]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Resizable virtual memory filesystem for Linux. | |
3 | * | |
4 | * Copyright (C) 2000 Linus Torvalds. | |
5 | * 2000 Transmeta Corp. | |
6 | * 2000-2001 Christoph Rohland | |
7 | * 2000-2001 SAP AG | |
8 | * 2002 Red Hat Inc. | |
6922c0c7 HD |
9 | * Copyright (C) 2002-2011 Hugh Dickins. |
10 | * Copyright (C) 2011 Google Inc. | |
0edd73b3 | 11 | * Copyright (C) 2002-2005 VERITAS Software Corporation. |
1da177e4 LT |
12 | * Copyright (C) 2004 Andi Kleen, SuSE Labs |
13 | * | |
14 | * Extended attribute support for tmpfs: | |
15 | * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net> | |
16 | * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com> | |
17 | * | |
853ac43a MM |
18 | * tiny-shmem: |
19 | * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com> | |
20 | * | |
1da177e4 LT |
21 | * This file is released under the GPL. |
22 | */ | |
23 | ||
853ac43a MM |
24 | #include <linux/fs.h> |
25 | #include <linux/init.h> | |
26 | #include <linux/vfs.h> | |
27 | #include <linux/mount.h> | |
250297ed | 28 | #include <linux/ramfs.h> |
caefba17 | 29 | #include <linux/pagemap.h> |
853ac43a MM |
30 | #include <linux/file.h> |
31 | #include <linux/mm.h> | |
46c9a946 | 32 | #include <linux/random.h> |
174cd4b1 | 33 | #include <linux/sched/signal.h> |
b95f1b31 | 34 | #include <linux/export.h> |
853ac43a | 35 | #include <linux/swap.h> |
e2e40f2c | 36 | #include <linux/uio.h> |
f3f0e1d2 | 37 | #include <linux/khugepaged.h> |
749df87b | 38 | #include <linux/hugetlb.h> |
b56a2d8a | 39 | #include <linux/frontswap.h> |
626c3920 | 40 | #include <linux/fs_parser.h> |
86a2f3f2 | 41 | #include <linux/swapfile.h> |
95cc09d6 | 42 | |
853ac43a MM |
43 | static struct vfsmount *shm_mnt; |
44 | ||
45 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
46 | /* |
47 | * This virtual memory filesystem is heavily based on the ramfs. It | |
48 | * extends ramfs by the ability to use swap and honor resource limits | |
49 | * which makes it a completely usable filesystem. | |
50 | */ | |
51 | ||
39f0247d | 52 | #include <linux/xattr.h> |
a5694255 | 53 | #include <linux/exportfs.h> |
1c7c474c | 54 | #include <linux/posix_acl.h> |
feda821e | 55 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 56 | #include <linux/mman.h> |
1da177e4 LT |
57 | #include <linux/string.h> |
58 | #include <linux/slab.h> | |
59 | #include <linux/backing-dev.h> | |
60 | #include <linux/shmem_fs.h> | |
1da177e4 | 61 | #include <linux/writeback.h> |
1da177e4 | 62 | #include <linux/blkdev.h> |
bda97eab | 63 | #include <linux/pagevec.h> |
41ffe5d5 | 64 | #include <linux/percpu_counter.h> |
83e4fa9c | 65 | #include <linux/falloc.h> |
708e3508 | 66 | #include <linux/splice.h> |
1da177e4 LT |
67 | #include <linux/security.h> |
68 | #include <linux/swapops.h> | |
69 | #include <linux/mempolicy.h> | |
70 | #include <linux/namei.h> | |
b00dc3ad | 71 | #include <linux/ctype.h> |
304dbdb7 | 72 | #include <linux/migrate.h> |
c1f60a5a | 73 | #include <linux/highmem.h> |
680d794b | 74 | #include <linux/seq_file.h> |
92562927 | 75 | #include <linux/magic.h> |
9183df25 | 76 | #include <linux/syscalls.h> |
40e041a2 | 77 | #include <linux/fcntl.h> |
9183df25 | 78 | #include <uapi/linux/memfd.h> |
cfda0526 | 79 | #include <linux/userfaultfd_k.h> |
4c27fe4c | 80 | #include <linux/rmap.h> |
2b4db796 | 81 | #include <linux/uuid.h> |
304dbdb7 | 82 | |
7c0f6ba6 | 83 | #include <linux/uaccess.h> |
1da177e4 | 84 | |
dd56b046 MG |
85 | #include "internal.h" |
86 | ||
09cbfeaf KS |
87 | #define BLOCKS_PER_PAGE (PAGE_SIZE/512) |
88 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT) | |
1da177e4 | 89 | |
1da177e4 LT |
90 | /* Pretend that each entry is of this size in directory's i_size */ |
91 | #define BOGO_DIRENT_SIZE 20 | |
92 | ||
69f07ec9 HD |
93 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
94 | #define SHORT_SYMLINK_LEN 128 | |
95 | ||
1aac1400 | 96 | /* |
f00cdc6d | 97 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
9608703e | 98 | * inode->i_private (with i_rwsem making sure that it has only one user at |
f00cdc6d | 99 | * a time): we would prefer not to enlarge the shmem inode just for that. |
1aac1400 HD |
100 | */ |
101 | struct shmem_falloc { | |
8e205f77 | 102 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
103 | pgoff_t start; /* start of range currently being fallocated */ |
104 | pgoff_t next; /* the next page offset to be fallocated */ | |
105 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
106 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
107 | }; | |
108 | ||
0b5071dd AV |
109 | struct shmem_options { |
110 | unsigned long long blocks; | |
111 | unsigned long long inodes; | |
112 | struct mempolicy *mpol; | |
113 | kuid_t uid; | |
114 | kgid_t gid; | |
115 | umode_t mode; | |
ea3271f7 | 116 | bool full_inums; |
0b5071dd AV |
117 | int huge; |
118 | int seen; | |
119 | #define SHMEM_SEEN_BLOCKS 1 | |
120 | #define SHMEM_SEEN_INODES 2 | |
121 | #define SHMEM_SEEN_HUGE 4 | |
ea3271f7 | 122 | #define SHMEM_SEEN_INUMS 8 |
0b5071dd AV |
123 | }; |
124 | ||
b76db735 | 125 | #ifdef CONFIG_TMPFS |
680d794b AM |
126 | static unsigned long shmem_default_max_blocks(void) |
127 | { | |
ca79b0c2 | 128 | return totalram_pages() / 2; |
680d794b AM |
129 | } |
130 | ||
131 | static unsigned long shmem_default_max_inodes(void) | |
132 | { | |
ca79b0c2 AK |
133 | unsigned long nr_pages = totalram_pages(); |
134 | ||
135 | return min(nr_pages - totalhigh_pages(), nr_pages / 2); | |
680d794b | 136 | } |
b76db735 | 137 | #endif |
680d794b | 138 | |
c5bf121e VRP |
139 | static int shmem_swapin_page(struct inode *inode, pgoff_t index, |
140 | struct page **pagep, enum sgp_type sgp, | |
141 | gfp_t gfp, struct vm_area_struct *vma, | |
142 | vm_fault_t *fault_type); | |
68da9f05 | 143 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 144 | struct page **pagep, enum sgp_type sgp, |
cfda0526 | 145 | gfp_t gfp, struct vm_area_struct *vma, |
2b740303 | 146 | struct vm_fault *vmf, vm_fault_t *fault_type); |
68da9f05 | 147 | |
f3f0e1d2 | 148 | int shmem_getpage(struct inode *inode, pgoff_t index, |
9e18eb29 | 149 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
150 | { |
151 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
cfda0526 | 152 | mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL); |
68da9f05 | 153 | } |
1da177e4 | 154 | |
1da177e4 LT |
155 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
156 | { | |
157 | return sb->s_fs_info; | |
158 | } | |
159 | ||
160 | /* | |
161 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
162 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
163 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
164 | * consistent with the pre-accounting of private mappings ... | |
165 | */ | |
166 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
167 | { | |
0b0a0806 | 168 | return (flags & VM_NORESERVE) ? |
191c5424 | 169 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
170 | } |
171 | ||
172 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
173 | { | |
0b0a0806 | 174 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
175 | vm_unacct_memory(VM_ACCT(size)); |
176 | } | |
177 | ||
77142517 KK |
178 | static inline int shmem_reacct_size(unsigned long flags, |
179 | loff_t oldsize, loff_t newsize) | |
180 | { | |
181 | if (!(flags & VM_NORESERVE)) { | |
182 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
183 | return security_vm_enough_memory_mm(current->mm, | |
184 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
185 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
186 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
187 | } | |
188 | return 0; | |
189 | } | |
190 | ||
1da177e4 LT |
191 | /* |
192 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 193 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
194 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
195 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
196 | */ | |
800d8c63 | 197 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 198 | { |
800d8c63 KS |
199 | if (!(flags & VM_NORESERVE)) |
200 | return 0; | |
201 | ||
202 | return security_vm_enough_memory_mm(current->mm, | |
203 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
204 | } |
205 | ||
206 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
207 | { | |
0b0a0806 | 208 | if (flags & VM_NORESERVE) |
09cbfeaf | 209 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
210 | } |
211 | ||
0f079694 MR |
212 | static inline bool shmem_inode_acct_block(struct inode *inode, long pages) |
213 | { | |
214 | struct shmem_inode_info *info = SHMEM_I(inode); | |
215 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
216 | ||
217 | if (shmem_acct_block(info->flags, pages)) | |
218 | return false; | |
219 | ||
220 | if (sbinfo->max_blocks) { | |
221 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
222 | sbinfo->max_blocks - pages) > 0) | |
223 | goto unacct; | |
224 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
225 | } | |
226 | ||
227 | return true; | |
228 | ||
229 | unacct: | |
230 | shmem_unacct_blocks(info->flags, pages); | |
231 | return false; | |
232 | } | |
233 | ||
234 | static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages) | |
235 | { | |
236 | struct shmem_inode_info *info = SHMEM_I(inode); | |
237 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
238 | ||
239 | if (sbinfo->max_blocks) | |
240 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
241 | shmem_unacct_blocks(info->flags, pages); | |
242 | } | |
243 | ||
759b9775 | 244 | static const struct super_operations shmem_ops; |
30e6a51d | 245 | const struct address_space_operations shmem_aops; |
15ad7cdc | 246 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
247 | static const struct inode_operations shmem_inode_operations; |
248 | static const struct inode_operations shmem_dir_inode_operations; | |
249 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 250 | static const struct vm_operations_struct shmem_vm_ops; |
779750d2 | 251 | static struct file_system_type shmem_fs_type; |
1da177e4 | 252 | |
b0506e48 MR |
253 | bool vma_is_shmem(struct vm_area_struct *vma) |
254 | { | |
255 | return vma->vm_ops == &shmem_vm_ops; | |
256 | } | |
257 | ||
1da177e4 | 258 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 259 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 260 | |
e809d5f0 CD |
261 | /* |
262 | * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and | |
263 | * produces a novel ino for the newly allocated inode. | |
264 | * | |
265 | * It may also be called when making a hard link to permit the space needed by | |
266 | * each dentry. However, in that case, no new inode number is needed since that | |
267 | * internally draws from another pool of inode numbers (currently global | |
268 | * get_next_ino()). This case is indicated by passing NULL as inop. | |
269 | */ | |
270 | #define SHMEM_INO_BATCH 1024 | |
271 | static int shmem_reserve_inode(struct super_block *sb, ino_t *inop) | |
5b04c689 PE |
272 | { |
273 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
e809d5f0 CD |
274 | ino_t ino; |
275 | ||
276 | if (!(sb->s_flags & SB_KERNMOUNT)) { | |
bf11b9a8 | 277 | raw_spin_lock(&sbinfo->stat_lock); |
bb3e96d6 BS |
278 | if (sbinfo->max_inodes) { |
279 | if (!sbinfo->free_inodes) { | |
bf11b9a8 | 280 | raw_spin_unlock(&sbinfo->stat_lock); |
bb3e96d6 BS |
281 | return -ENOSPC; |
282 | } | |
283 | sbinfo->free_inodes--; | |
5b04c689 | 284 | } |
e809d5f0 CD |
285 | if (inop) { |
286 | ino = sbinfo->next_ino++; | |
287 | if (unlikely(is_zero_ino(ino))) | |
288 | ino = sbinfo->next_ino++; | |
ea3271f7 CD |
289 | if (unlikely(!sbinfo->full_inums && |
290 | ino > UINT_MAX)) { | |
e809d5f0 CD |
291 | /* |
292 | * Emulate get_next_ino uint wraparound for | |
293 | * compatibility | |
294 | */ | |
ea3271f7 CD |
295 | if (IS_ENABLED(CONFIG_64BIT)) |
296 | pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n", | |
297 | __func__, MINOR(sb->s_dev)); | |
298 | sbinfo->next_ino = 1; | |
299 | ino = sbinfo->next_ino++; | |
e809d5f0 CD |
300 | } |
301 | *inop = ino; | |
302 | } | |
bf11b9a8 | 303 | raw_spin_unlock(&sbinfo->stat_lock); |
e809d5f0 CD |
304 | } else if (inop) { |
305 | /* | |
306 | * __shmem_file_setup, one of our callers, is lock-free: it | |
307 | * doesn't hold stat_lock in shmem_reserve_inode since | |
308 | * max_inodes is always 0, and is called from potentially | |
309 | * unknown contexts. As such, use a per-cpu batched allocator | |
310 | * which doesn't require the per-sb stat_lock unless we are at | |
311 | * the batch boundary. | |
ea3271f7 CD |
312 | * |
313 | * We don't need to worry about inode{32,64} since SB_KERNMOUNT | |
314 | * shmem mounts are not exposed to userspace, so we don't need | |
315 | * to worry about things like glibc compatibility. | |
e809d5f0 CD |
316 | */ |
317 | ino_t *next_ino; | |
bf11b9a8 | 318 | |
e809d5f0 CD |
319 | next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu()); |
320 | ino = *next_ino; | |
321 | if (unlikely(ino % SHMEM_INO_BATCH == 0)) { | |
bf11b9a8 | 322 | raw_spin_lock(&sbinfo->stat_lock); |
e809d5f0 CD |
323 | ino = sbinfo->next_ino; |
324 | sbinfo->next_ino += SHMEM_INO_BATCH; | |
bf11b9a8 | 325 | raw_spin_unlock(&sbinfo->stat_lock); |
e809d5f0 CD |
326 | if (unlikely(is_zero_ino(ino))) |
327 | ino++; | |
328 | } | |
329 | *inop = ino; | |
330 | *next_ino = ++ino; | |
331 | put_cpu(); | |
5b04c689 | 332 | } |
e809d5f0 | 333 | |
5b04c689 PE |
334 | return 0; |
335 | } | |
336 | ||
337 | static void shmem_free_inode(struct super_block *sb) | |
338 | { | |
339 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
340 | if (sbinfo->max_inodes) { | |
bf11b9a8 | 341 | raw_spin_lock(&sbinfo->stat_lock); |
5b04c689 | 342 | sbinfo->free_inodes++; |
bf11b9a8 | 343 | raw_spin_unlock(&sbinfo->stat_lock); |
5b04c689 PE |
344 | } |
345 | } | |
346 | ||
46711810 | 347 | /** |
41ffe5d5 | 348 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
349 | * @inode: inode to recalc |
350 | * | |
351 | * We have to calculate the free blocks since the mm can drop | |
352 | * undirtied hole pages behind our back. | |
353 | * | |
354 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
355 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
356 | * | |
357 | * It has to be called with the spinlock held. | |
358 | */ | |
359 | static void shmem_recalc_inode(struct inode *inode) | |
360 | { | |
361 | struct shmem_inode_info *info = SHMEM_I(inode); | |
362 | long freed; | |
363 | ||
364 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
365 | if (freed > 0) { | |
366 | info->alloced -= freed; | |
54af6042 | 367 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
0f079694 | 368 | shmem_inode_unacct_blocks(inode, freed); |
1da177e4 LT |
369 | } |
370 | } | |
371 | ||
800d8c63 KS |
372 | bool shmem_charge(struct inode *inode, long pages) |
373 | { | |
374 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 375 | unsigned long flags; |
800d8c63 | 376 | |
0f079694 | 377 | if (!shmem_inode_acct_block(inode, pages)) |
800d8c63 | 378 | return false; |
b1cc94ab | 379 | |
aaa52e34 HD |
380 | /* nrpages adjustment first, then shmem_recalc_inode() when balanced */ |
381 | inode->i_mapping->nrpages += pages; | |
382 | ||
4595ef88 | 383 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
384 | info->alloced += pages; |
385 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
386 | shmem_recalc_inode(inode); | |
4595ef88 | 387 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 | 388 | |
800d8c63 KS |
389 | return true; |
390 | } | |
391 | ||
392 | void shmem_uncharge(struct inode *inode, long pages) | |
393 | { | |
394 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 395 | unsigned long flags; |
800d8c63 | 396 | |
aaa52e34 HD |
397 | /* nrpages adjustment done by __delete_from_page_cache() or caller */ |
398 | ||
4595ef88 | 399 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
400 | info->alloced -= pages; |
401 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
402 | shmem_recalc_inode(inode); | |
4595ef88 | 403 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 | 404 | |
0f079694 | 405 | shmem_inode_unacct_blocks(inode, pages); |
800d8c63 KS |
406 | } |
407 | ||
7a5d0fbb | 408 | /* |
62f945b6 | 409 | * Replace item expected in xarray by a new item, while holding xa_lock. |
7a5d0fbb | 410 | */ |
62f945b6 | 411 | static int shmem_replace_entry(struct address_space *mapping, |
7a5d0fbb HD |
412 | pgoff_t index, void *expected, void *replacement) |
413 | { | |
62f945b6 | 414 | XA_STATE(xas, &mapping->i_pages, index); |
6dbaf22c | 415 | void *item; |
7a5d0fbb HD |
416 | |
417 | VM_BUG_ON(!expected); | |
6dbaf22c | 418 | VM_BUG_ON(!replacement); |
62f945b6 | 419 | item = xas_load(&xas); |
7a5d0fbb HD |
420 | if (item != expected) |
421 | return -ENOENT; | |
62f945b6 | 422 | xas_store(&xas, replacement); |
7a5d0fbb HD |
423 | return 0; |
424 | } | |
425 | ||
d1899228 HD |
426 | /* |
427 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
428 | * that an entry was not already brought back from swap by a racing thread. | |
429 | * | |
430 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
431 | * might be reused, and again be SwapCache, using the same swap as before. | |
432 | */ | |
433 | static bool shmem_confirm_swap(struct address_space *mapping, | |
434 | pgoff_t index, swp_entry_t swap) | |
435 | { | |
a12831bf | 436 | return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap); |
d1899228 HD |
437 | } |
438 | ||
5a6e75f8 KS |
439 | /* |
440 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
441 | * | |
442 | * SHMEM_HUGE_NEVER: | |
443 | * disables huge pages for the mount; | |
444 | * SHMEM_HUGE_ALWAYS: | |
445 | * enables huge pages for the mount; | |
446 | * SHMEM_HUGE_WITHIN_SIZE: | |
447 | * only allocate huge pages if the page will be fully within i_size, | |
448 | * also respect fadvise()/madvise() hints; | |
449 | * SHMEM_HUGE_ADVISE: | |
450 | * only allocate huge pages if requested with fadvise()/madvise(); | |
451 | */ | |
452 | ||
453 | #define SHMEM_HUGE_NEVER 0 | |
454 | #define SHMEM_HUGE_ALWAYS 1 | |
455 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
456 | #define SHMEM_HUGE_ADVISE 3 | |
457 | ||
458 | /* | |
459 | * Special values. | |
460 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
461 | * | |
462 | * SHMEM_HUGE_DENY: | |
463 | * disables huge on shm_mnt and all mounts, for emergency use; | |
464 | * SHMEM_HUGE_FORCE: | |
465 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
466 | * | |
467 | */ | |
468 | #define SHMEM_HUGE_DENY (-1) | |
469 | #define SHMEM_HUGE_FORCE (-2) | |
470 | ||
396bcc52 | 471 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5a6e75f8 KS |
472 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
473 | ||
5e6e5a12 | 474 | static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER; |
5a6e75f8 | 475 | |
5e6e5a12 HD |
476 | bool shmem_is_huge(struct vm_area_struct *vma, |
477 | struct inode *inode, pgoff_t index) | |
c852023e | 478 | { |
c852023e | 479 | loff_t i_size; |
c852023e | 480 | |
c852023e HD |
481 | if (shmem_huge == SHMEM_HUGE_DENY) |
482 | return false; | |
5e6e5a12 HD |
483 | if (vma && ((vma->vm_flags & VM_NOHUGEPAGE) || |
484 | test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))) | |
c852023e | 485 | return false; |
5e6e5a12 HD |
486 | if (shmem_huge == SHMEM_HUGE_FORCE) |
487 | return true; | |
488 | ||
489 | switch (SHMEM_SB(inode->i_sb)->huge) { | |
c852023e HD |
490 | case SHMEM_HUGE_ALWAYS: |
491 | return true; | |
492 | case SHMEM_HUGE_WITHIN_SIZE: | |
de6ee659 | 493 | index = round_up(index + 1, HPAGE_PMD_NR); |
c852023e | 494 | i_size = round_up(i_size_read(inode), PAGE_SIZE); |
de6ee659 | 495 | if (i_size >> PAGE_SHIFT >= index) |
c852023e HD |
496 | return true; |
497 | fallthrough; | |
498 | case SHMEM_HUGE_ADVISE: | |
5e6e5a12 HD |
499 | if (vma && (vma->vm_flags & VM_HUGEPAGE)) |
500 | return true; | |
501 | fallthrough; | |
c852023e | 502 | default: |
c852023e HD |
503 | return false; |
504 | } | |
505 | } | |
5a6e75f8 | 506 | |
e5f2249a | 507 | #if defined(CONFIG_SYSFS) |
5a6e75f8 KS |
508 | static int shmem_parse_huge(const char *str) |
509 | { | |
510 | if (!strcmp(str, "never")) | |
511 | return SHMEM_HUGE_NEVER; | |
512 | if (!strcmp(str, "always")) | |
513 | return SHMEM_HUGE_ALWAYS; | |
514 | if (!strcmp(str, "within_size")) | |
515 | return SHMEM_HUGE_WITHIN_SIZE; | |
516 | if (!strcmp(str, "advise")) | |
517 | return SHMEM_HUGE_ADVISE; | |
518 | if (!strcmp(str, "deny")) | |
519 | return SHMEM_HUGE_DENY; | |
520 | if (!strcmp(str, "force")) | |
521 | return SHMEM_HUGE_FORCE; | |
522 | return -EINVAL; | |
523 | } | |
e5f2249a | 524 | #endif |
5a6e75f8 | 525 | |
e5f2249a | 526 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
527 | static const char *shmem_format_huge(int huge) |
528 | { | |
529 | switch (huge) { | |
530 | case SHMEM_HUGE_NEVER: | |
531 | return "never"; | |
532 | case SHMEM_HUGE_ALWAYS: | |
533 | return "always"; | |
534 | case SHMEM_HUGE_WITHIN_SIZE: | |
535 | return "within_size"; | |
536 | case SHMEM_HUGE_ADVISE: | |
537 | return "advise"; | |
538 | case SHMEM_HUGE_DENY: | |
539 | return "deny"; | |
540 | case SHMEM_HUGE_FORCE: | |
541 | return "force"; | |
542 | default: | |
543 | VM_BUG_ON(1); | |
544 | return "bad_val"; | |
545 | } | |
546 | } | |
f1f5929c | 547 | #endif |
5a6e75f8 | 548 | |
779750d2 KS |
549 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
550 | struct shrink_control *sc, unsigned long nr_to_split) | |
551 | { | |
552 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 553 | LIST_HEAD(to_remove); |
779750d2 KS |
554 | struct inode *inode; |
555 | struct shmem_inode_info *info; | |
556 | struct page *page; | |
557 | unsigned long batch = sc ? sc->nr_to_scan : 128; | |
950c3043 | 558 | int split = 0; |
779750d2 KS |
559 | |
560 | if (list_empty(&sbinfo->shrinklist)) | |
561 | return SHRINK_STOP; | |
562 | ||
563 | spin_lock(&sbinfo->shrinklist_lock); | |
564 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
565 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
566 | ||
567 | /* pin the inode */ | |
568 | inode = igrab(&info->vfs_inode); | |
569 | ||
570 | /* inode is about to be evicted */ | |
571 | if (!inode) { | |
572 | list_del_init(&info->shrinklist); | |
779750d2 KS |
573 | goto next; |
574 | } | |
575 | ||
576 | /* Check if there's anything to gain */ | |
577 | if (round_up(inode->i_size, PAGE_SIZE) == | |
578 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 579 | list_move(&info->shrinklist, &to_remove); |
779750d2 KS |
580 | goto next; |
581 | } | |
582 | ||
583 | list_move(&info->shrinklist, &list); | |
584 | next: | |
950c3043 | 585 | sbinfo->shrinklist_len--; |
779750d2 KS |
586 | if (!--batch) |
587 | break; | |
588 | } | |
589 | spin_unlock(&sbinfo->shrinklist_lock); | |
590 | ||
253fd0f0 KS |
591 | list_for_each_safe(pos, next, &to_remove) { |
592 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
593 | inode = &info->vfs_inode; | |
594 | list_del_init(&info->shrinklist); | |
595 | iput(inode); | |
596 | } | |
597 | ||
779750d2 KS |
598 | list_for_each_safe(pos, next, &list) { |
599 | int ret; | |
600 | ||
601 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
602 | inode = &info->vfs_inode; | |
603 | ||
b3cd54b2 | 604 | if (nr_to_split && split >= nr_to_split) |
950c3043 | 605 | goto move_back; |
779750d2 | 606 | |
b3cd54b2 | 607 | page = find_get_page(inode->i_mapping, |
779750d2 KS |
608 | (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT); |
609 | if (!page) | |
610 | goto drop; | |
611 | ||
b3cd54b2 | 612 | /* No huge page at the end of the file: nothing to split */ |
779750d2 | 613 | if (!PageTransHuge(page)) { |
779750d2 KS |
614 | put_page(page); |
615 | goto drop; | |
616 | } | |
617 | ||
b3cd54b2 | 618 | /* |
950c3043 GL |
619 | * Move the inode on the list back to shrinklist if we failed |
620 | * to lock the page at this time. | |
b3cd54b2 KS |
621 | * |
622 | * Waiting for the lock may lead to deadlock in the | |
623 | * reclaim path. | |
624 | */ | |
625 | if (!trylock_page(page)) { | |
626 | put_page(page); | |
950c3043 | 627 | goto move_back; |
b3cd54b2 KS |
628 | } |
629 | ||
779750d2 KS |
630 | ret = split_huge_page(page); |
631 | unlock_page(page); | |
632 | put_page(page); | |
633 | ||
950c3043 | 634 | /* If split failed move the inode on the list back to shrinklist */ |
b3cd54b2 | 635 | if (ret) |
950c3043 | 636 | goto move_back; |
779750d2 KS |
637 | |
638 | split++; | |
639 | drop: | |
640 | list_del_init(&info->shrinklist); | |
950c3043 GL |
641 | goto put; |
642 | move_back: | |
643 | /* | |
644 | * Make sure the inode is either on the global list or deleted | |
645 | * from any local list before iput() since it could be deleted | |
646 | * in another thread once we put the inode (then the local list | |
647 | * is corrupted). | |
648 | */ | |
649 | spin_lock(&sbinfo->shrinklist_lock); | |
650 | list_move(&info->shrinklist, &sbinfo->shrinklist); | |
651 | sbinfo->shrinklist_len++; | |
652 | spin_unlock(&sbinfo->shrinklist_lock); | |
653 | put: | |
779750d2 KS |
654 | iput(inode); |
655 | } | |
656 | ||
779750d2 KS |
657 | return split; |
658 | } | |
659 | ||
660 | static long shmem_unused_huge_scan(struct super_block *sb, | |
661 | struct shrink_control *sc) | |
662 | { | |
663 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
664 | ||
665 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
666 | return SHRINK_STOP; | |
667 | ||
668 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
669 | } | |
670 | ||
671 | static long shmem_unused_huge_count(struct super_block *sb, | |
672 | struct shrink_control *sc) | |
673 | { | |
674 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
675 | return READ_ONCE(sbinfo->shrinklist_len); | |
676 | } | |
396bcc52 | 677 | #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ |
5a6e75f8 KS |
678 | |
679 | #define shmem_huge SHMEM_HUGE_DENY | |
680 | ||
5e6e5a12 HD |
681 | bool shmem_is_huge(struct vm_area_struct *vma, |
682 | struct inode *inode, pgoff_t index) | |
683 | { | |
684 | return false; | |
685 | } | |
686 | ||
779750d2 KS |
687 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
688 | struct shrink_control *sc, unsigned long nr_to_split) | |
689 | { | |
690 | return 0; | |
691 | } | |
396bcc52 | 692 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
5a6e75f8 | 693 | |
46f65ec1 HD |
694 | /* |
695 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
696 | */ | |
697 | static int shmem_add_to_page_cache(struct page *page, | |
698 | struct address_space *mapping, | |
3fea5a49 JW |
699 | pgoff_t index, void *expected, gfp_t gfp, |
700 | struct mm_struct *charge_mm) | |
46f65ec1 | 701 | { |
552446a4 MW |
702 | XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page)); |
703 | unsigned long i = 0; | |
d8c6546b | 704 | unsigned long nr = compound_nr(page); |
3fea5a49 | 705 | int error; |
46f65ec1 | 706 | |
800d8c63 KS |
707 | VM_BUG_ON_PAGE(PageTail(page), page); |
708 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
709 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
710 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 711 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 712 | |
800d8c63 | 713 | page_ref_add(page, nr); |
b065b432 HD |
714 | page->mapping = mapping; |
715 | page->index = index; | |
716 | ||
4c6355b2 | 717 | if (!PageSwapCache(page)) { |
d9eb1ea2 | 718 | error = mem_cgroup_charge(page, charge_mm, gfp); |
4c6355b2 JW |
719 | if (error) { |
720 | if (PageTransHuge(page)) { | |
721 | count_vm_event(THP_FILE_FALLBACK); | |
722 | count_vm_event(THP_FILE_FALLBACK_CHARGE); | |
723 | } | |
724 | goto error; | |
3fea5a49 | 725 | } |
3fea5a49 JW |
726 | } |
727 | cgroup_throttle_swaprate(page, gfp); | |
728 | ||
552446a4 MW |
729 | do { |
730 | void *entry; | |
731 | xas_lock_irq(&xas); | |
732 | entry = xas_find_conflict(&xas); | |
733 | if (entry != expected) | |
734 | xas_set_err(&xas, -EEXIST); | |
735 | xas_create_range(&xas); | |
736 | if (xas_error(&xas)) | |
737 | goto unlock; | |
738 | next: | |
4101196b | 739 | xas_store(&xas, page); |
552446a4 MW |
740 | if (++i < nr) { |
741 | xas_next(&xas); | |
742 | goto next; | |
800d8c63 | 743 | } |
552446a4 | 744 | if (PageTransHuge(page)) { |
800d8c63 | 745 | count_vm_event(THP_FILE_ALLOC); |
57b2847d | 746 | __mod_lruvec_page_state(page, NR_SHMEM_THPS, nr); |
800d8c63 | 747 | } |
800d8c63 | 748 | mapping->nrpages += nr; |
0d1c2072 JW |
749 | __mod_lruvec_page_state(page, NR_FILE_PAGES, nr); |
750 | __mod_lruvec_page_state(page, NR_SHMEM, nr); | |
552446a4 MW |
751 | unlock: |
752 | xas_unlock_irq(&xas); | |
753 | } while (xas_nomem(&xas, gfp)); | |
754 | ||
755 | if (xas_error(&xas)) { | |
3fea5a49 JW |
756 | error = xas_error(&xas); |
757 | goto error; | |
46f65ec1 | 758 | } |
552446a4 MW |
759 | |
760 | return 0; | |
3fea5a49 JW |
761 | error: |
762 | page->mapping = NULL; | |
763 | page_ref_sub(page, nr); | |
764 | return error; | |
46f65ec1 HD |
765 | } |
766 | ||
6922c0c7 HD |
767 | /* |
768 | * Like delete_from_page_cache, but substitutes swap for page. | |
769 | */ | |
770 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
771 | { | |
772 | struct address_space *mapping = page->mapping; | |
773 | int error; | |
774 | ||
800d8c63 KS |
775 | VM_BUG_ON_PAGE(PageCompound(page), page); |
776 | ||
b93b0163 | 777 | xa_lock_irq(&mapping->i_pages); |
62f945b6 | 778 | error = shmem_replace_entry(mapping, page->index, page, radswap); |
6922c0c7 HD |
779 | page->mapping = NULL; |
780 | mapping->nrpages--; | |
0d1c2072 JW |
781 | __dec_lruvec_page_state(page, NR_FILE_PAGES); |
782 | __dec_lruvec_page_state(page, NR_SHMEM); | |
b93b0163 | 783 | xa_unlock_irq(&mapping->i_pages); |
09cbfeaf | 784 | put_page(page); |
6922c0c7 HD |
785 | BUG_ON(error); |
786 | } | |
787 | ||
7a5d0fbb | 788 | /* |
c121d3bb | 789 | * Remove swap entry from page cache, free the swap and its page cache. |
7a5d0fbb HD |
790 | */ |
791 | static int shmem_free_swap(struct address_space *mapping, | |
792 | pgoff_t index, void *radswap) | |
793 | { | |
6dbaf22c | 794 | void *old; |
7a5d0fbb | 795 | |
55f3f7ea | 796 | old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0); |
6dbaf22c JW |
797 | if (old != radswap) |
798 | return -ENOENT; | |
799 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
800 | return 0; | |
7a5d0fbb HD |
801 | } |
802 | ||
6a15a370 VB |
803 | /* |
804 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 805 | * given offsets are swapped out. |
6a15a370 | 806 | * |
9608703e | 807 | * This is safe to call without i_rwsem or the i_pages lock thanks to RCU, |
6a15a370 VB |
808 | * as long as the inode doesn't go away and racy results are not a problem. |
809 | */ | |
48131e03 VB |
810 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
811 | pgoff_t start, pgoff_t end) | |
6a15a370 | 812 | { |
7ae3424f | 813 | XA_STATE(xas, &mapping->i_pages, start); |
6a15a370 | 814 | struct page *page; |
48131e03 | 815 | unsigned long swapped = 0; |
6a15a370 VB |
816 | |
817 | rcu_read_lock(); | |
7ae3424f MW |
818 | xas_for_each(&xas, page, end - 1) { |
819 | if (xas_retry(&xas, page)) | |
2cf938aa | 820 | continue; |
3159f943 | 821 | if (xa_is_value(page)) |
6a15a370 VB |
822 | swapped++; |
823 | ||
824 | if (need_resched()) { | |
7ae3424f | 825 | xas_pause(&xas); |
6a15a370 | 826 | cond_resched_rcu(); |
6a15a370 VB |
827 | } |
828 | } | |
829 | ||
830 | rcu_read_unlock(); | |
831 | ||
832 | return swapped << PAGE_SHIFT; | |
833 | } | |
834 | ||
48131e03 VB |
835 | /* |
836 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
837 | * given vma is swapped out. | |
838 | * | |
9608703e | 839 | * This is safe to call without i_rwsem or the i_pages lock thanks to RCU, |
48131e03 VB |
840 | * as long as the inode doesn't go away and racy results are not a problem. |
841 | */ | |
842 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
843 | { | |
844 | struct inode *inode = file_inode(vma->vm_file); | |
845 | struct shmem_inode_info *info = SHMEM_I(inode); | |
846 | struct address_space *mapping = inode->i_mapping; | |
847 | unsigned long swapped; | |
848 | ||
849 | /* Be careful as we don't hold info->lock */ | |
850 | swapped = READ_ONCE(info->swapped); | |
851 | ||
852 | /* | |
853 | * The easier cases are when the shmem object has nothing in swap, or | |
854 | * the vma maps it whole. Then we can simply use the stats that we | |
855 | * already track. | |
856 | */ | |
857 | if (!swapped) | |
858 | return 0; | |
859 | ||
860 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
861 | return swapped << PAGE_SHIFT; | |
862 | ||
863 | /* Here comes the more involved part */ | |
864 | return shmem_partial_swap_usage(mapping, | |
865 | linear_page_index(vma, vma->vm_start), | |
866 | linear_page_index(vma, vma->vm_end)); | |
867 | } | |
868 | ||
24513264 HD |
869 | /* |
870 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
871 | */ | |
872 | void shmem_unlock_mapping(struct address_space *mapping) | |
873 | { | |
874 | struct pagevec pvec; | |
24513264 HD |
875 | pgoff_t index = 0; |
876 | ||
86679820 | 877 | pagevec_init(&pvec); |
24513264 HD |
878 | /* |
879 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
880 | */ | |
881 | while (!mapping_unevictable(mapping)) { | |
96888e0a | 882 | if (!pagevec_lookup(&pvec, mapping, &index)) |
24513264 | 883 | break; |
64e3d12f | 884 | check_move_unevictable_pages(&pvec); |
24513264 HD |
885 | pagevec_release(&pvec); |
886 | cond_resched(); | |
887 | } | |
7a5d0fbb HD |
888 | } |
889 | ||
71725ed1 HD |
890 | /* |
891 | * Check whether a hole-punch or truncation needs to split a huge page, | |
892 | * returning true if no split was required, or the split has been successful. | |
893 | * | |
894 | * Eviction (or truncation to 0 size) should never need to split a huge page; | |
895 | * but in rare cases might do so, if shmem_undo_range() failed to trylock on | |
896 | * head, and then succeeded to trylock on tail. | |
897 | * | |
898 | * A split can only succeed when there are no additional references on the | |
899 | * huge page: so the split below relies upon find_get_entries() having stopped | |
900 | * when it found a subpage of the huge page, without getting further references. | |
901 | */ | |
902 | static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end) | |
903 | { | |
904 | if (!PageTransCompound(page)) | |
905 | return true; | |
906 | ||
907 | /* Just proceed to delete a huge page wholly within the range punched */ | |
908 | if (PageHead(page) && | |
909 | page->index >= start && page->index + HPAGE_PMD_NR <= end) | |
910 | return true; | |
911 | ||
912 | /* Try to split huge page, so we can truly punch the hole or truncate */ | |
913 | return split_huge_page(page) >= 0; | |
914 | } | |
915 | ||
7a5d0fbb | 916 | /* |
7f4446ee | 917 | * Remove range of pages and swap entries from page cache, and free them. |
1635f6a7 | 918 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 919 | */ |
1635f6a7 HD |
920 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
921 | bool unfalloc) | |
1da177e4 | 922 | { |
285b2c4f | 923 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 924 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
925 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
926 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
927 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
928 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 929 | struct pagevec pvec; |
7a5d0fbb HD |
930 | pgoff_t indices[PAGEVEC_SIZE]; |
931 | long nr_swaps_freed = 0; | |
285b2c4f | 932 | pgoff_t index; |
bda97eab HD |
933 | int i; |
934 | ||
83e4fa9c HD |
935 | if (lend == -1) |
936 | end = -1; /* unsigned, so actually very big */ | |
bda97eab | 937 | |
d144bf62 HD |
938 | if (info->fallocend > start && info->fallocend <= end && !unfalloc) |
939 | info->fallocend = start; | |
940 | ||
86679820 | 941 | pagevec_init(&pvec); |
bda97eab | 942 | index = start; |
5c211ba2 MWO |
943 | while (index < end && find_lock_entries(mapping, index, end - 1, |
944 | &pvec, indices)) { | |
bda97eab HD |
945 | for (i = 0; i < pagevec_count(&pvec); i++) { |
946 | struct page *page = pvec.pages[i]; | |
947 | ||
7a5d0fbb | 948 | index = indices[i]; |
bda97eab | 949 | |
3159f943 | 950 | if (xa_is_value(page)) { |
1635f6a7 HD |
951 | if (unfalloc) |
952 | continue; | |
7a5d0fbb HD |
953 | nr_swaps_freed += !shmem_free_swap(mapping, |
954 | index, page); | |
bda97eab | 955 | continue; |
7a5d0fbb | 956 | } |
5c211ba2 | 957 | index += thp_nr_pages(page) - 1; |
7a5d0fbb | 958 | |
5c211ba2 MWO |
959 | if (!unfalloc || !PageUptodate(page)) |
960 | truncate_inode_page(mapping, page); | |
bda97eab HD |
961 | unlock_page(page); |
962 | } | |
0cd6144a | 963 | pagevec_remove_exceptionals(&pvec); |
24513264 | 964 | pagevec_release(&pvec); |
bda97eab HD |
965 | cond_resched(); |
966 | index++; | |
967 | } | |
1da177e4 | 968 | |
83e4fa9c | 969 | if (partial_start) { |
bda97eab | 970 | struct page *page = NULL; |
9e18eb29 | 971 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 972 | if (page) { |
09cbfeaf | 973 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
974 | if (start > end) { |
975 | top = partial_end; | |
976 | partial_end = 0; | |
977 | } | |
978 | zero_user_segment(page, partial_start, top); | |
979 | set_page_dirty(page); | |
980 | unlock_page(page); | |
09cbfeaf | 981 | put_page(page); |
83e4fa9c HD |
982 | } |
983 | } | |
984 | if (partial_end) { | |
985 | struct page *page = NULL; | |
9e18eb29 | 986 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
987 | if (page) { |
988 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
989 | set_page_dirty(page); |
990 | unlock_page(page); | |
09cbfeaf | 991 | put_page(page); |
bda97eab HD |
992 | } |
993 | } | |
83e4fa9c HD |
994 | if (start >= end) |
995 | return; | |
bda97eab HD |
996 | |
997 | index = start; | |
b1a36650 | 998 | while (index < end) { |
bda97eab | 999 | cond_resched(); |
0cd6144a | 1000 | |
cf2039af MWO |
1001 | if (!find_get_entries(mapping, index, end - 1, &pvec, |
1002 | indices)) { | |
b1a36650 HD |
1003 | /* If all gone or hole-punch or unfalloc, we're done */ |
1004 | if (index == start || end != -1) | |
bda97eab | 1005 | break; |
b1a36650 | 1006 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
1007 | index = start; |
1008 | continue; | |
1009 | } | |
bda97eab HD |
1010 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1011 | struct page *page = pvec.pages[i]; | |
1012 | ||
7a5d0fbb | 1013 | index = indices[i]; |
3159f943 | 1014 | if (xa_is_value(page)) { |
1635f6a7 HD |
1015 | if (unfalloc) |
1016 | continue; | |
b1a36650 HD |
1017 | if (shmem_free_swap(mapping, index, page)) { |
1018 | /* Swap was replaced by page: retry */ | |
1019 | index--; | |
1020 | break; | |
1021 | } | |
1022 | nr_swaps_freed++; | |
7a5d0fbb HD |
1023 | continue; |
1024 | } | |
1025 | ||
bda97eab | 1026 | lock_page(page); |
800d8c63 | 1027 | |
1635f6a7 | 1028 | if (!unfalloc || !PageUptodate(page)) { |
71725ed1 | 1029 | if (page_mapping(page) != mapping) { |
b1a36650 HD |
1030 | /* Page was replaced by swap: retry */ |
1031 | unlock_page(page); | |
1032 | index--; | |
1033 | break; | |
1635f6a7 | 1034 | } |
71725ed1 HD |
1035 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1036 | if (shmem_punch_compound(page, start, end)) | |
1037 | truncate_inode_page(mapping, page); | |
0783ac95 | 1038 | else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { |
71725ed1 HD |
1039 | /* Wipe the page and don't get stuck */ |
1040 | clear_highpage(page); | |
1041 | flush_dcache_page(page); | |
1042 | set_page_dirty(page); | |
1043 | if (index < | |
1044 | round_up(start, HPAGE_PMD_NR)) | |
1045 | start = index + 1; | |
1046 | } | |
7a5d0fbb | 1047 | } |
bda97eab HD |
1048 | unlock_page(page); |
1049 | } | |
0cd6144a | 1050 | pagevec_remove_exceptionals(&pvec); |
24513264 | 1051 | pagevec_release(&pvec); |
bda97eab HD |
1052 | index++; |
1053 | } | |
94c1e62d | 1054 | |
4595ef88 | 1055 | spin_lock_irq(&info->lock); |
7a5d0fbb | 1056 | info->swapped -= nr_swaps_freed; |
1da177e4 | 1057 | shmem_recalc_inode(inode); |
4595ef88 | 1058 | spin_unlock_irq(&info->lock); |
1635f6a7 | 1059 | } |
1da177e4 | 1060 | |
1635f6a7 HD |
1061 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
1062 | { | |
1063 | shmem_undo_range(inode, lstart, lend, false); | |
078cd827 | 1064 | inode->i_ctime = inode->i_mtime = current_time(inode); |
1da177e4 | 1065 | } |
94c1e62d | 1066 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 1067 | |
549c7297 CB |
1068 | static int shmem_getattr(struct user_namespace *mnt_userns, |
1069 | const struct path *path, struct kstat *stat, | |
a528d35e | 1070 | u32 request_mask, unsigned int query_flags) |
44a30220 | 1071 | { |
a528d35e | 1072 | struct inode *inode = path->dentry->d_inode; |
44a30220 YZ |
1073 | struct shmem_inode_info *info = SHMEM_I(inode); |
1074 | ||
d0424c42 | 1075 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
4595ef88 | 1076 | spin_lock_irq(&info->lock); |
d0424c42 | 1077 | shmem_recalc_inode(inode); |
4595ef88 | 1078 | spin_unlock_irq(&info->lock); |
d0424c42 | 1079 | } |
0d56a451 | 1080 | generic_fillattr(&init_user_ns, inode, stat); |
89fdcd26 | 1081 | |
a7fddc36 | 1082 | if (shmem_is_huge(NULL, inode, 0)) |
89fdcd26 YS |
1083 | stat->blksize = HPAGE_PMD_SIZE; |
1084 | ||
44a30220 YZ |
1085 | return 0; |
1086 | } | |
1087 | ||
549c7297 CB |
1088 | static int shmem_setattr(struct user_namespace *mnt_userns, |
1089 | struct dentry *dentry, struct iattr *attr) | |
1da177e4 | 1090 | { |
75c3cfa8 | 1091 | struct inode *inode = d_inode(dentry); |
40e041a2 | 1092 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 LT |
1093 | int error; |
1094 | ||
2f221d6f | 1095 | error = setattr_prepare(&init_user_ns, dentry, attr); |
db78b877 CH |
1096 | if (error) |
1097 | return error; | |
1098 | ||
94c1e62d HD |
1099 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
1100 | loff_t oldsize = inode->i_size; | |
1101 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 1102 | |
9608703e | 1103 | /* protected by i_rwsem */ |
40e041a2 DH |
1104 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || |
1105 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
1106 | return -EPERM; | |
1107 | ||
94c1e62d | 1108 | if (newsize != oldsize) { |
77142517 KK |
1109 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
1110 | oldsize, newsize); | |
1111 | if (error) | |
1112 | return error; | |
94c1e62d | 1113 | i_size_write(inode, newsize); |
078cd827 | 1114 | inode->i_ctime = inode->i_mtime = current_time(inode); |
94c1e62d | 1115 | } |
afa2db2f | 1116 | if (newsize <= oldsize) { |
94c1e62d | 1117 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
1118 | if (oldsize > holebegin) |
1119 | unmap_mapping_range(inode->i_mapping, | |
1120 | holebegin, 0, 1); | |
1121 | if (info->alloced) | |
1122 | shmem_truncate_range(inode, | |
1123 | newsize, (loff_t)-1); | |
94c1e62d | 1124 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
1125 | if (oldsize > holebegin) |
1126 | unmap_mapping_range(inode->i_mapping, | |
1127 | holebegin, 0, 1); | |
94c1e62d | 1128 | } |
1da177e4 LT |
1129 | } |
1130 | ||
2f221d6f | 1131 | setattr_copy(&init_user_ns, inode, attr); |
db78b877 | 1132 | if (attr->ia_valid & ATTR_MODE) |
e65ce2a5 | 1133 | error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode); |
1da177e4 LT |
1134 | return error; |
1135 | } | |
1136 | ||
1f895f75 | 1137 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1138 | { |
1da177e4 | 1139 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1140 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 | 1141 | |
30e6a51d | 1142 | if (shmem_mapping(inode->i_mapping)) { |
1da177e4 LT |
1143 | shmem_unacct_size(info->flags, inode->i_size); |
1144 | inode->i_size = 0; | |
3889e6e7 | 1145 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1146 | if (!list_empty(&info->shrinklist)) { |
1147 | spin_lock(&sbinfo->shrinklist_lock); | |
1148 | if (!list_empty(&info->shrinklist)) { | |
1149 | list_del_init(&info->shrinklist); | |
1150 | sbinfo->shrinklist_len--; | |
1151 | } | |
1152 | spin_unlock(&sbinfo->shrinklist_lock); | |
1153 | } | |
af53d3e9 HD |
1154 | while (!list_empty(&info->swaplist)) { |
1155 | /* Wait while shmem_unuse() is scanning this inode... */ | |
1156 | wait_var_event(&info->stop_eviction, | |
1157 | !atomic_read(&info->stop_eviction)); | |
cb5f7b9a | 1158 | mutex_lock(&shmem_swaplist_mutex); |
af53d3e9 HD |
1159 | /* ...but beware of the race if we peeked too early */ |
1160 | if (!atomic_read(&info->stop_eviction)) | |
1161 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1162 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1163 | } |
3ed47db3 | 1164 | } |
b09e0fa4 | 1165 | |
38f38657 | 1166 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 1167 | WARN_ON(inode->i_blocks); |
5b04c689 | 1168 | shmem_free_inode(inode->i_sb); |
dbd5768f | 1169 | clear_inode(inode); |
1da177e4 LT |
1170 | } |
1171 | ||
b56a2d8a VRP |
1172 | static int shmem_find_swap_entries(struct address_space *mapping, |
1173 | pgoff_t start, unsigned int nr_entries, | |
1174 | struct page **entries, pgoff_t *indices, | |
87039546 | 1175 | unsigned int type, bool frontswap) |
478922e2 | 1176 | { |
b56a2d8a VRP |
1177 | XA_STATE(xas, &mapping->i_pages, start); |
1178 | struct page *page; | |
87039546 | 1179 | swp_entry_t entry; |
b56a2d8a VRP |
1180 | unsigned int ret = 0; |
1181 | ||
1182 | if (!nr_entries) | |
1183 | return 0; | |
478922e2 MW |
1184 | |
1185 | rcu_read_lock(); | |
b56a2d8a VRP |
1186 | xas_for_each(&xas, page, ULONG_MAX) { |
1187 | if (xas_retry(&xas, page)) | |
5b9c98f3 | 1188 | continue; |
b56a2d8a VRP |
1189 | |
1190 | if (!xa_is_value(page)) | |
478922e2 | 1191 | continue; |
b56a2d8a | 1192 | |
87039546 HD |
1193 | entry = radix_to_swp_entry(page); |
1194 | if (swp_type(entry) != type) | |
1195 | continue; | |
1196 | if (frontswap && | |
1197 | !frontswap_test(swap_info[type], swp_offset(entry))) | |
1198 | continue; | |
b56a2d8a VRP |
1199 | |
1200 | indices[ret] = xas.xa_index; | |
1201 | entries[ret] = page; | |
1202 | ||
1203 | if (need_resched()) { | |
1204 | xas_pause(&xas); | |
1205 | cond_resched_rcu(); | |
1206 | } | |
1207 | if (++ret == nr_entries) | |
1208 | break; | |
478922e2 | 1209 | } |
478922e2 | 1210 | rcu_read_unlock(); |
e21a2955 | 1211 | |
b56a2d8a | 1212 | return ret; |
478922e2 MW |
1213 | } |
1214 | ||
46f65ec1 | 1215 | /* |
b56a2d8a VRP |
1216 | * Move the swapped pages for an inode to page cache. Returns the count |
1217 | * of pages swapped in, or the error in case of failure. | |
46f65ec1 | 1218 | */ |
b56a2d8a VRP |
1219 | static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec, |
1220 | pgoff_t *indices) | |
1da177e4 | 1221 | { |
b56a2d8a VRP |
1222 | int i = 0; |
1223 | int ret = 0; | |
bde05d1c | 1224 | int error = 0; |
b56a2d8a | 1225 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 1226 | |
b56a2d8a VRP |
1227 | for (i = 0; i < pvec.nr; i++) { |
1228 | struct page *page = pvec.pages[i]; | |
2e0e26c7 | 1229 | |
b56a2d8a VRP |
1230 | if (!xa_is_value(page)) |
1231 | continue; | |
1232 | error = shmem_swapin_page(inode, indices[i], | |
1233 | &page, SGP_CACHE, | |
1234 | mapping_gfp_mask(mapping), | |
1235 | NULL, NULL); | |
1236 | if (error == 0) { | |
1237 | unlock_page(page); | |
1238 | put_page(page); | |
1239 | ret++; | |
1240 | } | |
1241 | if (error == -ENOMEM) | |
1242 | break; | |
1243 | error = 0; | |
bde05d1c | 1244 | } |
b56a2d8a VRP |
1245 | return error ? error : ret; |
1246 | } | |
bde05d1c | 1247 | |
b56a2d8a VRP |
1248 | /* |
1249 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1250 | */ | |
1251 | static int shmem_unuse_inode(struct inode *inode, unsigned int type, | |
1252 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1253 | { | |
1254 | struct address_space *mapping = inode->i_mapping; | |
1255 | pgoff_t start = 0; | |
1256 | struct pagevec pvec; | |
1257 | pgoff_t indices[PAGEVEC_SIZE]; | |
1258 | bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0); | |
1259 | int ret = 0; | |
1260 | ||
1261 | pagevec_init(&pvec); | |
1262 | do { | |
1263 | unsigned int nr_entries = PAGEVEC_SIZE; | |
1264 | ||
1265 | if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE) | |
1266 | nr_entries = *fs_pages_to_unuse; | |
1267 | ||
1268 | pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries, | |
1269 | pvec.pages, indices, | |
87039546 | 1270 | type, frontswap); |
b56a2d8a VRP |
1271 | if (pvec.nr == 0) { |
1272 | ret = 0; | |
1273 | break; | |
46f65ec1 | 1274 | } |
b56a2d8a VRP |
1275 | |
1276 | ret = shmem_unuse_swap_entries(inode, pvec, indices); | |
1277 | if (ret < 0) | |
1278 | break; | |
1279 | ||
1280 | if (frontswap_partial) { | |
1281 | *fs_pages_to_unuse -= ret; | |
1282 | if (*fs_pages_to_unuse == 0) { | |
1283 | ret = FRONTSWAP_PAGES_UNUSED; | |
1284 | break; | |
1285 | } | |
1286 | } | |
1287 | ||
1288 | start = indices[pvec.nr - 1]; | |
1289 | } while (true); | |
1290 | ||
1291 | return ret; | |
1da177e4 LT |
1292 | } |
1293 | ||
1294 | /* | |
b56a2d8a VRP |
1295 | * Read all the shared memory data that resides in the swap |
1296 | * device 'type' back into memory, so the swap device can be | |
1297 | * unused. | |
1da177e4 | 1298 | */ |
b56a2d8a VRP |
1299 | int shmem_unuse(unsigned int type, bool frontswap, |
1300 | unsigned long *fs_pages_to_unuse) | |
1da177e4 | 1301 | { |
b56a2d8a | 1302 | struct shmem_inode_info *info, *next; |
bde05d1c HD |
1303 | int error = 0; |
1304 | ||
b56a2d8a VRP |
1305 | if (list_empty(&shmem_swaplist)) |
1306 | return 0; | |
1307 | ||
1308 | mutex_lock(&shmem_swaplist_mutex); | |
b56a2d8a VRP |
1309 | list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) { |
1310 | if (!info->swapped) { | |
6922c0c7 | 1311 | list_del_init(&info->swaplist); |
b56a2d8a VRP |
1312 | continue; |
1313 | } | |
af53d3e9 HD |
1314 | /* |
1315 | * Drop the swaplist mutex while searching the inode for swap; | |
1316 | * but before doing so, make sure shmem_evict_inode() will not | |
1317 | * remove placeholder inode from swaplist, nor let it be freed | |
1318 | * (igrab() would protect from unlink, but not from unmount). | |
1319 | */ | |
1320 | atomic_inc(&info->stop_eviction); | |
b56a2d8a | 1321 | mutex_unlock(&shmem_swaplist_mutex); |
b56a2d8a | 1322 | |
af53d3e9 | 1323 | error = shmem_unuse_inode(&info->vfs_inode, type, frontswap, |
b56a2d8a | 1324 | fs_pages_to_unuse); |
cb5f7b9a | 1325 | cond_resched(); |
b56a2d8a VRP |
1326 | |
1327 | mutex_lock(&shmem_swaplist_mutex); | |
1328 | next = list_next_entry(info, swaplist); | |
1329 | if (!info->swapped) | |
1330 | list_del_init(&info->swaplist); | |
af53d3e9 HD |
1331 | if (atomic_dec_and_test(&info->stop_eviction)) |
1332 | wake_up_var(&info->stop_eviction); | |
b56a2d8a | 1333 | if (error) |
778dd893 | 1334 | break; |
1da177e4 | 1335 | } |
cb5f7b9a | 1336 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1337 | |
778dd893 | 1338 | return error; |
1da177e4 LT |
1339 | } |
1340 | ||
1341 | /* | |
1342 | * Move the page from the page cache to the swap cache. | |
1343 | */ | |
1344 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1345 | { | |
1346 | struct shmem_inode_info *info; | |
1da177e4 | 1347 | struct address_space *mapping; |
1da177e4 | 1348 | struct inode *inode; |
6922c0c7 HD |
1349 | swp_entry_t swap; |
1350 | pgoff_t index; | |
1da177e4 | 1351 | |
1e6decf3 HD |
1352 | /* |
1353 | * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or | |
1354 | * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages, | |
1355 | * and its shmem_writeback() needs them to be split when swapping. | |
1356 | */ | |
1357 | if (PageTransCompound(page)) { | |
1358 | /* Ensure the subpages are still dirty */ | |
1359 | SetPageDirty(page); | |
1360 | if (split_huge_page(page) < 0) | |
1361 | goto redirty; | |
1362 | ClearPageDirty(page); | |
1363 | } | |
1364 | ||
1da177e4 | 1365 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1366 | mapping = page->mapping; |
1367 | index = page->index; | |
1368 | inode = mapping->host; | |
1369 | info = SHMEM_I(inode); | |
1370 | if (info->flags & VM_LOCKED) | |
1371 | goto redirty; | |
d9fe526a | 1372 | if (!total_swap_pages) |
1da177e4 LT |
1373 | goto redirty; |
1374 | ||
d9fe526a | 1375 | /* |
97b713ba CH |
1376 | * Our capabilities prevent regular writeback or sync from ever calling |
1377 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1378 | * its underlying filesystem, in which case tmpfs should write out to | |
1379 | * swap only in response to memory pressure, and not for the writeback | |
1380 | * threads or sync. | |
d9fe526a | 1381 | */ |
48f170fb HD |
1382 | if (!wbc->for_reclaim) { |
1383 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1384 | goto redirty; | |
1385 | } | |
1635f6a7 HD |
1386 | |
1387 | /* | |
1388 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1389 | * value into swapfile.c, the only way we can correctly account for a | |
1390 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1391 | * |
1392 | * That's okay for a page already fallocated earlier, but if we have | |
1393 | * not yet completed the fallocation, then (a) we want to keep track | |
1394 | * of this page in case we have to undo it, and (b) it may not be a | |
1395 | * good idea to continue anyway, once we're pushing into swap. So | |
1396 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1397 | */ |
1398 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1399 | if (inode->i_private) { |
1400 | struct shmem_falloc *shmem_falloc; | |
1401 | spin_lock(&inode->i_lock); | |
1402 | shmem_falloc = inode->i_private; | |
1403 | if (shmem_falloc && | |
8e205f77 | 1404 | !shmem_falloc->waitq && |
1aac1400 HD |
1405 | index >= shmem_falloc->start && |
1406 | index < shmem_falloc->next) | |
1407 | shmem_falloc->nr_unswapped++; | |
1408 | else | |
1409 | shmem_falloc = NULL; | |
1410 | spin_unlock(&inode->i_lock); | |
1411 | if (shmem_falloc) | |
1412 | goto redirty; | |
1413 | } | |
1635f6a7 HD |
1414 | clear_highpage(page); |
1415 | flush_dcache_page(page); | |
1416 | SetPageUptodate(page); | |
1417 | } | |
1418 | ||
38d8b4e6 | 1419 | swap = get_swap_page(page); |
48f170fb HD |
1420 | if (!swap.val) |
1421 | goto redirty; | |
d9fe526a | 1422 | |
b1dea800 HD |
1423 | /* |
1424 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1425 | * if it's not already there. Do it now before the page is |
1426 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1427 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1428 | * we've incremented swapped, because shmem_unuse_inode() will |
1429 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1430 | */ |
48f170fb HD |
1431 | mutex_lock(&shmem_swaplist_mutex); |
1432 | if (list_empty(&info->swaplist)) | |
b56a2d8a | 1433 | list_add(&info->swaplist, &shmem_swaplist); |
b1dea800 | 1434 | |
4afab1cd | 1435 | if (add_to_swap_cache(page, swap, |
3852f676 JK |
1436 | __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN, |
1437 | NULL) == 0) { | |
4595ef88 | 1438 | spin_lock_irq(&info->lock); |
6922c0c7 | 1439 | shmem_recalc_inode(inode); |
267a4c76 | 1440 | info->swapped++; |
4595ef88 | 1441 | spin_unlock_irq(&info->lock); |
6922c0c7 | 1442 | |
267a4c76 HD |
1443 | swap_shmem_alloc(swap); |
1444 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1445 | ||
6922c0c7 | 1446 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1447 | BUG_ON(page_mapped(page)); |
9fab5619 | 1448 | swap_writepage(page, wbc); |
1da177e4 LT |
1449 | return 0; |
1450 | } | |
1451 | ||
6922c0c7 | 1452 | mutex_unlock(&shmem_swaplist_mutex); |
75f6d6d2 | 1453 | put_swap_page(page, swap); |
1da177e4 LT |
1454 | redirty: |
1455 | set_page_dirty(page); | |
d9fe526a HD |
1456 | if (wbc->for_reclaim) |
1457 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1458 | unlock_page(page); | |
1459 | return 0; | |
1da177e4 LT |
1460 | } |
1461 | ||
75edd345 | 1462 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1463 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1464 | { |
095f1fc4 | 1465 | char buffer[64]; |
680d794b | 1466 | |
71fe804b | 1467 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1468 | return; /* show nothing */ |
680d794b | 1469 | |
a7a88b23 | 1470 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1471 | |
1472 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1473 | } |
71fe804b LS |
1474 | |
1475 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1476 | { | |
1477 | struct mempolicy *mpol = NULL; | |
1478 | if (sbinfo->mpol) { | |
bf11b9a8 | 1479 | raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ |
71fe804b LS |
1480 | mpol = sbinfo->mpol; |
1481 | mpol_get(mpol); | |
bf11b9a8 | 1482 | raw_spin_unlock(&sbinfo->stat_lock); |
71fe804b LS |
1483 | } |
1484 | return mpol; | |
1485 | } | |
75edd345 HD |
1486 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1487 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1488 | { | |
1489 | } | |
1490 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1491 | { | |
1492 | return NULL; | |
1493 | } | |
1494 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1495 | #ifndef CONFIG_NUMA | |
1496 | #define vm_policy vm_private_data | |
1497 | #endif | |
680d794b | 1498 | |
800d8c63 KS |
1499 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1500 | struct shmem_inode_info *info, pgoff_t index) | |
1501 | { | |
1502 | /* Create a pseudo vma that just contains the policy */ | |
2c4541e2 | 1503 | vma_init(vma, NULL); |
800d8c63 KS |
1504 | /* Bias interleave by inode number to distribute better across nodes */ |
1505 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
800d8c63 KS |
1506 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
1507 | } | |
1508 | ||
1509 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1510 | { | |
1511 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1512 | mpol_cond_put(vma->vm_policy); | |
1513 | } | |
1514 | ||
41ffe5d5 HD |
1515 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1516 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1517 | { |
1da177e4 | 1518 | struct vm_area_struct pvma; |
18a2f371 | 1519 | struct page *page; |
8c63ca5b WD |
1520 | struct vm_fault vmf = { |
1521 | .vma = &pvma, | |
1522 | }; | |
52cd3b07 | 1523 | |
800d8c63 | 1524 | shmem_pseudo_vma_init(&pvma, info, index); |
e9e9b7ec | 1525 | page = swap_cluster_readahead(swap, gfp, &vmf); |
800d8c63 | 1526 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1527 | |
800d8c63 KS |
1528 | return page; |
1529 | } | |
1530 | ||
78cc8cdc RR |
1531 | /* |
1532 | * Make sure huge_gfp is always more limited than limit_gfp. | |
1533 | * Some of the flags set permissions, while others set limitations. | |
1534 | */ | |
1535 | static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) | |
1536 | { | |
1537 | gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM; | |
1538 | gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY; | |
187df5dd RR |
1539 | gfp_t zoneflags = limit_gfp & GFP_ZONEMASK; |
1540 | gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK); | |
1541 | ||
1542 | /* Allow allocations only from the originally specified zones. */ | |
1543 | result |= zoneflags; | |
78cc8cdc RR |
1544 | |
1545 | /* | |
1546 | * Minimize the result gfp by taking the union with the deny flags, | |
1547 | * and the intersection of the allow flags. | |
1548 | */ | |
1549 | result |= (limit_gfp & denyflags); | |
1550 | result |= (huge_gfp & limit_gfp) & allowflags; | |
1551 | ||
1552 | return result; | |
1553 | } | |
1554 | ||
800d8c63 KS |
1555 | static struct page *shmem_alloc_hugepage(gfp_t gfp, |
1556 | struct shmem_inode_info *info, pgoff_t index) | |
1557 | { | |
1558 | struct vm_area_struct pvma; | |
7b8d046f MW |
1559 | struct address_space *mapping = info->vfs_inode.i_mapping; |
1560 | pgoff_t hindex; | |
800d8c63 KS |
1561 | struct page *page; |
1562 | ||
4620a06e | 1563 | hindex = round_down(index, HPAGE_PMD_NR); |
7b8d046f MW |
1564 | if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1, |
1565 | XA_PRESENT)) | |
800d8c63 | 1566 | return NULL; |
18a2f371 | 1567 | |
800d8c63 | 1568 | shmem_pseudo_vma_init(&pvma, info, hindex); |
164cc4fe RR |
1569 | page = alloc_pages_vma(gfp, HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), |
1570 | true); | |
800d8c63 KS |
1571 | shmem_pseudo_vma_destroy(&pvma); |
1572 | if (page) | |
1573 | prep_transhuge_page(page); | |
dcdf11ee DR |
1574 | else |
1575 | count_vm_event(THP_FILE_FALLBACK); | |
18a2f371 | 1576 | return page; |
1da177e4 LT |
1577 | } |
1578 | ||
02098fea | 1579 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1580 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1581 | { |
1582 | struct vm_area_struct pvma; | |
18a2f371 | 1583 | struct page *page; |
1da177e4 | 1584 | |
800d8c63 KS |
1585 | shmem_pseudo_vma_init(&pvma, info, index); |
1586 | page = alloc_page_vma(gfp, &pvma, 0); | |
1587 | shmem_pseudo_vma_destroy(&pvma); | |
1588 | ||
1589 | return page; | |
1590 | } | |
1591 | ||
1592 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
0f079694 | 1593 | struct inode *inode, |
800d8c63 KS |
1594 | pgoff_t index, bool huge) |
1595 | { | |
0f079694 | 1596 | struct shmem_inode_info *info = SHMEM_I(inode); |
800d8c63 KS |
1597 | struct page *page; |
1598 | int nr; | |
1599 | int err = -ENOSPC; | |
52cd3b07 | 1600 | |
396bcc52 | 1601 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) |
800d8c63 KS |
1602 | huge = false; |
1603 | nr = huge ? HPAGE_PMD_NR : 1; | |
1604 | ||
0f079694 | 1605 | if (!shmem_inode_acct_block(inode, nr)) |
800d8c63 | 1606 | goto failed; |
800d8c63 KS |
1607 | |
1608 | if (huge) | |
1609 | page = shmem_alloc_hugepage(gfp, info, index); | |
1610 | else | |
1611 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1612 | if (page) { |
1613 | __SetPageLocked(page); | |
1614 | __SetPageSwapBacked(page); | |
800d8c63 | 1615 | return page; |
75edd345 | 1616 | } |
18a2f371 | 1617 | |
800d8c63 | 1618 | err = -ENOMEM; |
0f079694 | 1619 | shmem_inode_unacct_blocks(inode, nr); |
800d8c63 KS |
1620 | failed: |
1621 | return ERR_PTR(err); | |
1da177e4 | 1622 | } |
71fe804b | 1623 | |
bde05d1c HD |
1624 | /* |
1625 | * When a page is moved from swapcache to shmem filecache (either by the | |
1626 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1627 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1628 | * ignorance of the mapping it belongs to. If that mapping has special | |
1629 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1630 | * we may need to copy to a suitable page before moving to filecache. | |
1631 | * | |
1632 | * In a future release, this may well be extended to respect cpuset and | |
1633 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1634 | * but for now it is a simple matter of zone. | |
1635 | */ | |
1636 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1637 | { | |
1638 | return page_zonenum(page) > gfp_zone(gfp); | |
1639 | } | |
1640 | ||
1641 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1642 | struct shmem_inode_info *info, pgoff_t index) | |
1643 | { | |
1644 | struct page *oldpage, *newpage; | |
1645 | struct address_space *swap_mapping; | |
c1cb20d4 | 1646 | swp_entry_t entry; |
bde05d1c HD |
1647 | pgoff_t swap_index; |
1648 | int error; | |
1649 | ||
1650 | oldpage = *pagep; | |
c1cb20d4 YZ |
1651 | entry.val = page_private(oldpage); |
1652 | swap_index = swp_offset(entry); | |
bde05d1c HD |
1653 | swap_mapping = page_mapping(oldpage); |
1654 | ||
1655 | /* | |
1656 | * We have arrived here because our zones are constrained, so don't | |
1657 | * limit chance of success by further cpuset and node constraints. | |
1658 | */ | |
1659 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1660 | newpage = shmem_alloc_page(gfp, info, index); | |
1661 | if (!newpage) | |
1662 | return -ENOMEM; | |
bde05d1c | 1663 | |
09cbfeaf | 1664 | get_page(newpage); |
bde05d1c | 1665 | copy_highpage(newpage, oldpage); |
0142ef6c | 1666 | flush_dcache_page(newpage); |
bde05d1c | 1667 | |
9956edf3 HD |
1668 | __SetPageLocked(newpage); |
1669 | __SetPageSwapBacked(newpage); | |
bde05d1c | 1670 | SetPageUptodate(newpage); |
c1cb20d4 | 1671 | set_page_private(newpage, entry.val); |
bde05d1c HD |
1672 | SetPageSwapCache(newpage); |
1673 | ||
1674 | /* | |
1675 | * Our caller will very soon move newpage out of swapcache, but it's | |
1676 | * a nice clean interface for us to replace oldpage by newpage there. | |
1677 | */ | |
b93b0163 | 1678 | xa_lock_irq(&swap_mapping->i_pages); |
62f945b6 | 1679 | error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage); |
0142ef6c | 1680 | if (!error) { |
0d1c2072 JW |
1681 | mem_cgroup_migrate(oldpage, newpage); |
1682 | __inc_lruvec_page_state(newpage, NR_FILE_PAGES); | |
1683 | __dec_lruvec_page_state(oldpage, NR_FILE_PAGES); | |
0142ef6c | 1684 | } |
b93b0163 | 1685 | xa_unlock_irq(&swap_mapping->i_pages); |
bde05d1c | 1686 | |
0142ef6c HD |
1687 | if (unlikely(error)) { |
1688 | /* | |
1689 | * Is this possible? I think not, now that our callers check | |
1690 | * both PageSwapCache and page_private after getting page lock; | |
1691 | * but be defensive. Reverse old to newpage for clear and free. | |
1692 | */ | |
1693 | oldpage = newpage; | |
1694 | } else { | |
6058eaec | 1695 | lru_cache_add(newpage); |
0142ef6c HD |
1696 | *pagep = newpage; |
1697 | } | |
bde05d1c HD |
1698 | |
1699 | ClearPageSwapCache(oldpage); | |
1700 | set_page_private(oldpage, 0); | |
1701 | ||
1702 | unlock_page(oldpage); | |
09cbfeaf KS |
1703 | put_page(oldpage); |
1704 | put_page(oldpage); | |
0142ef6c | 1705 | return error; |
bde05d1c HD |
1706 | } |
1707 | ||
c5bf121e VRP |
1708 | /* |
1709 | * Swap in the page pointed to by *pagep. | |
1710 | * Caller has to make sure that *pagep contains a valid swapped page. | |
1711 | * Returns 0 and the page in pagep if success. On failure, returns the | |
af44c12f | 1712 | * error code and NULL in *pagep. |
c5bf121e VRP |
1713 | */ |
1714 | static int shmem_swapin_page(struct inode *inode, pgoff_t index, | |
1715 | struct page **pagep, enum sgp_type sgp, | |
1716 | gfp_t gfp, struct vm_area_struct *vma, | |
1717 | vm_fault_t *fault_type) | |
1718 | { | |
1719 | struct address_space *mapping = inode->i_mapping; | |
1720 | struct shmem_inode_info *info = SHMEM_I(inode); | |
04f94e3f | 1721 | struct mm_struct *charge_mm = vma ? vma->vm_mm : NULL; |
b1e1ef34 | 1722 | struct page *page; |
c5bf121e VRP |
1723 | swp_entry_t swap; |
1724 | int error; | |
1725 | ||
1726 | VM_BUG_ON(!*pagep || !xa_is_value(*pagep)); | |
1727 | swap = radix_to_swp_entry(*pagep); | |
1728 | *pagep = NULL; | |
1729 | ||
1730 | /* Look it up and read it in.. */ | |
1731 | page = lookup_swap_cache(swap, NULL, 0); | |
1732 | if (!page) { | |
1733 | /* Or update major stats only when swapin succeeds?? */ | |
1734 | if (fault_type) { | |
1735 | *fault_type |= VM_FAULT_MAJOR; | |
1736 | count_vm_event(PGMAJFAULT); | |
1737 | count_memcg_event_mm(charge_mm, PGMAJFAULT); | |
1738 | } | |
1739 | /* Here we actually start the io */ | |
1740 | page = shmem_swapin(swap, gfp, info, index); | |
1741 | if (!page) { | |
1742 | error = -ENOMEM; | |
1743 | goto failed; | |
1744 | } | |
1745 | } | |
1746 | ||
1747 | /* We have to do this with page locked to prevent races */ | |
1748 | lock_page(page); | |
1749 | if (!PageSwapCache(page) || page_private(page) != swap.val || | |
1750 | !shmem_confirm_swap(mapping, index, swap)) { | |
1751 | error = -EEXIST; | |
1752 | goto unlock; | |
1753 | } | |
1754 | if (!PageUptodate(page)) { | |
1755 | error = -EIO; | |
1756 | goto failed; | |
1757 | } | |
1758 | wait_on_page_writeback(page); | |
1759 | ||
8a84802e SP |
1760 | /* |
1761 | * Some architectures may have to restore extra metadata to the | |
1762 | * physical page after reading from swap. | |
1763 | */ | |
1764 | arch_swap_restore(swap, page); | |
1765 | ||
c5bf121e VRP |
1766 | if (shmem_should_replace_page(page, gfp)) { |
1767 | error = shmem_replace_page(&page, gfp, info, index); | |
1768 | if (error) | |
1769 | goto failed; | |
1770 | } | |
1771 | ||
14235ab3 | 1772 | error = shmem_add_to_page_cache(page, mapping, index, |
3fea5a49 JW |
1773 | swp_to_radix_entry(swap), gfp, |
1774 | charge_mm); | |
1775 | if (error) | |
14235ab3 | 1776 | goto failed; |
c5bf121e VRP |
1777 | |
1778 | spin_lock_irq(&info->lock); | |
1779 | info->swapped--; | |
1780 | shmem_recalc_inode(inode); | |
1781 | spin_unlock_irq(&info->lock); | |
1782 | ||
1783 | if (sgp == SGP_WRITE) | |
1784 | mark_page_accessed(page); | |
1785 | ||
1786 | delete_from_swap_cache(page); | |
1787 | set_page_dirty(page); | |
1788 | swap_free(swap); | |
1789 | ||
1790 | *pagep = page; | |
1791 | return 0; | |
1792 | failed: | |
1793 | if (!shmem_confirm_swap(mapping, index, swap)) | |
1794 | error = -EEXIST; | |
1795 | unlock: | |
1796 | if (page) { | |
1797 | unlock_page(page); | |
1798 | put_page(page); | |
1799 | } | |
1800 | ||
1801 | return error; | |
1802 | } | |
1803 | ||
1da177e4 | 1804 | /* |
68da9f05 | 1805 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1806 | * |
1807 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1808 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1809 | * entry since a page cannot live in both the swap and page cache. |
1810 | * | |
c949b097 | 1811 | * vma, vmf, and fault_type are only supplied by shmem_fault: |
9e18eb29 | 1812 | * otherwise they are NULL. |
1da177e4 | 1813 | */ |
41ffe5d5 | 1814 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 1815 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
2b740303 SJ |
1816 | struct vm_area_struct *vma, struct vm_fault *vmf, |
1817 | vm_fault_t *fault_type) | |
1da177e4 LT |
1818 | { |
1819 | struct address_space *mapping = inode->i_mapping; | |
23f919d4 | 1820 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 | 1821 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1822 | struct mm_struct *charge_mm; |
27ab7006 | 1823 | struct page *page; |
800d8c63 | 1824 | pgoff_t hindex = index; |
164cc4fe | 1825 | gfp_t huge_gfp; |
1da177e4 | 1826 | int error; |
54af6042 | 1827 | int once = 0; |
1635f6a7 | 1828 | int alloced = 0; |
1da177e4 | 1829 | |
09cbfeaf | 1830 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1831 | return -EFBIG; |
1da177e4 | 1832 | repeat: |
c5bf121e VRP |
1833 | if (sgp <= SGP_CACHE && |
1834 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { | |
1835 | return -EINVAL; | |
1836 | } | |
1837 | ||
1838 | sbinfo = SHMEM_SB(inode->i_sb); | |
04f94e3f | 1839 | charge_mm = vma ? vma->vm_mm : NULL; |
c5bf121e | 1840 | |
44835d20 MWO |
1841 | page = pagecache_get_page(mapping, index, |
1842 | FGP_ENTRY | FGP_HEAD | FGP_LOCK, 0); | |
c949b097 AR |
1843 | |
1844 | if (page && vma && userfaultfd_minor(vma)) { | |
1845 | if (!xa_is_value(page)) { | |
1846 | unlock_page(page); | |
1847 | put_page(page); | |
1848 | } | |
1849 | *fault_type = handle_userfault(vmf, VM_UFFD_MINOR); | |
1850 | return 0; | |
1851 | } | |
1852 | ||
3159f943 | 1853 | if (xa_is_value(page)) { |
c5bf121e VRP |
1854 | error = shmem_swapin_page(inode, index, &page, |
1855 | sgp, gfp, vma, fault_type); | |
1856 | if (error == -EEXIST) | |
1857 | goto repeat; | |
54af6042 | 1858 | |
c5bf121e VRP |
1859 | *pagep = page; |
1860 | return error; | |
54af6042 HD |
1861 | } |
1862 | ||
acdd9f8e | 1863 | if (page) { |
63ec1973 | 1864 | hindex = page->index; |
acdd9f8e HD |
1865 | if (sgp == SGP_WRITE) |
1866 | mark_page_accessed(page); | |
1867 | if (PageUptodate(page)) | |
1868 | goto out; | |
1869 | /* fallocated page */ | |
1635f6a7 HD |
1870 | if (sgp != SGP_READ) |
1871 | goto clear; | |
1872 | unlock_page(page); | |
09cbfeaf | 1873 | put_page(page); |
1635f6a7 | 1874 | } |
27ab7006 HD |
1875 | |
1876 | /* | |
acdd9f8e HD |
1877 | * SGP_READ: succeed on hole, with NULL page, letting caller zero. |
1878 | * SGP_NOALLOC: fail on hole, with NULL page, letting caller fail. | |
1879 | */ | |
1880 | *pagep = NULL; | |
1881 | if (sgp == SGP_READ) | |
1882 | return 0; | |
1883 | if (sgp == SGP_NOALLOC) | |
1884 | return -ENOENT; | |
1885 | ||
1886 | /* | |
1887 | * Fast cache lookup and swap lookup did not find it: allocate. | |
27ab7006 | 1888 | */ |
54af6042 | 1889 | |
c5bf121e VRP |
1890 | if (vma && userfaultfd_missing(vma)) { |
1891 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
1892 | return 0; | |
1893 | } | |
cfda0526 | 1894 | |
5e6e5a12 HD |
1895 | /* Never use a huge page for shmem_symlink() */ |
1896 | if (S_ISLNK(inode->i_mode)) | |
c5bf121e | 1897 | goto alloc_nohuge; |
5e6e5a12 | 1898 | if (!shmem_is_huge(vma, inode, index)) |
c5bf121e | 1899 | goto alloc_nohuge; |
1da177e4 | 1900 | |
164cc4fe | 1901 | huge_gfp = vma_thp_gfp_mask(vma); |
78cc8cdc | 1902 | huge_gfp = limit_gfp_mask(huge_gfp, gfp); |
164cc4fe | 1903 | page = shmem_alloc_and_acct_page(huge_gfp, inode, index, true); |
c5bf121e VRP |
1904 | if (IS_ERR(page)) { |
1905 | alloc_nohuge: | |
1906 | page = shmem_alloc_and_acct_page(gfp, inode, | |
1907 | index, false); | |
1908 | } | |
1909 | if (IS_ERR(page)) { | |
1910 | int retry = 5; | |
800d8c63 | 1911 | |
c5bf121e VRP |
1912 | error = PTR_ERR(page); |
1913 | page = NULL; | |
1914 | if (error != -ENOSPC) | |
1915 | goto unlock; | |
1916 | /* | |
1917 | * Try to reclaim some space by splitting a huge page | |
1918 | * beyond i_size on the filesystem. | |
1919 | */ | |
1920 | while (retry--) { | |
1921 | int ret; | |
66d2f4d2 | 1922 | |
c5bf121e VRP |
1923 | ret = shmem_unused_huge_shrink(sbinfo, NULL, 1); |
1924 | if (ret == SHRINK_STOP) | |
1925 | break; | |
1926 | if (ret) | |
1927 | goto alloc_nohuge; | |
b065b432 | 1928 | } |
c5bf121e VRP |
1929 | goto unlock; |
1930 | } | |
54af6042 | 1931 | |
c5bf121e VRP |
1932 | if (PageTransHuge(page)) |
1933 | hindex = round_down(index, HPAGE_PMD_NR); | |
1934 | else | |
1935 | hindex = index; | |
54af6042 | 1936 | |
c5bf121e VRP |
1937 | if (sgp == SGP_WRITE) |
1938 | __SetPageReferenced(page); | |
1939 | ||
c5bf121e | 1940 | error = shmem_add_to_page_cache(page, mapping, hindex, |
3fea5a49 JW |
1941 | NULL, gfp & GFP_RECLAIM_MASK, |
1942 | charge_mm); | |
1943 | if (error) | |
c5bf121e | 1944 | goto unacct; |
6058eaec | 1945 | lru_cache_add(page); |
779750d2 | 1946 | |
c5bf121e | 1947 | spin_lock_irq(&info->lock); |
d8c6546b | 1948 | info->alloced += compound_nr(page); |
c5bf121e VRP |
1949 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); |
1950 | shmem_recalc_inode(inode); | |
1951 | spin_unlock_irq(&info->lock); | |
1952 | alloced = true; | |
1953 | ||
1954 | if (PageTransHuge(page) && | |
1955 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < | |
1956 | hindex + HPAGE_PMD_NR - 1) { | |
ec9516fb | 1957 | /* |
c5bf121e VRP |
1958 | * Part of the huge page is beyond i_size: subject |
1959 | * to shrink under memory pressure. | |
1635f6a7 | 1960 | */ |
c5bf121e | 1961 | spin_lock(&sbinfo->shrinklist_lock); |
1635f6a7 | 1962 | /* |
c5bf121e VRP |
1963 | * _careful to defend against unlocked access to |
1964 | * ->shrink_list in shmem_unused_huge_shrink() | |
ec9516fb | 1965 | */ |
c5bf121e VRP |
1966 | if (list_empty_careful(&info->shrinklist)) { |
1967 | list_add_tail(&info->shrinklist, | |
1968 | &sbinfo->shrinklist); | |
1969 | sbinfo->shrinklist_len++; | |
1970 | } | |
1971 | spin_unlock(&sbinfo->shrinklist_lock); | |
1972 | } | |
800d8c63 | 1973 | |
c5bf121e VRP |
1974 | /* |
1975 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. | |
1976 | */ | |
1977 | if (sgp == SGP_FALLOC) | |
1978 | sgp = SGP_WRITE; | |
1979 | clear: | |
1980 | /* | |
1981 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1982 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1983 | * it now, lest undo on failure cancel our earlier guarantee. | |
1984 | */ | |
1985 | if (sgp != SGP_WRITE && !PageUptodate(page)) { | |
c5bf121e VRP |
1986 | int i; |
1987 | ||
63ec1973 MWO |
1988 | for (i = 0; i < compound_nr(page); i++) { |
1989 | clear_highpage(page + i); | |
1990 | flush_dcache_page(page + i); | |
ec9516fb | 1991 | } |
63ec1973 | 1992 | SetPageUptodate(page); |
1da177e4 | 1993 | } |
bde05d1c | 1994 | |
54af6042 | 1995 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1996 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1997 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1998 | if (alloced) { |
1999 | ClearPageDirty(page); | |
2000 | delete_from_page_cache(page); | |
4595ef88 | 2001 | spin_lock_irq(&info->lock); |
267a4c76 | 2002 | shmem_recalc_inode(inode); |
4595ef88 | 2003 | spin_unlock_irq(&info->lock); |
267a4c76 | 2004 | } |
54af6042 | 2005 | error = -EINVAL; |
267a4c76 | 2006 | goto unlock; |
e83c32e8 | 2007 | } |
63ec1973 | 2008 | out: |
800d8c63 | 2009 | *pagep = page + index - hindex; |
54af6042 | 2010 | return 0; |
1da177e4 | 2011 | |
59a16ead | 2012 | /* |
54af6042 | 2013 | * Error recovery. |
59a16ead | 2014 | */ |
54af6042 | 2015 | unacct: |
d8c6546b | 2016 | shmem_inode_unacct_blocks(inode, compound_nr(page)); |
800d8c63 KS |
2017 | |
2018 | if (PageTransHuge(page)) { | |
2019 | unlock_page(page); | |
2020 | put_page(page); | |
2021 | goto alloc_nohuge; | |
2022 | } | |
d1899228 | 2023 | unlock: |
27ab7006 | 2024 | if (page) { |
54af6042 | 2025 | unlock_page(page); |
09cbfeaf | 2026 | put_page(page); |
54af6042 HD |
2027 | } |
2028 | if (error == -ENOSPC && !once++) { | |
4595ef88 | 2029 | spin_lock_irq(&info->lock); |
54af6042 | 2030 | shmem_recalc_inode(inode); |
4595ef88 | 2031 | spin_unlock_irq(&info->lock); |
27ab7006 | 2032 | goto repeat; |
ff36b801 | 2033 | } |
7f4446ee | 2034 | if (error == -EEXIST) |
54af6042 HD |
2035 | goto repeat; |
2036 | return error; | |
1da177e4 LT |
2037 | } |
2038 | ||
10d20bd2 LT |
2039 | /* |
2040 | * This is like autoremove_wake_function, but it removes the wait queue | |
2041 | * entry unconditionally - even if something else had already woken the | |
2042 | * target. | |
2043 | */ | |
ac6424b9 | 2044 | static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
10d20bd2 LT |
2045 | { |
2046 | int ret = default_wake_function(wait, mode, sync, key); | |
2055da97 | 2047 | list_del_init(&wait->entry); |
10d20bd2 LT |
2048 | return ret; |
2049 | } | |
2050 | ||
20acce67 | 2051 | static vm_fault_t shmem_fault(struct vm_fault *vmf) |
1da177e4 | 2052 | { |
11bac800 | 2053 | struct vm_area_struct *vma = vmf->vma; |
496ad9aa | 2054 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 2055 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
20acce67 SJ |
2056 | int err; |
2057 | vm_fault_t ret = VM_FAULT_LOCKED; | |
1da177e4 | 2058 | |
f00cdc6d HD |
2059 | /* |
2060 | * Trinity finds that probing a hole which tmpfs is punching can | |
2061 | * prevent the hole-punch from ever completing: which in turn | |
9608703e | 2062 | * locks writers out with its hold on i_rwsem. So refrain from |
8e205f77 HD |
2063 | * faulting pages into the hole while it's being punched. Although |
2064 | * shmem_undo_range() does remove the additions, it may be unable to | |
2065 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
2066 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
2067 | * | |
2068 | * It does not matter if we sometimes reach this check just before the | |
2069 | * hole-punch begins, so that one fault then races with the punch: | |
2070 | * we just need to make racing faults a rare case. | |
2071 | * | |
2072 | * The implementation below would be much simpler if we just used a | |
9608703e | 2073 | * standard mutex or completion: but we cannot take i_rwsem in fault, |
8e205f77 | 2074 | * and bloating every shmem inode for this unlikely case would be sad. |
f00cdc6d HD |
2075 | */ |
2076 | if (unlikely(inode->i_private)) { | |
2077 | struct shmem_falloc *shmem_falloc; | |
2078 | ||
2079 | spin_lock(&inode->i_lock); | |
2080 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
2081 | if (shmem_falloc && |
2082 | shmem_falloc->waitq && | |
2083 | vmf->pgoff >= shmem_falloc->start && | |
2084 | vmf->pgoff < shmem_falloc->next) { | |
8897c1b1 | 2085 | struct file *fpin; |
8e205f77 | 2086 | wait_queue_head_t *shmem_falloc_waitq; |
10d20bd2 | 2087 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); |
8e205f77 HD |
2088 | |
2089 | ret = VM_FAULT_NOPAGE; | |
8897c1b1 KS |
2090 | fpin = maybe_unlock_mmap_for_io(vmf, NULL); |
2091 | if (fpin) | |
8e205f77 | 2092 | ret = VM_FAULT_RETRY; |
8e205f77 HD |
2093 | |
2094 | shmem_falloc_waitq = shmem_falloc->waitq; | |
2095 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
2096 | TASK_UNINTERRUPTIBLE); | |
2097 | spin_unlock(&inode->i_lock); | |
2098 | schedule(); | |
2099 | ||
2100 | /* | |
2101 | * shmem_falloc_waitq points into the shmem_fallocate() | |
2102 | * stack of the hole-punching task: shmem_falloc_waitq | |
2103 | * is usually invalid by the time we reach here, but | |
2104 | * finish_wait() does not dereference it in that case; | |
2105 | * though i_lock needed lest racing with wake_up_all(). | |
2106 | */ | |
2107 | spin_lock(&inode->i_lock); | |
2108 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
2109 | spin_unlock(&inode->i_lock); | |
8897c1b1 KS |
2110 | |
2111 | if (fpin) | |
2112 | fput(fpin); | |
8e205f77 | 2113 | return ret; |
f00cdc6d | 2114 | } |
8e205f77 | 2115 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
2116 | } |
2117 | ||
5e6e5a12 | 2118 | err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, SGP_CACHE, |
cfda0526 | 2119 | gfp, vma, vmf, &ret); |
20acce67 SJ |
2120 | if (err) |
2121 | return vmf_error(err); | |
68da9f05 | 2122 | return ret; |
1da177e4 LT |
2123 | } |
2124 | ||
c01d5b30 HD |
2125 | unsigned long shmem_get_unmapped_area(struct file *file, |
2126 | unsigned long uaddr, unsigned long len, | |
2127 | unsigned long pgoff, unsigned long flags) | |
2128 | { | |
2129 | unsigned long (*get_area)(struct file *, | |
2130 | unsigned long, unsigned long, unsigned long, unsigned long); | |
2131 | unsigned long addr; | |
2132 | unsigned long offset; | |
2133 | unsigned long inflated_len; | |
2134 | unsigned long inflated_addr; | |
2135 | unsigned long inflated_offset; | |
2136 | ||
2137 | if (len > TASK_SIZE) | |
2138 | return -ENOMEM; | |
2139 | ||
2140 | get_area = current->mm->get_unmapped_area; | |
2141 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2142 | ||
396bcc52 | 2143 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) |
c01d5b30 HD |
2144 | return addr; |
2145 | if (IS_ERR_VALUE(addr)) | |
2146 | return addr; | |
2147 | if (addr & ~PAGE_MASK) | |
2148 | return addr; | |
2149 | if (addr > TASK_SIZE - len) | |
2150 | return addr; | |
2151 | ||
2152 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2153 | return addr; | |
2154 | if (len < HPAGE_PMD_SIZE) | |
2155 | return addr; | |
2156 | if (flags & MAP_FIXED) | |
2157 | return addr; | |
2158 | /* | |
2159 | * Our priority is to support MAP_SHARED mapped hugely; | |
2160 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
99158997 KS |
2161 | * But if caller specified an address hint and we allocated area there |
2162 | * successfully, respect that as before. | |
c01d5b30 | 2163 | */ |
99158997 | 2164 | if (uaddr == addr) |
c01d5b30 HD |
2165 | return addr; |
2166 | ||
2167 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2168 | struct super_block *sb; | |
2169 | ||
2170 | if (file) { | |
2171 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2172 | sb = file_inode(file)->i_sb; | |
2173 | } else { | |
2174 | /* | |
2175 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2176 | * for "/dev/zero", to create a shared anonymous object. | |
2177 | */ | |
2178 | if (IS_ERR(shm_mnt)) | |
2179 | return addr; | |
2180 | sb = shm_mnt->mnt_sb; | |
2181 | } | |
3089bf61 | 2182 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2183 | return addr; |
2184 | } | |
2185 | ||
2186 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2187 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2188 | return addr; | |
2189 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2190 | return addr; | |
2191 | ||
2192 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2193 | if (inflated_len > TASK_SIZE) | |
2194 | return addr; | |
2195 | if (inflated_len < len) | |
2196 | return addr; | |
2197 | ||
99158997 | 2198 | inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags); |
c01d5b30 HD |
2199 | if (IS_ERR_VALUE(inflated_addr)) |
2200 | return addr; | |
2201 | if (inflated_addr & ~PAGE_MASK) | |
2202 | return addr; | |
2203 | ||
2204 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2205 | inflated_addr += offset - inflated_offset; | |
2206 | if (inflated_offset > offset) | |
2207 | inflated_addr += HPAGE_PMD_SIZE; | |
2208 | ||
2209 | if (inflated_addr > TASK_SIZE - len) | |
2210 | return addr; | |
2211 | return inflated_addr; | |
2212 | } | |
2213 | ||
1da177e4 | 2214 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2215 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2216 | { |
496ad9aa | 2217 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2218 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2219 | } |
2220 | ||
d8dc74f2 AB |
2221 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
2222 | unsigned long addr) | |
1da177e4 | 2223 | { |
496ad9aa | 2224 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2225 | pgoff_t index; |
1da177e4 | 2226 | |
41ffe5d5 HD |
2227 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2228 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
2229 | } |
2230 | #endif | |
2231 | ||
d7c9e99a | 2232 | int shmem_lock(struct file *file, int lock, struct ucounts *ucounts) |
1da177e4 | 2233 | { |
496ad9aa | 2234 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2235 | struct shmem_inode_info *info = SHMEM_I(inode); |
2236 | int retval = -ENOMEM; | |
2237 | ||
ea0dfeb4 HD |
2238 | /* |
2239 | * What serializes the accesses to info->flags? | |
2240 | * ipc_lock_object() when called from shmctl_do_lock(), | |
2241 | * no serialization needed when called from shm_destroy(). | |
2242 | */ | |
1da177e4 | 2243 | if (lock && !(info->flags & VM_LOCKED)) { |
d7c9e99a | 2244 | if (!user_shm_lock(inode->i_size, ucounts)) |
1da177e4 LT |
2245 | goto out_nomem; |
2246 | info->flags |= VM_LOCKED; | |
89e004ea | 2247 | mapping_set_unevictable(file->f_mapping); |
1da177e4 | 2248 | } |
d7c9e99a AG |
2249 | if (!lock && (info->flags & VM_LOCKED) && ucounts) { |
2250 | user_shm_unlock(inode->i_size, ucounts); | |
1da177e4 | 2251 | info->flags &= ~VM_LOCKED; |
89e004ea | 2252 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2253 | } |
2254 | retval = 0; | |
89e004ea | 2255 | |
1da177e4 | 2256 | out_nomem: |
1da177e4 LT |
2257 | return retval; |
2258 | } | |
2259 | ||
9b83a6a8 | 2260 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 | 2261 | { |
ab3948f5 | 2262 | struct shmem_inode_info *info = SHMEM_I(file_inode(file)); |
22247efd | 2263 | int ret; |
ab3948f5 | 2264 | |
22247efd PX |
2265 | ret = seal_check_future_write(info->seals, vma); |
2266 | if (ret) | |
2267 | return ret; | |
ab3948f5 | 2268 | |
51b0bff2 CM |
2269 | /* arm64 - allow memory tagging on RAM-based files */ |
2270 | vma->vm_flags |= VM_MTE_ALLOWED; | |
2271 | ||
1da177e4 LT |
2272 | file_accessed(file); |
2273 | vma->vm_ops = &shmem_vm_ops; | |
396bcc52 | 2274 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && |
f3f0e1d2 KS |
2275 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
2276 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
2277 | khugepaged_enter(vma, vma->vm_flags); | |
2278 | } | |
1da177e4 LT |
2279 | return 0; |
2280 | } | |
2281 | ||
454abafe | 2282 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 2283 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
2284 | { |
2285 | struct inode *inode; | |
2286 | struct shmem_inode_info *info; | |
2287 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
e809d5f0 | 2288 | ino_t ino; |
1da177e4 | 2289 | |
e809d5f0 | 2290 | if (shmem_reserve_inode(sb, &ino)) |
5b04c689 | 2291 | return NULL; |
1da177e4 LT |
2292 | |
2293 | inode = new_inode(sb); | |
2294 | if (inode) { | |
e809d5f0 | 2295 | inode->i_ino = ino; |
21cb47be | 2296 | inode_init_owner(&init_user_ns, inode, dir, mode); |
1da177e4 | 2297 | inode->i_blocks = 0; |
078cd827 | 2298 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
46c9a946 | 2299 | inode->i_generation = prandom_u32(); |
1da177e4 LT |
2300 | info = SHMEM_I(inode); |
2301 | memset(info, 0, (char *)inode - (char *)info); | |
2302 | spin_lock_init(&info->lock); | |
af53d3e9 | 2303 | atomic_set(&info->stop_eviction, 0); |
40e041a2 | 2304 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 2305 | info->flags = flags & VM_NORESERVE; |
779750d2 | 2306 | INIT_LIST_HEAD(&info->shrinklist); |
1da177e4 | 2307 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 2308 | simple_xattrs_init(&info->xattrs); |
72c04902 | 2309 | cache_no_acl(inode); |
1da177e4 LT |
2310 | |
2311 | switch (mode & S_IFMT) { | |
2312 | default: | |
39f0247d | 2313 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
2314 | init_special_inode(inode, mode, dev); |
2315 | break; | |
2316 | case S_IFREG: | |
14fcc23f | 2317 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
2318 | inode->i_op = &shmem_inode_operations; |
2319 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
2320 | mpol_shared_policy_init(&info->policy, |
2321 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
2322 | break; |
2323 | case S_IFDIR: | |
d8c76e6f | 2324 | inc_nlink(inode); |
1da177e4 LT |
2325 | /* Some things misbehave if size == 0 on a directory */ |
2326 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2327 | inode->i_op = &shmem_dir_inode_operations; | |
2328 | inode->i_fop = &simple_dir_operations; | |
2329 | break; | |
2330 | case S_IFLNK: | |
2331 | /* | |
2332 | * Must not load anything in the rbtree, | |
2333 | * mpol_free_shared_policy will not be called. | |
2334 | */ | |
71fe804b | 2335 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
2336 | break; |
2337 | } | |
b45d71fb JFG |
2338 | |
2339 | lockdep_annotate_inode_mutex_key(inode); | |
5b04c689 PE |
2340 | } else |
2341 | shmem_free_inode(sb); | |
1da177e4 LT |
2342 | return inode; |
2343 | } | |
2344 | ||
3460f6e5 AR |
2345 | #ifdef CONFIG_USERFAULTFD |
2346 | int shmem_mfill_atomic_pte(struct mm_struct *dst_mm, | |
2347 | pmd_t *dst_pmd, | |
2348 | struct vm_area_struct *dst_vma, | |
2349 | unsigned long dst_addr, | |
2350 | unsigned long src_addr, | |
2351 | bool zeropage, | |
2352 | struct page **pagep) | |
4c27fe4c MR |
2353 | { |
2354 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2355 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4c27fe4c MR |
2356 | struct address_space *mapping = inode->i_mapping; |
2357 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2358 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
4c27fe4c MR |
2359 | void *page_kaddr; |
2360 | struct page *page; | |
4c27fe4c | 2361 | int ret; |
3460f6e5 | 2362 | pgoff_t max_off; |
4c27fe4c | 2363 | |
7ed9d238 AR |
2364 | if (!shmem_inode_acct_block(inode, 1)) { |
2365 | /* | |
2366 | * We may have got a page, returned -ENOENT triggering a retry, | |
2367 | * and now we find ourselves with -ENOMEM. Release the page, to | |
2368 | * avoid a BUG_ON in our caller. | |
2369 | */ | |
2370 | if (unlikely(*pagep)) { | |
2371 | put_page(*pagep); | |
2372 | *pagep = NULL; | |
2373 | } | |
7d64ae3a | 2374 | return -ENOMEM; |
7ed9d238 | 2375 | } |
4c27fe4c | 2376 | |
cb658a45 | 2377 | if (!*pagep) { |
7d64ae3a | 2378 | ret = -ENOMEM; |
4c27fe4c MR |
2379 | page = shmem_alloc_page(gfp, info, pgoff); |
2380 | if (!page) | |
0f079694 | 2381 | goto out_unacct_blocks; |
4c27fe4c | 2382 | |
3460f6e5 | 2383 | if (!zeropage) { /* COPY */ |
8d103963 MR |
2384 | page_kaddr = kmap_atomic(page); |
2385 | ret = copy_from_user(page_kaddr, | |
2386 | (const void __user *)src_addr, | |
2387 | PAGE_SIZE); | |
2388 | kunmap_atomic(page_kaddr); | |
2389 | ||
c1e8d7c6 | 2390 | /* fallback to copy_from_user outside mmap_lock */ |
8d103963 MR |
2391 | if (unlikely(ret)) { |
2392 | *pagep = page; | |
7d64ae3a | 2393 | ret = -ENOENT; |
8d103963 | 2394 | /* don't free the page */ |
7d64ae3a | 2395 | goto out_unacct_blocks; |
8d103963 | 2396 | } |
3460f6e5 | 2397 | } else { /* ZEROPAGE */ |
8d103963 | 2398 | clear_highpage(page); |
4c27fe4c MR |
2399 | } |
2400 | } else { | |
2401 | page = *pagep; | |
2402 | *pagep = NULL; | |
2403 | } | |
2404 | ||
3460f6e5 AR |
2405 | VM_BUG_ON(PageLocked(page)); |
2406 | VM_BUG_ON(PageSwapBacked(page)); | |
9cc90c66 AA |
2407 | __SetPageLocked(page); |
2408 | __SetPageSwapBacked(page); | |
a425d358 | 2409 | __SetPageUptodate(page); |
9cc90c66 | 2410 | |
e2a50c1f | 2411 | ret = -EFAULT; |
e2a50c1f | 2412 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3460f6e5 | 2413 | if (unlikely(pgoff >= max_off)) |
e2a50c1f AA |
2414 | goto out_release; |
2415 | ||
552446a4 | 2416 | ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL, |
3fea5a49 | 2417 | gfp & GFP_RECLAIM_MASK, dst_mm); |
4c27fe4c | 2418 | if (ret) |
3fea5a49 | 2419 | goto out_release; |
4c27fe4c | 2420 | |
7d64ae3a AR |
2421 | ret = mfill_atomic_install_pte(dst_mm, dst_pmd, dst_vma, dst_addr, |
2422 | page, true, false); | |
2423 | if (ret) | |
2424 | goto out_delete_from_cache; | |
4c27fe4c | 2425 | |
94b7cc01 | 2426 | spin_lock_irq(&info->lock); |
4c27fe4c MR |
2427 | info->alloced++; |
2428 | inode->i_blocks += BLOCKS_PER_PAGE; | |
2429 | shmem_recalc_inode(inode); | |
94b7cc01 | 2430 | spin_unlock_irq(&info->lock); |
4c27fe4c | 2431 | |
7d64ae3a | 2432 | SetPageDirty(page); |
e2a50c1f | 2433 | unlock_page(page); |
7d64ae3a AR |
2434 | return 0; |
2435 | out_delete_from_cache: | |
e2a50c1f | 2436 | delete_from_page_cache(page); |
4c27fe4c | 2437 | out_release: |
9cc90c66 | 2438 | unlock_page(page); |
4c27fe4c | 2439 | put_page(page); |
4c27fe4c | 2440 | out_unacct_blocks: |
0f079694 | 2441 | shmem_inode_unacct_blocks(inode, 1); |
7d64ae3a | 2442 | return ret; |
8d103963 | 2443 | } |
3460f6e5 | 2444 | #endif /* CONFIG_USERFAULTFD */ |
8d103963 | 2445 | |
1da177e4 | 2446 | #ifdef CONFIG_TMPFS |
92e1d5be | 2447 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2448 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2449 | |
6d9d88d0 JS |
2450 | #ifdef CONFIG_TMPFS_XATTR |
2451 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2452 | #else | |
2453 | #define shmem_initxattrs NULL | |
2454 | #endif | |
2455 | ||
1da177e4 | 2456 | static int |
800d15a5 NP |
2457 | shmem_write_begin(struct file *file, struct address_space *mapping, |
2458 | loff_t pos, unsigned len, unsigned flags, | |
2459 | struct page **pagep, void **fsdata) | |
1da177e4 | 2460 | { |
800d15a5 | 2461 | struct inode *inode = mapping->host; |
40e041a2 | 2462 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2463 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 | 2464 | |
9608703e | 2465 | /* i_rwsem is held by caller */ |
ab3948f5 JFG |
2466 | if (unlikely(info->seals & (F_SEAL_GROW | |
2467 | F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) { | |
2468 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) | |
40e041a2 DH |
2469 | return -EPERM; |
2470 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2471 | return -EPERM; | |
2472 | } | |
2473 | ||
9e18eb29 | 2474 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
2475 | } |
2476 | ||
2477 | static int | |
2478 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2479 | loff_t pos, unsigned len, unsigned copied, | |
2480 | struct page *page, void *fsdata) | |
2481 | { | |
2482 | struct inode *inode = mapping->host; | |
2483 | ||
d3602444 HD |
2484 | if (pos + copied > inode->i_size) |
2485 | i_size_write(inode, pos + copied); | |
2486 | ||
ec9516fb | 2487 | if (!PageUptodate(page)) { |
800d8c63 KS |
2488 | struct page *head = compound_head(page); |
2489 | if (PageTransCompound(page)) { | |
2490 | int i; | |
2491 | ||
2492 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2493 | if (head + i == page) | |
2494 | continue; | |
2495 | clear_highpage(head + i); | |
2496 | flush_dcache_page(head + i); | |
2497 | } | |
2498 | } | |
09cbfeaf KS |
2499 | if (copied < PAGE_SIZE) { |
2500 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2501 | zero_user_segments(page, 0, from, |
09cbfeaf | 2502 | from + copied, PAGE_SIZE); |
ec9516fb | 2503 | } |
800d8c63 | 2504 | SetPageUptodate(head); |
ec9516fb | 2505 | } |
800d15a5 | 2506 | set_page_dirty(page); |
6746aff7 | 2507 | unlock_page(page); |
09cbfeaf | 2508 | put_page(page); |
800d15a5 | 2509 | |
800d15a5 | 2510 | return copied; |
1da177e4 LT |
2511 | } |
2512 | ||
2ba5bbed | 2513 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2514 | { |
6e58e79d AV |
2515 | struct file *file = iocb->ki_filp; |
2516 | struct inode *inode = file_inode(file); | |
1da177e4 | 2517 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2518 | pgoff_t index; |
2519 | unsigned long offset; | |
a0ee5ec5 | 2520 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2521 | int error = 0; |
cb66a7a1 | 2522 | ssize_t retval = 0; |
6e58e79d | 2523 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2524 | |
2525 | /* | |
2526 | * Might this read be for a stacking filesystem? Then when reading | |
2527 | * holes of a sparse file, we actually need to allocate those pages, | |
2528 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2529 | */ | |
777eda2c | 2530 | if (!iter_is_iovec(to)) |
75edd345 | 2531 | sgp = SGP_CACHE; |
1da177e4 | 2532 | |
09cbfeaf KS |
2533 | index = *ppos >> PAGE_SHIFT; |
2534 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2535 | |
2536 | for (;;) { | |
2537 | struct page *page = NULL; | |
41ffe5d5 HD |
2538 | pgoff_t end_index; |
2539 | unsigned long nr, ret; | |
1da177e4 LT |
2540 | loff_t i_size = i_size_read(inode); |
2541 | ||
09cbfeaf | 2542 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2543 | if (index > end_index) |
2544 | break; | |
2545 | if (index == end_index) { | |
09cbfeaf | 2546 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2547 | if (nr <= offset) |
2548 | break; | |
2549 | } | |
2550 | ||
9e18eb29 | 2551 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2552 | if (error) { |
2553 | if (error == -EINVAL) | |
2554 | error = 0; | |
1da177e4 LT |
2555 | break; |
2556 | } | |
75edd345 HD |
2557 | if (page) { |
2558 | if (sgp == SGP_CACHE) | |
2559 | set_page_dirty(page); | |
d3602444 | 2560 | unlock_page(page); |
75edd345 | 2561 | } |
1da177e4 LT |
2562 | |
2563 | /* | |
2564 | * We must evaluate after, since reads (unlike writes) | |
9608703e | 2565 | * are called without i_rwsem protection against truncate |
1da177e4 | 2566 | */ |
09cbfeaf | 2567 | nr = PAGE_SIZE; |
1da177e4 | 2568 | i_size = i_size_read(inode); |
09cbfeaf | 2569 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2570 | if (index == end_index) { |
09cbfeaf | 2571 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2572 | if (nr <= offset) { |
2573 | if (page) | |
09cbfeaf | 2574 | put_page(page); |
1da177e4 LT |
2575 | break; |
2576 | } | |
2577 | } | |
2578 | nr -= offset; | |
2579 | ||
2580 | if (page) { | |
2581 | /* | |
2582 | * If users can be writing to this page using arbitrary | |
2583 | * virtual addresses, take care about potential aliasing | |
2584 | * before reading the page on the kernel side. | |
2585 | */ | |
2586 | if (mapping_writably_mapped(mapping)) | |
2587 | flush_dcache_page(page); | |
2588 | /* | |
2589 | * Mark the page accessed if we read the beginning. | |
2590 | */ | |
2591 | if (!offset) | |
2592 | mark_page_accessed(page); | |
b5810039 | 2593 | } else { |
1da177e4 | 2594 | page = ZERO_PAGE(0); |
09cbfeaf | 2595 | get_page(page); |
b5810039 | 2596 | } |
1da177e4 LT |
2597 | |
2598 | /* | |
2599 | * Ok, we have the page, and it's up-to-date, so | |
2600 | * now we can copy it to user space... | |
1da177e4 | 2601 | */ |
2ba5bbed | 2602 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2603 | retval += ret; |
1da177e4 | 2604 | offset += ret; |
09cbfeaf KS |
2605 | index += offset >> PAGE_SHIFT; |
2606 | offset &= ~PAGE_MASK; | |
1da177e4 | 2607 | |
09cbfeaf | 2608 | put_page(page); |
2ba5bbed | 2609 | if (!iov_iter_count(to)) |
1da177e4 | 2610 | break; |
6e58e79d AV |
2611 | if (ret < nr) { |
2612 | error = -EFAULT; | |
2613 | break; | |
2614 | } | |
1da177e4 LT |
2615 | cond_resched(); |
2616 | } | |
2617 | ||
09cbfeaf | 2618 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2619 | file_accessed(file); |
2620 | return retval ? retval : error; | |
1da177e4 LT |
2621 | } |
2622 | ||
965c8e59 | 2623 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2624 | { |
2625 | struct address_space *mapping = file->f_mapping; | |
2626 | struct inode *inode = mapping->host; | |
220f2ac9 | 2627 | |
965c8e59 AM |
2628 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2629 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2630 | MAX_LFS_FILESIZE, i_size_read(inode)); |
41139aa4 MWO |
2631 | if (offset < 0) |
2632 | return -ENXIO; | |
2633 | ||
5955102c | 2634 | inode_lock(inode); |
9608703e | 2635 | /* We're holding i_rwsem so we can access i_size directly */ |
41139aa4 | 2636 | offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence); |
387aae6f HD |
2637 | if (offset >= 0) |
2638 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2639 | inode_unlock(inode); |
220f2ac9 HD |
2640 | return offset; |
2641 | } | |
2642 | ||
83e4fa9c HD |
2643 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2644 | loff_t len) | |
2645 | { | |
496ad9aa | 2646 | struct inode *inode = file_inode(file); |
e2d12e22 | 2647 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2648 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2649 | struct shmem_falloc shmem_falloc; |
d144bf62 | 2650 | pgoff_t start, index, end, undo_fallocend; |
e2d12e22 | 2651 | int error; |
83e4fa9c | 2652 | |
13ace4d0 HD |
2653 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2654 | return -EOPNOTSUPP; | |
2655 | ||
5955102c | 2656 | inode_lock(inode); |
83e4fa9c HD |
2657 | |
2658 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2659 | struct address_space *mapping = file->f_mapping; | |
2660 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2661 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2662 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2663 | |
9608703e | 2664 | /* protected by i_rwsem */ |
ab3948f5 | 2665 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { |
40e041a2 DH |
2666 | error = -EPERM; |
2667 | goto out; | |
2668 | } | |
2669 | ||
8e205f77 | 2670 | shmem_falloc.waitq = &shmem_falloc_waitq; |
aa71ecd8 | 2671 | shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT; |
f00cdc6d HD |
2672 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; |
2673 | spin_lock(&inode->i_lock); | |
2674 | inode->i_private = &shmem_falloc; | |
2675 | spin_unlock(&inode->i_lock); | |
2676 | ||
83e4fa9c HD |
2677 | if ((u64)unmap_end > (u64)unmap_start) |
2678 | unmap_mapping_range(mapping, unmap_start, | |
2679 | 1 + unmap_end - unmap_start, 0); | |
2680 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2681 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2682 | |
2683 | spin_lock(&inode->i_lock); | |
2684 | inode->i_private = NULL; | |
2685 | wake_up_all(&shmem_falloc_waitq); | |
2055da97 | 2686 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head)); |
8e205f77 | 2687 | spin_unlock(&inode->i_lock); |
83e4fa9c | 2688 | error = 0; |
8e205f77 | 2689 | goto out; |
e2d12e22 HD |
2690 | } |
2691 | ||
2692 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2693 | error = inode_newsize_ok(inode, offset + len); | |
2694 | if (error) | |
2695 | goto out; | |
2696 | ||
40e041a2 DH |
2697 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2698 | error = -EPERM; | |
2699 | goto out; | |
2700 | } | |
2701 | ||
09cbfeaf KS |
2702 | start = offset >> PAGE_SHIFT; |
2703 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2704 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2705 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2706 | error = -ENOSPC; | |
2707 | goto out; | |
83e4fa9c HD |
2708 | } |
2709 | ||
8e205f77 | 2710 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2711 | shmem_falloc.start = start; |
2712 | shmem_falloc.next = start; | |
2713 | shmem_falloc.nr_falloced = 0; | |
2714 | shmem_falloc.nr_unswapped = 0; | |
2715 | spin_lock(&inode->i_lock); | |
2716 | inode->i_private = &shmem_falloc; | |
2717 | spin_unlock(&inode->i_lock); | |
2718 | ||
d144bf62 HD |
2719 | /* |
2720 | * info->fallocend is only relevant when huge pages might be | |
2721 | * involved: to prevent split_huge_page() freeing fallocated | |
2722 | * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size. | |
2723 | */ | |
2724 | undo_fallocend = info->fallocend; | |
2725 | if (info->fallocend < end) | |
2726 | info->fallocend = end; | |
2727 | ||
050dcb5c | 2728 | for (index = start; index < end; ) { |
e2d12e22 HD |
2729 | struct page *page; |
2730 | ||
2731 | /* | |
2732 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2733 | * been interrupted because we are using up too much memory. | |
2734 | */ | |
2735 | if (signal_pending(current)) | |
2736 | error = -EINTR; | |
1aac1400 HD |
2737 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2738 | error = -ENOMEM; | |
e2d12e22 | 2739 | else |
9e18eb29 | 2740 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2741 | if (error) { |
d144bf62 | 2742 | info->fallocend = undo_fallocend; |
1635f6a7 | 2743 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2744 | if (index > start) { |
2745 | shmem_undo_range(inode, | |
2746 | (loff_t)start << PAGE_SHIFT, | |
2747 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2748 | } | |
1aac1400 | 2749 | goto undone; |
e2d12e22 HD |
2750 | } |
2751 | ||
050dcb5c HD |
2752 | index++; |
2753 | /* | |
2754 | * Here is a more important optimization than it appears: | |
2755 | * a second SGP_FALLOC on the same huge page will clear it, | |
2756 | * making it PageUptodate and un-undoable if we fail later. | |
2757 | */ | |
2758 | if (PageTransCompound(page)) { | |
2759 | index = round_up(index, HPAGE_PMD_NR); | |
2760 | /* Beware 32-bit wraparound */ | |
2761 | if (!index) | |
2762 | index--; | |
2763 | } | |
2764 | ||
1aac1400 HD |
2765 | /* |
2766 | * Inform shmem_writepage() how far we have reached. | |
2767 | * No need for lock or barrier: we have the page lock. | |
2768 | */ | |
1aac1400 | 2769 | if (!PageUptodate(page)) |
050dcb5c HD |
2770 | shmem_falloc.nr_falloced += index - shmem_falloc.next; |
2771 | shmem_falloc.next = index; | |
1aac1400 | 2772 | |
e2d12e22 | 2773 | /* |
1635f6a7 HD |
2774 | * If !PageUptodate, leave it that way so that freeable pages |
2775 | * can be recognized if we need to rollback on error later. | |
2776 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2777 | * than free the pages we are allocating (and SGP_CACHE pages |
2778 | * might still be clean: we now need to mark those dirty too). | |
2779 | */ | |
2780 | set_page_dirty(page); | |
2781 | unlock_page(page); | |
09cbfeaf | 2782 | put_page(page); |
e2d12e22 HD |
2783 | cond_resched(); |
2784 | } | |
2785 | ||
2786 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2787 | i_size_write(inode, offset + len); | |
078cd827 | 2788 | inode->i_ctime = current_time(inode); |
1aac1400 HD |
2789 | undone: |
2790 | spin_lock(&inode->i_lock); | |
2791 | inode->i_private = NULL; | |
2792 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2793 | out: |
5955102c | 2794 | inode_unlock(inode); |
83e4fa9c HD |
2795 | return error; |
2796 | } | |
2797 | ||
726c3342 | 2798 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2799 | { |
726c3342 | 2800 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2801 | |
2802 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2803 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2804 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2805 | if (sbinfo->max_blocks) { |
1da177e4 | 2806 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2807 | buf->f_bavail = |
2808 | buf->f_bfree = sbinfo->max_blocks - | |
2809 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2810 | } |
2811 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2812 | buf->f_files = sbinfo->max_inodes; |
2813 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2814 | } |
2815 | /* else leave those fields 0 like simple_statfs */ | |
59cda49e AG |
2816 | |
2817 | buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b); | |
2818 | ||
1da177e4 LT |
2819 | return 0; |
2820 | } | |
2821 | ||
2822 | /* | |
2823 | * File creation. Allocate an inode, and we're done.. | |
2824 | */ | |
2825 | static int | |
549c7297 CB |
2826 | shmem_mknod(struct user_namespace *mnt_userns, struct inode *dir, |
2827 | struct dentry *dentry, umode_t mode, dev_t dev) | |
1da177e4 | 2828 | { |
0b0a0806 | 2829 | struct inode *inode; |
1da177e4 LT |
2830 | int error = -ENOSPC; |
2831 | ||
454abafe | 2832 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2833 | if (inode) { |
feda821e CH |
2834 | error = simple_acl_create(dir, inode); |
2835 | if (error) | |
2836 | goto out_iput; | |
2a7dba39 | 2837 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2838 | &dentry->d_name, |
6d9d88d0 | 2839 | shmem_initxattrs, NULL); |
feda821e CH |
2840 | if (error && error != -EOPNOTSUPP) |
2841 | goto out_iput; | |
37ec43cd | 2842 | |
718deb6b | 2843 | error = 0; |
1da177e4 | 2844 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 2845 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
2846 | d_instantiate(dentry, inode); |
2847 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2848 | } |
2849 | return error; | |
feda821e CH |
2850 | out_iput: |
2851 | iput(inode); | |
2852 | return error; | |
1da177e4 LT |
2853 | } |
2854 | ||
60545d0d | 2855 | static int |
549c7297 CB |
2856 | shmem_tmpfile(struct user_namespace *mnt_userns, struct inode *dir, |
2857 | struct dentry *dentry, umode_t mode) | |
60545d0d AV |
2858 | { |
2859 | struct inode *inode; | |
2860 | int error = -ENOSPC; | |
2861 | ||
2862 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
2863 | if (inode) { | |
2864 | error = security_inode_init_security(inode, dir, | |
2865 | NULL, | |
2866 | shmem_initxattrs, NULL); | |
feda821e CH |
2867 | if (error && error != -EOPNOTSUPP) |
2868 | goto out_iput; | |
2869 | error = simple_acl_create(dir, inode); | |
2870 | if (error) | |
2871 | goto out_iput; | |
60545d0d AV |
2872 | d_tmpfile(dentry, inode); |
2873 | } | |
2874 | return error; | |
feda821e CH |
2875 | out_iput: |
2876 | iput(inode); | |
2877 | return error; | |
60545d0d AV |
2878 | } |
2879 | ||
549c7297 CB |
2880 | static int shmem_mkdir(struct user_namespace *mnt_userns, struct inode *dir, |
2881 | struct dentry *dentry, umode_t mode) | |
1da177e4 LT |
2882 | { |
2883 | int error; | |
2884 | ||
549c7297 CB |
2885 | if ((error = shmem_mknod(&init_user_ns, dir, dentry, |
2886 | mode | S_IFDIR, 0))) | |
1da177e4 | 2887 | return error; |
d8c76e6f | 2888 | inc_nlink(dir); |
1da177e4 LT |
2889 | return 0; |
2890 | } | |
2891 | ||
549c7297 CB |
2892 | static int shmem_create(struct user_namespace *mnt_userns, struct inode *dir, |
2893 | struct dentry *dentry, umode_t mode, bool excl) | |
1da177e4 | 2894 | { |
549c7297 | 2895 | return shmem_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0); |
1da177e4 LT |
2896 | } |
2897 | ||
2898 | /* | |
2899 | * Link a file.. | |
2900 | */ | |
2901 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2902 | { | |
75c3cfa8 | 2903 | struct inode *inode = d_inode(old_dentry); |
29b00e60 | 2904 | int ret = 0; |
1da177e4 LT |
2905 | |
2906 | /* | |
2907 | * No ordinary (disk based) filesystem counts links as inodes; | |
2908 | * but each new link needs a new dentry, pinning lowmem, and | |
2909 | * tmpfs dentries cannot be pruned until they are unlinked. | |
1062af92 DW |
2910 | * But if an O_TMPFILE file is linked into the tmpfs, the |
2911 | * first link must skip that, to get the accounting right. | |
1da177e4 | 2912 | */ |
1062af92 | 2913 | if (inode->i_nlink) { |
e809d5f0 | 2914 | ret = shmem_reserve_inode(inode->i_sb, NULL); |
1062af92 DW |
2915 | if (ret) |
2916 | goto out; | |
2917 | } | |
1da177e4 LT |
2918 | |
2919 | dir->i_size += BOGO_DIRENT_SIZE; | |
078cd827 | 2920 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
d8c76e6f | 2921 | inc_nlink(inode); |
7de9c6ee | 2922 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
2923 | dget(dentry); /* Extra pinning count for the created dentry */ |
2924 | d_instantiate(dentry, inode); | |
5b04c689 PE |
2925 | out: |
2926 | return ret; | |
1da177e4 LT |
2927 | } |
2928 | ||
2929 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
2930 | { | |
75c3cfa8 | 2931 | struct inode *inode = d_inode(dentry); |
1da177e4 | 2932 | |
5b04c689 PE |
2933 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2934 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
2935 | |
2936 | dir->i_size -= BOGO_DIRENT_SIZE; | |
078cd827 | 2937 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
9a53c3a7 | 2938 | drop_nlink(inode); |
1da177e4 LT |
2939 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
2940 | return 0; | |
2941 | } | |
2942 | ||
2943 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
2944 | { | |
2945 | if (!simple_empty(dentry)) | |
2946 | return -ENOTEMPTY; | |
2947 | ||
75c3cfa8 | 2948 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 2949 | drop_nlink(dir); |
1da177e4 LT |
2950 | return shmem_unlink(dir, dentry); |
2951 | } | |
2952 | ||
37456771 MS |
2953 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
2954 | { | |
e36cb0b8 DH |
2955 | bool old_is_dir = d_is_dir(old_dentry); |
2956 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
2957 | |
2958 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
2959 | if (old_is_dir) { | |
2960 | drop_nlink(old_dir); | |
2961 | inc_nlink(new_dir); | |
2962 | } else { | |
2963 | drop_nlink(new_dir); | |
2964 | inc_nlink(old_dir); | |
2965 | } | |
2966 | } | |
2967 | old_dir->i_ctime = old_dir->i_mtime = | |
2968 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 | 2969 | d_inode(old_dentry)->i_ctime = |
078cd827 | 2970 | d_inode(new_dentry)->i_ctime = current_time(old_dir); |
37456771 MS |
2971 | |
2972 | return 0; | |
2973 | } | |
2974 | ||
549c7297 CB |
2975 | static int shmem_whiteout(struct user_namespace *mnt_userns, |
2976 | struct inode *old_dir, struct dentry *old_dentry) | |
46fdb794 MS |
2977 | { |
2978 | struct dentry *whiteout; | |
2979 | int error; | |
2980 | ||
2981 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
2982 | if (!whiteout) | |
2983 | return -ENOMEM; | |
2984 | ||
549c7297 | 2985 | error = shmem_mknod(&init_user_ns, old_dir, whiteout, |
46fdb794 MS |
2986 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); |
2987 | dput(whiteout); | |
2988 | if (error) | |
2989 | return error; | |
2990 | ||
2991 | /* | |
2992 | * Cheat and hash the whiteout while the old dentry is still in | |
2993 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
2994 | * | |
2995 | * d_lookup() will consistently find one of them at this point, | |
2996 | * not sure which one, but that isn't even important. | |
2997 | */ | |
2998 | d_rehash(whiteout); | |
2999 | return 0; | |
3000 | } | |
3001 | ||
1da177e4 LT |
3002 | /* |
3003 | * The VFS layer already does all the dentry stuff for rename, | |
3004 | * we just have to decrement the usage count for the target if | |
3005 | * it exists so that the VFS layer correctly free's it when it | |
3006 | * gets overwritten. | |
3007 | */ | |
549c7297 CB |
3008 | static int shmem_rename2(struct user_namespace *mnt_userns, |
3009 | struct inode *old_dir, struct dentry *old_dentry, | |
3010 | struct inode *new_dir, struct dentry *new_dentry, | |
3011 | unsigned int flags) | |
1da177e4 | 3012 | { |
75c3cfa8 | 3013 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
3014 | int they_are_dirs = S_ISDIR(inode->i_mode); |
3015 | ||
46fdb794 | 3016 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
3017 | return -EINVAL; |
3018 | ||
37456771 MS |
3019 | if (flags & RENAME_EXCHANGE) |
3020 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
3021 | ||
1da177e4 LT |
3022 | if (!simple_empty(new_dentry)) |
3023 | return -ENOTEMPTY; | |
3024 | ||
46fdb794 MS |
3025 | if (flags & RENAME_WHITEOUT) { |
3026 | int error; | |
3027 | ||
549c7297 | 3028 | error = shmem_whiteout(&init_user_ns, old_dir, old_dentry); |
46fdb794 MS |
3029 | if (error) |
3030 | return error; | |
3031 | } | |
3032 | ||
75c3cfa8 | 3033 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 3034 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 3035 | if (they_are_dirs) { |
75c3cfa8 | 3036 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 3037 | drop_nlink(old_dir); |
b928095b | 3038 | } |
1da177e4 | 3039 | } else if (they_are_dirs) { |
9a53c3a7 | 3040 | drop_nlink(old_dir); |
d8c76e6f | 3041 | inc_nlink(new_dir); |
1da177e4 LT |
3042 | } |
3043 | ||
3044 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
3045 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
3046 | old_dir->i_ctime = old_dir->i_mtime = | |
3047 | new_dir->i_ctime = new_dir->i_mtime = | |
078cd827 | 3048 | inode->i_ctime = current_time(old_dir); |
1da177e4 LT |
3049 | return 0; |
3050 | } | |
3051 | ||
549c7297 CB |
3052 | static int shmem_symlink(struct user_namespace *mnt_userns, struct inode *dir, |
3053 | struct dentry *dentry, const char *symname) | |
1da177e4 LT |
3054 | { |
3055 | int error; | |
3056 | int len; | |
3057 | struct inode *inode; | |
9276aad6 | 3058 | struct page *page; |
1da177e4 LT |
3059 | |
3060 | len = strlen(symname) + 1; | |
09cbfeaf | 3061 | if (len > PAGE_SIZE) |
1da177e4 LT |
3062 | return -ENAMETOOLONG; |
3063 | ||
0825a6f9 JP |
3064 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0, |
3065 | VM_NORESERVE); | |
1da177e4 LT |
3066 | if (!inode) |
3067 | return -ENOSPC; | |
3068 | ||
9d8f13ba | 3069 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3070 | shmem_initxattrs, NULL); |
343c3d7f MN |
3071 | if (error && error != -EOPNOTSUPP) { |
3072 | iput(inode); | |
3073 | return error; | |
570bc1c2 SS |
3074 | } |
3075 | ||
1da177e4 | 3076 | inode->i_size = len-1; |
69f07ec9 | 3077 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3078 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3079 | if (!inode->i_link) { | |
69f07ec9 HD |
3080 | iput(inode); |
3081 | return -ENOMEM; | |
3082 | } | |
3083 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3084 | } else { |
e8ecde25 | 3085 | inode_nohighmem(inode); |
9e18eb29 | 3086 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
3087 | if (error) { |
3088 | iput(inode); | |
3089 | return error; | |
3090 | } | |
14fcc23f | 3091 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 3092 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 3093 | memcpy(page_address(page), symname, len); |
ec9516fb | 3094 | SetPageUptodate(page); |
1da177e4 | 3095 | set_page_dirty(page); |
6746aff7 | 3096 | unlock_page(page); |
09cbfeaf | 3097 | put_page(page); |
1da177e4 | 3098 | } |
1da177e4 | 3099 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3100 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3101 | d_instantiate(dentry, inode); |
3102 | dget(dentry); | |
3103 | return 0; | |
3104 | } | |
3105 | ||
fceef393 | 3106 | static void shmem_put_link(void *arg) |
1da177e4 | 3107 | { |
fceef393 AV |
3108 | mark_page_accessed(arg); |
3109 | put_page(arg); | |
1da177e4 LT |
3110 | } |
3111 | ||
6b255391 | 3112 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
3113 | struct inode *inode, |
3114 | struct delayed_call *done) | |
1da177e4 | 3115 | { |
1da177e4 | 3116 | struct page *page = NULL; |
6b255391 | 3117 | int error; |
6a6c9904 AV |
3118 | if (!dentry) { |
3119 | page = find_get_page(inode->i_mapping, 0); | |
3120 | if (!page) | |
3121 | return ERR_PTR(-ECHILD); | |
3122 | if (!PageUptodate(page)) { | |
3123 | put_page(page); | |
3124 | return ERR_PTR(-ECHILD); | |
3125 | } | |
3126 | } else { | |
9e18eb29 | 3127 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3128 | if (error) |
3129 | return ERR_PTR(error); | |
3130 | unlock_page(page); | |
3131 | } | |
fceef393 | 3132 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3133 | return page_address(page); |
1da177e4 LT |
3134 | } |
3135 | ||
b09e0fa4 | 3136 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3137 | /* |
b09e0fa4 EP |
3138 | * Superblocks without xattr inode operations may get some security.* xattr |
3139 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3140 | * like ACLs, we also need to implement the security.* handlers at |
3141 | * filesystem level, though. | |
3142 | */ | |
3143 | ||
6d9d88d0 JS |
3144 | /* |
3145 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3146 | */ | |
3147 | static int shmem_initxattrs(struct inode *inode, | |
3148 | const struct xattr *xattr_array, | |
3149 | void *fs_info) | |
3150 | { | |
3151 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3152 | const struct xattr *xattr; | |
38f38657 | 3153 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3154 | size_t len; |
3155 | ||
3156 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3157 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3158 | if (!new_xattr) |
3159 | return -ENOMEM; | |
3160 | ||
3161 | len = strlen(xattr->name) + 1; | |
3162 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3163 | GFP_KERNEL); | |
3164 | if (!new_xattr->name) { | |
3bef735a | 3165 | kvfree(new_xattr); |
6d9d88d0 JS |
3166 | return -ENOMEM; |
3167 | } | |
3168 | ||
3169 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3170 | XATTR_SECURITY_PREFIX_LEN); | |
3171 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3172 | xattr->name, len); | |
3173 | ||
38f38657 | 3174 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3175 | } |
3176 | ||
3177 | return 0; | |
3178 | } | |
3179 | ||
aa7c5241 | 3180 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3181 | struct dentry *unused, struct inode *inode, |
3182 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3183 | { |
b296821a | 3184 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3185 | |
aa7c5241 | 3186 | name = xattr_full_name(handler, name); |
38f38657 | 3187 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3188 | } |
3189 | ||
aa7c5241 | 3190 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
e65ce2a5 | 3191 | struct user_namespace *mnt_userns, |
59301226 AV |
3192 | struct dentry *unused, struct inode *inode, |
3193 | const char *name, const void *value, | |
3194 | size_t size, int flags) | |
b09e0fa4 | 3195 | { |
59301226 | 3196 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3197 | |
aa7c5241 | 3198 | name = xattr_full_name(handler, name); |
a46a2295 | 3199 | return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL); |
b09e0fa4 EP |
3200 | } |
3201 | ||
aa7c5241 AG |
3202 | static const struct xattr_handler shmem_security_xattr_handler = { |
3203 | .prefix = XATTR_SECURITY_PREFIX, | |
3204 | .get = shmem_xattr_handler_get, | |
3205 | .set = shmem_xattr_handler_set, | |
3206 | }; | |
b09e0fa4 | 3207 | |
aa7c5241 AG |
3208 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3209 | .prefix = XATTR_TRUSTED_PREFIX, | |
3210 | .get = shmem_xattr_handler_get, | |
3211 | .set = shmem_xattr_handler_set, | |
3212 | }; | |
b09e0fa4 | 3213 | |
aa7c5241 AG |
3214 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3215 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3216 | &posix_acl_access_xattr_handler, | |
3217 | &posix_acl_default_xattr_handler, | |
3218 | #endif | |
3219 | &shmem_security_xattr_handler, | |
3220 | &shmem_trusted_xattr_handler, | |
3221 | NULL | |
3222 | }; | |
b09e0fa4 EP |
3223 | |
3224 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3225 | { | |
75c3cfa8 | 3226 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3227 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3228 | } |
3229 | #endif /* CONFIG_TMPFS_XATTR */ | |
3230 | ||
69f07ec9 | 3231 | static const struct inode_operations shmem_short_symlink_operations = { |
6b255391 | 3232 | .get_link = simple_get_link, |
b09e0fa4 | 3233 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3234 | .listxattr = shmem_listxattr, |
b09e0fa4 EP |
3235 | #endif |
3236 | }; | |
3237 | ||
3238 | static const struct inode_operations shmem_symlink_inode_operations = { | |
6b255391 | 3239 | .get_link = shmem_get_link, |
b09e0fa4 | 3240 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3241 | .listxattr = shmem_listxattr, |
39f0247d | 3242 | #endif |
b09e0fa4 | 3243 | }; |
39f0247d | 3244 | |
91828a40 DG |
3245 | static struct dentry *shmem_get_parent(struct dentry *child) |
3246 | { | |
3247 | return ERR_PTR(-ESTALE); | |
3248 | } | |
3249 | ||
3250 | static int shmem_match(struct inode *ino, void *vfh) | |
3251 | { | |
3252 | __u32 *fh = vfh; | |
3253 | __u64 inum = fh[2]; | |
3254 | inum = (inum << 32) | fh[1]; | |
3255 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
3256 | } | |
3257 | ||
12ba780d AG |
3258 | /* Find any alias of inode, but prefer a hashed alias */ |
3259 | static struct dentry *shmem_find_alias(struct inode *inode) | |
3260 | { | |
3261 | struct dentry *alias = d_find_alias(inode); | |
3262 | ||
3263 | return alias ?: d_find_any_alias(inode); | |
3264 | } | |
3265 | ||
3266 | ||
480b116c CH |
3267 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3268 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3269 | { |
91828a40 | 3270 | struct inode *inode; |
480b116c | 3271 | struct dentry *dentry = NULL; |
35c2a7f4 | 3272 | u64 inum; |
480b116c CH |
3273 | |
3274 | if (fh_len < 3) | |
3275 | return NULL; | |
91828a40 | 3276 | |
35c2a7f4 HD |
3277 | inum = fid->raw[2]; |
3278 | inum = (inum << 32) | fid->raw[1]; | |
3279 | ||
480b116c CH |
3280 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
3281 | shmem_match, fid->raw); | |
91828a40 | 3282 | if (inode) { |
12ba780d | 3283 | dentry = shmem_find_alias(inode); |
91828a40 DG |
3284 | iput(inode); |
3285 | } | |
3286 | ||
480b116c | 3287 | return dentry; |
91828a40 DG |
3288 | } |
3289 | ||
b0b0382b AV |
3290 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3291 | struct inode *parent) | |
91828a40 | 3292 | { |
5fe0c237 AK |
3293 | if (*len < 3) { |
3294 | *len = 3; | |
94e07a75 | 3295 | return FILEID_INVALID; |
5fe0c237 | 3296 | } |
91828a40 | 3297 | |
1d3382cb | 3298 | if (inode_unhashed(inode)) { |
91828a40 DG |
3299 | /* Unfortunately insert_inode_hash is not idempotent, |
3300 | * so as we hash inodes here rather than at creation | |
3301 | * time, we need a lock to ensure we only try | |
3302 | * to do it once | |
3303 | */ | |
3304 | static DEFINE_SPINLOCK(lock); | |
3305 | spin_lock(&lock); | |
1d3382cb | 3306 | if (inode_unhashed(inode)) |
91828a40 DG |
3307 | __insert_inode_hash(inode, |
3308 | inode->i_ino + inode->i_generation); | |
3309 | spin_unlock(&lock); | |
3310 | } | |
3311 | ||
3312 | fh[0] = inode->i_generation; | |
3313 | fh[1] = inode->i_ino; | |
3314 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
3315 | ||
3316 | *len = 3; | |
3317 | return 1; | |
3318 | } | |
3319 | ||
39655164 | 3320 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3321 | .get_parent = shmem_get_parent, |
91828a40 | 3322 | .encode_fh = shmem_encode_fh, |
480b116c | 3323 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3324 | }; |
3325 | ||
626c3920 AV |
3326 | enum shmem_param { |
3327 | Opt_gid, | |
3328 | Opt_huge, | |
3329 | Opt_mode, | |
3330 | Opt_mpol, | |
3331 | Opt_nr_blocks, | |
3332 | Opt_nr_inodes, | |
3333 | Opt_size, | |
3334 | Opt_uid, | |
ea3271f7 CD |
3335 | Opt_inode32, |
3336 | Opt_inode64, | |
626c3920 AV |
3337 | }; |
3338 | ||
5eede625 | 3339 | static const struct constant_table shmem_param_enums_huge[] = { |
2710c957 AV |
3340 | {"never", SHMEM_HUGE_NEVER }, |
3341 | {"always", SHMEM_HUGE_ALWAYS }, | |
3342 | {"within_size", SHMEM_HUGE_WITHIN_SIZE }, | |
3343 | {"advise", SHMEM_HUGE_ADVISE }, | |
2710c957 AV |
3344 | {} |
3345 | }; | |
3346 | ||
d7167b14 | 3347 | const struct fs_parameter_spec shmem_fs_parameters[] = { |
626c3920 | 3348 | fsparam_u32 ("gid", Opt_gid), |
2710c957 | 3349 | fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge), |
626c3920 AV |
3350 | fsparam_u32oct("mode", Opt_mode), |
3351 | fsparam_string("mpol", Opt_mpol), | |
3352 | fsparam_string("nr_blocks", Opt_nr_blocks), | |
3353 | fsparam_string("nr_inodes", Opt_nr_inodes), | |
3354 | fsparam_string("size", Opt_size), | |
3355 | fsparam_u32 ("uid", Opt_uid), | |
ea3271f7 CD |
3356 | fsparam_flag ("inode32", Opt_inode32), |
3357 | fsparam_flag ("inode64", Opt_inode64), | |
626c3920 AV |
3358 | {} |
3359 | }; | |
3360 | ||
f3235626 | 3361 | static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param) |
1da177e4 | 3362 | { |
f3235626 | 3363 | struct shmem_options *ctx = fc->fs_private; |
626c3920 AV |
3364 | struct fs_parse_result result; |
3365 | unsigned long long size; | |
e04dc423 | 3366 | char *rest; |
626c3920 AV |
3367 | int opt; |
3368 | ||
d7167b14 | 3369 | opt = fs_parse(fc, shmem_fs_parameters, param, &result); |
f3235626 | 3370 | if (opt < 0) |
626c3920 | 3371 | return opt; |
1da177e4 | 3372 | |
626c3920 AV |
3373 | switch (opt) { |
3374 | case Opt_size: | |
3375 | size = memparse(param->string, &rest); | |
e04dc423 AV |
3376 | if (*rest == '%') { |
3377 | size <<= PAGE_SHIFT; | |
3378 | size *= totalram_pages(); | |
3379 | do_div(size, 100); | |
3380 | rest++; | |
3381 | } | |
3382 | if (*rest) | |
626c3920 | 3383 | goto bad_value; |
e04dc423 AV |
3384 | ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE); |
3385 | ctx->seen |= SHMEM_SEEN_BLOCKS; | |
626c3920 AV |
3386 | break; |
3387 | case Opt_nr_blocks: | |
3388 | ctx->blocks = memparse(param->string, &rest); | |
e04dc423 | 3389 | if (*rest) |
626c3920 | 3390 | goto bad_value; |
e04dc423 | 3391 | ctx->seen |= SHMEM_SEEN_BLOCKS; |
626c3920 AV |
3392 | break; |
3393 | case Opt_nr_inodes: | |
3394 | ctx->inodes = memparse(param->string, &rest); | |
e04dc423 | 3395 | if (*rest) |
626c3920 | 3396 | goto bad_value; |
e04dc423 | 3397 | ctx->seen |= SHMEM_SEEN_INODES; |
626c3920 AV |
3398 | break; |
3399 | case Opt_mode: | |
3400 | ctx->mode = result.uint_32 & 07777; | |
3401 | break; | |
3402 | case Opt_uid: | |
3403 | ctx->uid = make_kuid(current_user_ns(), result.uint_32); | |
e04dc423 | 3404 | if (!uid_valid(ctx->uid)) |
626c3920 AV |
3405 | goto bad_value; |
3406 | break; | |
3407 | case Opt_gid: | |
3408 | ctx->gid = make_kgid(current_user_ns(), result.uint_32); | |
e04dc423 | 3409 | if (!gid_valid(ctx->gid)) |
626c3920 AV |
3410 | goto bad_value; |
3411 | break; | |
3412 | case Opt_huge: | |
3413 | ctx->huge = result.uint_32; | |
3414 | if (ctx->huge != SHMEM_HUGE_NEVER && | |
396bcc52 | 3415 | !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && |
626c3920 AV |
3416 | has_transparent_hugepage())) |
3417 | goto unsupported_parameter; | |
e04dc423 | 3418 | ctx->seen |= SHMEM_SEEN_HUGE; |
626c3920 AV |
3419 | break; |
3420 | case Opt_mpol: | |
3421 | if (IS_ENABLED(CONFIG_NUMA)) { | |
3422 | mpol_put(ctx->mpol); | |
3423 | ctx->mpol = NULL; | |
3424 | if (mpol_parse_str(param->string, &ctx->mpol)) | |
3425 | goto bad_value; | |
3426 | break; | |
3427 | } | |
3428 | goto unsupported_parameter; | |
ea3271f7 CD |
3429 | case Opt_inode32: |
3430 | ctx->full_inums = false; | |
3431 | ctx->seen |= SHMEM_SEEN_INUMS; | |
3432 | break; | |
3433 | case Opt_inode64: | |
3434 | if (sizeof(ino_t) < 8) { | |
3435 | return invalfc(fc, | |
3436 | "Cannot use inode64 with <64bit inums in kernel\n"); | |
3437 | } | |
3438 | ctx->full_inums = true; | |
3439 | ctx->seen |= SHMEM_SEEN_INUMS; | |
3440 | break; | |
e04dc423 AV |
3441 | } |
3442 | return 0; | |
3443 | ||
626c3920 | 3444 | unsupported_parameter: |
f35aa2bc | 3445 | return invalfc(fc, "Unsupported parameter '%s'", param->key); |
626c3920 | 3446 | bad_value: |
f35aa2bc | 3447 | return invalfc(fc, "Bad value for '%s'", param->key); |
e04dc423 AV |
3448 | } |
3449 | ||
f3235626 | 3450 | static int shmem_parse_options(struct fs_context *fc, void *data) |
e04dc423 | 3451 | { |
f3235626 DH |
3452 | char *options = data; |
3453 | ||
33f37c64 AV |
3454 | if (options) { |
3455 | int err = security_sb_eat_lsm_opts(options, &fc->security); | |
3456 | if (err) | |
3457 | return err; | |
3458 | } | |
3459 | ||
b00dc3ad | 3460 | while (options != NULL) { |
626c3920 | 3461 | char *this_char = options; |
b00dc3ad HD |
3462 | for (;;) { |
3463 | /* | |
3464 | * NUL-terminate this option: unfortunately, | |
3465 | * mount options form a comma-separated list, | |
3466 | * but mpol's nodelist may also contain commas. | |
3467 | */ | |
3468 | options = strchr(options, ','); | |
3469 | if (options == NULL) | |
3470 | break; | |
3471 | options++; | |
3472 | if (!isdigit(*options)) { | |
3473 | options[-1] = '\0'; | |
3474 | break; | |
3475 | } | |
3476 | } | |
626c3920 | 3477 | if (*this_char) { |
68d68ff6 | 3478 | char *value = strchr(this_char, '='); |
f3235626 | 3479 | size_t len = 0; |
626c3920 AV |
3480 | int err; |
3481 | ||
3482 | if (value) { | |
3483 | *value++ = '\0'; | |
f3235626 | 3484 | len = strlen(value); |
626c3920 | 3485 | } |
f3235626 DH |
3486 | err = vfs_parse_fs_string(fc, this_char, value, len); |
3487 | if (err < 0) | |
3488 | return err; | |
1da177e4 | 3489 | } |
1da177e4 LT |
3490 | } |
3491 | return 0; | |
1da177e4 LT |
3492 | } |
3493 | ||
f3235626 DH |
3494 | /* |
3495 | * Reconfigure a shmem filesystem. | |
3496 | * | |
3497 | * Note that we disallow change from limited->unlimited blocks/inodes while any | |
3498 | * are in use; but we must separately disallow unlimited->limited, because in | |
3499 | * that case we have no record of how much is already in use. | |
3500 | */ | |
3501 | static int shmem_reconfigure(struct fs_context *fc) | |
1da177e4 | 3502 | { |
f3235626 DH |
3503 | struct shmem_options *ctx = fc->fs_private; |
3504 | struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb); | |
0edd73b3 | 3505 | unsigned long inodes; |
bf11b9a8 | 3506 | struct mempolicy *mpol = NULL; |
f3235626 | 3507 | const char *err; |
1da177e4 | 3508 | |
bf11b9a8 | 3509 | raw_spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3510 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
f3235626 DH |
3511 | if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) { |
3512 | if (!sbinfo->max_blocks) { | |
3513 | err = "Cannot retroactively limit size"; | |
0b5071dd | 3514 | goto out; |
f3235626 | 3515 | } |
0b5071dd | 3516 | if (percpu_counter_compare(&sbinfo->used_blocks, |
f3235626 DH |
3517 | ctx->blocks) > 0) { |
3518 | err = "Too small a size for current use"; | |
0b5071dd | 3519 | goto out; |
f3235626 | 3520 | } |
0b5071dd | 3521 | } |
f3235626 DH |
3522 | if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) { |
3523 | if (!sbinfo->max_inodes) { | |
3524 | err = "Cannot retroactively limit inodes"; | |
0b5071dd | 3525 | goto out; |
f3235626 DH |
3526 | } |
3527 | if (ctx->inodes < inodes) { | |
3528 | err = "Too few inodes for current use"; | |
0b5071dd | 3529 | goto out; |
f3235626 | 3530 | } |
0b5071dd | 3531 | } |
0edd73b3 | 3532 | |
ea3271f7 CD |
3533 | if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums && |
3534 | sbinfo->next_ino > UINT_MAX) { | |
3535 | err = "Current inum too high to switch to 32-bit inums"; | |
3536 | goto out; | |
3537 | } | |
3538 | ||
f3235626 DH |
3539 | if (ctx->seen & SHMEM_SEEN_HUGE) |
3540 | sbinfo->huge = ctx->huge; | |
ea3271f7 CD |
3541 | if (ctx->seen & SHMEM_SEEN_INUMS) |
3542 | sbinfo->full_inums = ctx->full_inums; | |
f3235626 DH |
3543 | if (ctx->seen & SHMEM_SEEN_BLOCKS) |
3544 | sbinfo->max_blocks = ctx->blocks; | |
3545 | if (ctx->seen & SHMEM_SEEN_INODES) { | |
3546 | sbinfo->max_inodes = ctx->inodes; | |
3547 | sbinfo->free_inodes = ctx->inodes - inodes; | |
0b5071dd | 3548 | } |
71fe804b | 3549 | |
5f00110f GT |
3550 | /* |
3551 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3552 | */ | |
f3235626 | 3553 | if (ctx->mpol) { |
bf11b9a8 | 3554 | mpol = sbinfo->mpol; |
f3235626 DH |
3555 | sbinfo->mpol = ctx->mpol; /* transfers initial ref */ |
3556 | ctx->mpol = NULL; | |
5f00110f | 3557 | } |
bf11b9a8 SAS |
3558 | raw_spin_unlock(&sbinfo->stat_lock); |
3559 | mpol_put(mpol); | |
f3235626 | 3560 | return 0; |
0edd73b3 | 3561 | out: |
bf11b9a8 | 3562 | raw_spin_unlock(&sbinfo->stat_lock); |
f35aa2bc | 3563 | return invalfc(fc, "%s", err); |
1da177e4 | 3564 | } |
680d794b | 3565 | |
34c80b1d | 3566 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3567 | { |
34c80b1d | 3568 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b AM |
3569 | |
3570 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3571 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3572 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b AM |
3573 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
3574 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
0825a6f9 | 3575 | if (sbinfo->mode != (0777 | S_ISVTX)) |
09208d15 | 3576 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3577 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3578 | seq_printf(seq, ",uid=%u", | |
3579 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3580 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3581 | seq_printf(seq, ",gid=%u", | |
3582 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
ea3271f7 CD |
3583 | |
3584 | /* | |
3585 | * Showing inode{64,32} might be useful even if it's the system default, | |
3586 | * since then people don't have to resort to checking both here and | |
3587 | * /proc/config.gz to confirm 64-bit inums were successfully applied | |
3588 | * (which may not even exist if IKCONFIG_PROC isn't enabled). | |
3589 | * | |
3590 | * We hide it when inode64 isn't the default and we are using 32-bit | |
3591 | * inodes, since that probably just means the feature isn't even under | |
3592 | * consideration. | |
3593 | * | |
3594 | * As such: | |
3595 | * | |
3596 | * +-----------------+-----------------+ | |
3597 | * | TMPFS_INODE64=y | TMPFS_INODE64=n | | |
3598 | * +------------------+-----------------+-----------------+ | |
3599 | * | full_inums=true | show | show | | |
3600 | * | full_inums=false | show | hide | | |
3601 | * +------------------+-----------------+-----------------+ | |
3602 | * | |
3603 | */ | |
3604 | if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums) | |
3605 | seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32)); | |
396bcc52 | 3606 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5a6e75f8 KS |
3607 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
3608 | if (sbinfo->huge) | |
3609 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3610 | #endif | |
71fe804b | 3611 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b AM |
3612 | return 0; |
3613 | } | |
9183df25 | 3614 | |
680d794b | 3615 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3616 | |
3617 | static void shmem_put_super(struct super_block *sb) | |
3618 | { | |
602586a8 HD |
3619 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3620 | ||
e809d5f0 | 3621 | free_percpu(sbinfo->ino_batch); |
602586a8 | 3622 | percpu_counter_destroy(&sbinfo->used_blocks); |
49cd0a5c | 3623 | mpol_put(sbinfo->mpol); |
602586a8 | 3624 | kfree(sbinfo); |
1da177e4 LT |
3625 | sb->s_fs_info = NULL; |
3626 | } | |
3627 | ||
f3235626 | 3628 | static int shmem_fill_super(struct super_block *sb, struct fs_context *fc) |
1da177e4 | 3629 | { |
f3235626 | 3630 | struct shmem_options *ctx = fc->fs_private; |
1da177e4 | 3631 | struct inode *inode; |
0edd73b3 | 3632 | struct shmem_sb_info *sbinfo; |
680d794b AM |
3633 | |
3634 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3635 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b AM |
3636 | L1_CACHE_BYTES), GFP_KERNEL); |
3637 | if (!sbinfo) | |
3638 | return -ENOMEM; | |
3639 | ||
680d794b | 3640 | sb->s_fs_info = sbinfo; |
1da177e4 | 3641 | |
0edd73b3 | 3642 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3643 | /* |
3644 | * Per default we only allow half of the physical ram per | |
3645 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3646 | * but the internal instance is left unlimited. | |
3647 | */ | |
1751e8a6 | 3648 | if (!(sb->s_flags & SB_KERNMOUNT)) { |
f3235626 DH |
3649 | if (!(ctx->seen & SHMEM_SEEN_BLOCKS)) |
3650 | ctx->blocks = shmem_default_max_blocks(); | |
3651 | if (!(ctx->seen & SHMEM_SEEN_INODES)) | |
3652 | ctx->inodes = shmem_default_max_inodes(); | |
ea3271f7 CD |
3653 | if (!(ctx->seen & SHMEM_SEEN_INUMS)) |
3654 | ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64); | |
ca4e0519 | 3655 | } else { |
1751e8a6 | 3656 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 3657 | } |
91828a40 | 3658 | sb->s_export_op = &shmem_export_ops; |
1751e8a6 | 3659 | sb->s_flags |= SB_NOSEC; |
1da177e4 | 3660 | #else |
1751e8a6 | 3661 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 3662 | #endif |
f3235626 DH |
3663 | sbinfo->max_blocks = ctx->blocks; |
3664 | sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes; | |
e809d5f0 CD |
3665 | if (sb->s_flags & SB_KERNMOUNT) { |
3666 | sbinfo->ino_batch = alloc_percpu(ino_t); | |
3667 | if (!sbinfo->ino_batch) | |
3668 | goto failed; | |
3669 | } | |
f3235626 DH |
3670 | sbinfo->uid = ctx->uid; |
3671 | sbinfo->gid = ctx->gid; | |
ea3271f7 | 3672 | sbinfo->full_inums = ctx->full_inums; |
f3235626 DH |
3673 | sbinfo->mode = ctx->mode; |
3674 | sbinfo->huge = ctx->huge; | |
3675 | sbinfo->mpol = ctx->mpol; | |
3676 | ctx->mpol = NULL; | |
1da177e4 | 3677 | |
bf11b9a8 | 3678 | raw_spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3679 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3680 | goto failed; |
779750d2 KS |
3681 | spin_lock_init(&sbinfo->shrinklist_lock); |
3682 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 3683 | |
285b2c4f | 3684 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3685 | sb->s_blocksize = PAGE_SIZE; |
3686 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3687 | sb->s_magic = TMPFS_MAGIC; |
3688 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3689 | sb->s_time_gran = 1; |
b09e0fa4 | 3690 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3691 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3692 | #endif |
3693 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
1751e8a6 | 3694 | sb->s_flags |= SB_POSIXACL; |
39f0247d | 3695 | #endif |
2b4db796 | 3696 | uuid_gen(&sb->s_uuid); |
0edd73b3 | 3697 | |
454abafe | 3698 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3699 | if (!inode) |
3700 | goto failed; | |
680d794b AM |
3701 | inode->i_uid = sbinfo->uid; |
3702 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3703 | sb->s_root = d_make_root(inode); |
3704 | if (!sb->s_root) | |
48fde701 | 3705 | goto failed; |
1da177e4 LT |
3706 | return 0; |
3707 | ||
1da177e4 LT |
3708 | failed: |
3709 | shmem_put_super(sb); | |
f2b346e4 | 3710 | return -ENOMEM; |
1da177e4 LT |
3711 | } |
3712 | ||
f3235626 DH |
3713 | static int shmem_get_tree(struct fs_context *fc) |
3714 | { | |
3715 | return get_tree_nodev(fc, shmem_fill_super); | |
3716 | } | |
3717 | ||
3718 | static void shmem_free_fc(struct fs_context *fc) | |
3719 | { | |
3720 | struct shmem_options *ctx = fc->fs_private; | |
3721 | ||
3722 | if (ctx) { | |
3723 | mpol_put(ctx->mpol); | |
3724 | kfree(ctx); | |
3725 | } | |
3726 | } | |
3727 | ||
3728 | static const struct fs_context_operations shmem_fs_context_ops = { | |
3729 | .free = shmem_free_fc, | |
3730 | .get_tree = shmem_get_tree, | |
3731 | #ifdef CONFIG_TMPFS | |
3732 | .parse_monolithic = shmem_parse_options, | |
3733 | .parse_param = shmem_parse_one, | |
3734 | .reconfigure = shmem_reconfigure, | |
3735 | #endif | |
3736 | }; | |
3737 | ||
fcc234f8 | 3738 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3739 | |
3740 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3741 | { | |
41ffe5d5 HD |
3742 | struct shmem_inode_info *info; |
3743 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3744 | if (!info) | |
1da177e4 | 3745 | return NULL; |
41ffe5d5 | 3746 | return &info->vfs_inode; |
1da177e4 LT |
3747 | } |
3748 | ||
74b1da56 | 3749 | static void shmem_free_in_core_inode(struct inode *inode) |
fa0d7e3d | 3750 | { |
84e710da AV |
3751 | if (S_ISLNK(inode->i_mode)) |
3752 | kfree(inode->i_link); | |
fa0d7e3d NP |
3753 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3754 | } | |
3755 | ||
1da177e4 LT |
3756 | static void shmem_destroy_inode(struct inode *inode) |
3757 | { | |
09208d15 | 3758 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3759 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
1da177e4 LT |
3760 | } |
3761 | ||
41ffe5d5 | 3762 | static void shmem_init_inode(void *foo) |
1da177e4 | 3763 | { |
41ffe5d5 HD |
3764 | struct shmem_inode_info *info = foo; |
3765 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3766 | } |
3767 | ||
9a8ec03e | 3768 | static void shmem_init_inodecache(void) |
1da177e4 LT |
3769 | { |
3770 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3771 | sizeof(struct shmem_inode_info), | |
5d097056 | 3772 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3773 | } |
3774 | ||
41ffe5d5 | 3775 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3776 | { |
1a1d92c1 | 3777 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3778 | } |
3779 | ||
30e6a51d | 3780 | const struct address_space_operations shmem_aops = { |
1da177e4 | 3781 | .writepage = shmem_writepage, |
76719325 | 3782 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3783 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3784 | .write_begin = shmem_write_begin, |
3785 | .write_end = shmem_write_end, | |
1da177e4 | 3786 | #endif |
1c93923c | 3787 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3788 | .migratepage = migrate_page, |
1c93923c | 3789 | #endif |
aa261f54 | 3790 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 3791 | }; |
30e6a51d | 3792 | EXPORT_SYMBOL(shmem_aops); |
1da177e4 | 3793 | |
15ad7cdc | 3794 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3795 | .mmap = shmem_mmap, |
c01d5b30 | 3796 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3797 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3798 | .llseek = shmem_file_llseek, |
2ba5bbed | 3799 | .read_iter = shmem_file_read_iter, |
8174202b | 3800 | .write_iter = generic_file_write_iter, |
1b061d92 | 3801 | .fsync = noop_fsync, |
82c156f8 | 3802 | .splice_read = generic_file_splice_read, |
f6cb85d0 | 3803 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3804 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3805 | #endif |
3806 | }; | |
3807 | ||
92e1d5be | 3808 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3809 | .getattr = shmem_getattr, |
94c1e62d | 3810 | .setattr = shmem_setattr, |
b09e0fa4 | 3811 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3812 | .listxattr = shmem_listxattr, |
feda821e | 3813 | .set_acl = simple_set_acl, |
b09e0fa4 | 3814 | #endif |
1da177e4 LT |
3815 | }; |
3816 | ||
92e1d5be | 3817 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3818 | #ifdef CONFIG_TMPFS |
3819 | .create = shmem_create, | |
3820 | .lookup = simple_lookup, | |
3821 | .link = shmem_link, | |
3822 | .unlink = shmem_unlink, | |
3823 | .symlink = shmem_symlink, | |
3824 | .mkdir = shmem_mkdir, | |
3825 | .rmdir = shmem_rmdir, | |
3826 | .mknod = shmem_mknod, | |
2773bf00 | 3827 | .rename = shmem_rename2, |
60545d0d | 3828 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3829 | #endif |
b09e0fa4 | 3830 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3831 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3832 | #endif |
39f0247d | 3833 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3834 | .setattr = shmem_setattr, |
feda821e | 3835 | .set_acl = simple_set_acl, |
39f0247d AG |
3836 | #endif |
3837 | }; | |
3838 | ||
92e1d5be | 3839 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3840 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3841 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3842 | #endif |
39f0247d | 3843 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3844 | .setattr = shmem_setattr, |
feda821e | 3845 | .set_acl = simple_set_acl, |
39f0247d | 3846 | #endif |
1da177e4 LT |
3847 | }; |
3848 | ||
759b9775 | 3849 | static const struct super_operations shmem_ops = { |
1da177e4 | 3850 | .alloc_inode = shmem_alloc_inode, |
74b1da56 | 3851 | .free_inode = shmem_free_in_core_inode, |
1da177e4 LT |
3852 | .destroy_inode = shmem_destroy_inode, |
3853 | #ifdef CONFIG_TMPFS | |
3854 | .statfs = shmem_statfs, | |
680d794b | 3855 | .show_options = shmem_show_options, |
1da177e4 | 3856 | #endif |
1f895f75 | 3857 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3858 | .drop_inode = generic_delete_inode, |
3859 | .put_super = shmem_put_super, | |
396bcc52 | 3860 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
779750d2 KS |
3861 | .nr_cached_objects = shmem_unused_huge_count, |
3862 | .free_cached_objects = shmem_unused_huge_scan, | |
3863 | #endif | |
1da177e4 LT |
3864 | }; |
3865 | ||
f0f37e2f | 3866 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3867 | .fault = shmem_fault, |
d7c17551 | 3868 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3869 | #ifdef CONFIG_NUMA |
3870 | .set_policy = shmem_set_policy, | |
3871 | .get_policy = shmem_get_policy, | |
3872 | #endif | |
3873 | }; | |
3874 | ||
f3235626 | 3875 | int shmem_init_fs_context(struct fs_context *fc) |
1da177e4 | 3876 | { |
f3235626 DH |
3877 | struct shmem_options *ctx; |
3878 | ||
3879 | ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL); | |
3880 | if (!ctx) | |
3881 | return -ENOMEM; | |
3882 | ||
3883 | ctx->mode = 0777 | S_ISVTX; | |
3884 | ctx->uid = current_fsuid(); | |
3885 | ctx->gid = current_fsgid(); | |
3886 | ||
3887 | fc->fs_private = ctx; | |
3888 | fc->ops = &shmem_fs_context_ops; | |
3889 | return 0; | |
1da177e4 LT |
3890 | } |
3891 | ||
41ffe5d5 | 3892 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3893 | .owner = THIS_MODULE, |
3894 | .name = "tmpfs", | |
f3235626 DH |
3895 | .init_fs_context = shmem_init_fs_context, |
3896 | #ifdef CONFIG_TMPFS | |
d7167b14 | 3897 | .parameters = shmem_fs_parameters, |
f3235626 | 3898 | #endif |
1da177e4 | 3899 | .kill_sb = kill_litter_super, |
01c70267 | 3900 | .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT, |
1da177e4 | 3901 | }; |
1da177e4 | 3902 | |
41ffe5d5 | 3903 | int __init shmem_init(void) |
1da177e4 LT |
3904 | { |
3905 | int error; | |
3906 | ||
9a8ec03e | 3907 | shmem_init_inodecache(); |
1da177e4 | 3908 | |
41ffe5d5 | 3909 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 3910 | if (error) { |
1170532b | 3911 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
3912 | goto out2; |
3913 | } | |
95dc112a | 3914 | |
ca4e0519 | 3915 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
3916 | if (IS_ERR(shm_mnt)) { |
3917 | error = PTR_ERR(shm_mnt); | |
1170532b | 3918 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
3919 | goto out1; |
3920 | } | |
5a6e75f8 | 3921 | |
396bcc52 | 3922 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
435c0b87 | 3923 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
3924 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
3925 | else | |
5e6e5a12 | 3926 | shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */ |
5a6e75f8 | 3927 | #endif |
1da177e4 LT |
3928 | return 0; |
3929 | ||
3930 | out1: | |
41ffe5d5 | 3931 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 3932 | out2: |
41ffe5d5 | 3933 | shmem_destroy_inodecache(); |
1da177e4 LT |
3934 | shm_mnt = ERR_PTR(error); |
3935 | return error; | |
3936 | } | |
853ac43a | 3937 | |
396bcc52 | 3938 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS) |
5a6e75f8 | 3939 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
79d4d38a | 3940 | struct kobj_attribute *attr, char *buf) |
5a6e75f8 | 3941 | { |
26083eb6 | 3942 | static const int values[] = { |
5a6e75f8 KS |
3943 | SHMEM_HUGE_ALWAYS, |
3944 | SHMEM_HUGE_WITHIN_SIZE, | |
3945 | SHMEM_HUGE_ADVISE, | |
3946 | SHMEM_HUGE_NEVER, | |
3947 | SHMEM_HUGE_DENY, | |
3948 | SHMEM_HUGE_FORCE, | |
3949 | }; | |
79d4d38a JP |
3950 | int len = 0; |
3951 | int i; | |
5a6e75f8 | 3952 | |
79d4d38a JP |
3953 | for (i = 0; i < ARRAY_SIZE(values); i++) { |
3954 | len += sysfs_emit_at(buf, len, | |
3955 | shmem_huge == values[i] ? "%s[%s]" : "%s%s", | |
3956 | i ? " " : "", | |
3957 | shmem_format_huge(values[i])); | |
5a6e75f8 | 3958 | } |
79d4d38a JP |
3959 | |
3960 | len += sysfs_emit_at(buf, len, "\n"); | |
3961 | ||
3962 | return len; | |
5a6e75f8 KS |
3963 | } |
3964 | ||
3965 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
3966 | struct kobj_attribute *attr, const char *buf, size_t count) | |
3967 | { | |
3968 | char tmp[16]; | |
3969 | int huge; | |
3970 | ||
3971 | if (count + 1 > sizeof(tmp)) | |
3972 | return -EINVAL; | |
3973 | memcpy(tmp, buf, count); | |
3974 | tmp[count] = '\0'; | |
3975 | if (count && tmp[count - 1] == '\n') | |
3976 | tmp[count - 1] = '\0'; | |
3977 | ||
3978 | huge = shmem_parse_huge(tmp); | |
3979 | if (huge == -EINVAL) | |
3980 | return -EINVAL; | |
3981 | if (!has_transparent_hugepage() && | |
3982 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
3983 | return -EINVAL; | |
3984 | ||
3985 | shmem_huge = huge; | |
435c0b87 | 3986 | if (shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
3987 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
3988 | return count; | |
3989 | } | |
3990 | ||
3991 | struct kobj_attribute shmem_enabled_attr = | |
3992 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
396bcc52 | 3993 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */ |
f3f0e1d2 | 3994 | |
853ac43a MM |
3995 | #else /* !CONFIG_SHMEM */ |
3996 | ||
3997 | /* | |
3998 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
3999 | * | |
4000 | * This is intended for small system where the benefits of the full | |
4001 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4002 | * their complexity. On systems without swap this code should be | |
4003 | * effectively equivalent, but much lighter weight. | |
4004 | */ | |
4005 | ||
41ffe5d5 | 4006 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4007 | .name = "tmpfs", |
f3235626 | 4008 | .init_fs_context = ramfs_init_fs_context, |
d7167b14 | 4009 | .parameters = ramfs_fs_parameters, |
853ac43a | 4010 | .kill_sb = kill_litter_super, |
2b8576cb | 4011 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4012 | }; |
4013 | ||
41ffe5d5 | 4014 | int __init shmem_init(void) |
853ac43a | 4015 | { |
41ffe5d5 | 4016 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4017 | |
41ffe5d5 | 4018 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
4019 | BUG_ON(IS_ERR(shm_mnt)); |
4020 | ||
4021 | return 0; | |
4022 | } | |
4023 | ||
b56a2d8a VRP |
4024 | int shmem_unuse(unsigned int type, bool frontswap, |
4025 | unsigned long *fs_pages_to_unuse) | |
853ac43a MM |
4026 | { |
4027 | return 0; | |
4028 | } | |
4029 | ||
d7c9e99a | 4030 | int shmem_lock(struct file *file, int lock, struct ucounts *ucounts) |
3f96b79a HD |
4031 | { |
4032 | return 0; | |
4033 | } | |
4034 | ||
24513264 HD |
4035 | void shmem_unlock_mapping(struct address_space *mapping) |
4036 | { | |
4037 | } | |
4038 | ||
c01d5b30 HD |
4039 | #ifdef CONFIG_MMU |
4040 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4041 | unsigned long addr, unsigned long len, | |
4042 | unsigned long pgoff, unsigned long flags) | |
4043 | { | |
4044 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4045 | } | |
4046 | #endif | |
4047 | ||
41ffe5d5 | 4048 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4049 | { |
41ffe5d5 | 4050 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4051 | } |
4052 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4053 | ||
0b0a0806 HD |
4054 | #define shmem_vm_ops generic_file_vm_ops |
4055 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 4056 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
4057 | #define shmem_acct_size(flags, size) 0 |
4058 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
4059 | |
4060 | #endif /* CONFIG_SHMEM */ | |
4061 | ||
4062 | /* common code */ | |
1da177e4 | 4063 | |
703321b6 | 4064 | static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size, |
c7277090 | 4065 | unsigned long flags, unsigned int i_flags) |
1da177e4 | 4066 | { |
1da177e4 | 4067 | struct inode *inode; |
93dec2da | 4068 | struct file *res; |
1da177e4 | 4069 | |
703321b6 MA |
4070 | if (IS_ERR(mnt)) |
4071 | return ERR_CAST(mnt); | |
1da177e4 | 4072 | |
285b2c4f | 4073 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4074 | return ERR_PTR(-EINVAL); |
4075 | ||
4076 | if (shmem_acct_size(flags, size)) | |
4077 | return ERR_PTR(-ENOMEM); | |
4078 | ||
93dec2da AV |
4079 | inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0, |
4080 | flags); | |
dac2d1f6 AV |
4081 | if (unlikely(!inode)) { |
4082 | shmem_unacct_size(flags, size); | |
4083 | return ERR_PTR(-ENOSPC); | |
4084 | } | |
c7277090 | 4085 | inode->i_flags |= i_flags; |
1da177e4 | 4086 | inode->i_size = size; |
6d6b77f1 | 4087 | clear_nlink(inode); /* It is unlinked */ |
26567cdb | 4088 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
93dec2da AV |
4089 | if (!IS_ERR(res)) |
4090 | res = alloc_file_pseudo(inode, mnt, name, O_RDWR, | |
4091 | &shmem_file_operations); | |
26567cdb | 4092 | if (IS_ERR(res)) |
93dec2da | 4093 | iput(inode); |
6b4d0b27 | 4094 | return res; |
1da177e4 | 4095 | } |
c7277090 EP |
4096 | |
4097 | /** | |
4098 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4099 | * kernel internal. There will be NO LSM permission checks against the | |
4100 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4101 | * higher layer. The users are the big_key and shm implementations. LSM |
4102 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4103 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4104 | * @size: size to be set for the file | |
4105 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4106 | */ | |
4107 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4108 | { | |
703321b6 | 4109 | return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE); |
c7277090 EP |
4110 | } |
4111 | ||
4112 | /** | |
4113 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4114 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4115 | * @size: size to be set for the file | |
4116 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4117 | */ | |
4118 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4119 | { | |
703321b6 | 4120 | return __shmem_file_setup(shm_mnt, name, size, flags, 0); |
c7277090 | 4121 | } |
395e0ddc | 4122 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4123 | |
703321b6 MA |
4124 | /** |
4125 | * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs | |
4126 | * @mnt: the tmpfs mount where the file will be created | |
4127 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4128 | * @size: size to be set for the file | |
4129 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4130 | */ | |
4131 | struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name, | |
4132 | loff_t size, unsigned long flags) | |
4133 | { | |
4134 | return __shmem_file_setup(mnt, name, size, flags, 0); | |
4135 | } | |
4136 | EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt); | |
4137 | ||
46711810 | 4138 | /** |
1da177e4 | 4139 | * shmem_zero_setup - setup a shared anonymous mapping |
45e55300 | 4140 | * @vma: the vma to be mmapped is prepared by do_mmap |
1da177e4 LT |
4141 | */ |
4142 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4143 | { | |
4144 | struct file *file; | |
4145 | loff_t size = vma->vm_end - vma->vm_start; | |
4146 | ||
66fc1303 | 4147 | /* |
c1e8d7c6 | 4148 | * Cloning a new file under mmap_lock leads to a lock ordering conflict |
66fc1303 HD |
4149 | * between XFS directory reading and selinux: since this file is only |
4150 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4151 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4152 | */ | |
703321b6 | 4153 | file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags); |
1da177e4 LT |
4154 | if (IS_ERR(file)) |
4155 | return PTR_ERR(file); | |
4156 | ||
4157 | if (vma->vm_file) | |
4158 | fput(vma->vm_file); | |
4159 | vma->vm_file = file; | |
4160 | vma->vm_ops = &shmem_vm_ops; | |
f3f0e1d2 | 4161 | |
396bcc52 | 4162 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && |
f3f0e1d2 KS |
4163 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
4164 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
4165 | khugepaged_enter(vma, vma->vm_flags); | |
4166 | } | |
4167 | ||
1da177e4 LT |
4168 | return 0; |
4169 | } | |
420a9fe5 | 4170 | EXPORT_SYMBOL_GPL(shmem_zero_setup); |
d9d90e5e HD |
4171 | |
4172 | /** | |
4173 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
4174 | * @mapping: the page's address_space | |
4175 | * @index: the page index | |
4176 | * @gfp: the page allocator flags to use if allocating | |
4177 | * | |
4178 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4179 | * with any new page allocations done using the specified allocation flags. | |
4180 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
4181 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
4182 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4183 | * | |
68da9f05 HD |
4184 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4185 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4186 | */ |
4187 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4188 | pgoff_t index, gfp_t gfp) | |
4189 | { | |
68da9f05 HD |
4190 | #ifdef CONFIG_SHMEM |
4191 | struct inode *inode = mapping->host; | |
9276aad6 | 4192 | struct page *page; |
68da9f05 HD |
4193 | int error; |
4194 | ||
30e6a51d | 4195 | BUG_ON(!shmem_mapping(mapping)); |
9e18eb29 | 4196 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
cfda0526 | 4197 | gfp, NULL, NULL, NULL); |
68da9f05 HD |
4198 | if (error) |
4199 | page = ERR_PTR(error); | |
4200 | else | |
4201 | unlock_page(page); | |
4202 | return page; | |
4203 | #else | |
4204 | /* | |
4205 | * The tiny !SHMEM case uses ramfs without swap | |
4206 | */ | |
d9d90e5e | 4207 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4208 | #endif |
d9d90e5e HD |
4209 | } |
4210 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |