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 */ | |
1635f6a7 HD |
104 | SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */ |
105 | SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */ | |
1da177e4 LT |
106 | }; |
107 | ||
b76db735 | 108 | #ifdef CONFIG_TMPFS |
680d794b | 109 | static unsigned long shmem_default_max_blocks(void) |
110 | { | |
111 | return totalram_pages / 2; | |
112 | } | |
113 | ||
114 | static unsigned long shmem_default_max_inodes(void) | |
115 | { | |
116 | return min(totalram_pages - totalhigh_pages, totalram_pages / 2); | |
117 | } | |
b76db735 | 118 | #endif |
680d794b | 119 | |
bde05d1c HD |
120 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
121 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
122 | struct shmem_inode_info *info, pgoff_t index); | |
68da9f05 HD |
123 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
124 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type); | |
125 | ||
126 | static inline int shmem_getpage(struct inode *inode, pgoff_t index, | |
127 | struct page **pagep, enum sgp_type sgp, int *fault_type) | |
128 | { | |
129 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
130 | mapping_gfp_mask(inode->i_mapping), fault_type); | |
131 | } | |
1da177e4 | 132 | |
1da177e4 LT |
133 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
134 | { | |
135 | return sb->s_fs_info; | |
136 | } | |
137 | ||
138 | /* | |
139 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
140 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
141 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
142 | * consistent with the pre-accounting of private mappings ... | |
143 | */ | |
144 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
145 | { | |
0b0a0806 | 146 | return (flags & VM_NORESERVE) ? |
191c5424 | 147 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
148 | } |
149 | ||
150 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
151 | { | |
0b0a0806 | 152 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
153 | vm_unacct_memory(VM_ACCT(size)); |
154 | } | |
155 | ||
77142517 KK |
156 | static inline int shmem_reacct_size(unsigned long flags, |
157 | loff_t oldsize, loff_t newsize) | |
158 | { | |
159 | if (!(flags & VM_NORESERVE)) { | |
160 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
161 | return security_vm_enough_memory_mm(current->mm, | |
162 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
163 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
164 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
165 | } | |
166 | return 0; | |
167 | } | |
168 | ||
1da177e4 LT |
169 | /* |
170 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 171 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
172 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
173 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
174 | */ | |
175 | static inline int shmem_acct_block(unsigned long flags) | |
176 | { | |
0b0a0806 | 177 | return (flags & VM_NORESERVE) ? |
09cbfeaf | 178 | security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_SIZE)) : 0; |
1da177e4 LT |
179 | } |
180 | ||
181 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
182 | { | |
0b0a0806 | 183 | if (flags & VM_NORESERVE) |
09cbfeaf | 184 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
185 | } |
186 | ||
759b9775 | 187 | static const struct super_operations shmem_ops; |
f5e54d6e | 188 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 189 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
190 | static const struct inode_operations shmem_inode_operations; |
191 | static const struct inode_operations shmem_dir_inode_operations; | |
192 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 193 | static const struct vm_operations_struct shmem_vm_ops; |
1da177e4 | 194 | |
1da177e4 | 195 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 196 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 197 | |
5b04c689 PE |
198 | static int shmem_reserve_inode(struct super_block *sb) |
199 | { | |
200 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
201 | if (sbinfo->max_inodes) { | |
202 | spin_lock(&sbinfo->stat_lock); | |
203 | if (!sbinfo->free_inodes) { | |
204 | spin_unlock(&sbinfo->stat_lock); | |
205 | return -ENOSPC; | |
206 | } | |
207 | sbinfo->free_inodes--; | |
208 | spin_unlock(&sbinfo->stat_lock); | |
209 | } | |
210 | return 0; | |
211 | } | |
212 | ||
213 | static void shmem_free_inode(struct super_block *sb) | |
214 | { | |
215 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
216 | if (sbinfo->max_inodes) { | |
217 | spin_lock(&sbinfo->stat_lock); | |
218 | sbinfo->free_inodes++; | |
219 | spin_unlock(&sbinfo->stat_lock); | |
220 | } | |
221 | } | |
222 | ||
46711810 | 223 | /** |
41ffe5d5 | 224 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
225 | * @inode: inode to recalc |
226 | * | |
227 | * We have to calculate the free blocks since the mm can drop | |
228 | * undirtied hole pages behind our back. | |
229 | * | |
230 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
231 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
232 | * | |
233 | * It has to be called with the spinlock held. | |
234 | */ | |
235 | static void shmem_recalc_inode(struct inode *inode) | |
236 | { | |
237 | struct shmem_inode_info *info = SHMEM_I(inode); | |
238 | long freed; | |
239 | ||
240 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
241 | if (freed > 0) { | |
54af6042 HD |
242 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
243 | if (sbinfo->max_blocks) | |
244 | percpu_counter_add(&sbinfo->used_blocks, -freed); | |
1da177e4 | 245 | info->alloced -= freed; |
54af6042 | 246 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
1da177e4 | 247 | shmem_unacct_blocks(info->flags, freed); |
1da177e4 LT |
248 | } |
249 | } | |
250 | ||
7a5d0fbb HD |
251 | /* |
252 | * Replace item expected in radix tree by a new item, while holding tree lock. | |
253 | */ | |
254 | static int shmem_radix_tree_replace(struct address_space *mapping, | |
255 | pgoff_t index, void *expected, void *replacement) | |
256 | { | |
257 | void **pslot; | |
6dbaf22c | 258 | void *item; |
7a5d0fbb HD |
259 | |
260 | VM_BUG_ON(!expected); | |
6dbaf22c | 261 | VM_BUG_ON(!replacement); |
7a5d0fbb | 262 | pslot = radix_tree_lookup_slot(&mapping->page_tree, index); |
6dbaf22c JW |
263 | if (!pslot) |
264 | return -ENOENT; | |
265 | item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock); | |
7a5d0fbb HD |
266 | if (item != expected) |
267 | return -ENOENT; | |
6dbaf22c | 268 | radix_tree_replace_slot(pslot, replacement); |
7a5d0fbb HD |
269 | return 0; |
270 | } | |
271 | ||
d1899228 HD |
272 | /* |
273 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
274 | * that an entry was not already brought back from swap by a racing thread. | |
275 | * | |
276 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
277 | * might be reused, and again be SwapCache, using the same swap as before. | |
278 | */ | |
279 | static bool shmem_confirm_swap(struct address_space *mapping, | |
280 | pgoff_t index, swp_entry_t swap) | |
281 | { | |
282 | void *item; | |
283 | ||
284 | rcu_read_lock(); | |
285 | item = radix_tree_lookup(&mapping->page_tree, index); | |
286 | rcu_read_unlock(); | |
287 | return item == swp_to_radix_entry(swap); | |
288 | } | |
289 | ||
46f65ec1 HD |
290 | /* |
291 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
292 | */ | |
293 | static int shmem_add_to_page_cache(struct page *page, | |
294 | struct address_space *mapping, | |
fed400a1 | 295 | pgoff_t index, void *expected) |
46f65ec1 | 296 | { |
b065b432 | 297 | int error; |
46f65ec1 | 298 | |
309381fe SL |
299 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
300 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
46f65ec1 | 301 | |
09cbfeaf | 302 | get_page(page); |
b065b432 HD |
303 | page->mapping = mapping; |
304 | page->index = index; | |
305 | ||
306 | spin_lock_irq(&mapping->tree_lock); | |
46f65ec1 | 307 | if (!expected) |
b065b432 HD |
308 | error = radix_tree_insert(&mapping->page_tree, index, page); |
309 | else | |
310 | error = shmem_radix_tree_replace(mapping, index, expected, | |
311 | page); | |
46f65ec1 | 312 | if (!error) { |
b065b432 HD |
313 | mapping->nrpages++; |
314 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
315 | __inc_zone_page_state(page, NR_SHMEM); | |
316 | spin_unlock_irq(&mapping->tree_lock); | |
317 | } else { | |
318 | page->mapping = NULL; | |
319 | spin_unlock_irq(&mapping->tree_lock); | |
09cbfeaf | 320 | put_page(page); |
46f65ec1 | 321 | } |
46f65ec1 HD |
322 | return error; |
323 | } | |
324 | ||
6922c0c7 HD |
325 | /* |
326 | * Like delete_from_page_cache, but substitutes swap for page. | |
327 | */ | |
328 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
329 | { | |
330 | struct address_space *mapping = page->mapping; | |
331 | int error; | |
332 | ||
333 | spin_lock_irq(&mapping->tree_lock); | |
334 | error = shmem_radix_tree_replace(mapping, page->index, page, radswap); | |
335 | page->mapping = NULL; | |
336 | mapping->nrpages--; | |
337 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
338 | __dec_zone_page_state(page, NR_SHMEM); | |
339 | spin_unlock_irq(&mapping->tree_lock); | |
09cbfeaf | 340 | put_page(page); |
6922c0c7 HD |
341 | BUG_ON(error); |
342 | } | |
343 | ||
7a5d0fbb HD |
344 | /* |
345 | * Remove swap entry from radix tree, free the swap and its page cache. | |
346 | */ | |
347 | static int shmem_free_swap(struct address_space *mapping, | |
348 | pgoff_t index, void *radswap) | |
349 | { | |
6dbaf22c | 350 | void *old; |
7a5d0fbb HD |
351 | |
352 | spin_lock_irq(&mapping->tree_lock); | |
6dbaf22c | 353 | old = radix_tree_delete_item(&mapping->page_tree, index, radswap); |
7a5d0fbb | 354 | spin_unlock_irq(&mapping->tree_lock); |
6dbaf22c JW |
355 | if (old != radswap) |
356 | return -ENOENT; | |
357 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
358 | return 0; | |
7a5d0fbb HD |
359 | } |
360 | ||
6a15a370 VB |
361 | /* |
362 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 363 | * given offsets are swapped out. |
6a15a370 VB |
364 | * |
365 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
366 | * as long as the inode doesn't go away and racy results are not a problem. | |
367 | */ | |
48131e03 VB |
368 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
369 | pgoff_t start, pgoff_t end) | |
6a15a370 | 370 | { |
6a15a370 VB |
371 | struct radix_tree_iter iter; |
372 | void **slot; | |
373 | struct page *page; | |
48131e03 | 374 | unsigned long swapped = 0; |
6a15a370 VB |
375 | |
376 | rcu_read_lock(); | |
377 | ||
6a15a370 VB |
378 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
379 | if (iter.index >= end) | |
380 | break; | |
381 | ||
382 | page = radix_tree_deref_slot(slot); | |
383 | ||
2cf938aa MW |
384 | if (radix_tree_deref_retry(page)) { |
385 | slot = radix_tree_iter_retry(&iter); | |
386 | continue; | |
387 | } | |
6a15a370 VB |
388 | |
389 | if (radix_tree_exceptional_entry(page)) | |
390 | swapped++; | |
391 | ||
392 | if (need_resched()) { | |
393 | cond_resched_rcu(); | |
7165092f | 394 | slot = radix_tree_iter_next(&iter); |
6a15a370 VB |
395 | } |
396 | } | |
397 | ||
398 | rcu_read_unlock(); | |
399 | ||
400 | return swapped << PAGE_SHIFT; | |
401 | } | |
402 | ||
48131e03 VB |
403 | /* |
404 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
405 | * given vma is swapped out. | |
406 | * | |
407 | * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU, | |
408 | * as long as the inode doesn't go away and racy results are not a problem. | |
409 | */ | |
410 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
411 | { | |
412 | struct inode *inode = file_inode(vma->vm_file); | |
413 | struct shmem_inode_info *info = SHMEM_I(inode); | |
414 | struct address_space *mapping = inode->i_mapping; | |
415 | unsigned long swapped; | |
416 | ||
417 | /* Be careful as we don't hold info->lock */ | |
418 | swapped = READ_ONCE(info->swapped); | |
419 | ||
420 | /* | |
421 | * The easier cases are when the shmem object has nothing in swap, or | |
422 | * the vma maps it whole. Then we can simply use the stats that we | |
423 | * already track. | |
424 | */ | |
425 | if (!swapped) | |
426 | return 0; | |
427 | ||
428 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
429 | return swapped << PAGE_SHIFT; | |
430 | ||
431 | /* Here comes the more involved part */ | |
432 | return shmem_partial_swap_usage(mapping, | |
433 | linear_page_index(vma, vma->vm_start), | |
434 | linear_page_index(vma, vma->vm_end)); | |
435 | } | |
436 | ||
24513264 HD |
437 | /* |
438 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
439 | */ | |
440 | void shmem_unlock_mapping(struct address_space *mapping) | |
441 | { | |
442 | struct pagevec pvec; | |
443 | pgoff_t indices[PAGEVEC_SIZE]; | |
444 | pgoff_t index = 0; | |
445 | ||
446 | pagevec_init(&pvec, 0); | |
447 | /* | |
448 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
449 | */ | |
450 | while (!mapping_unevictable(mapping)) { | |
451 | /* | |
452 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
453 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
454 | */ | |
0cd6144a JW |
455 | pvec.nr = find_get_entries(mapping, index, |
456 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
457 | if (!pvec.nr) |
458 | break; | |
459 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 460 | pagevec_remove_exceptionals(&pvec); |
24513264 HD |
461 | check_move_unevictable_pages(pvec.pages, pvec.nr); |
462 | pagevec_release(&pvec); | |
463 | cond_resched(); | |
464 | } | |
7a5d0fbb HD |
465 | } |
466 | ||
467 | /* | |
468 | * Remove range of pages and swap entries from radix tree, and free them. | |
1635f6a7 | 469 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 470 | */ |
1635f6a7 HD |
471 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
472 | bool unfalloc) | |
1da177e4 | 473 | { |
285b2c4f | 474 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 475 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
476 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
477 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
478 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
479 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 480 | struct pagevec pvec; |
7a5d0fbb HD |
481 | pgoff_t indices[PAGEVEC_SIZE]; |
482 | long nr_swaps_freed = 0; | |
285b2c4f | 483 | pgoff_t index; |
bda97eab HD |
484 | int i; |
485 | ||
83e4fa9c HD |
486 | if (lend == -1) |
487 | end = -1; /* unsigned, so actually very big */ | |
bda97eab HD |
488 | |
489 | pagevec_init(&pvec, 0); | |
490 | index = start; | |
83e4fa9c | 491 | while (index < end) { |
0cd6144a JW |
492 | pvec.nr = find_get_entries(mapping, index, |
493 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
494 | pvec.pages, indices); | |
7a5d0fbb HD |
495 | if (!pvec.nr) |
496 | break; | |
bda97eab HD |
497 | for (i = 0; i < pagevec_count(&pvec); i++) { |
498 | struct page *page = pvec.pages[i]; | |
499 | ||
7a5d0fbb | 500 | index = indices[i]; |
83e4fa9c | 501 | if (index >= end) |
bda97eab HD |
502 | break; |
503 | ||
7a5d0fbb | 504 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
505 | if (unfalloc) |
506 | continue; | |
7a5d0fbb HD |
507 | nr_swaps_freed += !shmem_free_swap(mapping, |
508 | index, page); | |
bda97eab | 509 | continue; |
7a5d0fbb HD |
510 | } |
511 | ||
512 | if (!trylock_page(page)) | |
bda97eab | 513 | continue; |
1635f6a7 HD |
514 | if (!unfalloc || !PageUptodate(page)) { |
515 | if (page->mapping == mapping) { | |
309381fe | 516 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 HD |
517 | truncate_inode_page(mapping, page); |
518 | } | |
bda97eab | 519 | } |
bda97eab HD |
520 | unlock_page(page); |
521 | } | |
0cd6144a | 522 | pagevec_remove_exceptionals(&pvec); |
24513264 | 523 | pagevec_release(&pvec); |
bda97eab HD |
524 | cond_resched(); |
525 | index++; | |
526 | } | |
1da177e4 | 527 | |
83e4fa9c | 528 | if (partial_start) { |
bda97eab HD |
529 | struct page *page = NULL; |
530 | shmem_getpage(inode, start - 1, &page, SGP_READ, NULL); | |
531 | if (page) { | |
09cbfeaf | 532 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
533 | if (start > end) { |
534 | top = partial_end; | |
535 | partial_end = 0; | |
536 | } | |
537 | zero_user_segment(page, partial_start, top); | |
538 | set_page_dirty(page); | |
539 | unlock_page(page); | |
09cbfeaf | 540 | put_page(page); |
83e4fa9c HD |
541 | } |
542 | } | |
543 | if (partial_end) { | |
544 | struct page *page = NULL; | |
545 | shmem_getpage(inode, end, &page, SGP_READ, NULL); | |
546 | if (page) { | |
547 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
548 | set_page_dirty(page); |
549 | unlock_page(page); | |
09cbfeaf | 550 | put_page(page); |
bda97eab HD |
551 | } |
552 | } | |
83e4fa9c HD |
553 | if (start >= end) |
554 | return; | |
bda97eab HD |
555 | |
556 | index = start; | |
b1a36650 | 557 | while (index < end) { |
bda97eab | 558 | cond_resched(); |
0cd6144a JW |
559 | |
560 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 561 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 562 | pvec.pages, indices); |
7a5d0fbb | 563 | if (!pvec.nr) { |
b1a36650 HD |
564 | /* If all gone or hole-punch or unfalloc, we're done */ |
565 | if (index == start || end != -1) | |
bda97eab | 566 | break; |
b1a36650 | 567 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
568 | index = start; |
569 | continue; | |
570 | } | |
bda97eab HD |
571 | for (i = 0; i < pagevec_count(&pvec); i++) { |
572 | struct page *page = pvec.pages[i]; | |
573 | ||
7a5d0fbb | 574 | index = indices[i]; |
83e4fa9c | 575 | if (index >= end) |
bda97eab HD |
576 | break; |
577 | ||
7a5d0fbb | 578 | if (radix_tree_exceptional_entry(page)) { |
1635f6a7 HD |
579 | if (unfalloc) |
580 | continue; | |
b1a36650 HD |
581 | if (shmem_free_swap(mapping, index, page)) { |
582 | /* Swap was replaced by page: retry */ | |
583 | index--; | |
584 | break; | |
585 | } | |
586 | nr_swaps_freed++; | |
7a5d0fbb HD |
587 | continue; |
588 | } | |
589 | ||
bda97eab | 590 | lock_page(page); |
1635f6a7 HD |
591 | if (!unfalloc || !PageUptodate(page)) { |
592 | if (page->mapping == mapping) { | |
309381fe | 593 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1635f6a7 | 594 | truncate_inode_page(mapping, page); |
b1a36650 HD |
595 | } else { |
596 | /* Page was replaced by swap: retry */ | |
597 | unlock_page(page); | |
598 | index--; | |
599 | break; | |
1635f6a7 | 600 | } |
7a5d0fbb | 601 | } |
bda97eab HD |
602 | unlock_page(page); |
603 | } | |
0cd6144a | 604 | pagevec_remove_exceptionals(&pvec); |
24513264 | 605 | pagevec_release(&pvec); |
bda97eab HD |
606 | index++; |
607 | } | |
94c1e62d | 608 | |
1da177e4 | 609 | spin_lock(&info->lock); |
7a5d0fbb | 610 | info->swapped -= nr_swaps_freed; |
1da177e4 LT |
611 | shmem_recalc_inode(inode); |
612 | spin_unlock(&info->lock); | |
1635f6a7 | 613 | } |
1da177e4 | 614 | |
1635f6a7 HD |
615 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
616 | { | |
617 | shmem_undo_range(inode, lstart, lend, false); | |
285b2c4f | 618 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1da177e4 | 619 | } |
94c1e62d | 620 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 621 | |
44a30220 YZ |
622 | static int shmem_getattr(struct vfsmount *mnt, struct dentry *dentry, |
623 | struct kstat *stat) | |
624 | { | |
625 | struct inode *inode = dentry->d_inode; | |
626 | struct shmem_inode_info *info = SHMEM_I(inode); | |
627 | ||
d0424c42 HD |
628 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
629 | spin_lock(&info->lock); | |
630 | shmem_recalc_inode(inode); | |
631 | spin_unlock(&info->lock); | |
632 | } | |
44a30220 | 633 | generic_fillattr(inode, stat); |
44a30220 YZ |
634 | return 0; |
635 | } | |
636 | ||
94c1e62d | 637 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 638 | { |
75c3cfa8 | 639 | struct inode *inode = d_inode(dentry); |
40e041a2 | 640 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 LT |
641 | int error; |
642 | ||
db78b877 CH |
643 | error = inode_change_ok(inode, attr); |
644 | if (error) | |
645 | return error; | |
646 | ||
94c1e62d HD |
647 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
648 | loff_t oldsize = inode->i_size; | |
649 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 650 | |
40e041a2 DH |
651 | /* protected by i_mutex */ |
652 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
653 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
654 | return -EPERM; | |
655 | ||
94c1e62d | 656 | if (newsize != oldsize) { |
77142517 KK |
657 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
658 | oldsize, newsize); | |
659 | if (error) | |
660 | return error; | |
94c1e62d HD |
661 | i_size_write(inode, newsize); |
662 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
663 | } | |
afa2db2f | 664 | if (newsize <= oldsize) { |
94c1e62d | 665 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
666 | if (oldsize > holebegin) |
667 | unmap_mapping_range(inode->i_mapping, | |
668 | holebegin, 0, 1); | |
669 | if (info->alloced) | |
670 | shmem_truncate_range(inode, | |
671 | newsize, (loff_t)-1); | |
94c1e62d | 672 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
673 | if (oldsize > holebegin) |
674 | unmap_mapping_range(inode->i_mapping, | |
675 | holebegin, 0, 1); | |
94c1e62d | 676 | } |
1da177e4 LT |
677 | } |
678 | ||
db78b877 | 679 | setattr_copy(inode, attr); |
db78b877 | 680 | if (attr->ia_valid & ATTR_MODE) |
feda821e | 681 | error = posix_acl_chmod(inode, inode->i_mode); |
1da177e4 LT |
682 | return error; |
683 | } | |
684 | ||
1f895f75 | 685 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 686 | { |
1da177e4 LT |
687 | struct shmem_inode_info *info = SHMEM_I(inode); |
688 | ||
3889e6e7 | 689 | if (inode->i_mapping->a_ops == &shmem_aops) { |
1da177e4 LT |
690 | shmem_unacct_size(info->flags, inode->i_size); |
691 | inode->i_size = 0; | |
3889e6e7 | 692 | shmem_truncate_range(inode, 0, (loff_t)-1); |
1da177e4 | 693 | if (!list_empty(&info->swaplist)) { |
cb5f7b9a | 694 | mutex_lock(&shmem_swaplist_mutex); |
1da177e4 | 695 | list_del_init(&info->swaplist); |
cb5f7b9a | 696 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 697 | } |
3ed47db3 | 698 | } |
b09e0fa4 | 699 | |
38f38657 | 700 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 701 | WARN_ON(inode->i_blocks); |
5b04c689 | 702 | shmem_free_inode(inode->i_sb); |
dbd5768f | 703 | clear_inode(inode); |
1da177e4 LT |
704 | } |
705 | ||
46f65ec1 HD |
706 | /* |
707 | * If swap found in inode, free it and move page from swapcache to filecache. | |
708 | */ | |
41ffe5d5 | 709 | static int shmem_unuse_inode(struct shmem_inode_info *info, |
bde05d1c | 710 | swp_entry_t swap, struct page **pagep) |
1da177e4 | 711 | { |
285b2c4f | 712 | struct address_space *mapping = info->vfs_inode.i_mapping; |
46f65ec1 | 713 | void *radswap; |
41ffe5d5 | 714 | pgoff_t index; |
bde05d1c HD |
715 | gfp_t gfp; |
716 | int error = 0; | |
1da177e4 | 717 | |
46f65ec1 | 718 | radswap = swp_to_radix_entry(swap); |
e504f3fd | 719 | index = radix_tree_locate_item(&mapping->page_tree, radswap); |
46f65ec1 | 720 | if (index == -1) |
00501b53 | 721 | return -EAGAIN; /* tell shmem_unuse we found nothing */ |
2e0e26c7 | 722 | |
1b1b32f2 HD |
723 | /* |
724 | * Move _head_ to start search for next from here. | |
1f895f75 | 725 | * But be careful: shmem_evict_inode checks list_empty without taking |
1b1b32f2 | 726 | * mutex, and there's an instant in list_move_tail when info->swaplist |
285b2c4f | 727 | * would appear empty, if it were the only one on shmem_swaplist. |
1b1b32f2 HD |
728 | */ |
729 | if (shmem_swaplist.next != &info->swaplist) | |
730 | list_move_tail(&shmem_swaplist, &info->swaplist); | |
2e0e26c7 | 731 | |
bde05d1c HD |
732 | gfp = mapping_gfp_mask(mapping); |
733 | if (shmem_should_replace_page(*pagep, gfp)) { | |
734 | mutex_unlock(&shmem_swaplist_mutex); | |
735 | error = shmem_replace_page(pagep, gfp, info, index); | |
736 | mutex_lock(&shmem_swaplist_mutex); | |
737 | /* | |
738 | * We needed to drop mutex to make that restrictive page | |
0142ef6c HD |
739 | * allocation, but the inode might have been freed while we |
740 | * dropped it: although a racing shmem_evict_inode() cannot | |
741 | * complete without emptying the radix_tree, our page lock | |
742 | * on this swapcache page is not enough to prevent that - | |
743 | * free_swap_and_cache() of our swap entry will only | |
744 | * trylock_page(), removing swap from radix_tree whatever. | |
745 | * | |
746 | * We must not proceed to shmem_add_to_page_cache() if the | |
747 | * inode has been freed, but of course we cannot rely on | |
748 | * inode or mapping or info to check that. However, we can | |
749 | * safely check if our swap entry is still in use (and here | |
750 | * it can't have got reused for another page): if it's still | |
751 | * in use, then the inode cannot have been freed yet, and we | |
752 | * can safely proceed (if it's no longer in use, that tells | |
753 | * nothing about the inode, but we don't need to unuse swap). | |
bde05d1c HD |
754 | */ |
755 | if (!page_swapcount(*pagep)) | |
756 | error = -ENOENT; | |
757 | } | |
758 | ||
d13d1443 | 759 | /* |
778dd893 HD |
760 | * We rely on shmem_swaplist_mutex, not only to protect the swaplist, |
761 | * but also to hold up shmem_evict_inode(): so inode cannot be freed | |
762 | * beneath us (pagelock doesn't help until the page is in pagecache). | |
d13d1443 | 763 | */ |
bde05d1c HD |
764 | if (!error) |
765 | error = shmem_add_to_page_cache(*pagep, mapping, index, | |
fed400a1 | 766 | radswap); |
48f170fb | 767 | if (error != -ENOMEM) { |
46f65ec1 HD |
768 | /* |
769 | * Truncation and eviction use free_swap_and_cache(), which | |
770 | * only does trylock page: if we raced, best clean up here. | |
771 | */ | |
bde05d1c HD |
772 | delete_from_swap_cache(*pagep); |
773 | set_page_dirty(*pagep); | |
46f65ec1 HD |
774 | if (!error) { |
775 | spin_lock(&info->lock); | |
776 | info->swapped--; | |
777 | spin_unlock(&info->lock); | |
778 | swap_free(swap); | |
779 | } | |
1da177e4 | 780 | } |
2e0e26c7 | 781 | return error; |
1da177e4 LT |
782 | } |
783 | ||
784 | /* | |
46f65ec1 | 785 | * Search through swapped inodes to find and replace swap by page. |
1da177e4 | 786 | */ |
41ffe5d5 | 787 | int shmem_unuse(swp_entry_t swap, struct page *page) |
1da177e4 | 788 | { |
41ffe5d5 | 789 | struct list_head *this, *next; |
1da177e4 | 790 | struct shmem_inode_info *info; |
00501b53 | 791 | struct mem_cgroup *memcg; |
bde05d1c HD |
792 | int error = 0; |
793 | ||
794 | /* | |
795 | * There's a faint possibility that swap page was replaced before | |
0142ef6c | 796 | * caller locked it: caller will come back later with the right page. |
bde05d1c | 797 | */ |
0142ef6c | 798 | if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val)) |
bde05d1c | 799 | goto out; |
778dd893 HD |
800 | |
801 | /* | |
802 | * Charge page using GFP_KERNEL while we can wait, before taking | |
803 | * the shmem_swaplist_mutex which might hold up shmem_writepage(). | |
804 | * Charged back to the user (not to caller) when swap account is used. | |
778dd893 | 805 | */ |
f627c2f5 KS |
806 | error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg, |
807 | false); | |
778dd893 HD |
808 | if (error) |
809 | goto out; | |
46f65ec1 | 810 | /* No radix_tree_preload: swap entry keeps a place for page in tree */ |
00501b53 | 811 | error = -EAGAIN; |
1da177e4 | 812 | |
cb5f7b9a | 813 | mutex_lock(&shmem_swaplist_mutex); |
41ffe5d5 HD |
814 | list_for_each_safe(this, next, &shmem_swaplist) { |
815 | info = list_entry(this, struct shmem_inode_info, swaplist); | |
285b2c4f | 816 | if (info->swapped) |
00501b53 | 817 | error = shmem_unuse_inode(info, swap, &page); |
6922c0c7 HD |
818 | else |
819 | list_del_init(&info->swaplist); | |
cb5f7b9a | 820 | cond_resched(); |
00501b53 | 821 | if (error != -EAGAIN) |
778dd893 | 822 | break; |
00501b53 | 823 | /* found nothing in this: move on to search the next */ |
1da177e4 | 824 | } |
cb5f7b9a | 825 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 826 | |
00501b53 JW |
827 | if (error) { |
828 | if (error != -ENOMEM) | |
829 | error = 0; | |
f627c2f5 | 830 | mem_cgroup_cancel_charge(page, memcg, false); |
00501b53 | 831 | } else |
f627c2f5 | 832 | mem_cgroup_commit_charge(page, memcg, true, false); |
778dd893 | 833 | out: |
aaa46865 | 834 | unlock_page(page); |
09cbfeaf | 835 | put_page(page); |
778dd893 | 836 | return error; |
1da177e4 LT |
837 | } |
838 | ||
839 | /* | |
840 | * Move the page from the page cache to the swap cache. | |
841 | */ | |
842 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
843 | { | |
844 | struct shmem_inode_info *info; | |
1da177e4 | 845 | struct address_space *mapping; |
1da177e4 | 846 | struct inode *inode; |
6922c0c7 HD |
847 | swp_entry_t swap; |
848 | pgoff_t index; | |
1da177e4 LT |
849 | |
850 | BUG_ON(!PageLocked(page)); | |
1da177e4 LT |
851 | mapping = page->mapping; |
852 | index = page->index; | |
853 | inode = mapping->host; | |
854 | info = SHMEM_I(inode); | |
855 | if (info->flags & VM_LOCKED) | |
856 | goto redirty; | |
d9fe526a | 857 | if (!total_swap_pages) |
1da177e4 LT |
858 | goto redirty; |
859 | ||
d9fe526a | 860 | /* |
97b713ba CH |
861 | * Our capabilities prevent regular writeback or sync from ever calling |
862 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
863 | * its underlying filesystem, in which case tmpfs should write out to | |
864 | * swap only in response to memory pressure, and not for the writeback | |
865 | * threads or sync. | |
d9fe526a | 866 | */ |
48f170fb HD |
867 | if (!wbc->for_reclaim) { |
868 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
869 | goto redirty; | |
870 | } | |
1635f6a7 HD |
871 | |
872 | /* | |
873 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
874 | * value into swapfile.c, the only way we can correctly account for a | |
875 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
876 | * |
877 | * That's okay for a page already fallocated earlier, but if we have | |
878 | * not yet completed the fallocation, then (a) we want to keep track | |
879 | * of this page in case we have to undo it, and (b) it may not be a | |
880 | * good idea to continue anyway, once we're pushing into swap. So | |
881 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
882 | */ |
883 | if (!PageUptodate(page)) { | |
1aac1400 HD |
884 | if (inode->i_private) { |
885 | struct shmem_falloc *shmem_falloc; | |
886 | spin_lock(&inode->i_lock); | |
887 | shmem_falloc = inode->i_private; | |
888 | if (shmem_falloc && | |
8e205f77 | 889 | !shmem_falloc->waitq && |
1aac1400 HD |
890 | index >= shmem_falloc->start && |
891 | index < shmem_falloc->next) | |
892 | shmem_falloc->nr_unswapped++; | |
893 | else | |
894 | shmem_falloc = NULL; | |
895 | spin_unlock(&inode->i_lock); | |
896 | if (shmem_falloc) | |
897 | goto redirty; | |
898 | } | |
1635f6a7 HD |
899 | clear_highpage(page); |
900 | flush_dcache_page(page); | |
901 | SetPageUptodate(page); | |
902 | } | |
903 | ||
48f170fb HD |
904 | swap = get_swap_page(); |
905 | if (!swap.val) | |
906 | goto redirty; | |
d9fe526a | 907 | |
37e84351 VD |
908 | if (mem_cgroup_try_charge_swap(page, swap)) |
909 | goto free_swap; | |
910 | ||
b1dea800 HD |
911 | /* |
912 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
913 | * if it's not already there. Do it now before the page is |
914 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 915 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
916 | * we've incremented swapped, because shmem_unuse_inode() will |
917 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 918 | */ |
48f170fb HD |
919 | mutex_lock(&shmem_swaplist_mutex); |
920 | if (list_empty(&info->swaplist)) | |
921 | list_add_tail(&info->swaplist, &shmem_swaplist); | |
b1dea800 | 922 | |
48f170fb | 923 | if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { |
6922c0c7 | 924 | spin_lock(&info->lock); |
6922c0c7 | 925 | shmem_recalc_inode(inode); |
267a4c76 | 926 | info->swapped++; |
826267cf | 927 | spin_unlock(&info->lock); |
6922c0c7 | 928 | |
267a4c76 HD |
929 | swap_shmem_alloc(swap); |
930 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
931 | ||
6922c0c7 | 932 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 933 | BUG_ON(page_mapped(page)); |
9fab5619 | 934 | swap_writepage(page, wbc); |
1da177e4 LT |
935 | return 0; |
936 | } | |
937 | ||
6922c0c7 | 938 | mutex_unlock(&shmem_swaplist_mutex); |
37e84351 | 939 | free_swap: |
0a31bc97 | 940 | swapcache_free(swap); |
1da177e4 LT |
941 | redirty: |
942 | set_page_dirty(page); | |
d9fe526a HD |
943 | if (wbc->for_reclaim) |
944 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
945 | unlock_page(page); | |
946 | return 0; | |
1da177e4 LT |
947 | } |
948 | ||
75edd345 | 949 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 950 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 951 | { |
095f1fc4 | 952 | char buffer[64]; |
680d794b | 953 | |
71fe804b | 954 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 955 | return; /* show nothing */ |
680d794b | 956 | |
a7a88b23 | 957 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
958 | |
959 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 960 | } |
71fe804b LS |
961 | |
962 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
963 | { | |
964 | struct mempolicy *mpol = NULL; | |
965 | if (sbinfo->mpol) { | |
966 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
967 | mpol = sbinfo->mpol; | |
968 | mpol_get(mpol); | |
969 | spin_unlock(&sbinfo->stat_lock); | |
970 | } | |
971 | return mpol; | |
972 | } | |
75edd345 HD |
973 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
974 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
975 | { | |
976 | } | |
977 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
978 | { | |
979 | return NULL; | |
980 | } | |
981 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
982 | #ifndef CONFIG_NUMA | |
983 | #define vm_policy vm_private_data | |
984 | #endif | |
680d794b | 985 | |
41ffe5d5 HD |
986 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
987 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 988 | { |
1da177e4 | 989 | struct vm_area_struct pvma; |
18a2f371 | 990 | struct page *page; |
52cd3b07 | 991 | |
1da177e4 | 992 | /* Create a pseudo vma that just contains the policy */ |
c4cc6d07 | 993 | pvma.vm_start = 0; |
09c231cb NZ |
994 | /* Bias interleave by inode number to distribute better across nodes */ |
995 | pvma.vm_pgoff = index + info->vfs_inode.i_ino; | |
c4cc6d07 | 996 | pvma.vm_ops = NULL; |
18a2f371 MG |
997 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
998 | ||
999 | page = swapin_readahead(swap, gfp, &pvma, 0); | |
1000 | ||
1001 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1002 | mpol_cond_put(pvma.vm_policy); | |
1003 | ||
1004 | return page; | |
1da177e4 LT |
1005 | } |
1006 | ||
02098fea | 1007 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1008 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1009 | { |
1010 | struct vm_area_struct pvma; | |
18a2f371 | 1011 | struct page *page; |
1da177e4 | 1012 | |
c4cc6d07 HD |
1013 | /* Create a pseudo vma that just contains the policy */ |
1014 | pvma.vm_start = 0; | |
09c231cb NZ |
1015 | /* Bias interleave by inode number to distribute better across nodes */ |
1016 | pvma.vm_pgoff = index + info->vfs_inode.i_ino; | |
c4cc6d07 | 1017 | pvma.vm_ops = NULL; |
41ffe5d5 | 1018 | pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
52cd3b07 | 1019 | |
75edd345 HD |
1020 | page = alloc_pages_vma(gfp, 0, &pvma, 0, numa_node_id(), false); |
1021 | if (page) { | |
1022 | __SetPageLocked(page); | |
1023 | __SetPageSwapBacked(page); | |
1024 | } | |
18a2f371 MG |
1025 | |
1026 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1027 | mpol_cond_put(pvma.vm_policy); | |
1028 | ||
1029 | return page; | |
1da177e4 | 1030 | } |
71fe804b | 1031 | |
bde05d1c HD |
1032 | /* |
1033 | * When a page is moved from swapcache to shmem filecache (either by the | |
1034 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1035 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1036 | * ignorance of the mapping it belongs to. If that mapping has special | |
1037 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1038 | * we may need to copy to a suitable page before moving to filecache. | |
1039 | * | |
1040 | * In a future release, this may well be extended to respect cpuset and | |
1041 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1042 | * but for now it is a simple matter of zone. | |
1043 | */ | |
1044 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1045 | { | |
1046 | return page_zonenum(page) > gfp_zone(gfp); | |
1047 | } | |
1048 | ||
1049 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1050 | struct shmem_inode_info *info, pgoff_t index) | |
1051 | { | |
1052 | struct page *oldpage, *newpage; | |
1053 | struct address_space *swap_mapping; | |
1054 | pgoff_t swap_index; | |
1055 | int error; | |
1056 | ||
1057 | oldpage = *pagep; | |
1058 | swap_index = page_private(oldpage); | |
1059 | swap_mapping = page_mapping(oldpage); | |
1060 | ||
1061 | /* | |
1062 | * We have arrived here because our zones are constrained, so don't | |
1063 | * limit chance of success by further cpuset and node constraints. | |
1064 | */ | |
1065 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1066 | newpage = shmem_alloc_page(gfp, info, index); | |
1067 | if (!newpage) | |
1068 | return -ENOMEM; | |
bde05d1c | 1069 | |
09cbfeaf | 1070 | get_page(newpage); |
bde05d1c | 1071 | copy_highpage(newpage, oldpage); |
0142ef6c | 1072 | flush_dcache_page(newpage); |
bde05d1c | 1073 | |
bde05d1c | 1074 | SetPageUptodate(newpage); |
bde05d1c | 1075 | set_page_private(newpage, swap_index); |
bde05d1c HD |
1076 | SetPageSwapCache(newpage); |
1077 | ||
1078 | /* | |
1079 | * Our caller will very soon move newpage out of swapcache, but it's | |
1080 | * a nice clean interface for us to replace oldpage by newpage there. | |
1081 | */ | |
1082 | spin_lock_irq(&swap_mapping->tree_lock); | |
1083 | error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage, | |
1084 | newpage); | |
0142ef6c HD |
1085 | if (!error) { |
1086 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
1087 | __dec_zone_page_state(oldpage, NR_FILE_PAGES); | |
1088 | } | |
bde05d1c | 1089 | spin_unlock_irq(&swap_mapping->tree_lock); |
bde05d1c | 1090 | |
0142ef6c HD |
1091 | if (unlikely(error)) { |
1092 | /* | |
1093 | * Is this possible? I think not, now that our callers check | |
1094 | * both PageSwapCache and page_private after getting page lock; | |
1095 | * but be defensive. Reverse old to newpage for clear and free. | |
1096 | */ | |
1097 | oldpage = newpage; | |
1098 | } else { | |
6a93ca8f | 1099 | mem_cgroup_migrate(oldpage, newpage); |
0142ef6c HD |
1100 | lru_cache_add_anon(newpage); |
1101 | *pagep = newpage; | |
1102 | } | |
bde05d1c HD |
1103 | |
1104 | ClearPageSwapCache(oldpage); | |
1105 | set_page_private(oldpage, 0); | |
1106 | ||
1107 | unlock_page(oldpage); | |
09cbfeaf KS |
1108 | put_page(oldpage); |
1109 | put_page(oldpage); | |
0142ef6c | 1110 | return error; |
bde05d1c HD |
1111 | } |
1112 | ||
1da177e4 | 1113 | /* |
68da9f05 | 1114 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1115 | * |
1116 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1117 | * vm. If we swap it in we mark it dirty since we also free the swap | |
1118 | * entry since a page cannot live in both the swap and page cache | |
1119 | */ | |
41ffe5d5 | 1120 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
68da9f05 | 1121 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type) |
1da177e4 LT |
1122 | { |
1123 | struct address_space *mapping = inode->i_mapping; | |
54af6042 | 1124 | struct shmem_inode_info *info; |
1da177e4 | 1125 | struct shmem_sb_info *sbinfo; |
00501b53 | 1126 | struct mem_cgroup *memcg; |
27ab7006 | 1127 | struct page *page; |
1da177e4 LT |
1128 | swp_entry_t swap; |
1129 | int error; | |
54af6042 | 1130 | int once = 0; |
1635f6a7 | 1131 | int alloced = 0; |
1da177e4 | 1132 | |
09cbfeaf | 1133 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1134 | return -EFBIG; |
1da177e4 | 1135 | repeat: |
54af6042 | 1136 | swap.val = 0; |
0cd6144a | 1137 | page = find_lock_entry(mapping, index); |
54af6042 HD |
1138 | if (radix_tree_exceptional_entry(page)) { |
1139 | swap = radix_to_swp_entry(page); | |
1140 | page = NULL; | |
1141 | } | |
1142 | ||
75edd345 | 1143 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1144 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 1145 | error = -EINVAL; |
267a4c76 | 1146 | goto unlock; |
54af6042 HD |
1147 | } |
1148 | ||
66d2f4d2 HD |
1149 | if (page && sgp == SGP_WRITE) |
1150 | mark_page_accessed(page); | |
1151 | ||
1635f6a7 HD |
1152 | /* fallocated page? */ |
1153 | if (page && !PageUptodate(page)) { | |
1154 | if (sgp != SGP_READ) | |
1155 | goto clear; | |
1156 | unlock_page(page); | |
09cbfeaf | 1157 | put_page(page); |
1635f6a7 HD |
1158 | page = NULL; |
1159 | } | |
54af6042 | 1160 | if (page || (sgp == SGP_READ && !swap.val)) { |
54af6042 HD |
1161 | *pagep = page; |
1162 | return 0; | |
27ab7006 HD |
1163 | } |
1164 | ||
1165 | /* | |
54af6042 HD |
1166 | * Fast cache lookup did not find it: |
1167 | * bring it back from swap or allocate. | |
27ab7006 | 1168 | */ |
54af6042 HD |
1169 | info = SHMEM_I(inode); |
1170 | sbinfo = SHMEM_SB(inode->i_sb); | |
1da177e4 | 1171 | |
1da177e4 LT |
1172 | if (swap.val) { |
1173 | /* Look it up and read it in.. */ | |
27ab7006 HD |
1174 | page = lookup_swap_cache(swap); |
1175 | if (!page) { | |
1da177e4 | 1176 | /* here we actually do the io */ |
68da9f05 HD |
1177 | if (fault_type) |
1178 | *fault_type |= VM_FAULT_MAJOR; | |
41ffe5d5 | 1179 | page = shmem_swapin(swap, gfp, info, index); |
27ab7006 | 1180 | if (!page) { |
54af6042 HD |
1181 | error = -ENOMEM; |
1182 | goto failed; | |
1da177e4 | 1183 | } |
1da177e4 LT |
1184 | } |
1185 | ||
1186 | /* We have to do this with page locked to prevent races */ | |
54af6042 | 1187 | lock_page(page); |
0142ef6c | 1188 | if (!PageSwapCache(page) || page_private(page) != swap.val || |
d1899228 | 1189 | !shmem_confirm_swap(mapping, index, swap)) { |
bde05d1c | 1190 | error = -EEXIST; /* try again */ |
d1899228 | 1191 | goto unlock; |
bde05d1c | 1192 | } |
27ab7006 | 1193 | if (!PageUptodate(page)) { |
1da177e4 | 1194 | error = -EIO; |
54af6042 | 1195 | goto failed; |
1da177e4 | 1196 | } |
54af6042 HD |
1197 | wait_on_page_writeback(page); |
1198 | ||
bde05d1c HD |
1199 | if (shmem_should_replace_page(page, gfp)) { |
1200 | error = shmem_replace_page(&page, gfp, info, index); | |
1201 | if (error) | |
1202 | goto failed; | |
1da177e4 | 1203 | } |
27ab7006 | 1204 | |
f627c2f5 KS |
1205 | error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg, |
1206 | false); | |
d1899228 | 1207 | if (!error) { |
aa3b1895 | 1208 | error = shmem_add_to_page_cache(page, mapping, index, |
fed400a1 | 1209 | swp_to_radix_entry(swap)); |
215c02bc HD |
1210 | /* |
1211 | * We already confirmed swap under page lock, and make | |
1212 | * no memory allocation here, so usually no possibility | |
1213 | * of error; but free_swap_and_cache() only trylocks a | |
1214 | * page, so it is just possible that the entry has been | |
1215 | * truncated or holepunched since swap was confirmed. | |
1216 | * shmem_undo_range() will have done some of the | |
1217 | * unaccounting, now delete_from_swap_cache() will do | |
93aa7d95 | 1218 | * the rest. |
215c02bc HD |
1219 | * Reset swap.val? No, leave it so "failed" goes back to |
1220 | * "repeat": reading a hole and writing should succeed. | |
1221 | */ | |
00501b53 | 1222 | if (error) { |
f627c2f5 | 1223 | mem_cgroup_cancel_charge(page, memcg, false); |
215c02bc | 1224 | delete_from_swap_cache(page); |
00501b53 | 1225 | } |
d1899228 | 1226 | } |
54af6042 HD |
1227 | if (error) |
1228 | goto failed; | |
1229 | ||
f627c2f5 | 1230 | mem_cgroup_commit_charge(page, memcg, true, false); |
00501b53 | 1231 | |
54af6042 | 1232 | spin_lock(&info->lock); |
285b2c4f | 1233 | info->swapped--; |
54af6042 | 1234 | shmem_recalc_inode(inode); |
27ab7006 | 1235 | spin_unlock(&info->lock); |
54af6042 | 1236 | |
66d2f4d2 HD |
1237 | if (sgp == SGP_WRITE) |
1238 | mark_page_accessed(page); | |
1239 | ||
54af6042 | 1240 | delete_from_swap_cache(page); |
27ab7006 HD |
1241 | set_page_dirty(page); |
1242 | swap_free(swap); | |
1243 | ||
54af6042 HD |
1244 | } else { |
1245 | if (shmem_acct_block(info->flags)) { | |
1246 | error = -ENOSPC; | |
1247 | goto failed; | |
1da177e4 | 1248 | } |
0edd73b3 | 1249 | if (sbinfo->max_blocks) { |
fc5da22a | 1250 | if (percpu_counter_compare(&sbinfo->used_blocks, |
54af6042 HD |
1251 | sbinfo->max_blocks) >= 0) { |
1252 | error = -ENOSPC; | |
1253 | goto unacct; | |
1254 | } | |
7e496299 | 1255 | percpu_counter_inc(&sbinfo->used_blocks); |
54af6042 | 1256 | } |
1da177e4 | 1257 | |
54af6042 HD |
1258 | page = shmem_alloc_page(gfp, info, index); |
1259 | if (!page) { | |
1260 | error = -ENOMEM; | |
1261 | goto decused; | |
1da177e4 | 1262 | } |
66d2f4d2 | 1263 | if (sgp == SGP_WRITE) |
eb39d618 | 1264 | __SetPageReferenced(page); |
66d2f4d2 | 1265 | |
f627c2f5 KS |
1266 | error = mem_cgroup_try_charge(page, current->mm, gfp, &memcg, |
1267 | false); | |
54af6042 HD |
1268 | if (error) |
1269 | goto decused; | |
5e4c0d97 | 1270 | error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK); |
b065b432 HD |
1271 | if (!error) { |
1272 | error = shmem_add_to_page_cache(page, mapping, index, | |
fed400a1 | 1273 | NULL); |
b065b432 HD |
1274 | radix_tree_preload_end(); |
1275 | } | |
1276 | if (error) { | |
f627c2f5 | 1277 | mem_cgroup_cancel_charge(page, memcg, false); |
b065b432 HD |
1278 | goto decused; |
1279 | } | |
f627c2f5 | 1280 | mem_cgroup_commit_charge(page, memcg, false, false); |
54af6042 HD |
1281 | lru_cache_add_anon(page); |
1282 | ||
1283 | spin_lock(&info->lock); | |
1da177e4 | 1284 | info->alloced++; |
54af6042 HD |
1285 | inode->i_blocks += BLOCKS_PER_PAGE; |
1286 | shmem_recalc_inode(inode); | |
1da177e4 | 1287 | spin_unlock(&info->lock); |
1635f6a7 | 1288 | alloced = true; |
54af6042 | 1289 | |
ec9516fb | 1290 | /* |
1635f6a7 HD |
1291 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. |
1292 | */ | |
1293 | if (sgp == SGP_FALLOC) | |
1294 | sgp = SGP_WRITE; | |
1295 | clear: | |
1296 | /* | |
1297 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1298 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1299 | * it now, lest undo on failure cancel our earlier guarantee. | |
ec9516fb HD |
1300 | */ |
1301 | if (sgp != SGP_WRITE) { | |
1302 | clear_highpage(page); | |
1303 | flush_dcache_page(page); | |
1304 | SetPageUptodate(page); | |
1305 | } | |
1da177e4 | 1306 | } |
bde05d1c | 1307 | |
54af6042 | 1308 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1309 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1310 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1311 | if (alloced) { |
1312 | ClearPageDirty(page); | |
1313 | delete_from_page_cache(page); | |
1314 | spin_lock(&info->lock); | |
1315 | shmem_recalc_inode(inode); | |
1316 | spin_unlock(&info->lock); | |
1317 | } | |
54af6042 | 1318 | error = -EINVAL; |
267a4c76 | 1319 | goto unlock; |
e83c32e8 | 1320 | } |
54af6042 HD |
1321 | *pagep = page; |
1322 | return 0; | |
1da177e4 | 1323 | |
59a16ead | 1324 | /* |
54af6042 | 1325 | * Error recovery. |
59a16ead | 1326 | */ |
54af6042 HD |
1327 | decused: |
1328 | if (sbinfo->max_blocks) | |
1329 | percpu_counter_add(&sbinfo->used_blocks, -1); | |
1330 | unacct: | |
1331 | shmem_unacct_blocks(info->flags, 1); | |
1332 | failed: | |
267a4c76 | 1333 | if (swap.val && !shmem_confirm_swap(mapping, index, swap)) |
d1899228 HD |
1334 | error = -EEXIST; |
1335 | unlock: | |
27ab7006 | 1336 | if (page) { |
54af6042 | 1337 | unlock_page(page); |
09cbfeaf | 1338 | put_page(page); |
54af6042 HD |
1339 | } |
1340 | if (error == -ENOSPC && !once++) { | |
1341 | info = SHMEM_I(inode); | |
1342 | spin_lock(&info->lock); | |
1343 | shmem_recalc_inode(inode); | |
1344 | spin_unlock(&info->lock); | |
27ab7006 | 1345 | goto repeat; |
ff36b801 | 1346 | } |
d1899228 | 1347 | if (error == -EEXIST) /* from above or from radix_tree_insert */ |
54af6042 HD |
1348 | goto repeat; |
1349 | return error; | |
1da177e4 LT |
1350 | } |
1351 | ||
d0217ac0 | 1352 | static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 | 1353 | { |
496ad9aa | 1354 | struct inode *inode = file_inode(vma->vm_file); |
1da177e4 | 1355 | int error; |
68da9f05 | 1356 | int ret = VM_FAULT_LOCKED; |
1da177e4 | 1357 | |
f00cdc6d HD |
1358 | /* |
1359 | * Trinity finds that probing a hole which tmpfs is punching can | |
1360 | * prevent the hole-punch from ever completing: which in turn | |
1361 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
1362 | * faulting pages into the hole while it's being punched. Although |
1363 | * shmem_undo_range() does remove the additions, it may be unable to | |
1364 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
1365 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
1366 | * | |
1367 | * It does not matter if we sometimes reach this check just before the | |
1368 | * hole-punch begins, so that one fault then races with the punch: | |
1369 | * we just need to make racing faults a rare case. | |
1370 | * | |
1371 | * The implementation below would be much simpler if we just used a | |
1372 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
1373 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
1374 | */ |
1375 | if (unlikely(inode->i_private)) { | |
1376 | struct shmem_falloc *shmem_falloc; | |
1377 | ||
1378 | spin_lock(&inode->i_lock); | |
1379 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
1380 | if (shmem_falloc && |
1381 | shmem_falloc->waitq && | |
1382 | vmf->pgoff >= shmem_falloc->start && | |
1383 | vmf->pgoff < shmem_falloc->next) { | |
1384 | wait_queue_head_t *shmem_falloc_waitq; | |
1385 | DEFINE_WAIT(shmem_fault_wait); | |
1386 | ||
1387 | ret = VM_FAULT_NOPAGE; | |
f00cdc6d HD |
1388 | if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) && |
1389 | !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) { | |
8e205f77 | 1390 | /* It's polite to up mmap_sem if we can */ |
f00cdc6d | 1391 | up_read(&vma->vm_mm->mmap_sem); |
8e205f77 | 1392 | ret = VM_FAULT_RETRY; |
f00cdc6d | 1393 | } |
8e205f77 HD |
1394 | |
1395 | shmem_falloc_waitq = shmem_falloc->waitq; | |
1396 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
1397 | TASK_UNINTERRUPTIBLE); | |
1398 | spin_unlock(&inode->i_lock); | |
1399 | schedule(); | |
1400 | ||
1401 | /* | |
1402 | * shmem_falloc_waitq points into the shmem_fallocate() | |
1403 | * stack of the hole-punching task: shmem_falloc_waitq | |
1404 | * is usually invalid by the time we reach here, but | |
1405 | * finish_wait() does not dereference it in that case; | |
1406 | * though i_lock needed lest racing with wake_up_all(). | |
1407 | */ | |
1408 | spin_lock(&inode->i_lock); | |
1409 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
1410 | spin_unlock(&inode->i_lock); | |
1411 | return ret; | |
f00cdc6d | 1412 | } |
8e205f77 | 1413 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
1414 | } |
1415 | ||
27d54b39 | 1416 | error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); |
d0217ac0 NP |
1417 | if (error) |
1418 | return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); | |
68da9f05 | 1419 | |
456f998e YH |
1420 | if (ret & VM_FAULT_MAJOR) { |
1421 | count_vm_event(PGMAJFAULT); | |
1422 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); | |
1423 | } | |
68da9f05 | 1424 | return ret; |
1da177e4 LT |
1425 | } |
1426 | ||
1da177e4 | 1427 | #ifdef CONFIG_NUMA |
41ffe5d5 | 1428 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 1429 | { |
496ad9aa | 1430 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1431 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
1432 | } |
1433 | ||
d8dc74f2 AB |
1434 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
1435 | unsigned long addr) | |
1da177e4 | 1436 | { |
496ad9aa | 1437 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 1438 | pgoff_t index; |
1da177e4 | 1439 | |
41ffe5d5 HD |
1440 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
1441 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
1442 | } |
1443 | #endif | |
1444 | ||
1445 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
1446 | { | |
496ad9aa | 1447 | struct inode *inode = file_inode(file); |
1da177e4 LT |
1448 | struct shmem_inode_info *info = SHMEM_I(inode); |
1449 | int retval = -ENOMEM; | |
1450 | ||
1451 | spin_lock(&info->lock); | |
1452 | if (lock && !(info->flags & VM_LOCKED)) { | |
1453 | if (!user_shm_lock(inode->i_size, user)) | |
1454 | goto out_nomem; | |
1455 | info->flags |= VM_LOCKED; | |
89e004ea | 1456 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
1457 | } |
1458 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
1459 | user_shm_unlock(inode->i_size, user); | |
1460 | info->flags &= ~VM_LOCKED; | |
89e004ea | 1461 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
1462 | } |
1463 | retval = 0; | |
89e004ea | 1464 | |
1da177e4 LT |
1465 | out_nomem: |
1466 | spin_unlock(&info->lock); | |
1467 | return retval; | |
1468 | } | |
1469 | ||
9b83a6a8 | 1470 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
1471 | { |
1472 | file_accessed(file); | |
1473 | vma->vm_ops = &shmem_vm_ops; | |
1474 | return 0; | |
1475 | } | |
1476 | ||
454abafe | 1477 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 1478 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
1479 | { |
1480 | struct inode *inode; | |
1481 | struct shmem_inode_info *info; | |
1482 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
1483 | ||
5b04c689 PE |
1484 | if (shmem_reserve_inode(sb)) |
1485 | return NULL; | |
1da177e4 LT |
1486 | |
1487 | inode = new_inode(sb); | |
1488 | if (inode) { | |
85fe4025 | 1489 | inode->i_ino = get_next_ino(); |
454abafe | 1490 | inode_init_owner(inode, dir, mode); |
1da177e4 | 1491 | inode->i_blocks = 0; |
1da177e4 | 1492 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
91828a40 | 1493 | inode->i_generation = get_seconds(); |
1da177e4 LT |
1494 | info = SHMEM_I(inode); |
1495 | memset(info, 0, (char *)inode - (char *)info); | |
1496 | spin_lock_init(&info->lock); | |
40e041a2 | 1497 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 1498 | info->flags = flags & VM_NORESERVE; |
1da177e4 | 1499 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 1500 | simple_xattrs_init(&info->xattrs); |
72c04902 | 1501 | cache_no_acl(inode); |
1da177e4 LT |
1502 | |
1503 | switch (mode & S_IFMT) { | |
1504 | default: | |
39f0247d | 1505 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
1506 | init_special_inode(inode, mode, dev); |
1507 | break; | |
1508 | case S_IFREG: | |
14fcc23f | 1509 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
1510 | inode->i_op = &shmem_inode_operations; |
1511 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
1512 | mpol_shared_policy_init(&info->policy, |
1513 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
1514 | break; |
1515 | case S_IFDIR: | |
d8c76e6f | 1516 | inc_nlink(inode); |
1da177e4 LT |
1517 | /* Some things misbehave if size == 0 on a directory */ |
1518 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
1519 | inode->i_op = &shmem_dir_inode_operations; | |
1520 | inode->i_fop = &simple_dir_operations; | |
1521 | break; | |
1522 | case S_IFLNK: | |
1523 | /* | |
1524 | * Must not load anything in the rbtree, | |
1525 | * mpol_free_shared_policy will not be called. | |
1526 | */ | |
71fe804b | 1527 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
1528 | break; |
1529 | } | |
5b04c689 PE |
1530 | } else |
1531 | shmem_free_inode(sb); | |
1da177e4 LT |
1532 | return inode; |
1533 | } | |
1534 | ||
0cd6144a JW |
1535 | bool shmem_mapping(struct address_space *mapping) |
1536 | { | |
f0774d88 SL |
1537 | if (!mapping->host) |
1538 | return false; | |
1539 | ||
97b713ba | 1540 | return mapping->host->i_sb->s_op == &shmem_ops; |
0cd6144a JW |
1541 | } |
1542 | ||
1da177e4 | 1543 | #ifdef CONFIG_TMPFS |
92e1d5be | 1544 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 1545 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 1546 | |
6d9d88d0 JS |
1547 | #ifdef CONFIG_TMPFS_XATTR |
1548 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
1549 | #else | |
1550 | #define shmem_initxattrs NULL | |
1551 | #endif | |
1552 | ||
1da177e4 | 1553 | static int |
800d15a5 NP |
1554 | shmem_write_begin(struct file *file, struct address_space *mapping, |
1555 | loff_t pos, unsigned len, unsigned flags, | |
1556 | struct page **pagep, void **fsdata) | |
1da177e4 | 1557 | { |
800d15a5 | 1558 | struct inode *inode = mapping->host; |
40e041a2 | 1559 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 1560 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
1561 | |
1562 | /* i_mutex is held by caller */ | |
1563 | if (unlikely(info->seals)) { | |
1564 | if (info->seals & F_SEAL_WRITE) | |
1565 | return -EPERM; | |
1566 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
1567 | return -EPERM; | |
1568 | } | |
1569 | ||
66d2f4d2 | 1570 | return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); |
800d15a5 NP |
1571 | } |
1572 | ||
1573 | static int | |
1574 | shmem_write_end(struct file *file, struct address_space *mapping, | |
1575 | loff_t pos, unsigned len, unsigned copied, | |
1576 | struct page *page, void *fsdata) | |
1577 | { | |
1578 | struct inode *inode = mapping->host; | |
1579 | ||
d3602444 HD |
1580 | if (pos + copied > inode->i_size) |
1581 | i_size_write(inode, pos + copied); | |
1582 | ||
ec9516fb | 1583 | if (!PageUptodate(page)) { |
09cbfeaf KS |
1584 | if (copied < PAGE_SIZE) { |
1585 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 1586 | zero_user_segments(page, 0, from, |
09cbfeaf | 1587 | from + copied, PAGE_SIZE); |
ec9516fb HD |
1588 | } |
1589 | SetPageUptodate(page); | |
1590 | } | |
800d15a5 | 1591 | set_page_dirty(page); |
6746aff7 | 1592 | unlock_page(page); |
09cbfeaf | 1593 | put_page(page); |
800d15a5 | 1594 | |
800d15a5 | 1595 | return copied; |
1da177e4 LT |
1596 | } |
1597 | ||
2ba5bbed | 1598 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 1599 | { |
6e58e79d AV |
1600 | struct file *file = iocb->ki_filp; |
1601 | struct inode *inode = file_inode(file); | |
1da177e4 | 1602 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
1603 | pgoff_t index; |
1604 | unsigned long offset; | |
a0ee5ec5 | 1605 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 1606 | int error = 0; |
cb66a7a1 | 1607 | ssize_t retval = 0; |
6e58e79d | 1608 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
1609 | |
1610 | /* | |
1611 | * Might this read be for a stacking filesystem? Then when reading | |
1612 | * holes of a sparse file, we actually need to allocate those pages, | |
1613 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
1614 | */ | |
777eda2c | 1615 | if (!iter_is_iovec(to)) |
75edd345 | 1616 | sgp = SGP_CACHE; |
1da177e4 | 1617 | |
09cbfeaf KS |
1618 | index = *ppos >> PAGE_SHIFT; |
1619 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
1620 | |
1621 | for (;;) { | |
1622 | struct page *page = NULL; | |
41ffe5d5 HD |
1623 | pgoff_t end_index; |
1624 | unsigned long nr, ret; | |
1da177e4 LT |
1625 | loff_t i_size = i_size_read(inode); |
1626 | ||
09cbfeaf | 1627 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
1628 | if (index > end_index) |
1629 | break; | |
1630 | if (index == end_index) { | |
09cbfeaf | 1631 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
1632 | if (nr <= offset) |
1633 | break; | |
1634 | } | |
1635 | ||
6e58e79d AV |
1636 | error = shmem_getpage(inode, index, &page, sgp, NULL); |
1637 | if (error) { | |
1638 | if (error == -EINVAL) | |
1639 | error = 0; | |
1da177e4 LT |
1640 | break; |
1641 | } | |
75edd345 HD |
1642 | if (page) { |
1643 | if (sgp == SGP_CACHE) | |
1644 | set_page_dirty(page); | |
d3602444 | 1645 | unlock_page(page); |
75edd345 | 1646 | } |
1da177e4 LT |
1647 | |
1648 | /* | |
1649 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 1650 | * are called without i_mutex protection against truncate |
1da177e4 | 1651 | */ |
09cbfeaf | 1652 | nr = PAGE_SIZE; |
1da177e4 | 1653 | i_size = i_size_read(inode); |
09cbfeaf | 1654 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 1655 | if (index == end_index) { |
09cbfeaf | 1656 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
1657 | if (nr <= offset) { |
1658 | if (page) | |
09cbfeaf | 1659 | put_page(page); |
1da177e4 LT |
1660 | break; |
1661 | } | |
1662 | } | |
1663 | nr -= offset; | |
1664 | ||
1665 | if (page) { | |
1666 | /* | |
1667 | * If users can be writing to this page using arbitrary | |
1668 | * virtual addresses, take care about potential aliasing | |
1669 | * before reading the page on the kernel side. | |
1670 | */ | |
1671 | if (mapping_writably_mapped(mapping)) | |
1672 | flush_dcache_page(page); | |
1673 | /* | |
1674 | * Mark the page accessed if we read the beginning. | |
1675 | */ | |
1676 | if (!offset) | |
1677 | mark_page_accessed(page); | |
b5810039 | 1678 | } else { |
1da177e4 | 1679 | page = ZERO_PAGE(0); |
09cbfeaf | 1680 | get_page(page); |
b5810039 | 1681 | } |
1da177e4 LT |
1682 | |
1683 | /* | |
1684 | * Ok, we have the page, and it's up-to-date, so | |
1685 | * now we can copy it to user space... | |
1da177e4 | 1686 | */ |
2ba5bbed | 1687 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 1688 | retval += ret; |
1da177e4 | 1689 | offset += ret; |
09cbfeaf KS |
1690 | index += offset >> PAGE_SHIFT; |
1691 | offset &= ~PAGE_MASK; | |
1da177e4 | 1692 | |
09cbfeaf | 1693 | put_page(page); |
2ba5bbed | 1694 | if (!iov_iter_count(to)) |
1da177e4 | 1695 | break; |
6e58e79d AV |
1696 | if (ret < nr) { |
1697 | error = -EFAULT; | |
1698 | break; | |
1699 | } | |
1da177e4 LT |
1700 | cond_resched(); |
1701 | } | |
1702 | ||
09cbfeaf | 1703 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
1704 | file_accessed(file); |
1705 | return retval ? retval : error; | |
1da177e4 LT |
1706 | } |
1707 | ||
708e3508 HD |
1708 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, |
1709 | struct pipe_inode_info *pipe, size_t len, | |
1710 | unsigned int flags) | |
1711 | { | |
1712 | struct address_space *mapping = in->f_mapping; | |
71f0e07a | 1713 | struct inode *inode = mapping->host; |
708e3508 HD |
1714 | unsigned int loff, nr_pages, req_pages; |
1715 | struct page *pages[PIPE_DEF_BUFFERS]; | |
1716 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
1717 | struct page *page; | |
1718 | pgoff_t index, end_index; | |
1719 | loff_t isize, left; | |
1720 | int error, page_nr; | |
1721 | struct splice_pipe_desc spd = { | |
1722 | .pages = pages, | |
1723 | .partial = partial, | |
047fe360 | 1724 | .nr_pages_max = PIPE_DEF_BUFFERS, |
708e3508 HD |
1725 | .flags = flags, |
1726 | .ops = &page_cache_pipe_buf_ops, | |
1727 | .spd_release = spd_release_page, | |
1728 | }; | |
1729 | ||
71f0e07a | 1730 | isize = i_size_read(inode); |
708e3508 HD |
1731 | if (unlikely(*ppos >= isize)) |
1732 | return 0; | |
1733 | ||
1734 | left = isize - *ppos; | |
1735 | if (unlikely(left < len)) | |
1736 | len = left; | |
1737 | ||
1738 | if (splice_grow_spd(pipe, &spd)) | |
1739 | return -ENOMEM; | |
1740 | ||
09cbfeaf KS |
1741 | index = *ppos >> PAGE_SHIFT; |
1742 | loff = *ppos & ~PAGE_MASK; | |
1743 | req_pages = (len + loff + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
a786c06d | 1744 | nr_pages = min(req_pages, spd.nr_pages_max); |
708e3508 | 1745 | |
708e3508 HD |
1746 | spd.nr_pages = find_get_pages_contig(mapping, index, |
1747 | nr_pages, spd.pages); | |
1748 | index += spd.nr_pages; | |
708e3508 | 1749 | error = 0; |
708e3508 | 1750 | |
71f0e07a | 1751 | while (spd.nr_pages < nr_pages) { |
71f0e07a HD |
1752 | error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL); |
1753 | if (error) | |
1754 | break; | |
1755 | unlock_page(page); | |
708e3508 HD |
1756 | spd.pages[spd.nr_pages++] = page; |
1757 | index++; | |
1758 | } | |
1759 | ||
09cbfeaf | 1760 | index = *ppos >> PAGE_SHIFT; |
708e3508 HD |
1761 | nr_pages = spd.nr_pages; |
1762 | spd.nr_pages = 0; | |
71f0e07a | 1763 | |
708e3508 HD |
1764 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { |
1765 | unsigned int this_len; | |
1766 | ||
1767 | if (!len) | |
1768 | break; | |
1769 | ||
09cbfeaf | 1770 | this_len = min_t(unsigned long, len, PAGE_SIZE - loff); |
708e3508 HD |
1771 | page = spd.pages[page_nr]; |
1772 | ||
71f0e07a | 1773 | if (!PageUptodate(page) || page->mapping != mapping) { |
71f0e07a HD |
1774 | error = shmem_getpage(inode, index, &page, |
1775 | SGP_CACHE, NULL); | |
1776 | if (error) | |
708e3508 | 1777 | break; |
71f0e07a | 1778 | unlock_page(page); |
09cbfeaf | 1779 | put_page(spd.pages[page_nr]); |
71f0e07a | 1780 | spd.pages[page_nr] = page; |
708e3508 | 1781 | } |
71f0e07a HD |
1782 | |
1783 | isize = i_size_read(inode); | |
09cbfeaf | 1784 | end_index = (isize - 1) >> PAGE_SHIFT; |
708e3508 HD |
1785 | if (unlikely(!isize || index > end_index)) |
1786 | break; | |
1787 | ||
708e3508 HD |
1788 | if (end_index == index) { |
1789 | unsigned int plen; | |
1790 | ||
09cbfeaf | 1791 | plen = ((isize - 1) & ~PAGE_MASK) + 1; |
708e3508 HD |
1792 | if (plen <= loff) |
1793 | break; | |
1794 | ||
708e3508 HD |
1795 | this_len = min(this_len, plen - loff); |
1796 | len = this_len; | |
1797 | } | |
1798 | ||
1799 | spd.partial[page_nr].offset = loff; | |
1800 | spd.partial[page_nr].len = this_len; | |
1801 | len -= this_len; | |
1802 | loff = 0; | |
1803 | spd.nr_pages++; | |
1804 | index++; | |
1805 | } | |
1806 | ||
708e3508 | 1807 | while (page_nr < nr_pages) |
09cbfeaf | 1808 | put_page(spd.pages[page_nr++]); |
708e3508 HD |
1809 | |
1810 | if (spd.nr_pages) | |
1811 | error = splice_to_pipe(pipe, &spd); | |
1812 | ||
047fe360 | 1813 | splice_shrink_spd(&spd); |
708e3508 HD |
1814 | |
1815 | if (error > 0) { | |
1816 | *ppos += error; | |
1817 | file_accessed(in); | |
1818 | } | |
1819 | return error; | |
1820 | } | |
1821 | ||
220f2ac9 HD |
1822 | /* |
1823 | * llseek SEEK_DATA or SEEK_HOLE through the radix_tree. | |
1824 | */ | |
1825 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 1826 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
1827 | { |
1828 | struct page *page; | |
1829 | struct pagevec pvec; | |
1830 | pgoff_t indices[PAGEVEC_SIZE]; | |
1831 | bool done = false; | |
1832 | int i; | |
1833 | ||
1834 | pagevec_init(&pvec, 0); | |
1835 | pvec.nr = 1; /* start small: we may be there already */ | |
1836 | while (!done) { | |
0cd6144a | 1837 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
1838 | pvec.nr, pvec.pages, indices); |
1839 | if (!pvec.nr) { | |
965c8e59 | 1840 | if (whence == SEEK_DATA) |
220f2ac9 HD |
1841 | index = end; |
1842 | break; | |
1843 | } | |
1844 | for (i = 0; i < pvec.nr; i++, index++) { | |
1845 | if (index < indices[i]) { | |
965c8e59 | 1846 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
1847 | done = true; |
1848 | break; | |
1849 | } | |
1850 | index = indices[i]; | |
1851 | } | |
1852 | page = pvec.pages[i]; | |
1853 | if (page && !radix_tree_exceptional_entry(page)) { | |
1854 | if (!PageUptodate(page)) | |
1855 | page = NULL; | |
1856 | } | |
1857 | if (index >= end || | |
965c8e59 AM |
1858 | (page && whence == SEEK_DATA) || |
1859 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
1860 | done = true; |
1861 | break; | |
1862 | } | |
1863 | } | |
0cd6144a | 1864 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
1865 | pagevec_release(&pvec); |
1866 | pvec.nr = PAGEVEC_SIZE; | |
1867 | cond_resched(); | |
1868 | } | |
1869 | return index; | |
1870 | } | |
1871 | ||
965c8e59 | 1872 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
1873 | { |
1874 | struct address_space *mapping = file->f_mapping; | |
1875 | struct inode *inode = mapping->host; | |
1876 | pgoff_t start, end; | |
1877 | loff_t new_offset; | |
1878 | ||
965c8e59 AM |
1879 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
1880 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 1881 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 1882 | inode_lock(inode); |
220f2ac9 HD |
1883 | /* We're holding i_mutex so we can access i_size directly */ |
1884 | ||
1885 | if (offset < 0) | |
1886 | offset = -EINVAL; | |
1887 | else if (offset >= inode->i_size) | |
1888 | offset = -ENXIO; | |
1889 | else { | |
09cbfeaf KS |
1890 | start = offset >> PAGE_SHIFT; |
1891 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 1892 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 1893 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
1894 | if (new_offset > offset) { |
1895 | if (new_offset < inode->i_size) | |
1896 | offset = new_offset; | |
965c8e59 | 1897 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
1898 | offset = -ENXIO; |
1899 | else | |
1900 | offset = inode->i_size; | |
1901 | } | |
1902 | } | |
1903 | ||
387aae6f HD |
1904 | if (offset >= 0) |
1905 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 1906 | inode_unlock(inode); |
220f2ac9 HD |
1907 | return offset; |
1908 | } | |
1909 | ||
05f65b5c DH |
1910 | /* |
1911 | * We need a tag: a new tag would expand every radix_tree_node by 8 bytes, | |
1912 | * so reuse a tag which we firmly believe is never set or cleared on shmem. | |
1913 | */ | |
1914 | #define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE | |
1915 | #define LAST_SCAN 4 /* about 150ms max */ | |
1916 | ||
1917 | static void shmem_tag_pins(struct address_space *mapping) | |
1918 | { | |
1919 | struct radix_tree_iter iter; | |
1920 | void **slot; | |
1921 | pgoff_t start; | |
1922 | struct page *page; | |
1923 | ||
1924 | lru_add_drain(); | |
1925 | start = 0; | |
1926 | rcu_read_lock(); | |
1927 | ||
05f65b5c DH |
1928 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
1929 | page = radix_tree_deref_slot(slot); | |
1930 | if (!page || radix_tree_exception(page)) { | |
2cf938aa MW |
1931 | if (radix_tree_deref_retry(page)) { |
1932 | slot = radix_tree_iter_retry(&iter); | |
1933 | continue; | |
1934 | } | |
05f65b5c DH |
1935 | } else if (page_count(page) - page_mapcount(page) > 1) { |
1936 | spin_lock_irq(&mapping->tree_lock); | |
1937 | radix_tree_tag_set(&mapping->page_tree, iter.index, | |
1938 | SHMEM_TAG_PINNED); | |
1939 | spin_unlock_irq(&mapping->tree_lock); | |
1940 | } | |
1941 | ||
1942 | if (need_resched()) { | |
1943 | cond_resched_rcu(); | |
7165092f | 1944 | slot = radix_tree_iter_next(&iter); |
05f65b5c DH |
1945 | } |
1946 | } | |
1947 | rcu_read_unlock(); | |
1948 | } | |
1949 | ||
1950 | /* | |
1951 | * Setting SEAL_WRITE requires us to verify there's no pending writer. However, | |
1952 | * via get_user_pages(), drivers might have some pending I/O without any active | |
1953 | * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages | |
1954 | * and see whether it has an elevated ref-count. If so, we tag them and wait for | |
1955 | * them to be dropped. | |
1956 | * The caller must guarantee that no new user will acquire writable references | |
1957 | * to those pages to avoid races. | |
1958 | */ | |
40e041a2 DH |
1959 | static int shmem_wait_for_pins(struct address_space *mapping) |
1960 | { | |
05f65b5c DH |
1961 | struct radix_tree_iter iter; |
1962 | void **slot; | |
1963 | pgoff_t start; | |
1964 | struct page *page; | |
1965 | int error, scan; | |
1966 | ||
1967 | shmem_tag_pins(mapping); | |
1968 | ||
1969 | error = 0; | |
1970 | for (scan = 0; scan <= LAST_SCAN; scan++) { | |
1971 | if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED)) | |
1972 | break; | |
1973 | ||
1974 | if (!scan) | |
1975 | lru_add_drain_all(); | |
1976 | else if (schedule_timeout_killable((HZ << scan) / 200)) | |
1977 | scan = LAST_SCAN; | |
1978 | ||
1979 | start = 0; | |
1980 | rcu_read_lock(); | |
05f65b5c DH |
1981 | radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, |
1982 | start, SHMEM_TAG_PINNED) { | |
1983 | ||
1984 | page = radix_tree_deref_slot(slot); | |
1985 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
1986 | if (radix_tree_deref_retry(page)) { |
1987 | slot = radix_tree_iter_retry(&iter); | |
1988 | continue; | |
1989 | } | |
05f65b5c DH |
1990 | |
1991 | page = NULL; | |
1992 | } | |
1993 | ||
1994 | if (page && | |
1995 | page_count(page) - page_mapcount(page) != 1) { | |
1996 | if (scan < LAST_SCAN) | |
1997 | goto continue_resched; | |
1998 | ||
1999 | /* | |
2000 | * On the last scan, we clean up all those tags | |
2001 | * we inserted; but make a note that we still | |
2002 | * found pages pinned. | |
2003 | */ | |
2004 | error = -EBUSY; | |
2005 | } | |
2006 | ||
2007 | spin_lock_irq(&mapping->tree_lock); | |
2008 | radix_tree_tag_clear(&mapping->page_tree, | |
2009 | iter.index, SHMEM_TAG_PINNED); | |
2010 | spin_unlock_irq(&mapping->tree_lock); | |
2011 | continue_resched: | |
2012 | if (need_resched()) { | |
2013 | cond_resched_rcu(); | |
7165092f | 2014 | slot = radix_tree_iter_next(&iter); |
05f65b5c DH |
2015 | } |
2016 | } | |
2017 | rcu_read_unlock(); | |
2018 | } | |
2019 | ||
2020 | return error; | |
40e041a2 DH |
2021 | } |
2022 | ||
2023 | #define F_ALL_SEALS (F_SEAL_SEAL | \ | |
2024 | F_SEAL_SHRINK | \ | |
2025 | F_SEAL_GROW | \ | |
2026 | F_SEAL_WRITE) | |
2027 | ||
2028 | int shmem_add_seals(struct file *file, unsigned int seals) | |
2029 | { | |
2030 | struct inode *inode = file_inode(file); | |
2031 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2032 | int error; | |
2033 | ||
2034 | /* | |
2035 | * SEALING | |
2036 | * Sealing allows multiple parties to share a shmem-file but restrict | |
2037 | * access to a specific subset of file operations. Seals can only be | |
2038 | * added, but never removed. This way, mutually untrusted parties can | |
2039 | * share common memory regions with a well-defined policy. A malicious | |
2040 | * peer can thus never perform unwanted operations on a shared object. | |
2041 | * | |
2042 | * Seals are only supported on special shmem-files and always affect | |
2043 | * the whole underlying inode. Once a seal is set, it may prevent some | |
2044 | * kinds of access to the file. Currently, the following seals are | |
2045 | * defined: | |
2046 | * SEAL_SEAL: Prevent further seals from being set on this file | |
2047 | * SEAL_SHRINK: Prevent the file from shrinking | |
2048 | * SEAL_GROW: Prevent the file from growing | |
2049 | * SEAL_WRITE: Prevent write access to the file | |
2050 | * | |
2051 | * As we don't require any trust relationship between two parties, we | |
2052 | * must prevent seals from being removed. Therefore, sealing a file | |
2053 | * only adds a given set of seals to the file, it never touches | |
2054 | * existing seals. Furthermore, the "setting seals"-operation can be | |
2055 | * sealed itself, which basically prevents any further seal from being | |
2056 | * added. | |
2057 | * | |
2058 | * Semantics of sealing are only defined on volatile files. Only | |
2059 | * anonymous shmem files support sealing. More importantly, seals are | |
2060 | * never written to disk. Therefore, there's no plan to support it on | |
2061 | * other file types. | |
2062 | */ | |
2063 | ||
2064 | if (file->f_op != &shmem_file_operations) | |
2065 | return -EINVAL; | |
2066 | if (!(file->f_mode & FMODE_WRITE)) | |
2067 | return -EPERM; | |
2068 | if (seals & ~(unsigned int)F_ALL_SEALS) | |
2069 | return -EINVAL; | |
2070 | ||
5955102c | 2071 | inode_lock(inode); |
40e041a2 DH |
2072 | |
2073 | if (info->seals & F_SEAL_SEAL) { | |
2074 | error = -EPERM; | |
2075 | goto unlock; | |
2076 | } | |
2077 | ||
2078 | if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) { | |
2079 | error = mapping_deny_writable(file->f_mapping); | |
2080 | if (error) | |
2081 | goto unlock; | |
2082 | ||
2083 | error = shmem_wait_for_pins(file->f_mapping); | |
2084 | if (error) { | |
2085 | mapping_allow_writable(file->f_mapping); | |
2086 | goto unlock; | |
2087 | } | |
2088 | } | |
2089 | ||
2090 | info->seals |= seals; | |
2091 | error = 0; | |
2092 | ||
2093 | unlock: | |
5955102c | 2094 | inode_unlock(inode); |
40e041a2 DH |
2095 | return error; |
2096 | } | |
2097 | EXPORT_SYMBOL_GPL(shmem_add_seals); | |
2098 | ||
2099 | int shmem_get_seals(struct file *file) | |
2100 | { | |
2101 | if (file->f_op != &shmem_file_operations) | |
2102 | return -EINVAL; | |
2103 | ||
2104 | return SHMEM_I(file_inode(file))->seals; | |
2105 | } | |
2106 | EXPORT_SYMBOL_GPL(shmem_get_seals); | |
2107 | ||
2108 | long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg) | |
2109 | { | |
2110 | long error; | |
2111 | ||
2112 | switch (cmd) { | |
2113 | case F_ADD_SEALS: | |
2114 | /* disallow upper 32bit */ | |
2115 | if (arg > UINT_MAX) | |
2116 | return -EINVAL; | |
2117 | ||
2118 | error = shmem_add_seals(file, arg); | |
2119 | break; | |
2120 | case F_GET_SEALS: | |
2121 | error = shmem_get_seals(file); | |
2122 | break; | |
2123 | default: | |
2124 | error = -EINVAL; | |
2125 | break; | |
2126 | } | |
2127 | ||
2128 | return error; | |
2129 | } | |
2130 | ||
83e4fa9c HD |
2131 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2132 | loff_t len) | |
2133 | { | |
496ad9aa | 2134 | struct inode *inode = file_inode(file); |
e2d12e22 | 2135 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2136 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2137 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2138 | pgoff_t start, index, end; |
2139 | int error; | |
83e4fa9c | 2140 | |
13ace4d0 HD |
2141 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2142 | return -EOPNOTSUPP; | |
2143 | ||
5955102c | 2144 | inode_lock(inode); |
83e4fa9c HD |
2145 | |
2146 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2147 | struct address_space *mapping = file->f_mapping; | |
2148 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2149 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2150 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2151 | |
40e041a2 DH |
2152 | /* protected by i_mutex */ |
2153 | if (info->seals & F_SEAL_WRITE) { | |
2154 | error = -EPERM; | |
2155 | goto out; | |
2156 | } | |
2157 | ||
8e205f77 | 2158 | shmem_falloc.waitq = &shmem_falloc_waitq; |
f00cdc6d HD |
2159 | shmem_falloc.start = unmap_start >> PAGE_SHIFT; |
2160 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; | |
2161 | spin_lock(&inode->i_lock); | |
2162 | inode->i_private = &shmem_falloc; | |
2163 | spin_unlock(&inode->i_lock); | |
2164 | ||
83e4fa9c HD |
2165 | if ((u64)unmap_end > (u64)unmap_start) |
2166 | unmap_mapping_range(mapping, unmap_start, | |
2167 | 1 + unmap_end - unmap_start, 0); | |
2168 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2169 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2170 | |
2171 | spin_lock(&inode->i_lock); | |
2172 | inode->i_private = NULL; | |
2173 | wake_up_all(&shmem_falloc_waitq); | |
2174 | spin_unlock(&inode->i_lock); | |
83e4fa9c | 2175 | error = 0; |
8e205f77 | 2176 | goto out; |
e2d12e22 HD |
2177 | } |
2178 | ||
2179 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2180 | error = inode_newsize_ok(inode, offset + len); | |
2181 | if (error) | |
2182 | goto out; | |
2183 | ||
40e041a2 DH |
2184 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2185 | error = -EPERM; | |
2186 | goto out; | |
2187 | } | |
2188 | ||
09cbfeaf KS |
2189 | start = offset >> PAGE_SHIFT; |
2190 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2191 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2192 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2193 | error = -ENOSPC; | |
2194 | goto out; | |
83e4fa9c HD |
2195 | } |
2196 | ||
8e205f77 | 2197 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2198 | shmem_falloc.start = start; |
2199 | shmem_falloc.next = start; | |
2200 | shmem_falloc.nr_falloced = 0; | |
2201 | shmem_falloc.nr_unswapped = 0; | |
2202 | spin_lock(&inode->i_lock); | |
2203 | inode->i_private = &shmem_falloc; | |
2204 | spin_unlock(&inode->i_lock); | |
2205 | ||
e2d12e22 HD |
2206 | for (index = start; index < end; index++) { |
2207 | struct page *page; | |
2208 | ||
2209 | /* | |
2210 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2211 | * been interrupted because we are using up too much memory. | |
2212 | */ | |
2213 | if (signal_pending(current)) | |
2214 | error = -EINTR; | |
1aac1400 HD |
2215 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2216 | error = -ENOMEM; | |
e2d12e22 | 2217 | else |
1635f6a7 | 2218 | error = shmem_getpage(inode, index, &page, SGP_FALLOC, |
e2d12e22 HD |
2219 | NULL); |
2220 | if (error) { | |
1635f6a7 HD |
2221 | /* Remove the !PageUptodate pages we added */ |
2222 | shmem_undo_range(inode, | |
09cbfeaf KS |
2223 | (loff_t)start << PAGE_SHIFT, |
2224 | (loff_t)index << PAGE_SHIFT, true); | |
1aac1400 | 2225 | goto undone; |
e2d12e22 HD |
2226 | } |
2227 | ||
1aac1400 HD |
2228 | /* |
2229 | * Inform shmem_writepage() how far we have reached. | |
2230 | * No need for lock or barrier: we have the page lock. | |
2231 | */ | |
2232 | shmem_falloc.next++; | |
2233 | if (!PageUptodate(page)) | |
2234 | shmem_falloc.nr_falloced++; | |
2235 | ||
e2d12e22 | 2236 | /* |
1635f6a7 HD |
2237 | * If !PageUptodate, leave it that way so that freeable pages |
2238 | * can be recognized if we need to rollback on error later. | |
2239 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2240 | * than free the pages we are allocating (and SGP_CACHE pages |
2241 | * might still be clean: we now need to mark those dirty too). | |
2242 | */ | |
2243 | set_page_dirty(page); | |
2244 | unlock_page(page); | |
09cbfeaf | 2245 | put_page(page); |
e2d12e22 HD |
2246 | cond_resched(); |
2247 | } | |
2248 | ||
2249 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2250 | i_size_write(inode, offset + len); | |
e2d12e22 | 2251 | inode->i_ctime = CURRENT_TIME; |
1aac1400 HD |
2252 | undone: |
2253 | spin_lock(&inode->i_lock); | |
2254 | inode->i_private = NULL; | |
2255 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2256 | out: |
5955102c | 2257 | inode_unlock(inode); |
83e4fa9c HD |
2258 | return error; |
2259 | } | |
2260 | ||
726c3342 | 2261 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2262 | { |
726c3342 | 2263 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2264 | |
2265 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2266 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2267 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2268 | if (sbinfo->max_blocks) { |
1da177e4 | 2269 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2270 | buf->f_bavail = |
2271 | buf->f_bfree = sbinfo->max_blocks - | |
2272 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2273 | } |
2274 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2275 | buf->f_files = sbinfo->max_inodes; |
2276 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2277 | } |
2278 | /* else leave those fields 0 like simple_statfs */ | |
2279 | return 0; | |
2280 | } | |
2281 | ||
2282 | /* | |
2283 | * File creation. Allocate an inode, and we're done.. | |
2284 | */ | |
2285 | static int | |
1a67aafb | 2286 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2287 | { |
0b0a0806 | 2288 | struct inode *inode; |
1da177e4 LT |
2289 | int error = -ENOSPC; |
2290 | ||
454abafe | 2291 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2292 | if (inode) { |
feda821e CH |
2293 | error = simple_acl_create(dir, inode); |
2294 | if (error) | |
2295 | goto out_iput; | |
2a7dba39 | 2296 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2297 | &dentry->d_name, |
6d9d88d0 | 2298 | shmem_initxattrs, NULL); |
feda821e CH |
2299 | if (error && error != -EOPNOTSUPP) |
2300 | goto out_iput; | |
37ec43cd | 2301 | |
718deb6b | 2302 | error = 0; |
1da177e4 LT |
2303 | dir->i_size += BOGO_DIRENT_SIZE; |
2304 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2305 | d_instantiate(dentry, inode); | |
2306 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2307 | } |
2308 | return error; | |
feda821e CH |
2309 | out_iput: |
2310 | iput(inode); | |
2311 | return error; | |
1da177e4 LT |
2312 | } |
2313 | ||
60545d0d AV |
2314 | static int |
2315 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2316 | { | |
2317 | struct inode *inode; | |
2318 | int error = -ENOSPC; | |
2319 | ||
2320 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
2321 | if (inode) { | |
2322 | error = security_inode_init_security(inode, dir, | |
2323 | NULL, | |
2324 | shmem_initxattrs, NULL); | |
feda821e CH |
2325 | if (error && error != -EOPNOTSUPP) |
2326 | goto out_iput; | |
2327 | error = simple_acl_create(dir, inode); | |
2328 | if (error) | |
2329 | goto out_iput; | |
60545d0d AV |
2330 | d_tmpfile(dentry, inode); |
2331 | } | |
2332 | return error; | |
feda821e CH |
2333 | out_iput: |
2334 | iput(inode); | |
2335 | return error; | |
60545d0d AV |
2336 | } |
2337 | ||
18bb1db3 | 2338 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2339 | { |
2340 | int error; | |
2341 | ||
2342 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
2343 | return error; | |
d8c76e6f | 2344 | inc_nlink(dir); |
1da177e4 LT |
2345 | return 0; |
2346 | } | |
2347 | ||
4acdaf27 | 2348 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2349 | bool excl) |
1da177e4 LT |
2350 | { |
2351 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
2352 | } | |
2353 | ||
2354 | /* | |
2355 | * Link a file.. | |
2356 | */ | |
2357 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2358 | { | |
75c3cfa8 | 2359 | struct inode *inode = d_inode(old_dentry); |
5b04c689 | 2360 | int ret; |
1da177e4 LT |
2361 | |
2362 | /* | |
2363 | * No ordinary (disk based) filesystem counts links as inodes; | |
2364 | * but each new link needs a new dentry, pinning lowmem, and | |
2365 | * tmpfs dentries cannot be pruned until they are unlinked. | |
2366 | */ | |
5b04c689 PE |
2367 | ret = shmem_reserve_inode(inode->i_sb); |
2368 | if (ret) | |
2369 | goto out; | |
1da177e4 LT |
2370 | |
2371 | dir->i_size += BOGO_DIRENT_SIZE; | |
2372 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
d8c76e6f | 2373 | inc_nlink(inode); |
7de9c6ee | 2374 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
2375 | dget(dentry); /* Extra pinning count for the created dentry */ |
2376 | d_instantiate(dentry, inode); | |
5b04c689 PE |
2377 | out: |
2378 | return ret; | |
1da177e4 LT |
2379 | } |
2380 | ||
2381 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
2382 | { | |
75c3cfa8 | 2383 | struct inode *inode = d_inode(dentry); |
1da177e4 | 2384 | |
5b04c689 PE |
2385 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2386 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
2387 | |
2388 | dir->i_size -= BOGO_DIRENT_SIZE; | |
2389 | inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
9a53c3a7 | 2390 | drop_nlink(inode); |
1da177e4 LT |
2391 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
2392 | return 0; | |
2393 | } | |
2394 | ||
2395 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
2396 | { | |
2397 | if (!simple_empty(dentry)) | |
2398 | return -ENOTEMPTY; | |
2399 | ||
75c3cfa8 | 2400 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 2401 | drop_nlink(dir); |
1da177e4 LT |
2402 | return shmem_unlink(dir, dentry); |
2403 | } | |
2404 | ||
37456771 MS |
2405 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
2406 | { | |
e36cb0b8 DH |
2407 | bool old_is_dir = d_is_dir(old_dentry); |
2408 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
2409 | |
2410 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
2411 | if (old_is_dir) { | |
2412 | drop_nlink(old_dir); | |
2413 | inc_nlink(new_dir); | |
2414 | } else { | |
2415 | drop_nlink(new_dir); | |
2416 | inc_nlink(old_dir); | |
2417 | } | |
2418 | } | |
2419 | old_dir->i_ctime = old_dir->i_mtime = | |
2420 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 DH |
2421 | d_inode(old_dentry)->i_ctime = |
2422 | d_inode(new_dentry)->i_ctime = CURRENT_TIME; | |
37456771 MS |
2423 | |
2424 | return 0; | |
2425 | } | |
2426 | ||
46fdb794 MS |
2427 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
2428 | { | |
2429 | struct dentry *whiteout; | |
2430 | int error; | |
2431 | ||
2432 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
2433 | if (!whiteout) | |
2434 | return -ENOMEM; | |
2435 | ||
2436 | error = shmem_mknod(old_dir, whiteout, | |
2437 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
2438 | dput(whiteout); | |
2439 | if (error) | |
2440 | return error; | |
2441 | ||
2442 | /* | |
2443 | * Cheat and hash the whiteout while the old dentry is still in | |
2444 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
2445 | * | |
2446 | * d_lookup() will consistently find one of them at this point, | |
2447 | * not sure which one, but that isn't even important. | |
2448 | */ | |
2449 | d_rehash(whiteout); | |
2450 | return 0; | |
2451 | } | |
2452 | ||
1da177e4 LT |
2453 | /* |
2454 | * The VFS layer already does all the dentry stuff for rename, | |
2455 | * we just have to decrement the usage count for the target if | |
2456 | * it exists so that the VFS layer correctly free's it when it | |
2457 | * gets overwritten. | |
2458 | */ | |
3b69ff51 | 2459 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 2460 | { |
75c3cfa8 | 2461 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
2462 | int they_are_dirs = S_ISDIR(inode->i_mode); |
2463 | ||
46fdb794 | 2464 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
2465 | return -EINVAL; |
2466 | ||
37456771 MS |
2467 | if (flags & RENAME_EXCHANGE) |
2468 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
2469 | ||
1da177e4 LT |
2470 | if (!simple_empty(new_dentry)) |
2471 | return -ENOTEMPTY; | |
2472 | ||
46fdb794 MS |
2473 | if (flags & RENAME_WHITEOUT) { |
2474 | int error; | |
2475 | ||
2476 | error = shmem_whiteout(old_dir, old_dentry); | |
2477 | if (error) | |
2478 | return error; | |
2479 | } | |
2480 | ||
75c3cfa8 | 2481 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 2482 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 2483 | if (they_are_dirs) { |
75c3cfa8 | 2484 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 2485 | drop_nlink(old_dir); |
b928095b | 2486 | } |
1da177e4 | 2487 | } else if (they_are_dirs) { |
9a53c3a7 | 2488 | drop_nlink(old_dir); |
d8c76e6f | 2489 | inc_nlink(new_dir); |
1da177e4 LT |
2490 | } |
2491 | ||
2492 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
2493 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
2494 | old_dir->i_ctime = old_dir->i_mtime = | |
2495 | new_dir->i_ctime = new_dir->i_mtime = | |
2496 | inode->i_ctime = CURRENT_TIME; | |
2497 | return 0; | |
2498 | } | |
2499 | ||
2500 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
2501 | { | |
2502 | int error; | |
2503 | int len; | |
2504 | struct inode *inode; | |
9276aad6 | 2505 | struct page *page; |
1da177e4 LT |
2506 | struct shmem_inode_info *info; |
2507 | ||
2508 | len = strlen(symname) + 1; | |
09cbfeaf | 2509 | if (len > PAGE_SIZE) |
1da177e4 LT |
2510 | return -ENAMETOOLONG; |
2511 | ||
454abafe | 2512 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); |
1da177e4 LT |
2513 | if (!inode) |
2514 | return -ENOSPC; | |
2515 | ||
9d8f13ba | 2516 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 2517 | shmem_initxattrs, NULL); |
570bc1c2 SS |
2518 | if (error) { |
2519 | if (error != -EOPNOTSUPP) { | |
2520 | iput(inode); | |
2521 | return error; | |
2522 | } | |
2523 | error = 0; | |
2524 | } | |
2525 | ||
1da177e4 LT |
2526 | info = SHMEM_I(inode); |
2527 | inode->i_size = len-1; | |
69f07ec9 | 2528 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
2529 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
2530 | if (!inode->i_link) { | |
69f07ec9 HD |
2531 | iput(inode); |
2532 | return -ENOMEM; | |
2533 | } | |
2534 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 2535 | } else { |
e8ecde25 | 2536 | inode_nohighmem(inode); |
1da177e4 LT |
2537 | error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); |
2538 | if (error) { | |
2539 | iput(inode); | |
2540 | return error; | |
2541 | } | |
14fcc23f | 2542 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 2543 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 2544 | memcpy(page_address(page), symname, len); |
ec9516fb | 2545 | SetPageUptodate(page); |
1da177e4 | 2546 | set_page_dirty(page); |
6746aff7 | 2547 | unlock_page(page); |
09cbfeaf | 2548 | put_page(page); |
1da177e4 | 2549 | } |
1da177e4 LT |
2550 | dir->i_size += BOGO_DIRENT_SIZE; |
2551 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
2552 | d_instantiate(dentry, inode); | |
2553 | dget(dentry); | |
2554 | return 0; | |
2555 | } | |
2556 | ||
fceef393 | 2557 | static void shmem_put_link(void *arg) |
1da177e4 | 2558 | { |
fceef393 AV |
2559 | mark_page_accessed(arg); |
2560 | put_page(arg); | |
1da177e4 LT |
2561 | } |
2562 | ||
6b255391 | 2563 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
2564 | struct inode *inode, |
2565 | struct delayed_call *done) | |
1da177e4 | 2566 | { |
1da177e4 | 2567 | struct page *page = NULL; |
6b255391 | 2568 | int error; |
6a6c9904 AV |
2569 | if (!dentry) { |
2570 | page = find_get_page(inode->i_mapping, 0); | |
2571 | if (!page) | |
2572 | return ERR_PTR(-ECHILD); | |
2573 | if (!PageUptodate(page)) { | |
2574 | put_page(page); | |
2575 | return ERR_PTR(-ECHILD); | |
2576 | } | |
2577 | } else { | |
2578 | error = shmem_getpage(inode, 0, &page, SGP_READ, NULL); | |
2579 | if (error) | |
2580 | return ERR_PTR(error); | |
2581 | unlock_page(page); | |
2582 | } | |
fceef393 | 2583 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 2584 | return page_address(page); |
1da177e4 LT |
2585 | } |
2586 | ||
b09e0fa4 | 2587 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 2588 | /* |
b09e0fa4 EP |
2589 | * Superblocks without xattr inode operations may get some security.* xattr |
2590 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
2591 | * like ACLs, we also need to implement the security.* handlers at |
2592 | * filesystem level, though. | |
2593 | */ | |
2594 | ||
6d9d88d0 JS |
2595 | /* |
2596 | * Callback for security_inode_init_security() for acquiring xattrs. | |
2597 | */ | |
2598 | static int shmem_initxattrs(struct inode *inode, | |
2599 | const struct xattr *xattr_array, | |
2600 | void *fs_info) | |
2601 | { | |
2602 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2603 | const struct xattr *xattr; | |
38f38657 | 2604 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
2605 | size_t len; |
2606 | ||
2607 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 2608 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
2609 | if (!new_xattr) |
2610 | return -ENOMEM; | |
2611 | ||
2612 | len = strlen(xattr->name) + 1; | |
2613 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
2614 | GFP_KERNEL); | |
2615 | if (!new_xattr->name) { | |
2616 | kfree(new_xattr); | |
2617 | return -ENOMEM; | |
2618 | } | |
2619 | ||
2620 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
2621 | XATTR_SECURITY_PREFIX_LEN); | |
2622 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
2623 | xattr->name, len); | |
2624 | ||
38f38657 | 2625 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
2626 | } |
2627 | ||
2628 | return 0; | |
2629 | } | |
2630 | ||
aa7c5241 | 2631 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
2632 | struct dentry *unused, struct inode *inode, |
2633 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 2634 | { |
b296821a | 2635 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 2636 | |
aa7c5241 | 2637 | name = xattr_full_name(handler, name); |
38f38657 | 2638 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
2639 | } |
2640 | ||
aa7c5241 AG |
2641 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
2642 | struct dentry *dentry, const char *name, | |
2643 | const void *value, size_t size, int flags) | |
b09e0fa4 | 2644 | { |
75c3cfa8 | 2645 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
b09e0fa4 | 2646 | |
aa7c5241 | 2647 | name = xattr_full_name(handler, name); |
38f38657 | 2648 | return simple_xattr_set(&info->xattrs, name, value, size, flags); |
b09e0fa4 EP |
2649 | } |
2650 | ||
aa7c5241 AG |
2651 | static const struct xattr_handler shmem_security_xattr_handler = { |
2652 | .prefix = XATTR_SECURITY_PREFIX, | |
2653 | .get = shmem_xattr_handler_get, | |
2654 | .set = shmem_xattr_handler_set, | |
2655 | }; | |
b09e0fa4 | 2656 | |
aa7c5241 AG |
2657 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
2658 | .prefix = XATTR_TRUSTED_PREFIX, | |
2659 | .get = shmem_xattr_handler_get, | |
2660 | .set = shmem_xattr_handler_set, | |
2661 | }; | |
b09e0fa4 | 2662 | |
aa7c5241 AG |
2663 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
2664 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
2665 | &posix_acl_access_xattr_handler, | |
2666 | &posix_acl_default_xattr_handler, | |
2667 | #endif | |
2668 | &shmem_security_xattr_handler, | |
2669 | &shmem_trusted_xattr_handler, | |
2670 | NULL | |
2671 | }; | |
b09e0fa4 EP |
2672 | |
2673 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
2674 | { | |
75c3cfa8 | 2675 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 2676 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
2677 | } |
2678 | #endif /* CONFIG_TMPFS_XATTR */ | |
2679 | ||
69f07ec9 | 2680 | static const struct inode_operations shmem_short_symlink_operations = { |
b09e0fa4 | 2681 | .readlink = generic_readlink, |
6b255391 | 2682 | .get_link = simple_get_link, |
b09e0fa4 | 2683 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
2684 | .setxattr = generic_setxattr, |
2685 | .getxattr = generic_getxattr, | |
b09e0fa4 | 2686 | .listxattr = shmem_listxattr, |
aa7c5241 | 2687 | .removexattr = generic_removexattr, |
b09e0fa4 EP |
2688 | #endif |
2689 | }; | |
2690 | ||
2691 | static const struct inode_operations shmem_symlink_inode_operations = { | |
2692 | .readlink = generic_readlink, | |
6b255391 | 2693 | .get_link = shmem_get_link, |
b09e0fa4 | 2694 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
2695 | .setxattr = generic_setxattr, |
2696 | .getxattr = generic_getxattr, | |
b09e0fa4 | 2697 | .listxattr = shmem_listxattr, |
aa7c5241 | 2698 | .removexattr = generic_removexattr, |
39f0247d | 2699 | #endif |
b09e0fa4 | 2700 | }; |
39f0247d | 2701 | |
91828a40 DG |
2702 | static struct dentry *shmem_get_parent(struct dentry *child) |
2703 | { | |
2704 | return ERR_PTR(-ESTALE); | |
2705 | } | |
2706 | ||
2707 | static int shmem_match(struct inode *ino, void *vfh) | |
2708 | { | |
2709 | __u32 *fh = vfh; | |
2710 | __u64 inum = fh[2]; | |
2711 | inum = (inum << 32) | fh[1]; | |
2712 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
2713 | } | |
2714 | ||
480b116c CH |
2715 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
2716 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 2717 | { |
91828a40 | 2718 | struct inode *inode; |
480b116c | 2719 | struct dentry *dentry = NULL; |
35c2a7f4 | 2720 | u64 inum; |
480b116c CH |
2721 | |
2722 | if (fh_len < 3) | |
2723 | return NULL; | |
91828a40 | 2724 | |
35c2a7f4 HD |
2725 | inum = fid->raw[2]; |
2726 | inum = (inum << 32) | fid->raw[1]; | |
2727 | ||
480b116c CH |
2728 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
2729 | shmem_match, fid->raw); | |
91828a40 | 2730 | if (inode) { |
480b116c | 2731 | dentry = d_find_alias(inode); |
91828a40 DG |
2732 | iput(inode); |
2733 | } | |
2734 | ||
480b116c | 2735 | return dentry; |
91828a40 DG |
2736 | } |
2737 | ||
b0b0382b AV |
2738 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
2739 | struct inode *parent) | |
91828a40 | 2740 | { |
5fe0c237 AK |
2741 | if (*len < 3) { |
2742 | *len = 3; | |
94e07a75 | 2743 | return FILEID_INVALID; |
5fe0c237 | 2744 | } |
91828a40 | 2745 | |
1d3382cb | 2746 | if (inode_unhashed(inode)) { |
91828a40 DG |
2747 | /* Unfortunately insert_inode_hash is not idempotent, |
2748 | * so as we hash inodes here rather than at creation | |
2749 | * time, we need a lock to ensure we only try | |
2750 | * to do it once | |
2751 | */ | |
2752 | static DEFINE_SPINLOCK(lock); | |
2753 | spin_lock(&lock); | |
1d3382cb | 2754 | if (inode_unhashed(inode)) |
91828a40 DG |
2755 | __insert_inode_hash(inode, |
2756 | inode->i_ino + inode->i_generation); | |
2757 | spin_unlock(&lock); | |
2758 | } | |
2759 | ||
2760 | fh[0] = inode->i_generation; | |
2761 | fh[1] = inode->i_ino; | |
2762 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
2763 | ||
2764 | *len = 3; | |
2765 | return 1; | |
2766 | } | |
2767 | ||
39655164 | 2768 | static const struct export_operations shmem_export_ops = { |
91828a40 | 2769 | .get_parent = shmem_get_parent, |
91828a40 | 2770 | .encode_fh = shmem_encode_fh, |
480b116c | 2771 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
2772 | }; |
2773 | ||
680d794b | 2774 | static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, |
2775 | bool remount) | |
1da177e4 LT |
2776 | { |
2777 | char *this_char, *value, *rest; | |
49cd0a5c | 2778 | struct mempolicy *mpol = NULL; |
8751e039 EB |
2779 | uid_t uid; |
2780 | gid_t gid; | |
1da177e4 | 2781 | |
b00dc3ad HD |
2782 | while (options != NULL) { |
2783 | this_char = options; | |
2784 | for (;;) { | |
2785 | /* | |
2786 | * NUL-terminate this option: unfortunately, | |
2787 | * mount options form a comma-separated list, | |
2788 | * but mpol's nodelist may also contain commas. | |
2789 | */ | |
2790 | options = strchr(options, ','); | |
2791 | if (options == NULL) | |
2792 | break; | |
2793 | options++; | |
2794 | if (!isdigit(*options)) { | |
2795 | options[-1] = '\0'; | |
2796 | break; | |
2797 | } | |
2798 | } | |
1da177e4 LT |
2799 | if (!*this_char) |
2800 | continue; | |
2801 | if ((value = strchr(this_char,'=')) != NULL) { | |
2802 | *value++ = 0; | |
2803 | } else { | |
1170532b JP |
2804 | pr_err("tmpfs: No value for mount option '%s'\n", |
2805 | this_char); | |
49cd0a5c | 2806 | goto error; |
1da177e4 LT |
2807 | } |
2808 | ||
2809 | if (!strcmp(this_char,"size")) { | |
2810 | unsigned long long size; | |
2811 | size = memparse(value,&rest); | |
2812 | if (*rest == '%') { | |
2813 | size <<= PAGE_SHIFT; | |
2814 | size *= totalram_pages; | |
2815 | do_div(size, 100); | |
2816 | rest++; | |
2817 | } | |
2818 | if (*rest) | |
2819 | goto bad_val; | |
680d794b | 2820 | sbinfo->max_blocks = |
09cbfeaf | 2821 | DIV_ROUND_UP(size, PAGE_SIZE); |
1da177e4 | 2822 | } else if (!strcmp(this_char,"nr_blocks")) { |
680d794b | 2823 | sbinfo->max_blocks = memparse(value, &rest); |
1da177e4 LT |
2824 | if (*rest) |
2825 | goto bad_val; | |
2826 | } else if (!strcmp(this_char,"nr_inodes")) { | |
680d794b | 2827 | sbinfo->max_inodes = memparse(value, &rest); |
1da177e4 LT |
2828 | if (*rest) |
2829 | goto bad_val; | |
2830 | } else if (!strcmp(this_char,"mode")) { | |
680d794b | 2831 | if (remount) |
1da177e4 | 2832 | continue; |
680d794b | 2833 | sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; |
1da177e4 LT |
2834 | if (*rest) |
2835 | goto bad_val; | |
2836 | } else if (!strcmp(this_char,"uid")) { | |
680d794b | 2837 | if (remount) |
1da177e4 | 2838 | continue; |
8751e039 | 2839 | uid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2840 | if (*rest) |
2841 | goto bad_val; | |
8751e039 EB |
2842 | sbinfo->uid = make_kuid(current_user_ns(), uid); |
2843 | if (!uid_valid(sbinfo->uid)) | |
2844 | goto bad_val; | |
1da177e4 | 2845 | } else if (!strcmp(this_char,"gid")) { |
680d794b | 2846 | if (remount) |
1da177e4 | 2847 | continue; |
8751e039 | 2848 | gid = simple_strtoul(value, &rest, 0); |
1da177e4 LT |
2849 | if (*rest) |
2850 | goto bad_val; | |
8751e039 EB |
2851 | sbinfo->gid = make_kgid(current_user_ns(), gid); |
2852 | if (!gid_valid(sbinfo->gid)) | |
2853 | goto bad_val; | |
7339ff83 | 2854 | } else if (!strcmp(this_char,"mpol")) { |
49cd0a5c GT |
2855 | mpol_put(mpol); |
2856 | mpol = NULL; | |
2857 | if (mpol_parse_str(value, &mpol)) | |
7339ff83 | 2858 | goto bad_val; |
1da177e4 | 2859 | } else { |
1170532b | 2860 | pr_err("tmpfs: Bad mount option %s\n", this_char); |
49cd0a5c | 2861 | goto error; |
1da177e4 LT |
2862 | } |
2863 | } | |
49cd0a5c | 2864 | sbinfo->mpol = mpol; |
1da177e4 LT |
2865 | return 0; |
2866 | ||
2867 | bad_val: | |
1170532b | 2868 | pr_err("tmpfs: Bad value '%s' for mount option '%s'\n", |
1da177e4 | 2869 | value, this_char); |
49cd0a5c GT |
2870 | error: |
2871 | mpol_put(mpol); | |
1da177e4 LT |
2872 | return 1; |
2873 | ||
2874 | } | |
2875 | ||
2876 | static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) | |
2877 | { | |
2878 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
680d794b | 2879 | struct shmem_sb_info config = *sbinfo; |
0edd73b3 HD |
2880 | unsigned long inodes; |
2881 | int error = -EINVAL; | |
2882 | ||
5f00110f | 2883 | config.mpol = NULL; |
680d794b | 2884 | if (shmem_parse_options(data, &config, true)) |
0edd73b3 | 2885 | return error; |
1da177e4 | 2886 | |
0edd73b3 | 2887 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 2888 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
7e496299 | 2889 | if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) |
0edd73b3 | 2890 | goto out; |
680d794b | 2891 | if (config.max_inodes < inodes) |
0edd73b3 HD |
2892 | goto out; |
2893 | /* | |
54af6042 | 2894 | * Those tests disallow limited->unlimited while any are in use; |
0edd73b3 HD |
2895 | * but we must separately disallow unlimited->limited, because |
2896 | * in that case we have no record of how much is already in use. | |
2897 | */ | |
680d794b | 2898 | if (config.max_blocks && !sbinfo->max_blocks) |
0edd73b3 | 2899 | goto out; |
680d794b | 2900 | if (config.max_inodes && !sbinfo->max_inodes) |
0edd73b3 HD |
2901 | goto out; |
2902 | ||
2903 | error = 0; | |
680d794b | 2904 | sbinfo->max_blocks = config.max_blocks; |
680d794b | 2905 | sbinfo->max_inodes = config.max_inodes; |
2906 | sbinfo->free_inodes = config.max_inodes - inodes; | |
71fe804b | 2907 | |
5f00110f GT |
2908 | /* |
2909 | * Preserve previous mempolicy unless mpol remount option was specified. | |
2910 | */ | |
2911 | if (config.mpol) { | |
2912 | mpol_put(sbinfo->mpol); | |
2913 | sbinfo->mpol = config.mpol; /* transfers initial ref */ | |
2914 | } | |
0edd73b3 HD |
2915 | out: |
2916 | spin_unlock(&sbinfo->stat_lock); | |
2917 | return error; | |
1da177e4 | 2918 | } |
680d794b | 2919 | |
34c80b1d | 2920 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 2921 | { |
34c80b1d | 2922 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b | 2923 | |
2924 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
2925 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 2926 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 2927 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
2928 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
2929 | if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) | |
09208d15 | 2930 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
2931 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
2932 | seq_printf(seq, ",uid=%u", | |
2933 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
2934 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
2935 | seq_printf(seq, ",gid=%u", | |
2936 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
71fe804b | 2937 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 2938 | return 0; |
2939 | } | |
9183df25 DH |
2940 | |
2941 | #define MFD_NAME_PREFIX "memfd:" | |
2942 | #define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1) | |
2943 | #define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN) | |
2944 | ||
2945 | #define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING) | |
2946 | ||
2947 | SYSCALL_DEFINE2(memfd_create, | |
2948 | const char __user *, uname, | |
2949 | unsigned int, flags) | |
2950 | { | |
2951 | struct shmem_inode_info *info; | |
2952 | struct file *file; | |
2953 | int fd, error; | |
2954 | char *name; | |
2955 | long len; | |
2956 | ||
2957 | if (flags & ~(unsigned int)MFD_ALL_FLAGS) | |
2958 | return -EINVAL; | |
2959 | ||
2960 | /* length includes terminating zero */ | |
2961 | len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1); | |
2962 | if (len <= 0) | |
2963 | return -EFAULT; | |
2964 | if (len > MFD_NAME_MAX_LEN + 1) | |
2965 | return -EINVAL; | |
2966 | ||
2967 | name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY); | |
2968 | if (!name) | |
2969 | return -ENOMEM; | |
2970 | ||
2971 | strcpy(name, MFD_NAME_PREFIX); | |
2972 | if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) { | |
2973 | error = -EFAULT; | |
2974 | goto err_name; | |
2975 | } | |
2976 | ||
2977 | /* terminating-zero may have changed after strnlen_user() returned */ | |
2978 | if (name[len + MFD_NAME_PREFIX_LEN - 1]) { | |
2979 | error = -EFAULT; | |
2980 | goto err_name; | |
2981 | } | |
2982 | ||
2983 | fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0); | |
2984 | if (fd < 0) { | |
2985 | error = fd; | |
2986 | goto err_name; | |
2987 | } | |
2988 | ||
2989 | file = shmem_file_setup(name, 0, VM_NORESERVE); | |
2990 | if (IS_ERR(file)) { | |
2991 | error = PTR_ERR(file); | |
2992 | goto err_fd; | |
2993 | } | |
2994 | info = SHMEM_I(file_inode(file)); | |
2995 | file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; | |
2996 | file->f_flags |= O_RDWR | O_LARGEFILE; | |
2997 | if (flags & MFD_ALLOW_SEALING) | |
2998 | info->seals &= ~F_SEAL_SEAL; | |
2999 | ||
3000 | fd_install(fd, file); | |
3001 | kfree(name); | |
3002 | return fd; | |
3003 | ||
3004 | err_fd: | |
3005 | put_unused_fd(fd); | |
3006 | err_name: | |
3007 | kfree(name); | |
3008 | return error; | |
3009 | } | |
3010 | ||
680d794b | 3011 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3012 | |
3013 | static void shmem_put_super(struct super_block *sb) | |
3014 | { | |
602586a8 HD |
3015 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3016 | ||
3017 | percpu_counter_destroy(&sbinfo->used_blocks); | |
49cd0a5c | 3018 | mpol_put(sbinfo->mpol); |
602586a8 | 3019 | kfree(sbinfo); |
1da177e4 LT |
3020 | sb->s_fs_info = NULL; |
3021 | } | |
3022 | ||
2b2af54a | 3023 | int shmem_fill_super(struct super_block *sb, void *data, int silent) |
1da177e4 LT |
3024 | { |
3025 | struct inode *inode; | |
0edd73b3 | 3026 | struct shmem_sb_info *sbinfo; |
680d794b | 3027 | int err = -ENOMEM; |
3028 | ||
3029 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3030 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 3031 | L1_CACHE_BYTES), GFP_KERNEL); |
3032 | if (!sbinfo) | |
3033 | return -ENOMEM; | |
3034 | ||
680d794b | 3035 | sbinfo->mode = S_IRWXUGO | S_ISVTX; |
76aac0e9 DH |
3036 | sbinfo->uid = current_fsuid(); |
3037 | sbinfo->gid = current_fsgid(); | |
680d794b | 3038 | sb->s_fs_info = sbinfo; |
1da177e4 | 3039 | |
0edd73b3 | 3040 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3041 | /* |
3042 | * Per default we only allow half of the physical ram per | |
3043 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3044 | * but the internal instance is left unlimited. | |
3045 | */ | |
ca4e0519 | 3046 | if (!(sb->s_flags & MS_KERNMOUNT)) { |
680d794b | 3047 | sbinfo->max_blocks = shmem_default_max_blocks(); |
3048 | sbinfo->max_inodes = shmem_default_max_inodes(); | |
3049 | if (shmem_parse_options(data, sbinfo, false)) { | |
3050 | err = -EINVAL; | |
3051 | goto failed; | |
3052 | } | |
ca4e0519 AV |
3053 | } else { |
3054 | sb->s_flags |= MS_NOUSER; | |
1da177e4 | 3055 | } |
91828a40 | 3056 | sb->s_export_op = &shmem_export_ops; |
2f6e38f3 | 3057 | sb->s_flags |= MS_NOSEC; |
1da177e4 LT |
3058 | #else |
3059 | sb->s_flags |= MS_NOUSER; | |
3060 | #endif | |
3061 | ||
0edd73b3 | 3062 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3063 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3064 | goto failed; |
680d794b | 3065 | sbinfo->free_inodes = sbinfo->max_inodes; |
0edd73b3 | 3066 | |
285b2c4f | 3067 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3068 | sb->s_blocksize = PAGE_SIZE; |
3069 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3070 | sb->s_magic = TMPFS_MAGIC; |
3071 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3072 | sb->s_time_gran = 1; |
b09e0fa4 | 3073 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3074 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3075 | #endif |
3076 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
39f0247d AG |
3077 | sb->s_flags |= MS_POSIXACL; |
3078 | #endif | |
0edd73b3 | 3079 | |
454abafe | 3080 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3081 | if (!inode) |
3082 | goto failed; | |
680d794b | 3083 | inode->i_uid = sbinfo->uid; |
3084 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3085 | sb->s_root = d_make_root(inode); |
3086 | if (!sb->s_root) | |
48fde701 | 3087 | goto failed; |
1da177e4 LT |
3088 | return 0; |
3089 | ||
1da177e4 LT |
3090 | failed: |
3091 | shmem_put_super(sb); | |
3092 | return err; | |
3093 | } | |
3094 | ||
fcc234f8 | 3095 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3096 | |
3097 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3098 | { | |
41ffe5d5 HD |
3099 | struct shmem_inode_info *info; |
3100 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3101 | if (!info) | |
1da177e4 | 3102 | return NULL; |
41ffe5d5 | 3103 | return &info->vfs_inode; |
1da177e4 LT |
3104 | } |
3105 | ||
41ffe5d5 | 3106 | static void shmem_destroy_callback(struct rcu_head *head) |
fa0d7e3d NP |
3107 | { |
3108 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
84e710da AV |
3109 | if (S_ISLNK(inode->i_mode)) |
3110 | kfree(inode->i_link); | |
fa0d7e3d NP |
3111 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3112 | } | |
3113 | ||
1da177e4 LT |
3114 | static void shmem_destroy_inode(struct inode *inode) |
3115 | { | |
09208d15 | 3116 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3117 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
41ffe5d5 | 3118 | call_rcu(&inode->i_rcu, shmem_destroy_callback); |
1da177e4 LT |
3119 | } |
3120 | ||
41ffe5d5 | 3121 | static void shmem_init_inode(void *foo) |
1da177e4 | 3122 | { |
41ffe5d5 HD |
3123 | struct shmem_inode_info *info = foo; |
3124 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3125 | } |
3126 | ||
41ffe5d5 | 3127 | static int shmem_init_inodecache(void) |
1da177e4 LT |
3128 | { |
3129 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3130 | sizeof(struct shmem_inode_info), | |
5d097056 | 3131 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3132 | return 0; |
3133 | } | |
3134 | ||
41ffe5d5 | 3135 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3136 | { |
1a1d92c1 | 3137 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3138 | } |
3139 | ||
f5e54d6e | 3140 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 3141 | .writepage = shmem_writepage, |
76719325 | 3142 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3143 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3144 | .write_begin = shmem_write_begin, |
3145 | .write_end = shmem_write_end, | |
1da177e4 | 3146 | #endif |
1c93923c | 3147 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3148 | .migratepage = migrate_page, |
1c93923c | 3149 | #endif |
aa261f54 | 3150 | .error_remove_page = generic_error_remove_page, |
1da177e4 LT |
3151 | }; |
3152 | ||
15ad7cdc | 3153 | static const struct file_operations shmem_file_operations = { |
1da177e4 LT |
3154 | .mmap = shmem_mmap, |
3155 | #ifdef CONFIG_TMPFS | |
220f2ac9 | 3156 | .llseek = shmem_file_llseek, |
2ba5bbed | 3157 | .read_iter = shmem_file_read_iter, |
8174202b | 3158 | .write_iter = generic_file_write_iter, |
1b061d92 | 3159 | .fsync = noop_fsync, |
708e3508 | 3160 | .splice_read = shmem_file_splice_read, |
f6cb85d0 | 3161 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3162 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3163 | #endif |
3164 | }; | |
3165 | ||
92e1d5be | 3166 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3167 | .getattr = shmem_getattr, |
94c1e62d | 3168 | .setattr = shmem_setattr, |
b09e0fa4 | 3169 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3170 | .setxattr = generic_setxattr, |
3171 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3172 | .listxattr = shmem_listxattr, |
aa7c5241 | 3173 | .removexattr = generic_removexattr, |
feda821e | 3174 | .set_acl = simple_set_acl, |
b09e0fa4 | 3175 | #endif |
1da177e4 LT |
3176 | }; |
3177 | ||
92e1d5be | 3178 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3179 | #ifdef CONFIG_TMPFS |
3180 | .create = shmem_create, | |
3181 | .lookup = simple_lookup, | |
3182 | .link = shmem_link, | |
3183 | .unlink = shmem_unlink, | |
3184 | .symlink = shmem_symlink, | |
3185 | .mkdir = shmem_mkdir, | |
3186 | .rmdir = shmem_rmdir, | |
3187 | .mknod = shmem_mknod, | |
3b69ff51 | 3188 | .rename2 = shmem_rename2, |
60545d0d | 3189 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3190 | #endif |
b09e0fa4 | 3191 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3192 | .setxattr = generic_setxattr, |
3193 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3194 | .listxattr = shmem_listxattr, |
aa7c5241 | 3195 | .removexattr = generic_removexattr, |
b09e0fa4 | 3196 | #endif |
39f0247d | 3197 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3198 | .setattr = shmem_setattr, |
feda821e | 3199 | .set_acl = simple_set_acl, |
39f0247d AG |
3200 | #endif |
3201 | }; | |
3202 | ||
92e1d5be | 3203 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3204 | #ifdef CONFIG_TMPFS_XATTR |
aa7c5241 AG |
3205 | .setxattr = generic_setxattr, |
3206 | .getxattr = generic_getxattr, | |
b09e0fa4 | 3207 | .listxattr = shmem_listxattr, |
aa7c5241 | 3208 | .removexattr = generic_removexattr, |
b09e0fa4 | 3209 | #endif |
39f0247d | 3210 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3211 | .setattr = shmem_setattr, |
feda821e | 3212 | .set_acl = simple_set_acl, |
39f0247d | 3213 | #endif |
1da177e4 LT |
3214 | }; |
3215 | ||
759b9775 | 3216 | static const struct super_operations shmem_ops = { |
1da177e4 LT |
3217 | .alloc_inode = shmem_alloc_inode, |
3218 | .destroy_inode = shmem_destroy_inode, | |
3219 | #ifdef CONFIG_TMPFS | |
3220 | .statfs = shmem_statfs, | |
3221 | .remount_fs = shmem_remount_fs, | |
680d794b | 3222 | .show_options = shmem_show_options, |
1da177e4 | 3223 | #endif |
1f895f75 | 3224 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3225 | .drop_inode = generic_delete_inode, |
3226 | .put_super = shmem_put_super, | |
3227 | }; | |
3228 | ||
f0f37e2f | 3229 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3230 | .fault = shmem_fault, |
d7c17551 | 3231 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3232 | #ifdef CONFIG_NUMA |
3233 | .set_policy = shmem_set_policy, | |
3234 | .get_policy = shmem_get_policy, | |
3235 | #endif | |
3236 | }; | |
3237 | ||
3c26ff6e AV |
3238 | static struct dentry *shmem_mount(struct file_system_type *fs_type, |
3239 | int flags, const char *dev_name, void *data) | |
1da177e4 | 3240 | { |
3c26ff6e | 3241 | return mount_nodev(fs_type, flags, data, shmem_fill_super); |
1da177e4 LT |
3242 | } |
3243 | ||
41ffe5d5 | 3244 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3245 | .owner = THIS_MODULE, |
3246 | .name = "tmpfs", | |
3c26ff6e | 3247 | .mount = shmem_mount, |
1da177e4 | 3248 | .kill_sb = kill_litter_super, |
2b8576cb | 3249 | .fs_flags = FS_USERNS_MOUNT, |
1da177e4 | 3250 | }; |
1da177e4 | 3251 | |
41ffe5d5 | 3252 | int __init shmem_init(void) |
1da177e4 LT |
3253 | { |
3254 | int error; | |
3255 | ||
16203a7a RL |
3256 | /* If rootfs called this, don't re-init */ |
3257 | if (shmem_inode_cachep) | |
3258 | return 0; | |
3259 | ||
41ffe5d5 | 3260 | error = shmem_init_inodecache(); |
1da177e4 LT |
3261 | if (error) |
3262 | goto out3; | |
3263 | ||
41ffe5d5 | 3264 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 3265 | if (error) { |
1170532b | 3266 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
3267 | goto out2; |
3268 | } | |
95dc112a | 3269 | |
ca4e0519 | 3270 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
3271 | if (IS_ERR(shm_mnt)) { |
3272 | error = PTR_ERR(shm_mnt); | |
1170532b | 3273 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
3274 | goto out1; |
3275 | } | |
3276 | return 0; | |
3277 | ||
3278 | out1: | |
41ffe5d5 | 3279 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 3280 | out2: |
41ffe5d5 | 3281 | shmem_destroy_inodecache(); |
1da177e4 LT |
3282 | out3: |
3283 | shm_mnt = ERR_PTR(error); | |
3284 | return error; | |
3285 | } | |
853ac43a MM |
3286 | |
3287 | #else /* !CONFIG_SHMEM */ | |
3288 | ||
3289 | /* | |
3290 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
3291 | * | |
3292 | * This is intended for small system where the benefits of the full | |
3293 | * shmem code (swap-backed and resource-limited) are outweighed by | |
3294 | * their complexity. On systems without swap this code should be | |
3295 | * effectively equivalent, but much lighter weight. | |
3296 | */ | |
3297 | ||
41ffe5d5 | 3298 | static struct file_system_type shmem_fs_type = { |
853ac43a | 3299 | .name = "tmpfs", |
3c26ff6e | 3300 | .mount = ramfs_mount, |
853ac43a | 3301 | .kill_sb = kill_litter_super, |
2b8576cb | 3302 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
3303 | }; |
3304 | ||
41ffe5d5 | 3305 | int __init shmem_init(void) |
853ac43a | 3306 | { |
41ffe5d5 | 3307 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 3308 | |
41ffe5d5 | 3309 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
3310 | BUG_ON(IS_ERR(shm_mnt)); |
3311 | ||
3312 | return 0; | |
3313 | } | |
3314 | ||
41ffe5d5 | 3315 | int shmem_unuse(swp_entry_t swap, struct page *page) |
853ac43a MM |
3316 | { |
3317 | return 0; | |
3318 | } | |
3319 | ||
3f96b79a HD |
3320 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
3321 | { | |
3322 | return 0; | |
3323 | } | |
3324 | ||
24513264 HD |
3325 | void shmem_unlock_mapping(struct address_space *mapping) |
3326 | { | |
3327 | } | |
3328 | ||
41ffe5d5 | 3329 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 3330 | { |
41ffe5d5 | 3331 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
3332 | } |
3333 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
3334 | ||
0b0a0806 HD |
3335 | #define shmem_vm_ops generic_file_vm_ops |
3336 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 3337 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
3338 | #define shmem_acct_size(flags, size) 0 |
3339 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
3340 | |
3341 | #endif /* CONFIG_SHMEM */ | |
3342 | ||
3343 | /* common code */ | |
1da177e4 | 3344 | |
3451538a | 3345 | static struct dentry_operations anon_ops = { |
118b2302 | 3346 | .d_dname = simple_dname |
3451538a AV |
3347 | }; |
3348 | ||
c7277090 EP |
3349 | static struct file *__shmem_file_setup(const char *name, loff_t size, |
3350 | unsigned long flags, unsigned int i_flags) | |
1da177e4 | 3351 | { |
6b4d0b27 | 3352 | struct file *res; |
1da177e4 | 3353 | struct inode *inode; |
2c48b9c4 | 3354 | struct path path; |
3451538a | 3355 | struct super_block *sb; |
1da177e4 LT |
3356 | struct qstr this; |
3357 | ||
3358 | if (IS_ERR(shm_mnt)) | |
6b4d0b27 | 3359 | return ERR_CAST(shm_mnt); |
1da177e4 | 3360 | |
285b2c4f | 3361 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
3362 | return ERR_PTR(-EINVAL); |
3363 | ||
3364 | if (shmem_acct_size(flags, size)) | |
3365 | return ERR_PTR(-ENOMEM); | |
3366 | ||
6b4d0b27 | 3367 | res = ERR_PTR(-ENOMEM); |
1da177e4 LT |
3368 | this.name = name; |
3369 | this.len = strlen(name); | |
3370 | this.hash = 0; /* will go */ | |
3451538a | 3371 | sb = shm_mnt->mnt_sb; |
66ee4b88 | 3372 | path.mnt = mntget(shm_mnt); |
3451538a | 3373 | path.dentry = d_alloc_pseudo(sb, &this); |
2c48b9c4 | 3374 | if (!path.dentry) |
1da177e4 | 3375 | goto put_memory; |
3451538a | 3376 | d_set_d_op(path.dentry, &anon_ops); |
1da177e4 | 3377 | |
6b4d0b27 | 3378 | res = ERR_PTR(-ENOSPC); |
3451538a | 3379 | inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); |
1da177e4 | 3380 | if (!inode) |
66ee4b88 | 3381 | goto put_memory; |
1da177e4 | 3382 | |
c7277090 | 3383 | inode->i_flags |= i_flags; |
2c48b9c4 | 3384 | d_instantiate(path.dentry, inode); |
1da177e4 | 3385 | inode->i_size = size; |
6d6b77f1 | 3386 | clear_nlink(inode); /* It is unlinked */ |
26567cdb AV |
3387 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
3388 | if (IS_ERR(res)) | |
66ee4b88 | 3389 | goto put_path; |
4b42af81 | 3390 | |
6b4d0b27 | 3391 | res = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
4b42af81 | 3392 | &shmem_file_operations); |
6b4d0b27 | 3393 | if (IS_ERR(res)) |
66ee4b88 | 3394 | goto put_path; |
4b42af81 | 3395 | |
6b4d0b27 | 3396 | return res; |
1da177e4 | 3397 | |
1da177e4 LT |
3398 | put_memory: |
3399 | shmem_unacct_size(flags, size); | |
66ee4b88 KK |
3400 | put_path: |
3401 | path_put(&path); | |
6b4d0b27 | 3402 | return res; |
1da177e4 | 3403 | } |
c7277090 EP |
3404 | |
3405 | /** | |
3406 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
3407 | * kernel internal. There will be NO LSM permission checks against the | |
3408 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
3409 | * higher layer. The users are the big_key and shm implementations. LSM |
3410 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
3411 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
3412 | * @size: size to be set for the file | |
3413 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3414 | */ | |
3415 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
3416 | { | |
3417 | return __shmem_file_setup(name, size, flags, S_PRIVATE); | |
3418 | } | |
3419 | ||
3420 | /** | |
3421 | * shmem_file_setup - get an unlinked file living in tmpfs | |
3422 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
3423 | * @size: size to be set for the file | |
3424 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
3425 | */ | |
3426 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
3427 | { | |
3428 | return __shmem_file_setup(name, size, flags, 0); | |
3429 | } | |
395e0ddc | 3430 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 3431 | |
46711810 | 3432 | /** |
1da177e4 | 3433 | * shmem_zero_setup - setup a shared anonymous mapping |
1da177e4 LT |
3434 | * @vma: the vma to be mmapped is prepared by do_mmap_pgoff |
3435 | */ | |
3436 | int shmem_zero_setup(struct vm_area_struct *vma) | |
3437 | { | |
3438 | struct file *file; | |
3439 | loff_t size = vma->vm_end - vma->vm_start; | |
3440 | ||
66fc1303 HD |
3441 | /* |
3442 | * Cloning a new file under mmap_sem leads to a lock ordering conflict | |
3443 | * between XFS directory reading and selinux: since this file is only | |
3444 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
3445 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
3446 | */ | |
3447 | file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE); | |
1da177e4 LT |
3448 | if (IS_ERR(file)) |
3449 | return PTR_ERR(file); | |
3450 | ||
3451 | if (vma->vm_file) | |
3452 | fput(vma->vm_file); | |
3453 | vma->vm_file = file; | |
3454 | vma->vm_ops = &shmem_vm_ops; | |
3455 | return 0; | |
3456 | } | |
d9d90e5e HD |
3457 | |
3458 | /** | |
3459 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
3460 | * @mapping: the page's address_space | |
3461 | * @index: the page index | |
3462 | * @gfp: the page allocator flags to use if allocating | |
3463 | * | |
3464 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
3465 | * with any new page allocations done using the specified allocation flags. | |
3466 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
3467 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
3468 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
3469 | * | |
68da9f05 HD |
3470 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
3471 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
3472 | */ |
3473 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
3474 | pgoff_t index, gfp_t gfp) | |
3475 | { | |
68da9f05 HD |
3476 | #ifdef CONFIG_SHMEM |
3477 | struct inode *inode = mapping->host; | |
9276aad6 | 3478 | struct page *page; |
68da9f05 HD |
3479 | int error; |
3480 | ||
3481 | BUG_ON(mapping->a_ops != &shmem_aops); | |
3482 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL); | |
3483 | if (error) | |
3484 | page = ERR_PTR(error); | |
3485 | else | |
3486 | unlock_page(page); | |
3487 | return page; | |
3488 | #else | |
3489 | /* | |
3490 | * The tiny !SHMEM case uses ramfs without swap | |
3491 | */ | |
d9d90e5e | 3492 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 3493 | #endif |
d9d90e5e HD |
3494 | } |
3495 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |