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> |
853ac43a MM |
35 | |
36 | static struct vfsmount *shm_mnt; | |
37 | ||
38 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
39 | /* |
40 | * This virtual memory filesystem is heavily based on the ramfs. It | |
41 | * extends ramfs by the ability to use swap and honor resource limits | |
42 | * which makes it a completely usable filesystem. | |
43 | */ | |
44 | ||
39f0247d | 45 | #include <linux/xattr.h> |
a5694255 | 46 | #include <linux/exportfs.h> |
1c7c474c | 47 | #include <linux/posix_acl.h> |
feda821e | 48 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 49 | #include <linux/mman.h> |
1da177e4 LT |
50 | #include <linux/string.h> |
51 | #include <linux/slab.h> | |
52 | #include <linux/backing-dev.h> | |
53 | #include <linux/shmem_fs.h> | |
1da177e4 | 54 | #include <linux/writeback.h> |
1da177e4 | 55 | #include <linux/blkdev.h> |
bda97eab | 56 | #include <linux/pagevec.h> |
41ffe5d5 | 57 | #include <linux/percpu_counter.h> |
83e4fa9c | 58 | #include <linux/falloc.h> |
708e3508 | 59 | #include <linux/splice.h> |
1da177e4 LT |
60 | #include <linux/security.h> |
61 | #include <linux/swapops.h> | |
62 | #include <linux/mempolicy.h> | |
63 | #include <linux/namei.h> | |
b00dc3ad | 64 | #include <linux/ctype.h> |
304dbdb7 | 65 | #include <linux/migrate.h> |
c1f60a5a | 66 | #include <linux/highmem.h> |
680d794b | 67 | #include <linux/seq_file.h> |
92562927 | 68 | #include <linux/magic.h> |
9183df25 | 69 | #include <linux/syscalls.h> |
40e041a2 | 70 | #include <linux/fcntl.h> |
9183df25 | 71 | #include <uapi/linux/memfd.h> |
304dbdb7 | 72 | |
1da177e4 | 73 | #include <asm/uaccess.h> |
1da177e4 LT |
74 | #include <asm/pgtable.h> |
75 | ||
dd56b046 MG |
76 | #include "internal.h" |
77 | ||
09cbfeaf KS |
78 | #define BLOCKS_PER_PAGE (PAGE_SIZE/512) |
79 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT) | |
1da177e4 | 80 | |
1da177e4 LT |
81 | /* Pretend that each entry is of this size in directory's i_size */ |
82 | #define BOGO_DIRENT_SIZE 20 | |
83 | ||
69f07ec9 HD |
84 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
85 | #define SHORT_SYMLINK_LEN 128 | |
86 | ||
1aac1400 | 87 | /* |
f00cdc6d HD |
88 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
89 | * inode->i_private (with i_mutex making sure that it has only one user at | |
90 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
91 | */ |
92 | struct shmem_falloc { | |
8e205f77 | 93 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
94 | pgoff_t start; /* start of range currently being fallocated */ |
95 | pgoff_t next; /* the next page offset to be fallocated */ | |
96 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
97 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
98 | }; | |
99 | ||
285b2c4f | 100 | /* Flag allocation requirements to shmem_getpage */ |
1da177e4 | 101 | enum sgp_type { |
1da177e4 LT |
102 | SGP_READ, /* don't exceed i_size, don't allocate page */ |
103 | SGP_CACHE, /* don't exceed i_size, may allocate page */ | |
657e3038 KS |
104 | SGP_NOHUGE, /* like SGP_CACHE, but no huge pages */ |
105 | SGP_HUGE, /* like SGP_CACHE, huge pages preferred */ | |
1635f6a7 HD |
106 | SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */ |
107 | SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */ | |
1da177e4 LT |
108 | }; |
109 | ||
b76db735 | 110 | #ifdef CONFIG_TMPFS |
680d794b | 111 | static unsigned long shmem_default_max_blocks(void) |
112 | { | |
113 | return totalram_pages / 2; | |
114 | } | |
115 | ||
116 | static unsigned long shmem_default_max_inodes(void) | |
117 | { | |
118 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
119 | } | |
b76db735 | 120 | #endif |
680d794b | 121 | |
bde05d1c HD |
122 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
123 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
124 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 | 125 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 ALC |
126 | struct page **pagep, enum sgp_type sgp, |
127 | gfp_t gfp, struct mm_struct *fault_mm, int *fault_type); | |
68da9f05 HD |
128 | |
129 | static inline int shmem_getpage(struct inode *inode, pgoff_t index, | |
9e18eb29 | 130 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
131 | { |
132 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
9e18eb29 | 133 | mapping_gfp_mask(inode->i_mapping), NULL, NULL); |
68da9f05 | 134 | } |
1da177e4 | 135 | |
1da177e4 LT |
136 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
137 | { | |
138 | return sb->s_fs_info; | |
139 | } | |
140 | ||
141 | /* | |
142 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
143 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
144 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
145 | * consistent with the pre-accounting of private mappings ... | |
146 | */ | |
147 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
148 | { | |
0b0a0806 | 149 | return (flags & VM_NORESERVE) ? |
191c5424 | 150 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
151 | } |
152 | ||
153 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
154 | { | |
0b0a0806 | 155 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
156 | vm_unacct_memory(VM_ACCT(size)); |
157 | } | |
158 | ||
77142517 KK |
159 | static inline int shmem_reacct_size(unsigned long flags, |
160 | loff_t oldsize, loff_t newsize) | |
161 | { | |
162 | if (!(flags & VM_NORESERVE)) { | |
163 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
164 | return security_vm_enough_memory_mm(current->mm, | |
165 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
166 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
167 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
168 | } | |
169 | return 0; | |
170 | } | |
171 | ||
1da177e4 LT |
172 | /* |
173 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 174 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
175 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
176 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
177 | */ | |
800d8c63 | 178 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 179 | { |
800d8c63 KS |
180 | if (!(flags & VM_NORESERVE)) |
181 | return 0; | |
182 | ||
183 | return security_vm_enough_memory_mm(current->mm, | |
184 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
185 | } |
186 | ||
187 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
188 | { | |
0b0a0806 | 189 | if (flags & VM_NORESERVE) |
09cbfeaf | 190 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
191 | } |
192 | ||
759b9775 | 193 | static const struct super_operations shmem_ops; |
f5e54d6e | 194 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 195 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
196 | static const struct inode_operations shmem_inode_operations; |
197 | static const struct inode_operations shmem_dir_inode_operations; | |
198 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 199 | static const struct vm_operations_struct shmem_vm_ops; |
1da177e4 | 200 | |
1da177e4 | 201 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 202 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 203 | |
5b04c689 PE |
204 | static int shmem_reserve_inode(struct super_block *sb) |
205 | { | |
206 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
207 | if (sbinfo->max_inodes) { | |
208 | spin_lock(&sbinfo->stat_lock); | |
209 | if (!sbinfo->free_inodes) { | |
210 | spin_unlock(&sbinfo->stat_lock); | |
211 | return -ENOSPC; | |
212 | } | |
213 | sbinfo->free_inodes--; | |
214 | spin_unlock(&sbinfo->stat_lock); | |
215 | } | |
216 | return 0; | |
217 | } | |
218 | ||
219 | static void shmem_free_inode(struct super_block *sb) | |
220 | { | |
221 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
222 | if (sbinfo->max_inodes) { | |
223 | spin_lock(&sbinfo->stat_lock); | |
224 | sbinfo->free_inodes++; | |
225 | spin_unlock(&sbinfo->stat_lock); | |
226 | } | |
227 | } | |
228 | ||
46711810 | 229 | /** |
41ffe5d5 | 230 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
231 | * @inode: inode to recalc |
232 | * | |
233 | * We have to calculate the free blocks since the mm can drop | |
234 | * undirtied hole pages behind our back. | |
235 | * | |
236 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
237 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
238 | * | |
239 | * It has to be called with the spinlock held. | |
240 | */ | |
241 | static void shmem_recalc_inode(struct inode *inode) | |
242 | { | |
243 | struct shmem_inode_info *info = SHMEM_I(inode); | |
244 | long freed; | |
245 | ||
246 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
247 | if (freed > 0) { | |
54af6042 HD |
248 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
249 | if (sbinfo->max_blocks) | |
250 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 251 | info->alloced -= freed; |
54af6042 | 252 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 253 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
254 | } |
255 | } | |
256 | ||
800d8c63 KS |
257 | bool shmem_charge(struct inode *inode, long pages) |
258 | { | |
259 | struct shmem_inode_info *info = SHMEM_I(inode); | |
260 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
261 | ||
262 | if (shmem_acct_block(info->flags, pages)) | |
263 | return false; | |
264 | spin_lock(&info->lock); | |
265 | info->alloced += pages; | |
266 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
267 | shmem_recalc_inode(inode); | |
268 | spin_unlock(&info->lock); | |
269 | inode->i_mapping->nrpages += pages; | |
270 | ||
271 | if (!sbinfo->max_blocks) | |
272 | return true; | |
273 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
274 | sbinfo->max_blocks - pages) > 0) { | |
275 | inode->i_mapping->nrpages -= pages; | |
276 | spin_lock(&info->lock); | |
277 | info->alloced -= pages; | |
278 | shmem_recalc_inode(inode); | |
279 | spin_unlock(&info->lock); | |
280 | ||
281 | return false; | |
282 | } | |
283 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
284 | return true; | |
285 | } | |
286 | ||
287 | void shmem_uncharge(struct inode *inode, long pages) | |
288 | { | |
289 | struct shmem_inode_info *info = SHMEM_I(inode); | |
290 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
291 | ||
292 | spin_lock(&info->lock); | |
293 | info->alloced -= pages; | |
294 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
295 | shmem_recalc_inode(inode); | |
296 | spin_unlock(&info->lock); | |
297 | ||
298 | if (sbinfo->max_blocks) | |
299 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
300 | } | |
301 | ||
7a5d0fbb HD |
302 | /* |
303 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
304 | */ | |
305 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
306 | pgoff_t index, void *expected, void *replacement) | |
307 | { | |
308 | void **pslot; | |
6dbaf22c | 309 | void *item; |
7a5d0fbb HD |
310 | |
311 | VM_BUG_ON(!expected); | |
6dbaf22c | 312 | VM_BUG_ON(!replacement); |
7a5d0fbb | 313 | pslot = radix_tree_lookup_slot(&mapping->page_tree, index); |
6dbaf22c JW |
314 | if (!pslot) |
315 | return -ENOENT; | |
316 | item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock); | |
7a5d0fbb HD |
317 | if (item != expected) |
318 | return -ENOENT; | |
6dbaf22c | 319 | radix_tree_replace_slot(pslot, replacement); |
7a5d0fbb HD |
320 | return 0; |
321 | } | |
322 | ||
d1899228 HD |
323 | /* |
324 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
325 | * that an entry was not already brought back from swap by a racing thread. | |
326 | * | |
327 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
328 | * might be reused, and again be SwapCache, using the same swap as before. | |
329 | */ | |
330 | static bool shmem_confirm_swap(struct address_space *mapping, | |
331 | pgoff_t index, swp_entry_t swap) | |
332 | { | |
333 | void *item; | |
334 | ||
335 | rcu_read_lock(); | |
336 | item = radix_tree_lookup(&mapping->page_tree, index); | |
337 | rcu_read_unlock(); | |
338 | return item == swp_to_radix_entry(swap); | |
339 | } | |
340 | ||
5a6e75f8 KS |
341 | /* |
342 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
343 | * | |
344 | * SHMEM_HUGE_NEVER: | |
345 | * disables huge pages for the mount; | |
346 | * SHMEM_HUGE_ALWAYS: | |
347 | * enables huge pages for the mount; | |
348 | * SHMEM_HUGE_WITHIN_SIZE: | |
349 | * only allocate huge pages if the page will be fully within i_size, | |
350 | * also respect fadvise()/madvise() hints; | |
351 | * SHMEM_HUGE_ADVISE: | |
352 | * only allocate huge pages if requested with fadvise()/madvise(); | |
353 | */ | |
354 | ||
355 | #define SHMEM_HUGE_NEVER 0 | |
356 | #define SHMEM_HUGE_ALWAYS 1 | |
357 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
358 | #define SHMEM_HUGE_ADVISE 3 | |
359 | ||
360 | /* | |
361 | * Special values. | |
362 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
363 | * | |
364 | * SHMEM_HUGE_DENY: | |
365 | * disables huge on shm_mnt and all mounts, for emergency use; | |
366 | * SHMEM_HUGE_FORCE: | |
367 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
368 | * | |
369 | */ | |
370 | #define SHMEM_HUGE_DENY (-1) | |
371 | #define SHMEM_HUGE_FORCE (-2) | |
372 | ||
373 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
374 | /* ifdef here to avoid bloating shmem.o when not necessary */ | |
375 | ||
376 | int shmem_huge __read_mostly; | |
377 | ||
378 | static int shmem_parse_huge(const char *str) | |
379 | { | |
380 | if (!strcmp(str, "never")) | |
381 | return SHMEM_HUGE_NEVER; | |
382 | if (!strcmp(str, "always")) | |
383 | return SHMEM_HUGE_ALWAYS; | |
384 | if (!strcmp(str, "within_size")) | |
385 | return SHMEM_HUGE_WITHIN_SIZE; | |
386 | if (!strcmp(str, "advise")) | |
387 | return SHMEM_HUGE_ADVISE; | |
388 | if (!strcmp(str, "deny")) | |
389 | return SHMEM_HUGE_DENY; | |
390 | if (!strcmp(str, "force")) | |
391 | return SHMEM_HUGE_FORCE; | |
392 | return -EINVAL; | |
393 | } | |
394 | ||
395 | static const char *shmem_format_huge(int huge) | |
396 | { | |
397 | switch (huge) { | |
398 | case SHMEM_HUGE_NEVER: | |
399 | return "never"; | |
400 | case SHMEM_HUGE_ALWAYS: | |
401 | return "always"; | |
402 | case SHMEM_HUGE_WITHIN_SIZE: | |
403 | return "within_size"; | |
404 | case SHMEM_HUGE_ADVISE: | |
405 | return "advise"; | |
406 | case SHMEM_HUGE_DENY: | |
407 | return "deny"; | |
408 | case SHMEM_HUGE_FORCE: | |
409 | return "force"; | |
410 | default: | |
411 | VM_BUG_ON(1); | |
412 | return "bad_val"; | |
413 | } | |
414 | } | |
415 | ||
416 | #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ | |
417 | ||
418 | #define shmem_huge SHMEM_HUGE_DENY | |
419 | ||
420 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
421 | ||
46f65ec1 HD |
422 | /* |
423 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
424 | */ | |
425 | static int shmem_add_to_page_cache(struct page *page, | |
426 | struct address_space *mapping, | |
fed400a1 | 427 | pgoff_t index, void *expected) |
46f65ec1 | 428 | { |
800d8c63 | 429 | int error, nr = hpage_nr_pages(page); |
46f65ec1 | 430 | |
800d8c63 KS |
431 | VM_BUG_ON_PAGE(PageTail(page), page); |
432 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
433 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
434 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 435 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 436 | |
800d8c63 | 437 | page_ref_add(page, nr); |
b065b432 HD |
438 | page->mapping = mapping; |
439 | page->index = index; | |
440 | ||
441 | spin_lock_irq(&mapping->tree_lock); | |
800d8c63 KS |
442 | if (PageTransHuge(page)) { |
443 | void __rcu **results; | |
444 | pgoff_t idx; | |
445 | int i; | |
446 | ||
447 | error = 0; | |
448 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, | |
449 | &results, &idx, index, 1) && | |
450 | idx < index + HPAGE_PMD_NR) { | |
451 | error = -EEXIST; | |
452 | } | |
453 | ||
454 | if (!error) { | |
455 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
456 | error = radix_tree_insert(&mapping->page_tree, | |
457 | index + i, page + i); | |
458 | VM_BUG_ON(error); | |
459 | } | |
460 | count_vm_event(THP_FILE_ALLOC); | |
461 | } | |
462 | } else if (!expected) { | |
b065b432 | 463 | error = radix_tree_insert(&mapping->page_tree, index, page); |
800d8c63 | 464 | } else { |
b065b432 HD |
465 | error = shmem_radix_tree_replace(mapping, index, expected, |
466 | page); | |
800d8c63 KS |
467 | } |
468 | ||
46f65ec1 | 469 | if (!error) { |
800d8c63 KS |
470 | mapping->nrpages += nr; |
471 | if (PageTransHuge(page)) | |
472 | __inc_zone_page_state(page, NR_SHMEM_THPS); | |
473 | __mod_zone_page_state(page_zone(page), NR_FILE_PAGES, nr); | |
474 | __mod_zone_page_state(page_zone(page), NR_SHMEM, nr); | |
b065b432 HD |
475 | spin_unlock_irq(&mapping->tree_lock); |
476 | } else { | |
477 | page->mapping = NULL; | |
478 | spin_unlock_irq(&mapping->tree_lock); | |
800d8c63 | 479 | page_ref_sub(page, nr); |
46f65ec1 | 480 | } |
46f65ec1 HD |
481 | return error; |
482 | } | |
483 | ||
6922c0c7 HD |
484 | /* |
485 | * Like delete_from_page_cache, but substitutes swap for page. | |
486 | */ | |
487 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
488 | { | |
489 | struct address_space *mapping = page->mapping; | |
490 | int error; | |
491 | ||
800d8c63 KS |
492 | VM_BUG_ON_PAGE(PageCompound(page), page); |
493 | ||
6922c0c7 HD |
494 | spin_lock_irq(&mapping->tree_lock); |
495 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
496 | page->mapping = NULL; | |
497 | mapping->nrpages--; | |
498 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
499 | __dec_zone_page_state(page, NR_SHMEM); | |
500 | spin_unlock_irq(&mapping->tree_lock); | |
09cbfeaf | 501 | put_page(page); |
6922c0c7 HD |
502 | BUG_ON(error); |
503 | } | |
504 | ||
7a5d0fbb HD |
505 | /* |
506 | * Remove swap entry from radix tree, free the swap and its page cache. | |
507 | */ | |
508 | static int shmem_free_swap(struct address_space *mapping, | |
509 | pgoff_t index, void *radswap) | |
510 | { | |
6dbaf22c | 511 | void *old; |
7a5d0fbb HD |
512 | |
513 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 514 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 515 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
516 | if (old != radswap) |
517 | return -ENOENT; | |
518 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
519 | return 0; | |
7a5d0fbb HD |
520 | } |
521 | ||
6a15a370 VB |
522 | /* |
523 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 524 | * given offsets are swapped out. |
6a15a370 VB |
525 | * |
526 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
527 | * as long as the inode doesn't go away and racy results are not a problem. | |
528 | */ | |
48131e03 VB |
529 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
530 | pgoff_t start, pgoff_t end) | |
6a15a370 | 531 | { |
6a15a370 VB |
532 | struct radix_tree_iter iter; |
533 | void **slot; | |
534 | struct page *page; | |
48131e03 | 535 | unsigned long swapped = 0; |
6a15a370 VB |
536 | |
537 | rcu_read_lock(); | |
538 | ||
6a15a370 VB |
539 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
540 | if (iter.index >= end) | |
541 | break; | |
542 | ||
543 | page = radix_tree_deref_slot(slot); | |
544 | ||
2cf938aa MW |
545 | if (radix_tree_deref_retry(page)) { |
546 | slot = radix_tree_iter_retry(&iter); | |
547 | continue; | |
548 | } | |
6a15a370 VB |
549 | |
550 | if (radix_tree_exceptional_entry(page)) | |
551 | swapped++; | |
552 | ||
553 | if (need_resched()) { | |
554 | cond_resched_rcu(); | |
7165092f | 555 | slot = radix_tree_iter_next(&iter); |
6a15a370 VB |
556 | } |
557 | } | |
558 | ||
559 | rcu_read_unlock(); | |
560 | ||
561 | return swapped << PAGE_SHIFT; | |
562 | } | |
563 | ||
48131e03 VB |
564 | /* |
565 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
566 | * given vma is swapped out. | |
567 | * | |
568 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
569 | * as long as the inode doesn't go away and racy results are not a problem. | |
570 | */ | |
571 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
572 | { | |
573 | struct inode *inode = file_inode(vma->vm_file); | |
574 | struct shmem_inode_info *info = SHMEM_I(inode); | |
575 | struct address_space *mapping = inode->i_mapping; | |
576 | unsigned long swapped; | |
577 | ||
578 | /* Be careful as we don't hold info->lock */ | |
579 | swapped = READ_ONCE(info->swapped); | |
580 | ||
581 | /* | |
582 | * The easier cases are when the shmem object has nothing in swap, or | |
583 | * the vma maps it whole. Then we can simply use the stats that we | |
584 | * already track. | |
585 | */ | |
586 | if (!swapped) | |
587 | return 0; | |
588 | ||
589 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
590 | return swapped << PAGE_SHIFT; | |
591 | ||
592 | /* Here comes the more involved part */ | |
593 | return shmem_partial_swap_usage(mapping, | |
594 | linear_page_index(vma, vma->vm_start), | |
595 | linear_page_index(vma, vma->vm_end)); | |
596 | } | |
597 | ||
24513264 HD |
598 | /* |
599 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
600 | */ | |
601 | void shmem_unlock_mapping(struct address_space *mapping) | |
602 | { | |
603 | struct pagevec pvec; | |
604 | pgoff_t indices[PAGEVEC_SIZE]; | |
605 | pgoff_t index = 0; | |
606 | ||
607 | pagevec_init(&pvec, 0); | |
608 | /* | |
609 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
610 | */ | |
611 | while (!mapping_unevictable(mapping)) { | |
612 | /* | |
613 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
614 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
615 | */ | |
0cd6144a JW |
616 | pvec.nr = find_get_entries(mapping, index, |
617 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
618 | if (!pvec.nr) |
619 | break; | |
620 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 621 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
622 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
623 | pagevec_release(&pvec); | |
624 | cond_resched(); | |
625 | } | |
7a5d0fbb HD |
626 | } |
627 | ||
628 | /* | |
629 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 630 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 631 | */ |
1635f6a7 HD |
632 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
633 | bool unfalloc) | |
1da177e4 | 634 | { |
285b2c4f | 635 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 636 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
637 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
638 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
639 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
640 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 641 | struct pagevec pvec; |
7a5d0fbb HD |
642 | pgoff_t indices[PAGEVEC_SIZE]; |
643 | long nr_swaps_freed = 0; | |
285b2c4f | 644 | pgoff_t index; |
bda97eab HD |
645 | int i; |
646 | ||
83e4fa9c HD |
647 | if (lend == -1) |
648 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
649 | |
650 | pagevec_init(&pvec, 0); | |
651 | index = start; | |
83e4fa9c | 652 | while (index < end) { |
0cd6144a JW |
653 | pvec.nr = find_get_entries(mapping, index, |
654 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
655 | pvec.pages, indices); | |
7a5d0fbb HD |
656 | if (!pvec.nr) |
657 | break; | |
bda97eab HD |
658 | for (i = 0; i < pagevec_count(&pvec); i++) { |
659 | struct page *page = pvec.pages[i]; | |
660 | ||
7a5d0fbb | 661 | index = indices[i]; |
83e4fa9c | 662 | if (index >= end) |
bda97eab HD |
663 | break; |
664 | ||
7a5d0fbb | 665 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
666 | if (unfalloc) |
667 | continue; | |
7a5d0fbb HD |
668 | nr_swaps_freed += !shmem_free_swap(mapping, |
669 | index, page); | |
bda97eab | 670 | continue; |
7a5d0fbb HD |
671 | } |
672 | ||
800d8c63 KS |
673 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
674 | ||
7a5d0fbb | 675 | if (!trylock_page(page)) |
bda97eab | 676 | continue; |
800d8c63 KS |
677 | |
678 | if (PageTransTail(page)) { | |
679 | /* Middle of THP: zero out the page */ | |
680 | clear_highpage(page); | |
681 | unlock_page(page); | |
682 | continue; | |
683 | } else if (PageTransHuge(page)) { | |
684 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
685 | /* | |
686 | * Range ends in the middle of THP: | |
687 | * zero out the page | |
688 | */ | |
689 | clear_highpage(page); | |
690 | unlock_page(page); | |
691 | continue; | |
692 | } | |
693 | index += HPAGE_PMD_NR - 1; | |
694 | i += HPAGE_PMD_NR - 1; | |
695 | } | |
696 | ||
1635f6a7 | 697 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
698 | VM_BUG_ON_PAGE(PageTail(page), page); |
699 | if (page_mapping(page) == mapping) { | |
309381fe | 700 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
701 | truncate_inode_page(mapping, page); |
702 | } | |
bda97eab | 703 | } |
bda97eab HD |
704 | unlock_page(page); |
705 | } | |
0cd6144a | 706 | pagevec_remove_exceptionals(&pvec); |
24513264 | 707 | pagevec_release(&pvec); |
bda97eab HD |
708 | cond_resched(); |
709 | index++; | |
710 | } | |
1da177e4 | 711 | |
83e4fa9c | 712 | if (partial_start) { |
bda97eab | 713 | struct page *page = NULL; |
9e18eb29 | 714 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 715 | if (page) { |
09cbfeaf | 716 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
717 | if (start > end) { |
718 | top = partial_end; | |
719 | partial_end = 0; | |
720 | } | |
721 | zero_user_segment(page, partial_start, top); | |
722 | set_page_dirty(page); | |
723 | unlock_page(page); | |
09cbfeaf | 724 | put_page(page); |
83e4fa9c HD |
725 | } |
726 | } | |
727 | if (partial_end) { | |
728 | struct page *page = NULL; | |
9e18eb29 | 729 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
730 | if (page) { |
731 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
732 | set_page_dirty(page); |
733 | unlock_page(page); | |
09cbfeaf | 734 | put_page(page); |
bda97eab HD |
735 | } |
736 | } | |
83e4fa9c HD |
737 | if (start >= end) |
738 | return; | |
bda97eab HD |
739 | |
740 | index = start; | |
b1a36650 | 741 | while (index < end) { |
bda97eab | 742 | cond_resched(); |
0cd6144a JW |
743 | |
744 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 745 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 746 | pvec.pages, indices); |
7a5d0fbb | 747 | if (!pvec.nr) { |
b1a36650 HD |
748 | /* If all gone or hole-punch or unfalloc, we're done */ |
749 | if (index == start || end != -1) | |
bda97eab | 750 | break; |
b1a36650 | 751 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
752 | index = start; |
753 | continue; | |
754 | } | |
bda97eab HD |
755 | for (i = 0; i < pagevec_count(&pvec); i++) { |
756 | struct page *page = pvec.pages[i]; | |
757 | ||
7a5d0fbb | 758 | index = indices[i]; |
83e4fa9c | 759 | if (index >= end) |
bda97eab HD |
760 | break; |
761 | ||
7a5d0fbb | 762 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
763 | if (unfalloc) |
764 | continue; | |
b1a36650 HD |
765 | if (shmem_free_swap(mapping, index, page)) { |
766 | /* Swap was replaced by page: retry */ | |
767 | index--; | |
768 | break; | |
769 | } | |
770 | nr_swaps_freed++; | |
7a5d0fbb HD |
771 | continue; |
772 | } | |
773 | ||
bda97eab | 774 | lock_page(page); |
800d8c63 KS |
775 | |
776 | if (PageTransTail(page)) { | |
777 | /* Middle of THP: zero out the page */ | |
778 | clear_highpage(page); | |
779 | unlock_page(page); | |
780 | /* | |
781 | * Partial thp truncate due 'start' in middle | |
782 | * of THP: don't need to look on these pages | |
783 | * again on !pvec.nr restart. | |
784 | */ | |
785 | if (index != round_down(end, HPAGE_PMD_NR)) | |
786 | start++; | |
787 | continue; | |
788 | } else if (PageTransHuge(page)) { | |
789 | if (index == round_down(end, HPAGE_PMD_NR)) { | |
790 | /* | |
791 | * Range ends in the middle of THP: | |
792 | * zero out the page | |
793 | */ | |
794 | clear_highpage(page); | |
795 | unlock_page(page); | |
796 | continue; | |
797 | } | |
798 | index += HPAGE_PMD_NR - 1; | |
799 | i += HPAGE_PMD_NR - 1; | |
800 | } | |
801 | ||
1635f6a7 | 802 | if (!unfalloc || !PageUptodate(page)) { |
800d8c63 KS |
803 | VM_BUG_ON_PAGE(PageTail(page), page); |
804 | if (page_mapping(page) == mapping) { | |
309381fe | 805 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 806 | truncate_inode_page(mapping, page); |
b1a36650 HD |
807 | } else { |
808 | /* Page was replaced by swap: retry */ | |
809 | unlock_page(page); | |
810 | index--; | |
811 | break; | |
1635f6a7 | 812 | } |
7a5d0fbb | 813 | } |
bda97eab HD |
814 | unlock_page(page); |
815 | } | |
0cd6144a | 816 | pagevec_remove_exceptionals(&pvec); |
24513264 | 817 | pagevec_release(&pvec); |
bda97eab HD |
818 | index++; |
819 | } | |
94c1e62d | 820 | |
1da177e4 | 821 | spin_lock(&info->lock); |
7a5d0fbb | 822 | info->swapped -= nr_swaps_freed; |
1da177e4 LT |
823 | shmem_recalc_inode(inode); |
824 | spin_unlock(&info->lock); | |
1635f6a7 | 825 | } |
1da177e4 | 826 | |
1635f6a7 HD |
827 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
828 | { | |
829 | shmem_undo_range(inode, lstart, lend, false); | |
285b2c4f | 830 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1da177e4 | 831 | } |
94c1e62d | 832 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 833 | |
44a30220 YZ |
834 | static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry, |
835 | struct kstat *stat) | |
836 | { | |
837 | struct inode *inode = dentry->d_inode; | |
838 | struct shmem_inode_info *info = SHMEM_I(inode); | |
839 | ||
d0424c42 HD |
840 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
841 | spin_lock(&info->lock); | |
842 | shmem_recalc_inode(inode); | |
843 | spin_unlock(&info->lock); | |
844 | } | |
44a30220 | 845 | generic_fillattr(inode, stat); |
44a30220 YZ |
846 | return 0; |
847 | } | |
848 | ||
94c1e62d | 849 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 850 | { |
75c3cfa8 | 851 | struct inode *inode = d_inode(dentry); |
40e041a2 | 852 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 LT |
853 | int error; |
854 | ||
db78b877 CH |
855 | error = inode_change_ok(inode, attr); |
856 | if (error) | |
857 | return error; | |
858 | ||
94c1e62d HD |
859 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
860 | loff_t oldsize = inode->i_size; | |
861 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 862 | |
40e041a2 DH |
863 | /* protected by i_mutex */ |
864 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
865 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
866 | return -EPERM; | |
867 | ||
94c1e62d | 868 | if (newsize != oldsize) { |
77142517 KK |
869 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
870 | oldsize, newsize); | |
871 | if (error) | |
872 | return error; | |
94c1e62d HD |
873 | i_size_write(inode, newsize); |
874 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
875 | } | |
afa2db2f | 876 | if (newsize <= oldsize) { |
94c1e62d | 877 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
878 | if (oldsize > holebegin) |
879 | unmap_mapping_range(inode->i_mapping, | |
880 | holebegin, 0, 1); | |
881 | if (info->alloced) | |
882 | shmem_truncate_range(inode, | |
883 | newsize, (loff_t)-1); | |
94c1e62d | 884 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
885 | if (oldsize > holebegin) |
886 | unmap_mapping_range(inode->i_mapping, | |
887 | holebegin, 0, 1); | |
94c1e62d | 888 | } |
1da177e4 LT |
889 | } |
890 | ||
db78b877 | 891 | setattr_copy(inode, attr); |
db78b877 | 892 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 893 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
894 | return error; |
895 | } | |
896 | ||
1f895f75 | 897 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 898 | { |
1da177e4 LT |
899 | struct shmem_inode_info *info = SHMEM_I(inode); |
900 | ||
3889e6e7 | 901 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
902 | shmem_unacct_size(info->flags, inode->i_size); |
903 | inode->i_size = 0; | |
3889e6e7 | 904 | shmem_truncate_range(inode, 0, (loff_t)-1); |
1da177e4 | 905 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 906 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 907 | list_del_init(&info->swaplist); |
cb5f7b9a | 908 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 909 | } |
3ed47db3 | 910 | } |
b09e0fa4 | 911 | |
38f38657 | 912 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 913 | WARN_ON(inode->i_blocks); |
5b04c689 | 914 | shmem_free_inode(inode->i_sb); |
dbd5768f | 915 | clear_inode(inode); |
1da177e4 LT |
916 | } |
917 | ||
46f65ec1 HD |
918 | /* |
919 | * If swap found in inode, free it and move page from swapcache to filecache. | |
920 | */ | |
41ffe5d5 | 921 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 922 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 923 | { |
285b2c4f | 924 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 925 | void *radswap; |
41ffe5d5 | 926 | pgoff_t index; |
bde05d1c HD |
927 | gfp_t gfp; |
928 | int error = 0; | |
1da177e4 | 929 | |
46f65ec1 | 930 | radswap = swp_to_radix_entry(swap); |
e504f3fd | 931 | index = radix_tree_locate_item(&mapping->page_tree, radswap); |
46f65ec1 | 932 | if (index == -1) |
00501b53 | 933 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 934 | |
1b1b32f2 HD |
935 | /* |
936 | * Move _head_ to start search for next from here. | |
1f895f75 | 937 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 938 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 939 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
940 | */ |
941 | if (shmem_swaplist.next != &info->swaplist) | |
942 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 943 | |
bde05d1c HD |
944 | gfp = mapping_gfp_mask(mapping); |
945 | if (shmem_should_replace_page(*pagep, gfp)) { | |
946 | mutex_unlock(&shmem_swaplist_mutex); | |
947 | error = shmem_replace_page(pagep, gfp, info, index); | |
948 | mutex_lock(&shmem_swaplist_mutex); | |
949 | /* | |
950 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
951 | * allocation, but the inode might have been freed while we |
952 | * dropped it: although a racing shmem_evict_inode() cannot | |
953 | * complete without emptying the radix_tree, our page lock | |
954 | * on this swapcache page is not enough to prevent that - | |
955 | * free_swap_and_cache() of our swap entry will only | |
956 | * trylock_page(), removing swap from radix_tree whatever. | |
957 | * | |
958 | * We must not proceed to shmem_add_to_page_cache() if the | |
959 | * inode has been freed, but of course we cannot rely on | |
960 | * inode or mapping or info to check that. However, we can | |
961 | * safely check if our swap entry is still in use (and here | |
962 | * it can't have got reused for another page): if it's still | |
963 | * in use, then the inode cannot have been freed yet, and we | |
964 | * can safely proceed (if it's no longer in use, that tells | |
965 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
966 | */ |
967 | if (!page_swapcount(*pagep)) | |
968 | error = -ENOENT; | |
969 | } | |
970 | ||
d13d1443 | 971 | /* |
778dd893 HD |
972 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
973 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
974 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 975 | */ |
bde05d1c HD |
976 | if (!error) |
977 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 978 | radswap); |
48f170fb | 979 | if (error != -ENOMEM) { |
46f65ec1 HD |
980 | /* |
981 | * Truncation and eviction use free_swap_and_cache(), which | |
982 | * only does trylock page: if we raced, best clean up here. | |
983 | */ | |
bde05d1c HD |
984 | delete_from_swap_cache(*pagep); |
985 | set_page_dirty(*pagep); | |
46f65ec1 HD |
986 | if (!error) { |
987 | spin_lock(&info->lock); | |
988 | info->swapped--; | |
989 | spin_unlock(&info->lock); | |
990 | swap_free(swap); | |
991 | } | |
1da177e4 | 992 | } |
2e0e26c7 | 993 | return error; |
1da177e4 LT |
994 | } |
995 | ||
996 | /* | |
46f65ec1 | 997 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 998 | */ |
41ffe5d5 | 999 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 1000 | { |
41ffe5d5 | 1001 | struct list_head *this, *next; |
1da177e4 | 1002 | struct shmem_inode_info *info; |
00501b53 | 1003 | struct mem_cgroup *memcg; |
bde05d1c HD |
1004 | int error = 0; |
1005 | ||
1006 | /* | |
1007 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 1008 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 1009 | */ |
0142ef6c | 1010 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 1011 | goto out; |
778dd893 HD |
1012 | |
1013 | /* | |
1014 | * Charge page using GFP_KERNEL while we can wait, before taking | |
1015 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
1016 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 1017 | */ |
f627c2f5 KS |
1018 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg, |
1019 | false); | |
778dd893 HD |
1020 | if (error) |
1021 | goto out; | |
46f65ec1 | 1022 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 1023 | error = -EAGAIN; |
1da177e4 | 1024 | |
cb5f7b9a | 1025 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
1026 | list_for_each_safe(this, next, &shmem_swaplist) { |
1027 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 1028 | if (info->swapped) |
00501b53 | 1029 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
1030 | else |
1031 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1032 | cond_resched(); |
00501b53 | 1033 | if (error != -EAGAIN) |
778dd893 | 1034 | break; |
00501b53 | 1035 | /* found nothing in this: move on to search the next */ |
1da177e4 | 1036 | } |
cb5f7b9a | 1037 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1038 | |
00501b53 JW |
1039 | if (error) { |
1040 | if (error != -ENOMEM) | |
1041 | error = 0; | |
f627c2f5 | 1042 | mem_cgroup_cancel_charge(page, memcg, false); |
00501b53 | 1043 | } else |
f627c2f5 | 1044 | mem_cgroup_commit_charge(page, memcg, true, false); |
778dd893 | 1045 | out: |
aaa46865 | 1046 | unlock_page(page); |
09cbfeaf | 1047 | put_page(page); |
778dd893 | 1048 | return error; |
1da177e4 LT |
1049 | } |
1050 | ||
1051 | /* | |
1052 | * Move the page from the page cache to the swap cache. | |
1053 | */ | |
1054 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1055 | { | |
1056 | struct shmem_inode_info *info; | |
1da177e4 | 1057 | struct address_space *mapping; |
1da177e4 | 1058 | struct inode *inode; |
6922c0c7 HD |
1059 | swp_entry_t swap; |
1060 | pgoff_t index; | |
1da177e4 | 1061 | |
800d8c63 | 1062 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1063 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1064 | mapping = page->mapping; |
1065 | index = page->index; | |
1066 | inode = mapping->host; | |
1067 | info = SHMEM_I(inode); | |
1068 | if (info->flags & VM_LOCKED) | |
1069 | goto redirty; | |
d9fe526a | 1070 | if (!total_swap_pages) |
1da177e4 LT |
1071 | goto redirty; |
1072 | ||
d9fe526a | 1073 | /* |
97b713ba CH |
1074 | * Our capabilities prevent regular writeback or sync from ever calling |
1075 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1076 | * its underlying filesystem, in which case tmpfs should write out to | |
1077 | * swap only in response to memory pressure, and not for the writeback | |
1078 | * threads or sync. | |
d9fe526a | 1079 | */ |
48f170fb HD |
1080 | if (!wbc->for_reclaim) { |
1081 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1082 | goto redirty; | |
1083 | } | |
1635f6a7 HD |
1084 | |
1085 | /* | |
1086 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1087 | * value into swapfile.c, the only way we can correctly account for a | |
1088 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1089 | * |
1090 | * That's okay for a page already fallocated earlier, but if we have | |
1091 | * not yet completed the fallocation, then (a) we want to keep track | |
1092 | * of this page in case we have to undo it, and (b) it may not be a | |
1093 | * good idea to continue anyway, once we're pushing into swap. So | |
1094 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1095 | */ |
1096 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1097 | if (inode->i_private) { |
1098 | struct shmem_falloc *shmem_falloc; | |
1099 | spin_lock(&inode->i_lock); | |
1100 | shmem_falloc = inode->i_private; | |
1101 | if (shmem_falloc && | |
8e205f77 | 1102 | !shmem_falloc->waitq && |
1aac1400 HD |
1103 | index >= shmem_falloc->start && |
1104 | index < shmem_falloc->next) | |
1105 | shmem_falloc->nr_unswapped++; | |
1106 | else | |
1107 | shmem_falloc = NULL; | |
1108 | spin_unlock(&inode->i_lock); | |
1109 | if (shmem_falloc) | |
1110 | goto redirty; | |
1111 | } | |
1635f6a7 HD |
1112 | clear_highpage(page); |
1113 | flush_dcache_page(page); | |
1114 | SetPageUptodate(page); | |
1115 | } | |
1116 | ||
48f170fb HD |
1117 | swap = get_swap_page(); |
1118 | if (!swap.val) | |
1119 | goto redirty; | |
d9fe526a | 1120 | |
37e84351 VD |
1121 | if (mem_cgroup_try_charge_swap(page, swap)) |
1122 | goto free_swap; | |
1123 | ||
b1dea800 HD |
1124 | /* |
1125 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1126 | * if it's not already there. Do it now before the page is |
1127 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1128 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1129 | * we've incremented swapped, because shmem_unuse_inode() will |
1130 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1131 | */ |
48f170fb HD |
1132 | mutex_lock(&shmem_swaplist_mutex); |
1133 | if (list_empty(&info->swaplist)) | |
1134 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 1135 | |
48f170fb | 1136 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
6922c0c7 | 1137 | spin_lock(&info->lock); |
6922c0c7 | 1138 | shmem_recalc_inode(inode); |
267a4c76 | 1139 | info->swapped++; |
826267cf | 1140 | spin_unlock(&info->lock); |
6922c0c7 | 1141 | |
267a4c76 HD |
1142 | swap_shmem_alloc(swap); |
1143 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1144 | ||
6922c0c7 | 1145 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1146 | BUG_ON(page_mapped(page)); |
9fab5619 | 1147 | swap_writepage(page, wbc); |
1da177e4 LT |
1148 | return 0; |
1149 | } | |
1150 | ||
6922c0c7 | 1151 | mutex_unlock(&shmem_swaplist_mutex); |
37e84351 | 1152 | free_swap: |
0a31bc97 | 1153 | swapcache_free(swap); |
1da177e4 LT |
1154 | redirty: |
1155 | set_page_dirty(page); | |
d9fe526a HD |
1156 | if (wbc->for_reclaim) |
1157 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1158 | unlock_page(page); | |
1159 | return 0; | |
1da177e4 LT |
1160 | } |
1161 | ||
75edd345 | 1162 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1163 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1164 | { |
095f1fc4 | 1165 | char buffer[64]; |
680d794b | 1166 | |
71fe804b | 1167 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1168 | return; /* show nothing */ |
680d794b | 1169 | |
a7a88b23 | 1170 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1171 | |
1172 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1173 | } |
71fe804b LS |
1174 | |
1175 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1176 | { | |
1177 | struct mempolicy *mpol = NULL; | |
1178 | if (sbinfo->mpol) { | |
1179 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1180 | mpol = sbinfo->mpol; | |
1181 | mpol_get(mpol); | |
1182 | spin_unlock(&sbinfo->stat_lock); | |
1183 | } | |
1184 | return mpol; | |
1185 | } | |
75edd345 HD |
1186 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1187 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1188 | { | |
1189 | } | |
1190 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1191 | { | |
1192 | return NULL; | |
1193 | } | |
1194 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1195 | #ifndef CONFIG_NUMA | |
1196 | #define vm_policy vm_private_data | |
1197 | #endif | |
680d794b | 1198 | |
800d8c63 KS |
1199 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1200 | struct shmem_inode_info *info, pgoff_t index) | |
1201 | { | |
1202 | /* Create a pseudo vma that just contains the policy */ | |
1203 | vma->vm_start = 0; | |
1204 | /* Bias interleave by inode number to distribute better across nodes */ | |
1205 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
1206 | vma->vm_ops = NULL; | |
1207 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); | |
1208 | } | |
1209 | ||
1210 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1211 | { | |
1212 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1213 | mpol_cond_put(vma->vm_policy); | |
1214 | } | |
1215 | ||
41ffe5d5 HD |
1216 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1217 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1218 | { |
1da177e4 | 1219 | struct vm_area_struct pvma; |
18a2f371 | 1220 | struct page *page; |
52cd3b07 | 1221 | |
800d8c63 | 1222 | shmem_pseudo_vma_init(&pvma, info, index); |
18a2f371 | 1223 | page = swapin_readahead(swap, gfp, &pvma, 0); |
800d8c63 | 1224 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1225 | |
800d8c63 KS |
1226 | return page; |
1227 | } | |
1228 | ||
1229 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1230 | struct shmem_inode_info *info, pgoff_t index) | |
1231 | { | |
1232 | struct vm_area_struct pvma; | |
1233 | struct inode *inode = &info->vfs_inode; | |
1234 | struct address_space *mapping = inode->i_mapping; | |
1235 | pgoff_t idx, hindex = round_down(index, HPAGE_PMD_NR); | |
1236 | void __rcu **results; | |
1237 | struct page *page; | |
1238 | ||
1239 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) | |
1240 | return NULL; | |
1241 | ||
1242 | rcu_read_lock(); | |
1243 | if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx, | |
1244 | hindex, 1) && idx < hindex + HPAGE_PMD_NR) { | |
1245 | rcu_read_unlock(); | |
1246 | return NULL; | |
1247 | } | |
1248 | rcu_read_unlock(); | |
18a2f371 | 1249 | |
800d8c63 KS |
1250 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1251 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
1252 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); | |
1253 | shmem_pseudo_vma_destroy(&pvma); | |
1254 | if (page) | |
1255 | prep_transhuge_page(page); | |
18a2f371 | 1256 | return page; |
1da177e4 LT |
1257 | } |
1258 | ||
02098fea | 1259 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1260 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1261 | { |
1262 | struct vm_area_struct pvma; | |
18a2f371 | 1263 | struct page *page; |
1da177e4 | 1264 | |
800d8c63 KS |
1265 | shmem_pseudo_vma_init(&pvma, info, index); |
1266 | page = alloc_page_vma(gfp, &pvma, 0); | |
1267 | shmem_pseudo_vma_destroy(&pvma); | |
1268 | ||
1269 | return page; | |
1270 | } | |
1271 | ||
1272 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
1273 | struct shmem_inode_info *info, struct shmem_sb_info *sbinfo, | |
1274 | pgoff_t index, bool huge) | |
1275 | { | |
1276 | struct page *page; | |
1277 | int nr; | |
1278 | int err = -ENOSPC; | |
52cd3b07 | 1279 | |
800d8c63 KS |
1280 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) |
1281 | huge = false; | |
1282 | nr = huge ? HPAGE_PMD_NR : 1; | |
1283 | ||
1284 | if (shmem_acct_block(info->flags, nr)) | |
1285 | goto failed; | |
1286 | if (sbinfo->max_blocks) { | |
1287 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
1288 | sbinfo->max_blocks - nr) > 0) | |
1289 | goto unacct; | |
1290 | percpu_counter_add(&sbinfo->used_blocks, nr); | |
1291 | } | |
1292 | ||
1293 | if (huge) | |
1294 | page = shmem_alloc_hugepage(gfp, info, index); | |
1295 | else | |
1296 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1297 | if (page) { |
1298 | __SetPageLocked(page); | |
1299 | __SetPageSwapBacked(page); | |
800d8c63 | 1300 | return page; |
75edd345 | 1301 | } |
18a2f371 | 1302 | |
800d8c63 KS |
1303 | err = -ENOMEM; |
1304 | if (sbinfo->max_blocks) | |
1305 | percpu_counter_add(&sbinfo->used_blocks, -nr); | |
1306 | unacct: | |
1307 | shmem_unacct_blocks(info->flags, nr); | |
1308 | failed: | |
1309 | return ERR_PTR(err); | |
1da177e4 | 1310 | } |
71fe804b | 1311 | |
bde05d1c HD |
1312 | /* |
1313 | * When a page is moved from swapcache to shmem filecache (either by the | |
1314 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1315 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1316 | * ignorance of the mapping it belongs to. If that mapping has special | |
1317 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1318 | * we may need to copy to a suitable page before moving to filecache. | |
1319 | * | |
1320 | * In a future release, this may well be extended to respect cpuset and | |
1321 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1322 | * but for now it is a simple matter of zone. | |
1323 | */ | |
1324 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1325 | { | |
1326 | return page_zonenum(page) > gfp_zone(gfp); | |
1327 | } | |
1328 | ||
1329 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1330 | struct shmem_inode_info *info, pgoff_t index) | |
1331 | { | |
1332 | struct page *oldpage, *newpage; | |
1333 | struct address_space *swap_mapping; | |
1334 | pgoff_t swap_index; | |
1335 | int error; | |
1336 | ||
1337 | oldpage = *pagep; | |
1338 | swap_index = page_private(oldpage); | |
1339 | swap_mapping = page_mapping(oldpage); | |
1340 | ||
1341 | /* | |
1342 | * We have arrived here because our zones are constrained, so don't | |
1343 | * limit chance of success by further cpuset and node constraints. | |
1344 | */ | |
1345 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1346 | newpage = shmem_alloc_page(gfp, info, index); | |
1347 | if (!newpage) | |
1348 | return -ENOMEM; | |
bde05d1c | 1349 | |
09cbfeaf | 1350 | get_page(newpage); |
bde05d1c | 1351 | copy_highpage(newpage, oldpage); |
0142ef6c | 1352 | flush_dcache_page(newpage); |
bde05d1c | 1353 | |
bde05d1c | 1354 | SetPageUptodate(newpage); |
bde05d1c | 1355 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1356 | SetPageSwapCache(newpage); |
1357 | ||
1358 | /* | |
1359 | * Our caller will very soon move newpage out of swapcache, but it's | |
1360 | * a nice clean interface for us to replace oldpage by newpage there. | |
1361 | */ | |
1362 | spin_lock_irq(&swap_mapping->tree_lock); | |
1363 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1364 | newpage); | |
0142ef6c HD |
1365 | if (!error) { |
1366 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
1367 | __dec_zone_page_state(oldpage, NR_FILE_PAGES); | |
1368 | } | |
bde05d1c | 1369 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1370 | |
0142ef6c HD |
1371 | if (unlikely(error)) { |
1372 | /* | |
1373 | * Is this possible? I think not, now that our callers check | |
1374 | * both PageSwapCache and page_private after getting page lock; | |
1375 | * but be defensive. Reverse old to newpage for clear and free. | |
1376 | */ | |
1377 | oldpage = newpage; | |
1378 | } else { | |
6a93ca8f | 1379 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1380 | lru_cache_add_anon(newpage); |
1381 | *pagep = newpage; | |
1382 | } | |
bde05d1c HD |
1383 | |
1384 | ClearPageSwapCache(oldpage); | |
1385 | set_page_private(oldpage, 0); | |
1386 | ||
1387 | unlock_page(oldpage); | |
09cbfeaf KS |
1388 | put_page(oldpage); |
1389 | put_page(oldpage); | |
0142ef6c | 1390 | return error; |
bde05d1c HD |
1391 | } |
1392 | ||
1da177e4 | 1393 | /* |
68da9f05 | 1394 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1395 | * |
1396 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1397 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1398 | * entry since a page cannot live in both the swap and page cache. |
1399 | * | |
1400 | * fault_mm and fault_type are only supplied by shmem_fault: | |
1401 | * otherwise they are NULL. | |
1da177e4 | 1402 | */ |
41ffe5d5 | 1403 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 ALC |
1404 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
1405 | struct mm_struct *fault_mm, int *fault_type) | |
1da177e4 LT |
1406 | { |
1407 | struct address_space *mapping = inode->i_mapping; | |
54af6042 | 1408 | struct shmem_inode_info *info; |
1da177e4 | 1409 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1410 | struct mm_struct *charge_mm; |
00501b53 | 1411 | struct mem_cgroup *memcg; |
27ab7006 | 1412 | struct page *page; |
1da177e4 | 1413 | swp_entry_t swap; |
657e3038 | 1414 | enum sgp_type sgp_huge = sgp; |
800d8c63 | 1415 | pgoff_t hindex = index; |
1da177e4 | 1416 | int error; |
54af6042 | 1417 | int once = 0; |
1635f6a7 | 1418 | int alloced = 0; |
1da177e4 | 1419 | |
09cbfeaf | 1420 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1421 | return -EFBIG; |
657e3038 KS |
1422 | if (sgp == SGP_NOHUGE || sgp == SGP_HUGE) |
1423 | sgp = SGP_CACHE; | |
1da177e4 | 1424 | repeat: |
54af6042 | 1425 | swap.val = 0; |
0cd6144a | 1426 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1427 | if (radix_tree_exceptional_entry(page)) { |
1428 | swap = radix_to_swp_entry(page); | |
1429 | page = NULL; | |
1430 | } | |
1431 | ||
75edd345 | 1432 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1433 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 1434 | error = -EINVAL; |
267a4c76 | 1435 | goto unlock; |
54af6042 HD |
1436 | } |
1437 | ||
66d2f4d2 HD |
1438 | if (page && sgp == SGP_WRITE) |
1439 | mark_page_accessed(page); | |
1440 | ||
1635f6a7 HD |
1441 | /* fallocated page? */ |
1442 | if (page && !PageUptodate(page)) { | |
1443 | if (sgp != SGP_READ) | |
1444 | goto clear; | |
1445 | unlock_page(page); | |
09cbfeaf | 1446 | put_page(page); |
1635f6a7 HD |
1447 | page = NULL; |
1448 | } | |
54af6042 | 1449 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1450 | *pagep = page; |
1451 | return 0; | |
27ab7006 HD |
1452 | } |
1453 | ||
1454 | /* | |
54af6042 HD |
1455 | * Fast cache lookup did not find it: |
1456 | * bring it back from swap or allocate. | |
27ab7006 | 1457 | */ |
54af6042 HD |
1458 | info = SHMEM_I(inode); |
1459 | sbinfo = SHMEM_SB(inode->i_sb); | |
9e18eb29 | 1460 | charge_mm = fault_mm ? : current->mm; |
1da177e4 | 1461 | |
1da177e4 LT |
1462 | if (swap.val) { |
1463 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1464 | page = lookup_swap_cache(swap); |
1465 | if (!page) { | |
9e18eb29 ALC |
1466 | /* Or update major stats only when swapin succeeds?? */ |
1467 | if (fault_type) { | |
68da9f05 | 1468 | *fault_type |= VM_FAULT_MAJOR; |
9e18eb29 ALC |
1469 | count_vm_event(PGMAJFAULT); |
1470 | mem_cgroup_count_vm_event(fault_mm, PGMAJFAULT); | |
1471 | } | |
1472 | /* Here we actually start the io */ | |
41ffe5d5 | 1473 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1474 | if (!page) { |
54af6042 HD |
1475 | error = -ENOMEM; |
1476 | goto failed; | |
1da177e4 | 1477 | } |
1da177e4 LT |
1478 | } |
1479 | ||
1480 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1481 | lock_page(page); |
0142ef6c | 1482 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1483 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1484 | error = -EEXIST; /* try again */ |
d1899228 | 1485 | goto unlock; |
bde05d1c | 1486 | } |
27ab7006 | 1487 | if (!PageUptodate(page)) { |
1da177e4 | 1488 | error = -EIO; |
54af6042 | 1489 | goto failed; |
1da177e4 | 1490 | } |
54af6042 HD |
1491 | wait_on_page_writeback(page); |
1492 | ||
bde05d1c HD |
1493 | if (shmem_should_replace_page(page, gfp)) { |
1494 | error = shmem_replace_page(&page, gfp, info, index); | |
1495 | if (error) | |
1496 | goto failed; | |
1da177e4 | 1497 | } |
27ab7006 | 1498 | |
9e18eb29 | 1499 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
f627c2f5 | 1500 | false); |
d1899228 | 1501 | if (!error) { |
aa3b1895 | 1502 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1503 | swp_to_radix_entry(swap)); |
215c02bc HD |
1504 | /* |
1505 | * We already confirmed swap under page lock, and make | |
1506 | * no memory allocation here, so usually no possibility | |
1507 | * of error; but free_swap_and_cache() only trylocks a | |
1508 | * page, so it is just possible that the entry has been | |
1509 | * truncated or holepunched since swap was confirmed. | |
1510 | * shmem_undo_range() will have done some of the | |
1511 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1512 | * the rest. |
215c02bc HD |
1513 | * Reset swap.val? No, leave it so "failed" goes back to |
1514 | * "repeat": reading a hole and writing should succeed. | |
1515 | */ | |
00501b53 | 1516 | if (error) { |
f627c2f5 | 1517 | mem_cgroup_cancel_charge(page, memcg, false); |
215c02bc | 1518 | delete_from_swap_cache(page); |
00501b53 | 1519 | } |
d1899228 | 1520 | } |
54af6042 HD |
1521 | if (error) |
1522 | goto failed; | |
1523 | ||
f627c2f5 | 1524 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1525 | |
54af6042 | 1526 | spin_lock(&info->lock); |
285b2c4f | 1527 | info->swapped--; |
54af6042 | 1528 | shmem_recalc_inode(inode); |
27ab7006 | 1529 | spin_unlock(&info->lock); |
54af6042 | 1530 | |
66d2f4d2 HD |
1531 | if (sgp == SGP_WRITE) |
1532 | mark_page_accessed(page); | |
1533 | ||
54af6042 | 1534 | delete_from_swap_cache(page); |
27ab7006 HD |
1535 | set_page_dirty(page); |
1536 | swap_free(swap); | |
1537 | ||
54af6042 | 1538 | } else { |
800d8c63 KS |
1539 | /* shmem_symlink() */ |
1540 | if (mapping->a_ops != &shmem_aops) | |
1541 | goto alloc_nohuge; | |
657e3038 | 1542 | if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE) |
800d8c63 KS |
1543 | goto alloc_nohuge; |
1544 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1545 | goto alloc_huge; | |
1546 | switch (sbinfo->huge) { | |
1547 | loff_t i_size; | |
1548 | pgoff_t off; | |
1549 | case SHMEM_HUGE_NEVER: | |
1550 | goto alloc_nohuge; | |
1551 | case SHMEM_HUGE_WITHIN_SIZE: | |
1552 | off = round_up(index, HPAGE_PMD_NR); | |
1553 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1554 | if (i_size >= HPAGE_PMD_SIZE && | |
1555 | i_size >> PAGE_SHIFT >= off) | |
1556 | goto alloc_huge; | |
1557 | /* fallthrough */ | |
1558 | case SHMEM_HUGE_ADVISE: | |
657e3038 KS |
1559 | if (sgp_huge == SGP_HUGE) |
1560 | goto alloc_huge; | |
1561 | /* TODO: implement fadvise() hints */ | |
800d8c63 | 1562 | goto alloc_nohuge; |
54af6042 | 1563 | } |
1da177e4 | 1564 | |
800d8c63 KS |
1565 | alloc_huge: |
1566 | page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1567 | index, true); | |
1568 | if (IS_ERR(page)) { | |
1569 | alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo, | |
1570 | index, false); | |
1da177e4 | 1571 | } |
800d8c63 KS |
1572 | if (IS_ERR(page)) { |
1573 | error = PTR_ERR(page); | |
1574 | page = NULL; | |
1575 | goto failed; | |
1576 | } | |
1577 | ||
1578 | if (PageTransHuge(page)) | |
1579 | hindex = round_down(index, HPAGE_PMD_NR); | |
1580 | else | |
1581 | hindex = index; | |
1582 | ||
66d2f4d2 | 1583 | if (sgp == SGP_WRITE) |
eb39d618 | 1584 | __SetPageReferenced(page); |
66d2f4d2 | 1585 | |
9e18eb29 | 1586 | error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg, |
800d8c63 | 1587 | PageTransHuge(page)); |
54af6042 | 1588 | if (error) |
800d8c63 KS |
1589 | goto unacct; |
1590 | error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK, | |
1591 | compound_order(page)); | |
b065b432 | 1592 | if (!error) { |
800d8c63 | 1593 | error = shmem_add_to_page_cache(page, mapping, hindex, |
fed400a1 | 1594 | NULL); |
b065b432 HD |
1595 | radix_tree_preload_end(); |
1596 | } | |
1597 | if (error) { | |
800d8c63 KS |
1598 | mem_cgroup_cancel_charge(page, memcg, |
1599 | PageTransHuge(page)); | |
1600 | goto unacct; | |
b065b432 | 1601 | } |
800d8c63 KS |
1602 | mem_cgroup_commit_charge(page, memcg, false, |
1603 | PageTransHuge(page)); | |
54af6042 HD |
1604 | lru_cache_add_anon(page); |
1605 | ||
1606 | spin_lock(&info->lock); | |
800d8c63 KS |
1607 | info->alloced += 1 << compound_order(page); |
1608 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); | |
54af6042 | 1609 | shmem_recalc_inode(inode); |
1da177e4 | 1610 | spin_unlock(&info->lock); |
1635f6a7 | 1611 | alloced = true; |
54af6042 | 1612 | |
ec9516fb | 1613 | /* |
1635f6a7 HD |
1614 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1615 | */ | |
1616 | if (sgp == SGP_FALLOC) | |
1617 | sgp = SGP_WRITE; | |
1618 | clear: | |
1619 | /* | |
1620 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1621 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1622 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb | 1623 | */ |
800d8c63 KS |
1624 | if (sgp != SGP_WRITE && !PageUptodate(page)) { |
1625 | struct page *head = compound_head(page); | |
1626 | int i; | |
1627 | ||
1628 | for (i = 0; i < (1 << compound_order(head)); i++) { | |
1629 | clear_highpage(head + i); | |
1630 | flush_dcache_page(head + i); | |
1631 | } | |
1632 | SetPageUptodate(head); | |
ec9516fb | 1633 | } |
1da177e4 | 1634 | } |
bde05d1c | 1635 | |
54af6042 | 1636 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1637 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1638 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1639 | if (alloced) { |
1640 | ClearPageDirty(page); | |
1641 | delete_from_page_cache(page); | |
1642 | spin_lock(&info->lock); | |
1643 | shmem_recalc_inode(inode); | |
1644 | spin_unlock(&info->lock); | |
1645 | } | |
54af6042 | 1646 | error = -EINVAL; |
267a4c76 | 1647 | goto unlock; |
e83c32e8 | 1648 | } |
800d8c63 | 1649 | *pagep = page + index - hindex; |
54af6042 | 1650 | return 0; |
1da177e4 | 1651 | |
59a16ead | 1652 | /* |
54af6042 | 1653 | * Error recovery. |
59a16ead | 1654 | */ |
54af6042 | 1655 | unacct: |
800d8c63 KS |
1656 | if (sbinfo->max_blocks) |
1657 | percpu_counter_sub(&sbinfo->used_blocks, | |
1658 | 1 << compound_order(page)); | |
1659 | shmem_unacct_blocks(info->flags, 1 << compound_order(page)); | |
1660 | ||
1661 | if (PageTransHuge(page)) { | |
1662 | unlock_page(page); | |
1663 | put_page(page); | |
1664 | goto alloc_nohuge; | |
1665 | } | |
54af6042 | 1666 | failed: |
267a4c76 | 1667 | if (swap.val && !shmem_confirm_swap(mapping, index, swap)) |
d1899228 HD |
1668 | error = -EEXIST; |
1669 | unlock: | |
27ab7006 | 1670 | if (page) { |
54af6042 | 1671 | unlock_page(page); |
09cbfeaf | 1672 | put_page(page); |
54af6042 HD |
1673 | } |
1674 | if (error == -ENOSPC && !once++) { | |
1675 | info = SHMEM_I(inode); | |
1676 | spin_lock(&info->lock); | |
1677 | shmem_recalc_inode(inode); | |
1678 | spin_unlock(&info->lock); | |
27ab7006 | 1679 | goto repeat; |
ff36b801 | 1680 | } |
d1899228 | 1681 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1682 | goto repeat; |
1683 | return error; | |
1da177e4 LT |
1684 | } |
1685 | ||
d0217ac0 | 1686 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1687 | { |
496ad9aa | 1688 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 1689 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
657e3038 | 1690 | enum sgp_type sgp; |
1da177e4 | 1691 | int error; |
68da9f05 | 1692 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1693 | |
f00cdc6d HD |
1694 | /* |
1695 | * Trinity finds that probing a hole which tmpfs is punching can | |
1696 | * prevent the hole-punch from ever completing: which in turn | |
1697 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1698 | * faulting pages into the hole while it's being punched. Although |
1699 | * shmem_undo_range() does remove the additions, it may be unable to | |
1700 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1701 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1702 | * | |
1703 | * It does not matter if we sometimes reach this check just before the | |
1704 | * hole-punch begins, so that one fault then races with the punch: | |
1705 | * we just need to make racing faults a rare case. | |
1706 | * | |
1707 | * The implementation below would be much simpler if we just used a | |
1708 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1709 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1710 | */ |
1711 | if (unlikely(inode->i_private)) { | |
1712 | struct shmem_falloc *shmem_falloc; | |
1713 | ||
1714 | spin_lock(&inode->i_lock); | |
1715 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1716 | if (shmem_falloc && |
1717 | shmem_falloc->waitq && | |
1718 | vmf->pgoff >= shmem_falloc->start && | |
1719 | vmf->pgoff < shmem_falloc->next) { | |
1720 | wait_queue_head_t *shmem_falloc_waitq; | |
1721 | DEFINE_WAIT(shmem_fault_wait); | |
1722 | ||
1723 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1724 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1725 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1726 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1727 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1728 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1729 | } |
8e205f77 HD |
1730 | |
1731 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1732 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1733 | TASK_UNINTERRUPTIBLE); | |
1734 | spin_unlock(&inode->i_lock); | |
1735 | schedule(); | |
1736 | ||
1737 | /* | |
1738 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1739 | * stack of the hole-punching task: shmem_falloc_waitq | |
1740 | * is usually invalid by the time we reach here, but | |
1741 | * finish_wait() does not dereference it in that case; | |
1742 | * though i_lock needed lest racing with wake_up_all(). | |
1743 | */ | |
1744 | spin_lock(&inode->i_lock); | |
1745 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1746 | spin_unlock(&inode->i_lock); | |
1747 | return ret; | |
f00cdc6d | 1748 | } |
8e205f77 | 1749 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1750 | } |
1751 | ||
657e3038 KS |
1752 | sgp = SGP_CACHE; |
1753 | if (vma->vm_flags & VM_HUGEPAGE) | |
1754 | sgp = SGP_HUGE; | |
1755 | else if (vma->vm_flags & VM_NOHUGEPAGE) | |
1756 | sgp = SGP_NOHUGE; | |
1757 | ||
1758 | error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp, | |
9e18eb29 | 1759 | gfp, vma->vm_mm, &ret); |
d0217ac0 NP |
1760 | if (error) |
1761 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1762 | return ret; |
1da177e4 LT |
1763 | } |
1764 | ||
c01d5b30 HD |
1765 | unsigned long shmem_get_unmapped_area(struct file *file, |
1766 | unsigned long uaddr, unsigned long len, | |
1767 | unsigned long pgoff, unsigned long flags) | |
1768 | { | |
1769 | unsigned long (*get_area)(struct file *, | |
1770 | unsigned long, unsigned long, unsigned long, unsigned long); | |
1771 | unsigned long addr; | |
1772 | unsigned long offset; | |
1773 | unsigned long inflated_len; | |
1774 | unsigned long inflated_addr; | |
1775 | unsigned long inflated_offset; | |
1776 | ||
1777 | if (len > TASK_SIZE) | |
1778 | return -ENOMEM; | |
1779 | ||
1780 | get_area = current->mm->get_unmapped_area; | |
1781 | addr = get_area(file, uaddr, len, pgoff, flags); | |
1782 | ||
1783 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) | |
1784 | return addr; | |
1785 | if (IS_ERR_VALUE(addr)) | |
1786 | return addr; | |
1787 | if (addr & ~PAGE_MASK) | |
1788 | return addr; | |
1789 | if (addr > TASK_SIZE - len) | |
1790 | return addr; | |
1791 | ||
1792 | if (shmem_huge == SHMEM_HUGE_DENY) | |
1793 | return addr; | |
1794 | if (len < HPAGE_PMD_SIZE) | |
1795 | return addr; | |
1796 | if (flags & MAP_FIXED) | |
1797 | return addr; | |
1798 | /* | |
1799 | * Our priority is to support MAP_SHARED mapped hugely; | |
1800 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
1801 | * But if caller specified an address hint, respect that as before. | |
1802 | */ | |
1803 | if (uaddr) | |
1804 | return addr; | |
1805 | ||
1806 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
1807 | struct super_block *sb; | |
1808 | ||
1809 | if (file) { | |
1810 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
1811 | sb = file_inode(file)->i_sb; | |
1812 | } else { | |
1813 | /* | |
1814 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
1815 | * for "/dev/zero", to create a shared anonymous object. | |
1816 | */ | |
1817 | if (IS_ERR(shm_mnt)) | |
1818 | return addr; | |
1819 | sb = shm_mnt->mnt_sb; | |
1820 | } | |
1821 | if (SHMEM_SB(sb)->huge != SHMEM_HUGE_NEVER) | |
1822 | return addr; | |
1823 | } | |
1824 | ||
1825 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
1826 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
1827 | return addr; | |
1828 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
1829 | return addr; | |
1830 | ||
1831 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
1832 | if (inflated_len > TASK_SIZE) | |
1833 | return addr; | |
1834 | if (inflated_len < len) | |
1835 | return addr; | |
1836 | ||
1837 | inflated_addr = get_area(NULL, 0, inflated_len, 0, flags); | |
1838 | if (IS_ERR_VALUE(inflated_addr)) | |
1839 | return addr; | |
1840 | if (inflated_addr & ~PAGE_MASK) | |
1841 | return addr; | |
1842 | ||
1843 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
1844 | inflated_addr += offset - inflated_offset; | |
1845 | if (inflated_offset > offset) | |
1846 | inflated_addr += HPAGE_PMD_SIZE; | |
1847 | ||
1848 | if (inflated_addr > TASK_SIZE - len) | |
1849 | return addr; | |
1850 | return inflated_addr; | |
1851 | } | |
1852 | ||
1da177e4 | 1853 | #ifdef CONFIG_NUMA |
41ffe5d5 | 1854 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 1855 | { |
496ad9aa | 1856 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1857 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
1858 | } |
1859 | ||
d8dc74f2 AB |
1860 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1861 | unsigned long addr) | |
1da177e4 | 1862 | { |
496ad9aa | 1863 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1864 | pgoff_t index; |
1da177e4 | 1865 | |
41ffe5d5 HD |
1866 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
1867 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
1868 | } |
1869 | #endif | |
1870 | ||
1871 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1872 | { | |
496ad9aa | 1873 | struct inode *inode = file_inode(file); |
1da177e4 LT |
1874 | struct shmem_inode_info *info = SHMEM_I(inode); |
1875 | int retval = -ENOMEM; | |
1876 | ||
1877 | spin_lock(&info->lock); | |
1878 | if (lock && !(info->flags & VM_LOCKED)) { | |
1879 | if (!user_shm_lock(inode->i_size, user)) | |
1880 | goto out_nomem; | |
1881 | info->flags |= VM_LOCKED; | |
89e004ea | 1882 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
1883 | } |
1884 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1885 | user_shm_unlock(inode->i_size, user); | |
1886 | info->flags &= ~VM_LOCKED; | |
89e004ea | 1887 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
1888 | } |
1889 | retval = 0; | |
89e004ea | 1890 | |
1da177e4 LT |
1891 | out_nomem: |
1892 | spin_unlock(&info->lock); | |
1893 | return retval; | |
1894 | } | |
1895 | ||
9b83a6a8 | 1896 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1897 | { |
1898 | file_accessed(file); | |
1899 | vma->vm_ops = &shmem_vm_ops; | |
1900 | return 0; | |
1901 | } | |
1902 | ||
454abafe | 1903 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 1904 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
1905 | { |
1906 | struct inode *inode; | |
1907 | struct shmem_inode_info *info; | |
1908 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1909 | ||
5b04c689 PE |
1910 | if (shmem_reserve_inode(sb)) |
1911 | return NULL; | |
1da177e4 LT |
1912 | |
1913 | inode = new_inode(sb); | |
1914 | if (inode) { | |
85fe4025 | 1915 | inode->i_ino = get_next_ino(); |
454abafe | 1916 | inode_init_owner(inode, dir, mode); |
1da177e4 | 1917 | inode->i_blocks = 0; |
1da177e4 | 1918 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
91828a40 | 1919 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1920 | info = SHMEM_I(inode); |
1921 | memset(info, 0, (char *)inode - (char *)info); | |
1922 | spin_lock_init(&info->lock); | |
40e041a2 | 1923 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 1924 | info->flags = flags & VM_NORESERVE; |
1da177e4 | 1925 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 1926 | simple_xattrs_init(&info->xattrs); |
72c04902 | 1927 | cache_no_acl(inode); |
1da177e4 LT |
1928 | |
1929 | switch (mode & S_IFMT) { | |
1930 | default: | |
39f0247d | 1931 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1932 | init_special_inode(inode, mode, dev); |
1933 | break; | |
1934 | case S_IFREG: | |
14fcc23f | 1935 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1936 | inode->i_op = &shmem_inode_operations; |
1937 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
1938 | mpol_shared_policy_init(&info->policy, |
1939 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
1940 | break; |
1941 | case S_IFDIR: | |
d8c76e6f | 1942 | inc_nlink(inode); |
1da177e4 LT |
1943 | /* Some things misbehave if size == 0 on a directory */ |
1944 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1945 | inode->i_op = &shmem_dir_inode_operations; | |
1946 | inode->i_fop = &simple_dir_operations; | |
1947 | break; | |
1948 | case S_IFLNK: | |
1949 | /* | |
1950 | * Must not load anything in the rbtree, | |
1951 | * mpol_free_shared_policy will not be called. | |
1952 | */ | |
71fe804b | 1953 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
1954 | break; |
1955 | } | |
5b04c689 PE |
1956 | } else |
1957 | shmem_free_inode(sb); | |
1da177e4 LT |
1958 | return inode; |
1959 | } | |
1960 | ||
0cd6144a JW |
1961 | bool shmem_mapping(struct address_space *mapping) |
1962 | { | |
f0774d88 SL |
1963 | if (!mapping->host) |
1964 | return false; | |
1965 | ||
97b713ba | 1966 | return mapping->host->i_sb->s_op == &shmem_ops; |
0cd6144a JW |
1967 | } |
1968 | ||
1da177e4 | 1969 | #ifdef CONFIG_TMPFS |
92e1d5be | 1970 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 1971 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 1972 | |
6d9d88d0 JS |
1973 | #ifdef CONFIG_TMPFS_XATTR |
1974 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
1975 | #else | |
1976 | #define shmem_initxattrs NULL | |
1977 | #endif | |
1978 | ||
1da177e4 | 1979 | static int |
800d15a5 NP |
1980 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1981 | loff_t pos, unsigned len, unsigned flags, | |
1982 | struct page **pagep, void **fsdata) | |
1da177e4 | 1983 | { |
800d15a5 | 1984 | struct inode *inode = mapping->host; |
40e041a2 | 1985 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 1986 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
1987 | |
1988 | /* i_mutex is held by caller */ | |
1989 | if (unlikely(info->seals)) { | |
1990 | if (info->seals & F_SEAL_WRITE) | |
1991 | return -EPERM; | |
1992 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
1993 | return -EPERM; | |
1994 | } | |
1995 | ||
9e18eb29 | 1996 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
1997 | } |
1998 | ||
1999 | static int | |
2000 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2001 | loff_t pos, unsigned len, unsigned copied, | |
2002 | struct page *page, void *fsdata) | |
2003 | { | |
2004 | struct inode *inode = mapping->host; | |
2005 | ||
d3602444 HD |
2006 | if (pos + copied > inode->i_size) |
2007 | i_size_write(inode, pos + copied); | |
2008 | ||
ec9516fb | 2009 | if (!PageUptodate(page)) { |
800d8c63 KS |
2010 | struct page *head = compound_head(page); |
2011 | if (PageTransCompound(page)) { | |
2012 | int i; | |
2013 | ||
2014 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2015 | if (head + i == page) | |
2016 | continue; | |
2017 | clear_highpage(head + i); | |
2018 | flush_dcache_page(head + i); | |
2019 | } | |
2020 | } | |
09cbfeaf KS |
2021 | if (copied < PAGE_SIZE) { |
2022 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2023 | zero_user_segments(page, 0, from, |
09cbfeaf | 2024 | from + copied, PAGE_SIZE); |
ec9516fb | 2025 | } |
800d8c63 | 2026 | SetPageUptodate(head); |
ec9516fb | 2027 | } |
800d15a5 | 2028 | set_page_dirty(page); |
6746aff7 | 2029 | unlock_page(page); |
09cbfeaf | 2030 | put_page(page); |
800d15a5 | 2031 | |
800d15a5 | 2032 | return copied; |
1da177e4 LT |
2033 | } |
2034 | ||
2ba5bbed | 2035 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2036 | { |
6e58e79d AV |
2037 | struct file *file = iocb->ki_filp; |
2038 | struct inode *inode = file_inode(file); | |
1da177e4 | 2039 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2040 | pgoff_t index; |
2041 | unsigned long offset; | |
a0ee5ec5 | 2042 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2043 | int error = 0; |
cb66a7a1 | 2044 | ssize_t retval = 0; |
6e58e79d | 2045 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2046 | |
2047 | /* | |
2048 | * Might this read be for a stacking filesystem? Then when reading | |
2049 | * holes of a sparse file, we actually need to allocate those pages, | |
2050 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2051 | */ | |
777eda2c | 2052 | if (!iter_is_iovec(to)) |
75edd345 | 2053 | sgp = SGP_CACHE; |
1da177e4 | 2054 | |
09cbfeaf KS |
2055 | index = *ppos >> PAGE_SHIFT; |
2056 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2057 | |
2058 | for (;;) { | |
2059 | struct page *page = NULL; | |
41ffe5d5 HD |
2060 | pgoff_t end_index; |
2061 | unsigned long nr, ret; | |
1da177e4 LT |
2062 | loff_t i_size = i_size_read(inode); |
2063 | ||
09cbfeaf | 2064 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2065 | if (index > end_index) |
2066 | break; | |
2067 | if (index == end_index) { | |
09cbfeaf | 2068 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2069 | if (nr <= offset) |
2070 | break; | |
2071 | } | |
2072 | ||
9e18eb29 | 2073 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2074 | if (error) { |
2075 | if (error == -EINVAL) | |
2076 | error = 0; | |
1da177e4 LT |
2077 | break; |
2078 | } | |
75edd345 HD |
2079 | if (page) { |
2080 | if (sgp == SGP_CACHE) | |
2081 | set_page_dirty(page); | |
d3602444 | 2082 | unlock_page(page); |
75edd345 | 2083 | } |
1da177e4 LT |
2084 | |
2085 | /* | |
2086 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2087 | * are called without i_mutex protection against truncate |
1da177e4 | 2088 | */ |
09cbfeaf | 2089 | nr = PAGE_SIZE; |
1da177e4 | 2090 | i_size = i_size_read(inode); |
09cbfeaf | 2091 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2092 | if (index == end_index) { |
09cbfeaf | 2093 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2094 | if (nr <= offset) { |
2095 | if (page) | |
09cbfeaf | 2096 | put_page(page); |
1da177e4 LT |
2097 | break; |
2098 | } | |
2099 | } | |
2100 | nr -= offset; | |
2101 | ||
2102 | if (page) { | |
2103 | /* | |
2104 | * If users can be writing to this page using arbitrary | |
2105 | * virtual addresses, take care about potential aliasing | |
2106 | * before reading the page on the kernel side. | |
2107 | */ | |
2108 | if (mapping_writably_mapped(mapping)) | |
2109 | flush_dcache_page(page); | |
2110 | /* | |
2111 | * Mark the page accessed if we read the beginning. | |
2112 | */ | |
2113 | if (!offset) | |
2114 | mark_page_accessed(page); | |
b5810039 | 2115 | } else { |
1da177e4 | 2116 | page = ZERO_PAGE(0); |
09cbfeaf | 2117 | get_page(page); |
b5810039 | 2118 | } |
1da177e4 LT |
2119 | |
2120 | /* | |
2121 | * Ok, we have the page, and it's up-to-date, so | |
2122 | * now we can copy it to user space... | |
1da177e4 | 2123 | */ |
2ba5bbed | 2124 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2125 | retval += ret; |
1da177e4 | 2126 | offset += ret; |
09cbfeaf KS |
2127 | index += offset >> PAGE_SHIFT; |
2128 | offset &= ~PAGE_MASK; | |
1da177e4 | 2129 | |
09cbfeaf | 2130 | put_page(page); |
2ba5bbed | 2131 | if (!iov_iter_count(to)) |
1da177e4 | 2132 | break; |
6e58e79d AV |
2133 | if (ret < nr) { |
2134 | error = -EFAULT; | |
2135 | break; | |
2136 | } | |
1da177e4 LT |
2137 | cond_resched(); |
2138 | } | |
2139 | ||
09cbfeaf | 2140 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2141 | file_accessed(file); |
2142 | return retval ? retval : error; | |
1da177e4 LT |
2143 | } |
2144 | ||
708e3508 HD |
2145 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, |
2146 | struct pipe_inode_info *pipe, size_t len, | |
2147 | unsigned int flags) | |
2148 | { | |
2149 | struct address_space *mapping = in->f_mapping; | |
71f0e07a | 2150 | struct inode *inode = mapping->host; |
708e3508 HD |
2151 | unsigned int loff, nr_pages, req_pages; |
2152 | struct page *pages[PIPE_DEF_BUFFERS]; | |
2153 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
2154 | struct page *page; | |
2155 | pgoff_t index, end_index; | |
2156 | loff_t isize, left; | |
2157 | int error, page_nr; | |
2158 | struct splice_pipe_desc spd = { | |
2159 | .pages = pages, | |
2160 | .partial = partial, | |
047fe360 | 2161 | .nr_pages_max = PIPE_DEF_BUFFERS, |
708e3508 HD |
2162 | .flags = flags, |
2163 | .ops = &page_cache_pipe_buf_ops, | |
2164 | .spd_release = spd_release_page, | |
2165 | }; | |
2166 | ||
71f0e07a | 2167 | isize = i_size_read(inode); |
708e3508 HD |
2168 | if (unlikely(*ppos >= isize)) |
2169 | return 0; | |
2170 | ||
2171 | left = isize - *ppos; | |
2172 | if (unlikely(left < len)) | |
2173 | len = left; | |
2174 | ||
2175 | if (splice_grow_spd(pipe, &spd)) | |
2176 | return -ENOMEM; | |
2177 | ||
09cbfeaf KS |
2178 | index = *ppos >> PAGE_SHIFT; |
2179 | loff = *ppos & ~PAGE_MASK; | |
2180 | req_pages = (len + loff + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
a786c06d | 2181 | nr_pages = min(req_pages, spd.nr_pages_max); |
708e3508 | 2182 | |
708e3508 HD |
2183 | spd.nr_pages = find_get_pages_contig(mapping, index, |
2184 | nr_pages, spd.pages); | |
2185 | index += spd.nr_pages; | |
708e3508 | 2186 | error = 0; |
708e3508 | 2187 | |
71f0e07a | 2188 | while (spd.nr_pages < nr_pages) { |
9e18eb29 | 2189 | error = shmem_getpage(inode, index, &page, SGP_CACHE); |
71f0e07a HD |
2190 | if (error) |
2191 | break; | |
2192 | unlock_page(page); | |
708e3508 HD |
2193 | spd.pages[spd.nr_pages++] = page; |
2194 | index++; | |
2195 | } | |
2196 | ||
09cbfeaf | 2197 | index = *ppos >> PAGE_SHIFT; |
708e3508 HD |
2198 | nr_pages = spd.nr_pages; |
2199 | spd.nr_pages = 0; | |
71f0e07a | 2200 | |
708e3508 HD |
2201 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { |
2202 | unsigned int this_len; | |
2203 | ||
2204 | if (!len) | |
2205 | break; | |
2206 | ||
09cbfeaf | 2207 | this_len = min_t(unsigned long, len, PAGE_SIZE - loff); |
708e3508 HD |
2208 | page = spd.pages[page_nr]; |
2209 | ||
71f0e07a | 2210 | if (!PageUptodate(page) || page->mapping != mapping) { |
9e18eb29 | 2211 | error = shmem_getpage(inode, index, &page, SGP_CACHE); |
71f0e07a | 2212 | if (error) |
708e3508 | 2213 | break; |
71f0e07a | 2214 | unlock_page(page); |
09cbfeaf | 2215 | put_page(spd.pages[page_nr]); |
71f0e07a | 2216 | spd.pages[page_nr] = page; |
708e3508 | 2217 | } |
71f0e07a HD |
2218 | |
2219 | isize = i_size_read(inode); | |
09cbfeaf | 2220 | end_index = (isize - 1) >> PAGE_SHIFT; |
708e3508 HD |
2221 | if (unlikely(!isize || index > end_index)) |
2222 | break; | |
2223 | ||
708e3508 HD |
2224 | if (end_index == index) { |
2225 | unsigned int plen; | |
2226 | ||
09cbfeaf | 2227 | plen = ((isize - 1) & ~PAGE_MASK) + 1; |
708e3508 HD |
2228 | if (plen <= loff) |
2229 | break; | |
2230 | ||
708e3508 HD |
2231 | this_len = min(this_len, plen - loff); |
2232 | len = this_len; | |
2233 | } | |
2234 | ||
2235 | spd.partial[page_nr].offset = loff; | |
2236 | spd.partial[page_nr].len = this_len; | |
2237 | len -= this_len; | |
2238 | loff = 0; | |
2239 | spd.nr_pages++; | |
2240 | index++; | |
2241 | } | |
2242 | ||
708e3508 | 2243 | while (page_nr < nr_pages) |
09cbfeaf | 2244 | put_page(spd.pages[page_nr++]); |
708e3508 HD |
2245 | |
2246 | if (spd.nr_pages) | |
2247 | error = splice_to_pipe(pipe, &spd); | |
2248 | ||
047fe360 | 2249 | splice_shrink_spd(&spd); |
708e3508 HD |
2250 | |
2251 | if (error > 0) { | |
2252 | *ppos += error; | |
2253 | file_accessed(in); | |
2254 | } | |
2255 | return error; | |
2256 | } | |
2257 | ||
220f2ac9 HD |
2258 | /* |
2259 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
2260 | */ | |
2261 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2262 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2263 | { |
2264 | struct page *page; | |
2265 | struct pagevec pvec; | |
2266 | pgoff_t indices[PAGEVEC_SIZE]; | |
2267 | bool done = false; | |
2268 | int i; | |
2269 | ||
2270 | pagevec_init(&pvec, 0); | |
2271 | pvec.nr = 1; /* start small: we may be there already */ | |
2272 | while (!done) { | |
0cd6144a | 2273 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2274 | pvec.nr, pvec.pages, indices); |
2275 | if (!pvec.nr) { | |
965c8e59 | 2276 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2277 | index = end; |
2278 | break; | |
2279 | } | |
2280 | for (i = 0; i < pvec.nr; i++, index++) { | |
2281 | if (index < indices[i]) { | |
965c8e59 | 2282 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2283 | done = true; |
2284 | break; | |
2285 | } | |
2286 | index = indices[i]; | |
2287 | } | |
2288 | page = pvec.pages[i]; | |
2289 | if (page && !radix_tree_exceptional_entry(page)) { | |
2290 | if (!PageUptodate(page)) | |
2291 | page = NULL; | |
2292 | } | |
2293 | if (index >= end || | |
965c8e59 AM |
2294 | (page && whence == SEEK_DATA) || |
2295 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2296 | done = true; |
2297 | break; | |
2298 | } | |
2299 | } | |
0cd6144a | 2300 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2301 | pagevec_release(&pvec); |
2302 | pvec.nr = PAGEVEC_SIZE; | |
2303 | cond_resched(); | |
2304 | } | |
2305 | return index; | |
2306 | } | |
2307 | ||
965c8e59 | 2308 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2309 | { |
2310 | struct address_space *mapping = file->f_mapping; | |
2311 | struct inode *inode = mapping->host; | |
2312 | pgoff_t start, end; | |
2313 | loff_t new_offset; | |
2314 | ||
965c8e59 AM |
2315 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2316 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2317 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2318 | inode_lock(inode); |
220f2ac9 HD |
2319 | /* We're holding i_mutex so we can access i_size directly */ |
2320 | ||
2321 | if (offset < 0) | |
2322 | offset = -EINVAL; | |
2323 | else if (offset >= inode->i_size) | |
2324 | offset = -ENXIO; | |
2325 | else { | |
09cbfeaf KS |
2326 | start = offset >> PAGE_SHIFT; |
2327 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2328 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2329 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2330 | if (new_offset > offset) { |
2331 | if (new_offset < inode->i_size) | |
2332 | offset = new_offset; | |
965c8e59 | 2333 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2334 | offset = -ENXIO; |
2335 | else | |
2336 | offset = inode->i_size; | |
2337 | } | |
2338 | } | |
2339 | ||
387aae6f HD |
2340 | if (offset >= 0) |
2341 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2342 | inode_unlock(inode); |
220f2ac9 HD |
2343 | return offset; |
2344 | } | |
2345 | ||
05f65b5c DH |
2346 | /* |
2347 | * We need a tag: a new tag would expand every radix_tree_node by 8 bytes, | |
2348 | * so reuse a tag which we firmly believe is never set or cleared on shmem. | |
2349 | */ | |
2350 | #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE | |
2351 | #define LAST_SCAN 4 /* about 150ms max */ | |
2352 | ||
2353 | static void shmem_tag_pins(struct address_space *mapping) | |
2354 | { | |
2355 | struct radix_tree_iter iter; | |
2356 | void **slot; | |
2357 | pgoff_t start; | |
2358 | struct page *page; | |
2359 | ||
2360 | lru_add_drain(); | |
2361 | start = 0; | |
2362 | rcu_read_lock(); | |
2363 | ||
05f65b5c DH |
2364 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
2365 | page = radix_tree_deref_slot(slot); | |
2366 | if (!page || radix_tree_exception(page)) { | |
2cf938aa MW |
2367 | if (radix_tree_deref_retry(page)) { |
2368 | slot = radix_tree_iter_retry(&iter); | |
2369 | continue; | |
2370 | } | |
05f65b5c DH |
2371 | } else if (page_count(page) - page_mapcount(page) > 1) { |
2372 | spin_lock_irq(&mapping->tree_lock); | |
2373 | radix_tree_tag_set(&mapping->page_tree, iter.index, | |
2374 | SHMEM_TAG_PINNED); | |
2375 | spin_unlock_irq(&mapping->tree_lock); | |
2376 | } | |
2377 | ||
2378 | if (need_resched()) { | |
2379 | cond_resched_rcu(); | |
7165092f | 2380 | slot = radix_tree_iter_next(&iter); |
05f65b5c DH |
2381 | } |
2382 | } | |
2383 | rcu_read_unlock(); | |
2384 | } | |
2385 | ||
2386 | /* | |
2387 | * Setting SEAL_WRITE requires us to verify there's no pending writer. However, | |
2388 | * via get_user_pages(), drivers might have some pending I/O without any active | |
2389 | * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages | |
2390 | * and see whether it has an elevated ref-count. If so, we tag them and wait for | |
2391 | * them to be dropped. | |
2392 | * The caller must guarantee that no new user will acquire writable references | |
2393 | * to those pages to avoid races. | |
2394 | */ | |
40e041a2 DH |
2395 | static int shmem_wait_for_pins(struct address_space *mapping) |
2396 | { | |
05f65b5c DH |
2397 | struct radix_tree_iter iter; |
2398 | void **slot; | |
2399 | pgoff_t start; | |
2400 | struct page *page; | |
2401 | int error, scan; | |
2402 | ||
2403 | shmem_tag_pins(mapping); | |
2404 | ||
2405 | error = 0; | |
2406 | for (scan = 0; scan <= LAST_SCAN; scan++) { | |
2407 | if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED)) | |
2408 | break; | |
2409 | ||
2410 | if (!scan) | |
2411 | lru_add_drain_all(); | |
2412 | else if (schedule_timeout_killable((HZ << scan) / 200)) | |
2413 | scan = LAST_SCAN; | |
2414 | ||
2415 | start = 0; | |
2416 | rcu_read_lock(); | |
05f65b5c DH |
2417 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, |
2418 | start, SHMEM_TAG_PINNED) { | |
2419 | ||
2420 | page = radix_tree_deref_slot(slot); | |
2421 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2422 | if (radix_tree_deref_retry(page)) { |
2423 | slot = radix_tree_iter_retry(&iter); | |
2424 | continue; | |
2425 | } | |
05f65b5c DH |
2426 | |
2427 | page = NULL; | |
2428 | } | |
2429 | ||
2430 | if (page && | |
2431 | page_count(page) - page_mapcount(page) != 1) { | |
2432 | if (scan < LAST_SCAN) | |
2433 | goto continue_resched; | |
2434 | ||
2435 | /* | |
2436 | * On the last scan, we clean up all those tags | |
2437 | * we inserted; but make a note that we still | |
2438 | * found pages pinned. | |
2439 | */ | |
2440 | error = -EBUSY; | |
2441 | } | |
2442 | ||
2443 | spin_lock_irq(&mapping->tree_lock); | |
2444 | radix_tree_tag_clear(&mapping->page_tree, | |
2445 | iter.index, SHMEM_TAG_PINNED); | |
2446 | spin_unlock_irq(&mapping->tree_lock); | |
2447 | continue_resched: | |
2448 | if (need_resched()) { | |
2449 | cond_resched_rcu(); | |
7165092f | 2450 | slot = radix_tree_iter_next(&iter); |
05f65b5c DH |
2451 | } |
2452 | } | |
2453 | rcu_read_unlock(); | |
2454 | } | |
2455 | ||
2456 | return error; | |
40e041a2 DH |
2457 | } |
2458 | ||
2459 | #define F_ALL_SEALS (F_SEAL_SEAL | \ | |
2460 | F_SEAL_SHRINK | \ | |
2461 | F_SEAL_GROW | \ | |
2462 | F_SEAL_WRITE) | |
2463 | ||
2464 | int shmem_add_seals(struct file *file, unsigned int seals) | |
2465 | { | |
2466 | struct inode *inode = file_inode(file); | |
2467 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2468 | int error; | |
2469 | ||
2470 | /* | |
2471 | * SEALING | |
2472 | * Sealing allows multiple parties to share a shmem-file but restrict | |
2473 | * access to a specific subset of file operations. Seals can only be | |
2474 | * added, but never removed. This way, mutually untrusted parties can | |
2475 | * share common memory regions with a well-defined policy. A malicious | |
2476 | * peer can thus never perform unwanted operations on a shared object. | |
2477 | * | |
2478 | * Seals are only supported on special shmem-files and always affect | |
2479 | * the whole underlying inode. Once a seal is set, it may prevent some | |
2480 | * kinds of access to the file. Currently, the following seals are | |
2481 | * defined: | |
2482 | * SEAL_SEAL: Prevent further seals from being set on this file | |
2483 | * SEAL_SHRINK: Prevent the file from shrinking | |
2484 | * SEAL_GROW: Prevent the file from growing | |
2485 | * SEAL_WRITE: Prevent write access to the file | |
2486 | * | |
2487 | * As we don't require any trust relationship between two parties, we | |
2488 | * must prevent seals from being removed. Therefore, sealing a file | |
2489 | * only adds a given set of seals to the file, it never touches | |
2490 | * existing seals. Furthermore, the "setting seals"-operation can be | |
2491 | * sealed itself, which basically prevents any further seal from being | |
2492 | * added. | |
2493 | * | |
2494 | * Semantics of sealing are only defined on volatile files. Only | |
2495 | * anonymous shmem files support sealing. More importantly, seals are | |
2496 | * never written to disk. Therefore, there's no plan to support it on | |
2497 | * other file types. | |
2498 | */ | |
2499 | ||
2500 | if (file->f_op != &shmem_file_operations) | |
2501 | return -EINVAL; | |
2502 | if (!(file->f_mode & FMODE_WRITE)) | |
2503 | return -EPERM; | |
2504 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
2505 | return -EINVAL; | |
2506 | ||
5955102c | 2507 | inode_lock(inode); |
40e041a2 DH |
2508 | |
2509 | if (info->seals & F_SEAL_SEAL) { | |
2510 | error = -EPERM; | |
2511 | goto unlock; | |
2512 | } | |
2513 | ||
2514 | if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) { | |
2515 | error = mapping_deny_writable(file->f_mapping); | |
2516 | if (error) | |
2517 | goto unlock; | |
2518 | ||
2519 | error = shmem_wait_for_pins(file->f_mapping); | |
2520 | if (error) { | |
2521 | mapping_allow_writable(file->f_mapping); | |
2522 | goto unlock; | |
2523 | } | |
2524 | } | |
2525 | ||
2526 | info->seals |= seals; | |
2527 | error = 0; | |
2528 | ||
2529 | unlock: | |
5955102c | 2530 | inode_unlock(inode); |
40e041a2 DH |
2531 | return error; |
2532 | } | |
2533 | EXPORT_SYMBOL_GPL(shmem_add_seals); | |
2534 | ||
2535 | int shmem_get_seals(struct file *file) | |
2536 | { | |
2537 | if (file->f_op != &shmem_file_operations) | |
2538 | return -EINVAL; | |
2539 | ||
2540 | return SHMEM_I(file_inode(file))->seals; | |
2541 | } | |
2542 | EXPORT_SYMBOL_GPL(shmem_get_seals); | |
2543 | ||
2544 | long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg) | |
2545 | { | |
2546 | long error; | |
2547 | ||
2548 | switch (cmd) { | |
2549 | case F_ADD_SEALS: | |
2550 | /* disallow upper 32bit */ | |
2551 | if (arg > UINT_MAX) | |
2552 | return -EINVAL; | |
2553 | ||
2554 | error = shmem_add_seals(file, arg); | |
2555 | break; | |
2556 | case F_GET_SEALS: | |
2557 | error = shmem_get_seals(file); | |
2558 | break; | |
2559 | default: | |
2560 | error = -EINVAL; | |
2561 | break; | |
2562 | } | |
2563 | ||
2564 | return error; | |
2565 | } | |
2566 | ||
83e4fa9c HD |
2567 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2568 | loff_t len) | |
2569 | { | |
496ad9aa | 2570 | struct inode *inode = file_inode(file); |
e2d12e22 | 2571 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2572 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2573 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2574 | pgoff_t start, index, end; |
2575 | int error; | |
83e4fa9c | 2576 | |
13ace4d0 HD |
2577 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2578 | return -EOPNOTSUPP; | |
2579 | ||
5955102c | 2580 | inode_lock(inode); |
83e4fa9c HD |
2581 | |
2582 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2583 | struct address_space *mapping = file->f_mapping; | |
2584 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2585 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2586 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2587 | |
40e041a2 DH |
2588 | /* protected by i_mutex */ |
2589 | if (info->seals & F_SEAL_WRITE) { | |
2590 | error = -EPERM; | |
2591 | goto out; | |
2592 | } | |
2593 | ||
8e205f77 | 2594 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
2595 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
2596 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
2597 | spin_lock(&inode->i_lock); | |
2598 | inode->i_private = &shmem_falloc; | |
2599 | spin_unlock(&inode->i_lock); | |
2600 | ||
83e4fa9c HD |
2601 | if ((u64)unmap_end > (u64)unmap_start) |
2602 | unmap_mapping_range(mapping, unmap_start, | |
2603 | 1 + unmap_end - unmap_start, 0); | |
2604 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2605 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2606 | |
2607 | spin_lock(&inode->i_lock); | |
2608 | inode->i_private = NULL; | |
2609 | wake_up_all(&shmem_falloc_waitq); | |
2610 | spin_unlock(&inode->i_lock); | |
83e4fa9c | 2611 | error = 0; |
8e205f77 | 2612 | goto out; |
e2d12e22 HD |
2613 | } |
2614 | ||
2615 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2616 | error = inode_newsize_ok(inode, offset + len); | |
2617 | if (error) | |
2618 | goto out; | |
2619 | ||
40e041a2 DH |
2620 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2621 | error = -EPERM; | |
2622 | goto out; | |
2623 | } | |
2624 | ||
09cbfeaf KS |
2625 | start = offset >> PAGE_SHIFT; |
2626 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2627 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2628 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2629 | error = -ENOSPC; | |
2630 | goto out; | |
83e4fa9c HD |
2631 | } |
2632 | ||
8e205f77 | 2633 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2634 | shmem_falloc.start = start; |
2635 | shmem_falloc.next = start; | |
2636 | shmem_falloc.nr_falloced = 0; | |
2637 | shmem_falloc.nr_unswapped = 0; | |
2638 | spin_lock(&inode->i_lock); | |
2639 | inode->i_private = &shmem_falloc; | |
2640 | spin_unlock(&inode->i_lock); | |
2641 | ||
e2d12e22 HD |
2642 | for (index = start; index < end; index++) { |
2643 | struct page *page; | |
2644 | ||
2645 | /* | |
2646 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2647 | * been interrupted because we are using up too much memory. | |
2648 | */ | |
2649 | if (signal_pending(current)) | |
2650 | error = -EINTR; | |
1aac1400 HD |
2651 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2652 | error = -ENOMEM; | |
e2d12e22 | 2653 | else |
9e18eb29 | 2654 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2655 | if (error) { |
1635f6a7 | 2656 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2657 | if (index > start) { |
2658 | shmem_undo_range(inode, | |
2659 | (loff_t)start << PAGE_SHIFT, | |
2660 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2661 | } | |
1aac1400 | 2662 | goto undone; |
e2d12e22 HD |
2663 | } |
2664 | ||
1aac1400 HD |
2665 | /* |
2666 | * Inform shmem_writepage() how far we have reached. | |
2667 | * No need for lock or barrier: we have the page lock. | |
2668 | */ | |
2669 | shmem_falloc.next++; | |
2670 | if (!PageUptodate(page)) | |
2671 | shmem_falloc.nr_falloced++; | |
2672 | ||
e2d12e22 | 2673 | /* |
1635f6a7 HD |
2674 | * If !PageUptodate, leave it that way so that freeable pages |
2675 | * can be recognized if we need to rollback on error later. | |
2676 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2677 | * than free the pages we are allocating (and SGP_CACHE pages |
2678 | * might still be clean: we now need to mark those dirty too). | |
2679 | */ | |
2680 | set_page_dirty(page); | |
2681 | unlock_page(page); | |
09cbfeaf | 2682 | put_page(page); |
e2d12e22 HD |
2683 | cond_resched(); |
2684 | } | |
2685 | ||
2686 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2687 | i_size_write(inode, offset + len); | |
e2d12e22 | 2688 | inode->i_ctime = CURRENT_TIME; |
1aac1400 HD |
2689 | undone: |
2690 | spin_lock(&inode->i_lock); | |
2691 | inode->i_private = NULL; | |
2692 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2693 | out: |
5955102c | 2694 | inode_unlock(inode); |
83e4fa9c HD |
2695 | return error; |
2696 | } | |
2697 | ||
726c3342 | 2698 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2699 | { |
726c3342 | 2700 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2701 | |
2702 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2703 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2704 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2705 | if (sbinfo->max_blocks) { |
1da177e4 | 2706 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2707 | buf->f_bavail = |
2708 | buf->f_bfree = sbinfo->max_blocks - | |
2709 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2710 | } |
2711 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2712 | buf->f_files = sbinfo->max_inodes; |
2713 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2714 | } |
2715 | /* else leave those fields 0 like simple_statfs */ | |
2716 | return 0; | |
2717 | } | |
2718 | ||
2719 | /* | |
2720 | * File creation. Allocate an inode, and we're done.. | |
2721 | */ | |
2722 | static int | |
1a67aafb | 2723 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2724 | { |
0b0a0806 | 2725 | struct inode *inode; |
1da177e4 LT |
2726 | int error = -ENOSPC; |
2727 | ||
454abafe | 2728 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2729 | if (inode) { |
feda821e CH |
2730 | error = simple_acl_create(dir, inode); |
2731 | if (error) | |
2732 | goto out_iput; | |
2a7dba39 | 2733 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2734 | &dentry->d_name, |
6d9d88d0 | 2735 | shmem_initxattrs, NULL); |
feda821e CH |
2736 | if (error && error != -EOPNOTSUPP) |
2737 | goto out_iput; | |
37ec43cd | 2738 | |
718deb6b | 2739 | error = 0; |
1da177e4 LT |
2740 | dir->i_size += BOGO_DIRENT_SIZE; |
2741 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2742 | d_instantiate(dentry, inode); | |
2743 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2744 | } |
2745 | return error; | |
feda821e CH |
2746 | out_iput: |
2747 | iput(inode); | |
2748 | return error; | |
1da177e4 LT |
2749 | } |
2750 | ||
60545d0d AV |
2751 | static int |
2752 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2753 | { | |
2754 | struct inode *inode; | |
2755 | int error = -ENOSPC; | |
2756 | ||
2757 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
2758 | if (inode) { | |
2759 | error = security_inode_init_security(inode, dir, | |
2760 | NULL, | |
2761 | shmem_initxattrs, NULL); | |
feda821e CH |
2762 | if (error && error != -EOPNOTSUPP) |
2763 | goto out_iput; | |
2764 | error = simple_acl_create(dir, inode); | |
2765 | if (error) | |
2766 | goto out_iput; | |
60545d0d AV |
2767 | d_tmpfile(dentry, inode); |
2768 | } | |
2769 | return error; | |
feda821e CH |
2770 | out_iput: |
2771 | iput(inode); | |
2772 | return error; | |
60545d0d AV |
2773 | } |
2774 | ||
18bb1db3 | 2775 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2776 | { |
2777 | int error; | |
2778 | ||
2779 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
2780 | return error; | |
d8c76e6f | 2781 | inc_nlink(dir); |
1da177e4 LT |
2782 | return 0; |
2783 | } | |
2784 | ||
4acdaf27 | 2785 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2786 | bool excl) |
1da177e4 LT |
2787 | { |
2788 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
2789 | } | |
2790 | ||
2791 | /* | |
2792 | * Link a file.. | |
2793 | */ | |
2794 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2795 | { | |
75c3cfa8 | 2796 | struct inode *inode = d_inode(old_dentry); |
5b04c689 | 2797 | int ret; |
1da177e4 LT |
2798 | |
2799 | /* | |
2800 | * No ordinary (disk based) filesystem counts links as inodes; | |
2801 | * but each new link needs a new dentry, pinning lowmem, and | |
2802 | * tmpfs dentries cannot be pruned until they are unlinked. | |
2803 | */ | |
5b04c689 PE |
2804 | ret = shmem_reserve_inode(inode->i_sb); |
2805 | if (ret) | |
2806 | goto out; | |
1da177e4 LT |
2807 | |
2808 | dir->i_size += BOGO_DIRENT_SIZE; | |
2809 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 2810 | inc_nlink(inode); |
7de9c6ee | 2811 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
2812 | dget(dentry); /* Extra pinning count for the created dentry */ |
2813 | d_instantiate(dentry, inode); | |
5b04c689 PE |
2814 | out: |
2815 | return ret; | |
1da177e4 LT |
2816 | } |
2817 | ||
2818 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
2819 | { | |
75c3cfa8 | 2820 | struct inode *inode = d_inode(dentry); |
1da177e4 | 2821 | |
5b04c689 PE |
2822 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2823 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
2824 | |
2825 | dir->i_size -= BOGO_DIRENT_SIZE; | |
2826 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 2827 | drop_nlink(inode); |
1da177e4 LT |
2828 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
2829 | return 0; | |
2830 | } | |
2831 | ||
2832 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
2833 | { | |
2834 | if (!simple_empty(dentry)) | |
2835 | return -ENOTEMPTY; | |
2836 | ||
75c3cfa8 | 2837 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 2838 | drop_nlink(dir); |
1da177e4 LT |
2839 | return shmem_unlink(dir, dentry); |
2840 | } | |
2841 | ||
37456771 MS |
2842 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
2843 | { | |
e36cb0b8 DH |
2844 | bool old_is_dir = d_is_dir(old_dentry); |
2845 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
2846 | |
2847 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
2848 | if (old_is_dir) { | |
2849 | drop_nlink(old_dir); | |
2850 | inc_nlink(new_dir); | |
2851 | } else { | |
2852 | drop_nlink(new_dir); | |
2853 | inc_nlink(old_dir); | |
2854 | } | |
2855 | } | |
2856 | old_dir->i_ctime = old_dir->i_mtime = | |
2857 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 DH |
2858 | d_inode(old_dentry)->i_ctime = |
2859 | d_inode(new_dentry)->i_ctime = CURRENT_TIME; | |
37456771 MS |
2860 | |
2861 | return 0; | |
2862 | } | |
2863 | ||
46fdb794 MS |
2864 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
2865 | { | |
2866 | struct dentry *whiteout; | |
2867 | int error; | |
2868 | ||
2869 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
2870 | if (!whiteout) | |
2871 | return -ENOMEM; | |
2872 | ||
2873 | error = shmem_mknod(old_dir, whiteout, | |
2874 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
2875 | dput(whiteout); | |
2876 | if (error) | |
2877 | return error; | |
2878 | ||
2879 | /* | |
2880 | * Cheat and hash the whiteout while the old dentry is still in | |
2881 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
2882 | * | |
2883 | * d_lookup() will consistently find one of them at this point, | |
2884 | * not sure which one, but that isn't even important. | |
2885 | */ | |
2886 | d_rehash(whiteout); | |
2887 | return 0; | |
2888 | } | |
2889 | ||
1da177e4 LT |
2890 | /* |
2891 | * The VFS layer already does all the dentry stuff for rename, | |
2892 | * we just have to decrement the usage count for the target if | |
2893 | * it exists so that the VFS layer correctly free's it when it | |
2894 | * gets overwritten. | |
2895 | */ | |
3b69ff51 | 2896 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 2897 | { |
75c3cfa8 | 2898 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
2899 | int they_are_dirs = S_ISDIR(inode->i_mode); |
2900 | ||
46fdb794 | 2901 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
2902 | return -EINVAL; |
2903 | ||
37456771 MS |
2904 | if (flags & RENAME_EXCHANGE) |
2905 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
2906 | ||
1da177e4 LT |
2907 | if (!simple_empty(new_dentry)) |
2908 | return -ENOTEMPTY; | |
2909 | ||
46fdb794 MS |
2910 | if (flags & RENAME_WHITEOUT) { |
2911 | int error; | |
2912 | ||
2913 | error = shmem_whiteout(old_dir, old_dentry); | |
2914 | if (error) | |
2915 | return error; | |
2916 | } | |
2917 | ||
75c3cfa8 | 2918 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 2919 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 2920 | if (they_are_dirs) { |
75c3cfa8 | 2921 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 2922 | drop_nlink(old_dir); |
b928095b | 2923 | } |
1da177e4 | 2924 | } else if (they_are_dirs) { |
9a53c3a7 | 2925 | drop_nlink(old_dir); |
d8c76e6f | 2926 | inc_nlink(new_dir); |
1da177e4 LT |
2927 | } |
2928 | ||
2929 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
2930 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
2931 | old_dir->i_ctime = old_dir->i_mtime = | |
2932 | new_dir->i_ctime = new_dir->i_mtime = | |
2933 | inode->i_ctime = CURRENT_TIME; | |
2934 | return 0; | |
2935 | } | |
2936 | ||
2937 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
2938 | { | |
2939 | int error; | |
2940 | int len; | |
2941 | struct inode *inode; | |
9276aad6 | 2942 | struct page *page; |
1da177e4 LT |
2943 | struct shmem_inode_info *info; |
2944 | ||
2945 | len = strlen(symname) + 1; | |
09cbfeaf | 2946 | if (len > PAGE_SIZE) |
1da177e4 LT |
2947 | return -ENAMETOOLONG; |
2948 | ||
454abafe | 2949 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
2950 | if (!inode) |
2951 | return -ENOSPC; | |
2952 | ||
9d8f13ba | 2953 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 2954 | shmem_initxattrs, NULL); |
570bc1c2 SS |
2955 | if (error) { |
2956 | if (error != -EOPNOTSUPP) { | |
2957 | iput(inode); | |
2958 | return error; | |
2959 | } | |
2960 | error = 0; | |
2961 | } | |
2962 | ||
1da177e4 LT |
2963 | info = SHMEM_I(inode); |
2964 | inode->i_size = len-1; | |
69f07ec9 | 2965 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
2966 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
2967 | if (!inode->i_link) { | |
69f07ec9 HD |
2968 | iput(inode); |
2969 | return -ENOMEM; | |
2970 | } | |
2971 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 2972 | } else { |
e8ecde25 | 2973 | inode_nohighmem(inode); |
9e18eb29 | 2974 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
2975 | if (error) { |
2976 | iput(inode); | |
2977 | return error; | |
2978 | } | |
14fcc23f | 2979 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 2980 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 2981 | memcpy(page_address(page), symname, len); |
ec9516fb | 2982 | SetPageUptodate(page); |
1da177e4 | 2983 | set_page_dirty(page); |
6746aff7 | 2984 | unlock_page(page); |
09cbfeaf | 2985 | put_page(page); |
1da177e4 | 2986 | } |
1da177e4 LT |
2987 | dir->i_size += BOGO_DIRENT_SIZE; |
2988 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2989 | d_instantiate(dentry, inode); | |
2990 | dget(dentry); | |
2991 | return 0; | |
2992 | } | |
2993 | ||
fceef393 | 2994 | static void shmem_put_link(void *arg) |
1da177e4 | 2995 | { |
fceef393 AV |
2996 | mark_page_accessed(arg); |
2997 | put_page(arg); | |
1da177e4 LT |
2998 | } |
2999 | ||
6b255391 | 3000 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
3001 | struct inode *inode, |
3002 | struct delayed_call *done) | |
1da177e4 | 3003 | { |
1da177e4 | 3004 | struct page *page = NULL; |
6b255391 | 3005 | int error; |
6a6c9904 AV |
3006 | if (!dentry) { |
3007 | page = find_get_page(inode->i_mapping, 0); | |
3008 | if (!page) | |
3009 | return ERR_PTR(-ECHILD); | |
3010 | if (!PageUptodate(page)) { | |
3011 | put_page(page); | |
3012 | return ERR_PTR(-ECHILD); | |
3013 | } | |
3014 | } else { | |
9e18eb29 | 3015 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3016 | if (error) |
3017 | return ERR_PTR(error); | |
3018 | unlock_page(page); | |
3019 | } | |
fceef393 | 3020 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3021 | return page_address(page); |
1da177e4 LT |
3022 | } |
3023 | ||
b09e0fa4 | 3024 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3025 | /* |
b09e0fa4 EP |
3026 | * Superblocks without xattr inode operations may get some security.* xattr |
3027 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3028 | * like ACLs, we also need to implement the security.* handlers at |
3029 | * filesystem level, though. | |
3030 | */ | |
3031 | ||
6d9d88d0 JS |
3032 | /* |
3033 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3034 | */ | |
3035 | static int shmem_initxattrs(struct inode *inode, | |
3036 | const struct xattr *xattr_array, | |
3037 | void *fs_info) | |
3038 | { | |
3039 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3040 | const struct xattr *xattr; | |
38f38657 | 3041 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3042 | size_t len; |
3043 | ||
3044 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3045 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3046 | if (!new_xattr) |
3047 | return -ENOMEM; | |
3048 | ||
3049 | len = strlen(xattr->name) + 1; | |
3050 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3051 | GFP_KERNEL); | |
3052 | if (!new_xattr->name) { | |
3053 | kfree(new_xattr); | |
3054 | return -ENOMEM; | |
3055 | } | |
3056 | ||
3057 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3058 | XATTR_SECURITY_PREFIX_LEN); | |
3059 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3060 | xattr->name, len); | |
3061 | ||
38f38657 | 3062 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3063 | } |
3064 | ||
3065 | return 0; | |
3066 | } | |
3067 | ||
aa7c5241 | 3068 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3069 | struct dentry *unused, struct inode *inode, |
3070 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3071 | { |
b296821a | 3072 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3073 | |
aa7c5241 | 3074 | name = xattr_full_name(handler, name); |
38f38657 | 3075 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3076 | } |
3077 | ||
aa7c5241 | 3078 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
59301226 AV |
3079 | struct dentry *unused, struct inode *inode, |
3080 | const char *name, const void *value, | |
3081 | size_t size, int flags) | |
b09e0fa4 | 3082 | { |
59301226 | 3083 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3084 | |
aa7c5241 | 3085 | name = xattr_full_name(handler, name); |
38f38657 | 3086 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
3087 | } |
3088 | ||
aa7c5241 AG |
3089 | static const struct xattr_handler shmem_security_xattr_handler = { |
3090 | .prefix = XATTR_SECURITY_PREFIX, | |
3091 | .get = shmem_xattr_handler_get, | |
3092 | .set = shmem_xattr_handler_set, | |
3093 | }; | |
b09e0fa4 | 3094 | |
aa7c5241 AG |
3095 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3096 | .prefix = XATTR_TRUSTED_PREFIX, | |
3097 | .get = shmem_xattr_handler_get, | |
3098 | .set = shmem_xattr_handler_set, | |
3099 | }; | |
b09e0fa4 | 3100 | |
aa7c5241 AG |
3101 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3102 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3103 | &posix_acl_access_xattr_handler, | |
3104 | &posix_acl_default_xattr_handler, | |
3105 | #endif | |
3106 | &shmem_security_xattr_handler, | |
3107 | &shmem_trusted_xattr_handler, | |
3108 | NULL | |
3109 | }; | |
b09e0fa4 EP |
3110 | |
3111 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3112 | { | |
75c3cfa8 | 3113 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3114 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3115 | } |
3116 | #endif /* CONFIG_TMPFS_XATTR */ | |
3117 | ||
69f07ec9 | 3118 | static const struct inode_operations shmem_short_symlink_operations = { |
b09e0fa4 | 3119 | .readlink = generic_readlink, |
6b255391 | 3120 | .get_link = simple_get_link, |
b09e0fa4 | 3121 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3122 | .setxattr = generic_setxattr, |
3123 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3124 | .listxattr = shmem_listxattr, |
aa7c5241 | 3125 | .removexattr = generic_removexattr, |
b09e0fa4 EP |
3126 | #endif |
3127 | }; | |
3128 | ||
3129 | static const struct inode_operations shmem_symlink_inode_operations = { | |
3130 | .readlink = generic_readlink, | |
6b255391 | 3131 | .get_link = shmem_get_link, |
b09e0fa4 | 3132 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3133 | .setxattr = generic_setxattr, |
3134 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3135 | .listxattr = shmem_listxattr, |
aa7c5241 | 3136 | .removexattr = generic_removexattr, |
39f0247d | 3137 | #endif |
b09e0fa4 | 3138 | }; |
39f0247d | 3139 | |
91828a40 DG |
3140 | static struct dentry *shmem_get_parent(struct dentry *child) |
3141 | { | |
3142 | return ERR_PTR(-ESTALE); | |
3143 | } | |
3144 | ||
3145 | static int shmem_match(struct inode *ino, void *vfh) | |
3146 | { | |
3147 | __u32 *fh = vfh; | |
3148 | __u64 inum = fh[2]; | |
3149 | inum = (inum << 32) | fh[1]; | |
3150 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
3151 | } | |
3152 | ||
480b116c CH |
3153 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3154 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3155 | { |
91828a40 | 3156 | struct inode *inode; |
480b116c | 3157 | struct dentry *dentry = NULL; |
35c2a7f4 | 3158 | u64 inum; |
480b116c CH |
3159 | |
3160 | if (fh_len < 3) | |
3161 | return NULL; | |
91828a40 | 3162 | |
35c2a7f4 HD |
3163 | inum = fid->raw[2]; |
3164 | inum = (inum << 32) | fid->raw[1]; | |
3165 | ||
480b116c CH |
3166 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
3167 | shmem_match, fid->raw); | |
91828a40 | 3168 | if (inode) { |
480b116c | 3169 | dentry = d_find_alias(inode); |
91828a40 DG |
3170 | iput(inode); |
3171 | } | |
3172 | ||
480b116c | 3173 | return dentry; |
91828a40 DG |
3174 | } |
3175 | ||
b0b0382b AV |
3176 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3177 | struct inode *parent) | |
91828a40 | 3178 | { |
5fe0c237 AK |
3179 | if (*len < 3) { |
3180 | *len = 3; | |
94e07a75 | 3181 | return FILEID_INVALID; |
5fe0c237 | 3182 | } |
91828a40 | 3183 | |
1d3382cb | 3184 | if (inode_unhashed(inode)) { |
91828a40 DG |
3185 | /* Unfortunately insert_inode_hash is not idempotent, |
3186 | * so as we hash inodes here rather than at creation | |
3187 | * time, we need a lock to ensure we only try | |
3188 | * to do it once | |
3189 | */ | |
3190 | static DEFINE_SPINLOCK(lock); | |
3191 | spin_lock(&lock); | |
1d3382cb | 3192 | if (inode_unhashed(inode)) |
91828a40 DG |
3193 | __insert_inode_hash(inode, |
3194 | inode->i_ino + inode->i_generation); | |
3195 | spin_unlock(&lock); | |
3196 | } | |
3197 | ||
3198 | fh[0] = inode->i_generation; | |
3199 | fh[1] = inode->i_ino; | |
3200 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
3201 | ||
3202 | *len = 3; | |
3203 | return 1; | |
3204 | } | |
3205 | ||
39655164 | 3206 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3207 | .get_parent = shmem_get_parent, |
91828a40 | 3208 | .encode_fh = shmem_encode_fh, |
480b116c | 3209 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3210 | }; |
3211 | ||
680d794b | 3212 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
3213 | bool remount) | |
1da177e4 LT |
3214 | { |
3215 | char *this_char, *value, *rest; | |
49cd0a5c | 3216 | struct mempolicy *mpol = NULL; |
8751e039 EB |
3217 | uid_t uid; |
3218 | gid_t gid; | |
1da177e4 | 3219 | |
b00dc3ad HD |
3220 | while (options != NULL) { |
3221 | this_char = options; | |
3222 | for (;;) { | |
3223 | /* | |
3224 | * NUL-terminate this option: unfortunately, | |
3225 | * mount options form a comma-separated list, | |
3226 | * but mpol's nodelist may also contain commas. | |
3227 | */ | |
3228 | options = strchr(options, ','); | |
3229 | if (options == NULL) | |
3230 | break; | |
3231 | options++; | |
3232 | if (!isdigit(*options)) { | |
3233 | options[-1] = '\0'; | |
3234 | break; | |
3235 | } | |
3236 | } | |
1da177e4 LT |
3237 | if (!*this_char) |
3238 | continue; | |
3239 | if ((value = strchr(this_char,'=')) != NULL) { | |
3240 | *value++ = 0; | |
3241 | } else { | |
1170532b JP |
3242 | pr_err("tmpfs: No value for mount option '%s'\n", |
3243 | this_char); | |
49cd0a5c | 3244 | goto error; |
1da177e4 LT |
3245 | } |
3246 | ||
3247 | if (!strcmp(this_char,"size")) { | |
3248 | unsigned long long size; | |
3249 | size = memparse(value,&rest); | |
3250 | if (*rest == '%') { | |
3251 | size <<= PAGE_SHIFT; | |
3252 | size *= totalram_pages; | |
3253 | do_div(size, 100); | |
3254 | rest++; | |
3255 | } | |
3256 | if (*rest) | |
3257 | goto bad_val; | |
680d794b | 3258 | sbinfo->max_blocks = |
09cbfeaf | 3259 | DIV_ROUND_UP(size, PAGE_SIZE); |
1da177e4 | 3260 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 3261 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
3262 | if (*rest) |
3263 | goto bad_val; | |
3264 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 3265 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
3266 | if (*rest) |
3267 | goto bad_val; | |
3268 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 3269 | if (remount) |
1da177e4 | 3270 | continue; |
680d794b | 3271 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
3272 | if (*rest) |
3273 | goto bad_val; | |
3274 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 3275 | if (remount) |
1da177e4 | 3276 | continue; |
8751e039 | 3277 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3278 | if (*rest) |
3279 | goto bad_val; | |
8751e039 EB |
3280 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
3281 | if (!uid_valid(sbinfo->uid)) | |
3282 | goto bad_val; | |
1da177e4 | 3283 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 3284 | if (remount) |
1da177e4 | 3285 | continue; |
8751e039 | 3286 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
3287 | if (*rest) |
3288 | goto bad_val; | |
8751e039 EB |
3289 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
3290 | if (!gid_valid(sbinfo->gid)) | |
3291 | goto bad_val; | |
5a6e75f8 KS |
3292 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3293 | } else if (!strcmp(this_char, "huge")) { | |
3294 | int huge; | |
3295 | huge = shmem_parse_huge(value); | |
3296 | if (huge < 0) | |
3297 | goto bad_val; | |
3298 | if (!has_transparent_hugepage() && | |
3299 | huge != SHMEM_HUGE_NEVER) | |
3300 | goto bad_val; | |
3301 | sbinfo->huge = huge; | |
3302 | #endif | |
3303 | #ifdef CONFIG_NUMA | |
7339ff83 | 3304 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
3305 | mpol_put(mpol); |
3306 | mpol = NULL; | |
3307 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 3308 | goto bad_val; |
5a6e75f8 | 3309 | #endif |
1da177e4 | 3310 | } else { |
1170532b | 3311 | pr_err("tmpfs: Bad mount option %s\n", this_char); |
49cd0a5c | 3312 | goto error; |
1da177e4 LT |
3313 | } |
3314 | } | |
49cd0a5c | 3315 | sbinfo->mpol = mpol; |
1da177e4 LT |
3316 | return 0; |
3317 | ||
3318 | bad_val: | |
1170532b | 3319 | pr_err("tmpfs: Bad value '%s' for mount option '%s'\n", |
1da177e4 | 3320 | value, this_char); |
49cd0a5c GT |
3321 | error: |
3322 | mpol_put(mpol); | |
1da177e4 LT |
3323 | return 1; |
3324 | ||
3325 | } | |
3326 | ||
3327 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
3328 | { | |
3329 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 3330 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
3331 | unsigned long inodes; |
3332 | int error = -EINVAL; | |
3333 | ||
5f00110f | 3334 | config.mpol = NULL; |
680d794b | 3335 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 3336 | return error; |
1da177e4 | 3337 | |
0edd73b3 | 3338 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3339 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 3340 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 3341 | goto out; |
680d794b | 3342 | if (config.max_inodes < inodes) |
0edd73b3 HD |
3343 | goto out; |
3344 | /* | |
54af6042 | 3345 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
3346 | * but we must separately disallow unlimited->limited, because |
3347 | * in that case we have no record of how much is already in use. | |
3348 | */ | |
680d794b | 3349 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 3350 | goto out; |
680d794b | 3351 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
3352 | goto out; |
3353 | ||
3354 | error = 0; | |
5a6e75f8 | 3355 | sbinfo->huge = config.huge; |
680d794b | 3356 | sbinfo->max_blocks = config.max_blocks; |
680d794b | 3357 | sbinfo->max_inodes = config.max_inodes; |
3358 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 3359 | |
5f00110f GT |
3360 | /* |
3361 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3362 | */ | |
3363 | if (config.mpol) { | |
3364 | mpol_put(sbinfo->mpol); | |
3365 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
3366 | } | |
0edd73b3 HD |
3367 | out: |
3368 | spin_unlock(&sbinfo->stat_lock); | |
3369 | return error; | |
1da177e4 | 3370 | } |
680d794b | 3371 | |
34c80b1d | 3372 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3373 | { |
34c80b1d | 3374 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b | 3375 | |
3376 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3377 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3378 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 3379 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
3380 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
3381 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 3382 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3383 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3384 | seq_printf(seq, ",uid=%u", | |
3385 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3386 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3387 | seq_printf(seq, ",gid=%u", | |
3388 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
5a6e75f8 KS |
3389 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3390 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ | |
3391 | if (sbinfo->huge) | |
3392 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3393 | #endif | |
71fe804b | 3394 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 3395 | return 0; |
3396 | } | |
9183df25 DH |
3397 | |
3398 | #define MFD_NAME_PREFIX "memfd:" | |
3399 | #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1) | |
3400 | #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN) | |
3401 | ||
3402 | #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING) | |
3403 | ||
3404 | SYSCALL_DEFINE2(memfd_create, | |
3405 | const char __user *, uname, | |
3406 | unsigned int, flags) | |
3407 | { | |
3408 | struct shmem_inode_info *info; | |
3409 | struct file *file; | |
3410 | int fd, error; | |
3411 | char *name; | |
3412 | long len; | |
3413 | ||
3414 | if (flags & ~(unsigned int)MFD_ALL_FLAGS) | |
3415 | return -EINVAL; | |
3416 | ||
3417 | /* length includes terminating zero */ | |
3418 | len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1); | |
3419 | if (len <= 0) | |
3420 | return -EFAULT; | |
3421 | if (len > MFD_NAME_MAX_LEN + 1) | |
3422 | return -EINVAL; | |
3423 | ||
3424 | name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY); | |
3425 | if (!name) | |
3426 | return -ENOMEM; | |
3427 | ||
3428 | strcpy(name, MFD_NAME_PREFIX); | |
3429 | if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) { | |
3430 | error = -EFAULT; | |
3431 | goto err_name; | |
3432 | } | |
3433 | ||
3434 | /* terminating-zero may have changed after strnlen_user() returned */ | |
3435 | if (name[len + MFD_NAME_PREFIX_LEN - 1]) { | |
3436 | error = -EFAULT; | |
3437 | goto err_name; | |
3438 | } | |
3439 | ||
3440 | fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0); | |
3441 | if (fd < 0) { | |
3442 | error = fd; | |
3443 | goto err_name; | |
3444 | } | |
3445 | ||
3446 | file = shmem_file_setup(name, 0, VM_NORESERVE); | |
3447 | if (IS_ERR(file)) { | |
3448 | error = PTR_ERR(file); | |
3449 | goto err_fd; | |
3450 | } | |
3451 | info = SHMEM_I(file_inode(file)); | |
3452 | file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; | |
3453 | file->f_flags |= O_RDWR | O_LARGEFILE; | |
3454 | if (flags & MFD_ALLOW_SEALING) | |
3455 | info->seals &= ~F_SEAL_SEAL; | |
3456 | ||
3457 | fd_install(fd, file); | |
3458 | kfree(name); | |
3459 | return fd; | |
3460 | ||
3461 | err_fd: | |
3462 | put_unused_fd(fd); | |
3463 | err_name: | |
3464 | kfree(name); | |
3465 | return error; | |
3466 | } | |
3467 | ||
680d794b | 3468 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3469 | |
3470 | static void shmem_put_super(struct super_block *sb) | |
3471 | { | |
602586a8 HD |
3472 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3473 | ||
3474 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 3475 | mpol_put(sbinfo->mpol); |
602586a8 | 3476 | kfree(sbinfo); |
1da177e4 LT |
3477 | sb->s_fs_info = NULL; |
3478 | } | |
3479 | ||
2b2af54a | 3480 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3481 | { |
3482 | struct inode *inode; | |
0edd73b3 | 3483 | struct shmem_sb_info *sbinfo; |
680d794b | 3484 | int err = -ENOMEM; |
3485 | ||
3486 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3487 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 3488 | L1_CACHE_BYTES), GFP_KERNEL); |
3489 | if (!sbinfo) | |
3490 | return -ENOMEM; | |
3491 | ||
680d794b | 3492 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
3493 | sbinfo->uid = current_fsuid(); |
3494 | sbinfo->gid = current_fsgid(); | |
680d794b | 3495 | sb->s_fs_info = sbinfo; |
1da177e4 | 3496 | |
0edd73b3 | 3497 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3498 | /* |
3499 | * Per default we only allow half of the physical ram per | |
3500 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3501 | * but the internal instance is left unlimited. | |
3502 | */ | |
ca4e0519 | 3503 | if (!(sb->s_flags & MS_KERNMOUNT)) { |
680d794b | 3504 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3505 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3506 | if (shmem_parse_options(data, sbinfo, false)) { | |
3507 | err = -EINVAL; | |
3508 | goto failed; | |
3509 | } | |
ca4e0519 AV |
3510 | } else { |
3511 | sb->s_flags |= MS_NOUSER; | |
1da177e4 | 3512 | } |
91828a40 | 3513 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 3514 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
3515 | #else |
3516 | sb->s_flags |= MS_NOUSER; | |
3517 | #endif | |
3518 | ||
0edd73b3 | 3519 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3520 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3521 | goto failed; |
680d794b | 3522 | sbinfo->free_inodes = sbinfo->max_inodes; |
0edd73b3 | 3523 | |
285b2c4f | 3524 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3525 | sb->s_blocksize = PAGE_SIZE; |
3526 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3527 | sb->s_magic = TMPFS_MAGIC; |
3528 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3529 | sb->s_time_gran = 1; |
b09e0fa4 | 3530 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3531 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3532 | #endif |
3533 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
3534 | sb->s_flags |= MS_POSIXACL; |
3535 | #endif | |
0edd73b3 | 3536 | |
454abafe | 3537 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3538 | if (!inode) |
3539 | goto failed; | |
680d794b | 3540 | inode->i_uid = sbinfo->uid; |
3541 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3542 | sb->s_root = d_make_root(inode); |
3543 | if (!sb->s_root) | |
48fde701 | 3544 | goto failed; |
1da177e4 LT |
3545 | return 0; |
3546 | ||
1da177e4 LT |
3547 | failed: |
3548 | shmem_put_super(sb); | |
3549 | return err; | |
3550 | } | |
3551 | ||
fcc234f8 | 3552 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3553 | |
3554 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3555 | { | |
41ffe5d5 HD |
3556 | struct shmem_inode_info *info; |
3557 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3558 | if (!info) | |
1da177e4 | 3559 | return NULL; |
41ffe5d5 | 3560 | return &info->vfs_inode; |
1da177e4 LT |
3561 | } |
3562 | ||
41ffe5d5 | 3563 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3564 | { |
3565 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84e710da AV |
3566 | if (S_ISLNK(inode->i_mode)) |
3567 | kfree(inode->i_link); | |
fa0d7e3d NP |
3568 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3569 | } | |
3570 | ||
1da177e4 LT |
3571 | static void shmem_destroy_inode(struct inode *inode) |
3572 | { | |
09208d15 | 3573 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3574 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3575 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3576 | } |
3577 | ||
41ffe5d5 | 3578 | static void shmem_init_inode(void *foo) |
1da177e4 | 3579 | { |
41ffe5d5 HD |
3580 | struct shmem_inode_info *info = foo; |
3581 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3582 | } |
3583 | ||
41ffe5d5 | 3584 | static int shmem_init_inodecache(void) |
1da177e4 LT |
3585 | { |
3586 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3587 | sizeof(struct shmem_inode_info), | |
5d097056 | 3588 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3589 | return 0; |
3590 | } | |
3591 | ||
41ffe5d5 | 3592 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3593 | { |
1a1d92c1 | 3594 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3595 | } |
3596 | ||
f5e54d6e | 3597 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3598 | .writepage = shmem_writepage, |
76719325 | 3599 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3600 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3601 | .write_begin = shmem_write_begin, |
3602 | .write_end = shmem_write_end, | |
1da177e4 | 3603 | #endif |
1c93923c | 3604 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3605 | .migratepage = migrate_page, |
1c93923c | 3606 | #endif |
aa261f54 | 3607 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3608 | }; |
3609 | ||
15ad7cdc | 3610 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3611 | .mmap = shmem_mmap, |
c01d5b30 | 3612 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3613 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3614 | .llseek = shmem_file_llseek, |
2ba5bbed | 3615 | .read_iter = shmem_file_read_iter, |
8174202b | 3616 | .write_iter = generic_file_write_iter, |
1b061d92 | 3617 | .fsync = noop_fsync, |
708e3508 | 3618 | .splice_read = shmem_file_splice_read, |
f6cb85d0 | 3619 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3620 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3621 | #endif |
3622 | }; | |
3623 | ||
92e1d5be | 3624 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3625 | .getattr = shmem_getattr, |
94c1e62d | 3626 | .setattr = shmem_setattr, |
b09e0fa4 | 3627 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3628 | .setxattr = generic_setxattr, |
3629 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3630 | .listxattr = shmem_listxattr, |
aa7c5241 | 3631 | .removexattr = generic_removexattr, |
feda821e | 3632 | .set_acl = simple_set_acl, |
b09e0fa4 | 3633 | #endif |
1da177e4 LT |
3634 | }; |
3635 | ||
92e1d5be | 3636 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3637 | #ifdef CONFIG_TMPFS |
3638 | .create = shmem_create, | |
3639 | .lookup = simple_lookup, | |
3640 | .link = shmem_link, | |
3641 | .unlink = shmem_unlink, | |
3642 | .symlink = shmem_symlink, | |
3643 | .mkdir = shmem_mkdir, | |
3644 | .rmdir = shmem_rmdir, | |
3645 | .mknod = shmem_mknod, | |
3b69ff51 | 3646 | .rename2 = shmem_rename2, |
60545d0d | 3647 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3648 | #endif |
b09e0fa4 | 3649 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3650 | .setxattr = generic_setxattr, |
3651 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3652 | .listxattr = shmem_listxattr, |
aa7c5241 | 3653 | .removexattr = generic_removexattr, |
b09e0fa4 | 3654 | #endif |
39f0247d | 3655 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3656 | .setattr = shmem_setattr, |
feda821e | 3657 | .set_acl = simple_set_acl, |
39f0247d AG |
3658 | #endif |
3659 | }; | |
3660 | ||
92e1d5be | 3661 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3662 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3663 | .setxattr = generic_setxattr, |
3664 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3665 | .listxattr = shmem_listxattr, |
aa7c5241 | 3666 | .removexattr = generic_removexattr, |
b09e0fa4 | 3667 | #endif |
39f0247d | 3668 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3669 | .setattr = shmem_setattr, |
feda821e | 3670 | .set_acl = simple_set_acl, |
39f0247d | 3671 | #endif |
1da177e4 LT |
3672 | }; |
3673 | ||
759b9775 | 3674 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
3675 | .alloc_inode = shmem_alloc_inode, |
3676 | .destroy_inode = shmem_destroy_inode, | |
3677 | #ifdef CONFIG_TMPFS | |
3678 | .statfs = shmem_statfs, | |
3679 | .remount_fs = shmem_remount_fs, | |
680d794b | 3680 | .show_options = shmem_show_options, |
1da177e4 | 3681 | #endif |
1f895f75 | 3682 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3683 | .drop_inode = generic_delete_inode, |
3684 | .put_super = shmem_put_super, | |
3685 | }; | |
3686 | ||
f0f37e2f | 3687 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3688 | .fault = shmem_fault, |
d7c17551 | 3689 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3690 | #ifdef CONFIG_NUMA |
3691 | .set_policy = shmem_set_policy, | |
3692 | .get_policy = shmem_get_policy, | |
3693 | #endif | |
3694 | }; | |
3695 | ||
3c26ff6e AV |
3696 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
3697 | int flags, const char *dev_name, void *data) | |
1da177e4 | 3698 | { |
3c26ff6e | 3699 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
3700 | } |
3701 | ||
41ffe5d5 | 3702 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3703 | .owner = THIS_MODULE, |
3704 | .name = "tmpfs", | |
3c26ff6e | 3705 | .mount = shmem_mount, |
1da177e4 | 3706 | .kill_sb = kill_litter_super, |
2b8576cb | 3707 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 3708 | }; |
1da177e4 | 3709 | |
41ffe5d5 | 3710 | int __init shmem_init(void) |
1da177e4 LT |
3711 | { |
3712 | int error; | |
3713 | ||
16203a7a RL |
3714 | /* If rootfs called this, don't re-init */ |
3715 | if (shmem_inode_cachep) | |
3716 | return 0; | |
3717 | ||
41ffe5d5 | 3718 | error = shmem_init_inodecache(); |
1da177e4 LT |
3719 | if (error) |
3720 | goto out3; | |
3721 | ||
41ffe5d5 | 3722 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 3723 | if (error) { |
1170532b | 3724 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
3725 | goto out2; |
3726 | } | |
95dc112a | 3727 | |
ca4e0519 | 3728 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
3729 | if (IS_ERR(shm_mnt)) { |
3730 | error = PTR_ERR(shm_mnt); | |
1170532b | 3731 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
3732 | goto out1; |
3733 | } | |
5a6e75f8 KS |
3734 | |
3735 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3736 | if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY) | |
3737 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; | |
3738 | else | |
3739 | shmem_huge = 0; /* just in case it was patched */ | |
3740 | #endif | |
1da177e4 LT |
3741 | return 0; |
3742 | ||
3743 | out1: | |
41ffe5d5 | 3744 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 3745 | out2: |
41ffe5d5 | 3746 | shmem_destroy_inodecache(); |
1da177e4 LT |
3747 | out3: |
3748 | shm_mnt = ERR_PTR(error); | |
3749 | return error; | |
3750 | } | |
853ac43a | 3751 | |
5a6e75f8 KS |
3752 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS) |
3753 | static ssize_t shmem_enabled_show(struct kobject *kobj, | |
3754 | struct kobj_attribute *attr, char *buf) | |
3755 | { | |
3756 | int values[] = { | |
3757 | SHMEM_HUGE_ALWAYS, | |
3758 | SHMEM_HUGE_WITHIN_SIZE, | |
3759 | SHMEM_HUGE_ADVISE, | |
3760 | SHMEM_HUGE_NEVER, | |
3761 | SHMEM_HUGE_DENY, | |
3762 | SHMEM_HUGE_FORCE, | |
3763 | }; | |
3764 | int i, count; | |
3765 | ||
3766 | for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) { | |
3767 | const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s "; | |
3768 | ||
3769 | count += sprintf(buf + count, fmt, | |
3770 | shmem_format_huge(values[i])); | |
3771 | } | |
3772 | buf[count - 1] = '\n'; | |
3773 | return count; | |
3774 | } | |
3775 | ||
3776 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
3777 | struct kobj_attribute *attr, const char *buf, size_t count) | |
3778 | { | |
3779 | char tmp[16]; | |
3780 | int huge; | |
3781 | ||
3782 | if (count + 1 > sizeof(tmp)) | |
3783 | return -EINVAL; | |
3784 | memcpy(tmp, buf, count); | |
3785 | tmp[count] = '\0'; | |
3786 | if (count && tmp[count - 1] == '\n') | |
3787 | tmp[count - 1] = '\0'; | |
3788 | ||
3789 | huge = shmem_parse_huge(tmp); | |
3790 | if (huge == -EINVAL) | |
3791 | return -EINVAL; | |
3792 | if (!has_transparent_hugepage() && | |
3793 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
3794 | return -EINVAL; | |
3795 | ||
3796 | shmem_huge = huge; | |
3797 | if (shmem_huge < SHMEM_HUGE_DENY) | |
3798 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; | |
3799 | return count; | |
3800 | } | |
3801 | ||
3802 | struct kobj_attribute shmem_enabled_attr = | |
3803 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
3804 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */ | |
3805 | ||
853ac43a MM |
3806 | #else /* !CONFIG_SHMEM */ |
3807 | ||
3808 | /* | |
3809 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
3810 | * | |
3811 | * This is intended for small system where the benefits of the full | |
3812 | * shmem code (swap-backed and resource-limited) are outweighed by | |
3813 | * their complexity. On systems without swap this code should be | |
3814 | * effectively equivalent, but much lighter weight. | |
3815 | */ | |
3816 | ||
41ffe5d5 | 3817 | static struct file_system_type shmem_fs_type = { |
853ac43a | 3818 | .name = "tmpfs", |
3c26ff6e | 3819 | .mount = ramfs_mount, |
853ac43a | 3820 | .kill_sb = kill_litter_super, |
2b8576cb | 3821 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
3822 | }; |
3823 | ||
41ffe5d5 | 3824 | int __init shmem_init(void) |
853ac43a | 3825 | { |
41ffe5d5 | 3826 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 3827 | |
41ffe5d5 | 3828 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
3829 | BUG_ON(IS_ERR(shm_mnt)); |
3830 | ||
3831 | return 0; | |
3832 | } | |
3833 | ||
41ffe5d5 | 3834 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
3835 | { |
3836 | return 0; | |
3837 | } | |
3838 | ||
3f96b79a HD |
3839 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
3840 | { | |
3841 | return 0; | |
3842 | } | |
3843 | ||
24513264 HD |
3844 | void shmem_unlock_mapping(struct address_space *mapping) |
3845 | { | |
3846 | } | |
3847 | ||
c01d5b30 HD |
3848 | #ifdef CONFIG_MMU |
3849 | unsigned long shmem_get_unmapped_area(struct file *file, | |
3850 | unsigned long addr, unsigned long len, | |
3851 | unsigned long pgoff, unsigned long flags) | |
3852 | { | |
3853 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
3854 | } | |
3855 | #endif | |
3856 | ||
41ffe5d5 | 3857 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 3858 | { |
41ffe5d5 | 3859 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
3860 | } |
3861 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
3862 | ||
0b0a0806 HD |
3863 | #define shmem_vm_ops generic_file_vm_ops |
3864 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 3865 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
3866 | #define shmem_acct_size(flags, size) 0 |
3867 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
3868 | |
3869 | #endif /* CONFIG_SHMEM */ | |
3870 | ||
3871 | /* common code */ | |
1da177e4 | 3872 | |
3451538a | 3873 | static struct dentry_operations anon_ops = { |
118b2302 | 3874 | .d_dname = simple_dname |
3451538a AV |
3875 | }; |
3876 | ||
c7277090 EP |
3877 | static struct file *__shmem_file_setup(const char *name, loff_t size, |
3878 | unsigned long flags, unsigned int i_flags) | |
1da177e4 | 3879 | { |
6b4d0b27 | 3880 | struct file *res; |
1da177e4 | 3881 | struct inode *inode; |
2c48b9c4 | 3882 | struct path path; |
3451538a | 3883 | struct super_block *sb; |
1da177e4 LT |
3884 | struct qstr this; |
3885 | ||
3886 | if (IS_ERR(shm_mnt)) | |
6b4d0b27 | 3887 | return ERR_CAST(shm_mnt); |
1da177e4 | 3888 | |
285b2c4f | 3889 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
3890 | return ERR_PTR(-EINVAL); |
3891 | ||
3892 | if (shmem_acct_size(flags, size)) | |
3893 | return ERR_PTR(-ENOMEM); | |
3894 | ||
6b4d0b27 | 3895 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
3896 | this.name = name; |
3897 | this.len = strlen(name); | |
3898 | this.hash = 0; /* will go */ | |
3451538a | 3899 | sb = shm_mnt->mnt_sb; |
66ee4b88 | 3900 | path.mnt = mntget(shm_mnt); |
3451538a | 3901 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 3902 | if (!path.dentry) |
1da177e4 | 3903 | goto put_memory; |
3451538a | 3904 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 3905 | |
6b4d0b27 | 3906 | res = ERR_PTR(-ENOSPC); |
3451538a | 3907 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 3908 | if (!inode) |
66ee4b88 | 3909 | goto put_memory; |
1da177e4 | 3910 | |
c7277090 | 3911 | inode->i_flags |= i_flags; |
2c48b9c4 | 3912 | d_instantiate(path.dentry, inode); |
1da177e4 | 3913 | inode->i_size = size; |
6d6b77f1 | 3914 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
3915 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
3916 | if (IS_ERR(res)) | |
66ee4b88 | 3917 | goto put_path; |
4b42af81 | 3918 | |
6b4d0b27 | 3919 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 3920 | &shmem_file_operations); |
6b4d0b27 | 3921 | if (IS_ERR(res)) |
66ee4b88 | 3922 | goto put_path; |
4b42af81 | 3923 | |
6b4d0b27 | 3924 | return res; |
1da177e4 | 3925 | |
1da177e4 LT |
3926 | put_memory: |
3927 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
3928 | put_path: |
3929 | path_put(&path); | |
6b4d0b27 | 3930 | return res; |
1da177e4 | 3931 | } |
c7277090 EP |
3932 | |
3933 | /** | |
3934 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
3935 | * kernel internal. There will be NO LSM permission checks against the | |
3936 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
3937 | * higher layer. The users are the big_key and shm implementations. LSM |
3938 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
3939 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
3940 | * @size: size to be set for the file | |
3941 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3942 | */ | |
3943 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
3944 | { | |
3945 | return __shmem_file_setup(name, size, flags, S_PRIVATE); | |
3946 | } | |
3947 | ||
3948 | /** | |
3949 | * shmem_file_setup - get an unlinked file living in tmpfs | |
3950 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
3951 | * @size: size to be set for the file | |
3952 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3953 | */ | |
3954 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
3955 | { | |
3956 | return __shmem_file_setup(name, size, flags, 0); | |
3957 | } | |
395e0ddc | 3958 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 3959 | |
46711810 | 3960 | /** |
1da177e4 | 3961 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
3962 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
3963 | */ | |
3964 | int shmem_zero_setup(struct vm_area_struct *vma) | |
3965 | { | |
3966 | struct file *file; | |
3967 | loff_t size = vma->vm_end - vma->vm_start; | |
3968 | ||
66fc1303 HD |
3969 | /* |
3970 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
3971 | * between XFS directory reading and selinux: since this file is only | |
3972 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
3973 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
3974 | */ | |
3975 | file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE); | |
1da177e4 LT |
3976 | if (IS_ERR(file)) |
3977 | return PTR_ERR(file); | |
3978 | ||
3979 | if (vma->vm_file) | |
3980 | fput(vma->vm_file); | |
3981 | vma->vm_file = file; | |
3982 | vma->vm_ops = &shmem_vm_ops; | |
3983 | return 0; | |
3984 | } | |
d9d90e5e HD |
3985 | |
3986 | /** | |
3987 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
3988 | * @mapping: the page's address_space | |
3989 | * @index: the page index | |
3990 | * @gfp: the page allocator flags to use if allocating | |
3991 | * | |
3992 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
3993 | * with any new page allocations done using the specified allocation flags. | |
3994 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
3995 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
3996 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
3997 | * | |
68da9f05 HD |
3998 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
3999 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4000 | */ |
4001 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4002 | pgoff_t index, gfp_t gfp) | |
4003 | { | |
68da9f05 HD |
4004 | #ifdef CONFIG_SHMEM |
4005 | struct inode *inode = mapping->host; | |
9276aad6 | 4006 | struct page *page; |
68da9f05 HD |
4007 | int error; |
4008 | ||
4009 | BUG_ON(mapping->a_ops != &shmem_aops); | |
9e18eb29 ALC |
4010 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
4011 | gfp, NULL, NULL); | |
68da9f05 HD |
4012 | if (error) |
4013 | page = ERR_PTR(error); | |
4014 | else | |
4015 | unlock_page(page); | |
4016 | return page; | |
4017 | #else | |
4018 | /* | |
4019 | * The tiny !SHMEM case uses ramfs without swap | |
4020 | */ | |
d9d90e5e | 4021 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4022 | #endif |
d9d90e5e HD |
4023 | } |
4024 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |