
1897 | if (PageTransHuge(page)) { | |
1898 | unlock_page(page); | |
1899 | put_page(page); | |
1900 | goto alloc_nohuge; | |
1901 | } | |
54af6042 | 1902 | failed: |
267a4c76 | 1903 | if (swap.val && !shmem_confirm_swap(mapping, index, swap)) |
d1899228 HD |
1904 | error = -EEXIST; |
1905 | unlock: | |
27ab7006 | 1906 | if (page) { |
54af6042 | 1907 | unlock_page(page); |
09cbfeaf | 1908 | put_page(page); |
54af6042 HD |
1909 | } |
1910 | if (error == -ENOSPC && !once++) { | |
4595ef88 | 1911 | spin_lock_irq(&info->lock); |
54af6042 | 1912 | shmem_recalc_inode(inode); |
4595ef88 | 1913 | spin_unlock_irq(&info->lock); |
27ab7006 | 1914 | goto repeat; |
ff36b801 | 1915 | } |
d1899228 | 1916 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1917 | goto repeat; |
1918 | return error; | |
1da177e4 LT |
1919 | } |
1920 | ||
10d20bd2 LT |
1921 | /* |
1922 | * This is like autoremove_wake_function, but it removes the wait queue | |
1923 | * entry unconditionally - even if something else had already woken the | |
1924 | * target. | |
1925 | */ | |
ac6424b9 | 1926 | static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
10d20bd2 LT |
1927 | { |
1928 | int ret = default_wake_function(wait, mode, sync, key); | |
2055da97 | 1929 | list_del_init(&wait->entry); |
10d20bd2 LT |
1930 | return ret; |
1931 | } | |
1932 | ||
20acce67 | 1933 | static vm_fault_t shmem_fault(struct vm_fault *vmf) |
1da177e4 | 1934 | { |
11bac800 | 1935 | struct vm_area_struct *vma = vmf->vma; |
496ad9aa | 1936 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 1937 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
657e3038 | 1938 | enum sgp_type sgp; |
20acce67 SJ |
1939 | int err; |
1940 | vm_fault_t ret = VM_FAULT_LOCKED; | |
1da177e4 | 1941 | |
f00cdc6d HD |
1942 | /* |
1943 | * Trinity finds that probing a hole which tmpfs is punching can | |
1944 | * prevent the hole-punch from ever completing: which in turn | |
1945 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1946 | * faulting pages into the hole while it's being punched. Although |
1947 | * shmem_undo_range() does remove the additions, it may be unable to | |
1948 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1949 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1950 | * | |
1951 | * It does not matter if we sometimes reach this check just before the | |
1952 | * hole-punch begins, so that one fault then races with the punch: | |
1953 | * we just need to make racing faults a rare case. | |
1954 | * | |
1955 | * The implementation below would be much simpler if we just used a | |
1956 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1957 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1958 | */ |
1959 | if (unlikely(inode->i_private)) { | |
1960 | struct shmem_falloc *shmem_falloc; | |
1961 | ||
1962 | spin_lock(&inode->i_lock); | |
1963 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1964 | if (shmem_falloc && |
1965 | shmem_falloc->waitq && | |
1966 | vmf->pgoff >= shmem_falloc->start && | |
1967 | vmf->pgoff < shmem_falloc->next) { | |
1968 | wait_queue_head_t *shmem_falloc_waitq; | |
10d20bd2 | 1969 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); |
8e205f77 HD |
1970 | |
1971 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1972 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1973 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1974 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1975 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1976 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1977 | } |
8e205f77 HD |
1978 | |
1979 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1980 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1981 | TASK_UNINTERRUPTIBLE); | |
1982 | spin_unlock(&inode->i_lock); | |
1983 | schedule(); | |
1984 | ||
1985 | /* | |
1986 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1987 | * stack of the hole-punching task: shmem_falloc_waitq | |
1988 | * is usually invalid by the time we reach here, but | |
1989 | * finish_wait() does not dereference it in that case; | |
1990 | * though i_lock needed lest racing with wake_up_all(). | |
1991 | */ | |
1992 | spin_lock(&inode->i_lock); | |
1993 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1994 | spin_unlock(&inode->i_lock); | |
1995 | return ret; | |
f00cdc6d | 1996 | } |
8e205f77 | 1997 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1998 | } |
1999 | ||
657e3038 | 2000 | sgp = SGP_CACHE; |
18600332 MH |
2001 | |
2002 | if ((vma->vm_flags & VM_NOHUGEPAGE) || | |
2003 | test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) | |
657e3038 | 2004 | sgp = SGP_NOHUGE; |
18600332 MH |
2005 | else if (vma->vm_flags & VM_HUGEPAGE) |
2006 | sgp = SGP_HUGE; | |
657e3038 | 2007 | |
20acce67 | 2008 | err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp, |
cfda0526 | 2009 | gfp, vma, vmf, &ret); |
20acce67 SJ |
2010 | if (err) |
2011 | return vmf_error(err); | |
68da9f05 | 2012 | return ret; |
1da177e4 LT |
2013 | } |
2014 | ||
c01d5b30 HD |
2015 | unsigned long shmem_get_unmapped_area(struct file *file, |
2016 | unsigned long uaddr, unsigned long len, | |
2017 | unsigned long pgoff, unsigned long flags) | |
2018 | { | |
2019 | unsigned long (*get_area)(struct file *, | |
2020 | unsigned long, unsigned long, unsigned long, unsigned long); | |
2021 | unsigned long addr; | |
2022 | unsigned long offset; | |
2023 | unsigned long inflated_len; | |
2024 | unsigned long inflated_addr; | |
2025 | unsigned long inflated_offset; | |
2026 | ||
2027 | if (len > TASK_SIZE) | |
2028 | return -ENOMEM; | |
2029 | ||
2030 | get_area = current->mm->get_unmapped_area; | |
2031 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2032 | ||
e496cf3d | 2033 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
c01d5b30 HD |
2034 | return addr; |
2035 | if (IS_ERR_VALUE(addr)) | |
2036 | return addr; | |
2037 | if (addr & ~PAGE_MASK) | |
2038 | return addr; | |
2039 | if (addr > TASK_SIZE - len) | |
2040 | return addr; | |
2041 | ||
2042 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2043 | return addr; | |
2044 | if (len < HPAGE_PMD_SIZE) | |
2045 | return addr; | |
2046 | if (flags & MAP_FIXED) | |
2047 | return addr; | |
2048 | /* | |
2049 | * Our priority is to support MAP_SHARED mapped hugely; | |
2050 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
2051 | * But if caller specified an address hint, respect that as before. | |
2052 | */ | |
2053 | if (uaddr) | |
2054 | return addr; | |
2055 | ||
2056 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2057 | struct super_block *sb; | |
2058 | ||
2059 | if (file) { | |
2060 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2061 | sb = file_inode(file)->i_sb; | |
2062 | } else { | |
2063 | /* | |
2064 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2065 | * for "/dev/zero", to create a shared anonymous object. | |
2066 | */ | |
2067 | if (IS_ERR(shm_mnt)) | |
2068 | return addr; | |
2069 | sb = shm_mnt->mnt_sb; | |
2070 | } | |
3089bf61 | 2071 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2072 | return addr; |
2073 | } | |
2074 | ||
2075 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2076 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2077 | return addr; | |
2078 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2079 | return addr; | |
2080 | ||
2081 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2082 | if (inflated_len > TASK_SIZE) | |
2083 | return addr; | |
2084 | if (inflated_len < len) | |
2085 | return addr; | |
2086 | ||
2087 | inflated_addr = get_area(NULL, 0, inflated_len, 0, flags); | |
2088 | if (IS_ERR_VALUE(inflated_addr)) | |
2089 | return addr; | |
2090 | if (inflated_addr & ~PAGE_MASK) | |
2091 | return addr; | |
2092 | ||
2093 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2094 | inflated_addr += offset - inflated_offset; | |
2095 | if (inflated_offset > offset) | |
2096 | inflated_addr += HPAGE_PMD_SIZE; | |
2097 | ||
2098 | if (inflated_addr > TASK_SIZE - len) | |
2099 | return addr; | |
2100 | return inflated_addr; | |
2101 | } | |
2102 | ||
1da177e4 | 2103 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2104 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2105 | { |
496ad9aa | 2106 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2107 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2108 | } |
2109 | ||
d8dc74f2 AB |
2110 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
2111 | unsigned long addr) | |
1da177e4 | 2112 | { |
496ad9aa | 2113 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2114 | pgoff_t index; |
1da177e4 | 2115 | |
41ffe5d5 HD |
2116 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2117 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
2118 | } |
2119 | #endif | |
2120 | ||
2121 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
2122 | { | |
496ad9aa | 2123 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2124 | struct shmem_inode_info *info = SHMEM_I(inode); |
2125 | int retval = -ENOMEM; | |
2126 | ||
4595ef88 | 2127 | spin_lock_irq(&info->lock); |
1da177e4 LT |
2128 | if (lock && !(info->flags & VM_LOCKED)) { |
2129 | if (!user_shm_lock(inode->i_size, user)) | |
2130 | goto out_nomem; | |
2131 | info->flags |= VM_LOCKED; | |
89e004ea | 2132 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
2133 | } |
2134 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
2135 | user_shm_unlock(inode->i_size, user); | |
2136 | info->flags &= ~VM_LOCKED; | |
89e004ea | 2137 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2138 | } |
2139 | retval = 0; | |
89e004ea | 2140 | |
1da177e4 | 2141 | out_nomem: |
4595ef88 | 2142 | spin_unlock_irq(&info->lock); |
1da177e4 LT |
2143 | return retval; |
2144 | } | |
2145 | ||
9b83a6a8 | 2146 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
2147 | { |
2148 | file_accessed(file); | |
2149 | vma->vm_ops = &shmem_vm_ops; | |
e496cf3d | 2150 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
2151 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
2152 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
2153 | khugepaged_enter(vma, vma->vm_flags); | |
2154 | } | |
1da177e4 LT |
2155 | return 0; |
2156 | } | |
2157 | ||
454abafe | 2158 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 2159 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
2160 | { |
2161 | struct inode *inode; | |
2162 | struct shmem_inode_info *info; | |
2163 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
2164 | ||
5b04c689 PE |
2165 | if (shmem_reserve_inode(sb)) |
2166 | return NULL; | |
1da177e4 LT |
2167 | |
2168 | inode = new_inode(sb); | |
2169 | if (inode) { | |
85fe4025 | 2170 | inode->i_ino = get_next_ino(); |
454abafe | 2171 | inode_init_owner(inode, dir, mode); |
1da177e4 | 2172 | inode->i_blocks = 0; |
078cd827 | 2173 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
46c9a946 | 2174 | inode->i_generation = prandom_u32(); |
1da177e4 LT |
2175 | info = SHMEM_I(inode); |
2176 | memset(info, 0, (char *)inode - (char *)info); | |
2177 | spin_lock_init(&info->lock); | |
40e041a2 | 2178 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 2179 | info->flags = flags & VM_NORESERVE; |
779750d2 | 2180 | INIT_LIST_HEAD(&info->shrinklist); |
1da177e4 | 2181 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 2182 | simple_xattrs_init(&info->xattrs); |
72c04902 | 2183 | cache_no_acl(inode); |
1da177e4 LT |
2184 | |
2185 | switch (mode & S_IFMT) { | |
2186 | default: | |
39f0247d | 2187 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
2188 | init_special_inode(inode, mode, dev); |
2189 | break; | |
2190 | case S_IFREG: | |
14fcc23f | 2191 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
2192 | inode->i_op = &shmem_inode_operations; |
2193 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
2194 | mpol_shared_policy_init(&info->policy, |
2195 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
2196 | break; |
2197 | case S_IFDIR: | |
d8c76e6f | 2198 | inc_nlink(inode); |
1da177e4 LT |
2199 | /* Some things misbehave if size == 0 on a directory */ |
2200 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2201 | inode->i_op = &shmem_dir_inode_operations; | |
2202 | inode->i_fop = &simple_dir_operations; | |
2203 | break; | |
2204 | case S_IFLNK: | |
2205 | /* | |
2206 | * Must not load anything in the rbtree, | |
2207 | * mpol_free_shared_policy will not be called. | |
2208 | */ | |
71fe804b | 2209 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
2210 | break; |
2211 | } | |
b45d71fb JFG |
2212 | |
2213 | lockdep_annotate_inode_mutex_key(inode); | |
5b04c689 PE |
2214 | } else |
2215 | shmem_free_inode(sb); | |
1da177e4 LT |
2216 | return inode; |
2217 | } | |
2218 | ||
0cd6144a JW |
2219 | bool shmem_mapping(struct address_space *mapping) |
2220 | { | |
f8005451 | 2221 | return mapping->a_ops == &shmem_aops; |
0cd6144a JW |
2222 | } |
2223 | ||
8d103963 MR |
2224 | static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm, |
2225 | pmd_t *dst_pmd, | |
2226 | struct vm_area_struct *dst_vma, | |
2227 | unsigned long dst_addr, | |
2228 | unsigned long src_addr, | |
2229 | bool zeropage, | |
2230 | struct page **pagep) | |
4c27fe4c MR |
2231 | { |
2232 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2233 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4c27fe4c MR |
2234 | struct address_space *mapping = inode->i_mapping; |
2235 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2236 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
2237 | struct mem_cgroup *memcg; | |
2238 | spinlock_t *ptl; | |
2239 | void *page_kaddr; | |
2240 | struct page *page; | |
2241 | pte_t _dst_pte, *dst_pte; | |
2242 | int ret; | |
2243 | ||
cb658a45 | 2244 | ret = -ENOMEM; |
0f079694 | 2245 | if (!shmem_inode_acct_block(inode, 1)) |
cb658a45 | 2246 | goto out; |
4c27fe4c | 2247 | |
cb658a45 | 2248 | if (!*pagep) { |
4c27fe4c MR |
2249 | page = shmem_alloc_page(gfp, info, pgoff); |
2250 | if (!page) | |
0f079694 | 2251 | goto out_unacct_blocks; |
4c27fe4c | 2252 | |
8d103963 MR |
2253 | if (!zeropage) { /* mcopy_atomic */ |
2254 | page_kaddr = kmap_atomic(page); | |
2255 | ret = copy_from_user(page_kaddr, | |
2256 | (const void __user *)src_addr, | |
2257 | PAGE_SIZE); | |
2258 | kunmap_atomic(page_kaddr); | |
2259 | ||
2260 | /* fallback to copy_from_user outside mmap_sem */ | |
2261 | if (unlikely(ret)) { | |
2262 | *pagep = page; | |
2263 | shmem_inode_unacct_blocks(inode, 1); | |
2264 | /* don't free the page */ | |
2265 | return -EFAULT; | |
2266 | } | |
2267 | } else { /* mfill_zeropage_atomic */ | |
2268 | clear_highpage(page); | |
4c27fe4c MR |
2269 | } |
2270 | } else { | |
2271 | page = *pagep; | |
2272 | *pagep = NULL; | |
2273 | } | |
2274 | ||
9cc90c66 AA |
2275 | VM_BUG_ON(PageLocked(page) || PageSwapBacked(page)); |
2276 | __SetPageLocked(page); | |
2277 | __SetPageSwapBacked(page); | |
a425d358 | 2278 | __SetPageUptodate(page); |
9cc90c66 | 2279 | |
2cf85583 | 2280 | ret = mem_cgroup_try_charge_delay(page, dst_mm, gfp, &memcg, false); |
4c27fe4c MR |
2281 | if (ret) |
2282 | goto out_release; | |
2283 | ||
2284 | ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK); | |
2285 | if (!ret) { | |
2286 | ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL); | |
2287 | radix_tree_preload_end(); | |
2288 | } | |
2289 | if (ret) | |
2290 | goto out_release_uncharge; | |
2291 | ||
2292 | mem_cgroup_commit_charge(page, memcg, false, false); | |
2293 | ||
2294 | _dst_pte = mk_pte(page, dst_vma->vm_page_prot); | |
2295 | if (dst_vma->vm_flags & VM_WRITE) | |
2296 | _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte)); | |
2297 | ||
2298 | ret = -EEXIST; | |
2299 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); | |
2300 | if (!pte_none(*dst_pte)) | |
2301 | goto out_release_uncharge_unlock; | |
2302 | ||
4c27fe4c MR |
2303 | lru_cache_add_anon(page); |
2304 | ||
2305 | spin_lock(&info->lock); | |
2306 | info->alloced++; | |
2307 | inode->i_blocks += BLOCKS_PER_PAGE; | |
2308 | shmem_recalc_inode(inode); | |
2309 | spin_unlock(&info->lock); | |
2310 | ||
2311 | inc_mm_counter(dst_mm, mm_counter_file(page)); | |
2312 | page_add_file_rmap(page, false); | |
2313 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
2314 | ||
2315 | /* No need to invalidate - it was non-present before */ | |
2316 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
2317 | unlock_page(page); | |
2318 | pte_unmap_unlock(dst_pte, ptl); | |
2319 | ret = 0; | |
2320 | out: | |
2321 | return ret; | |
2322 | out_release_uncharge_unlock: | |
2323 | pte_unmap_unlock(dst_pte, ptl); | |
2324 | out_release_uncharge: | |
2325 | mem_cgroup_cancel_charge(page, memcg, false); | |
2326 | out_release: | |
9cc90c66 | 2327 | unlock_page(page); |
4c27fe4c | 2328 | put_page(page); |
4c27fe4c | 2329 | out_unacct_blocks: |
0f079694 | 2330 | shmem_inode_unacct_blocks(inode, 1); |
4c27fe4c MR |
2331 | goto out; |
2332 | } | |
2333 | ||
8d103963 MR |
2334 | int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm, |
2335 | pmd_t *dst_pmd, | |
2336 | struct vm_area_struct *dst_vma, | |
2337 | unsigned long dst_addr, | |
2338 | unsigned long src_addr, | |
2339 | struct page **pagep) | |
2340 | { | |
2341 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2342 | dst_addr, src_addr, false, pagep); | |
2343 | } | |
2344 | ||
2345 | int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm, | |
2346 | pmd_t *dst_pmd, | |
2347 | struct vm_area_struct *dst_vma, | |
2348 | unsigned long dst_addr) | |
2349 | { | |
2350 | struct page *page = NULL; | |
2351 | ||
2352 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2353 | dst_addr, 0, true, &page); | |
2354 | } | |
2355 | ||
1da177e4 | 2356 | #ifdef CONFIG_TMPFS |
92e1d5be | 2357 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2358 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2359 | |
6d9d88d0 JS |
2360 | #ifdef CONFIG_TMPFS_XATTR |
2361 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2362 | #else | |
2363 | #define shmem_initxattrs NULL | |
2364 | #endif | |
2365 | ||
1da177e4 | 2366 | static int |
800d15a5 NP |
2367 | shmem_write_begin(struct file *file, struct address_space *mapping, |
2368 | loff_t pos, unsigned len, unsigned flags, | |
2369 | struct page **pagep, void **fsdata) | |
1da177e4 | 2370 | { |
800d15a5 | 2371 | struct inode *inode = mapping->host; |
40e041a2 | 2372 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2373 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
2374 | |
2375 | /* i_mutex is held by caller */ | |
3f472cc9 | 2376 | if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) { |
40e041a2 DH |
2377 | if (info->seals & F_SEAL_WRITE) |
2378 | return -EPERM; | |
2379 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2380 | return -EPERM; | |
2381 | } | |
2382 | ||
9e18eb29 | 2383 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
2384 | } |
2385 | ||
2386 | static int | |
2387 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2388 | loff_t pos, unsigned len, unsigned copied, | |
2389 | struct page *page, void *fsdata) | |
2390 | { | |
2391 | struct inode *inode = mapping->host; | |
2392 | ||
d3602444 HD |
2393 | if (pos + copied > inode->i_size) |
2394 | i_size_write(inode, pos + copied); | |
2395 | ||
ec9516fb | 2396 | if (!PageUptodate(page)) { |
800d8c63 KS |
2397 | struct page *head = compound_head(page); |
2398 | if (PageTransCompound(page)) { | |
2399 | int i; | |
2400 | ||
2401 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2402 | if (head + i == page) | |
2403 | continue; | |
2404 | clear_highpage(head + i); | |
2405 | flush_dcache_page(head + i); | |
2406 | } | |
2407 | } | |
09cbfeaf KS |
2408 | if (copied < PAGE_SIZE) { |
2409 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2410 | zero_user_segments(page, 0, from, |
09cbfeaf | 2411 | from + copied, PAGE_SIZE); |
ec9516fb | 2412 | } |
800d8c63 | 2413 | SetPageUptodate(head); |
ec9516fb | 2414 | } |
800d15a5 | 2415 | set_page_dirty(page); |
6746aff7 | 2416 | unlock_page(page); |
09cbfeaf | 2417 | put_page(page); |
800d15a5 | 2418 | |
800d15a5 | 2419 | return copied; |
1da177e4 LT |
2420 | } |
2421 | ||
2ba5bbed | 2422 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2423 | { |
6e58e79d AV |
2424 | struct file *file = iocb->ki_filp; |
2425 | struct inode *inode = file_inode(file); | |
1da177e4 | 2426 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2427 | pgoff_t index; |
2428 | unsigned long offset; | |
a0ee5ec5 | 2429 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2430 | int error = 0; |
cb66a7a1 | 2431 | ssize_t retval = 0; |
6e58e79d | 2432 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2433 | |
2434 | /* | |
2435 | * Might this read be for a stacking filesystem? Then when reading | |
2436 | * holes of a sparse file, we actually need to allocate those pages, | |
2437 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2438 | */ | |
777eda2c | 2439 | if (!iter_is_iovec(to)) |
75edd345 | 2440 | sgp = SGP_CACHE; |
1da177e4 | 2441 | |
09cbfeaf KS |
2442 | index = *ppos >> PAGE_SHIFT; |
2443 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2444 | |
2445 | for (;;) { | |
2446 | struct page *page = NULL; | |
41ffe5d5 HD |
2447 | pgoff_t end_index; |
2448 | unsigned long nr, ret; | |
1da177e4 LT |
2449 | loff_t i_size = i_size_read(inode); |
2450 | ||
09cbfeaf | 2451 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2452 | if (index > end_index) |
2453 | break; | |
2454 | if (index == end_index) { | |
09cbfeaf | 2455 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2456 | if (nr <= offset) |
2457 | break; | |
2458 | } | |
2459 | ||
9e18eb29 | 2460 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2461 | if (error) { |
2462 | if (error == -EINVAL) | |
2463 | error = 0; | |
1da177e4 LT |
2464 | break; |
2465 | } | |
75edd345 HD |
2466 | if (page) { |
2467 | if (sgp == SGP_CACHE) | |
2468 | set_page_dirty(page); | |
d3602444 | 2469 | unlock_page(page); |
75edd345 | 2470 | } |
1da177e4 LT |
2471 | |
2472 | /* | |
2473 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2474 | * are called without i_mutex protection against truncate |
1da177e4 | 2475 | */ |
09cbfeaf | 2476 | nr = PAGE_SIZE; |
1da177e4 | 2477 | i_size = i_size_read(inode); |
09cbfeaf | 2478 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2479 | if (index == end_index) { |
09cbfeaf | 2480 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2481 | if (nr <= offset) { |
2482 | if (page) | |
09cbfeaf | 2483 | put_page(page); |
1da177e4 LT |
2484 | break; |
2485 | } | |
2486 | } | |
2487 | nr -= offset; | |
2488 | ||
2489 | if (page) { | |
2490 | /* | |
2491 | * If users can be writing to this page using arbitrary | |
2492 | * virtual addresses, take care about potential aliasing | |
2493 | * before reading the page on the kernel side. | |
2494 | */ | |
2495 | if (mapping_writably_mapped(mapping)) | |
2496 | flush_dcache_page(page); | |
2497 | /* | |
2498 | * Mark the page accessed if we read the beginning. | |
2499 | */ | |
2500 | if (!offset) | |
2501 | mark_page_accessed(page); | |
b5810039 | 2502 | } else { |
1da177e4 | 2503 | page = ZERO_PAGE(0); |
09cbfeaf | 2504 | get_page(page); |
b5810039 | 2505 | } |
1da177e4 LT |
2506 | |
2507 | /* | |
2508 | * Ok, we have the page, and it's up-to-date, so | |
2509 | * now we can copy it to user space... | |
1da177e4 | 2510 | */ |
2ba5bbed | 2511 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2512 | retval += ret; |
1da177e4 | 2513 | offset += ret; |
09cbfeaf KS |
2514 | index += offset >> PAGE_SHIFT; |
2515 | offset &= ~PAGE_MASK; | |
1da177e4 | 2516 | |
09cbfeaf | 2517 | put_page(page); |
2ba5bbed | 2518 | if (!iov_iter_count(to)) |
1da177e4 | 2519 | break; |
6e58e79d AV |
2520 | if (ret < nr) { |
2521 | error = -EFAULT; | |
2522 | break; | |
2523 | } | |
1da177e4 LT |
2524 | cond_resched(); |
2525 | } | |
2526 | ||
09cbfeaf | 2527 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2528 | file_accessed(file); |
2529 | return retval ? retval : error; | |
1da177e4 LT |
2530 | } |
2531 | ||
220f2ac9 HD |
2532 | /* |
2533 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
2534 | */ | |
2535 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2536 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2537 | { |
2538 | struct page *page; | |
2539 | struct pagevec pvec; | |
2540 | pgoff_t indices[PAGEVEC_SIZE]; | |
2541 | bool done = false; | |
2542 | int i; | |
2543 | ||
86679820 | 2544 | pagevec_init(&pvec); |
220f2ac9 HD |
2545 | pvec.nr = 1; /* start small: we may be there already */ |
2546 | while (!done) { | |
0cd6144a | 2547 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2548 | pvec.nr, pvec.pages, indices); |
2549 | if (!pvec.nr) { | |
965c8e59 | 2550 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2551 | index = end; |
2552 | break; | |
2553 | } | |
2554 | for (i = 0; i < pvec.nr; i++, index++) { | |
2555 | if (index < indices[i]) { | |
965c8e59 | 2556 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2557 | done = true; |
2558 | break; | |
2559 | } | |
2560 | index = indices[i]; | |
2561 | } | |
2562 | page = pvec.pages[i]; | |
3159f943 | 2563 | if (page && !xa_is_value(page)) { |
220f2ac9 HD |
2564 | if (!PageUptodate(page)) |
2565 | page = NULL; | |
2566 | } | |
2567 | if (index >= end || | |
965c8e59 AM |
2568 | (page && whence == SEEK_DATA) || |
2569 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2570 | done = true; |
2571 | break; | |
2572 | } | |
2573 | } | |
0cd6144a | 2574 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2575 | pagevec_release(&pvec); |
2576 | pvec.nr = PAGEVEC_SIZE; | |
2577 | cond_resched(); | |
2578 | } | |
2579 | return index; | |
2580 | } | |
2581 | ||
965c8e59 | 2582 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2583 | { |
2584 | struct address_space *mapping = file->f_mapping; | |
2585 | struct inode *inode = mapping->host; | |
2586 | pgoff_t start, end; | |
2587 | loff_t new_offset; | |
2588 | ||
965c8e59 AM |
2589 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2590 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2591 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2592 | inode_lock(inode); |
220f2ac9 HD |
2593 | /* We're holding i_mutex so we can access i_size directly */ |
2594 | ||
2595 | if (offset < 0) | |
2596 | offset = -EINVAL; | |
2597 | else if (offset >= inode->i_size) | |
2598 | offset = -ENXIO; | |
2599 | else { | |
09cbfeaf KS |
2600 | start = offset >> PAGE_SHIFT; |
2601 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2602 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2603 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2604 | if (new_offset > offset) { |
2605 | if (new_offset < inode->i_size) | |
2606 | offset = new_offset; | |
965c8e59 | 2607 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2608 | offset = -ENXIO; |
2609 | else | |
2610 | offset = inode->i_size; | |
2611 | } | |
2612 | } | |
2613 | ||
387aae6f HD |
2614 | if (offset >= 0) |
2615 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2616 | inode_unlock(inode); |
220f2ac9 HD |
2617 | return offset; |
2618 | } | |
2619 | ||
83e4fa9c HD |
2620 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2621 | loff_t len) | |
2622 | { | |
496ad9aa | 2623 | struct inode *inode = file_inode(file); |
e2d12e22 | 2624 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2625 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2626 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2627 | pgoff_t start, index, end; |
2628 | int error; | |
83e4fa9c | 2629 | |
13ace4d0 HD |
2630 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2631 | return -EOPNOTSUPP; | |
2632 | ||
5955102c | 2633 | inode_lock(inode); |
83e4fa9c HD |
2634 | |
2635 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2636 | struct address_space *mapping = file->f_mapping; | |
2637 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2638 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2639 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2640 | |
40e041a2 DH |
2641 | /* protected by i_mutex */ |
2642 | if (info->seals & F_SEAL_WRITE) { | |
2643 | error = -EPERM; | |
2644 | goto out; | |
2645 | } | |
2646 | ||
8e205f77 | 2647 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
2648 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
2649 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
2650 | spin_lock(&inode->i_lock); | |
2651 | inode->i_private = &shmem_falloc; | |
2652 | spin_unlock(&inode->i_lock); | |
2653 | ||
83e4fa9c HD |
2654 | if ((u64)unmap_end > (u64)unmap_start) |
2655 | unmap_mapping_range(mapping, unmap_start, | |
2656 | 1 + unmap_end - unmap_start, 0); | |
2657 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2658 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2659 | |
2660 | spin_lock(&inode->i_lock); | |
2661 | inode->i_private = NULL; | |
2662 | wake_up_all(&shmem_falloc_waitq); | |
2055da97 | 2663 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head)); |
8e205f77 | 2664 | spin_unlock(&inode->i_lock); |
83e4fa9c | 2665 | error = 0; |
8e205f77 | 2666 | goto out; |
e2d12e22 HD |
2667 | } |
2668 | ||
2669 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2670 | error = inode_newsize_ok(inode, offset + len); | |
2671 | if (error) | |
2672 | goto out; | |
2673 | ||
40e041a2 DH |
2674 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2675 | error = -EPERM; | |
2676 | goto out; | |
2677 | } | |
2678 | ||
09cbfeaf KS |
2679 | start = offset >> PAGE_SHIFT; |
2680 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2681 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2682 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2683 | error = -ENOSPC; | |
2684 | goto out; | |
83e4fa9c HD |
2685 | } |
2686 | ||
8e205f77 | 2687 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2688 | shmem_falloc.start = start; |
2689 | shmem_falloc.next = start; | |
2690 | shmem_falloc.nr_falloced = 0; | |
2691 | shmem_falloc.nr_unswapped = 0; | |
2692 | spin_lock(&inode->i_lock); | |
2693 | inode->i_private = &shmem_falloc; | |
2694 | spin_unlock(&inode->i_lock); | |
2695 | ||
e2d12e22 HD |
2696 | for (index = start; index < end; index++) { |
2697 | struct page *page; | |
2698 | ||
2699 | /* | |
2700 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2701 | * been interrupted because we are using up too much memory. | |
2702 | */ | |
2703 | if (signal_pending(current)) | |
2704 | error = -EINTR; | |
1aac1400 HD |
2705 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2706 | error = -ENOMEM; | |
e2d12e22 | 2707 | else |
9e18eb29 | 2708 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2709 | if (error) { |
1635f6a7 | 2710 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2711 | if (index > start) { |
2712 | shmem_undo_range(inode, | |
2713 | (loff_t)start << PAGE_SHIFT, | |
2714 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2715 | } | |
1aac1400 | 2716 | goto undone; |
e2d12e22 HD |
2717 | } |
2718 | ||
1aac1400 HD |
2719 | /* |
2720 | * Inform shmem_writepage() how far we have reached. | |
2721 | * No need for lock or barrier: we have the page lock. | |
2722 | */ | |
2723 | shmem_falloc.next++; | |
2724 | if (!PageUptodate(page)) | |
2725 | shmem_falloc.nr_falloced++; | |
2726 | ||
e2d12e22 | 2727 | /* |
1635f6a7 HD |
2728 | * If !PageUptodate, leave it that way so that freeable pages |
2729 | * can be recognized if we need to rollback on error later. | |
2730 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2731 | * than free the pages we are allocating (and SGP_CACHE pages |
2732 | * might still be clean: we now need to mark those dirty too). | |
2733 | */ | |
2734 | set_page_dirty(page); | |
2735 | unlock_page(page); | |
09cbfeaf | 2736 | put_page(page); |
e2d12e22 HD |
2737 | cond_resched(); |
2738 | } | |
2739 | ||
2740 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2741 | i_size_write(inode, offset + len); | |
078cd827 | 2742 | inode->i_ctime = current_time(inode); |
1aac1400 HD |
2743 | undone: |
2744 | spin_lock(&inode->i_lock); | |
2745 | inode->i_private = NULL; | |
2746 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2747 | out: |
5955102c | 2748 | inode_unlock(inode); |
83e4fa9c HD |
2749 | return error; |
2750 | } | |
2751 | ||
726c3342 | 2752 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2753 | { |
726c3342 | 2754 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2755 | |
2756 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2757 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2758 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2759 | if (sbinfo->max_blocks) { |
1da177e4 | 2760 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2761 | buf->f_bavail = |
2762 | buf->f_bfree = sbinfo->max_blocks - | |
2763 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2764 | } |
2765 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2766 | buf->f_files = sbinfo->max_inodes; |
2767 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2768 | } |
2769 | /* else leave those fields 0 like simple_statfs */ | |
2770 | return 0; | |
2771 | } | |
2772 | ||
2773 | /* | |
2774 | * File creation. Allocate an inode, and we're done.. | |
2775 | */ | |
2776 | static int | |
1a67aafb | 2777 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2778 | { |
0b0a0806 | 2779 | struct inode *inode; |
1da177e4 LT |
2780 | int error = -ENOSPC; |
2781 | ||
454abafe | 2782 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2783 | if (inode) { |
feda821e CH |
2784 | error = simple_acl_create(dir, inode); |
2785 | if (error) | |
2786 | goto out_iput; | |
2a7dba39 | 2787 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2788 | &dentry->d_name, |
6d9d88d0 | 2789 | shmem_initxattrs, NULL); |
feda821e CH |
2790 | if (error && error != -EOPNOTSUPP) |
2791 | goto out_iput; | |
37ec43cd | 2792 | |
718deb6b | 2793 | error = 0; |
1da177e4 | 2794 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 2795 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
2796 | d_instantiate(dentry, inode); |
2797 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2798 | } |
2799 | return error; | |
feda821e CH |
2800 | out_iput: |
2801 | iput(inode); | |
2802 | return error; | |
1da177e4 LT |
2803 | } |
2804 | ||
60545d0d AV |
2805 | static int |
2806 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2807 | { | |
2808 | struct inode *inode; | |
2809 | int error = -ENOSPC; | |
2810 | ||
2811 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
2812 | if (inode) { | |
2813 | error = security_inode_init_security(inode, dir, | |
2814 | NULL, | |
2815 | shmem_initxattrs, NULL); | |
feda821e CH |
2816 | if (error && error != -EOPNOTSUPP) |
2817 | goto out_iput; | |
2818 | error = simple_acl_create(dir, inode); | |
2819 | if (error) | |
2820 | goto out_iput; | |
60545d0d AV |
2821 | d_tmpfile(dentry, inode); |
2822 | } | |
2823 | return error; | |
feda821e CH |
2824 | out_iput: |
2825 | iput(inode); | |
2826 | return error; | |
60545d0d AV |
2827 | } |
2828 | ||
18bb1db3 | 2829 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2830 | { |
2831 | int error; | |
2832 | ||
2833 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
2834 | return error; | |
d8c76e6f | 2835 | inc_nlink(dir); |
1da177e4 LT |
2836 | return 0; |
2837 | } | |
2838 | ||
4acdaf27 | 2839 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2840 | bool excl) |
1da177e4 LT |
2841 | { |
2842 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
2843 | } | |
2844 | ||
2845 | /* | |
2846 | * Link a file.. | |
2847 | */ | |
2848 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2849 | { | |
75c3cfa8 | 2850 | struct inode *inode = d_inode(old_dentry); |
5b04c689 | 2851 | int ret; |
1da177e4 LT |
2852 | |
2853 | /* | |
2854 | * No ordinary (disk based) filesystem counts links as inodes; | |
2855 | * but each new link needs a new dentry, pinning lowmem, and | |
2856 | * tmpfs dentries cannot be pruned until they are unlinked. | |
2857 | */ | |
5b04c689 PE |
2858 | ret = shmem_reserve_inode(inode->i_sb); |
2859 | if (ret) | |
2860 | goto out; | |
1da177e4 LT |
2861 | |
2862 | dir->i_size += BOGO_DIRENT_SIZE; | |
078cd827 | 2863 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
d8c76e6f | 2864 | inc_nlink(inode); |
7de9c6ee | 2865 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
2866 | dget(dentry); /* Extra pinning count for the created dentry */ |
2867 | d_instantiate(dentry, inode); | |
5b04c689 PE |
2868 | out: |
2869 | return ret; | |
1da177e4 LT |
2870 | } |
2871 | ||
2872 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
2873 | { | |
75c3cfa8 | 2874 | struct inode *inode = d_inode(dentry); |
1da177e4 | 2875 | |
5b04c689 PE |
2876 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2877 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
2878 | |
2879 | dir->i_size -= BOGO_DIRENT_SIZE; | |
078cd827 | 2880 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
9a53c3a7 | 2881 | drop_nlink(inode); |
1da177e4 LT |
2882 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
2883 | return 0; | |
2884 | } | |
2885 | ||
2886 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
2887 | { | |
2888 | if (!simple_empty(dentry)) | |
2889 | return -ENOTEMPTY; | |
2890 | ||
75c3cfa8 | 2891 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 2892 | drop_nlink(dir); |
1da177e4 LT |
2893 | return shmem_unlink(dir, dentry); |
2894 | } | |
2895 | ||
37456771 MS |
2896 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
2897 | { | |
e36cb0b8 DH |
2898 | bool old_is_dir = d_is_dir(old_dentry); |
2899 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
2900 | |
2901 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
2902 | if (old_is_dir) { | |
2903 | drop_nlink(old_dir); | |
2904 | inc_nlink(new_dir); | |
2905 | } else { | |
2906 | drop_nlink(new_dir); | |
2907 | inc_nlink(old_dir); | |
2908 | } | |
2909 | } | |
2910 | old_dir->i_ctime = old_dir->i_mtime = | |
2911 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 | 2912 | d_inode(old_dentry)->i_ctime = |
078cd827 | 2913 | d_inode(new_dentry)->i_ctime = current_time(old_dir); |
37456771 MS |
2914 | |
2915 | return 0; | |
2916 | } | |
2917 | ||
46fdb794 MS |
2918 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
2919 | { | |
2920 | struct dentry *whiteout; | |
2921 | int error; | |
2922 | ||
2923 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
2924 | if (!whiteout) | |
2925 | return -ENOMEM; | |
2926 | ||
2927 | error = shmem_mknod(old_dir, whiteout, | |
2928 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
2929 | dput(whiteout); | |
2930 | if (error) | |
2931 | return error; | |
2932 | ||
2933 | /* | |
2934 | * Cheat and hash the whiteout while the old dentry is still in | |
2935 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
2936 | * | |
2937 | * d_lookup() will consistently find one of them at this point, | |
2938 | * not sure which one, but that isn't even important. | |
2939 | */ | |
2940 | d_rehash(whiteout); | |
2941 | return 0; | |
2942 | } | |
2943 | ||
1da177e4 LT |
2944 | /* |
2945 | * The VFS layer already does all the dentry stuff for rename, | |
2946 | * we just have to decrement the usage count for the target if | |
2947 | * it exists so that the VFS layer correctly free's it when it | |
2948 | * gets overwritten. | |
2949 | */ | |
3b69ff51 | 2950 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 2951 | { |
75c3cfa8 | 2952 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
2953 | int they_are_dirs = S_ISDIR(inode->i_mode); |
2954 | ||
46fdb794 | 2955 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
2956 | return -EINVAL; |
2957 | ||
37456771 MS |
2958 | if (flags & RENAME_EXCHANGE) |
2959 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
2960 | ||
1da177e4 LT |
2961 | if (!simple_empty(new_dentry)) |
2962 | return -ENOTEMPTY; | |
2963 | ||
46fdb794 MS |
2964 | if (flags & RENAME_WHITEOUT) { |
2965 | int error; | |
2966 | ||
2967 | error = shmem_whiteout(old_dir, old_dentry); | |
2968 | if (error) | |
2969 | return error; | |
2970 | } | |
2971 | ||
75c3cfa8 | 2972 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 2973 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 2974 | if (they_are_dirs) { |
75c3cfa8 | 2975 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 2976 | drop_nlink(old_dir); |
b928095b | 2977 | } |
1da177e4 | 2978 | } else if (they_are_dirs) { |
9a53c3a7 | 2979 | drop_nlink(old_dir); |
d8c76e6f | 2980 | inc_nlink(new_dir); |
1da177e4 LT |
2981 | } |
2982 | ||
2983 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
2984 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
2985 | old_dir->i_ctime = old_dir->i_mtime = | |
2986 | new_dir->i_ctime = new_dir->i_mtime = | |
078cd827 | 2987 | inode->i_ctime = current_time(old_dir); |
1da177e4 LT |
2988 | return 0; |
2989 | } | |
2990 | ||
2991 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
2992 | { | |
2993 | int error; | |
2994 | int len; | |
2995 | struct inode *inode; | |
9276aad6 | 2996 | struct page *page; |
1da177e4 LT |
2997 | |
2998 | len = strlen(symname) + 1; | |
09cbfeaf | 2999 | if (len > PAGE_SIZE) |
1da177e4 LT |
3000 | return -ENAMETOOLONG; |
3001 | ||
0825a6f9 JP |
3002 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0, |
3003 | VM_NORESERVE); | |
1da177e4 LT |
3004 | if (!inode) |
3005 | return -ENOSPC; | |
3006 | ||
9d8f13ba | 3007 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3008 | shmem_initxattrs, NULL); |
570bc1c2 SS |
3009 | if (error) { |
3010 | if (error != -EOPNOTSUPP) { | |
3011 | iput(inode); | |
3012 | return error; | |
3013 | } | |
3014 | error = 0; | |
3015 | } | |
3016 | ||
1da177e4 | 3017 | inode->i_size = len-1; |
69f07ec9 | 3018 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3019 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3020 | if (!inode->i_link) { | |
69f07ec9 HD |
3021 | iput(inode); |
3022 | return -ENOMEM; | |
3023 | } | |
3024 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3025 | } else { |
e8ecde25 | 3026 | inode_nohighmem(inode); |
9e18eb29 | 3027 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
3028 | if (error) { |
3029 | iput(inode); | |
3030 | return error; | |
3031 | } | |
14fcc23f | 3032 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 3033 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 3034 | memcpy(page_address(page), symname, len); |
ec9516fb | 3035 | SetPageUptodate(page); |
1da177e4 | 3036 | set_page_dirty(page); |
6746aff7 | 3037 | unlock_page(page); |
09cbfeaf | 3038 | put_page(page); |
1da177e4 | 3039 | } |
1da177e4 | 3040 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3041 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3042 | d_instantiate(dentry, inode); |
3043 | dget(dentry); | |
3044 | return 0; | |
3045 | } | |
3046 | ||
fceef393 | 3047 | static void shmem_put_link(void *arg) |
1da177e4 | 3048 | { |
fceef393 AV |
3049 | mark_page_accessed(arg); |
3050 | put_page(arg); | |
1da177e4 LT |
3051 | } |
3052 | ||
6b255391 | 3053 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
3054 | struct inode *inode, |
3055 | struct delayed_call *done) | |
1da177e4 | 3056 | { |
1da177e4 | 3057 | struct page *page = NULL; |
6b255391 | 3058 | int error; |
6a6c9904 AV |
3059 | if (!dentry) { |
3060 | page = find_get_page(inode->i_mapping, 0); | |
3061 | if (!page) | |
3062 | return ERR_PTR(-ECHILD); | |
3063 | if (!PageUptodate(page)) { | |
3064 | put_page(page); | |
3065 | return ERR_PTR(-ECHILD); | |
3066 | } | |
3067 | } else { | |
9e18eb29 | 3068 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3069 | if (error) |
3070 | return ERR_PTR(error); | |
3071 | unlock_page(page); | |
3072 | } | |
fceef393 | 3073 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3074 | return page_address(page); |
1da177e4 LT |
3075 | } |
3076 | ||
b09e0fa4 | 3077 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3078 | /* |
b09e0fa4 EP |
3079 | * Superblocks without xattr inode operations may get some security.* xattr |
3080 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3081 | * like ACLs, we also need to implement the security.* handlers at |
3082 | * filesystem level, though. | |
3083 | */ | |
3084 | ||
6d9d88d0 JS |
3085 | /* |
3086 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3087 | */ | |
3088 | static int shmem_initxattrs(struct inode *inode, | |
3089 | const struct xattr *xattr_array, | |
3090 | void *fs_info) | |
3091 | { | |
3092 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3093 | const struct xattr *xattr; | |
38f38657 | 3094 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3095 | size_t len; |
3096 | ||
3097 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3098 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3099 | if (!new_xattr) |
3100 | return -ENOMEM; | |
3101 | ||
3102 | len = strlen(xattr->name) + 1; | |
3103 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3104 | GFP_KERNEL); | |
3105 | if (!new_xattr->name) { | |
3106 | kfree(new_xattr); | |
3107 | return -ENOMEM; | |
3108 | } | |
3109 | ||
3110 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3111 | XATTR_SECURITY_PREFIX_LEN); | |
3112 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3113 | xattr->name, len); | |
3114 | ||
38f38657 | 3115 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3116 | } |
3117 | ||
3118 | return 0; | |
3119 | } | |
3120 | ||
aa7c5241 | 3121 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3122 | struct dentry *unused, struct inode *inode, |
3123 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3124 | { |
b296821a | 3125 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3126 | |
aa7c5241 | 3127 | name = xattr_full_name(handler, name); |
38f38657 | 3128 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3129 | } |
3130 | ||
aa7c5241 | 3131 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
59301226 AV |
3132 | struct dentry *unused, struct inode *inode, |
3133 | const char *name, const void *value, | |
3134 | size_t size, int flags) | |
b09e0fa4 | 3135 | { |
59301226 | 3136 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3137 | |
aa7c5241 | 3138 | name = xattr_full_name(handler, name); |
38f38657 | 3139 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
3140 | } |
3141 | ||
aa7c5241 AG |
3142 | static const struct xattr_handler shmem_security_xattr_handler = { |
3143 | .prefix = XATTR_SECURITY_PREFIX, | |
3144 | .get = shmem_xattr_handler_get, | |
3145 | .set = shmem_xattr_handler_set, | |
3146 | }; | |
b09e0fa4 | 3147 | |
aa7c5241 AG |
3148 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3149 | .prefix = XATTR_TRUSTED_PREFIX, | |
3150 | .get = shmem_xattr_handler_get, | |
3151 | .set = shmem_xattr_handler_set, | |
3152 | }; | |
b09e0fa4 | 3153 | |
aa7c5241 AG |
3154 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3155 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3156 | &posix_acl_access_xattr_handler, | |
3157 | &posix_acl_default_xattr_handler, | |
3158 | #endif | |
3159 | &shmem_security_xattr_handler, | |
3160 | &shmem_trusted_xattr_handler, | |
3161 | NULL | |
3162 | }; | |
b09e0fa4 EP |
3163 | |
3164 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3165 | { | |
75c3cfa8 | 3166 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3167 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3168 | } |
3169 | #endif /* CONFIG_TMPFS_XATTR */ | |
3170 | ||
69f07ec9 | 3171 | static const struct inode_operations shmem_short_symlink_operations = { |
6b255391 | 3172 | .get_link = simple_get_link, |
b09e0fa4 | 3173 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3174 | .listxattr = shmem_listxattr, |
b09e0fa4 EP |
3175 | #endif |
3176 | }; | |
3177 | ||
3178 | static const struct inode_operations shmem_symlink_inode_operations = { | |
6b255391 | 3179 | .get_link = shmem_get_link, |
b09e0fa4 | 3180 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3181 | .listxattr = shmem_listxattr, |
39f0247d | 3182 | #endif |
b09e0fa4 | 3183 | }; |
39f0247d | 3184 | |
91828a40 DG |
3185 | static struct dentry *shmem_get_parent(struct dentry *child) |
3186 | { | |
3187 | return ERR_PTR(-ESTALE); | |
3188 | } | |
3189 | ||
3190 | static int shmem_match(struct inode *ino, void *vfh) | |
3191 | { | |
3192 | __u32 *fh = vfh; | |
3193 | __u64 inum = fh[2]; | |
3194 | inum = (inum << 32) | fh[1]; | |
3195 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
3196 | } | |
3197 | ||
12ba780d AG |
3198 | /* Find any alias of inode, but prefer a hashed alias */ |
3199 | static struct dentry *shmem_find_alias(struct inode *inode) | |
3200 | { | |
3201 | struct dentry *alias = d_find_alias(inode); | |
3202 | ||
3203 | return alias ?: d_find_any_alias(inode); | |
3204 | } | |
3205 | ||
3206 | ||
480b116c CH |
3207 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3208 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3209 | { |
91828a40 | 3210 | struct inode *inode; |
480b116c | 3211 | struct dentry *dentry = NULL; |
35c2a7f4 | 3212 | u64 inum; |
480b116c CH |
3213 | |
3214 | if (fh_len < 3) | |
3215 | return NULL; | |
91828a40 | 3216 | |
35c2a7f4 HD |
3217 | inum = fid->raw[2]; |
3218 | inum = (inum << 32) | fid->raw[1]; | |
3219 | ||
480b116c CH |
3220 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
3221 | shmem_match, fid->raw); | |
91828a40 | 3222 | if (inode) { |
12ba780d | 3223 | dentry = shmem_find_alias(inode); |
91828a40 DG |
3224 | iput(inode); |
3225 | } | |
3226 | ||
480b116c | 3227 | return dentry; |
91828a40 DG |
3228 | } |
3229 | ||
b0b0382b AV |
3230 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3231 | struct inode *parent) | |
91828a40 | 3232 | { |
5fe0c237 AK |
3233 | if (*len < 3) { |
3234 | *len = 3; | |
94e07a75 | 3235 | return FILEID_INVALID; |
5fe0c237 | 3236 | } |
91828a40 | 3237 | |
1d3382cb | 3238 | if (inode_unhashed(inode)) { |
91828a40 DG |
3239 | /* Unfortunately insert_inode_hash is not idempotent, |
3240 | * so as we hash inodes here rather than at creation | |
3241 | * time, we need a lock to ensure we only try | |
3242 | * to do it once | |
3243 | */ | |
3244 | static DEFINE_SPINLOCK(lock); | |
3245 | spin_lock(&lock); | |
1d3382cb | 3246 | if (inode_unhashed(inode)) |
91828a40 DG |
3247 | __insert_inode_hash(inode, |
3248 | inode->i_ino + inode->i_generation); | |
3249 | spin_unlock(&lock); | |
3250 | } | |
3251 | ||
3252 | fh[0] = inode->i_generation; | |
3253 | fh[1] = inode->i_ino; | |
3254 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
3255 | ||
3256 | *len = 3; | |
3257 | return 1; | |
3258 | } | |
3259 | ||
39655164 | 3260 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3261 | .get_parent = shmem_get_parent, |
91828a40 | 3262 | .encode_fh = shmem_encode_fh, |
480b116c | 3263 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3264 | }; |
3265 | ||
680d794b | 3266 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
3267 | bool remount) | |
1da177e4 LT |
3268 | { |
3269 | char *this_char, *value, *rest; | |
49cd0a5c | 3270 | struct mempolicy *mpol = NULL; |
8751e039 EB |
3271 | uid_t uid; |
3272 | gid_t gid; | |
1da177e4 | 3273 | |
b00dc3ad HD |
3274 | while (options != NULL) { |
3275 | this_char = options; | |
3276 | for (;;) { | |
3277 | /* | |
3278 | * NUL-terminate this option: unfortunately, | |
3279 | * mount options form a comma-separated list, | |
3280 | * but mpol's nodelist may also contain commas. | |
3281 | */ | |
3282 | options = strchr(options, ','); | |
3283 | if (options == NULL) | |
3284 | break; | |
3285 | options++; | |
3286 | if (!isdigit(*options)) { | |
3287 | options[-1] = '\0'; | |
3288 | break; | |
3289 | } | |
3290 | } | |
1da177e4 LT |
3291 | if (!*this_char) |
3292 | continue; | |
3293 | if ((value = strchr(this_char,'=')) != NULL) { | |
3294 | *value++ = 0; | |
3295 | } else { | |
1170532b JP |
3296 | pr_err("tmpfs: No value for mount option '%s'\n", |
3297 | this_char); | |
49cd0a5c | 3298 | goto error; |
1da177e4 LT |
3299 | } |
3300 | ||
3301 | if (!strcmp(this_char,"size")) { | |
3302 | unsigned long long size; | |
3303 | size = memparse(value,&rest); | |
3304 | if (*rest == '%') { | |
3305 | size <<= PAGE_SHIFT; | |
3306 | size *= totalram_pages; | |
3307 | do_div(size, 100); | |
3308 | rest++; | |
3309 | } | |
3310 | if (*rest) | |
3311 | goto bad_val; | |
680d794b | 3312 | sbinfo->max_blocks = |
09cbfeaf | 3313 | DIV_ROUND_UP(size, PAGE_SIZE); |
1da177e4 | 3314 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 3315 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
3316 | if (*rest) |
3317 | goto bad_val; | |
3318 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 3319 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
3320 | if (*rest) |
3321 | goto bad_val; | |
3322 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 3323 | if (remount) |
1da177e4 | 3324 | continue; |
680d794b | 3325 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
3326 | if (*rest) |
3327 | goto bad_val; | |
3328 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 3329 | if (remount) |
1da177e4 | 3330 | continue; |
8751e039 | 3331 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3332 | if (*rest) |
3333 | goto bad_val; | |
8751e039 EB |
3334 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
3335 | if (!uid_valid(sbinfo->uid)) | |
3336 | goto bad_val; | |
1da177e4 | 3337 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 3338 | if (remount) |
1da177e4 | 3339 | continue; |
8751e039 | 3340 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3341 | if (*rest) |
3342 | goto bad_val; | |
8751e039 EB |
3343 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
3344 | if (!gid_valid(sbinfo->gid)) | |
3345 | goto bad_val; | |
e496cf3d | 3346 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3347 | } else if (!strcmp(this_char, "huge")) { |
3348 | int huge; | |
3349 | huge = shmem_parse_huge(value); | |
3350 | if (huge < 0) | |
3351 | goto bad_val; | |
3352 | if (!has_transparent_hugepage() && | |
3353 | huge != SHMEM_HUGE_NEVER) | |
3354 | goto bad_val; | |
3355 | sbinfo->huge = huge; | |
3356 | #endif | |
3357 | #ifdef CONFIG_NUMA | |
7339ff83 | 3358 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
3359 | mpol_put(mpol); |
3360 | mpol = NULL; | |
3361 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 3362 | goto bad_val; |
5a6e75f8 | 3363 | #endif |
1da177e4 | 3364 | } else { |
1170532b | 3365 | pr_err("tmpfs: Bad mount option %s\n", this_char); |
49cd0a5c | 3366 | goto error; |
1da177e4 LT |
3367 | } |
3368 | } | |
49cd0a5c | 3369 | sbinfo->mpol = mpol; |
1da177e4 LT |
3370 | return 0; |
3371 | ||
3372 | bad_val: | |
1170532b | 3373 | pr_err("tmpfs: Bad value '%s' for mount option '%s'\n", |
1da177e4 | 3374 | value, this_char); |
49cd0a5c GT |
3375 | error: |
3376 | mpol_put(mpol); | |
1da177e4 LT |
3377 | return 1; |
3378 | ||
3379 | } | |
3380 | ||
3381 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
3382 | { | |
3383 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 3384 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
3385 | unsigned long inodes; |
3386 | int error = -EINVAL; | |
3387 | ||
5f00110f | 3388 | config.mpol = NULL; |
680d794b | 3389 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 3390 | return error; |
1da177e4 | 3391 | |
0edd73b3 | 3392 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3393 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 3394 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 3395 | goto out; |
680d794b | 3396 | if (config.max_inodes < inodes) |
0edd73b3 HD |
3397 | goto out; |
3398 | /* | |
54af6042 | 3399 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
3400 | * but we must separately disallow unlimited->limited, because |
3401 | * in that case we have no record of how much is already in use. | |
3402 | */ | |
680d794b | 3403 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 3404 | goto out; |
680d794b | 3405 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
3406 | goto out; |
3407 | ||
3408 | error = 0; | |
5a6e75f8 | 3409 | sbinfo->huge = config.huge; |
680d794b | 3410 | sbinfo->max_blocks = config.max_blocks; |
680d794b | 3411 | sbinfo->max_inodes = config.max_inodes; |
3412 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 3413 | |
5f00110f GT |
3414 | /* |
3415 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3416 | */ | |
3417 | if (config.mpol) { | |
3418 | mpol_put(sbinfo->mpol); | |
3419 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
3420 | } | |
0edd73b3 HD |
3421 | out: |
3422 | spin_unlock(&sbinfo->stat_lock); | |
3423 | return error; | |
1da177e4 | 3424 | } |
680d794b | 3425 | |
34c80b1d | 3426 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3427 | { |
34c80b1d | 3428 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b | 3429 | |
3430 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3431 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3432 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 3433 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
3434 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
0825a6f9 | 3435 | if (sbinfo->mode != (0777 | S_ISVTX)) |
09208d15 | 3436 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3437 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3438 | seq_printf(seq, ",uid=%u", | |
3439 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3440 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3441 | seq_printf(seq, ",gid=%u", | |
3442 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
e496cf3d | 3443 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3444 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
3445 | if (sbinfo->huge) | |
3446 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3447 | #endif | |
71fe804b | 3448 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 3449 | return 0; |
3450 | } | |
9183df25 | 3451 | |
680d794b | 3452 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3453 | |
3454 | static void shmem_put_super(struct super_block *sb) | |
3455 | { | |
602586a8 HD |
3456 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3457 | ||
3458 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 3459 | mpol_put(sbinfo->mpol); |
602586a8 | 3460 | kfree(sbinfo); |
1da177e4 LT |
3461 | sb->s_fs_info = NULL; |
3462 | } | |
3463 | ||
2b2af54a | 3464 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3465 | { |
3466 | struct inode *inode; | |
0edd73b3 | 3467 | struct shmem_sb_info *sbinfo; |
680d794b | 3468 | int err = -ENOMEM; |
3469 | ||
3470 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3471 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 3472 | L1_CACHE_BYTES), GFP_KERNEL); |
3473 | if (!sbinfo) | |
3474 | return -ENOMEM; | |
3475 | ||
0825a6f9 | 3476 | sbinfo->mode = 0777 | S_ISVTX; |
76aac0e9 DH |
3477 | sbinfo->uid = current_fsuid(); |
3478 | sbinfo->gid = current_fsgid(); | |
680d794b | 3479 | sb->s_fs_info = sbinfo; |
1da177e4 | 3480 | |
0edd73b3 | 3481 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3482 | /* |
3483 | * Per default we only allow half of the physical ram per | |
3484 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3485 | * but the internal instance is left unlimited. | |
3486 | */ | |
1751e8a6 | 3487 | if (!(sb->s_flags & SB_KERNMOUNT)) { |
680d794b | 3488 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3489 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3490 | if (shmem_parse_options(data, sbinfo, false)) { | |
3491 | err = -EINVAL; | |
3492 | goto failed; | |
3493 | } | |
ca4e0519 | 3494 | } else { |
1751e8a6 | 3495 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 3496 | } |
91828a40 | 3497 | sb->s_export_op = &shmem_export_ops; |
1751e8a6 | 3498 | sb->s_flags |= SB_NOSEC; |
1da177e4 | 3499 | #else |
1751e8a6 | 3500 | sb->s_flags |= SB_NOUSER; |
1da177e4 LT |
3501 | #endif |
3502 | ||
0edd73b3 | 3503 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3504 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3505 | goto failed; |
680d794b | 3506 | sbinfo->free_inodes = sbinfo->max_inodes; |
779750d2 KS |
3507 | spin_lock_init(&sbinfo->shrinklist_lock); |
3508 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 3509 | |
285b2c4f | 3510 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3511 | sb->s_blocksize = PAGE_SIZE; |
3512 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3513 | sb->s_magic = TMPFS_MAGIC; |
3514 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3515 | sb->s_time_gran = 1; |
b09e0fa4 | 3516 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3517 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3518 | #endif |
3519 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
1751e8a6 | 3520 | sb->s_flags |= SB_POSIXACL; |
39f0247d | 3521 | #endif |
2b4db796 | 3522 | uuid_gen(&sb->s_uuid); |
0edd73b3 | 3523 | |
454abafe | 3524 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3525 | if (!inode) |
3526 | goto failed; | |
680d794b | 3527 | inode->i_uid = sbinfo->uid; |
3528 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3529 | sb->s_root = d_make_root(inode); |
3530 | if (!sb->s_root) | |
48fde701 | 3531 | goto failed; |
1da177e4 LT |
3532 | return 0; |
3533 | ||
1da177e4 LT |
3534 | failed: |
3535 | shmem_put_super(sb); | |
3536 | return err; | |
3537 | } | |
3538 | ||
fcc234f8 | 3539 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3540 | |
3541 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3542 | { | |
41ffe5d5 HD |
3543 | struct shmem_inode_info *info; |
3544 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3545 | if (!info) | |
1da177e4 | 3546 | return NULL; |
41ffe5d5 | 3547 | return &info->vfs_inode; |
1da177e4 LT |
3548 | } |
3549 | ||
41ffe5d5 | 3550 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3551 | { |
3552 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84e710da AV |
3553 | if (S_ISLNK(inode->i_mode)) |
3554 | kfree(inode->i_link); | |
fa0d7e3d NP |
3555 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3556 | } | |
3557 | ||
1da177e4 LT |
3558 | static void shmem_destroy_inode(struct inode *inode) |
3559 | { | |
09208d15 | 3560 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3561 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3562 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3563 | } |
3564 | ||
41ffe5d5 | 3565 | static void shmem_init_inode(void *foo) |
1da177e4 | 3566 | { |
41ffe5d5 HD |
3567 | struct shmem_inode_info *info = foo; |
3568 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3569 | } |
3570 | ||
9a8ec03e | 3571 | static void shmem_init_inodecache(void) |
1da177e4 LT |
3572 | { |
3573 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3574 | sizeof(struct shmem_inode_info), | |
5d097056 | 3575 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3576 | } |
3577 | ||
41ffe5d5 | 3578 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3579 | { |
1a1d92c1 | 3580 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3581 | } |
3582 | ||
f5e54d6e | 3583 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3584 | .writepage = shmem_writepage, |
76719325 | 3585 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3586 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3587 | .write_begin = shmem_write_begin, |
3588 | .write_end = shmem_write_end, | |
1da177e4 | 3589 | #endif |
1c93923c | 3590 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3591 | .migratepage = migrate_page, |
1c93923c | 3592 | #endif |
aa261f54 | 3593 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3594 | }; |
3595 | ||
15ad7cdc | 3596 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3597 | .mmap = shmem_mmap, |
c01d5b30 | 3598 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3599 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3600 | .llseek = shmem_file_llseek, |
2ba5bbed | 3601 | .read_iter = shmem_file_read_iter, |
8174202b | 3602 | .write_iter = generic_file_write_iter, |
1b061d92 | 3603 | .fsync = noop_fsync, |
82c156f8 | 3604 | .splice_read = generic_file_splice_read, |
f6cb85d0 | 3605 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3606 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3607 | #endif |
3608 | }; | |
3609 | ||
92e1d5be | 3610 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3611 | .getattr = shmem_getattr, |
94c1e62d | 3612 | .setattr = shmem_setattr, |
b09e0fa4 | 3613 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3614 | .listxattr = shmem_listxattr, |
feda821e | 3615 | .set_acl = simple_set_acl, |
b09e0fa4 | 3616 | #endif |
1da177e4 LT |
3617 | }; |
3618 | ||
92e1d5be | 3619 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3620 | #ifdef CONFIG_TMPFS |
3621 | .create = shmem_create, | |
3622 | .lookup = simple_lookup, | |
3623 | .link = shmem_link, | |
3624 | .unlink = shmem_unlink, | |
3625 | .symlink = shmem_symlink, | |
3626 | .mkdir = shmem_mkdir, | |
3627 | .rmdir = shmem_rmdir, | |
3628 | .mknod = shmem_mknod, | |
2773bf00 | 3629 | .rename = shmem_rename2, |
60545d0d | 3630 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3631 | #endif |
b09e0fa4 | 3632 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3633 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3634 | #endif |
39f0247d | 3635 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3636 | .setattr = shmem_setattr, |
feda821e | 3637 | .set_acl = simple_set_acl, |
39f0247d AG |
3638 | #endif |
3639 | }; | |
3640 | ||
92e1d5be | 3641 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3642 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3643 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3644 | #endif |
39f0247d | 3645 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3646 | .setattr = shmem_setattr, |
feda821e | 3647 | .set_acl = simple_set_acl, |
39f0247d | 3648 | #endif |
1da177e4 LT |
3649 | }; |
3650 | ||
759b9775 | 3651 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
3652 | .alloc_inode = shmem_alloc_inode, |
3653 | .destroy_inode = shmem_destroy_inode, | |
3654 | #ifdef CONFIG_TMPFS | |
3655 | .statfs = shmem_statfs, | |
3656 | .remount_fs = shmem_remount_fs, | |
680d794b | 3657 | .show_options = shmem_show_options, |
1da177e4 | 3658 | #endif |
1f895f75 | 3659 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3660 | .drop_inode = generic_delete_inode, |
3661 | .put_super = shmem_put_super, | |
779750d2 KS |
3662 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
3663 | .nr_cached_objects = shmem_unused_huge_count, | |
3664 | .free_cached_objects = shmem_unused_huge_scan, | |
3665 | #endif | |
1da177e4 LT |
3666 | }; |
3667 | ||
f0f37e2f | 3668 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3669 | .fault = shmem_fault, |
d7c17551 | 3670 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3671 | #ifdef CONFIG_NUMA |
3672 | .set_policy = shmem_set_policy, | |
3673 | .get_policy = shmem_get_policy, | |
3674 | #endif | |
3675 | }; | |
3676 | ||
3c26ff6e AV |
3677 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
3678 | int flags, const char *dev_name, void *data) | |
1da177e4 | 3679 | { |
3c26ff6e | 3680 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
3681 | } |
3682 | ||
41ffe5d5 | 3683 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3684 | .owner = THIS_MODULE, |
3685 | .name = "tmpfs", | |
3c26ff6e | 3686 | .mount = shmem_mount, |
1da177e4 | 3687 | .kill_sb = kill_litter_super, |
2b8576cb | 3688 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 3689 | }; |
1da177e4 | 3690 | |
41ffe5d5 | 3691 | int __init shmem_init(void) |
1da177e4 LT |
3692 | { |
3693 | int error; | |
3694 | ||
16203a7a RL |
3695 | /* If rootfs called this, don't re-init */ |
3696 | if (shmem_inode_cachep) | |
3697 | return 0; | |
3698 | ||
9a8ec03e | 3699 | shmem_init_inodecache(); |
1da177e4 | 3700 | |
41ffe5d5 | 3701 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 3702 | if (error) { |
1170532b | 3703 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
3704 | goto out2; |
3705 | } | |
95dc112a | 3706 | |
ca4e0519 | 3707 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
3708 | if (IS_ERR(shm_mnt)) { |
3709 | error = PTR_ERR(shm_mnt); | |
1170532b | 3710 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
3711 | goto out1; |
3712 | } | |
5a6e75f8 | 3713 | |
e496cf3d | 3714 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
435c0b87 | 3715 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
3716 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
3717 | else | |
3718 | shmem_huge = 0; /* just in case it was patched */ | |
3719 | #endif | |
1da177e4 LT |
3720 | return 0; |
3721 | ||
3722 | out1: | |
41ffe5d5 | 3723 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 3724 | out2: |
41ffe5d5 | 3725 | shmem_destroy_inodecache(); |
1da177e4 LT |
3726 | shm_mnt = ERR_PTR(error); |
3727 | return error; | |
3728 | } | |
853ac43a | 3729 | |
e496cf3d | 3730 | #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS) |
5a6e75f8 KS |
3731 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
3732 | struct kobj_attribute *attr, char *buf) | |
3733 | { | |
3734 | int values[] = { | |
3735 | SHMEM_HUGE_ALWAYS, | |
3736 | SHMEM_HUGE_WITHIN_SIZE, | |
3737 | SHMEM_HUGE_ADVISE, | |
3738 | SHMEM_HUGE_NEVER, | |
3739 | SHMEM_HUGE_DENY, | |
3740 | SHMEM_HUGE_FORCE, | |
3741 | }; | |
3742 | int i, count; | |
3743 | ||
3744 | for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) { | |
3745 | const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s "; | |
3746 | ||
3747 | count += sprintf(buf + count, fmt, | |
3748 | shmem_format_huge(values[i])); | |
3749 | } | |
3750 | buf[count - 1] = '\n'; | |
3751 | return count; | |
3752 | } | |
3753 | ||
3754 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
3755 | struct kobj_attribute *attr, const char *buf, size_t count) | |
3756 | { | |
3757 | char tmp[16]; | |
3758 | int huge; | |
3759 | ||
3760 | if (count + 1 > sizeof(tmp)) | |
3761 | return -EINVAL; | |
3762 | memcpy(tmp, buf, count); | |
3763 | tmp[count] = '\0'; | |
3764 | if (count && tmp[count - 1] == '\n') | |
3765 | tmp[count - 1] = '\0'; | |
3766 | ||
3767 | huge = shmem_parse_huge(tmp); | |
3768 | if (huge == -EINVAL) | |
3769 | return -EINVAL; | |
3770 | if (!has_transparent_hugepage() && | |
3771 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
3772 | return -EINVAL; | |
3773 | ||
3774 | shmem_huge = huge; | |
435c0b87 | 3775 | if (shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
3776 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
3777 | return count; | |
3778 | } | |
3779 | ||
3780 | struct kobj_attribute shmem_enabled_attr = | |
3781 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
3b33719c | 3782 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */ |
f3f0e1d2 | 3783 | |
3b33719c | 3784 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
f3f0e1d2 KS |
3785 | bool shmem_huge_enabled(struct vm_area_struct *vma) |
3786 | { | |
3787 | struct inode *inode = file_inode(vma->vm_file); | |
3788 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
3789 | loff_t i_size; | |
3790 | pgoff_t off; | |
3791 | ||
3792 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
3793 | return true; | |
3794 | if (shmem_huge == SHMEM_HUGE_DENY) | |
3795 | return false; | |
3796 | switch (sbinfo->huge) { | |
3797 | case SHMEM_HUGE_NEVER: | |
3798 | return false; | |
3799 | case SHMEM_HUGE_ALWAYS: | |
3800 | return true; | |
3801 | case SHMEM_HUGE_WITHIN_SIZE: | |
3802 | off = round_up(vma->vm_pgoff, HPAGE_PMD_NR); | |
3803 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
3804 | if (i_size >= HPAGE_PMD_SIZE && | |
3805 | i_size >> PAGE_SHIFT >= off) | |
3806 | return true; | |
c8402871 | 3807 | /* fall through */ |
f3f0e1d2 KS |
3808 | case SHMEM_HUGE_ADVISE: |
3809 | /* TODO: implement fadvise() hints */ | |
3810 | return (vma->vm_flags & VM_HUGEPAGE); | |
3811 | default: | |
3812 | VM_BUG_ON(1); | |
3813 | return false; | |
3814 | } | |
3815 | } | |
3b33719c | 3816 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 3817 | |
853ac43a MM |
3818 | #else /* !CONFIG_SHMEM */ |
3819 | ||
3820 | /* | |
3821 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
3822 | * | |
3823 | * This is intended for small system where the benefits of the full | |
3824 | * shmem code (swap-backed and resource-limited) are outweighed by | |
3825 | * their complexity. On systems without swap this code should be | |
3826 | * effectively equivalent, but much lighter weight. | |
3827 | */ | |
3828 | ||
41ffe5d5 | 3829 | static struct file_system_type shmem_fs_type = { |
853ac43a | 3830 | .name = "tmpfs", |
3c26ff6e | 3831 | .mount = ramfs_mount, |
853ac43a | 3832 | .kill_sb = kill_litter_super, |
2b8576cb | 3833 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
3834 | }; |
3835 | ||
41ffe5d5 | 3836 | int __init shmem_init(void) |
853ac43a | 3837 | { |
41ffe5d5 | 3838 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 3839 | |
41ffe5d5 | 3840 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
3841 | BUG_ON(IS_ERR(shm_mnt)); |
3842 | ||
3843 | return 0; | |
3844 | } | |
3845 | ||
41ffe5d5 | 3846 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
3847 | { |
3848 | return 0; | |
3849 | } | |
3850 | ||
3f96b79a HD |
3851 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
3852 | { | |
3853 | return 0; | |
3854 | } | |
3855 | ||
24513264 HD |
3856 | void shmem_unlock_mapping(struct address_space *mapping) |
3857 | { | |
3858 | } | |
3859 | ||
c01d5b30 HD |
3860 | #ifdef CONFIG_MMU |
3861 | unsigned long shmem_get_unmapped_area(struct file *file, | |
3862 | unsigned long addr, unsigned long len, | |
3863 | unsigned long pgoff, unsigned long flags) | |
3864 | { | |
3865 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
3866 | } | |
3867 | #endif | |
3868 | ||
41ffe5d5 | 3869 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 3870 | { |
41ffe5d5 | 3871 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
3872 | } |
3873 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
3874 | ||
0b0a0806 HD |
3875 | #define shmem_vm_ops generic_file_vm_ops |
3876 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 3877 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
3878 | #define shmem_acct_size(flags, size) 0 |
3879 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
3880 | |
3881 | #endif /* CONFIG_SHMEM */ | |
3882 | ||
3883 | /* common code */ | |
1da177e4 | 3884 | |
703321b6 | 3885 | static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size, |
c7277090 | 3886 | unsigned long flags, unsigned int i_flags) |
1da177e4 | 3887 | { |
1da177e4 | 3888 | struct inode *inode; |
93dec2da | 3889 | struct file *res; |
1da177e4 | 3890 | |
703321b6 MA |
3891 | if (IS_ERR(mnt)) |
3892 | return ERR_CAST(mnt); | |
1da177e4 | 3893 | |
285b2c4f | 3894 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
3895 | return ERR_PTR(-EINVAL); |
3896 | ||
3897 | if (shmem_acct_size(flags, size)) | |
3898 | return ERR_PTR(-ENOMEM); | |
3899 | ||
93dec2da AV |
3900 | inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0, |
3901 | flags); | |
dac2d1f6 AV |
3902 | if (unlikely(!inode)) { |
3903 | shmem_unacct_size(flags, size); | |
3904 | return ERR_PTR(-ENOSPC); | |
3905 | } | |
c7277090 | 3906 | inode->i_flags |= i_flags; |
1da177e4 | 3907 | inode->i_size = size; |
6d6b77f1 | 3908 | clear_nlink(inode); /* It is unlinked */ |
26567cdb | 3909 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
93dec2da AV |
3910 | if (!IS_ERR(res)) |
3911 | res = alloc_file_pseudo(inode, mnt, name, O_RDWR, | |
3912 | &shmem_file_operations); | |
26567cdb | 3913 | if (IS_ERR(res)) |
93dec2da | 3914 | iput(inode); |
6b4d0b27 | 3915 | return res; |
1da177e4 | 3916 | } |
c7277090 EP |
3917 | |
3918 | /** | |
3919 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
3920 | * kernel internal. There will be NO LSM permission checks against the | |
3921 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
3922 | * higher layer. The users are the big_key and shm implementations. LSM |
3923 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
3924 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
3925 | * @size: size to be set for the file | |
3926 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3927 | */ | |
3928 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
3929 | { | |
703321b6 | 3930 | return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE); |
c7277090 EP |
3931 | } |
3932 | ||
3933 | /** | |
3934 | * shmem_file_setup - get an unlinked file living in tmpfs | |
3935 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
3936 | * @size: size to be set for the file | |
3937 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3938 | */ | |
3939 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
3940 | { | |
703321b6 | 3941 | return __shmem_file_setup(shm_mnt, name, size, flags, 0); |
c7277090 | 3942 | } |
395e0ddc | 3943 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 3944 | |
703321b6 MA |
3945 | /** |
3946 | * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs | |
3947 | * @mnt: the tmpfs mount where the file will be created | |
3948 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
3949 | * @size: size to be set for the file | |
3950 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3951 | */ | |
3952 | struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name, | |
3953 | loff_t size, unsigned long flags) | |
3954 | { | |
3955 | return __shmem_file_setup(mnt, name, size, flags, 0); | |
3956 | } | |
3957 | EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt); | |
3958 | ||
46711810 | 3959 | /** |
1da177e4 | 3960 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
3961 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
3962 | */ | |
3963 | int shmem_zero_setup(struct vm_area_struct *vma) | |
3964 | { | |
3965 | struct file *file; | |
3966 | loff_t size = vma->vm_end - vma->vm_start; | |
3967 | ||
66fc1303 HD |
3968 | /* |
3969 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
3970 | * between XFS directory reading and selinux: since this file is only | |
3971 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
3972 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
3973 | */ | |
703321b6 | 3974 | file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags); |
1da177e4 LT |
3975 | if (IS_ERR(file)) |
3976 | return PTR_ERR(file); | |
3977 | ||
3978 | if (vma->vm_file) | |
3979 | fput(vma->vm_file); | |
3980 | vma->vm_file = file; | |
3981 | vma->vm_ops = &shmem_vm_ops; | |
f3f0e1d2 | 3982 | |
e496cf3d | 3983 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
3984 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
3985 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
3986 | khugepaged_enter(vma, vma->vm_flags); | |
3987 | } | |
3988 | ||
1da177e4 LT |
3989 | return 0; |
3990 | } | |
d9d90e5e HD |
3991 | |
3992 | /** | |
3993 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
3994 | * @mapping: the page's address_space | |
3995 | * @index: the page index | |
3996 | * @gfp: the page allocator flags to use if allocating | |
3997 | * | |
3998 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
3999 | * with any new page allocations done using the specified allocation flags. | |
4000 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
4001 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
4002 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4003 | * | |
68da9f05 HD |
4004 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4005 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4006 | */ |
4007 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4008 | pgoff_t index, gfp_t gfp) | |
4009 | { | |
68da9f05 HD |
4010 | #ifdef CONFIG_SHMEM |
4011 | struct inode *inode = mapping->host; | |
9276aad6 | 4012 | struct page *page; |
68da9f05 HD |
4013 | int error; |
4014 | ||
4015 | BUG_ON(mapping->a_ops != &shmem_aops); | |
9e18eb29 | 4016 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
cfda0526 | 4017 | gfp, NULL, NULL, NULL); |
68da9f05 HD |
4018 | if (error) |
4019 | page = ERR_PTR(error); | |
4020 | else | |
4021 | unlock_page(page); | |
4022 | return page; | |
4023 | #else | |
4024 | /* | |
4025 | * The tiny !SHMEM case uses ramfs without swap | |
4026 | */ | |
d9d90e5e | 4027 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4028 | #endif |
d9d90e5e HD |
4029 | } |
4030 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |