Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Resizable virtual memory filesystem for Linux. | |
3 | * | |
4 | * Copyright (C) 2000 Linus Torvalds. | |
5 | * 2000 Transmeta Corp. | |
6 | * 2000-2001 Christoph Rohland | |
7 | * 2000-2001 SAP AG | |
8 | * 2002 Red Hat Inc. | |
6922c0c7 HD |
9 | * Copyright (C) 2002-2011 Hugh Dickins. |
10 | * Copyright (C) 2011 Google Inc. | |
0edd73b3 | 11 | * Copyright (C) 2002-2005 VERITAS Software Corporation. |
1da177e4 LT |
12 | * Copyright (C) 2004 Andi Kleen, SuSE Labs |
13 | * | |
14 | * Extended attribute support for tmpfs: | |
15 | * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net> | |
16 | * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com> | |
17 | * | |
853ac43a MM |
18 | * tiny-shmem: |
19 | * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com> | |
20 | * | |
1da177e4 LT |
21 | * This file is released under the GPL. |
22 | */ | |
23 | ||
853ac43a MM |
24 | #include <linux/fs.h> |
25 | #include <linux/init.h> | |
26 | #include <linux/vfs.h> | |
27 | #include <linux/mount.h> | |
250297ed | 28 | #include <linux/ramfs.h> |
caefba17 | 29 | #include <linux/pagemap.h> |
853ac43a MM |
30 | #include <linux/file.h> |
31 | #include <linux/mm.h> | |
b95f1b31 | 32 | #include <linux/export.h> |
853ac43a | 33 | #include <linux/swap.h> |
e2e40f2c | 34 | #include <linux/uio.h> |
f3f0e1d2 | 35 | #include <linux/khugepaged.h> |
853ac43a MM |
36 | |
37 | static struct vfsmount *shm_mnt; | |
38 | ||
39 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
40 | /* |
41 | * This virtual memory filesystem is heavily based on the ramfs. It | |
42 | * extends ramfs by the ability to use swap and honor resource limits | |
43 | * which makes it a completely usable filesystem. | |
44 | */ | |
45 | ||
39f0247d | 46 | #include <linux/xattr.h> |
a5694255 | 47 | #include <linux/exportfs.h> |
1c7c474c | 48 | #include <linux/posix_acl.h> |
feda821e | 49 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 50 | #include <linux/mman.h> |
1da177e4 LT |
51 | #include <linux/string.h> |
52 | #include <linux/slab.h> | |
53 | #include <linux/backing-dev.h> | |
54 | #include <linux/shmem_fs.h> | |
1da177e4 | 55 | #include <linux/writeback.h> |
1da177e4 | 56 | #include <linux/blkdev.h> |
bda97eab | 57 | #include <linux/pagevec.h> |
41ffe5d5 | 58 | #include <linux/percpu_counter.h> |
83e4fa9c | 59 | #include <linux/falloc.h> |
708e3508 | 60 | #include <linux/splice.h> |
1da177e4 LT |
61 | #include <linux/security.h> |
62 | #include <linux/swapops.h> | |
63 | #include <linux/mempolicy.h> | |
64 | #include <linux/namei.h> | |
b00dc3ad | 65 | #include <linux/ctype.h> |
304dbdb7 | 66 | #include <linux/migrate.h> |
c1f60a5a | 67 | #include <linux/highmem.h> |
680d794b | 68 | #include <linux/seq_file.h> |
92562927 | 69 | #include <linux/magic.h> |
9183df25 | 70 | #include <linux/syscalls.h> |
40e041a2 | 71 | #include <linux/fcntl.h> |
9183df25 | 72 | #include <uapi/linux/memfd.h> |
304dbdb7 | 73 | |
1da177e4 | 74 | #include <asm/uaccess.h> |
1da177e4 LT |
75 | #include <asm/pgtable.h> |
76 | ||
dd56b046 MG |
77 | #include "internal.h" |
78 | ||
09cbfeaf KS |
79 | #define BLOCKS_PER_PAGE (PAGE_SIZE/512) |
80 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT) | |
1da177e4 | 81 | |
1da177e4 LT |
82 | /* Pretend that each entry is of this size in directory's i_size */ |
83 | #define BOGO_DIRENT_SIZE 20 | |
84 | ||
69f07ec9 HD |
85 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
86 | #define SHORT_SYMLINK_LEN 128 | |
87 | ||
1aac1400 | 88 | /* |
f00cdc6d HD |
89 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
90 | * inode->i_private (with i_mutex making sure that it has only one user at | |
91 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
92 | */ |
93 | struct shmem_falloc { | |
8e205f77 | 94 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
95 | pgoff_t start; /* start of range currently being fallocated */ |
96 | pgoff_t next; /* the next page offset to be fallocated */ | |
97 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
98 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
99 | }; | |
100 | ||
b76db735 | 101 | #ifdef CONFIG_TMPFS |
680d794b | 102 | static unsigned long shmem_default_max_blocks(void) |
103 | { | |
104 | return totalram_pages / 2; | |
105 | } | |
106 | ||
107 | static unsigned long shmem_default_max_inodes(void) | |
108 | { | |
109 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
110 | } | |
b76db735 | 111 | #endif |
680d794b | 112 | |
bde05d1c HD |
113 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
114 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
115 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 | 116 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 ALC |
117 | struct page **pagep, enum sgp_type sgp, |
118 | gfp_t gfp, struct mm_struct *fault_mm, int *fault_type); | |
68da9f05 | 119 | |
f3f0e1d2 | 120 | int shmem_getpage(struct inode *inode, pgoff_t index, |
9e18eb29 | 121 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
122 | { |
123 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
9e18eb29 | 124 | mapping_gfp_mask(inode->i_mapping), NULL, NULL); |
68da9f05 | 125 | } |
1da177e4 | 126 | |
1da177e4 LT |
127 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
128 | { | |
129 | return sb->s_fs_info; | |
130 | } | |
131 | ||
132 | /* | |
133 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
134 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
135 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
136 | * consistent with the pre-accounting of private mappings ... | |
137 | */ | |
138 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
139 | { | |
0b0a0806 | 140 | return (flags & VM_NORESERVE) ? |
191c5424 | 141 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
142 | } |
143 | ||
144 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
145 | { | |
0b0a0806 | 146 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
147 | vm_unacct_memory(VM_ACCT(size)); |
148 | } | |
149 | ||
77142517 KK |
150 | static inline int shmem_reacct_size(unsigned long flags, |
151 | loff_t oldsize, loff_t newsize) | |
152 | { | |
153 | if (!(flags & VM_NORESERVE)) { | |
154 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
155 | return security_vm_enough_memory_mm(current->mm, | |
156 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
157 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
158 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
159 | } | |
160 | return 0; | |
161 | } | |
162 | ||
1da177e4 LT |
163 | /* |
164 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 165 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
166 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
167 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
168 | */ | |
800d8c63 | 169 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 170 | { |
800d8c63 KS |
171 | if (!(flags & VM_NORESERVE)) |
172 | return 0; | |
173 | ||
174 | return security_vm_enough_memory_mm(current->mm, | |
175 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
176 | } |
177 | ||
178 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
179 | { | |
0b0a0806 | 180 | if (flags & VM_NORESERVE) |
09cbfeaf | 181 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
182 | } |
183 | ||
759b9775 | 184 | static const struct super_operations shmem_ops; |
f5e54d6e | 185 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 186 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
187 | static const struct inode_operations shmem_inode_operations; |
188 | static const struct inode_operations shmem_dir_inode_operations; | |
189 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 190 | static const struct vm_operations_struct shmem_vm_ops; |
779750d2 | 191 | static struct file_system_type shmem_fs_type; |
1da177e4 | 192 | |
1da177e4 | 193 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 194 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 195 | |
5b04c689 PE |
196 | static int shmem_reserve_inode(struct super_block *sb) |
197 | { | |
198 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
199 | if (sbinfo->max_inodes) { | |
200 | spin_lock(&sbinfo->stat_lock); | |
201 | if (!sbinfo->free_inodes) { | |
202 | spin_unlock(&sbinfo->stat_lock); | |
203 | return -ENOSPC; | |
204 | } | |
205 | sbinfo->free_inodes--; | |
206 | spin_unlock(&sbinfo->stat_lock); | |
207 | } | |
208 | return 0; | |
209 | } | |
210 | ||
211 | static void shmem_free_inode(struct super_block *sb) | |
212 | { | |
213 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
214 | if (sbinfo->max_inodes) { | |
215 | spin_lock(&sbinfo->stat_lock); | |
216 | sbinfo->free_inodes++; | |
217 | spin_unlock(&sbinfo->stat_lock); | |
218 | } | |
219 | } | |
220 | ||
46711810 | 221 | /** |
41ffe5d5 | 222 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
223 | * @inode: inode to recalc |
224 | * | |
225 | * We have to calculate the free blocks since the mm can drop | |
226 | * undirtied hole pages behind our back. | |
227 | * | |
228 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
229 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
230 | * | |
231 | * It has to be called with the spinlock held. | |
232 | */ | |
233 | static void shmem_recalc_inode(struct inode *inode) | |
234 | { | |
235 | struct shmem_inode_info *info = SHMEM_I(inode); | |
236 | long freed; | |
237 | ||
238 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
239 | if (freed > 0) { | |
54af6042 HD |
240 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
241 | if (sbinfo->max_blocks) | |
242 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 243 | info->alloced -= freed; |
54af6042 | 244 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 245 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
246 | } |
247 | } | |
248 | ||
800d8c63 KS |
249 | bool shmem_charge(struct inode *inode, long pages) |
250 | { | |
251 | struct shmem_inode_info *info = SHMEM_I(inode); | |
252 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4595ef88 | 253 | unsigned long flags; |
800d8c63 KS |
254 | |
255 | if (shmem_acct_block(info->flags, pages)) | |
256 | return false; | |
4595ef88 | 257 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
258 | info->alloced += pages; |
259 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
260 | shmem_recalc_inode(inode); | |
4595ef88 | 261 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
262 | inode->i_mapping->nrpages += pages; |
263 | ||
264 | if (!sbinfo->max_blocks) | |
265 | return true; | |
266 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
267 | sbinfo->max_blocks - pages) > 0) { | |
268 | inode->i_mapping->nrpages -= pages; | |
4595ef88 | 269 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
270 | info->alloced -= pages; |
271 | shmem_recalc_inode(inode); | |
4595ef88 | 272 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
273 | |
274 | return false; | |
275 | } | |
276 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
277 | return true; | |
278 | } | |
279 | ||
280 | void shmem_uncharge(struct inode *inode, long pages) | |
281 | { | |
282 | struct shmem_inode_info *info = SHMEM_I(inode); | |
283 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4595ef88 | 284 | unsigned long flags; |
800d8c63 | 285 | |
4595ef88 | 286 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
287 | info->alloced -= pages; |
288 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
289 | shmem_recalc_inode(inode); | |
4595ef88 | 290 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 KS |
291 | |
292 | if (sbinfo->max_blocks) | |
293 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
294 | } | |
295 | ||
7a5d0fbb HD |
296 | /* |
297 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
298 | */ | |
299 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
300 | pgoff_t index, void *expected, void *replacement) | |
301 | { | |
302 | void **pslot; | |
6dbaf22c | 303 | void *item; |
7a5d0fbb HD |
304 | |
305 | VM_BUG_ON(!expected); | |
6dbaf22c | 306 | VM_BUG_ON(!replacement); |
7a5d0fbb | 307 | pslot = radix_tree_lookup_slot(&mapping->page_tree, index); |
6dbaf22c JW |
308 | if (!pslot) |
309 | return -ENOENT; | |
310 | item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock); | |
7a5d0fbb HD |
311 | if (item != expected) |
312 | return -ENOENT; | |
6dbaf22c | 313 | radix_tree_replace_slot(pslot, replacement); |
7a5d0fbb HD |
314 | return 0; |
315 | } | |
316 | ||
d1899228 HD |
317 | /* |
318 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
319 | * that an entry was not already brought back from swap by a racing thread. | |
320 | * | |
321 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
322 | * might be reused, and again be SwapCache, using the same swap as before. | |
323 | */ | |
324 | static bool shmem_confirm_swap(struct address_space *mapping, | |
325 | pgoff_t index, swp_entry_t swap) | |
326 | { | |
327 | void *item; | |
328 | ||
329 | rcu_read_lock(); | |
330 | item = radix_tree_lookup(&mapping->page_tree, index); | |
331 | rcu_read_unlock(); | |
332 | return item == swp_to_radix_entry(swap); | |
333 | } | |
334 | ||
5a6e75f8 KS |
335 | /* |
336 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
337 | * | |
338 | * SHMEM_HUGE_NEVER: | |
339 | * disables huge pages for the mount; | |
340 | * SHMEM_HUGE_ALWAYS: | |
341 | * enables huge pages for the mount; | |
342 | * SHMEM_HUGE_WITHIN_SIZE: | |
343 | * only allocate huge pages if the page will be fully within i_size, | |
344 | * also respect fadvise()/madvise() hints; | |
345 | * SHMEM_HUGE_ADVISE: | |
346 | * only allocate huge pages if requested with fadvise()/madvise(); | |
347 | */ | |
348 | ||
349 | #define SHMEM_HUGE_NEVER 0 | |
350 | #define SHMEM_HUGE_ALWAYS 1 | |
351 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
352 | #define SHMEM_HUGE_ADVISE 3 | |
353 | ||
354 | /* | |
355 | * Special values. | |
356 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
357 | * | |
358 | * SHMEM_HUGE_DENY: | |
359 | * disables huge on shm_mnt and all mounts, for emergency use; | |
360 | * SHMEM_HUGE_FORCE: | |
361 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
362 | * | |
363 | */ | |
364 | #define SHMEM_HUGE_DENY (-1) | |
365 | #define SHMEM_HUGE_FORCE (-2) | |
366 | ||
e496cf3d | 367 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
368 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
369 | ||
370 | int shmem_huge __read_mostly; | |
371 | ||
372 | static int shmem_parse_huge(const char *str) | |
373 | { | |
374 | if (!strcmp(str, "never")) | |
375 | return SHMEM_HUGE_NEVER; | |
376 | if (!strcmp(str, "always")) | |
377 | return SHMEM_HUGE_ALWAYS; | |
378 | if (!strcmp(str, "within_size")) | |
379 | return SHMEM_HUGE_WITHIN_SIZE; | |
380 | if (!strcmp(str, "advise")) | |
381 | return SHMEM_HUGE_ADVISE; | |
382 | if (!strcmp(str, "deny")) | |
383 | return SHMEM_HUGE_DENY; | |
384 | if (!strcmp(str, "force")) | |
385 | return SHMEM_HUGE_FORCE; | |
386 | return -EINVAL; | |
387 | } | |
388 | ||
389 | static const char *shmem_format_huge(int huge) | |
390 | { | |
391 | switch (huge) { | |
392 | case SHMEM_HUGE_NEVER: | |
393 | return "never"; | |
394 | case SHMEM_HUGE_ALWAYS: | |
395 | return "always"; | |
396 | case SHMEM_HUGE_WITHIN_SIZE: | |
397 | return "within_size"; | |
398 | case SHMEM_HUGE_ADVISE: | |
399 | return "advise"; | |
400 | case SHMEM_HUGE_DENY: | |
401 | return "deny"; | |
402 | case SHMEM_HUGE_FORCE: | |
403 | return "force"; | |
404 | default: | |
405 | VM_BUG_ON(1); | |
406 | return "bad_val"; | |
407 | } | |
408 | } | |
409 | ||
779750d2 KS |
410 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
411 | struct shrink_control *sc, unsigned long nr_to_split) | |
412 | { | |
413 | LIST_HEAD(list), *pos, *next; | |
414 | struct inode *inode; | |
415 | struct shmem_inode_info *info; | |
416 | struct page *page; | |
417 | unsigned long batch = sc ? sc->nr_to_scan : 128; | |
418 | int removed = 0, split = 0; | |
419 | ||
420 | if (list_empty(&sbinfo->shrinklist)) | |
421 | return SHRINK_STOP; | |
422 | ||
423 | spin_lock(&sbinfo->shrinklist_lock); | |
424 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
425 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
426 | ||
427 | /* pin the inode */ | |
428 | inode = igrab(&info->vfs_inode); | |
429 | ||
430 | /* inode is about to be evicted */ | |
431 | if (!inode) { | |
432 | list_del_init(&info->shrinklist); | |
433 | removed++; | |
434 | goto next; | |
435 | } | |
436 | ||
437 | /* Check if there's anything to gain */ | |
438 | if (round_up(inode->i_size, PAGE_SIZE) == | |
439 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
440 | list_del_init(&info->shrinklist); | |
441 | removed++; | |
442 | iput(inode); | |
443 | goto next; | |
444 | } | |
445 | ||
446 | list_move(&info->shrinklist, &list); | |
447 | next: | |
448 | if (!--batch) | |
449 | break; | |
450 | } | |
451 | spin_unlock(&sbinfo->shrinklist_lock); | |
452 | ||
453 | list_for_each_safe(pos, next, &list) { | |
454 | int ret; | |
455 | ||
456 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
457 | inode = &info->vfs_inode; | |
458 | ||
459 | if (nr_to_split && split >= nr_to_split) { | |
460 | iput(inode); | |
461 | continue; | |
462 | } | |
463 | ||
464 | page = find_lock_page(inode->i_mapping, | |
465 | (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT); | |
466 | if (!page) | |
467 | goto drop; | |
468 | ||
469 | if (!PageTransHuge(page)) { | |
470 | unlock_page(page); | |
471 | put_page(page); | |
472 | goto drop; | |
473 | } | |
474 | ||
475 | ret = split_huge_page(page); | |
476 | unlock_page(page); | |
477 | put_page(page); | |
478 | ||
479 | if (ret) { | |
480 | /* split failed: leave it on the list */ | |
481 | iput(inode); | |
482 | continue; | |
483 | } | |
484 | ||
485 | split++; | |
486 | drop: | |
487 | list_del_init(&info->shrinklist); | |
488 | removed++; | |
489 | iput(inode); | |
490 | } | |
491 | ||
492 | spin_lock(&sbinfo->shrinklist_lock); | |
493 | list_splice_tail(&list, &sbinfo->shrinklist); | |
494 | sbinfo->shrinklist_len -= removed; | |
495 | spin_unlock(&sbinfo->shrinklist_lock); | |
496 | ||
497 | return split; | |
498 | } | |
499 | ||
500 | static long shmem_unused_huge_scan(struct super_block *sb, | |
501 | struct shrink_control *sc) | |
502 | { | |
503 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
504 | ||
505 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
506 | return SHRINK_STOP; | |
507 | ||
508 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
509 | } | |
510 | ||
511 | static long shmem_unused_huge_count(struct super_block *sb, | |
512 | struct shrink_control *sc) | |
513 | { | |
514 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
515 | return READ_ONCE(sbinfo->shrinklist_len); | |
516 | } | |
e496cf3d | 517 | #else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 KS |
518 | |
519 | #define shmem_huge SHMEM_HUGE_DENY | |
520 | ||
779750d2 KS |
521 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
522 | struct shrink_control *sc, unsigned long nr_to_split) | |
523 | { | |
524 | return 0; | |
525 | } | |
e496cf3d | 526 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */ |
5a6e75f8 | 527 | |
46f65ec1 HD |
528 | /* |
529 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
530 | */ | |
531 | static int shmem_add_to_page_cache(struct page *page, | |
532 | struct address_space *mapping, | |
fed400a1 | 533 | pgoff_t index, void *expected) |
46f65ec1 | 534 | { |
800d8c63 | 535 | int error, nr = hpage_nr_pages(page); |
46f65ec1 | 536 | |
800d8c63 KS |
537 | VM_BUG_ON_PAGE(PageTail(page), page); |
538 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
539 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
540 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 541 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 542 | |
800d8c63 | 543 | page_ref_add(page, nr); |
b065b432 HD |
544 | page->mapping = mapping; |
545 | page->index = index; | |
546 | ||
547 | spin_lock_irq(&mapping->tree_lock); | |
800d8c63 KS |
548 | if (PageTransHuge(page)) { |
549 | void __rcu **results; | |
550 | pgoff_t idx; | |
551 | int i; | |
552 | ||
553 | error = 0; | |
554 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, | |
555 | &results, &idx, index, 1) && | |
556 | idx < index + HPAGE_PMD_NR) { | |
557 | error = -EEXIST; | |
558 | } | |
559 | ||
560 | if (!error) { | |
561 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
562 | error = radix_tree_insert(&mapping->page_tree, | |
563 | index + i, page + i); | |
564 | VM_BUG_ON(error); | |
565 | } | |
566 | count_vm_event(THP_FILE_ALLOC); | |
567 | } | |
568 | } else if (!expected) { | |
b065b432 | 569 | error = radix_tree_insert(&mapping->page_tree, index, page); |
800d8c63 | 570 | } else { |
b065b432 HD |
571 | error = shmem_radix_tree_replace(mapping, index, expected, |
572 | page); | |
800d8c63 KS |
573 | } |
574 | ||
46f65ec1 | 575 | if (!error) { |
800d8c63 KS |
576 | mapping->nrpages += nr; |
577 | if (PageTransHuge(page)) | |
578 | __inc_zone_page_state(page, NR_SHMEM_THPS); | |
579 | __mod_zone_page_state(page_zone(page), NR_FILE_PAGES, nr); | |
580 | __mod_zone_page_state(page_zone(page), NR_SHMEM, nr); | |
b065b432 HD |
581 | spin_unlock_irq(&mapping->tree_lock); |
582 | } else { | |
583 | page->mapping = NULL; | |
584 | spin_unlock_irq(&mapping->tree_lock); | |
800d8c63 | 585 | page_ref_sub(page, nr); |
46f65ec1 | 586 | } |
46f65ec1 HD |
587 | return error; |
588 | } | |
589 | ||
6922c0c7 HD |
590 | /* |
591 | * Like delete_from_page_cache, but substitutes swap for page. | |
592 | */ | |
593 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
594 | { | |
595 | struct address_space *mapping = page->mapping; | |
596 | int error; | |
597 | ||
800d8c63 KS |
598 | VM_BUG_ON_PAGE(PageCompound(page), page); |
599 | ||
6922c0c7 HD |
600 | spin_lock_irq(&mapping->tree_lock); |
601 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
602 | page->mapping = NULL; | |
603 | mapping->nrpages--; | |
604 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
605 | __dec_zone_page_state(page, NR_SHMEM); | |
606 | spin_unlock_irq(&mapping->tree_lock); | |
09cbfeaf | 607 | put_page(page); |
6922c0c7 HD |
608 | BUG_ON(error); |
609 | } | |
610 | ||
7a5d0fbb HD |
611 | /* |
612 | * Remove swap entry from radix tree, free the swap and its page cache. | |
613 | */ | |
614 | static int shmem_free_swap(struct address_space *mapping, | |
615 | pgoff_t index, void *radswap) | |
616 | { | |
6dbaf22c | 617 | void *old; |
7a5d0fbb HD |
618 | |
619 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 620 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 621 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
622 | if (old != radswap) |
623 | return -ENOENT; | |
624 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
625 | return 0; | |
7a5d0fbb HD |
626 | } |
627 | ||
6a15a370 VB |
628 | /* |
629 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 630 | * given offsets are swapped out. |
6a15a370 VB |
631 | * |
632 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
633 | * as long as the inode doesn't go away and racy results are not a problem. | |
634 | */ | |
48131e03 VB |
635 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
636 | pgoff_t start, pgoff_t end) | |
6a15a370 | 637 | { |
6a15a370 VB |
638 | struct radix_tree_iter iter; |
639 | void **slot; | |
640 | struct page *page; | |
48131e03 | 641 | unsigned long swapped = 0; |
6a15a370 VB |
642 | |
643 | rcu_read_lock(); | |
644 | ||
6a15a370 VB |
645 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
646 | if (iter.index >= end) | |
647 | break; | |
648 | ||
649 | page = radix_tree_deref_slot(slot); | |
650 | ||
2cf938aa MW |
651 | if (radix_tree_deref_retry(page)) { |
652 | slot = radix_tree_iter_retry(&iter); | |
653 | continue; | |
654 | } | |
6a15a370 VB |
655 | |
656 | if (radix_tree_exceptional_entry(page)) | |
657 | swapped++; | |
658 | ||
659 | if (need_resched()) { | |
660 | cond_resched_rcu(); | |
7165092f | 661 | slot = radix_tree_iter_next(&iter); |
6a15a370 VB |
662 | } |
663 | } | |
664 | ||
665 | rcu_read_unlock(); | |
666 | ||
667 | return swapped << PAGE_SHIFT; | |
668 | } | |
669 | ||
48131e03 VB |
670 | /* |
671 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
672 | * given vma is swapped out. | |
673 | * | |
674 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
675 | * as long as the inode doesn't go away and racy results are not a problem. | |
676 | */ | |
677 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
678 | { | |
679 | struct inode *inode = file_inode(vma->vm_file); | |
680 | struct shmem_inode_info *info = SHMEM_I(inode); | |
681 | struct address_space *mapping = inode->i_mapping; | |
682 | unsigned long swapped; | |
683 | ||
684 | /* Be careful as we don't hold info->lock */ | |
685 | swapped = READ_ONCE(info->swapped); | |
686 | ||
687 | /* | |
688 | * The easier cases are when the shmem object has nothing in swap, or | |
689 | * the vma maps it whole. Then we can simply use the stats that we | |
690 | * already track. | |
691 | */ | |
692 | if (!swapped) | |
693 | return 0; | |
694 | ||
695 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
696 | return swapped << PAGE_SHIFT; | |
697 | ||
698 | /* Here comes the more involved part */ | |
699 | return shmem_partial_swap_usage(mapping, | |
700 | linear_page_index(vma, vma->vm_start), | |
701 | linear_page_index(vma, vma->vm_end)); | |
702 | } | |
703 | ||
24513264 HD |
704 | /* |
705 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
706 | */ | |
707 | void shmem_unlock_mapping(struct address_space *mapping) | |
708 | { | |
709 | struct pagevec pvec; | |
710 | pgoff_t indices[PAGEVEC_SIZE]; | |
711 | pgoff_t index = 0; | |
712 | ||
713 | pagevec_init(&pvec, 0); | |
714 | /* | |
715 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
716 | */ | |
717 | while (!mapping_unevictable(mapping)) { | |
718 | /* | |
719 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
720 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
721 | */ | |
0cd6144a JW |
722 | pvec.nr = find_get_entries(mapping, index, |
723 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
724 | if (!pvec.nr) |
725 | break; | |
726 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 727 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
728 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
729 | pagevec_release(&pvec); | |
730 | cond_resched(); | |
731 | } | |
7a5d0fbb HD |
732 | } |
733 | ||
734 | /* | |
735 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 736 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 737 | */ |
1635f6a7 HD |
738 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
739 | bool unfalloc) | |
1da177e4 | 740 | { |
285b2c4f | 741 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 742 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
743 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
744 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
745 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
746 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 747 | struct pagevec pvec; |
7a5d0fbb HD |
748 | pgoff_t indices[PAGEVEC_SIZE]; |
749 | long nr_swaps_freed = 0; | |
285b2c4f | 750 | pgoff_t index; |
bda97eab HD |
751 | int i; |
752 | ||
83e4fa9c HD |
753 | if (lend == -1) |
754 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
755 | |
756 | pagevec_init(&pvec, 0); | |
757 | index = start; | |
83e4fa9c | 758 | while (index < end) { |
0cd6144a JW |
759 | pvec.nr = find_get_entries(mapping, index, |
760 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
761 | pvec.pages, indices); | |
7a5d0fbb HD |
762 | if (!pvec.nr) |
763 | break; | |
bda97eab HD |
764 | for (i = 0; i < pagevec_count(&pvec); i++) { |
765 | struct page *page = pvec.pages[i]; | |
766 | ||
7a5d0fbb | 767 | index = indices[i]; |
83e4fa9c | 768 | if (index >= end) |
bda97eab HD |
769 | break; |
770 | ||
7a5d0fbb | 771 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
772 | if (unfalloc) |
773 | continue; | |
7a5d0fbb HD |
774 | nr_swaps_freed += !shmem_free_swap(mapping, |
775 | index, page); | |
bda97eab | 776 | continue; |
7a5d0fbb HD |
777 | } |
778 | ||
800d8c63 KS |
779 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
780 | ||
7a5d0fbb | 781 | if (!trylock_page(page)) |
bda97eab | 782 | continue; |
800d8c63 KS |
783 | |
784 | if (PageTransTail(page)) { | |
785 | /* Middle of THP: zero out the page */ | |
786 | clear_highpage(page); | |
787 | unlock_page(page); | |
788 | continue; | |
789 | } else if (PageTransHuge(page)) { | |
790 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
791 | /* | |
792 | * Range ends in the middle of THP: | |
793 | * zero out the page | |
794 | */ | |
795 | clear_highpage(page); | |
796 | unlock_page(page); | |
797 | continue; | |
798 | } | |
799 | index += HPAGE_PMD_NR - 1; | |
800 | i += HPAGE_PMD_NR - 1; | |
801 | } | |
802 | ||
1635f6a7 | 803 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
804 | VM_BUG_ON_PAGE(PageTail(page), page); |
805 | if (page_mapping(page) == mapping) { | |
309381fe | 806 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
807 | truncate_inode_page(mapping, page); |
808 | } | |
bda97eab | 809 | } |
bda97eab HD |
810 | unlock_page(page); |
811 | } | |
0cd6144a | 812 | pagevec_remove_exceptionals(&pvec); |
24513264 | 813 | pagevec_release(&pvec); |
bda97eab HD |
814 | cond_resched(); |
815 | index++; | |
816 | } | |
1da177e4 | 817 | |
83e4fa9c | 818 | if (partial_start) { |
bda97eab | 819 | struct page *page = NULL; |
9e18eb29 | 820 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 821 | if (page) { |
09cbfeaf | 822 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
823 | if (start > end) { |
824 | top = partial_end; | |
825 | partial_end = 0; | |
826 | } | |
827 | zero_user_segment(page, partial_start, top); | |
828 | set_page_dirty(page); | |
829 | unlock_page(page); | |
09cbfeaf | 830 | put_page(page); |
83e4fa9c HD |
831 | } |
832 | } | |
833 | if (partial_end) { | |
834 | struct page *page = NULL; | |
9e18eb29 | 835 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
836 | if (page) { |
837 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
838 | set_page_dirty(page); |
839 | unlock_page(page); | |
09cbfeaf | 840 | put_page(page); |
bda97eab HD |
841 | } |
842 | } | |
83e4fa9c HD |
843 | if (start >= end) |
844 | return; | |
bda97eab HD |
845 | |
846 | index = start; | |
b1a36650 | 847 | while (index < end) { |
bda97eab | 848 | cond_resched(); |
0cd6144a JW |
849 | |
850 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 851 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 852 | pvec.pages, indices); |
7a5d0fbb | 853 | if (!pvec.nr) { |
b1a36650 HD |
854 | /* If all gone or hole-punch or unfalloc, we're done */ |
855 | if (index == start || end != -1) | |
bda97eab | 856 | break; |
b1a36650 | 857 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
858 | index = start; |
859 | continue; | |
860 | } | |
bda97eab HD |
861 | for (i = 0; i < pagevec_count(&pvec); i++) { |
862 | struct page *page = pvec.pages[i]; | |
863 | ||
7a5d0fbb | 864 | index = indices[i]; |
83e4fa9c | 865 | if (index >= end) |
bda97eab HD |
866 | break; |
867 | ||
7a5d0fbb | 868 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
869 | if (unfalloc) |
870 | continue; | |
b1a36650 HD |
871 | if (shmem_free_swap(mapping, index, page)) { |
872 | /* Swap was replaced by page: retry */ | |
873 | index--; | |
874 | break; | |
875 | } | |
876 | nr_swaps_freed++; | |
7a5d0fbb HD |
877 | continue; |
878 | } | |
879 | ||
bda97eab | 880 | lock_page(page); |
800d8c63 KS |
881 | |
882 | if (PageTransTail(page)) { | |
883 | /* Middle of THP: zero out the page */ | |
884 | clear_highpage(page); | |
885 | unlock_page(page); | |
886 | /* | |
887 | * Partial thp truncate due 'start' in middle | |
888 | * of THP: don't need to look on these pages | |
889 | * again on !pvec.nr restart. | |
890 | */ | |
891 | if (index != round_down(end, HPAGE_PMD_NR)) | |
892 | start++; | |
893 | continue; | |
894 | } else if (PageTransHuge(page)) { | |
895 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
896 | /* | |
897 | * Range ends in the middle of THP: | |
898 | * zero out the page | |
899 | */ | |
900 | clear_highpage(page); | |
901 | unlock_page(page); | |
902 | continue; | |
903 | } | |
904 | index += HPAGE_PMD_NR - 1; | |
905 | i += HPAGE_PMD_NR - 1; | |
906 | } | |
907 | ||
1635f6a7 | 908 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
909 | VM_BUG_ON_PAGE(PageTail(page), page); |
910 | if (page_mapping(page) == mapping) { | |
309381fe | 911 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 912 | truncate_inode_page(mapping, page); |
b1a36650 HD |
913 | } else { |
914 | /* Page was replaced by swap: retry */ | |
915 | unlock_page(page); | |
916 | index--; | |
917 | break; | |
1635f6a7 | 918 | } |
7a5d0fbb | 919 | } |
bda97eab HD |
920 | unlock_page(page); |
921 | } | |
0cd6144a | 922 | pagevec_remove_exceptionals(&pvec); |
24513264 | 923 | pagevec_release(&pvec); |
bda97eab HD |
924 | index++; |
925 | } | |
94c1e62d | 926 | |
4595ef88 | 927 | spin_lock_irq(&info->lock); |
7a5d0fbb | 928 | info->swapped -= nr_swaps_freed; |
1da177e4 | 929 | shmem_recalc_inode(inode); |
4595ef88 | 930 | spin_unlock_irq(&info->lock); |
1635f6a7 | 931 | } |
1da177e4 | 932 | |
1635f6a7 HD |
933 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
934 | { | |
935 | shmem_undo_range(inode, lstart, lend, false); | |
285b2c4f | 936 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1da177e4 | 937 | } |
94c1e62d | 938 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 939 | |
44a30220 YZ |
940 | static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry, |
941 | struct kstat *stat) | |
942 | { | |
943 | struct inode *inode = dentry->d_inode; | |
944 | struct shmem_inode_info *info = SHMEM_I(inode); | |
945 | ||
d0424c42 | 946 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
4595ef88 | 947 | spin_lock_irq(&info->lock); |
d0424c42 | 948 | shmem_recalc_inode(inode); |
4595ef88 | 949 | spin_unlock_irq(&info->lock); |
d0424c42 | 950 | } |
44a30220 | 951 | generic_fillattr(inode, stat); |
44a30220 YZ |
952 | return 0; |
953 | } | |
954 | ||
94c1e62d | 955 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 956 | { |
75c3cfa8 | 957 | struct inode *inode = d_inode(dentry); |
40e041a2 | 958 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 959 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 LT |
960 | int error; |
961 | ||
db78b877 CH |
962 | error = inode_change_ok(inode, attr); |
963 | if (error) | |
964 | return error; | |
965 | ||
94c1e62d HD |
966 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
967 | loff_t oldsize = inode->i_size; | |
968 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 969 | |
40e041a2 DH |
970 | /* protected by i_mutex */ |
971 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
972 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
973 | return -EPERM; | |
974 | ||
94c1e62d | 975 | if (newsize != oldsize) { |
77142517 KK |
976 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
977 | oldsize, newsize); | |
978 | if (error) | |
979 | return error; | |
94c1e62d HD |
980 | i_size_write(inode, newsize); |
981 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
982 | } | |
afa2db2f | 983 | if (newsize <= oldsize) { |
94c1e62d | 984 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
985 | if (oldsize > holebegin) |
986 | unmap_mapping_range(inode->i_mapping, | |
987 | holebegin, 0, 1); | |
988 | if (info->alloced) | |
989 | shmem_truncate_range(inode, | |
990 | newsize, (loff_t)-1); | |
94c1e62d | 991 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
992 | if (oldsize > holebegin) |
993 | unmap_mapping_range(inode->i_mapping, | |
994 | holebegin, 0, 1); | |
779750d2 KS |
995 | |
996 | /* | |
997 | * Part of the huge page can be beyond i_size: subject | |
998 | * to shrink under memory pressure. | |
999 | */ | |
1000 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) { | |
1001 | spin_lock(&sbinfo->shrinklist_lock); | |
1002 | if (list_empty(&info->shrinklist)) { | |
1003 | list_add_tail(&info->shrinklist, | |
1004 | &sbinfo->shrinklist); | |
1005 | sbinfo->shrinklist_len++; | |
1006 | } | |
1007 | spin_unlock(&sbinfo->shrinklist_lock); | |
1008 | } | |
94c1e62d | 1009 | } |
1da177e4 LT |
1010 | } |
1011 | ||
db78b877 | 1012 | setattr_copy(inode, attr); |
db78b877 | 1013 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 1014 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
1015 | return error; |
1016 | } | |
1017 | ||
1f895f75 | 1018 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1019 | { |
1da177e4 | 1020 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1021 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 | 1022 | |
3889e6e7 | 1023 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
1024 | shmem_unacct_size(info->flags, inode->i_size); |
1025 | inode->i_size = 0; | |
3889e6e7 | 1026 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1027 | if (!list_empty(&info->shrinklist)) { |
1028 | spin_lock(&sbinfo->shrinklist_lock); | |
1029 | if (!list_empty(&info->shrinklist)) { | |
1030 | list_del_init(&info->shrinklist); | |
1031 | sbinfo->shrinklist_len--; | |
1032 | } | |
1033 | spin_unlock(&sbinfo->shrinklist_lock); | |
1034 | } | |
1da177e4 | 1035 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 1036 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 1037 | list_del_init(&info->swaplist); |
cb5f7b9a | 1038 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1039 | } |
3ed47db3 | 1040 | } |
b09e0fa4 | 1041 | |
38f38657 | 1042 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 1043 | WARN_ON(inode->i_blocks); |
5b04c689 | 1044 | shmem_free_inode(inode->i_sb); |
dbd5768f | 1045 | clear_inode(inode); |
1da177e4 LT |
1046 | } |
1047 | ||
46f65ec1 HD |
1048 | /* |
1049 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1050 | */ | |
41ffe5d5 | 1051 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 1052 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 1053 | { |
285b2c4f | 1054 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 1055 | void *radswap; |
41ffe5d5 | 1056 | pgoff_t index; |
bde05d1c HD |
1057 | gfp_t gfp; |
1058 | int error = 0; | |
1da177e4 | 1059 | |
46f65ec1 | 1060 | radswap = swp_to_radix_entry(swap); |
e504f3fd | 1061 | index = radix_tree_locate_item(&mapping->page_tree, radswap); |
46f65ec1 | 1062 | if (index == -1) |
00501b53 | 1063 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 1064 | |
1b1b32f2 HD |
1065 | /* |
1066 | * Move _head_ to start search for next from here. | |
1f895f75 | 1067 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 1068 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 1069 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
1070 | */ |
1071 | if (shmem_swaplist.next != &info->swaplist) | |
1072 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 1073 | |
bde05d1c HD |
1074 | gfp = mapping_gfp_mask(mapping); |
1075 | if (shmem_should_replace_page(*pagep, gfp)) { | |
1076 | mutex_unlock(&shmem_swaplist_mutex); | |
1077 | error = shmem_replace_page(pagep, gfp, info, index); | |
1078 | mutex_lock(&shmem_swaplist_mutex); | |
1079 | /* | |
1080 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
1081 | * allocation, but the inode might have been freed while we |
1082 | * dropped it: although a racing shmem_evict_inode() cannot | |
1083 | * complete without emptying the radix_tree, our page lock | |
1084 | * on this swapcache page is not enough to prevent that - | |
1085 | * free_swap_and_cache() of our swap entry will only | |
1086 | * trylock_page(), removing swap from radix_tree whatever. | |
1087 | * | |
1088 | * We must not proceed to shmem_add_to_page_cache() if the | |
1089 | * inode has been freed, but of course we cannot rely on | |
1090 | * inode or mapping or info to check that. However, we can | |
1091 | * safely check if our swap entry is still in use (and here | |
1092 | * it can't have got reused for another page): if it's still | |
1093 | * in use, then the inode cannot have been freed yet, and we | |
1094 | * can safely proceed (if it's no longer in use, that tells | |
1095 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
1096 | */ |
1097 | if (!page_swapcount(*pagep)) | |
1098 | error = -ENOENT; | |
1099 | } | |
1100 | ||
d13d1443 | 1101 | /* |
778dd893 HD |
1102 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
1103 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
1104 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 1105 | */ |
bde05d1c HD |
1106 | if (!error) |
1107 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 1108 | radswap); |
48f170fb | 1109 | if (error != -ENOMEM) { |
46f65ec1 HD |
1110 | /* |
1111 | * Truncation and eviction use free_swap_and_cache(), which | |
1112 | * only does trylock page: if we raced, best clean up here. | |
1113 | */ | |
bde05d1c HD |
1114 | delete_from_swap_cache(*pagep); |
1115 | set_page_dirty(*pagep); | |
46f65ec1 | 1116 | if (!error) { |
4595ef88 | 1117 | spin_lock_irq(&info->lock); |
46f65ec1 | 1118 | info->swapped--; |
4595ef88 | 1119 | spin_unlock_irq(&info->lock); |
46f65ec1 HD |
1120 | swap_free(swap); |
1121 | } | |
1da177e4 | 1122 | } |
2e0e26c7 | 1123 | return error; |
1da177e4 LT |
1124 | } |
1125 | ||
1126 | /* | |
46f65ec1 | 1127 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 1128 | */ |
41ffe5d5 | 1129 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 1130 | { |
41ffe5d5 | 1131 | struct list_head *this, *next; |
1da177e4 | 1132 | struct shmem_inode_info *info; |
00501b53 | 1133 | struct mem_cgroup *memcg; |
bde05d1c HD |
1134 | int error = 0; |
1135 | ||
1136 | /* | |
1137 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 1138 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 1139 | */ |
0142ef6c | 1140 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 1141 | goto out; |
778dd893 HD |
1142 | |
1143 | /* | |
1144 | * Charge page using GFP_KERNEL while we can wait, before taking | |
1145 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
1146 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 1147 | */ |
f627c2f5 KS |
1148 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg, |
1149 | false); | |
778dd893 HD |
1150 | if (error) |
1151 | goto out; | |
46f65ec1 | 1152 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 1153 | error = -EAGAIN; |
1da177e4 | 1154 | |
cb5f7b9a | 1155 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
1156 | list_for_each_safe(this, next, &shmem_swaplist) { |
1157 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 1158 | if (info->swapped) |
00501b53 | 1159 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
1160 | else |
1161 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1162 | cond_resched(); |
00501b53 | 1163 | if (error != -EAGAIN) |
778dd893 | 1164 | break; |
00501b53 | 1165 | /* found nothing in this: move on to search the next */ |
1da177e4 | 1166 | } |
cb5f7b9a | 1167 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1168 | |
00501b53 JW |
1169 | if (error) { |
1170 | if (error != -ENOMEM) | |
1171 | error = 0; | |
f627c2f5 | 1172 | mem_cgroup_cancel_charge(page, memcg, false); |
00501b53 | 1173 | } else |
f627c2f5 | 1174 | mem_cgroup_commit_charge(page, memcg, true, false); |
778dd893 | 1175 | out: |
aaa46865 | 1176 | unlock_page(page); |
09cbfeaf | 1177 | put_page(page); |
778dd893 | 1178 | return error; |
1da177e4 LT |
1179 | } |
1180 | ||
1181 | /* | |
1182 | * Move the page from the page cache to the swap cache. | |
1183 | */ | |
1184 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1185 | { | |
1186 | struct shmem_inode_info *info; | |
1da177e4 | 1187 | struct address_space *mapping; |
1da177e4 | 1188 | struct inode *inode; |
6922c0c7 HD |
1189 | swp_entry_t swap; |
1190 | pgoff_t index; | |
1da177e4 | 1191 | |
800d8c63 | 1192 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1193 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1194 | mapping = page->mapping; |
1195 | index = page->index; | |
1196 | inode = mapping->host; | |
1197 | info = SHMEM_I(inode); | |
1198 | if (info->flags & VM_LOCKED) | |
1199 | goto redirty; | |
d9fe526a | 1200 | if (!total_swap_pages) |
1da177e4 LT |
1201 | goto redirty; |
1202 | ||
d9fe526a | 1203 | /* |
97b713ba CH |
1204 | * Our capabilities prevent regular writeback or sync from ever calling |
1205 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1206 | * its underlying filesystem, in which case tmpfs should write out to | |
1207 | * swap only in response to memory pressure, and not for the writeback | |
1208 | * threads or sync. | |
d9fe526a | 1209 | */ |
48f170fb HD |
1210 | if (!wbc->for_reclaim) { |
1211 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1212 | goto redirty; | |
1213 | } | |
1635f6a7 HD |
1214 | |
1215 | /* | |
1216 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1217 | * value into swapfile.c, the only way we can correctly account for a | |
1218 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1219 | * |
1220 | * That's okay for a page already fallocated earlier, but if we have | |
1221 | * not yet completed the fallocation, then (a) we want to keep track | |
1222 | * of this page in case we have to undo it, and (b) it may not be a | |
1223 | * good idea to continue anyway, once we're pushing into swap. So | |
1224 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1225 | */ |
1226 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1227 | if (inode->i_private) { |
1228 | struct shmem_falloc *shmem_falloc; | |
1229 | spin_lock(&inode->i_lock); | |
1230 | shmem_falloc = inode->i_private; | |
1231 | if (shmem_falloc && | |
8e205f77 | 1232 | !shmem_falloc->waitq && |
1aac1400 HD |
1233 | index >= shmem_falloc->start && |
1234 | index < shmem_falloc->next) | |
1235 | shmem_falloc->nr_unswapped++; | |
1236 | else | |
1237 | shmem_falloc = NULL; | |
1238 | spin_unlock(&inode->i_lock); | |
1239 | if (shmem_falloc) | |
1240 | goto redirty; | |
1241 | } | |
1635f6a7 HD |
1242 | clear_highpage(page); |
1243 | flush_dcache_page(page); | |
1244 | SetPageUptodate(page); | |
1245 | } | |
1246 | ||
48f170fb HD |
1247 | swap = get_swap_page(); |
1248 | if (!swap.val) | |
1249 | goto redirty; | |
d9fe526a | 1250 | |
37e84351 VD |
1251 | if (mem_cgroup_try_charge_swap(page, swap)) |
1252 | goto free_swap; | |
1253 | ||
b1dea800 HD |
1254 | /* |
1255 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1256 | * if it's not already there. Do it now before the page is |
1257 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1258 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1259 | * we've incremented swapped, because shmem_unuse_inode() will |
1260 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1261 | */ |
48f170fb HD |
1262 | mutex_lock(&shmem_swaplist_mutex); |
1263 | if (list_empty(&info->swaplist)) | |
1264 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 1265 | |
48f170fb | 1266 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
4595ef88 | 1267 | spin_lock_irq(&info->lock); |
6922c0c7 | 1268 | shmem_recalc_inode(inode); |
267a4c76 | 1269 | info->swapped++; |
4595ef88 | 1270 | spin_unlock_irq(&info->lock); |
6922c0c7 | 1271 | |
267a4c76 HD |
1272 | swap_shmem_alloc(swap); |
1273 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1274 | ||
6922c0c7 | 1275 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1276 | BUG_ON(page_mapped(page)); |
9fab5619 | 1277 | swap_writepage(page, wbc); |
1da177e4 LT |
1278 | return 0; |
1279 | } | |
1280 | ||
6922c0c7 | 1281 | mutex_unlock(&shmem_swaplist_mutex); |
37e84351 | 1282 | free_swap: |
0a31bc97 | 1283 | swapcache_free(swap); |
1da177e4 LT |
1284 | redirty: |
1285 | set_page_dirty(page); | |
d9fe526a HD |
1286 | if (wbc->for_reclaim) |
1287 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1288 | unlock_page(page); | |
1289 | return 0; | |
1da177e4 LT |
1290 | } |
1291 | ||
75edd345 | 1292 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1293 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1294 | { |
095f1fc4 | 1295 | char buffer[64]; |
680d794b | 1296 | |
71fe804b | 1297 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1298 | return; /* show nothing */ |
680d794b | 1299 | |
a7a88b23 | 1300 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1301 | |
1302 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1303 | } |
71fe804b LS |
1304 | |
1305 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1306 | { | |
1307 | struct mempolicy *mpol = NULL; | |
1308 | if (sbinfo->mpol) { | |
1309 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1310 | mpol = sbinfo->mpol; | |
1311 | mpol_get(mpol); | |
1312 | spin_unlock(&sbinfo->stat_lock); | |
1313 | } | |
1314 | return mpol; | |
1315 | } | |
75edd345 HD |
1316 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1317 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1318 | { | |
1319 | } | |
1320 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1321 | { | |
1322 | return NULL; | |
1323 | } | |
1324 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1325 | #ifndef CONFIG_NUMA | |
1326 | #define vm_policy vm_private_data | |
1327 | #endif | |
680d794b | 1328 | |
800d8c63 KS |
1329 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1330 | struct shmem_inode_info *info, pgoff_t index) | |
1331 | { | |
1332 | /* Create a pseudo vma that just contains the policy */ | |
1333 | vma->vm_start = 0; | |
1334 | /* Bias interleave by inode number to distribute better across nodes */ | |
1335 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
1336 | vma->vm_ops = NULL; | |
1337 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); | |
1338 | } | |
1339 | ||
1340 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1341 | { | |
1342 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1343 | mpol_cond_put(vma->vm_policy); | |
1344 | } | |
1345 | ||
41ffe5d5 HD |
1346 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1347 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1348 | { |
1da177e4 | 1349 | struct vm_area_struct pvma; |
18a2f371 | 1350 | struct page *page; |
52cd3b07 | 1351 | |
800d8c63 | 1352 | shmem_pseudo_vma_init(&pvma, info, index); |
18a2f371 | 1353 | page = swapin_readahead(swap, gfp, &pvma, 0); |
800d8c63 | 1354 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1355 | |
800d8c63 KS |
1356 | return page; |
1357 | } | |
1358 | ||
1359 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1360 | struct shmem_inode_info *info, pgoff_t index) | |
1361 | { | |
1362 | struct vm_area_struct pvma; | |
1363 | struct inode *inode = &info->vfs_inode; | |
1364 | struct address_space *mapping = inode->i_mapping; | |
1365 | pgoff_t idx, hindex = round_down(index, HPAGE_PMD_NR); | |
1366 | void __rcu **results; | |
1367 | struct page *page; | |
1368 | ||
e496cf3d | 1369 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1370 | return NULL; |
1371 | ||
1372 | rcu_read_lock(); | |
1373 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx, | |
1374 | hindex, 1) && idx < hindex + HPAGE_PMD_NR) { | |
1375 | rcu_read_unlock(); | |
1376 | return NULL; | |
1377 | } | |
1378 | rcu_read_unlock(); | |
18a2f371 | 1379 | |
800d8c63 KS |
1380 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1381 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
1382 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); | |
1383 | shmem_pseudo_vma_destroy(&pvma); | |
1384 | if (page) | |
1385 | prep_transhuge_page(page); | |
18a2f371 | 1386 | return page; |
1da177e4 LT |
1387 | } |
1388 | ||
02098fea | 1389 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1390 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1391 | { |
1392 | struct vm_area_struct pvma; | |
18a2f371 | 1393 | struct page *page; |
1da177e4 | 1394 | |
800d8c63 KS |
1395 | shmem_pseudo_vma_init(&pvma, info, index); |
1396 | page = alloc_page_vma(gfp, &pvma, 0); | |
1397 | shmem_pseudo_vma_destroy(&pvma); | |
1398 | ||
1399 | return page; | |
1400 | } | |
1401 | ||
1402 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
1403 | struct shmem_inode_info *info, struct shmem_sb_info *sbinfo, | |
1404 | pgoff_t index, bool huge) | |
1405 | { | |
1406 | struct page *page; | |
1407 | int nr; | |
1408 | int err = -ENOSPC; | |
52cd3b07 | 1409 | |
e496cf3d | 1410 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
800d8c63 KS |
1411 | huge = false; |
1412 | nr = huge ? HPAGE_PMD_NR : 1; | |
1413 | ||
1414 | if (shmem_acct_block(info->flags, nr)) | |
1415 | goto failed; | |
1416 | if (sbinfo->max_blocks) { | |
1417 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
1418 | sbinfo->max_blocks - nr) > 0) | |
1419 | goto unacct; | |
1420 | percpu_counter_add(&sbinfo->used_blocks, nr); | |
1421 | } | |
1422 | ||
1423 | if (huge) | |
1424 | page = shmem_alloc_hugepage(gfp, info, index); | |
1425 | else | |
1426 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1427 | if (page) { |
1428 | __SetPageLocked(page); | |
1429 | __SetPageSwapBacked(page); | |
800d8c63 | 1430 | return page; |
75edd345 | 1431 | } |
18a2f371 | 1432 | |
800d8c63 KS |
1433 | err = -ENOMEM; |
1434 | if (sbinfo->max_blocks) | |
1435 | percpu_counter_add(&sbinfo->used_blocks, -nr); | |
1436 | unacct: | |
1437 | shmem_unacct_blocks(info->flags, nr); | |
1438 | failed: | |
1439 | return ERR_PTR(err); | |
1da177e4 | 1440 | } |
71fe804b | 1441 | |
bde05d1c HD |
1442 | /* |
1443 | * When a page is moved from swapcache to shmem filecache (either by the | |
1444 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1445 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1446 | * ignorance of the mapping it belongs to. If that mapping has special | |
1447 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1448 | * we may need to copy to a suitable page before moving to filecache. | |
1449 | * | |
1450 | * In a future release, this may well be extended to respect cpuset and | |
1451 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1452 | * but for now it is a simple matter of zone. | |
1453 | */ | |
1454 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1455 | { | |
1456 | return page_zonenum(page) > gfp_zone(gfp); | |
1457 | } | |
1458 | ||
1459 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1460 | struct shmem_inode_info *info, pgoff_t index) | |
1461 | { | |
1462 | struct page *oldpage, *newpage; | |
1463 | struct address_space *swap_mapping; | |
1464 | pgoff_t swap_index; | |
1465 | int error; | |
1466 | ||
1467 | oldpage = *pagep; | |
1468 | swap_index = page_private(oldpage); | |
1469 | swap_mapping = page_mapping(oldpage); | |
1470 | ||
1471 | /* | |
1472 | * We have arrived here because our zones are constrained, so don't | |
1473 | * limit chance of success by further cpuset and node constraints. | |
1474 | */ | |
1475 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1476 | newpage = shmem_alloc_page(gfp, info, index); | |
1477 | if (!newpage) | |
1478 | return -ENOMEM; | |
bde05d1c | 1479 | |
09cbfeaf | 1480 | get_page(newpage); |
bde05d1c | 1481 | copy_highpage(newpage, oldpage); |
0142ef6c | 1482 | flush_dcache_page(newpage); |
bde05d1c | 1483 | |
bde05d1c | 1484 | SetPageUptodate(newpage); |
bde05d1c | 1485 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1486 | SetPageSwapCache(newpage); |
1487 | ||
1488 | /* | |
1489 | * Our caller will very soon move newpage out of swapcache, but it's | |
1490 | * a nice clean interface for us to replace oldpage by newpage there. | |
1491 | */ | |
1492 | spin_lock_irq(&swap_mapping->tree_lock); | |
1493 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1494 | newpage); | |
0142ef6c HD |
1495 | if (!error) { |
1496 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
1497 | __dec_zone_page_state(oldpage, NR_FILE_PAGES); | |
1498 | } | |
bde05d1c | 1499 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1500 | |
0142ef6c HD |
1501 | if (unlikely(error)) { |
1502 | /* | |
1503 | * Is this possible? I think not, now that our callers check | |
1504 | * both PageSwapCache and page_private after getting page lock; | |
1505 | * but be defensive. Reverse old to newpage for clear and free. | |
1506 | */ | |
1507 | oldpage = newpage; | |
1508 | } else { | |
6a93ca8f | 1509 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1510 | lru_cache_add_anon(newpage); |
1511 | *pagep = newpage; | |
1512 | } | |
bde05d1c HD |
1513 | |
1514 | ClearPageSwapCache(oldpage); | |
1515 | set_page_private(oldpage, 0); | |
1516 | ||
1517 | unlock_page(oldpage); | |
09cbfeaf KS |
1518 | put_page(oldpage); |
1519 | put_page(oldpage); | |
0142ef6c | 1520 | return error; |
bde05d1c HD |
1521 | } |
1522 | ||
1da177e4 | 1523 | /* |
68da9f05 | 1524 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1525 | * |
1526 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1527 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1528 | * entry since a page cannot live in both the swap and page cache. |
1529 | * | |
1530 | * fault_mm and fault_type are only supplied by shmem_fault: | |
1531 | * otherwise they are NULL. | |
1da177e4 | 1532 | */ |
41ffe5d5 | 1533 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 ALC |
1534 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
1535 | struct mm_struct *fault_mm, int *fault_type) | |
1da177e4 LT |
1536 | { |
1537 | struct address_space *mapping = inode->i_mapping; | |
54af6042 | 1538 | struct shmem_inode_info *info; |
1da177e4 | 1539 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1540 | struct mm_struct *charge_mm; |
00501b53 | 1541 | struct mem_cgroup *memcg; |
27ab7006 | 1542 | struct page *page; |
1da177e4 | 1543 | swp_entry_t swap; |
657e3038 | 1544 | enum sgp_type sgp_huge = sgp; |
800d8c63 | 1545 | pgoff_t hindex = index; |
1da177e4 | 1546 | int error; |
54af6042 | 1547 | int once = 0; |
1635f6a7 | 1548 | int alloced = 0; |
1da177e4 | 1549 | |
09cbfeaf | 1550 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1551 | return -EFBIG; |
657e3038 KS |
1552 | if (sgp == SGP_NOHUGE || sgp == SGP_HUGE) |
1553 | sgp = SGP_CACHE; | |
1da177e4 | 1554 | repeat: |
54af6042 | 1555 | swap.val = 0; |
0cd6144a | 1556 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1557 | if (radix_tree_exceptional_entry(page)) { |
1558 | swap = radix_to_swp_entry(page); | |
1559 | page = NULL; | |
1560 | } | |
1561 | ||
75edd345 | 1562 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1563 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 1564 | error = -EINVAL; |
267a4c76 | 1565 | goto unlock; |
54af6042 HD |
1566 | } |
1567 | ||
66d2f4d2 HD |
1568 | if (page && sgp == SGP_WRITE) |
1569 | mark_page_accessed(page); | |
1570 | ||
1635f6a7 HD |
1571 | /* fallocated page? */ |
1572 | if (page && !PageUptodate(page)) { | |
1573 | if (sgp != SGP_READ) | |
1574 | goto clear; | |
1575 | unlock_page(page); | |
09cbfeaf | 1576 | put_page(page); |
1635f6a7 HD |
1577 | page = NULL; |
1578 | } | |
54af6042 | 1579 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1580 | *pagep = page; |
1581 | return 0; | |
27ab7006 HD |
1582 | } |
1583 | ||
1584 | /* | |
54af6042 HD |
1585 | * Fast cache lookup did not find it: |
1586 | * bring it back from swap or allocate. | |
27ab7006 | 1587 | */ |
54af6042 HD |
1588 | info = SHMEM_I(inode); |
1589 | sbinfo = SHMEM_SB(inode->i_sb); | |
9e18eb29 | 1590 | charge_mm = fault_mm ? : current->mm; |
1da177e4 | 1591 | |
1da177e4 LT |
1592 | if (swap.val) { |
1593 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1594 | page = lookup_swap_cache(swap); |
1595 | if (!page) { | |
9e18eb29 ALC |
1596 | /* Or update major stats only when swapin succeeds?? */ |
1597 | if (fault_type) { | |
68da9f05 | 1598 | *fault_type |= VM_FAULT_MAJOR; |
9e18eb29 ALC |
1599 | count_vm_event(PGMAJFAULT); |
1600 | mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT); | |
1601 | } | |
1602 | /* Here we actually start the io */ | |
41ffe5d5 | 1603 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1604 | if (!page) { |
54af6042 HD |
1605 | error = -ENOMEM; |
1606 | goto failed; | |
1da177e4 | 1607 | } |
1da177e4 LT |
1608 | } |
1609 | ||
1610 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1611 | lock_page(page); |
0142ef6c | 1612 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1613 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1614 | error = -EEXIST; /* try again */ |
d1899228 | 1615 | goto unlock; |
bde05d1c | 1616 | } |
27ab7006 | 1617 | if (!PageUptodate(page)) { |
1da177e4 | 1618 | error = -EIO; |
54af6042 | 1619 | goto failed; |
1da177e4 | 1620 | } |
54af6042 HD |
1621 | wait_on_page_writeback(page); |
1622 | ||
bde05d1c HD |
1623 | if (shmem_should_replace_page(page, gfp)) { |
1624 | error = shmem_replace_page(&page, gfp, info, index); | |
1625 | if (error) | |
1626 | goto failed; | |
1da177e4 | 1627 | } |
27ab7006 | 1628 | |
9e18eb29 | 1629 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
f627c2f5 | 1630 | false); |
d1899228 | 1631 | if (!error) { |
aa3b1895 | 1632 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1633 | swp_to_radix_entry(swap)); |
215c02bc HD |
1634 | /* |
1635 | * We already confirmed swap under page lock, and make | |
1636 | * no memory allocation here, so usually no possibility | |
1637 | * of error; but free_swap_and_cache() only trylocks a | |
1638 | * page, so it is just possible that the entry has been | |
1639 | * truncated or holepunched since swap was confirmed. | |
1640 | * shmem_undo_range() will have done some of the | |
1641 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1642 | * the rest. |
215c02bc HD |
1643 | * Reset swap.val? No, leave it so "failed" goes back to |
1644 | * "repeat": reading a hole and writing should succeed. | |
1645 | */ | |
00501b53 | 1646 | if (error) { |
f627c2f5 | 1647 | mem_cgroup_cancel_charge(page, memcg, false); |
215c02bc | 1648 | delete_from_swap_cache(page); |
00501b53 | 1649 | } |
d1899228 | 1650 | } |
54af6042 HD |
1651 | if (error) |
1652 | goto failed; | |
1653 | ||
f627c2f5 | 1654 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1655 | |
4595ef88 | 1656 | spin_lock_irq(&info->lock); |
285b2c4f | 1657 | info->swapped--; |
54af6042 | 1658 | shmem_recalc_inode(inode); |
4595ef88 | 1659 | spin_unlock_irq(&info->lock); |
54af6042 | 1660 | |
66d2f4d2 HD |
1661 | if (sgp == SGP_WRITE) |
1662 | mark_page_accessed(page); | |
1663 | ||
54af6042 | 1664 | delete_from_swap_cache(page); |
27ab7006 HD |
1665 | set_page_dirty(page); |
1666 | swap_free(swap); | |
1667 | ||
54af6042 | 1668 | } else { |
800d8c63 KS |
1669 | /* shmem_symlink() */ |
1670 | if (mapping->a_ops != &shmem_aops) | |
1671 | goto alloc_nohuge; | |
657e3038 | 1672 | if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE) |
800d8c63 KS |
1673 | goto alloc_nohuge; |
1674 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1675 | goto alloc_huge; | |
1676 | switch (sbinfo->huge) { | |
1677 | loff_t i_size; | |
1678 | pgoff_t off; | |
1679 | case SHMEM_HUGE_NEVER: | |
1680 | goto alloc_nohuge; | |
1681 | case SHMEM_HUGE_WITHIN_SIZE: | |
1682 | off = round_up(index, HPAGE_PMD_NR); | |
1683 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1684 | if (i_size >= HPAGE_PMD_SIZE && | |
1685 | i_size >> PAGE_SHIFT >= off) | |
1686 | goto alloc_huge; | |
1687 | /* fallthrough */ | |
1688 | case SHMEM_HUGE_ADVISE: | |
657e3038 KS |
1689 | if (sgp_huge == SGP_HUGE) |
1690 | goto alloc_huge; | |
1691 | /* TODO: implement fadvise() hints */ | |
800d8c63 | 1692 | goto alloc_nohuge; |
54af6042 | 1693 | } |
1da177e4 | 1694 | |
800d8c63 KS |
1695 | alloc_huge: |
1696 | page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1697 | index, true); | |
1698 | if (IS_ERR(page)) { | |
1699 | alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1700 | index, false); | |
1da177e4 | 1701 | } |
800d8c63 | 1702 | if (IS_ERR(page)) { |
779750d2 | 1703 | int retry = 5; |
800d8c63 KS |
1704 | error = PTR_ERR(page); |
1705 | page = NULL; | |
779750d2 KS |
1706 | if (error != -ENOSPC) |
1707 | goto failed; | |
1708 | /* | |
1709 | * Try to reclaim some spece by splitting a huge page | |
1710 | * beyond i_size on the filesystem. | |
1711 | */ | |
1712 | while (retry--) { | |
1713 | int ret; | |
1714 | ret = shmem_unused_huge_shrink(sbinfo, NULL, 1); | |
1715 | if (ret == SHRINK_STOP) | |
1716 | break; | |
1717 | if (ret) | |
1718 | goto alloc_nohuge; | |
1719 | } | |
800d8c63 KS |
1720 | goto failed; |
1721 | } | |
1722 | ||
1723 | if (PageTransHuge(page)) | |
1724 | hindex = round_down(index, HPAGE_PMD_NR); | |
1725 | else | |
1726 | hindex = index; | |
1727 | ||
66d2f4d2 | 1728 | if (sgp == SGP_WRITE) |
eb39d618 | 1729 | __SetPageReferenced(page); |
66d2f4d2 | 1730 | |
9e18eb29 | 1731 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
800d8c63 | 1732 | PageTransHuge(page)); |
54af6042 | 1733 | if (error) |
800d8c63 KS |
1734 | goto unacct; |
1735 | error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK, | |
1736 | compound_order(page)); | |
b065b432 | 1737 | if (!error) { |
800d8c63 | 1738 | error = shmem_add_to_page_cache(page, mapping, hindex, |
fed400a1 | 1739 | NULL); |
b065b432 HD |
1740 | radix_tree_preload_end(); |
1741 | } | |
1742 | if (error) { | |
800d8c63 KS |
1743 | mem_cgroup_cancel_charge(page, memcg, |
1744 | PageTransHuge(page)); | |
1745 | goto unacct; | |
b065b432 | 1746 | } |
800d8c63 KS |
1747 | mem_cgroup_commit_charge(page, memcg, false, |
1748 | PageTransHuge(page)); | |
54af6042 HD |
1749 | lru_cache_add_anon(page); |
1750 | ||
4595ef88 | 1751 | spin_lock_irq(&info->lock); |
800d8c63 KS |
1752 | info->alloced += 1 << compound_order(page); |
1753 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); | |
54af6042 | 1754 | shmem_recalc_inode(inode); |
4595ef88 | 1755 | spin_unlock_irq(&info->lock); |
1635f6a7 | 1756 | alloced = true; |
54af6042 | 1757 | |
779750d2 KS |
1758 | if (PageTransHuge(page) && |
1759 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < | |
1760 | hindex + HPAGE_PMD_NR - 1) { | |
1761 | /* | |
1762 | * Part of the huge page is beyond i_size: subject | |
1763 | * to shrink under memory pressure. | |
1764 | */ | |
1765 | spin_lock(&sbinfo->shrinklist_lock); | |
1766 | if (list_empty(&info->shrinklist)) { | |
1767 | list_add_tail(&info->shrinklist, | |
1768 | &sbinfo->shrinklist); | |
1769 | sbinfo->shrinklist_len++; | |
1770 | } | |
1771 | spin_unlock(&sbinfo->shrinklist_lock); | |
1772 | } | |
1773 | ||
ec9516fb | 1774 | /* |
1635f6a7 HD |
1775 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1776 | */ | |
1777 | if (sgp == SGP_FALLOC) | |
1778 | sgp = SGP_WRITE; | |
1779 | clear: | |
1780 | /* | |
1781 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1782 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1783 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb | 1784 | */ |
800d8c63 KS |
1785 | if (sgp != SGP_WRITE && !PageUptodate(page)) { |
1786 | struct page *head = compound_head(page); | |
1787 | int i; | |
1788 | ||
1789 | for (i = 0; i < (1 << compound_order(head)); i++) { | |
1790 | clear_highpage(head + i); | |
1791 | flush_dcache_page(head + i); | |
1792 | } | |
1793 | SetPageUptodate(head); | |
ec9516fb | 1794 | } |
1da177e4 | 1795 | } |
bde05d1c | 1796 | |
54af6042 | 1797 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1798 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1799 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1800 | if (alloced) { |
1801 | ClearPageDirty(page); | |
1802 | delete_from_page_cache(page); | |
4595ef88 | 1803 | spin_lock_irq(&info->lock); |
267a4c76 | 1804 | shmem_recalc_inode(inode); |
4595ef88 | 1805 | spin_unlock_irq(&info->lock); |
267a4c76 | 1806 | } |
54af6042 | 1807 | error = -EINVAL; |
267a4c76 | 1808 | goto unlock; |
e83c32e8 | 1809 | } |
800d8c63 | 1810 | *pagep = page + index - hindex; |
54af6042 | 1811 | return 0; |
1da177e4 | 1812 | |
59a16ead | 1813 | /* |
54af6042 | 1814 | * Error recovery. |
59a16ead | 1815 | */ |
54af6042 | 1816 | unacct: |
800d8c63 KS |
1817 | if (sbinfo->max_blocks) |
1818 | percpu_counter_sub(&sbinfo->used_blocks, | |
1819 | 1 << compound_order(page)); | |
1820 | shmem_unacct_blocks(info->flags, 1 << compound_order(page)); | |
1821 | ||
1822 | if (PageTransHuge(page)) { | |
1823 | unlock_page(page); | |
1824 | put_page(page); | |
1825 | goto alloc_nohuge; | |
1826 | } | |
54af6042 | 1827 | failed: |
267a4c76 | 1828 | if (swap.val && !shmem_confirm_swap(mapping, index, swap)) |
d1899228 HD |
1829 | error = -EEXIST; |
1830 | unlock: | |
27ab7006 | 1831 | if (page) { |
54af6042 | 1832 | unlock_page(page); |
09cbfeaf | 1833 | put_page(page); |
54af6042 HD |
1834 | } |
1835 | if (error == -ENOSPC && !once++) { | |
1836 | info = SHMEM_I(inode); | |
4595ef88 | 1837 | spin_lock_irq(&info->lock); |
54af6042 | 1838 | shmem_recalc_inode(inode); |
4595ef88 | 1839 | spin_unlock_irq(&info->lock); |
27ab7006 | 1840 | goto repeat; |
ff36b801 | 1841 | } |
d1899228 | 1842 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1843 | goto repeat; |
1844 | return error; | |
1da177e4 LT |
1845 | } |
1846 | ||
d0217ac0 | 1847 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1848 | { |
496ad9aa | 1849 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 1850 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
657e3038 | 1851 | enum sgp_type sgp; |
1da177e4 | 1852 | int error; |
68da9f05 | 1853 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1854 | |
f00cdc6d HD |
1855 | /* |
1856 | * Trinity finds that probing a hole which tmpfs is punching can | |
1857 | * prevent the hole-punch from ever completing: which in turn | |
1858 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1859 | * faulting pages into the hole while it's being punched. Although |
1860 | * shmem_undo_range() does remove the additions, it may be unable to | |
1861 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1862 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1863 | * | |
1864 | * It does not matter if we sometimes reach this check just before the | |
1865 | * hole-punch begins, so that one fault then races with the punch: | |
1866 | * we just need to make racing faults a rare case. | |
1867 | * | |
1868 | * The implementation below would be much simpler if we just used a | |
1869 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1870 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1871 | */ |
1872 | if (unlikely(inode->i_private)) { | |
1873 | struct shmem_falloc *shmem_falloc; | |
1874 | ||
1875 | spin_lock(&inode->i_lock); | |
1876 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1877 | if (shmem_falloc && |
1878 | shmem_falloc->waitq && | |
1879 | vmf->pgoff >= shmem_falloc->start && | |
1880 | vmf->pgoff < shmem_falloc->next) { | |
1881 | wait_queue_head_t *shmem_falloc_waitq; | |
1882 | DEFINE_WAIT(shmem_fault_wait); | |
1883 | ||
1884 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1885 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1886 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1887 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1888 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1889 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1890 | } |
8e205f77 HD |
1891 | |
1892 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1893 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1894 | TASK_UNINTERRUPTIBLE); | |
1895 | spin_unlock(&inode->i_lock); | |
1896 | schedule(); | |
1897 | ||
1898 | /* | |
1899 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1900 | * stack of the hole-punching task: shmem_falloc_waitq | |
1901 | * is usually invalid by the time we reach here, but | |
1902 | * finish_wait() does not dereference it in that case; | |
1903 | * though i_lock needed lest racing with wake_up_all(). | |
1904 | */ | |
1905 | spin_lock(&inode->i_lock); | |
1906 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1907 | spin_unlock(&inode->i_lock); | |
1908 | return ret; | |
f00cdc6d | 1909 | } |
8e205f77 | 1910 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1911 | } |
1912 | ||
657e3038 KS |
1913 | sgp = SGP_CACHE; |
1914 | if (vma->vm_flags & VM_HUGEPAGE) | |
1915 | sgp = SGP_HUGE; | |
1916 | else if (vma->vm_flags & VM_NOHUGEPAGE) | |
1917 | sgp = SGP_NOHUGE; | |
1918 | ||
1919 | error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp, | |
9e18eb29 | 1920 | gfp, vma->vm_mm, &ret); |
d0217ac0 NP |
1921 | if (error) |
1922 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1923 | return ret; |
1da177e4 LT |
1924 | } |
1925 | ||
c01d5b30 HD |
1926 | unsigned long shmem_get_unmapped_area(struct file *file, |
1927 | unsigned long uaddr, unsigned long len, | |
1928 | unsigned long pgoff, unsigned long flags) | |
1929 | { | |
1930 | unsigned long (*get_area)(struct file *, | |
1931 | unsigned long, unsigned long, unsigned long, unsigned long); | |
1932 | unsigned long addr; | |
1933 | unsigned long offset; | |
1934 | unsigned long inflated_len; | |
1935 | unsigned long inflated_addr; | |
1936 | unsigned long inflated_offset; | |
1937 | ||
1938 | if (len > TASK_SIZE) | |
1939 | return -ENOMEM; | |
1940 | ||
1941 | get_area = current->mm->get_unmapped_area; | |
1942 | addr = get_area(file, uaddr, len, pgoff, flags); | |
1943 | ||
e496cf3d | 1944 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) |
c01d5b30 HD |
1945 | return addr; |
1946 | if (IS_ERR_VALUE(addr)) | |
1947 | return addr; | |
1948 | if (addr & ~PAGE_MASK) | |
1949 | return addr; | |
1950 | if (addr > TASK_SIZE - len) | |
1951 | return addr; | |
1952 | ||
1953 | if (shmem_huge == SHMEM_HUGE_DENY) | |
1954 | return addr; | |
1955 | if (len < HPAGE_PMD_SIZE) | |
1956 | return addr; | |
1957 | if (flags & MAP_FIXED) | |
1958 | return addr; | |
1959 | /* | |
1960 | * Our priority is to support MAP_SHARED mapped hugely; | |
1961 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
1962 | * But if caller specified an address hint, respect that as before. | |
1963 | */ | |
1964 | if (uaddr) | |
1965 | return addr; | |
1966 | ||
1967 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
1968 | struct super_block *sb; | |
1969 | ||
1970 | if (file) { | |
1971 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
1972 | sb = file_inode(file)->i_sb; | |
1973 | } else { | |
1974 | /* | |
1975 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
1976 | * for "/dev/zero", to create a shared anonymous object. | |
1977 | */ | |
1978 | if (IS_ERR(shm_mnt)) | |
1979 | return addr; | |
1980 | sb = shm_mnt->mnt_sb; | |
1981 | } | |
1982 | if (SHMEM_SB(sb)->huge != SHMEM_HUGE_NEVER) | |
1983 | return addr; | |
1984 | } | |
1985 | ||
1986 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
1987 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
1988 | return addr; | |
1989 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
1990 | return addr; | |
1991 | ||
1992 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
1993 | if (inflated_len > TASK_SIZE) | |
1994 | return addr; | |
1995 | if (inflated_len < len) | |
1996 | return addr; | |
1997 | ||
1998 | inflated_addr = get_area(NULL, 0, inflated_len, 0, flags); | |
1999 | if (IS_ERR_VALUE(inflated_addr)) | |
2000 | return addr; | |
2001 | if (inflated_addr & ~PAGE_MASK) | |
2002 | return addr; | |
2003 | ||
2004 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2005 | inflated_addr += offset - inflated_offset; | |
2006 | if (inflated_offset > offset) | |
2007 | inflated_addr += HPAGE_PMD_SIZE; | |
2008 | ||
2009 | if (inflated_addr > TASK_SIZE - len) | |
2010 | return addr; | |
2011 | return inflated_addr; | |
2012 | } | |
2013 | ||
1da177e4 | 2014 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2015 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2016 | { |
496ad9aa | 2017 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2018 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2019 | } |
2020 | ||
d8dc74f2 AB |
2021 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
2022 | unsigned long addr) | |
1da177e4 | 2023 | { |
496ad9aa | 2024 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2025 | pgoff_t index; |
1da177e4 | 2026 | |
41ffe5d5 HD |
2027 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2028 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
2029 | } |
2030 | #endif | |
2031 | ||
2032 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
2033 | { | |
496ad9aa | 2034 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2035 | struct shmem_inode_info *info = SHMEM_I(inode); |
2036 | int retval = -ENOMEM; | |
2037 | ||
4595ef88 | 2038 | spin_lock_irq(&info->lock); |
1da177e4 LT |
2039 | if (lock && !(info->flags & VM_LOCKED)) { |
2040 | if (!user_shm_lock(inode->i_size, user)) | |
2041 | goto out_nomem; | |
2042 | info->flags |= VM_LOCKED; | |
89e004ea | 2043 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
2044 | } |
2045 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
2046 | user_shm_unlock(inode->i_size, user); | |
2047 | info->flags &= ~VM_LOCKED; | |
89e004ea | 2048 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2049 | } |
2050 | retval = 0; | |
89e004ea | 2051 | |
1da177e4 | 2052 | out_nomem: |
4595ef88 | 2053 | spin_unlock_irq(&info->lock); |
1da177e4 LT |
2054 | return retval; |
2055 | } | |
2056 | ||
9b83a6a8 | 2057 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
2058 | { |
2059 | file_accessed(file); | |
2060 | vma->vm_ops = &shmem_vm_ops; | |
e496cf3d | 2061 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
2062 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
2063 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
2064 | khugepaged_enter(vma, vma->vm_flags); | |
2065 | } | |
1da177e4 LT |
2066 | return 0; |
2067 | } | |
2068 | ||
454abafe | 2069 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 2070 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
2071 | { |
2072 | struct inode *inode; | |
2073 | struct shmem_inode_info *info; | |
2074 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
2075 | ||
5b04c689 PE |
2076 | if (shmem_reserve_inode(sb)) |
2077 | return NULL; | |
1da177e4 LT |
2078 | |
2079 | inode = new_inode(sb); | |
2080 | if (inode) { | |
85fe4025 | 2081 | inode->i_ino = get_next_ino(); |
454abafe | 2082 | inode_init_owner(inode, dir, mode); |
1da177e4 | 2083 | inode->i_blocks = 0; |
1da177e4 | 2084 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
91828a40 | 2085 | inode->i_generation = get_seconds(); |
1da177e4 LT |
2086 | info = SHMEM_I(inode); |
2087 | memset(info, 0, (char *)inode - (char *)info); | |
2088 | spin_lock_init(&info->lock); | |
40e041a2 | 2089 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 2090 | info->flags = flags & VM_NORESERVE; |
779750d2 | 2091 | INIT_LIST_HEAD(&info->shrinklist); |
1da177e4 | 2092 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 2093 | simple_xattrs_init(&info->xattrs); |
72c04902 | 2094 | cache_no_acl(inode); |
1da177e4 LT |
2095 | |
2096 | switch (mode & S_IFMT) { | |
2097 | default: | |
39f0247d | 2098 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
2099 | init_special_inode(inode, mode, dev); |
2100 | break; | |
2101 | case S_IFREG: | |
14fcc23f | 2102 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
2103 | inode->i_op = &shmem_inode_operations; |
2104 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
2105 | mpol_shared_policy_init(&info->policy, |
2106 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
2107 | break; |
2108 | case S_IFDIR: | |
d8c76e6f | 2109 | inc_nlink(inode); |
1da177e4 LT |
2110 | /* Some things misbehave if size == 0 on a directory */ |
2111 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2112 | inode->i_op = &shmem_dir_inode_operations; | |
2113 | inode->i_fop = &simple_dir_operations; | |
2114 | break; | |
2115 | case S_IFLNK: | |
2116 | /* | |
2117 | * Must not load anything in the rbtree, | |
2118 | * mpol_free_shared_policy will not be called. | |
2119 | */ | |
71fe804b | 2120 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
2121 | break; |
2122 | } | |
5b04c689 PE |
2123 | } else |
2124 | shmem_free_inode(sb); | |
1da177e4 LT |
2125 | return inode; |
2126 | } | |
2127 | ||
0cd6144a JW |
2128 | bool shmem_mapping(struct address_space *mapping) |
2129 | { | |
f0774d88 SL |
2130 | if (!mapping->host) |
2131 | return false; | |
2132 | ||
97b713ba | 2133 | return mapping->host->i_sb->s_op == &shmem_ops; |
0cd6144a JW |
2134 | } |
2135 | ||
1da177e4 | 2136 | #ifdef CONFIG_TMPFS |
92e1d5be | 2137 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2138 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2139 | |
6d9d88d0 JS |
2140 | #ifdef CONFIG_TMPFS_XATTR |
2141 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2142 | #else | |
2143 | #define shmem_initxattrs NULL | |
2144 | #endif | |
2145 | ||
1da177e4 | 2146 | static int |
800d15a5 NP |
2147 | shmem_write_begin(struct file *file, struct address_space *mapping, |
2148 | loff_t pos, unsigned len, unsigned flags, | |
2149 | struct page **pagep, void **fsdata) | |
1da177e4 | 2150 | { |
800d15a5 | 2151 | struct inode *inode = mapping->host; |
40e041a2 | 2152 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2153 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
2154 | |
2155 | /* i_mutex is held by caller */ | |
2156 | if (unlikely(info->seals)) { | |
2157 | if (info->seals & F_SEAL_WRITE) | |
2158 | return -EPERM; | |
2159 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2160 | return -EPERM; | |
2161 | } | |
2162 | ||
9e18eb29 | 2163 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
2164 | } |
2165 | ||
2166 | static int | |
2167 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2168 | loff_t pos, unsigned len, unsigned copied, | |
2169 | struct page *page, void *fsdata) | |
2170 | { | |
2171 | struct inode *inode = mapping->host; | |
2172 | ||
d3602444 HD |
2173 | if (pos + copied > inode->i_size) |
2174 | i_size_write(inode, pos + copied); | |
2175 | ||
ec9516fb | 2176 | if (!PageUptodate(page)) { |
800d8c63 KS |
2177 | struct page *head = compound_head(page); |
2178 | if (PageTransCompound(page)) { | |
2179 | int i; | |
2180 | ||
2181 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2182 | if (head + i == page) | |
2183 | continue; | |
2184 | clear_highpage(head + i); | |
2185 | flush_dcache_page(head + i); | |
2186 | } | |
2187 | } | |
09cbfeaf KS |
2188 | if (copied < PAGE_SIZE) { |
2189 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2190 | zero_user_segments(page, 0, from, |
09cbfeaf | 2191 | from + copied, PAGE_SIZE); |
ec9516fb | 2192 | } |
800d8c63 | 2193 | SetPageUptodate(head); |
ec9516fb | 2194 | } |
800d15a5 | 2195 | set_page_dirty(page); |
6746aff7 | 2196 | unlock_page(page); |
09cbfeaf | 2197 | put_page(page); |
800d15a5 | 2198 | |
800d15a5 | 2199 | return copied; |
1da177e4 LT |
2200 | } |
2201 | ||
2ba5bbed | 2202 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2203 | { |
6e58e79d AV |
2204 | struct file *file = iocb->ki_filp; |
2205 | struct inode *inode = file_inode(file); | |
1da177e4 | 2206 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2207 | pgoff_t index; |
2208 | unsigned long offset; | |
a0ee5ec5 | 2209 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2210 | int error = 0; |
cb66a7a1 | 2211 | ssize_t retval = 0; |
6e58e79d | 2212 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2213 | |
2214 | /* | |
2215 | * Might this read be for a stacking filesystem? Then when reading | |
2216 | * holes of a sparse file, we actually need to allocate those pages, | |
2217 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2218 | */ | |
777eda2c | 2219 | if (!iter_is_iovec(to)) |
75edd345 | 2220 | sgp = SGP_CACHE; |
1da177e4 | 2221 | |
09cbfeaf KS |
2222 | index = *ppos >> PAGE_SHIFT; |
2223 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2224 | |
2225 | for (;;) { | |
2226 | struct page *page = NULL; | |
41ffe5d5 HD |
2227 | pgoff_t end_index; |
2228 | unsigned long nr, ret; | |
1da177e4 LT |
2229 | loff_t i_size = i_size_read(inode); |
2230 | ||
09cbfeaf | 2231 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2232 | if (index > end_index) |
2233 | break; | |
2234 | if (index == end_index) { | |
09cbfeaf | 2235 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2236 | if (nr <= offset) |
2237 | break; | |
2238 | } | |
2239 | ||
9e18eb29 | 2240 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2241 | if (error) { |
2242 | if (error == -EINVAL) | |
2243 | error = 0; | |
1da177e4 LT |
2244 | break; |
2245 | } | |
75edd345 HD |
2246 | if (page) { |
2247 | if (sgp == SGP_CACHE) | |
2248 | set_page_dirty(page); | |
d3602444 | 2249 | unlock_page(page); |
75edd345 | 2250 | } |
1da177e4 LT |
2251 | |
2252 | /* | |
2253 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2254 | * are called without i_mutex protection against truncate |
1da177e4 | 2255 | */ |
09cbfeaf | 2256 | nr = PAGE_SIZE; |
1da177e4 | 2257 | i_size = i_size_read(inode); |
09cbfeaf | 2258 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2259 | if (index == end_index) { |
09cbfeaf | 2260 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2261 | if (nr <= offset) { |
2262 | if (page) | |
09cbfeaf | 2263 | put_page(page); |
1da177e4 LT |
2264 | break; |
2265 | } | |
2266 | } | |
2267 | nr -= offset; | |
2268 | ||
2269 | if (page) { | |
2270 | /* | |
2271 | * If users can be writing to this page using arbitrary | |
2272 | * virtual addresses, take care about potential aliasing | |
2273 | * before reading the page on the kernel side. | |
2274 | */ | |
2275 | if (mapping_writably_mapped(mapping)) | |
2276 | flush_dcache_page(page); | |
2277 | /* | |
2278 | * Mark the page accessed if we read the beginning. | |
2279 | */ | |
2280 | if (!offset) | |
2281 | mark_page_accessed(page); | |
b5810039 | 2282 | } else { |
1da177e4 | 2283 | page = ZERO_PAGE(0); |
09cbfeaf | 2284 | get_page(page); |
b5810039 | 2285 | } |
1da177e4 LT |
2286 | |
2287 | /* | |
2288 | * Ok, we have the page, and it's up-to-date, so | |
2289 | * now we can copy it to user space... | |
1da177e4 | 2290 | */ |
2ba5bbed | 2291 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2292 | retval += ret; |
1da177e4 | 2293 | offset += ret; |
09cbfeaf KS |
2294 | index += offset >> PAGE_SHIFT; |
2295 | offset &= ~PAGE_MASK; | |
1da177e4 | 2296 | |
09cbfeaf | 2297 | put_page(page); |
2ba5bbed | 2298 | if (!iov_iter_count(to)) |
1da177e4 | 2299 | break; |
6e58e79d AV |
2300 | if (ret < nr) { |
2301 | error = -EFAULT; | |
2302 | break; | |
2303 | } | |
1da177e4 LT |
2304 | cond_resched(); |
2305 | } | |
2306 | ||
09cbfeaf | 2307 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2308 | file_accessed(file); |
2309 | return retval ? retval : error; | |
1da177e4 LT |
2310 | } |
2311 | ||
708e3508 HD |
2312 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, |
2313 | struct pipe_inode_info *pipe, size_t len, | |
2314 | unsigned int flags) | |
2315 | { | |
2316 | struct address_space *mapping = in->f_mapping; | |
71f0e07a | 2317 | struct inode *inode = mapping->host; |
708e3508 HD |
2318 | unsigned int loff, nr_pages, req_pages; |
2319 | struct page *pages[PIPE_DEF_BUFFERS]; | |
2320 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
2321 | struct page *page; | |
2322 | pgoff_t index, end_index; | |
2323 | loff_t isize, left; | |
2324 | int error, page_nr; | |
2325 | struct splice_pipe_desc spd = { | |
2326 | .pages = pages, | |
2327 | .partial = partial, | |
047fe360 | 2328 | .nr_pages_max = PIPE_DEF_BUFFERS, |
708e3508 HD |
2329 | .flags = flags, |
2330 | .ops = &page_cache_pipe_buf_ops, | |
2331 | .spd_release = spd_release_page, | |
2332 | }; | |
2333 | ||
71f0e07a | 2334 | isize = i_size_read(inode); |
708e3508 HD |
2335 | if (unlikely(*ppos >= isize)) |
2336 | return 0; | |
2337 | ||
2338 | left = isize - *ppos; | |
2339 | if (unlikely(left < len)) | |
2340 | len = left; | |
2341 | ||
2342 | if (splice_grow_spd(pipe, &spd)) | |
2343 | return -ENOMEM; | |
2344 | ||
09cbfeaf KS |
2345 | index = *ppos >> PAGE_SHIFT; |
2346 | loff = *ppos & ~PAGE_MASK; | |
2347 | req_pages = (len + loff + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
a786c06d | 2348 | nr_pages = min(req_pages, spd.nr_pages_max); |
708e3508 | 2349 | |
708e3508 HD |
2350 | spd.nr_pages = find_get_pages_contig(mapping, index, |
2351 | nr_pages, spd.pages); | |
2352 | index += spd.nr_pages; | |
708e3508 | 2353 | error = 0; |
708e3508 | 2354 | |
71f0e07a | 2355 | while (spd.nr_pages < nr_pages) { |
9e18eb29 | 2356 | error = shmem_getpage(inode, index, &page, SGP_CACHE); |
71f0e07a HD |
2357 | if (error) |
2358 | break; | |
2359 | unlock_page(page); | |
708e3508 HD |
2360 | spd.pages[spd.nr_pages++] = page; |
2361 | index++; | |
2362 | } | |
2363 | ||
09cbfeaf | 2364 | index = *ppos >> PAGE_SHIFT; |
708e3508 HD |
2365 | nr_pages = spd.nr_pages; |
2366 | spd.nr_pages = 0; | |
71f0e07a | 2367 | |
708e3508 HD |
2368 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { |
2369 | unsigned int this_len; | |
2370 | ||
2371 | if (!len) | |
2372 | break; | |
2373 | ||
09cbfeaf | 2374 | this_len = min_t(unsigned long, len, PAGE_SIZE - loff); |
708e3508 HD |
2375 | page = spd.pages[page_nr]; |
2376 | ||
71f0e07a | 2377 | if (!PageUptodate(page) || page->mapping != mapping) { |
9e18eb29 | 2378 | error = shmem_getpage(inode, index, &page, SGP_CACHE); |
71f0e07a | 2379 | if (error) |
708e3508 | 2380 | break; |
71f0e07a | 2381 | unlock_page(page); |
09cbfeaf | 2382 | put_page(spd.pages[page_nr]); |
71f0e07a | 2383 | spd.pages[page_nr] = page; |
708e3508 | 2384 | } |
71f0e07a HD |
2385 | |
2386 | isize = i_size_read(inode); | |
09cbfeaf | 2387 | end_index = (isize - 1) >> PAGE_SHIFT; |
708e3508 HD |
2388 | if (unlikely(!isize || index > end_index)) |
2389 | break; | |
2390 | ||
708e3508 HD |
2391 | if (end_index == index) { |
2392 | unsigned int plen; | |
2393 | ||
09cbfeaf | 2394 | plen = ((isize - 1) & ~PAGE_MASK) + 1; |
708e3508 HD |
2395 | if (plen <= loff) |
2396 | break; | |
2397 | ||
708e3508 HD |
2398 | this_len = min(this_len, plen - loff); |
2399 | len = this_len; | |
2400 | } | |
2401 | ||
2402 | spd.partial[page_nr].offset = loff; | |
2403 | spd.partial[page_nr].len = this_len; | |
2404 | len -= this_len; | |
2405 | loff = 0; | |
2406 | spd.nr_pages++; | |
2407 | index++; | |
2408 | } | |
2409 | ||
708e3508 | 2410 | while (page_nr < nr_pages) |
09cbfeaf | 2411 | put_page(spd.pages[page_nr++]); |
708e3508 HD |
2412 | |
2413 | if (spd.nr_pages) | |
2414 | error = splice_to_pipe(pipe, &spd); | |
2415 | ||
047fe360 | 2416 | splice_shrink_spd(&spd); |
708e3508 HD |
2417 | |
2418 | if (error > 0) { | |
2419 | *ppos += error; | |
2420 | file_accessed(in); | |
2421 | } | |
2422 | return error; | |
2423 | } | |
2424 | ||
220f2ac9 HD |
2425 | /* |
2426 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
2427 | */ | |
2428 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2429 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2430 | { |
2431 | struct page *page; | |
2432 | struct pagevec pvec; | |
2433 | pgoff_t indices[PAGEVEC_SIZE]; | |
2434 | bool done = false; | |
2435 | int i; | |
2436 | ||
2437 | pagevec_init(&pvec, 0); | |
2438 | pvec.nr = 1; /* start small: we may be there already */ | |
2439 | while (!done) { | |
0cd6144a | 2440 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2441 | pvec.nr, pvec.pages, indices); |
2442 | if (!pvec.nr) { | |
965c8e59 | 2443 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2444 | index = end; |
2445 | break; | |
2446 | } | |
2447 | for (i = 0; i < pvec.nr; i++, index++) { | |
2448 | if (index < indices[i]) { | |
965c8e59 | 2449 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2450 | done = true; |
2451 | break; | |
2452 | } | |
2453 | index = indices[i]; | |
2454 | } | |
2455 | page = pvec.pages[i]; | |
2456 | if (page && !radix_tree_exceptional_entry(page)) { | |
2457 | if (!PageUptodate(page)) | |
2458 | page = NULL; | |
2459 | } | |
2460 | if (index >= end || | |
965c8e59 AM |
2461 | (page && whence == SEEK_DATA) || |
2462 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2463 | done = true; |
2464 | break; | |
2465 | } | |
2466 | } | |
0cd6144a | 2467 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2468 | pagevec_release(&pvec); |
2469 | pvec.nr = PAGEVEC_SIZE; | |
2470 | cond_resched(); | |
2471 | } | |
2472 | return index; | |
2473 | } | |
2474 | ||
965c8e59 | 2475 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2476 | { |
2477 | struct address_space *mapping = file->f_mapping; | |
2478 | struct inode *inode = mapping->host; | |
2479 | pgoff_t start, end; | |
2480 | loff_t new_offset; | |
2481 | ||
965c8e59 AM |
2482 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2483 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2484 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2485 | inode_lock(inode); |
220f2ac9 HD |
2486 | /* We're holding i_mutex so we can access i_size directly */ |
2487 | ||
2488 | if (offset < 0) | |
2489 | offset = -EINVAL; | |
2490 | else if (offset >= inode->i_size) | |
2491 | offset = -ENXIO; | |
2492 | else { | |
09cbfeaf KS |
2493 | start = offset >> PAGE_SHIFT; |
2494 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2495 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2496 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2497 | if (new_offset > offset) { |
2498 | if (new_offset < inode->i_size) | |
2499 | offset = new_offset; | |
965c8e59 | 2500 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2501 | offset = -ENXIO; |
2502 | else | |
2503 | offset = inode->i_size; | |
2504 | } | |
2505 | } | |
2506 | ||
387aae6f HD |
2507 | if (offset >= 0) |
2508 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2509 | inode_unlock(inode); |
220f2ac9 HD |
2510 | return offset; |
2511 | } | |
2512 | ||
05f65b5c DH |
2513 | /* |
2514 | * We need a tag: a new tag would expand every radix_tree_node by 8 bytes, | |
2515 | * so reuse a tag which we firmly believe is never set or cleared on shmem. | |
2516 | */ | |
2517 | #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE | |
2518 | #define LAST_SCAN 4 /* about 150ms max */ | |
2519 | ||
2520 | static void shmem_tag_pins(struct address_space *mapping) | |
2521 | { | |
2522 | struct radix_tree_iter iter; | |
2523 | void **slot; | |
2524 | pgoff_t start; | |
2525 | struct page *page; | |
2526 | ||
2527 | lru_add_drain(); | |
2528 | start = 0; | |
2529 | rcu_read_lock(); | |
2530 | ||
05f65b5c DH |
2531 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
2532 | page = radix_tree_deref_slot(slot); | |
2533 | if (!page || radix_tree_exception(page)) { | |
2cf938aa MW |
2534 | if (radix_tree_deref_retry(page)) { |
2535 | slot = radix_tree_iter_retry(&iter); | |
2536 | continue; | |
2537 | } | |
05f65b5c DH |
2538 | } else if (page_count(page) - page_mapcount(page) > 1) { |
2539 | spin_lock_irq(&mapping->tree_lock); | |
2540 | radix_tree_tag_set(&mapping->page_tree, iter.index, | |
2541 | SHMEM_TAG_PINNED); | |
2542 | spin_unlock_irq(&mapping->tree_lock); | |
2543 | } | |
2544 | ||
2545 | if (need_resched()) { | |
2546 | cond_resched_rcu(); | |
7165092f | 2547 | slot = radix_tree_iter_next(&iter); |
05f65b5c DH |
2548 | } |
2549 | } | |
2550 | rcu_read_unlock(); | |
2551 | } | |
2552 | ||
2553 | /* | |
2554 | * Setting SEAL_WRITE requires us to verify there's no pending writer. However, | |
2555 | * via get_user_pages(), drivers might have some pending I/O without any active | |
2556 | * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages | |
2557 | * and see whether it has an elevated ref-count. If so, we tag them and wait for | |
2558 | * them to be dropped. | |
2559 | * The caller must guarantee that no new user will acquire writable references | |
2560 | * to those pages to avoid races. | |
2561 | */ | |
40e041a2 DH |
2562 | static int shmem_wait_for_pins(struct address_space *mapping) |
2563 | { | |
05f65b5c DH |
2564 | struct radix_tree_iter iter; |
2565 | void **slot; | |
2566 | pgoff_t start; | |
2567 | struct page *page; | |
2568 | int error, scan; | |
2569 | ||
2570 | shmem_tag_pins(mapping); | |
2571 | ||
2572 | error = 0; | |
2573 | for (scan = 0; scan <= LAST_SCAN; scan++) { | |
2574 | if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED)) | |
2575 | break; | |
2576 | ||
2577 | if (!scan) | |
2578 | lru_add_drain_all(); | |
2579 | else if (schedule_timeout_killable((HZ << scan) / 200)) | |
2580 | scan = LAST_SCAN; | |
2581 | ||
2582 | start = 0; | |
2583 | rcu_read_lock(); | |
05f65b5c DH |
2584 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, |
2585 | start, SHMEM_TAG_PINNED) { | |
2586 | ||
2587 | page = radix_tree_deref_slot(slot); | |
2588 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2589 | if (radix_tree_deref_retry(page)) { |
2590 | slot = radix_tree_iter_retry(&iter); | |
2591 | continue; | |
2592 | } | |
05f65b5c DH |
2593 | |
2594 | page = NULL; | |
2595 | } | |
2596 | ||
2597 | if (page && | |
2598 | page_count(page) - page_mapcount(page) != 1) { | |
2599 | if (scan < LAST_SCAN) | |
2600 | goto continue_resched; | |
2601 | ||
2602 | /* | |
2603 | * On the last scan, we clean up all those tags | |
2604 | * we inserted; but make a note that we still | |
2605 | * found pages pinned. | |
2606 | */ | |
2607 | error = -EBUSY; | |
2608 | } | |
2609 | ||
2610 | spin_lock_irq(&mapping->tree_lock); | |
2611 | radix_tree_tag_clear(&mapping->page_tree, | |
2612 | iter.index, SHMEM_TAG_PINNED); | |
2613 | spin_unlock_irq(&mapping->tree_lock); | |
2614 | continue_resched: | |
2615 | if (need_resched()) { | |
2616 | cond_resched_rcu(); | |
7165092f | 2617 | slot = radix_tree_iter_next(&iter); |
05f65b5c DH |
2618 | } |
2619 | } | |
2620 | rcu_read_unlock(); | |
2621 | } | |
2622 | ||
2623 | return error; | |
40e041a2 DH |
2624 | } |
2625 | ||
2626 | #define F_ALL_SEALS (F_SEAL_SEAL | \ | |
2627 | F_SEAL_SHRINK | \ | |
2628 | F_SEAL_GROW | \ | |
2629 | F_SEAL_WRITE) | |
2630 | ||
2631 | int shmem_add_seals(struct file *file, unsigned int seals) | |
2632 | { | |
2633 | struct inode *inode = file_inode(file); | |
2634 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2635 | int error; | |
2636 | ||
2637 | /* | |
2638 | * SEALING | |
2639 | * Sealing allows multiple parties to share a shmem-file but restrict | |
2640 | * access to a specific subset of file operations. Seals can only be | |
2641 | * added, but never removed. This way, mutually untrusted parties can | |
2642 | * share common memory regions with a well-defined policy. A malicious | |
2643 | * peer can thus never perform unwanted operations on a shared object. | |
2644 | * | |
2645 | * Seals are only supported on special shmem-files and always affect | |
2646 | * the whole underlying inode. Once a seal is set, it may prevent some | |
2647 | * kinds of access to the file. Currently, the following seals are | |
2648 | * defined: | |
2649 | * SEAL_SEAL: Prevent further seals from being set on this file | |
2650 | * SEAL_SHRINK: Prevent the file from shrinking | |
2651 | * SEAL_GROW: Prevent the file from growing | |
2652 | * SEAL_WRITE: Prevent write access to the file | |
2653 | * | |
2654 | * As we don't require any trust relationship between two parties, we | |
2655 | * must prevent seals from being removed. Therefore, sealing a file | |
2656 | * only adds a given set of seals to the file, it never touches | |
2657 | * existing seals. Furthermore, the "setting seals"-operation can be | |
2658 | * sealed itself, which basically prevents any further seal from being | |
2659 | * added. | |
2660 | * | |
2661 | * Semantics of sealing are only defined on volatile files. Only | |
2662 | * anonymous shmem files support sealing. More importantly, seals are | |
2663 | * never written to disk. Therefore, there's no plan to support it on | |
2664 | * other file types. | |
2665 | */ | |
2666 | ||
2667 | if (file->f_op != &shmem_file_operations) | |
2668 | return -EINVAL; | |
2669 | if (!(file->f_mode & FMODE_WRITE)) | |
2670 | return -EPERM; | |
2671 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
2672 | return -EINVAL; | |
2673 | ||
5955102c | 2674 | inode_lock(inode); |
40e041a2 DH |
2675 | |
2676 | if (info->seals & F_SEAL_SEAL) { | |
2677 | error = -EPERM; | |
2678 | goto unlock; | |
2679 | } | |
2680 | ||
2681 | if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) { | |
2682 | error = mapping_deny_writable(file->f_mapping); | |
2683 | if (error) | |
2684 | goto unlock; | |
2685 | ||
2686 | error = shmem_wait_for_pins(file->f_mapping); | |
2687 | if (error) { | |
2688 | mapping_allow_writable(file->f_mapping); | |
2689 | goto unlock; | |
2690 | } | |
2691 | } | |
2692 | ||
2693 | info->seals |= seals; | |
2694 | error = 0; | |
2695 | ||
2696 | unlock: | |
5955102c | 2697 | inode_unlock(inode); |
40e041a2 DH |
2698 | return error; |
2699 | } | |
2700 | EXPORT_SYMBOL_GPL(shmem_add_seals); | |
2701 | ||
2702 | int shmem_get_seals(struct file *file) | |
2703 | { | |
2704 | if (file->f_op != &shmem_file_operations) | |
2705 | return -EINVAL; | |
2706 | ||
2707 | return SHMEM_I(file_inode(file))->seals; | |
2708 | } | |
2709 | EXPORT_SYMBOL_GPL(shmem_get_seals); | |
2710 | ||
2711 | long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg) | |
2712 | { | |
2713 | long error; | |
2714 | ||
2715 | switch (cmd) { | |
2716 | case F_ADD_SEALS: | |
2717 | /* disallow upper 32bit */ | |
2718 | if (arg > UINT_MAX) | |
2719 | return -EINVAL; | |
2720 | ||
2721 | error = shmem_add_seals(file, arg); | |
2722 | break; | |
2723 | case F_GET_SEALS: | |
2724 | error = shmem_get_seals(file); | |
2725 | break; | |
2726 | default: | |
2727 | error = -EINVAL; | |
2728 | break; | |
2729 | } | |
2730 | ||
2731 | return error; | |
2732 | } | |
2733 | ||
83e4fa9c HD |
2734 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2735 | loff_t len) | |
2736 | { | |
496ad9aa | 2737 | struct inode *inode = file_inode(file); |
e2d12e22 | 2738 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2739 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2740 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2741 | pgoff_t start, index, end; |
2742 | int error; | |
83e4fa9c | 2743 | |
13ace4d0 HD |
2744 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2745 | return -EOPNOTSUPP; | |
2746 | ||
5955102c | 2747 | inode_lock(inode); |
83e4fa9c HD |
2748 | |
2749 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2750 | struct address_space *mapping = file->f_mapping; | |
2751 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2752 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2753 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2754 | |
40e041a2 DH |
2755 | /* protected by i_mutex */ |
2756 | if (info->seals & F_SEAL_WRITE) { | |
2757 | error = -EPERM; | |
2758 | goto out; | |
2759 | } | |
2760 | ||
8e205f77 | 2761 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
2762 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
2763 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
2764 | spin_lock(&inode->i_lock); | |
2765 | inode->i_private = &shmem_falloc; | |
2766 | spin_unlock(&inode->i_lock); | |
2767 | ||
83e4fa9c HD |
2768 | if ((u64)unmap_end > (u64)unmap_start) |
2769 | unmap_mapping_range(mapping, unmap_start, | |
2770 | 1 + unmap_end - unmap_start, 0); | |
2771 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2772 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2773 | |
2774 | spin_lock(&inode->i_lock); | |
2775 | inode->i_private = NULL; | |
2776 | wake_up_all(&shmem_falloc_waitq); | |
2777 | spin_unlock(&inode->i_lock); | |
83e4fa9c | 2778 | error = 0; |
8e205f77 | 2779 | goto out; |
e2d12e22 HD |
2780 | } |
2781 | ||
2782 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2783 | error = inode_newsize_ok(inode, offset + len); | |
2784 | if (error) | |
2785 | goto out; | |
2786 | ||
40e041a2 DH |
2787 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2788 | error = -EPERM; | |
2789 | goto out; | |
2790 | } | |
2791 | ||
09cbfeaf KS |
2792 | start = offset >> PAGE_SHIFT; |
2793 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2794 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2795 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2796 | error = -ENOSPC; | |
2797 | goto out; | |
83e4fa9c HD |
2798 | } |
2799 | ||
8e205f77 | 2800 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2801 | shmem_falloc.start = start; |
2802 | shmem_falloc.next = start; | |
2803 | shmem_falloc.nr_falloced = 0; | |
2804 | shmem_falloc.nr_unswapped = 0; | |
2805 | spin_lock(&inode->i_lock); | |
2806 | inode->i_private = &shmem_falloc; | |
2807 | spin_unlock(&inode->i_lock); | |
2808 | ||
e2d12e22 HD |
2809 | for (index = start; index < end; index++) { |
2810 | struct page *page; | |
2811 | ||
2812 | /* | |
2813 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2814 | * been interrupted because we are using up too much memory. | |
2815 | */ | |
2816 | if (signal_pending(current)) | |
2817 | error = -EINTR; | |
1aac1400 HD |
2818 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2819 | error = -ENOMEM; | |
e2d12e22 | 2820 | else |
9e18eb29 | 2821 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2822 | if (error) { |
1635f6a7 | 2823 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2824 | if (index > start) { |
2825 | shmem_undo_range(inode, | |
2826 | (loff_t)start << PAGE_SHIFT, | |
2827 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2828 | } | |
1aac1400 | 2829 | goto undone; |
e2d12e22 HD |
2830 | } |
2831 | ||
1aac1400 HD |
2832 | /* |
2833 | * Inform shmem_writepage() how far we have reached. | |
2834 | * No need for lock or barrier: we have the page lock. | |
2835 | */ | |
2836 | shmem_falloc.next++; | |
2837 | if (!PageUptodate(page)) | |
2838 | shmem_falloc.nr_falloced++; | |
2839 | ||
e2d12e22 | 2840 | /* |
1635f6a7 HD |
2841 | * If !PageUptodate, leave it that way so that freeable pages |
2842 | * can be recognized if we need to rollback on error later. | |
2843 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2844 | * than free the pages we are allocating (and SGP_CACHE pages |
2845 | * might still be clean: we now need to mark those dirty too). | |
2846 | */ | |
2847 | set_page_dirty(page); | |
2848 | unlock_page(page); | |
09cbfeaf | 2849 | put_page(page); |
e2d12e22 HD |
2850 | cond_resched(); |
2851 | } | |
2852 | ||
2853 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2854 | i_size_write(inode, offset + len); | |
e2d12e22 | 2855 | inode->i_ctime = CURRENT_TIME; |
1aac1400 HD |
2856 | undone: |
2857 | spin_lock(&inode->i_lock); | |
2858 | inode->i_private = NULL; | |
2859 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2860 | out: |
5955102c | 2861 | inode_unlock(inode); |
83e4fa9c HD |
2862 | return error; |
2863 | } | |
2864 | ||
726c3342 | 2865 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2866 | { |
726c3342 | 2867 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2868 | |
2869 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2870 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2871 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2872 | if (sbinfo->max_blocks) { |
1da177e4 | 2873 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2874 | buf->f_bavail = |
2875 | buf->f_bfree = sbinfo->max_blocks - | |
2876 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2877 | } |
2878 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2879 | buf->f_files = sbinfo->max_inodes; |
2880 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2881 | } |
2882 | /* else leave those fields 0 like simple_statfs */ | |
2883 | return 0; | |
2884 | } | |
2885 | ||
2886 | /* | |
2887 | * File creation. Allocate an inode, and we're done.. | |
2888 | */ | |
2889 | static int | |
1a67aafb | 2890 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2891 | { |
0b0a0806 | 2892 | struct inode *inode; |
1da177e4 LT |
2893 | int error = -ENOSPC; |
2894 | ||
454abafe | 2895 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2896 | if (inode) { |
feda821e CH |
2897 | error = simple_acl_create(dir, inode); |
2898 | if (error) | |
2899 | goto out_iput; | |
2a7dba39 | 2900 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2901 | &dentry->d_name, |
6d9d88d0 | 2902 | shmem_initxattrs, NULL); |
feda821e CH |
2903 | if (error && error != -EOPNOTSUPP) |
2904 | goto out_iput; | |
37ec43cd | 2905 | |
718deb6b | 2906 | error = 0; |
1da177e4 LT |
2907 | dir->i_size += BOGO_DIRENT_SIZE; |
2908 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2909 | d_instantiate(dentry, inode); | |
2910 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2911 | } |
2912 | return error; | |
feda821e CH |
2913 | out_iput: |
2914 | iput(inode); | |
2915 | return error; | |
1da177e4 LT |
2916 | } |
2917 | ||
60545d0d AV |
2918 | static int |
2919 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2920 | { | |
2921 | struct inode *inode; | |
2922 | int error = -ENOSPC; | |
2923 | ||
2924 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
2925 | if (inode) { | |
2926 | error = security_inode_init_security(inode, dir, | |
2927 | NULL, | |
2928 | shmem_initxattrs, NULL); | |
feda821e CH |
2929 | if (error && error != -EOPNOTSUPP) |
2930 | goto out_iput; | |
2931 | error = simple_acl_create(dir, inode); | |
2932 | if (error) | |
2933 | goto out_iput; | |
60545d0d AV |
2934 | d_tmpfile(dentry, inode); |
2935 | } | |
2936 | return error; | |
feda821e CH |
2937 | out_iput: |
2938 | iput(inode); | |
2939 | return error; | |
60545d0d AV |
2940 | } |
2941 | ||
18bb1db3 | 2942 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2943 | { |
2944 | int error; | |
2945 | ||
2946 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
2947 | return error; | |
d8c76e6f | 2948 | inc_nlink(dir); |
1da177e4 LT |
2949 | return 0; |
2950 | } | |
2951 | ||
4acdaf27 | 2952 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2953 | bool excl) |
1da177e4 LT |
2954 | { |
2955 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
2956 | } | |
2957 | ||
2958 | /* | |
2959 | * Link a file.. | |
2960 | */ | |
2961 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2962 | { | |
75c3cfa8 | 2963 | struct inode *inode = d_inode(old_dentry); |
5b04c689 | 2964 | int ret; |
1da177e4 LT |
2965 | |
2966 | /* | |
2967 | * No ordinary (disk based) filesystem counts links as inodes; | |
2968 | * but each new link needs a new dentry, pinning lowmem, and | |
2969 | * tmpfs dentries cannot be pruned until they are unlinked. | |
2970 | */ | |
5b04c689 PE |
2971 | ret = shmem_reserve_inode(inode->i_sb); |
2972 | if (ret) | |
2973 | goto out; | |
1da177e4 LT |
2974 | |
2975 | dir->i_size += BOGO_DIRENT_SIZE; | |
2976 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 2977 | inc_nlink(inode); |
7de9c6ee | 2978 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
2979 | dget(dentry); /* Extra pinning count for the created dentry */ |
2980 | d_instantiate(dentry, inode); | |
5b04c689 PE |
2981 | out: |
2982 | return ret; | |
1da177e4 LT |
2983 | } |
2984 | ||
2985 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
2986 | { | |
75c3cfa8 | 2987 | struct inode *inode = d_inode(dentry); |
1da177e4 | 2988 | |
5b04c689 PE |
2989 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2990 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
2991 | |
2992 | dir->i_size -= BOGO_DIRENT_SIZE; | |
2993 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 2994 | drop_nlink(inode); |
1da177e4 LT |
2995 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
2996 | return 0; | |
2997 | } | |
2998 | ||
2999 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
3000 | { | |
3001 | if (!simple_empty(dentry)) | |
3002 | return -ENOTEMPTY; | |
3003 | ||
75c3cfa8 | 3004 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 3005 | drop_nlink(dir); |
1da177e4 LT |
3006 | return shmem_unlink(dir, dentry); |
3007 | } | |
3008 | ||
37456771 MS |
3009 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
3010 | { | |
e36cb0b8 DH |
3011 | bool old_is_dir = d_is_dir(old_dentry); |
3012 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
3013 | |
3014 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
3015 | if (old_is_dir) { | |
3016 | drop_nlink(old_dir); | |
3017 | inc_nlink(new_dir); | |
3018 | } else { | |
3019 | drop_nlink(new_dir); | |
3020 | inc_nlink(old_dir); | |
3021 | } | |
3022 | } | |
3023 | old_dir->i_ctime = old_dir->i_mtime = | |
3024 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 DH |
3025 | d_inode(old_dentry)->i_ctime = |
3026 | d_inode(new_dentry)->i_ctime = CURRENT_TIME; | |
37456771 MS |
3027 | |
3028 | return 0; | |
3029 | } | |
3030 | ||
46fdb794 MS |
3031 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
3032 | { | |
3033 | struct dentry *whiteout; | |
3034 | int error; | |
3035 | ||
3036 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
3037 | if (!whiteout) | |
3038 | return -ENOMEM; | |
3039 | ||
3040 | error = shmem_mknod(old_dir, whiteout, | |
3041 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
3042 | dput(whiteout); | |
3043 | if (error) | |
3044 | return error; | |
3045 | ||
3046 | /* | |
3047 | * Cheat and hash the whiteout while the old dentry is still in | |
3048 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
3049 | * | |
3050 | * d_lookup() will consistently find one of them at this point, | |
3051 | * not sure which one, but that isn't even important. | |
3052 | */ | |
3053 | d_rehash(whiteout); | |
3054 | return 0; | |
3055 | } | |
3056 | ||
1da177e4 LT |
3057 | /* |
3058 | * The VFS layer already does all the dentry stuff for rename, | |
3059 | * we just have to decrement the usage count for the target if | |
3060 | * it exists so that the VFS layer correctly free's it when it | |
3061 | * gets overwritten. | |
3062 | */ | |
3b69ff51 | 3063 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 3064 | { |
75c3cfa8 | 3065 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
3066 | int they_are_dirs = S_ISDIR(inode->i_mode); |
3067 | ||
46fdb794 | 3068 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
3069 | return -EINVAL; |
3070 | ||
37456771 MS |
3071 | if (flags & RENAME_EXCHANGE) |
3072 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
3073 | ||
1da177e4 LT |
3074 | if (!simple_empty(new_dentry)) |
3075 | return -ENOTEMPTY; | |
3076 | ||
46fdb794 MS |
3077 | if (flags & RENAME_WHITEOUT) { |
3078 | int error; | |
3079 | ||
3080 | error = shmem_whiteout(old_dir, old_dentry); | |
3081 | if (error) | |
3082 | return error; | |
3083 | } | |
3084 | ||
75c3cfa8 | 3085 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 3086 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 3087 | if (they_are_dirs) { |
75c3cfa8 | 3088 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 3089 | drop_nlink(old_dir); |
b928095b | 3090 | } |
1da177e4 | 3091 | } else if (they_are_dirs) { |
9a53c3a7 | 3092 | drop_nlink(old_dir); |
d8c76e6f | 3093 | inc_nlink(new_dir); |
1da177e4 LT |
3094 | } |
3095 | ||
3096 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
3097 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
3098 | old_dir->i_ctime = old_dir->i_mtime = | |
3099 | new_dir->i_ctime = new_dir->i_mtime = | |
3100 | inode->i_ctime = CURRENT_TIME; | |
3101 | return 0; | |
3102 | } | |
3103 | ||
3104 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
3105 | { | |
3106 | int error; | |
3107 | int len; | |
3108 | struct inode *inode; | |
9276aad6 | 3109 | struct page *page; |
1da177e4 LT |
3110 | struct shmem_inode_info *info; |
3111 | ||
3112 | len = strlen(symname) + 1; | |
09cbfeaf | 3113 | if (len > PAGE_SIZE) |
1da177e4 LT |
3114 | return -ENAMETOOLONG; |
3115 | ||
454abafe | 3116 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
3117 | if (!inode) |
3118 | return -ENOSPC; | |
3119 | ||
9d8f13ba | 3120 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3121 | shmem_initxattrs, NULL); |
570bc1c2 SS |
3122 | if (error) { |
3123 | if (error != -EOPNOTSUPP) { | |
3124 | iput(inode); | |
3125 | return error; | |
3126 | } | |
3127 | error = 0; | |
3128 | } | |
3129 | ||
1da177e4 LT |
3130 | info = SHMEM_I(inode); |
3131 | inode->i_size = len-1; | |
69f07ec9 | 3132 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3133 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3134 | if (!inode->i_link) { | |
69f07ec9 HD |
3135 | iput(inode); |
3136 | return -ENOMEM; | |
3137 | } | |
3138 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3139 | } else { |
e8ecde25 | 3140 | inode_nohighmem(inode); |
9e18eb29 | 3141 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
3142 | if (error) { |
3143 | iput(inode); | |
3144 | return error; | |
3145 | } | |
14fcc23f | 3146 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 3147 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 3148 | memcpy(page_address(page), symname, len); |
ec9516fb | 3149 | SetPageUptodate(page); |
1da177e4 | 3150 | set_page_dirty(page); |
6746aff7 | 3151 | unlock_page(page); |
09cbfeaf | 3152 | put_page(page); |
1da177e4 | 3153 | } |
1da177e4 LT |
3154 | dir->i_size += BOGO_DIRENT_SIZE; |
3155 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
3156 | d_instantiate(dentry, inode); | |
3157 | dget(dentry); | |
3158 | return 0; | |
3159 | } | |
3160 | ||
fceef393 | 3161 | static void shmem_put_link(void *arg) |
1da177e4 | 3162 | { |
fceef393 AV |
3163 | mark_page_accessed(arg); |
3164 | put_page(arg); | |
1da177e4 LT |
3165 | } |
3166 | ||
6b255391 | 3167 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
3168 | struct inode *inode, |
3169 | struct delayed_call *done) | |
1da177e4 | 3170 | { |
1da177e4 | 3171 | struct page *page = NULL; |
6b255391 | 3172 | int error; |
6a6c9904 AV |
3173 | if (!dentry) { |
3174 | page = find_get_page(inode->i_mapping, 0); | |
3175 | if (!page) | |
3176 | return ERR_PTR(-ECHILD); | |
3177 | if (!PageUptodate(page)) { | |
3178 | put_page(page); | |
3179 | return ERR_PTR(-ECHILD); | |
3180 | } | |
3181 | } else { | |
9e18eb29 | 3182 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3183 | if (error) |
3184 | return ERR_PTR(error); | |
3185 | unlock_page(page); | |
3186 | } | |
fceef393 | 3187 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3188 | return page_address(page); |
1da177e4 LT |
3189 | } |
3190 | ||
b09e0fa4 | 3191 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3192 | /* |
b09e0fa4 EP |
3193 | * Superblocks without xattr inode operations may get some security.* xattr |
3194 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3195 | * like ACLs, we also need to implement the security.* handlers at |
3196 | * filesystem level, though. | |
3197 | */ | |
3198 | ||
6d9d88d0 JS |
3199 | /* |
3200 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3201 | */ | |
3202 | static int shmem_initxattrs(struct inode *inode, | |
3203 | const struct xattr *xattr_array, | |
3204 | void *fs_info) | |
3205 | { | |
3206 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3207 | const struct xattr *xattr; | |
38f38657 | 3208 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3209 | size_t len; |
3210 | ||
3211 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3212 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3213 | if (!new_xattr) |
3214 | return -ENOMEM; | |
3215 | ||
3216 | len = strlen(xattr->name) + 1; | |
3217 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3218 | GFP_KERNEL); | |
3219 | if (!new_xattr->name) { | |
3220 | kfree(new_xattr); | |
3221 | return -ENOMEM; | |
3222 | } | |
3223 | ||
3224 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3225 | XATTR_SECURITY_PREFIX_LEN); | |
3226 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3227 | xattr->name, len); | |
3228 | ||
38f38657 | 3229 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3230 | } |
3231 | ||
3232 | return 0; | |
3233 | } | |
3234 | ||
aa7c5241 | 3235 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3236 | struct dentry *unused, struct inode *inode, |
3237 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3238 | { |
b296821a | 3239 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3240 | |
aa7c5241 | 3241 | name = xattr_full_name(handler, name); |
38f38657 | 3242 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3243 | } |
3244 | ||
aa7c5241 | 3245 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
59301226 AV |
3246 | struct dentry *unused, struct inode *inode, |
3247 | const char *name, const void *value, | |
3248 | size_t size, int flags) | |
b09e0fa4 | 3249 | { |
59301226 | 3250 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3251 | |
aa7c5241 | 3252 | name = xattr_full_name(handler, name); |
38f38657 | 3253 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
3254 | } |
3255 | ||
aa7c5241 AG |
3256 | static const struct xattr_handler shmem_security_xattr_handler = { |
3257 | .prefix = XATTR_SECURITY_PREFIX, | |
3258 | .get = shmem_xattr_handler_get, | |
3259 | .set = shmem_xattr_handler_set, | |
3260 | }; | |
b09e0fa4 | 3261 | |
aa7c5241 AG |
3262 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3263 | .prefix = XATTR_TRUSTED_PREFIX, | |
3264 | .get = shmem_xattr_handler_get, | |
3265 | .set = shmem_xattr_handler_set, | |
3266 | }; | |
b09e0fa4 | 3267 | |
aa7c5241 AG |
3268 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3269 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3270 | &posix_acl_access_xattr_handler, | |
3271 | &posix_acl_default_xattr_handler, | |
3272 | #endif | |
3273 | &shmem_security_xattr_handler, | |
3274 | &shmem_trusted_xattr_handler, | |
3275 | NULL | |
3276 | }; | |
b09e0fa4 EP |
3277 | |
3278 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3279 | { | |
75c3cfa8 | 3280 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3281 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3282 | } |
3283 | #endif /* CONFIG_TMPFS_XATTR */ | |
3284 | ||
69f07ec9 | 3285 | static const struct inode_operations shmem_short_symlink_operations = { |
b09e0fa4 | 3286 | .readlink = generic_readlink, |
6b255391 | 3287 | .get_link = simple_get_link, |
b09e0fa4 | 3288 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3289 | .setxattr = generic_setxattr, |
3290 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3291 | .listxattr = shmem_listxattr, |
aa7c5241 | 3292 | .removexattr = generic_removexattr, |
b09e0fa4 EP |
3293 | #endif |
3294 | }; | |
3295 | ||
3296 | static const struct inode_operations shmem_symlink_inode_operations = { | |
3297 | .readlink = generic_readlink, | |
6b255391 | 3298 | .get_link = shmem_get_link, |
b09e0fa4 | 3299 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3300 | .setxattr = generic_setxattr, |
3301 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3302 | .listxattr = shmem_listxattr, |
aa7c5241 | 3303 | .removexattr = generic_removexattr, |
39f0247d | 3304 | #endif |
b09e0fa4 | 3305 | }; |
39f0247d | 3306 | |
91828a40 DG |
3307 | static struct dentry *shmem_get_parent(struct dentry *child) |
3308 | { | |
3309 | return ERR_PTR(-ESTALE); | |
3310 | } | |
3311 | ||
3312 | static int shmem_match(struct inode *ino, void *vfh) | |
3313 | { | |
3314 | __u32 *fh = vfh; | |
3315 | __u64 inum = fh[2]; | |
3316 | inum = (inum << 32) | fh[1]; | |
3317 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
3318 | } | |
3319 | ||
480b116c CH |
3320 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3321 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3322 | { |
91828a40 | 3323 | struct inode *inode; |
480b116c | 3324 | struct dentry *dentry = NULL; |
35c2a7f4 | 3325 | u64 inum; |
480b116c CH |
3326 | |
3327 | if (fh_len < 3) | |
3328 | return NULL; | |
91828a40 | 3329 | |
35c2a7f4 HD |
3330 | inum = fid->raw[2]; |
3331 | inum = (inum << 32) | fid->raw[1]; | |
3332 | ||
480b116c CH |
3333 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
3334 | shmem_match, fid->raw); | |
91828a40 | 3335 | if (inode) { |
480b116c | 3336 | dentry = d_find_alias(inode); |
91828a40 DG |
3337 | iput(inode); |
3338 | } | |
3339 | ||
480b116c | 3340 | return dentry; |
91828a40 DG |
3341 | } |
3342 | ||
b0b0382b AV |
3343 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3344 | struct inode *parent) | |
91828a40 | 3345 | { |
5fe0c237 AK |
3346 | if (*len < 3) { |
3347 | *len = 3; | |
94e07a75 | 3348 | return FILEID_INVALID; |
5fe0c237 | 3349 | } |
91828a40 | 3350 | |
1d3382cb | 3351 | if (inode_unhashed(inode)) { |
91828a40 DG |
3352 | /* Unfortunately insert_inode_hash is not idempotent, |
3353 | * so as we hash inodes here rather than at creation | |
3354 | * time, we need a lock to ensure we only try | |
3355 | * to do it once | |
3356 | */ | |
3357 | static DEFINE_SPINLOCK(lock); | |
3358 | spin_lock(&lock); | |
1d3382cb | 3359 | if (inode_unhashed(inode)) |
91828a40 DG |
3360 | __insert_inode_hash(inode, |
3361 | inode->i_ino + inode->i_generation); | |
3362 | spin_unlock(&lock); | |
3363 | } | |
3364 | ||
3365 | fh[0] = inode->i_generation; | |
3366 | fh[1] = inode->i_ino; | |
3367 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
3368 | ||
3369 | *len = 3; | |
3370 | return 1; | |
3371 | } | |
3372 | ||
39655164 | 3373 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3374 | .get_parent = shmem_get_parent, |
91828a40 | 3375 | .encode_fh = shmem_encode_fh, |
480b116c | 3376 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3377 | }; |
3378 | ||
680d794b | 3379 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
3380 | bool remount) | |
1da177e4 LT |
3381 | { |
3382 | char *this_char, *value, *rest; | |
49cd0a5c | 3383 | struct mempolicy *mpol = NULL; |
8751e039 EB |
3384 | uid_t uid; |
3385 | gid_t gid; | |
1da177e4 | 3386 | |
b00dc3ad HD |
3387 | while (options != NULL) { |
3388 | this_char = options; | |
3389 | for (;;) { | |
3390 | /* | |
3391 | * NUL-terminate this option: unfortunately, | |
3392 | * mount options form a comma-separated list, | |
3393 | * but mpol's nodelist may also contain commas. | |
3394 | */ | |
3395 | options = strchr(options, ','); | |
3396 | if (options == NULL) | |
3397 | break; | |
3398 | options++; | |
3399 | if (!isdigit(*options)) { | |
3400 | options[-1] = '\0'; | |
3401 | break; | |
3402 | } | |
3403 | } | |
1da177e4 LT |
3404 | if (!*this_char) |
3405 | continue; | |
3406 | if ((value = strchr(this_char,'=')) != NULL) { | |
3407 | *value++ = 0; | |
3408 | } else { | |
1170532b JP |
3409 | pr_err("tmpfs: No value for mount option '%s'\n", |
3410 | this_char); | |
49cd0a5c | 3411 | goto error; |
1da177e4 LT |
3412 | } |
3413 | ||
3414 | if (!strcmp(this_char,"size")) { | |
3415 | unsigned long long size; | |
3416 | size = memparse(value,&rest); | |
3417 | if (*rest == '%') { | |
3418 | size <<= PAGE_SHIFT; | |
3419 | size *= totalram_pages; | |
3420 | do_div(size, 100); | |
3421 | rest++; | |
3422 | } | |
3423 | if (*rest) | |
3424 | goto bad_val; | |
680d794b | 3425 | sbinfo->max_blocks = |
09cbfeaf | 3426 | DIV_ROUND_UP(size, PAGE_SIZE); |
1da177e4 | 3427 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 3428 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
3429 | if (*rest) |
3430 | goto bad_val; | |
3431 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 3432 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
3433 | if (*rest) |
3434 | goto bad_val; | |
3435 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 3436 | if (remount) |
1da177e4 | 3437 | continue; |
680d794b | 3438 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
3439 | if (*rest) |
3440 | goto bad_val; | |
3441 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 3442 | if (remount) |
1da177e4 | 3443 | continue; |
8751e039 | 3444 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3445 | if (*rest) |
3446 | goto bad_val; | |
8751e039 EB |
3447 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
3448 | if (!uid_valid(sbinfo->uid)) | |
3449 | goto bad_val; | |
1da177e4 | 3450 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 3451 | if (remount) |
1da177e4 | 3452 | continue; |
8751e039 | 3453 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3454 | if (*rest) |
3455 | goto bad_val; | |
8751e039 EB |
3456 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
3457 | if (!gid_valid(sbinfo->gid)) | |
3458 | goto bad_val; | |
e496cf3d | 3459 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3460 | } else if (!strcmp(this_char, "huge")) { |
3461 | int huge; | |
3462 | huge = shmem_parse_huge(value); | |
3463 | if (huge < 0) | |
3464 | goto bad_val; | |
3465 | if (!has_transparent_hugepage() && | |
3466 | huge != SHMEM_HUGE_NEVER) | |
3467 | goto bad_val; | |
3468 | sbinfo->huge = huge; | |
3469 | #endif | |
3470 | #ifdef CONFIG_NUMA | |
7339ff83 | 3471 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
3472 | mpol_put(mpol); |
3473 | mpol = NULL; | |
3474 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 3475 | goto bad_val; |
5a6e75f8 | 3476 | #endif |
1da177e4 | 3477 | } else { |
1170532b | 3478 | pr_err("tmpfs: Bad mount option %s\n", this_char); |
49cd0a5c | 3479 | goto error; |
1da177e4 LT |
3480 | } |
3481 | } | |
49cd0a5c | 3482 | sbinfo->mpol = mpol; |
1da177e4 LT |
3483 | return 0; |
3484 | ||
3485 | bad_val: | |
1170532b | 3486 | pr_err("tmpfs: Bad value '%s' for mount option '%s'\n", |
1da177e4 | 3487 | value, this_char); |
49cd0a5c GT |
3488 | error: |
3489 | mpol_put(mpol); | |
1da177e4 LT |
3490 | return 1; |
3491 | ||
3492 | } | |
3493 | ||
3494 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
3495 | { | |
3496 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 3497 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
3498 | unsigned long inodes; |
3499 | int error = -EINVAL; | |
3500 | ||
5f00110f | 3501 | config.mpol = NULL; |
680d794b | 3502 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 3503 | return error; |
1da177e4 | 3504 | |
0edd73b3 | 3505 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3506 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 3507 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 3508 | goto out; |
680d794b | 3509 | if (config.max_inodes < inodes) |
0edd73b3 HD |
3510 | goto out; |
3511 | /* | |
54af6042 | 3512 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
3513 | * but we must separately disallow unlimited->limited, because |
3514 | * in that case we have no record of how much is already in use. | |
3515 | */ | |
680d794b | 3516 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 3517 | goto out; |
680d794b | 3518 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
3519 | goto out; |
3520 | ||
3521 | error = 0; | |
5a6e75f8 | 3522 | sbinfo->huge = config.huge; |
680d794b | 3523 | sbinfo->max_blocks = config.max_blocks; |
680d794b | 3524 | sbinfo->max_inodes = config.max_inodes; |
3525 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 3526 | |
5f00110f GT |
3527 | /* |
3528 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3529 | */ | |
3530 | if (config.mpol) { | |
3531 | mpol_put(sbinfo->mpol); | |
3532 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
3533 | } | |
0edd73b3 HD |
3534 | out: |
3535 | spin_unlock(&sbinfo->stat_lock); | |
3536 | return error; | |
1da177e4 | 3537 | } |
680d794b | 3538 | |
34c80b1d | 3539 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3540 | { |
34c80b1d | 3541 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b | 3542 | |
3543 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3544 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3545 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 3546 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
3547 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
3548 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 3549 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3550 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3551 | seq_printf(seq, ",uid=%u", | |
3552 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3553 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3554 | seq_printf(seq, ",gid=%u", | |
3555 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
e496cf3d | 3556 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3557 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
3558 | if (sbinfo->huge) | |
3559 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3560 | #endif | |
71fe804b | 3561 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 3562 | return 0; |
3563 | } | |
9183df25 DH |
3564 | |
3565 | #define MFD_NAME_PREFIX "memfd:" | |
3566 | #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1) | |
3567 | #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN) | |
3568 | ||
3569 | #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING) | |
3570 | ||
3571 | SYSCALL_DEFINE2(memfd_create, | |
3572 | const char __user *, uname, | |
3573 | unsigned int, flags) | |
3574 | { | |
3575 | struct shmem_inode_info *info; | |
3576 | struct file *file; | |
3577 | int fd, error; | |
3578 | char *name; | |
3579 | long len; | |
3580 | ||
3581 | if (flags & ~(unsigned int)MFD_ALL_FLAGS) | |
3582 | return -EINVAL; | |
3583 | ||
3584 | /* length includes terminating zero */ | |
3585 | len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1); | |
3586 | if (len <= 0) | |
3587 | return -EFAULT; | |
3588 | if (len > MFD_NAME_MAX_LEN + 1) | |
3589 | return -EINVAL; | |
3590 | ||
3591 | name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY); | |
3592 | if (!name) | |
3593 | return -ENOMEM; | |
3594 | ||
3595 | strcpy(name, MFD_NAME_PREFIX); | |
3596 | if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) { | |
3597 | error = -EFAULT; | |
3598 | goto err_name; | |
3599 | } | |
3600 | ||
3601 | /* terminating-zero may have changed after strnlen_user() returned */ | |
3602 | if (name[len + MFD_NAME_PREFIX_LEN - 1]) { | |
3603 | error = -EFAULT; | |
3604 | goto err_name; | |
3605 | } | |
3606 | ||
3607 | fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0); | |
3608 | if (fd < 0) { | |
3609 | error = fd; | |
3610 | goto err_name; | |
3611 | } | |
3612 | ||
3613 | file = shmem_file_setup(name, 0, VM_NORESERVE); | |
3614 | if (IS_ERR(file)) { | |
3615 | error = PTR_ERR(file); | |
3616 | goto err_fd; | |
3617 | } | |
3618 | info = SHMEM_I(file_inode(file)); | |
3619 | file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; | |
3620 | file->f_flags |= O_RDWR | O_LARGEFILE; | |
3621 | if (flags & MFD_ALLOW_SEALING) | |
3622 | info->seals &= ~F_SEAL_SEAL; | |
3623 | ||
3624 | fd_install(fd, file); | |
3625 | kfree(name); | |
3626 | return fd; | |
3627 | ||
3628 | err_fd: | |
3629 | put_unused_fd(fd); | |
3630 | err_name: | |
3631 | kfree(name); | |
3632 | return error; | |
3633 | } | |
3634 | ||
680d794b | 3635 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3636 | |
3637 | static void shmem_put_super(struct super_block *sb) | |
3638 | { | |
602586a8 HD |
3639 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3640 | ||
3641 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 3642 | mpol_put(sbinfo->mpol); |
602586a8 | 3643 | kfree(sbinfo); |
1da177e4 LT |
3644 | sb->s_fs_info = NULL; |
3645 | } | |
3646 | ||
2b2af54a | 3647 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3648 | { |
3649 | struct inode *inode; | |
0edd73b3 | 3650 | struct shmem_sb_info *sbinfo; |
680d794b | 3651 | int err = -ENOMEM; |
3652 | ||
3653 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3654 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 3655 | L1_CACHE_BYTES), GFP_KERNEL); |
3656 | if (!sbinfo) | |
3657 | return -ENOMEM; | |
3658 | ||
680d794b | 3659 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
3660 | sbinfo->uid = current_fsuid(); |
3661 | sbinfo->gid = current_fsgid(); | |
680d794b | 3662 | sb->s_fs_info = sbinfo; |
1da177e4 | 3663 | |
0edd73b3 | 3664 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3665 | /* |
3666 | * Per default we only allow half of the physical ram per | |
3667 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3668 | * but the internal instance is left unlimited. | |
3669 | */ | |
ca4e0519 | 3670 | if (!(sb->s_flags & MS_KERNMOUNT)) { |
680d794b | 3671 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3672 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3673 | if (shmem_parse_options(data, sbinfo, false)) { | |
3674 | err = -EINVAL; | |
3675 | goto failed; | |
3676 | } | |
ca4e0519 AV |
3677 | } else { |
3678 | sb->s_flags |= MS_NOUSER; | |
1da177e4 | 3679 | } |
91828a40 | 3680 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 3681 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
3682 | #else |
3683 | sb->s_flags |= MS_NOUSER; | |
3684 | #endif | |
3685 | ||
0edd73b3 | 3686 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3687 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3688 | goto failed; |
680d794b | 3689 | sbinfo->free_inodes = sbinfo->max_inodes; |
779750d2 KS |
3690 | spin_lock_init(&sbinfo->shrinklist_lock); |
3691 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 3692 | |
285b2c4f | 3693 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3694 | sb->s_blocksize = PAGE_SIZE; |
3695 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3696 | sb->s_magic = TMPFS_MAGIC; |
3697 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3698 | sb->s_time_gran = 1; |
b09e0fa4 | 3699 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3700 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3701 | #endif |
3702 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
3703 | sb->s_flags |= MS_POSIXACL; |
3704 | #endif | |
0edd73b3 | 3705 | |
454abafe | 3706 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3707 | if (!inode) |
3708 | goto failed; | |
680d794b | 3709 | inode->i_uid = sbinfo->uid; |
3710 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3711 | sb->s_root = d_make_root(inode); |
3712 | if (!sb->s_root) | |
48fde701 | 3713 | goto failed; |
1da177e4 LT |
3714 | return 0; |
3715 | ||
1da177e4 LT |
3716 | failed: |
3717 | shmem_put_super(sb); | |
3718 | return err; | |
3719 | } | |
3720 | ||
fcc234f8 | 3721 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3722 | |
3723 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3724 | { | |
41ffe5d5 HD |
3725 | struct shmem_inode_info *info; |
3726 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3727 | if (!info) | |
1da177e4 | 3728 | return NULL; |
41ffe5d5 | 3729 | return &info->vfs_inode; |
1da177e4 LT |
3730 | } |
3731 | ||
41ffe5d5 | 3732 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3733 | { |
3734 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84e710da AV |
3735 | if (S_ISLNK(inode->i_mode)) |
3736 | kfree(inode->i_link); | |
fa0d7e3d NP |
3737 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3738 | } | |
3739 | ||
1da177e4 LT |
3740 | static void shmem_destroy_inode(struct inode *inode) |
3741 | { | |
09208d15 | 3742 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3743 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3744 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3745 | } |
3746 | ||
41ffe5d5 | 3747 | static void shmem_init_inode(void *foo) |
1da177e4 | 3748 | { |
41ffe5d5 HD |
3749 | struct shmem_inode_info *info = foo; |
3750 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3751 | } |
3752 | ||
41ffe5d5 | 3753 | static int shmem_init_inodecache(void) |
1da177e4 LT |
3754 | { |
3755 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3756 | sizeof(struct shmem_inode_info), | |
5d097056 | 3757 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3758 | return 0; |
3759 | } | |
3760 | ||
41ffe5d5 | 3761 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3762 | { |
1a1d92c1 | 3763 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3764 | } |
3765 | ||
f5e54d6e | 3766 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3767 | .writepage = shmem_writepage, |
76719325 | 3768 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3769 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3770 | .write_begin = shmem_write_begin, |
3771 | .write_end = shmem_write_end, | |
1da177e4 | 3772 | #endif |
1c93923c | 3773 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3774 | .migratepage = migrate_page, |
1c93923c | 3775 | #endif |
aa261f54 | 3776 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3777 | }; |
3778 | ||
15ad7cdc | 3779 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3780 | .mmap = shmem_mmap, |
c01d5b30 | 3781 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3782 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3783 | .llseek = shmem_file_llseek, |
2ba5bbed | 3784 | .read_iter = shmem_file_read_iter, |
8174202b | 3785 | .write_iter = generic_file_write_iter, |
1b061d92 | 3786 | .fsync = noop_fsync, |
708e3508 | 3787 | .splice_read = shmem_file_splice_read, |
f6cb85d0 | 3788 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3789 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3790 | #endif |
3791 | }; | |
3792 | ||
92e1d5be | 3793 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3794 | .getattr = shmem_getattr, |
94c1e62d | 3795 | .setattr = shmem_setattr, |
b09e0fa4 | 3796 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3797 | .setxattr = generic_setxattr, |
3798 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3799 | .listxattr = shmem_listxattr, |
aa7c5241 | 3800 | .removexattr = generic_removexattr, |
feda821e | 3801 | .set_acl = simple_set_acl, |
b09e0fa4 | 3802 | #endif |
1da177e4 LT |
3803 | }; |
3804 | ||
92e1d5be | 3805 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3806 | #ifdef CONFIG_TMPFS |
3807 | .create = shmem_create, | |
3808 | .lookup = simple_lookup, | |
3809 | .link = shmem_link, | |
3810 | .unlink = shmem_unlink, | |
3811 | .symlink = shmem_symlink, | |
3812 | .mkdir = shmem_mkdir, | |
3813 | .rmdir = shmem_rmdir, | |
3814 | .mknod = shmem_mknod, | |
3b69ff51 | 3815 | .rename2 = shmem_rename2, |
60545d0d | 3816 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3817 | #endif |
b09e0fa4 | 3818 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3819 | .setxattr = generic_setxattr, |
3820 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3821 | .listxattr = shmem_listxattr, |
aa7c5241 | 3822 | .removexattr = generic_removexattr, |
b09e0fa4 | 3823 | #endif |
39f0247d | 3824 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3825 | .setattr = shmem_setattr, |
feda821e | 3826 | .set_acl = simple_set_acl, |
39f0247d AG |
3827 | #endif |
3828 | }; | |
3829 | ||
92e1d5be | 3830 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3831 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3832 | .setxattr = generic_setxattr, |
3833 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3834 | .listxattr = shmem_listxattr, |
aa7c5241 | 3835 | .removexattr = generic_removexattr, |
b09e0fa4 | 3836 | #endif |
39f0247d | 3837 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3838 | .setattr = shmem_setattr, |
feda821e | 3839 | .set_acl = simple_set_acl, |
39f0247d | 3840 | #endif |
1da177e4 LT |
3841 | }; |
3842 | ||
759b9775 | 3843 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
3844 | .alloc_inode = shmem_alloc_inode, |
3845 | .destroy_inode = shmem_destroy_inode, | |
3846 | #ifdef CONFIG_TMPFS | |
3847 | .statfs = shmem_statfs, | |
3848 | .remount_fs = shmem_remount_fs, | |
680d794b | 3849 | .show_options = shmem_show_options, |
1da177e4 | 3850 | #endif |
1f895f75 | 3851 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3852 | .drop_inode = generic_delete_inode, |
3853 | .put_super = shmem_put_super, | |
779750d2 KS |
3854 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
3855 | .nr_cached_objects = shmem_unused_huge_count, | |
3856 | .free_cached_objects = shmem_unused_huge_scan, | |
3857 | #endif | |
1da177e4 LT |
3858 | }; |
3859 | ||
f0f37e2f | 3860 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3861 | .fault = shmem_fault, |
d7c17551 | 3862 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3863 | #ifdef CONFIG_NUMA |
3864 | .set_policy = shmem_set_policy, | |
3865 | .get_policy = shmem_get_policy, | |
3866 | #endif | |
3867 | }; | |
3868 | ||
3c26ff6e AV |
3869 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
3870 | int flags, const char *dev_name, void *data) | |
1da177e4 | 3871 | { |
3c26ff6e | 3872 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
3873 | } |
3874 | ||
41ffe5d5 | 3875 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3876 | .owner = THIS_MODULE, |
3877 | .name = "tmpfs", | |
3c26ff6e | 3878 | .mount = shmem_mount, |
1da177e4 | 3879 | .kill_sb = kill_litter_super, |
2b8576cb | 3880 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 3881 | }; |
1da177e4 | 3882 | |
41ffe5d5 | 3883 | int __init shmem_init(void) |
1da177e4 LT |
3884 | { |
3885 | int error; | |
3886 | ||
16203a7a RL |
3887 | /* If rootfs called this, don't re-init */ |
3888 | if (shmem_inode_cachep) | |
3889 | return 0; | |
3890 | ||
41ffe5d5 | 3891 | error = shmem_init_inodecache(); |
1da177e4 LT |
3892 | if (error) |
3893 | goto out3; | |
3894 | ||
41ffe5d5 | 3895 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 3896 | if (error) { |
1170532b | 3897 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
3898 | goto out2; |
3899 | } | |
95dc112a | 3900 | |
ca4e0519 | 3901 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
3902 | if (IS_ERR(shm_mnt)) { |
3903 | error = PTR_ERR(shm_mnt); | |
1170532b | 3904 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
3905 | goto out1; |
3906 | } | |
5a6e75f8 | 3907 | |
e496cf3d | 3908 | #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE |
5a6e75f8 KS |
3909 | if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY) |
3910 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; | |
3911 | else | |
3912 | shmem_huge = 0; /* just in case it was patched */ | |
3913 | #endif | |
1da177e4 LT |
3914 | return 0; |
3915 | ||
3916 | out1: | |
41ffe5d5 | 3917 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 3918 | out2: |
41ffe5d5 | 3919 | shmem_destroy_inodecache(); |
1da177e4 LT |
3920 | out3: |
3921 | shm_mnt = ERR_PTR(error); | |
3922 | return error; | |
3923 | } | |
853ac43a | 3924 | |
e496cf3d | 3925 | #if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS) |
5a6e75f8 KS |
3926 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
3927 | struct kobj_attribute *attr, char *buf) | |
3928 | { | |
3929 | int values[] = { | |
3930 | SHMEM_HUGE_ALWAYS, | |
3931 | SHMEM_HUGE_WITHIN_SIZE, | |
3932 | SHMEM_HUGE_ADVISE, | |
3933 | SHMEM_HUGE_NEVER, | |
3934 | SHMEM_HUGE_DENY, | |
3935 | SHMEM_HUGE_FORCE, | |
3936 | }; | |
3937 | int i, count; | |
3938 | ||
3939 | for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) { | |
3940 | const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s "; | |
3941 | ||
3942 | count += sprintf(buf + count, fmt, | |
3943 | shmem_format_huge(values[i])); | |
3944 | } | |
3945 | buf[count - 1] = '\n'; | |
3946 | return count; | |
3947 | } | |
3948 | ||
3949 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
3950 | struct kobj_attribute *attr, const char *buf, size_t count) | |
3951 | { | |
3952 | char tmp[16]; | |
3953 | int huge; | |
3954 | ||
3955 | if (count + 1 > sizeof(tmp)) | |
3956 | return -EINVAL; | |
3957 | memcpy(tmp, buf, count); | |
3958 | tmp[count] = '\0'; | |
3959 | if (count && tmp[count - 1] == '\n') | |
3960 | tmp[count - 1] = '\0'; | |
3961 | ||
3962 | huge = shmem_parse_huge(tmp); | |
3963 | if (huge == -EINVAL) | |
3964 | return -EINVAL; | |
3965 | if (!has_transparent_hugepage() && | |
3966 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
3967 | return -EINVAL; | |
3968 | ||
3969 | shmem_huge = huge; | |
3970 | if (shmem_huge < SHMEM_HUGE_DENY) | |
3971 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; | |
3972 | return count; | |
3973 | } | |
3974 | ||
3975 | struct kobj_attribute shmem_enabled_attr = | |
3976 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
f3f0e1d2 KS |
3977 | |
3978 | bool shmem_huge_enabled(struct vm_area_struct *vma) | |
3979 | { | |
3980 | struct inode *inode = file_inode(vma->vm_file); | |
3981 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
3982 | loff_t i_size; | |
3983 | pgoff_t off; | |
3984 | ||
3985 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
3986 | return true; | |
3987 | if (shmem_huge == SHMEM_HUGE_DENY) | |
3988 | return false; | |
3989 | switch (sbinfo->huge) { | |
3990 | case SHMEM_HUGE_NEVER: | |
3991 | return false; | |
3992 | case SHMEM_HUGE_ALWAYS: | |
3993 | return true; | |
3994 | case SHMEM_HUGE_WITHIN_SIZE: | |
3995 | off = round_up(vma->vm_pgoff, HPAGE_PMD_NR); | |
3996 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
3997 | if (i_size >= HPAGE_PMD_SIZE && | |
3998 | i_size >> PAGE_SHIFT >= off) | |
3999 | return true; | |
4000 | case SHMEM_HUGE_ADVISE: | |
4001 | /* TODO: implement fadvise() hints */ | |
4002 | return (vma->vm_flags & VM_HUGEPAGE); | |
4003 | default: | |
4004 | VM_BUG_ON(1); | |
4005 | return false; | |
4006 | } | |
4007 | } | |
e496cf3d | 4008 | #endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */ |
5a6e75f8 | 4009 | |
853ac43a MM |
4010 | #else /* !CONFIG_SHMEM */ |
4011 | ||
4012 | /* | |
4013 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
4014 | * | |
4015 | * This is intended for small system where the benefits of the full | |
4016 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4017 | * their complexity. On systems without swap this code should be | |
4018 | * effectively equivalent, but much lighter weight. | |
4019 | */ | |
4020 | ||
41ffe5d5 | 4021 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4022 | .name = "tmpfs", |
3c26ff6e | 4023 | .mount = ramfs_mount, |
853ac43a | 4024 | .kill_sb = kill_litter_super, |
2b8576cb | 4025 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4026 | }; |
4027 | ||
41ffe5d5 | 4028 | int __init shmem_init(void) |
853ac43a | 4029 | { |
41ffe5d5 | 4030 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4031 | |
41ffe5d5 | 4032 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
4033 | BUG_ON(IS_ERR(shm_mnt)); |
4034 | ||
4035 | return 0; | |
4036 | } | |
4037 | ||
41ffe5d5 | 4038 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
4039 | { |
4040 | return 0; | |
4041 | } | |
4042 | ||
3f96b79a HD |
4043 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
4044 | { | |
4045 | return 0; | |
4046 | } | |
4047 | ||
24513264 HD |
4048 | void shmem_unlock_mapping(struct address_space *mapping) |
4049 | { | |
4050 | } | |
4051 | ||
c01d5b30 HD |
4052 | #ifdef CONFIG_MMU |
4053 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4054 | unsigned long addr, unsigned long len, | |
4055 | unsigned long pgoff, unsigned long flags) | |
4056 | { | |
4057 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4058 | } | |
4059 | #endif | |
4060 | ||
41ffe5d5 | 4061 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4062 | { |
41ffe5d5 | 4063 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4064 | } |
4065 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4066 | ||
0b0a0806 HD |
4067 | #define shmem_vm_ops generic_file_vm_ops |
4068 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 4069 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
4070 | #define shmem_acct_size(flags, size) 0 |
4071 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
4072 | |
4073 | #endif /* CONFIG_SHMEM */ | |
4074 | ||
4075 | /* common code */ | |
1da177e4 | 4076 | |
3451538a | 4077 | static struct dentry_operations anon_ops = { |
118b2302 | 4078 | .d_dname = simple_dname |
3451538a AV |
4079 | }; |
4080 | ||
c7277090 EP |
4081 | static struct file *__shmem_file_setup(const char *name, loff_t size, |
4082 | unsigned long flags, unsigned int i_flags) | |
1da177e4 | 4083 | { |
6b4d0b27 | 4084 | struct file *res; |
1da177e4 | 4085 | struct inode *inode; |
2c48b9c4 | 4086 | struct path path; |
3451538a | 4087 | struct super_block *sb; |
1da177e4 LT |
4088 | struct qstr this; |
4089 | ||
4090 | if (IS_ERR(shm_mnt)) | |
6b4d0b27 | 4091 | return ERR_CAST(shm_mnt); |
1da177e4 | 4092 | |
285b2c4f | 4093 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4094 | return ERR_PTR(-EINVAL); |
4095 | ||
4096 | if (shmem_acct_size(flags, size)) | |
4097 | return ERR_PTR(-ENOMEM); | |
4098 | ||
6b4d0b27 | 4099 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
4100 | this.name = name; |
4101 | this.len = strlen(name); | |
4102 | this.hash = 0; /* will go */ | |
3451538a | 4103 | sb = shm_mnt->mnt_sb; |
66ee4b88 | 4104 | path.mnt = mntget(shm_mnt); |
3451538a | 4105 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 4106 | if (!path.dentry) |
1da177e4 | 4107 | goto put_memory; |
3451538a | 4108 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 4109 | |
6b4d0b27 | 4110 | res = ERR_PTR(-ENOSPC); |
3451538a | 4111 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 4112 | if (!inode) |
66ee4b88 | 4113 | goto put_memory; |
1da177e4 | 4114 | |
c7277090 | 4115 | inode->i_flags |= i_flags; |
2c48b9c4 | 4116 | d_instantiate(path.dentry, inode); |
1da177e4 | 4117 | inode->i_size = size; |
6d6b77f1 | 4118 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
4119 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
4120 | if (IS_ERR(res)) | |
66ee4b88 | 4121 | goto put_path; |
4b42af81 | 4122 | |
6b4d0b27 | 4123 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 4124 | &shmem_file_operations); |
6b4d0b27 | 4125 | if (IS_ERR(res)) |
66ee4b88 | 4126 | goto put_path; |
4b42af81 | 4127 | |
6b4d0b27 | 4128 | return res; |
1da177e4 | 4129 | |
1da177e4 LT |
4130 | put_memory: |
4131 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
4132 | put_path: |
4133 | path_put(&path); | |
6b4d0b27 | 4134 | return res; |
1da177e4 | 4135 | } |
c7277090 EP |
4136 | |
4137 | /** | |
4138 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4139 | * kernel internal. There will be NO LSM permission checks against the | |
4140 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4141 | * higher layer. The users are the big_key and shm implementations. LSM |
4142 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4143 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4144 | * @size: size to be set for the file | |
4145 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4146 | */ | |
4147 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4148 | { | |
4149 | return __shmem_file_setup(name, size, flags, S_PRIVATE); | |
4150 | } | |
4151 | ||
4152 | /** | |
4153 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4154 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4155 | * @size: size to be set for the file | |
4156 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4157 | */ | |
4158 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4159 | { | |
4160 | return __shmem_file_setup(name, size, flags, 0); | |
4161 | } | |
395e0ddc | 4162 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4163 | |
46711810 | 4164 | /** |
1da177e4 | 4165 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
4166 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
4167 | */ | |
4168 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4169 | { | |
4170 | struct file *file; | |
4171 | loff_t size = vma->vm_end - vma->vm_start; | |
4172 | ||
66fc1303 HD |
4173 | /* |
4174 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
4175 | * between XFS directory reading and selinux: since this file is only | |
4176 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4177 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4178 | */ | |
4179 | file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE); | |
1da177e4 LT |
4180 | if (IS_ERR(file)) |
4181 | return PTR_ERR(file); | |
4182 | ||
4183 | if (vma->vm_file) | |
4184 | fput(vma->vm_file); | |
4185 | vma->vm_file = file; | |
4186 | vma->vm_ops = &shmem_vm_ops; | |
f3f0e1d2 | 4187 | |
e496cf3d | 4188 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && |
f3f0e1d2 KS |
4189 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
4190 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
4191 | khugepaged_enter(vma, vma->vm_flags); | |
4192 | } | |
4193 | ||
1da177e4 LT |
4194 | return 0; |
4195 | } | |
d9d90e5e HD |
4196 | |
4197 | /** | |
4198 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
4199 | * @mapping: the page's address_space | |
4200 | * @index: the page index | |
4201 | * @gfp: the page allocator flags to use if allocating | |
4202 | * | |
4203 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4204 | * with any new page allocations done using the specified allocation flags. | |
4205 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
4206 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
4207 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4208 | * | |
68da9f05 HD |
4209 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4210 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4211 | */ |
4212 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4213 | pgoff_t index, gfp_t gfp) | |
4214 | { | |
68da9f05 HD |
4215 | #ifdef CONFIG_SHMEM |
4216 | struct inode *inode = mapping->host; | |
9276aad6 | 4217 | struct page *page; |
68da9f05 HD |
4218 | int error; |
4219 | ||
4220 | BUG_ON(mapping->a_ops != &shmem_aops); | |
9e18eb29 ALC |
4221 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
4222 | gfp, NULL, NULL); | |
68da9f05 HD |
4223 | if (error) |
4224 | page = ERR_PTR(error); | |
4225 | else | |
4226 | unlock_page(page); | |
4227 | return page; | |
4228 | #else | |
4229 | /* | |
4230 | * The tiny !SHMEM case uses ramfs without swap | |
4231 | */ | |
d9d90e5e | 4232 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4233 | #endif |
d9d90e5e HD |
4234 | } |
4235 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |