]>
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 <[email protected]> | |
16 | * Copyright (c) 2004 Red Hat, Inc., James Morris <[email protected]> | |
17 | * | |
853ac43a MM |
18 | * tiny-shmem: |
19 | * Copyright (c) 2004, 2008 Matt Mackall <[email protected]> | |
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 | 30 | #include <linux/file.h> |
e408e695 | 31 | #include <linux/fileattr.h> |
853ac43a | 32 | #include <linux/mm.h> |
46c9a946 | 33 | #include <linux/random.h> |
174cd4b1 | 34 | #include <linux/sched/signal.h> |
b95f1b31 | 35 | #include <linux/export.h> |
5ff2121a | 36 | #include <linux/shmem_fs.h> |
853ac43a | 37 | #include <linux/swap.h> |
e2e40f2c | 38 | #include <linux/uio.h> |
749df87b | 39 | #include <linux/hugetlb.h> |
626c3920 | 40 | #include <linux/fs_parser.h> |
86a2f3f2 | 41 | #include <linux/swapfile.h> |
36f05cab | 42 | #include <linux/iversion.h> |
014bb1de | 43 | #include "swap.h" |
95cc09d6 | 44 | |
68279f9c | 45 | static struct vfsmount *shm_mnt __ro_after_init; |
853ac43a MM |
46 | |
47 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
48 | /* |
49 | * This virtual memory filesystem is heavily based on the ramfs. It | |
50 | * extends ramfs by the ability to use swap and honor resource limits | |
51 | * which makes it a completely usable filesystem. | |
52 | */ | |
53 | ||
39f0247d | 54 | #include <linux/xattr.h> |
a5694255 | 55 | #include <linux/exportfs.h> |
1c7c474c | 56 | #include <linux/posix_acl.h> |
feda821e | 57 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 58 | #include <linux/mman.h> |
1da177e4 LT |
59 | #include <linux/string.h> |
60 | #include <linux/slab.h> | |
61 | #include <linux/backing-dev.h> | |
1da177e4 | 62 | #include <linux/writeback.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> |
4c27fe4c | 79 | #include <linux/rmap.h> |
2b4db796 | 80 | #include <linux/uuid.h> |
e09764cf | 81 | #include <linux/quotaops.h> |
1e2f2d31 | 82 | #include <linux/rcupdate_wait.h> |
304dbdb7 | 83 | |
7c0f6ba6 | 84 | #include <linux/uaccess.h> |
1da177e4 | 85 | |
dd56b046 MG |
86 | #include "internal.h" |
87 | ||
09cbfeaf KS |
88 | #define BLOCKS_PER_PAGE (PAGE_SIZE/512) |
89 | #define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT) | |
1da177e4 | 90 | |
1da177e4 LT |
91 | /* Pretend that each entry is of this size in directory's i_size */ |
92 | #define BOGO_DIRENT_SIZE 20 | |
93 | ||
e07c469e HD |
94 | /* Pretend that one inode + its dentry occupy this much memory */ |
95 | #define BOGO_INODE_SIZE 1024 | |
96 | ||
69f07ec9 HD |
97 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
98 | #define SHORT_SYMLINK_LEN 128 | |
99 | ||
1aac1400 | 100 | /* |
f00cdc6d | 101 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
9608703e | 102 | * inode->i_private (with i_rwsem making sure that it has only one user at |
f00cdc6d | 103 | * a time): we would prefer not to enlarge the shmem inode just for that. |
1aac1400 HD |
104 | */ |
105 | struct shmem_falloc { | |
8e205f77 | 106 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
107 | pgoff_t start; /* start of range currently being fallocated */ |
108 | pgoff_t next; /* the next page offset to be fallocated */ | |
109 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
110 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
111 | }; | |
112 | ||
0b5071dd AV |
113 | struct shmem_options { |
114 | unsigned long long blocks; | |
115 | unsigned long long inodes; | |
116 | struct mempolicy *mpol; | |
117 | kuid_t uid; | |
118 | kgid_t gid; | |
119 | umode_t mode; | |
ea3271f7 | 120 | bool full_inums; |
0b5071dd AV |
121 | int huge; |
122 | int seen; | |
2c6efe9c | 123 | bool noswap; |
e09764cf | 124 | unsigned short quota_types; |
de4c0e7c | 125 | struct shmem_quota_limits qlimits; |
0b5071dd AV |
126 | #define SHMEM_SEEN_BLOCKS 1 |
127 | #define SHMEM_SEEN_INODES 2 | |
128 | #define SHMEM_SEEN_HUGE 4 | |
ea3271f7 | 129 | #define SHMEM_SEEN_INUMS 8 |
2c6efe9c | 130 | #define SHMEM_SEEN_NOSWAP 16 |
e09764cf | 131 | #define SHMEM_SEEN_QUOTA 32 |
0b5071dd AV |
132 | }; |
133 | ||
b76db735 | 134 | #ifdef CONFIG_TMPFS |
680d794b AM |
135 | static unsigned long shmem_default_max_blocks(void) |
136 | { | |
ca79b0c2 | 137 | return totalram_pages() / 2; |
680d794b AM |
138 | } |
139 | ||
140 | static unsigned long shmem_default_max_inodes(void) | |
141 | { | |
ca79b0c2 AK |
142 | unsigned long nr_pages = totalram_pages(); |
143 | ||
e07c469e HD |
144 | return min3(nr_pages - totalhigh_pages(), nr_pages / 2, |
145 | ULONG_MAX / BOGO_INODE_SIZE); | |
680d794b | 146 | } |
b76db735 | 147 | #endif |
680d794b | 148 | |
da08e9b7 | 149 | static int shmem_swapin_folio(struct inode *inode, pgoff_t index, |
054a9f7c HD |
150 | struct folio **foliop, enum sgp_type sgp, gfp_t gfp, |
151 | struct mm_struct *fault_mm, vm_fault_t *fault_type); | |
1da177e4 | 152 | |
1da177e4 LT |
153 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
154 | { | |
155 | return sb->s_fs_info; | |
156 | } | |
157 | ||
158 | /* | |
159 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
160 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
161 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
162 | * consistent with the pre-accounting of private mappings ... | |
163 | */ | |
164 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
165 | { | |
0b0a0806 | 166 | return (flags & VM_NORESERVE) ? |
191c5424 | 167 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
168 | } |
169 | ||
170 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
171 | { | |
0b0a0806 | 172 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
173 | vm_unacct_memory(VM_ACCT(size)); |
174 | } | |
175 | ||
77142517 KK |
176 | static inline int shmem_reacct_size(unsigned long flags, |
177 | loff_t oldsize, loff_t newsize) | |
178 | { | |
179 | if (!(flags & VM_NORESERVE)) { | |
180 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
181 | return security_vm_enough_memory_mm(current->mm, | |
182 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
183 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
184 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
185 | } | |
186 | return 0; | |
187 | } | |
188 | ||
1da177e4 LT |
189 | /* |
190 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 191 | * pages are allocated, in order to allow large sparse files. |
4199f51a | 192 | * shmem_get_folio reports shmem_acct_blocks failure as -ENOSPC not -ENOMEM, |
1da177e4 LT |
193 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. |
194 | */ | |
4199f51a | 195 | static inline int shmem_acct_blocks(unsigned long flags, long pages) |
1da177e4 | 196 | { |
800d8c63 KS |
197 | if (!(flags & VM_NORESERVE)) |
198 | return 0; | |
199 | ||
200 | return security_vm_enough_memory_mm(current->mm, | |
201 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
202 | } |
203 | ||
204 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
205 | { | |
0b0a0806 | 206 | if (flags & VM_NORESERVE) |
09cbfeaf | 207 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
208 | } |
209 | ||
4199f51a | 210 | static int shmem_inode_acct_blocks(struct inode *inode, long pages) |
0f079694 MR |
211 | { |
212 | struct shmem_inode_info *info = SHMEM_I(inode); | |
213 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
c7e263ab | 214 | int err = -ENOSPC; |
0f079694 | 215 | |
4199f51a | 216 | if (shmem_acct_blocks(info->flags, pages)) |
c7e263ab | 217 | return err; |
0f079694 | 218 | |
3c1b7528 | 219 | might_sleep(); /* when quotas */ |
0f079694 | 220 | if (sbinfo->max_blocks) { |
beb98686 HD |
221 | if (!percpu_counter_limited_add(&sbinfo->used_blocks, |
222 | sbinfo->max_blocks, pages)) | |
0f079694 | 223 | goto unacct; |
e09764cf CM |
224 | |
225 | err = dquot_alloc_block_nodirty(inode, pages); | |
beb98686 HD |
226 | if (err) { |
227 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
e09764cf | 228 | goto unacct; |
beb98686 | 229 | } |
e09764cf CM |
230 | } else { |
231 | err = dquot_alloc_block_nodirty(inode, pages); | |
232 | if (err) | |
233 | goto unacct; | |
0f079694 MR |
234 | } |
235 | ||
c7e263ab | 236 | return 0; |
0f079694 MR |
237 | |
238 | unacct: | |
239 | shmem_unacct_blocks(info->flags, pages); | |
c7e263ab | 240 | return err; |
0f079694 MR |
241 | } |
242 | ||
3c1b7528 | 243 | static void shmem_inode_unacct_blocks(struct inode *inode, long pages) |
0f079694 MR |
244 | { |
245 | struct shmem_inode_info *info = SHMEM_I(inode); | |
246 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
247 | ||
3c1b7528 | 248 | might_sleep(); /* when quotas */ |
e09764cf CM |
249 | dquot_free_block_nodirty(inode, pages); |
250 | ||
0f079694 MR |
251 | if (sbinfo->max_blocks) |
252 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
253 | shmem_unacct_blocks(info->flags, pages); | |
254 | } | |
255 | ||
759b9775 | 256 | static const struct super_operations shmem_ops; |
aefacb20 | 257 | static const struct address_space_operations shmem_aops; |
15ad7cdc | 258 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
259 | static const struct inode_operations shmem_inode_operations; |
260 | static const struct inode_operations shmem_dir_inode_operations; | |
261 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 262 | static const struct vm_operations_struct shmem_vm_ops; |
d09e8ca6 | 263 | static const struct vm_operations_struct shmem_anon_vm_ops; |
779750d2 | 264 | static struct file_system_type shmem_fs_type; |
1da177e4 | 265 | |
aefacb20 CH |
266 | bool shmem_mapping(struct address_space *mapping) |
267 | { | |
268 | return mapping->a_ops == &shmem_aops; | |
269 | } | |
270 | EXPORT_SYMBOL_GPL(shmem_mapping); | |
271 | ||
d09e8ca6 PT |
272 | bool vma_is_anon_shmem(struct vm_area_struct *vma) |
273 | { | |
274 | return vma->vm_ops == &shmem_anon_vm_ops; | |
275 | } | |
276 | ||
b0506e48 MR |
277 | bool vma_is_shmem(struct vm_area_struct *vma) |
278 | { | |
d09e8ca6 | 279 | return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops; |
b0506e48 MR |
280 | } |
281 | ||
1da177e4 | 282 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 283 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 284 | |
e09764cf CM |
285 | #ifdef CONFIG_TMPFS_QUOTA |
286 | ||
287 | static int shmem_enable_quotas(struct super_block *sb, | |
288 | unsigned short quota_types) | |
289 | { | |
290 | int type, err = 0; | |
291 | ||
292 | sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; | |
293 | for (type = 0; type < SHMEM_MAXQUOTAS; type++) { | |
294 | if (!(quota_types & (1 << type))) | |
295 | continue; | |
296 | err = dquot_load_quota_sb(sb, type, QFMT_SHMEM, | |
297 | DQUOT_USAGE_ENABLED | | |
298 | DQUOT_LIMITS_ENABLED); | |
299 | if (err) | |
300 | goto out_err; | |
301 | } | |
302 | return 0; | |
303 | ||
304 | out_err: | |
305 | pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n", | |
306 | type, err); | |
307 | for (type--; type >= 0; type--) | |
308 | dquot_quota_off(sb, type); | |
309 | return err; | |
310 | } | |
311 | ||
312 | static void shmem_disable_quotas(struct super_block *sb) | |
313 | { | |
314 | int type; | |
315 | ||
316 | for (type = 0; type < SHMEM_MAXQUOTAS; type++) | |
317 | dquot_quota_off(sb, type); | |
318 | } | |
319 | ||
ccb49011 | 320 | static struct dquot __rcu **shmem_get_dquots(struct inode *inode) |
e09764cf CM |
321 | { |
322 | return SHMEM_I(inode)->i_dquot; | |
323 | } | |
324 | #endif /* CONFIG_TMPFS_QUOTA */ | |
325 | ||
e809d5f0 CD |
326 | /* |
327 | * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and | |
328 | * produces a novel ino for the newly allocated inode. | |
329 | * | |
330 | * It may also be called when making a hard link to permit the space needed by | |
331 | * each dentry. However, in that case, no new inode number is needed since that | |
332 | * internally draws from another pool of inode numbers (currently global | |
333 | * get_next_ino()). This case is indicated by passing NULL as inop. | |
334 | */ | |
335 | #define SHMEM_INO_BATCH 1024 | |
336 | static int shmem_reserve_inode(struct super_block *sb, ino_t *inop) | |
5b04c689 PE |
337 | { |
338 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
e809d5f0 CD |
339 | ino_t ino; |
340 | ||
341 | if (!(sb->s_flags & SB_KERNMOUNT)) { | |
bf11b9a8 | 342 | raw_spin_lock(&sbinfo->stat_lock); |
bb3e96d6 | 343 | if (sbinfo->max_inodes) { |
e07c469e | 344 | if (sbinfo->free_ispace < BOGO_INODE_SIZE) { |
bf11b9a8 | 345 | raw_spin_unlock(&sbinfo->stat_lock); |
bb3e96d6 BS |
346 | return -ENOSPC; |
347 | } | |
e07c469e | 348 | sbinfo->free_ispace -= BOGO_INODE_SIZE; |
5b04c689 | 349 | } |
e809d5f0 CD |
350 | if (inop) { |
351 | ino = sbinfo->next_ino++; | |
352 | if (unlikely(is_zero_ino(ino))) | |
353 | ino = sbinfo->next_ino++; | |
ea3271f7 CD |
354 | if (unlikely(!sbinfo->full_inums && |
355 | ino > UINT_MAX)) { | |
e809d5f0 CD |
356 | /* |
357 | * Emulate get_next_ino uint wraparound for | |
358 | * compatibility | |
359 | */ | |
ea3271f7 CD |
360 | if (IS_ENABLED(CONFIG_64BIT)) |
361 | pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n", | |
362 | __func__, MINOR(sb->s_dev)); | |
363 | sbinfo->next_ino = 1; | |
364 | ino = sbinfo->next_ino++; | |
e809d5f0 CD |
365 | } |
366 | *inop = ino; | |
367 | } | |
bf11b9a8 | 368 | raw_spin_unlock(&sbinfo->stat_lock); |
e809d5f0 CD |
369 | } else if (inop) { |
370 | /* | |
371 | * __shmem_file_setup, one of our callers, is lock-free: it | |
372 | * doesn't hold stat_lock in shmem_reserve_inode since | |
373 | * max_inodes is always 0, and is called from potentially | |
374 | * unknown contexts. As such, use a per-cpu batched allocator | |
375 | * which doesn't require the per-sb stat_lock unless we are at | |
376 | * the batch boundary. | |
ea3271f7 CD |
377 | * |
378 | * We don't need to worry about inode{32,64} since SB_KERNMOUNT | |
379 | * shmem mounts are not exposed to userspace, so we don't need | |
380 | * to worry about things like glibc compatibility. | |
e809d5f0 CD |
381 | */ |
382 | ino_t *next_ino; | |
bf11b9a8 | 383 | |
e809d5f0 CD |
384 | next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu()); |
385 | ino = *next_ino; | |
386 | if (unlikely(ino % SHMEM_INO_BATCH == 0)) { | |
bf11b9a8 | 387 | raw_spin_lock(&sbinfo->stat_lock); |
e809d5f0 CD |
388 | ino = sbinfo->next_ino; |
389 | sbinfo->next_ino += SHMEM_INO_BATCH; | |
bf11b9a8 | 390 | raw_spin_unlock(&sbinfo->stat_lock); |
e809d5f0 CD |
391 | if (unlikely(is_zero_ino(ino))) |
392 | ino++; | |
393 | } | |
394 | *inop = ino; | |
395 | *next_ino = ++ino; | |
396 | put_cpu(); | |
5b04c689 | 397 | } |
e809d5f0 | 398 | |
5b04c689 PE |
399 | return 0; |
400 | } | |
401 | ||
2daf18a7 | 402 | static void shmem_free_inode(struct super_block *sb, size_t freed_ispace) |
5b04c689 PE |
403 | { |
404 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
405 | if (sbinfo->max_inodes) { | |
bf11b9a8 | 406 | raw_spin_lock(&sbinfo->stat_lock); |
2daf18a7 | 407 | sbinfo->free_ispace += BOGO_INODE_SIZE + freed_ispace; |
bf11b9a8 | 408 | raw_spin_unlock(&sbinfo->stat_lock); |
5b04c689 PE |
409 | } |
410 | } | |
411 | ||
46711810 | 412 | /** |
41ffe5d5 | 413 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 | 414 | * @inode: inode to recalc |
3c1b7528 HD |
415 | * @alloced: the change in number of pages allocated to inode |
416 | * @swapped: the change in number of pages swapped from inode | |
1da177e4 LT |
417 | * |
418 | * We have to calculate the free blocks since the mm can drop | |
419 | * undirtied hole pages behind our back. | |
420 | * | |
421 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
422 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
1da177e4 | 423 | */ |
3c1b7528 | 424 | static void shmem_recalc_inode(struct inode *inode, long alloced, long swapped) |
1da177e4 LT |
425 | { |
426 | struct shmem_inode_info *info = SHMEM_I(inode); | |
427 | long freed; | |
428 | ||
3c1b7528 HD |
429 | spin_lock(&info->lock); |
430 | info->alloced += alloced; | |
431 | info->swapped += swapped; | |
432 | freed = info->alloced - info->swapped - | |
433 | READ_ONCE(inode->i_mapping->nrpages); | |
434 | /* | |
435 | * Special case: whereas normally shmem_recalc_inode() is called | |
436 | * after i_mapping->nrpages has already been adjusted (up or down), | |
437 | * shmem_writepage() has to raise swapped before nrpages is lowered - | |
438 | * to stop a racing shmem_recalc_inode() from thinking that a page has | |
439 | * been freed. Compensate here, to avoid the need for a followup call. | |
440 | */ | |
441 | if (swapped > 0) | |
442 | freed += swapped; | |
443 | if (freed > 0) | |
1da177e4 | 444 | info->alloced -= freed; |
3c1b7528 HD |
445 | spin_unlock(&info->lock); |
446 | ||
447 | /* The quota case may block */ | |
448 | if (freed > 0) | |
0f079694 | 449 | shmem_inode_unacct_blocks(inode, freed); |
1da177e4 LT |
450 | } |
451 | ||
800d8c63 KS |
452 | bool shmem_charge(struct inode *inode, long pages) |
453 | { | |
509f0069 | 454 | struct address_space *mapping = inode->i_mapping; |
800d8c63 | 455 | |
4199f51a | 456 | if (shmem_inode_acct_blocks(inode, pages)) |
800d8c63 | 457 | return false; |
b1cc94ab | 458 | |
aaa52e34 | 459 | /* nrpages adjustment first, then shmem_recalc_inode() when balanced */ |
509f0069 HD |
460 | xa_lock_irq(&mapping->i_pages); |
461 | mapping->nrpages += pages; | |
462 | xa_unlock_irq(&mapping->i_pages); | |
800d8c63 | 463 | |
3c1b7528 | 464 | shmem_recalc_inode(inode, pages, 0); |
800d8c63 KS |
465 | return true; |
466 | } | |
467 | ||
468 | void shmem_uncharge(struct inode *inode, long pages) | |
469 | { | |
3c1b7528 | 470 | /* pages argument is currently unused: keep it to help debugging */ |
6ffcd825 | 471 | /* nrpages adjustment done by __filemap_remove_folio() or caller */ |
aaa52e34 | 472 | |
3c1b7528 | 473 | shmem_recalc_inode(inode, 0, 0); |
800d8c63 KS |
474 | } |
475 | ||
7a5d0fbb | 476 | /* |
62f945b6 | 477 | * Replace item expected in xarray by a new item, while holding xa_lock. |
7a5d0fbb | 478 | */ |
62f945b6 | 479 | static int shmem_replace_entry(struct address_space *mapping, |
7a5d0fbb HD |
480 | pgoff_t index, void *expected, void *replacement) |
481 | { | |
62f945b6 | 482 | XA_STATE(xas, &mapping->i_pages, index); |
6dbaf22c | 483 | void *item; |
7a5d0fbb HD |
484 | |
485 | VM_BUG_ON(!expected); | |
6dbaf22c | 486 | VM_BUG_ON(!replacement); |
62f945b6 | 487 | item = xas_load(&xas); |
7a5d0fbb HD |
488 | if (item != expected) |
489 | return -ENOENT; | |
62f945b6 | 490 | xas_store(&xas, replacement); |
7a5d0fbb HD |
491 | return 0; |
492 | } | |
493 | ||
d1899228 HD |
494 | /* |
495 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
496 | * that an entry was not already brought back from swap by a racing thread. | |
497 | * | |
498 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
499 | * might be reused, and again be SwapCache, using the same swap as before. | |
500 | */ | |
501 | static bool shmem_confirm_swap(struct address_space *mapping, | |
502 | pgoff_t index, swp_entry_t swap) | |
503 | { | |
a12831bf | 504 | return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap); |
d1899228 HD |
505 | } |
506 | ||
5a6e75f8 KS |
507 | /* |
508 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
509 | * | |
510 | * SHMEM_HUGE_NEVER: | |
511 | * disables huge pages for the mount; | |
512 | * SHMEM_HUGE_ALWAYS: | |
513 | * enables huge pages for the mount; | |
514 | * SHMEM_HUGE_WITHIN_SIZE: | |
515 | * only allocate huge pages if the page will be fully within i_size, | |
516 | * also respect fadvise()/madvise() hints; | |
517 | * SHMEM_HUGE_ADVISE: | |
518 | * only allocate huge pages if requested with fadvise()/madvise(); | |
519 | */ | |
520 | ||
521 | #define SHMEM_HUGE_NEVER 0 | |
522 | #define SHMEM_HUGE_ALWAYS 1 | |
523 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
524 | #define SHMEM_HUGE_ADVISE 3 | |
525 | ||
526 | /* | |
527 | * Special values. | |
528 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
529 | * | |
530 | * SHMEM_HUGE_DENY: | |
531 | * disables huge on shm_mnt and all mounts, for emergency use; | |
532 | * SHMEM_HUGE_FORCE: | |
533 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
534 | * | |
535 | */ | |
536 | #define SHMEM_HUGE_DENY (-1) | |
537 | #define SHMEM_HUGE_FORCE (-2) | |
538 | ||
396bcc52 | 539 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5a6e75f8 KS |
540 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
541 | ||
5e6e5a12 | 542 | static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER; |
5a6e75f8 | 543 | |
2cf13384 DS |
544 | bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force, |
545 | struct mm_struct *mm, unsigned long vm_flags) | |
c852023e | 546 | { |
c852023e | 547 | loff_t i_size; |
c852023e | 548 | |
f7cd16a5 XR |
549 | if (!S_ISREG(inode->i_mode)) |
550 | return false; | |
2cf13384 | 551 | if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags))) |
c852023e | 552 | return false; |
7c6c6cc4 ZK |
553 | if (shmem_huge == SHMEM_HUGE_DENY) |
554 | return false; | |
3de0c269 ZK |
555 | if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE) |
556 | return true; | |
5e6e5a12 HD |
557 | |
558 | switch (SHMEM_SB(inode->i_sb)->huge) { | |
c852023e HD |
559 | case SHMEM_HUGE_ALWAYS: |
560 | return true; | |
561 | case SHMEM_HUGE_WITHIN_SIZE: | |
de6ee659 | 562 | index = round_up(index + 1, HPAGE_PMD_NR); |
c852023e | 563 | i_size = round_up(i_size_read(inode), PAGE_SIZE); |
de6ee659 | 564 | if (i_size >> PAGE_SHIFT >= index) |
c852023e HD |
565 | return true; |
566 | fallthrough; | |
567 | case SHMEM_HUGE_ADVISE: | |
2cf13384 | 568 | if (mm && (vm_flags & VM_HUGEPAGE)) |
5e6e5a12 HD |
569 | return true; |
570 | fallthrough; | |
c852023e | 571 | default: |
c852023e HD |
572 | return false; |
573 | } | |
574 | } | |
5a6e75f8 | 575 | |
e5f2249a | 576 | #if defined(CONFIG_SYSFS) |
5a6e75f8 KS |
577 | static int shmem_parse_huge(const char *str) |
578 | { | |
579 | if (!strcmp(str, "never")) | |
580 | return SHMEM_HUGE_NEVER; | |
581 | if (!strcmp(str, "always")) | |
582 | return SHMEM_HUGE_ALWAYS; | |
583 | if (!strcmp(str, "within_size")) | |
584 | return SHMEM_HUGE_WITHIN_SIZE; | |
585 | if (!strcmp(str, "advise")) | |
586 | return SHMEM_HUGE_ADVISE; | |
587 | if (!strcmp(str, "deny")) | |
588 | return SHMEM_HUGE_DENY; | |
589 | if (!strcmp(str, "force")) | |
590 | return SHMEM_HUGE_FORCE; | |
591 | return -EINVAL; | |
592 | } | |
e5f2249a | 593 | #endif |
5a6e75f8 | 594 | |
e5f2249a | 595 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
596 | static const char *shmem_format_huge(int huge) |
597 | { | |
598 | switch (huge) { | |
599 | case SHMEM_HUGE_NEVER: | |
600 | return "never"; | |
601 | case SHMEM_HUGE_ALWAYS: | |
602 | return "always"; | |
603 | case SHMEM_HUGE_WITHIN_SIZE: | |
604 | return "within_size"; | |
605 | case SHMEM_HUGE_ADVISE: | |
606 | return "advise"; | |
607 | case SHMEM_HUGE_DENY: | |
608 | return "deny"; | |
609 | case SHMEM_HUGE_FORCE: | |
610 | return "force"; | |
611 | default: | |
612 | VM_BUG_ON(1); | |
613 | return "bad_val"; | |
614 | } | |
615 | } | |
f1f5929c | 616 | #endif |
5a6e75f8 | 617 | |
779750d2 KS |
618 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
619 | struct shrink_control *sc, unsigned long nr_to_split) | |
620 | { | |
621 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 622 | LIST_HEAD(to_remove); |
779750d2 KS |
623 | struct inode *inode; |
624 | struct shmem_inode_info *info; | |
05624571 | 625 | struct folio *folio; |
779750d2 | 626 | unsigned long batch = sc ? sc->nr_to_scan : 128; |
62c9827c | 627 | int split = 0; |
779750d2 KS |
628 | |
629 | if (list_empty(&sbinfo->shrinklist)) | |
630 | return SHRINK_STOP; | |
631 | ||
632 | spin_lock(&sbinfo->shrinklist_lock); | |
633 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
634 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
635 | ||
636 | /* pin the inode */ | |
637 | inode = igrab(&info->vfs_inode); | |
638 | ||
639 | /* inode is about to be evicted */ | |
640 | if (!inode) { | |
641 | list_del_init(&info->shrinklist); | |
779750d2 KS |
642 | goto next; |
643 | } | |
644 | ||
645 | /* Check if there's anything to gain */ | |
646 | if (round_up(inode->i_size, PAGE_SIZE) == | |
647 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 648 | list_move(&info->shrinklist, &to_remove); |
779750d2 KS |
649 | goto next; |
650 | } | |
651 | ||
652 | list_move(&info->shrinklist, &list); | |
653 | next: | |
62c9827c | 654 | sbinfo->shrinklist_len--; |
779750d2 KS |
655 | if (!--batch) |
656 | break; | |
657 | } | |
658 | spin_unlock(&sbinfo->shrinklist_lock); | |
659 | ||
253fd0f0 KS |
660 | list_for_each_safe(pos, next, &to_remove) { |
661 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
662 | inode = &info->vfs_inode; | |
663 | list_del_init(&info->shrinklist); | |
664 | iput(inode); | |
665 | } | |
666 | ||
779750d2 KS |
667 | list_for_each_safe(pos, next, &list) { |
668 | int ret; | |
05624571 | 669 | pgoff_t index; |
779750d2 KS |
670 | |
671 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
672 | inode = &info->vfs_inode; | |
673 | ||
b3cd54b2 | 674 | if (nr_to_split && split >= nr_to_split) |
62c9827c | 675 | goto move_back; |
779750d2 | 676 | |
05624571 MWO |
677 | index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT; |
678 | folio = filemap_get_folio(inode->i_mapping, index); | |
66dabbb6 | 679 | if (IS_ERR(folio)) |
779750d2 KS |
680 | goto drop; |
681 | ||
b3cd54b2 | 682 | /* No huge page at the end of the file: nothing to split */ |
05624571 MWO |
683 | if (!folio_test_large(folio)) { |
684 | folio_put(folio); | |
779750d2 KS |
685 | goto drop; |
686 | } | |
687 | ||
b3cd54b2 | 688 | /* |
62c9827c GL |
689 | * Move the inode on the list back to shrinklist if we failed |
690 | * to lock the page at this time. | |
b3cd54b2 KS |
691 | * |
692 | * Waiting for the lock may lead to deadlock in the | |
693 | * reclaim path. | |
694 | */ | |
05624571 MWO |
695 | if (!folio_trylock(folio)) { |
696 | folio_put(folio); | |
62c9827c | 697 | goto move_back; |
b3cd54b2 KS |
698 | } |
699 | ||
d788f5b3 | 700 | ret = split_folio(folio); |
05624571 MWO |
701 | folio_unlock(folio); |
702 | folio_put(folio); | |
779750d2 | 703 | |
62c9827c | 704 | /* If split failed move the inode on the list back to shrinklist */ |
b3cd54b2 | 705 | if (ret) |
62c9827c | 706 | goto move_back; |
779750d2 KS |
707 | |
708 | split++; | |
709 | drop: | |
710 | list_del_init(&info->shrinklist); | |
62c9827c GL |
711 | goto put; |
712 | move_back: | |
713 | /* | |
714 | * Make sure the inode is either on the global list or deleted | |
715 | * from any local list before iput() since it could be deleted | |
716 | * in another thread once we put the inode (then the local list | |
717 | * is corrupted). | |
718 | */ | |
719 | spin_lock(&sbinfo->shrinklist_lock); | |
720 | list_move(&info->shrinklist, &sbinfo->shrinklist); | |
721 | sbinfo->shrinklist_len++; | |
722 | spin_unlock(&sbinfo->shrinklist_lock); | |
723 | put: | |
779750d2 KS |
724 | iput(inode); |
725 | } | |
726 | ||
779750d2 KS |
727 | return split; |
728 | } | |
729 | ||
730 | static long shmem_unused_huge_scan(struct super_block *sb, | |
731 | struct shrink_control *sc) | |
732 | { | |
733 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
734 | ||
735 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
736 | return SHRINK_STOP; | |
737 | ||
738 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
739 | } | |
740 | ||
741 | static long shmem_unused_huge_count(struct super_block *sb, | |
742 | struct shrink_control *sc) | |
743 | { | |
744 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
745 | return READ_ONCE(sbinfo->shrinklist_len); | |
746 | } | |
396bcc52 | 747 | #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ |
5a6e75f8 KS |
748 | |
749 | #define shmem_huge SHMEM_HUGE_DENY | |
750 | ||
779750d2 KS |
751 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
752 | struct shrink_control *sc, unsigned long nr_to_split) | |
753 | { | |
754 | return 0; | |
755 | } | |
396bcc52 | 756 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
5a6e75f8 | 757 | |
46f65ec1 | 758 | /* |
9be7d5b0 | 759 | * Somewhat like filemap_add_folio, but error if expected item has gone. |
46f65ec1 | 760 | */ |
b7dd44a1 | 761 | static int shmem_add_to_page_cache(struct folio *folio, |
46f65ec1 | 762 | struct address_space *mapping, |
054a9f7c | 763 | pgoff_t index, void *expected, gfp_t gfp) |
46f65ec1 | 764 | { |
b7dd44a1 MWO |
765 | XA_STATE_ORDER(xas, &mapping->i_pages, index, folio_order(folio)); |
766 | long nr = folio_nr_pages(folio); | |
46f65ec1 | 767 | |
b7dd44a1 MWO |
768 | VM_BUG_ON_FOLIO(index != round_down(index, nr), folio); |
769 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); | |
770 | VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio); | |
771 | VM_BUG_ON(expected && folio_test_large(folio)); | |
46f65ec1 | 772 | |
b7dd44a1 MWO |
773 | folio_ref_add(folio, nr); |
774 | folio->mapping = mapping; | |
775 | folio->index = index; | |
b065b432 | 776 | |
054a9f7c | 777 | gfp &= GFP_RECLAIM_MASK; |
b7dd44a1 | 778 | folio_throttle_swaprate(folio, gfp); |
3fea5a49 | 779 | |
552446a4 | 780 | do { |
552446a4 | 781 | xas_lock_irq(&xas); |
6b24ca4a MWO |
782 | if (expected != xas_find_conflict(&xas)) { |
783 | xas_set_err(&xas, -EEXIST); | |
784 | goto unlock; | |
785 | } | |
786 | if (expected && xas_find_conflict(&xas)) { | |
552446a4 | 787 | xas_set_err(&xas, -EEXIST); |
552446a4 | 788 | goto unlock; |
800d8c63 | 789 | } |
b7dd44a1 | 790 | xas_store(&xas, folio); |
6b24ca4a MWO |
791 | if (xas_error(&xas)) |
792 | goto unlock; | |
3022fd7a | 793 | if (folio_test_pmd_mappable(folio)) |
b7dd44a1 | 794 | __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, nr); |
b7dd44a1 MWO |
795 | __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr); |
796 | __lruvec_stat_mod_folio(folio, NR_SHMEM, nr); | |
3022fd7a | 797 | mapping->nrpages += nr; |
552446a4 MW |
798 | unlock: |
799 | xas_unlock_irq(&xas); | |
800 | } while (xas_nomem(&xas, gfp)); | |
801 | ||
802 | if (xas_error(&xas)) { | |
054a9f7c HD |
803 | folio->mapping = NULL; |
804 | folio_ref_sub(folio, nr); | |
805 | return xas_error(&xas); | |
46f65ec1 | 806 | } |
552446a4 MW |
807 | |
808 | return 0; | |
46f65ec1 HD |
809 | } |
810 | ||
6922c0c7 | 811 | /* |
9be7d5b0 | 812 | * Somewhat like filemap_remove_folio, but substitutes swap for @folio. |
6922c0c7 | 813 | */ |
4cd400fd | 814 | static void shmem_delete_from_page_cache(struct folio *folio, void *radswap) |
6922c0c7 | 815 | { |
4cd400fd MWO |
816 | struct address_space *mapping = folio->mapping; |
817 | long nr = folio_nr_pages(folio); | |
6922c0c7 HD |
818 | int error; |
819 | ||
b93b0163 | 820 | xa_lock_irq(&mapping->i_pages); |
4cd400fd MWO |
821 | error = shmem_replace_entry(mapping, folio->index, folio, radswap); |
822 | folio->mapping = NULL; | |
823 | mapping->nrpages -= nr; | |
824 | __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr); | |
825 | __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr); | |
b93b0163 | 826 | xa_unlock_irq(&mapping->i_pages); |
4cd400fd | 827 | folio_put(folio); |
6922c0c7 HD |
828 | BUG_ON(error); |
829 | } | |
830 | ||
7a5d0fbb | 831 | /* |
c121d3bb | 832 | * Remove swap entry from page cache, free the swap and its page cache. |
7a5d0fbb HD |
833 | */ |
834 | static int shmem_free_swap(struct address_space *mapping, | |
835 | pgoff_t index, void *radswap) | |
836 | { | |
6dbaf22c | 837 | void *old; |
7a5d0fbb | 838 | |
55f3f7ea | 839 | old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0); |
6dbaf22c JW |
840 | if (old != radswap) |
841 | return -ENOENT; | |
842 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
843 | return 0; | |
7a5d0fbb HD |
844 | } |
845 | ||
6a15a370 VB |
846 | /* |
847 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 848 | * given offsets are swapped out. |
6a15a370 | 849 | * |
9608703e | 850 | * This is safe to call without i_rwsem or the i_pages lock thanks to RCU, |
6a15a370 VB |
851 | * as long as the inode doesn't go away and racy results are not a problem. |
852 | */ | |
48131e03 VB |
853 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
854 | pgoff_t start, pgoff_t end) | |
6a15a370 | 855 | { |
7ae3424f | 856 | XA_STATE(xas, &mapping->i_pages, start); |
6a15a370 | 857 | struct page *page; |
48131e03 | 858 | unsigned long swapped = 0; |
e5548f85 | 859 | unsigned long max = end - 1; |
6a15a370 VB |
860 | |
861 | rcu_read_lock(); | |
e5548f85 | 862 | xas_for_each(&xas, page, max) { |
7ae3424f | 863 | if (xas_retry(&xas, page)) |
2cf938aa | 864 | continue; |
3159f943 | 865 | if (xa_is_value(page)) |
6a15a370 | 866 | swapped++; |
e5548f85 HD |
867 | if (xas.xa_index == max) |
868 | break; | |
6a15a370 | 869 | if (need_resched()) { |
7ae3424f | 870 | xas_pause(&xas); |
6a15a370 | 871 | cond_resched_rcu(); |
6a15a370 VB |
872 | } |
873 | } | |
6a15a370 VB |
874 | rcu_read_unlock(); |
875 | ||
876 | return swapped << PAGE_SHIFT; | |
877 | } | |
878 | ||
48131e03 VB |
879 | /* |
880 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
881 | * given vma is swapped out. | |
882 | * | |
9608703e | 883 | * This is safe to call without i_rwsem or the i_pages lock thanks to RCU, |
48131e03 VB |
884 | * as long as the inode doesn't go away and racy results are not a problem. |
885 | */ | |
886 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
887 | { | |
888 | struct inode *inode = file_inode(vma->vm_file); | |
889 | struct shmem_inode_info *info = SHMEM_I(inode); | |
890 | struct address_space *mapping = inode->i_mapping; | |
891 | unsigned long swapped; | |
892 | ||
893 | /* Be careful as we don't hold info->lock */ | |
894 | swapped = READ_ONCE(info->swapped); | |
895 | ||
896 | /* | |
897 | * The easier cases are when the shmem object has nothing in swap, or | |
898 | * the vma maps it whole. Then we can simply use the stats that we | |
899 | * already track. | |
900 | */ | |
901 | if (!swapped) | |
902 | return 0; | |
903 | ||
904 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
905 | return swapped << PAGE_SHIFT; | |
906 | ||
907 | /* Here comes the more involved part */ | |
02399c88 PX |
908 | return shmem_partial_swap_usage(mapping, vma->vm_pgoff, |
909 | vma->vm_pgoff + vma_pages(vma)); | |
48131e03 VB |
910 | } |
911 | ||
24513264 HD |
912 | /* |
913 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
914 | */ | |
915 | void shmem_unlock_mapping(struct address_space *mapping) | |
916 | { | |
105c988f | 917 | struct folio_batch fbatch; |
24513264 HD |
918 | pgoff_t index = 0; |
919 | ||
105c988f | 920 | folio_batch_init(&fbatch); |
24513264 HD |
921 | /* |
922 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
923 | */ | |
105c988f MWO |
924 | while (!mapping_unevictable(mapping) && |
925 | filemap_get_folios(mapping, &index, ~0UL, &fbatch)) { | |
926 | check_move_unevictable_folios(&fbatch); | |
927 | folio_batch_release(&fbatch); | |
24513264 HD |
928 | cond_resched(); |
929 | } | |
7a5d0fbb HD |
930 | } |
931 | ||
b9a8a419 | 932 | static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index) |
71725ed1 | 933 | { |
b9a8a419 | 934 | struct folio *folio; |
71725ed1 | 935 | |
b9a8a419 | 936 | /* |
a7f5862c | 937 | * At first avoid shmem_get_folio(,,,SGP_READ): that fails |
81914aff | 938 | * beyond i_size, and reports fallocated folios as holes. |
b9a8a419 | 939 | */ |
81914aff HD |
940 | folio = filemap_get_entry(inode->i_mapping, index); |
941 | if (!folio) | |
b9a8a419 | 942 | return folio; |
81914aff HD |
943 | if (!xa_is_value(folio)) { |
944 | folio_lock(folio); | |
945 | if (folio->mapping == inode->i_mapping) | |
946 | return folio; | |
947 | /* The folio has been swapped out */ | |
948 | folio_unlock(folio); | |
949 | folio_put(folio); | |
950 | } | |
b9a8a419 | 951 | /* |
81914aff | 952 | * But read a folio back from swap if any of it is within i_size |
b9a8a419 MWO |
953 | * (although in some cases this is just a waste of time). |
954 | */ | |
a7f5862c MWO |
955 | folio = NULL; |
956 | shmem_get_folio(inode, index, &folio, SGP_READ); | |
957 | return folio; | |
71725ed1 HD |
958 | } |
959 | ||
7a5d0fbb | 960 | /* |
7f4446ee | 961 | * Remove range of pages and swap entries from page cache, and free them. |
1635f6a7 | 962 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 963 | */ |
1635f6a7 HD |
964 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
965 | bool unfalloc) | |
1da177e4 | 966 | { |
285b2c4f | 967 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 968 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
969 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
970 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
0e499ed3 | 971 | struct folio_batch fbatch; |
7a5d0fbb | 972 | pgoff_t indices[PAGEVEC_SIZE]; |
b9a8a419 MWO |
973 | struct folio *folio; |
974 | bool same_folio; | |
7a5d0fbb | 975 | long nr_swaps_freed = 0; |
285b2c4f | 976 | pgoff_t index; |
bda97eab HD |
977 | int i; |
978 | ||
83e4fa9c HD |
979 | if (lend == -1) |
980 | end = -1; /* unsigned, so actually very big */ | |
bda97eab | 981 | |
d144bf62 HD |
982 | if (info->fallocend > start && info->fallocend <= end && !unfalloc) |
983 | info->fallocend = start; | |
984 | ||
51dcbdac | 985 | folio_batch_init(&fbatch); |
bda97eab | 986 | index = start; |
3392ca12 | 987 | while (index < end && find_lock_entries(mapping, &index, end - 1, |
51dcbdac MWO |
988 | &fbatch, indices)) { |
989 | for (i = 0; i < folio_batch_count(&fbatch); i++) { | |
b9a8a419 | 990 | folio = fbatch.folios[i]; |
bda97eab | 991 | |
7b774aab | 992 | if (xa_is_value(folio)) { |
1635f6a7 HD |
993 | if (unfalloc) |
994 | continue; | |
7a5d0fbb | 995 | nr_swaps_freed += !shmem_free_swap(mapping, |
3392ca12 | 996 | indices[i], folio); |
bda97eab | 997 | continue; |
7a5d0fbb HD |
998 | } |
999 | ||
7b774aab | 1000 | if (!unfalloc || !folio_test_uptodate(folio)) |
1e84a3d9 | 1001 | truncate_inode_folio(mapping, folio); |
7b774aab | 1002 | folio_unlock(folio); |
bda97eab | 1003 | } |
51dcbdac MWO |
1004 | folio_batch_remove_exceptionals(&fbatch); |
1005 | folio_batch_release(&fbatch); | |
bda97eab | 1006 | cond_resched(); |
bda97eab | 1007 | } |
1da177e4 | 1008 | |
44bcabd7 HD |
1009 | /* |
1010 | * When undoing a failed fallocate, we want none of the partial folio | |
1011 | * zeroing and splitting below, but shall want to truncate the whole | |
1012 | * folio when !uptodate indicates that it was added by this fallocate, | |
1013 | * even when [lstart, lend] covers only a part of the folio. | |
1014 | */ | |
1015 | if (unfalloc) | |
1016 | goto whole_folios; | |
1017 | ||
b9a8a419 MWO |
1018 | same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT); |
1019 | folio = shmem_get_partial_folio(inode, lstart >> PAGE_SHIFT); | |
1020 | if (folio) { | |
1021 | same_folio = lend < folio_pos(folio) + folio_size(folio); | |
1022 | folio_mark_dirty(folio); | |
1023 | if (!truncate_inode_partial_folio(folio, lstart, lend)) { | |
87b11f86 | 1024 | start = folio_next_index(folio); |
b9a8a419 MWO |
1025 | if (same_folio) |
1026 | end = folio->index; | |
83e4fa9c | 1027 | } |
b9a8a419 MWO |
1028 | folio_unlock(folio); |
1029 | folio_put(folio); | |
1030 | folio = NULL; | |
83e4fa9c | 1031 | } |
b9a8a419 MWO |
1032 | |
1033 | if (!same_folio) | |
1034 | folio = shmem_get_partial_folio(inode, lend >> PAGE_SHIFT); | |
1035 | if (folio) { | |
1036 | folio_mark_dirty(folio); | |
1037 | if (!truncate_inode_partial_folio(folio, lstart, lend)) | |
1038 | end = folio->index; | |
1039 | folio_unlock(folio); | |
1040 | folio_put(folio); | |
bda97eab HD |
1041 | } |
1042 | ||
44bcabd7 HD |
1043 | whole_folios: |
1044 | ||
bda97eab | 1045 | index = start; |
b1a36650 | 1046 | while (index < end) { |
bda97eab | 1047 | cond_resched(); |
0cd6144a | 1048 | |
9fb6beea | 1049 | if (!find_get_entries(mapping, &index, end - 1, &fbatch, |
cf2039af | 1050 | indices)) { |
b1a36650 HD |
1051 | /* If all gone or hole-punch or unfalloc, we're done */ |
1052 | if (index == start || end != -1) | |
bda97eab | 1053 | break; |
b1a36650 | 1054 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
1055 | index = start; |
1056 | continue; | |
1057 | } | |
0e499ed3 | 1058 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
b9a8a419 | 1059 | folio = fbatch.folios[i]; |
bda97eab | 1060 | |
0e499ed3 | 1061 | if (xa_is_value(folio)) { |
1635f6a7 HD |
1062 | if (unfalloc) |
1063 | continue; | |
9fb6beea | 1064 | if (shmem_free_swap(mapping, indices[i], folio)) { |
b1a36650 | 1065 | /* Swap was replaced by page: retry */ |
9fb6beea | 1066 | index = indices[i]; |
b1a36650 HD |
1067 | break; |
1068 | } | |
1069 | nr_swaps_freed++; | |
7a5d0fbb HD |
1070 | continue; |
1071 | } | |
1072 | ||
0e499ed3 | 1073 | folio_lock(folio); |
800d8c63 | 1074 | |
0e499ed3 | 1075 | if (!unfalloc || !folio_test_uptodate(folio)) { |
0e499ed3 | 1076 | if (folio_mapping(folio) != mapping) { |
b1a36650 | 1077 | /* Page was replaced by swap: retry */ |
0e499ed3 | 1078 | folio_unlock(folio); |
9fb6beea | 1079 | index = indices[i]; |
b1a36650 | 1080 | break; |
1635f6a7 | 1081 | } |
0e499ed3 MWO |
1082 | VM_BUG_ON_FOLIO(folio_test_writeback(folio), |
1083 | folio); | |
55ac8bbe DS |
1084 | |
1085 | if (!folio_test_large(folio)) { | |
1086 | truncate_inode_folio(mapping, folio); | |
1087 | } else if (truncate_inode_partial_folio(folio, lstart, lend)) { | |
1088 | /* | |
1089 | * If we split a page, reset the loop so | |
1090 | * that we pick up the new sub pages. | |
1091 | * Otherwise the THP was entirely | |
1092 | * dropped or the target range was | |
1093 | * zeroed, so just continue the loop as | |
1094 | * is. | |
1095 | */ | |
1096 | if (!folio_test_large(folio)) { | |
1097 | folio_unlock(folio); | |
1098 | index = start; | |
1099 | break; | |
1100 | } | |
1101 | } | |
7a5d0fbb | 1102 | } |
0e499ed3 | 1103 | folio_unlock(folio); |
bda97eab | 1104 | } |
0e499ed3 MWO |
1105 | folio_batch_remove_exceptionals(&fbatch); |
1106 | folio_batch_release(&fbatch); | |
bda97eab | 1107 | } |
94c1e62d | 1108 | |
3c1b7528 | 1109 | shmem_recalc_inode(inode, 0, -nr_swaps_freed); |
1635f6a7 | 1110 | } |
1da177e4 | 1111 | |
1635f6a7 HD |
1112 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
1113 | { | |
1114 | shmem_undo_range(inode, lstart, lend, false); | |
cf2766bb | 1115 | inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); |
36f05cab | 1116 | inode_inc_iversion(inode); |
1da177e4 | 1117 | } |
94c1e62d | 1118 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 1119 | |
b74d24f7 | 1120 | static int shmem_getattr(struct mnt_idmap *idmap, |
549c7297 | 1121 | const struct path *path, struct kstat *stat, |
a528d35e | 1122 | u32 request_mask, unsigned int query_flags) |
44a30220 | 1123 | { |
a528d35e | 1124 | struct inode *inode = path->dentry->d_inode; |
44a30220 YZ |
1125 | struct shmem_inode_info *info = SHMEM_I(inode); |
1126 | ||
3c1b7528 HD |
1127 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) |
1128 | shmem_recalc_inode(inode, 0, 0); | |
1129 | ||
e408e695 TT |
1130 | if (info->fsflags & FS_APPEND_FL) |
1131 | stat->attributes |= STATX_ATTR_APPEND; | |
1132 | if (info->fsflags & FS_IMMUTABLE_FL) | |
1133 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
1134 | if (info->fsflags & FS_NODUMP_FL) | |
1135 | stat->attributes |= STATX_ATTR_NODUMP; | |
1136 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
1137 | STATX_ATTR_IMMUTABLE | | |
1138 | STATX_ATTR_NODUMP); | |
0d72b928 | 1139 | generic_fillattr(idmap, request_mask, inode, stat); |
89fdcd26 | 1140 | |
2cf13384 | 1141 | if (shmem_is_huge(inode, 0, false, NULL, 0)) |
89fdcd26 YS |
1142 | stat->blksize = HPAGE_PMD_SIZE; |
1143 | ||
f7cd16a5 XR |
1144 | if (request_mask & STATX_BTIME) { |
1145 | stat->result_mask |= STATX_BTIME; | |
1146 | stat->btime.tv_sec = info->i_crtime.tv_sec; | |
1147 | stat->btime.tv_nsec = info->i_crtime.tv_nsec; | |
1148 | } | |
1149 | ||
44a30220 YZ |
1150 | return 0; |
1151 | } | |
1152 | ||
c1632a0f | 1153 | static int shmem_setattr(struct mnt_idmap *idmap, |
549c7297 | 1154 | struct dentry *dentry, struct iattr *attr) |
1da177e4 | 1155 | { |
75c3cfa8 | 1156 | struct inode *inode = d_inode(dentry); |
40e041a2 | 1157 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 | 1158 | int error; |
36f05cab JL |
1159 | bool update_mtime = false; |
1160 | bool update_ctime = true; | |
1da177e4 | 1161 | |
7a80e5b8 | 1162 | error = setattr_prepare(idmap, dentry, attr); |
db78b877 CH |
1163 | if (error) |
1164 | return error; | |
1165 | ||
6fd73538 DV |
1166 | if ((info->seals & F_SEAL_EXEC) && (attr->ia_valid & ATTR_MODE)) { |
1167 | if ((inode->i_mode ^ attr->ia_mode) & 0111) { | |
1168 | return -EPERM; | |
1169 | } | |
1170 | } | |
1171 | ||
94c1e62d HD |
1172 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
1173 | loff_t oldsize = inode->i_size; | |
1174 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 1175 | |
9608703e | 1176 | /* protected by i_rwsem */ |
40e041a2 DR |
1177 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || |
1178 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
1179 | return -EPERM; | |
1180 | ||
94c1e62d | 1181 | if (newsize != oldsize) { |
77142517 KK |
1182 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
1183 | oldsize, newsize); | |
1184 | if (error) | |
1185 | return error; | |
94c1e62d | 1186 | i_size_write(inode, newsize); |
36f05cab JL |
1187 | update_mtime = true; |
1188 | } else { | |
1189 | update_ctime = false; | |
94c1e62d | 1190 | } |
afa2db2f | 1191 | if (newsize <= oldsize) { |
94c1e62d | 1192 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
1193 | if (oldsize > holebegin) |
1194 | unmap_mapping_range(inode->i_mapping, | |
1195 | holebegin, 0, 1); | |
1196 | if (info->alloced) | |
1197 | shmem_truncate_range(inode, | |
1198 | newsize, (loff_t)-1); | |
94c1e62d | 1199 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
1200 | if (oldsize > holebegin) |
1201 | unmap_mapping_range(inode->i_mapping, | |
1202 | holebegin, 0, 1); | |
94c1e62d | 1203 | } |
1da177e4 LT |
1204 | } |
1205 | ||
e09764cf CM |
1206 | if (is_quota_modification(idmap, inode, attr)) { |
1207 | error = dquot_initialize(inode); | |
1208 | if (error) | |
1209 | return error; | |
1210 | } | |
1211 | ||
1212 | /* Transfer quota accounting */ | |
1213 | if (i_uid_needs_update(idmap, attr, inode) || | |
1214 | i_gid_needs_update(idmap, attr, inode)) { | |
1215 | error = dquot_transfer(idmap, inode, attr); | |
e09764cf CM |
1216 | if (error) |
1217 | return error; | |
1218 | } | |
1219 | ||
7a80e5b8 | 1220 | setattr_copy(idmap, inode, attr); |
db78b877 | 1221 | if (attr->ia_valid & ATTR_MODE) |
7a80e5b8 | 1222 | error = posix_acl_chmod(idmap, dentry, inode->i_mode); |
36f05cab | 1223 | if (!error && update_ctime) { |
65287334 | 1224 | inode_set_ctime_current(inode); |
36f05cab | 1225 | if (update_mtime) |
cf2766bb | 1226 | inode_set_mtime_to_ts(inode, inode_get_ctime(inode)); |
36f05cab JL |
1227 | inode_inc_iversion(inode); |
1228 | } | |
1da177e4 LT |
1229 | return error; |
1230 | } | |
1231 | ||
1f895f75 | 1232 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1233 | { |
1da177e4 | 1234 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1235 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
2daf18a7 | 1236 | size_t freed = 0; |
1da177e4 | 1237 | |
30e6a51d | 1238 | if (shmem_mapping(inode->i_mapping)) { |
1da177e4 LT |
1239 | shmem_unacct_size(info->flags, inode->i_size); |
1240 | inode->i_size = 0; | |
bc786390 | 1241 | mapping_set_exiting(inode->i_mapping); |
3889e6e7 | 1242 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1243 | if (!list_empty(&info->shrinklist)) { |
1244 | spin_lock(&sbinfo->shrinklist_lock); | |
1245 | if (!list_empty(&info->shrinklist)) { | |
1246 | list_del_init(&info->shrinklist); | |
1247 | sbinfo->shrinklist_len--; | |
1248 | } | |
1249 | spin_unlock(&sbinfo->shrinklist_lock); | |
1250 | } | |
af53d3e9 HD |
1251 | while (!list_empty(&info->swaplist)) { |
1252 | /* Wait while shmem_unuse() is scanning this inode... */ | |
1253 | wait_var_event(&info->stop_eviction, | |
1254 | !atomic_read(&info->stop_eviction)); | |
cb5f7b9a | 1255 | mutex_lock(&shmem_swaplist_mutex); |
af53d3e9 HD |
1256 | /* ...but beware of the race if we peeked too early */ |
1257 | if (!atomic_read(&info->stop_eviction)) | |
1258 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1259 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1260 | } |
3ed47db3 | 1261 | } |
b09e0fa4 | 1262 | |
2daf18a7 HD |
1263 | simple_xattrs_free(&info->xattrs, sbinfo->max_inodes ? &freed : NULL); |
1264 | shmem_free_inode(inode->i_sb, freed); | |
0f3c42f5 | 1265 | WARN_ON(inode->i_blocks); |
dbd5768f | 1266 | clear_inode(inode); |
e09764cf CM |
1267 | #ifdef CONFIG_TMPFS_QUOTA |
1268 | dquot_free_inode(inode); | |
1269 | dquot_drop(inode); | |
1270 | #endif | |
1da177e4 LT |
1271 | } |
1272 | ||
b56a2d8a | 1273 | static int shmem_find_swap_entries(struct address_space *mapping, |
da08e9b7 MWO |
1274 | pgoff_t start, struct folio_batch *fbatch, |
1275 | pgoff_t *indices, unsigned int type) | |
478922e2 | 1276 | { |
b56a2d8a | 1277 | XA_STATE(xas, &mapping->i_pages, start); |
da08e9b7 | 1278 | struct folio *folio; |
87039546 | 1279 | swp_entry_t entry; |
478922e2 MW |
1280 | |
1281 | rcu_read_lock(); | |
da08e9b7 MWO |
1282 | xas_for_each(&xas, folio, ULONG_MAX) { |
1283 | if (xas_retry(&xas, folio)) | |
5b9c98f3 | 1284 | continue; |
b56a2d8a | 1285 | |
da08e9b7 | 1286 | if (!xa_is_value(folio)) |
478922e2 | 1287 | continue; |
b56a2d8a | 1288 | |
da08e9b7 | 1289 | entry = radix_to_swp_entry(folio); |
6cec2b95 ML |
1290 | /* |
1291 | * swapin error entries can be found in the mapping. But they're | |
1292 | * deliberately ignored here as we've done everything we can do. | |
1293 | */ | |
87039546 HD |
1294 | if (swp_type(entry) != type) |
1295 | continue; | |
b56a2d8a | 1296 | |
e384200e | 1297 | indices[folio_batch_count(fbatch)] = xas.xa_index; |
da08e9b7 MWO |
1298 | if (!folio_batch_add(fbatch, folio)) |
1299 | break; | |
b56a2d8a VRP |
1300 | |
1301 | if (need_resched()) { | |
1302 | xas_pause(&xas); | |
1303 | cond_resched_rcu(); | |
1304 | } | |
478922e2 | 1305 | } |
478922e2 | 1306 | rcu_read_unlock(); |
e21a2955 | 1307 | |
da08e9b7 | 1308 | return xas.xa_index; |
478922e2 MW |
1309 | } |
1310 | ||
46f65ec1 | 1311 | /* |
b56a2d8a VRP |
1312 | * Move the swapped pages for an inode to page cache. Returns the count |
1313 | * of pages swapped in, or the error in case of failure. | |
46f65ec1 | 1314 | */ |
da08e9b7 MWO |
1315 | static int shmem_unuse_swap_entries(struct inode *inode, |
1316 | struct folio_batch *fbatch, pgoff_t *indices) | |
1da177e4 | 1317 | { |
b56a2d8a VRP |
1318 | int i = 0; |
1319 | int ret = 0; | |
bde05d1c | 1320 | int error = 0; |
b56a2d8a | 1321 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 1322 | |
da08e9b7 MWO |
1323 | for (i = 0; i < folio_batch_count(fbatch); i++) { |
1324 | struct folio *folio = fbatch->folios[i]; | |
2e0e26c7 | 1325 | |
da08e9b7 | 1326 | if (!xa_is_value(folio)) |
b56a2d8a | 1327 | continue; |
054a9f7c HD |
1328 | error = shmem_swapin_folio(inode, indices[i], &folio, SGP_CACHE, |
1329 | mapping_gfp_mask(mapping), NULL, NULL); | |
b56a2d8a | 1330 | if (error == 0) { |
da08e9b7 MWO |
1331 | folio_unlock(folio); |
1332 | folio_put(folio); | |
b56a2d8a VRP |
1333 | ret++; |
1334 | } | |
1335 | if (error == -ENOMEM) | |
1336 | break; | |
1337 | error = 0; | |
bde05d1c | 1338 | } |
b56a2d8a VRP |
1339 | return error ? error : ret; |
1340 | } | |
bde05d1c | 1341 | |
b56a2d8a VRP |
1342 | /* |
1343 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1344 | */ | |
10a9c496 | 1345 | static int shmem_unuse_inode(struct inode *inode, unsigned int type) |
b56a2d8a VRP |
1346 | { |
1347 | struct address_space *mapping = inode->i_mapping; | |
1348 | pgoff_t start = 0; | |
da08e9b7 | 1349 | struct folio_batch fbatch; |
b56a2d8a | 1350 | pgoff_t indices[PAGEVEC_SIZE]; |
b56a2d8a VRP |
1351 | int ret = 0; |
1352 | ||
b56a2d8a | 1353 | do { |
da08e9b7 MWO |
1354 | folio_batch_init(&fbatch); |
1355 | shmem_find_swap_entries(mapping, start, &fbatch, indices, type); | |
1356 | if (folio_batch_count(&fbatch) == 0) { | |
b56a2d8a VRP |
1357 | ret = 0; |
1358 | break; | |
46f65ec1 | 1359 | } |
b56a2d8a | 1360 | |
da08e9b7 | 1361 | ret = shmem_unuse_swap_entries(inode, &fbatch, indices); |
b56a2d8a VRP |
1362 | if (ret < 0) |
1363 | break; | |
1364 | ||
da08e9b7 | 1365 | start = indices[folio_batch_count(&fbatch) - 1]; |
b56a2d8a VRP |
1366 | } while (true); |
1367 | ||
1368 | return ret; | |
1da177e4 LT |
1369 | } |
1370 | ||
1371 | /* | |
b56a2d8a VRP |
1372 | * Read all the shared memory data that resides in the swap |
1373 | * device 'type' back into memory, so the swap device can be | |
1374 | * unused. | |
1da177e4 | 1375 | */ |
10a9c496 | 1376 | int shmem_unuse(unsigned int type) |
1da177e4 | 1377 | { |
b56a2d8a | 1378 | struct shmem_inode_info *info, *next; |
bde05d1c HD |
1379 | int error = 0; |
1380 | ||
b56a2d8a VRP |
1381 | if (list_empty(&shmem_swaplist)) |
1382 | return 0; | |
1383 | ||
1384 | mutex_lock(&shmem_swaplist_mutex); | |
b56a2d8a VRP |
1385 | list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) { |
1386 | if (!info->swapped) { | |
6922c0c7 | 1387 | list_del_init(&info->swaplist); |
b56a2d8a VRP |
1388 | continue; |
1389 | } | |
af53d3e9 HD |
1390 | /* |
1391 | * Drop the swaplist mutex while searching the inode for swap; | |
1392 | * but before doing so, make sure shmem_evict_inode() will not | |
1393 | * remove placeholder inode from swaplist, nor let it be freed | |
1394 | * (igrab() would protect from unlink, but not from unmount). | |
1395 | */ | |
1396 | atomic_inc(&info->stop_eviction); | |
b56a2d8a | 1397 | mutex_unlock(&shmem_swaplist_mutex); |
b56a2d8a | 1398 | |
10a9c496 | 1399 | error = shmem_unuse_inode(&info->vfs_inode, type); |
cb5f7b9a | 1400 | cond_resched(); |
b56a2d8a VRP |
1401 | |
1402 | mutex_lock(&shmem_swaplist_mutex); | |
1403 | next = list_next_entry(info, swaplist); | |
1404 | if (!info->swapped) | |
1405 | list_del_init(&info->swaplist); | |
af53d3e9 HD |
1406 | if (atomic_dec_and_test(&info->stop_eviction)) |
1407 | wake_up_var(&info->stop_eviction); | |
b56a2d8a | 1408 | if (error) |
778dd893 | 1409 | break; |
1da177e4 | 1410 | } |
cb5f7b9a | 1411 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1412 | |
778dd893 | 1413 | return error; |
1da177e4 LT |
1414 | } |
1415 | ||
1416 | /* | |
1417 | * Move the page from the page cache to the swap cache. | |
1418 | */ | |
1419 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1420 | { | |
e2e3fdc7 | 1421 | struct folio *folio = page_folio(page); |
8ccee8c1 LC |
1422 | struct address_space *mapping = folio->mapping; |
1423 | struct inode *inode = mapping->host; | |
1424 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2c6efe9c | 1425 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
6922c0c7 HD |
1426 | swp_entry_t swap; |
1427 | pgoff_t index; | |
1da177e4 | 1428 | |
cf7992bf LC |
1429 | /* |
1430 | * Our capabilities prevent regular writeback or sync from ever calling | |
1431 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1432 | * its underlying filesystem, in which case tmpfs should write out to | |
1433 | * swap only in response to memory pressure, and not for the writeback | |
1434 | * threads or sync. | |
1435 | */ | |
1436 | if (WARN_ON_ONCE(!wbc->for_reclaim)) | |
1437 | goto redirty; | |
1438 | ||
2c6efe9c | 1439 | if (WARN_ON_ONCE((info->flags & VM_LOCKED) || sbinfo->noswap)) |
9a976f0c LC |
1440 | goto redirty; |
1441 | ||
1442 | if (!total_swap_pages) | |
1443 | goto redirty; | |
1444 | ||
1e6decf3 HD |
1445 | /* |
1446 | * If /sys/kernel/mm/transparent_hugepage/shmem_enabled is "always" or | |
1447 | * "force", drivers/gpu/drm/i915/gem/i915_gem_shmem.c gets huge pages, | |
1448 | * and its shmem_writeback() needs them to be split when swapping. | |
1449 | */ | |
f530ed0e | 1450 | if (folio_test_large(folio)) { |
1e6decf3 | 1451 | /* Ensure the subpages are still dirty */ |
f530ed0e | 1452 | folio_test_set_dirty(folio); |
1e6decf3 HD |
1453 | if (split_huge_page(page) < 0) |
1454 | goto redirty; | |
f530ed0e MWO |
1455 | folio = page_folio(page); |
1456 | folio_clear_dirty(folio); | |
1e6decf3 HD |
1457 | } |
1458 | ||
f530ed0e | 1459 | index = folio->index; |
1635f6a7 HD |
1460 | |
1461 | /* | |
1462 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1463 | * value into swapfile.c, the only way we can correctly account for a | |
f530ed0e | 1464 | * fallocated folio arriving here is now to initialize it and write it. |
1aac1400 | 1465 | * |
f530ed0e | 1466 | * That's okay for a folio already fallocated earlier, but if we have |
1aac1400 | 1467 | * not yet completed the fallocation, then (a) we want to keep track |
f530ed0e | 1468 | * of this folio in case we have to undo it, and (b) it may not be a |
1aac1400 | 1469 | * good idea to continue anyway, once we're pushing into swap. So |
f530ed0e | 1470 | * reactivate the folio, and let shmem_fallocate() quit when too many. |
1635f6a7 | 1471 | */ |
f530ed0e | 1472 | if (!folio_test_uptodate(folio)) { |
1aac1400 HD |
1473 | if (inode->i_private) { |
1474 | struct shmem_falloc *shmem_falloc; | |
1475 | spin_lock(&inode->i_lock); | |
1476 | shmem_falloc = inode->i_private; | |
1477 | if (shmem_falloc && | |
8e205f77 | 1478 | !shmem_falloc->waitq && |
1aac1400 HD |
1479 | index >= shmem_falloc->start && |
1480 | index < shmem_falloc->next) | |
1481 | shmem_falloc->nr_unswapped++; | |
1482 | else | |
1483 | shmem_falloc = NULL; | |
1484 | spin_unlock(&inode->i_lock); | |
1485 | if (shmem_falloc) | |
1486 | goto redirty; | |
1487 | } | |
f530ed0e MWO |
1488 | folio_zero_range(folio, 0, folio_size(folio)); |
1489 | flush_dcache_folio(folio); | |
1490 | folio_mark_uptodate(folio); | |
1635f6a7 HD |
1491 | } |
1492 | ||
e2e3fdc7 | 1493 | swap = folio_alloc_swap(folio); |
48f170fb HD |
1494 | if (!swap.val) |
1495 | goto redirty; | |
d9fe526a | 1496 | |
b1dea800 HD |
1497 | /* |
1498 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
f530ed0e | 1499 | * if it's not already there. Do it now before the folio is |
6922c0c7 | 1500 | * moved to swap cache, when its pagelock no longer protects |
b1dea800 | 1501 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1502 | * we've incremented swapped, because shmem_unuse_inode() will |
1503 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1504 | */ |
48f170fb HD |
1505 | mutex_lock(&shmem_swaplist_mutex); |
1506 | if (list_empty(&info->swaplist)) | |
b56a2d8a | 1507 | list_add(&info->swaplist, &shmem_swaplist); |
b1dea800 | 1508 | |
a4c366f0 | 1509 | if (add_to_swap_cache(folio, swap, |
3852f676 JK |
1510 | __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN, |
1511 | NULL) == 0) { | |
3c1b7528 | 1512 | shmem_recalc_inode(inode, 0, 1); |
267a4c76 | 1513 | swap_shmem_alloc(swap); |
4cd400fd | 1514 | shmem_delete_from_page_cache(folio, swp_to_radix_entry(swap)); |
267a4c76 | 1515 | |
6922c0c7 | 1516 | mutex_unlock(&shmem_swaplist_mutex); |
f530ed0e | 1517 | BUG_ON(folio_mapped(folio)); |
501a06fe | 1518 | return swap_writepage(&folio->page, wbc); |
1da177e4 LT |
1519 | } |
1520 | ||
6922c0c7 | 1521 | mutex_unlock(&shmem_swaplist_mutex); |
4081f744 | 1522 | put_swap_folio(folio, swap); |
1da177e4 | 1523 | redirty: |
f530ed0e | 1524 | folio_mark_dirty(folio); |
d9fe526a | 1525 | if (wbc->for_reclaim) |
f530ed0e MWO |
1526 | return AOP_WRITEPAGE_ACTIVATE; /* Return with folio locked */ |
1527 | folio_unlock(folio); | |
d9fe526a | 1528 | return 0; |
1da177e4 LT |
1529 | } |
1530 | ||
75edd345 | 1531 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1532 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1533 | { |
095f1fc4 | 1534 | char buffer[64]; |
680d794b | 1535 | |
71fe804b | 1536 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1537 | return; /* show nothing */ |
680d794b | 1538 | |
a7a88b23 | 1539 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1540 | |
1541 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1542 | } |
71fe804b LS |
1543 | |
1544 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1545 | { | |
1546 | struct mempolicy *mpol = NULL; | |
1547 | if (sbinfo->mpol) { | |
bf11b9a8 | 1548 | raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ |
71fe804b LS |
1549 | mpol = sbinfo->mpol; |
1550 | mpol_get(mpol); | |
bf11b9a8 | 1551 | raw_spin_unlock(&sbinfo->stat_lock); |
71fe804b LS |
1552 | } |
1553 | return mpol; | |
1554 | } | |
75edd345 HD |
1555 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1556 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1557 | { | |
1558 | } | |
1559 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1560 | { | |
1561 | return NULL; | |
1562 | } | |
1563 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
680d794b | 1564 | |
ddc1a5cb HD |
1565 | static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info, |
1566 | pgoff_t index, unsigned int order, pgoff_t *ilx); | |
800d8c63 | 1567 | |
ddc1a5cb | 1568 | static struct folio *shmem_swapin_cluster(swp_entry_t swap, gfp_t gfp, |
41ffe5d5 | 1569 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 | 1570 | { |
ddc1a5cb HD |
1571 | struct mempolicy *mpol; |
1572 | pgoff_t ilx; | |
a4575c41 | 1573 | struct folio *folio; |
52cd3b07 | 1574 | |
ddc1a5cb | 1575 | mpol = shmem_get_pgoff_policy(info, index, 0, &ilx); |
a4575c41 | 1576 | folio = swap_cluster_readahead(swap, gfp, mpol, ilx); |
ddc1a5cb | 1577 | mpol_cond_put(mpol); |
18a2f371 | 1578 | |
a4575c41 | 1579 | return folio; |
800d8c63 KS |
1580 | } |
1581 | ||
78cc8cdc RR |
1582 | /* |
1583 | * Make sure huge_gfp is always more limited than limit_gfp. | |
1584 | * Some of the flags set permissions, while others set limitations. | |
1585 | */ | |
1586 | static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) | |
1587 | { | |
1588 | gfp_t allowflags = __GFP_IO | __GFP_FS | __GFP_RECLAIM; | |
1589 | gfp_t denyflags = __GFP_NOWARN | __GFP_NORETRY; | |
187df5dd RR |
1590 | gfp_t zoneflags = limit_gfp & GFP_ZONEMASK; |
1591 | gfp_t result = huge_gfp & ~(allowflags | GFP_ZONEMASK); | |
1592 | ||
1593 | /* Allow allocations only from the originally specified zones. */ | |
1594 | result |= zoneflags; | |
78cc8cdc RR |
1595 | |
1596 | /* | |
1597 | * Minimize the result gfp by taking the union with the deny flags, | |
1598 | * and the intersection of the allow flags. | |
1599 | */ | |
1600 | result |= (limit_gfp & denyflags); | |
1601 | result |= (huge_gfp & limit_gfp) & allowflags; | |
1602 | ||
1603 | return result; | |
1604 | } | |
1605 | ||
72827e5c | 1606 | static struct folio *shmem_alloc_hugefolio(gfp_t gfp, |
800d8c63 KS |
1607 | struct shmem_inode_info *info, pgoff_t index) |
1608 | { | |
ddc1a5cb HD |
1609 | struct mempolicy *mpol; |
1610 | pgoff_t ilx; | |
1611 | struct page *page; | |
800d8c63 | 1612 | |
ddc1a5cb HD |
1613 | mpol = shmem_get_pgoff_policy(info, index, HPAGE_PMD_ORDER, &ilx); |
1614 | page = alloc_pages_mpol(gfp, HPAGE_PMD_ORDER, mpol, ilx, numa_node_id()); | |
1615 | mpol_cond_put(mpol); | |
18a2f371 | 1616 | |
ddc1a5cb | 1617 | return page_rmappable_folio(page); |
1da177e4 LT |
1618 | } |
1619 | ||
0c023ef5 | 1620 | static struct folio *shmem_alloc_folio(gfp_t gfp, |
3022fd7a | 1621 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 | 1622 | { |
ddc1a5cb HD |
1623 | struct mempolicy *mpol; |
1624 | pgoff_t ilx; | |
1625 | struct page *page; | |
1da177e4 | 1626 | |
ddc1a5cb HD |
1627 | mpol = shmem_get_pgoff_policy(info, index, 0, &ilx); |
1628 | page = alloc_pages_mpol(gfp, 0, mpol, ilx, numa_node_id()); | |
1629 | mpol_cond_put(mpol); | |
800d8c63 | 1630 | |
ddc1a5cb | 1631 | return (struct folio *)page; |
0c023ef5 MWO |
1632 | } |
1633 | ||
3022fd7a HD |
1634 | static struct folio *shmem_alloc_and_add_folio(gfp_t gfp, |
1635 | struct inode *inode, pgoff_t index, | |
1636 | struct mm_struct *fault_mm, bool huge) | |
800d8c63 | 1637 | { |
3022fd7a | 1638 | struct address_space *mapping = inode->i_mapping; |
0f079694 | 1639 | struct shmem_inode_info *info = SHMEM_I(inode); |
72827e5c | 1640 | struct folio *folio; |
3022fd7a HD |
1641 | long pages; |
1642 | int error; | |
52cd3b07 | 1643 | |
396bcc52 | 1644 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) |
800d8c63 | 1645 | huge = false; |
800d8c63 | 1646 | |
3022fd7a HD |
1647 | if (huge) { |
1648 | pages = HPAGE_PMD_NR; | |
1649 | index = round_down(index, HPAGE_PMD_NR); | |
1650 | ||
1651 | /* | |
1652 | * Check for conflict before waiting on a huge allocation. | |
1653 | * Conflict might be that a huge page has just been allocated | |
1654 | * and added to page cache by a racing thread, or that there | |
1655 | * is already at least one small page in the huge extent. | |
1656 | * Be careful to retry when appropriate, but not forever! | |
1657 | * Elsewhere -EEXIST would be the right code, but not here. | |
1658 | */ | |
1659 | if (xa_find(&mapping->i_pages, &index, | |
1660 | index + HPAGE_PMD_NR - 1, XA_PRESENT)) | |
1661 | return ERR_PTR(-E2BIG); | |
800d8c63 | 1662 | |
72827e5c | 1663 | folio = shmem_alloc_hugefolio(gfp, info, index); |
3022fd7a HD |
1664 | if (!folio) |
1665 | count_vm_event(THP_FILE_FALLBACK); | |
1666 | } else { | |
1667 | pages = 1; | |
72827e5c | 1668 | folio = shmem_alloc_folio(gfp, info, index); |
75edd345 | 1669 | } |
3022fd7a HD |
1670 | if (!folio) |
1671 | return ERR_PTR(-ENOMEM); | |
18a2f371 | 1672 | |
3022fd7a HD |
1673 | __folio_set_locked(folio); |
1674 | __folio_set_swapbacked(folio); | |
1675 | ||
1676 | gfp &= GFP_RECLAIM_MASK; | |
1677 | error = mem_cgroup_charge(folio, fault_mm, gfp); | |
1678 | if (error) { | |
1679 | if (xa_find(&mapping->i_pages, &index, | |
1680 | index + pages - 1, XA_PRESENT)) { | |
1681 | error = -EEXIST; | |
1682 | } else if (huge) { | |
1683 | count_vm_event(THP_FILE_FALLBACK); | |
1684 | count_vm_event(THP_FILE_FALLBACK_CHARGE); | |
1685 | } | |
1686 | goto unlock; | |
1687 | } | |
1688 | ||
1689 | error = shmem_add_to_page_cache(folio, mapping, index, NULL, gfp); | |
1690 | if (error) | |
1691 | goto unlock; | |
1692 | ||
1693 | error = shmem_inode_acct_blocks(inode, pages); | |
1694 | if (error) { | |
1695 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
1696 | long freed; | |
1697 | /* | |
1698 | * Try to reclaim some space by splitting a few | |
1699 | * large folios beyond i_size on the filesystem. | |
1700 | */ | |
1701 | shmem_unused_huge_shrink(sbinfo, NULL, 2); | |
1702 | /* | |
1703 | * And do a shmem_recalc_inode() to account for freed pages: | |
1704 | * except our folio is there in cache, so not quite balanced. | |
1705 | */ | |
1706 | spin_lock(&info->lock); | |
1707 | freed = pages + info->alloced - info->swapped - | |
1708 | READ_ONCE(mapping->nrpages); | |
1709 | if (freed > 0) | |
1710 | info->alloced -= freed; | |
1711 | spin_unlock(&info->lock); | |
1712 | if (freed > 0) | |
1713 | shmem_inode_unacct_blocks(inode, freed); | |
1714 | error = shmem_inode_acct_blocks(inode, pages); | |
1715 | if (error) { | |
1716 | filemap_remove_folio(folio); | |
1717 | goto unlock; | |
1718 | } | |
1719 | } | |
1720 | ||
1721 | shmem_recalc_inode(inode, pages, 0); | |
1722 | folio_add_lru(folio); | |
1723 | return folio; | |
1724 | ||
1725 | unlock: | |
1726 | folio_unlock(folio); | |
1727 | folio_put(folio); | |
1728 | return ERR_PTR(error); | |
1da177e4 | 1729 | } |
71fe804b | 1730 | |
bde05d1c HD |
1731 | /* |
1732 | * When a page is moved from swapcache to shmem filecache (either by the | |
fc26babb | 1733 | * usual swapin of shmem_get_folio_gfp(), or by the less common swapoff of |
bde05d1c HD |
1734 | * shmem_unuse_inode()), it may have been read in earlier from swap, in |
1735 | * ignorance of the mapping it belongs to. If that mapping has special | |
1736 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1737 | * we may need to copy to a suitable page before moving to filecache. | |
1738 | * | |
1739 | * In a future release, this may well be extended to respect cpuset and | |
1740 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1741 | * but for now it is a simple matter of zone. | |
1742 | */ | |
069d849c | 1743 | static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp) |
bde05d1c | 1744 | { |
069d849c | 1745 | return folio_zonenum(folio) > gfp_zone(gfp); |
bde05d1c HD |
1746 | } |
1747 | ||
0d698e25 | 1748 | static int shmem_replace_folio(struct folio **foliop, gfp_t gfp, |
bde05d1c HD |
1749 | struct shmem_inode_info *info, pgoff_t index) |
1750 | { | |
d21bba2b | 1751 | struct folio *old, *new; |
bde05d1c | 1752 | struct address_space *swap_mapping; |
c1cb20d4 | 1753 | swp_entry_t entry; |
bde05d1c HD |
1754 | pgoff_t swap_index; |
1755 | int error; | |
1756 | ||
0d698e25 | 1757 | old = *foliop; |
3d2c9087 | 1758 | entry = old->swap; |
c1cb20d4 | 1759 | swap_index = swp_offset(entry); |
907ea17e | 1760 | swap_mapping = swap_address_space(entry); |
bde05d1c HD |
1761 | |
1762 | /* | |
1763 | * We have arrived here because our zones are constrained, so don't | |
1764 | * limit chance of success by further cpuset and node constraints. | |
1765 | */ | |
1766 | gfp &= ~GFP_CONSTRAINT_MASK; | |
907ea17e MWO |
1767 | VM_BUG_ON_FOLIO(folio_test_large(old), old); |
1768 | new = shmem_alloc_folio(gfp, info, index); | |
1769 | if (!new) | |
bde05d1c | 1770 | return -ENOMEM; |
bde05d1c | 1771 | |
907ea17e MWO |
1772 | folio_get(new); |
1773 | folio_copy(new, old); | |
1774 | flush_dcache_folio(new); | |
bde05d1c | 1775 | |
907ea17e MWO |
1776 | __folio_set_locked(new); |
1777 | __folio_set_swapbacked(new); | |
1778 | folio_mark_uptodate(new); | |
3d2c9087 | 1779 | new->swap = entry; |
907ea17e | 1780 | folio_set_swapcache(new); |
bde05d1c HD |
1781 | |
1782 | /* | |
1783 | * Our caller will very soon move newpage out of swapcache, but it's | |
1784 | * a nice clean interface for us to replace oldpage by newpage there. | |
1785 | */ | |
b93b0163 | 1786 | xa_lock_irq(&swap_mapping->i_pages); |
907ea17e | 1787 | error = shmem_replace_entry(swap_mapping, swap_index, old, new); |
0142ef6c | 1788 | if (!error) { |
d21bba2b | 1789 | mem_cgroup_migrate(old, new); |
907ea17e MWO |
1790 | __lruvec_stat_mod_folio(new, NR_FILE_PAGES, 1); |
1791 | __lruvec_stat_mod_folio(new, NR_SHMEM, 1); | |
1792 | __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -1); | |
1793 | __lruvec_stat_mod_folio(old, NR_SHMEM, -1); | |
0142ef6c | 1794 | } |
b93b0163 | 1795 | xa_unlock_irq(&swap_mapping->i_pages); |
bde05d1c | 1796 | |
0142ef6c HD |
1797 | if (unlikely(error)) { |
1798 | /* | |
1799 | * Is this possible? I think not, now that our callers check | |
1800 | * both PageSwapCache and page_private after getting page lock; | |
1801 | * but be defensive. Reverse old to newpage for clear and free. | |
1802 | */ | |
907ea17e | 1803 | old = new; |
0142ef6c | 1804 | } else { |
907ea17e | 1805 | folio_add_lru(new); |
0d698e25 | 1806 | *foliop = new; |
0142ef6c | 1807 | } |
bde05d1c | 1808 | |
907ea17e MWO |
1809 | folio_clear_swapcache(old); |
1810 | old->private = NULL; | |
bde05d1c | 1811 | |
907ea17e MWO |
1812 | folio_unlock(old); |
1813 | folio_put_refs(old, 2); | |
0142ef6c | 1814 | return error; |
bde05d1c HD |
1815 | } |
1816 | ||
6cec2b95 ML |
1817 | static void shmem_set_folio_swapin_error(struct inode *inode, pgoff_t index, |
1818 | struct folio *folio, swp_entry_t swap) | |
1819 | { | |
1820 | struct address_space *mapping = inode->i_mapping; | |
6cec2b95 ML |
1821 | swp_entry_t swapin_error; |
1822 | void *old; | |
1823 | ||
af19487f | 1824 | swapin_error = make_poisoned_swp_entry(); |
6cec2b95 ML |
1825 | old = xa_cmpxchg_irq(&mapping->i_pages, index, |
1826 | swp_to_radix_entry(swap), | |
1827 | swp_to_radix_entry(swapin_error), 0); | |
1828 | if (old != swp_to_radix_entry(swap)) | |
1829 | return; | |
1830 | ||
1831 | folio_wait_writeback(folio); | |
75fa68a5 | 1832 | delete_from_swap_cache(folio); |
6cec2b95 | 1833 | /* |
3c1b7528 HD |
1834 | * Don't treat swapin error folio as alloced. Otherwise inode->i_blocks |
1835 | * won't be 0 when inode is released and thus trigger WARN_ON(i_blocks) | |
1836 | * in shmem_evict_inode(). | |
6cec2b95 | 1837 | */ |
3c1b7528 | 1838 | shmem_recalc_inode(inode, -1, -1); |
6cec2b95 ML |
1839 | swap_free(swap); |
1840 | } | |
1841 | ||
c5bf121e | 1842 | /* |
833de10f ML |
1843 | * Swap in the folio pointed to by *foliop. |
1844 | * Caller has to make sure that *foliop contains a valid swapped folio. | |
1845 | * Returns 0 and the folio in foliop if success. On failure, returns the | |
1846 | * error code and NULL in *foliop. | |
c5bf121e | 1847 | */ |
da08e9b7 MWO |
1848 | static int shmem_swapin_folio(struct inode *inode, pgoff_t index, |
1849 | struct folio **foliop, enum sgp_type sgp, | |
054a9f7c | 1850 | gfp_t gfp, struct mm_struct *fault_mm, |
c5bf121e VRP |
1851 | vm_fault_t *fault_type) |
1852 | { | |
1853 | struct address_space *mapping = inode->i_mapping; | |
1854 | struct shmem_inode_info *info = SHMEM_I(inode); | |
cbc2bd98 | 1855 | struct swap_info_struct *si; |
da08e9b7 | 1856 | struct folio *folio = NULL; |
c5bf121e VRP |
1857 | swp_entry_t swap; |
1858 | int error; | |
1859 | ||
da08e9b7 MWO |
1860 | VM_BUG_ON(!*foliop || !xa_is_value(*foliop)); |
1861 | swap = radix_to_swp_entry(*foliop); | |
1862 | *foliop = NULL; | |
c5bf121e | 1863 | |
af19487f | 1864 | if (is_poisoned_swp_entry(swap)) |
6cec2b95 ML |
1865 | return -EIO; |
1866 | ||
cbc2bd98 KS |
1867 | si = get_swap_device(swap); |
1868 | if (!si) { | |
1869 | if (!shmem_confirm_swap(mapping, index, swap)) | |
1870 | return -EEXIST; | |
1871 | else | |
1872 | return -EINVAL; | |
1873 | } | |
1874 | ||
c5bf121e | 1875 | /* Look it up and read it in.. */ |
5739a81c MWO |
1876 | folio = swap_cache_get_folio(swap, NULL, 0); |
1877 | if (!folio) { | |
c5bf121e VRP |
1878 | /* Or update major stats only when swapin succeeds?? */ |
1879 | if (fault_type) { | |
1880 | *fault_type |= VM_FAULT_MAJOR; | |
1881 | count_vm_event(PGMAJFAULT); | |
054a9f7c | 1882 | count_memcg_event_mm(fault_mm, PGMAJFAULT); |
c5bf121e VRP |
1883 | } |
1884 | /* Here we actually start the io */ | |
ddc1a5cb | 1885 | folio = shmem_swapin_cluster(swap, gfp, info, index); |
5739a81c | 1886 | if (!folio) { |
c5bf121e VRP |
1887 | error = -ENOMEM; |
1888 | goto failed; | |
1889 | } | |
1890 | } | |
1891 | ||
833de10f | 1892 | /* We have to do this with folio locked to prevent races */ |
da08e9b7 MWO |
1893 | folio_lock(folio); |
1894 | if (!folio_test_swapcache(folio) || | |
3d2c9087 | 1895 | folio->swap.val != swap.val || |
c5bf121e VRP |
1896 | !shmem_confirm_swap(mapping, index, swap)) { |
1897 | error = -EEXIST; | |
1898 | goto unlock; | |
1899 | } | |
da08e9b7 | 1900 | if (!folio_test_uptodate(folio)) { |
c5bf121e VRP |
1901 | error = -EIO; |
1902 | goto failed; | |
1903 | } | |
da08e9b7 | 1904 | folio_wait_writeback(folio); |
c5bf121e | 1905 | |
8a84802e SP |
1906 | /* |
1907 | * Some architectures may have to restore extra metadata to the | |
da08e9b7 | 1908 | * folio after reading from swap. |
8a84802e | 1909 | */ |
da08e9b7 | 1910 | arch_swap_restore(swap, folio); |
8a84802e | 1911 | |
069d849c | 1912 | if (shmem_should_replace_folio(folio, gfp)) { |
0d698e25 | 1913 | error = shmem_replace_folio(&folio, gfp, info, index); |
c5bf121e VRP |
1914 | if (error) |
1915 | goto failed; | |
1916 | } | |
1917 | ||
b7dd44a1 | 1918 | error = shmem_add_to_page_cache(folio, mapping, index, |
054a9f7c | 1919 | swp_to_radix_entry(swap), gfp); |
3fea5a49 | 1920 | if (error) |
14235ab3 | 1921 | goto failed; |
c5bf121e | 1922 | |
3c1b7528 | 1923 | shmem_recalc_inode(inode, 0, -1); |
c5bf121e VRP |
1924 | |
1925 | if (sgp == SGP_WRITE) | |
da08e9b7 | 1926 | folio_mark_accessed(folio); |
c5bf121e | 1927 | |
75fa68a5 | 1928 | delete_from_swap_cache(folio); |
da08e9b7 | 1929 | folio_mark_dirty(folio); |
c5bf121e | 1930 | swap_free(swap); |
cbc2bd98 | 1931 | put_swap_device(si); |
c5bf121e | 1932 | |
da08e9b7 | 1933 | *foliop = folio; |
c5bf121e VRP |
1934 | return 0; |
1935 | failed: | |
1936 | if (!shmem_confirm_swap(mapping, index, swap)) | |
1937 | error = -EEXIST; | |
6cec2b95 ML |
1938 | if (error == -EIO) |
1939 | shmem_set_folio_swapin_error(inode, index, folio, swap); | |
c5bf121e | 1940 | unlock: |
da08e9b7 MWO |
1941 | if (folio) { |
1942 | folio_unlock(folio); | |
1943 | folio_put(folio); | |
c5bf121e | 1944 | } |
cbc2bd98 | 1945 | put_swap_device(si); |
c5bf121e VRP |
1946 | |
1947 | return error; | |
1948 | } | |
1949 | ||
1da177e4 | 1950 | /* |
fc26babb | 1951 | * shmem_get_folio_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1952 | * |
1953 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1954 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1955 | * entry since a page cannot live in both the swap and page cache. |
1956 | * | |
e3e1a506 | 1957 | * vmf and fault_type are only supplied by shmem_fault: otherwise they are NULL. |
1da177e4 | 1958 | */ |
fc26babb MWO |
1959 | static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index, |
1960 | struct folio **foliop, enum sgp_type sgp, gfp_t gfp, | |
e3e1a506 | 1961 | struct vm_fault *vmf, vm_fault_t *fault_type) |
1da177e4 | 1962 | { |
e3e1a506 | 1963 | struct vm_area_struct *vma = vmf ? vmf->vma : NULL; |
054a9f7c | 1964 | struct mm_struct *fault_mm; |
b7dd44a1 | 1965 | struct folio *folio; |
1da177e4 | 1966 | int error; |
3022fd7a | 1967 | bool alloced; |
1da177e4 | 1968 | |
1cd81faa CH |
1969 | if (WARN_ON_ONCE(!shmem_mapping(inode->i_mapping))) |
1970 | return -EINVAL; | |
1971 | ||
09cbfeaf | 1972 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1973 | return -EFBIG; |
1da177e4 | 1974 | repeat: |
c5bf121e | 1975 | if (sgp <= SGP_CACHE && |
3022fd7a | 1976 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) |
c5bf121e | 1977 | return -EINVAL; |
c5bf121e | 1978 | |
3022fd7a | 1979 | alloced = false; |
054a9f7c | 1980 | fault_mm = vma ? vma->vm_mm : NULL; |
c5bf121e | 1981 | |
3022fd7a | 1982 | folio = filemap_get_entry(inode->i_mapping, index); |
b1d0ec3a | 1983 | if (folio && vma && userfaultfd_minor(vma)) { |
aaeb94eb | 1984 | if (!xa_is_value(folio)) |
b1d0ec3a | 1985 | folio_put(folio); |
c949b097 AR |
1986 | *fault_type = handle_userfault(vmf, VM_UFFD_MINOR); |
1987 | return 0; | |
1988 | } | |
1989 | ||
b1d0ec3a | 1990 | if (xa_is_value(folio)) { |
da08e9b7 | 1991 | error = shmem_swapin_folio(inode, index, &folio, |
054a9f7c | 1992 | sgp, gfp, fault_mm, fault_type); |
c5bf121e VRP |
1993 | if (error == -EEXIST) |
1994 | goto repeat; | |
54af6042 | 1995 | |
fc26babb | 1996 | *foliop = folio; |
c5bf121e | 1997 | return error; |
54af6042 HD |
1998 | } |
1999 | ||
b1d0ec3a | 2000 | if (folio) { |
aaeb94eb CH |
2001 | folio_lock(folio); |
2002 | ||
2003 | /* Has the folio been truncated or swapped out? */ | |
3022fd7a | 2004 | if (unlikely(folio->mapping != inode->i_mapping)) { |
aaeb94eb CH |
2005 | folio_unlock(folio); |
2006 | folio_put(folio); | |
2007 | goto repeat; | |
2008 | } | |
acdd9f8e | 2009 | if (sgp == SGP_WRITE) |
b1d0ec3a MWO |
2010 | folio_mark_accessed(folio); |
2011 | if (folio_test_uptodate(folio)) | |
acdd9f8e | 2012 | goto out; |
fc26babb | 2013 | /* fallocated folio */ |
1635f6a7 HD |
2014 | if (sgp != SGP_READ) |
2015 | goto clear; | |
b1d0ec3a MWO |
2016 | folio_unlock(folio); |
2017 | folio_put(folio); | |
1635f6a7 | 2018 | } |
27ab7006 HD |
2019 | |
2020 | /* | |
fc26babb MWO |
2021 | * SGP_READ: succeed on hole, with NULL folio, letting caller zero. |
2022 | * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail. | |
acdd9f8e | 2023 | */ |
fc26babb | 2024 | *foliop = NULL; |
acdd9f8e HD |
2025 | if (sgp == SGP_READ) |
2026 | return 0; | |
2027 | if (sgp == SGP_NOALLOC) | |
2028 | return -ENOENT; | |
2029 | ||
2030 | /* | |
2031 | * Fast cache lookup and swap lookup did not find it: allocate. | |
27ab7006 | 2032 | */ |
54af6042 | 2033 | |
c5bf121e VRP |
2034 | if (vma && userfaultfd_missing(vma)) { |
2035 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
2036 | return 0; | |
2037 | } | |
cfda0526 | 2038 | |
3022fd7a HD |
2039 | if (shmem_is_huge(inode, index, false, fault_mm, |
2040 | vma ? vma->vm_flags : 0)) { | |
2041 | gfp_t huge_gfp; | |
1da177e4 | 2042 | |
3022fd7a HD |
2043 | huge_gfp = vma_thp_gfp_mask(vma); |
2044 | huge_gfp = limit_gfp_mask(huge_gfp, gfp); | |
2045 | folio = shmem_alloc_and_add_folio(huge_gfp, | |
2046 | inode, index, fault_mm, true); | |
2047 | if (!IS_ERR(folio)) { | |
2048 | count_vm_event(THP_FILE_ALLOC); | |
2049 | goto alloced; | |
2050 | } | |
2051 | if (PTR_ERR(folio) == -EEXIST) | |
2052 | goto repeat; | |
c5bf121e | 2053 | } |
800d8c63 | 2054 | |
3022fd7a HD |
2055 | folio = shmem_alloc_and_add_folio(gfp, inode, index, fault_mm, false); |
2056 | if (IS_ERR(folio)) { | |
b1d0ec3a | 2057 | error = PTR_ERR(folio); |
3022fd7a HD |
2058 | if (error == -EEXIST) |
2059 | goto repeat; | |
b1d0ec3a | 2060 | folio = NULL; |
c5bf121e VRP |
2061 | goto unlock; |
2062 | } | |
54af6042 | 2063 | |
3022fd7a | 2064 | alloced: |
c5bf121e | 2065 | alloced = true; |
b1d0ec3a | 2066 | if (folio_test_pmd_mappable(folio) && |
c5bf121e | 2067 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < |
fc26babb | 2068 | folio_next_index(folio) - 1) { |
3022fd7a HD |
2069 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
2070 | struct shmem_inode_info *info = SHMEM_I(inode); | |
ec9516fb | 2071 | /* |
fc26babb | 2072 | * Part of the large folio is beyond i_size: subject |
c5bf121e | 2073 | * to shrink under memory pressure. |
1635f6a7 | 2074 | */ |
c5bf121e | 2075 | spin_lock(&sbinfo->shrinklist_lock); |
1635f6a7 | 2076 | /* |
c5bf121e VRP |
2077 | * _careful to defend against unlocked access to |
2078 | * ->shrink_list in shmem_unused_huge_shrink() | |
ec9516fb | 2079 | */ |
c5bf121e VRP |
2080 | if (list_empty_careful(&info->shrinklist)) { |
2081 | list_add_tail(&info->shrinklist, | |
2082 | &sbinfo->shrinklist); | |
2083 | sbinfo->shrinklist_len++; | |
2084 | } | |
2085 | spin_unlock(&sbinfo->shrinklist_lock); | |
2086 | } | |
800d8c63 | 2087 | |
3022fd7a HD |
2088 | if (sgp == SGP_WRITE) |
2089 | folio_set_referenced(folio); | |
c5bf121e | 2090 | /* |
fc26babb | 2091 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio. |
c5bf121e VRP |
2092 | */ |
2093 | if (sgp == SGP_FALLOC) | |
2094 | sgp = SGP_WRITE; | |
2095 | clear: | |
2096 | /* | |
fc26babb MWO |
2097 | * Let SGP_WRITE caller clear ends if write does not fill folio; |
2098 | * but SGP_FALLOC on a folio fallocated earlier must initialize | |
c5bf121e VRP |
2099 | * it now, lest undo on failure cancel our earlier guarantee. |
2100 | */ | |
b1d0ec3a MWO |
2101 | if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { |
2102 | long i, n = folio_nr_pages(folio); | |
c5bf121e | 2103 | |
b1d0ec3a MWO |
2104 | for (i = 0; i < n; i++) |
2105 | clear_highpage(folio_page(folio, i)); | |
2106 | flush_dcache_folio(folio); | |
2107 | folio_mark_uptodate(folio); | |
1da177e4 | 2108 | } |
bde05d1c | 2109 | |
54af6042 | 2110 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 2111 | if (sgp <= SGP_CACHE && |
09cbfeaf | 2112 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 2113 | error = -EINVAL; |
267a4c76 | 2114 | goto unlock; |
e83c32e8 | 2115 | } |
63ec1973 | 2116 | out: |
fc26babb | 2117 | *foliop = folio; |
54af6042 | 2118 | return 0; |
1da177e4 | 2119 | |
59a16ead | 2120 | /* |
54af6042 | 2121 | * Error recovery. |
59a16ead | 2122 | */ |
d1899228 | 2123 | unlock: |
3022fd7a HD |
2124 | if (alloced) |
2125 | filemap_remove_folio(folio); | |
2126 | shmem_recalc_inode(inode, 0, 0); | |
b1d0ec3a MWO |
2127 | if (folio) { |
2128 | folio_unlock(folio); | |
2129 | folio_put(folio); | |
54af6042 | 2130 | } |
54af6042 | 2131 | return error; |
1da177e4 LT |
2132 | } |
2133 | ||
d7468609 CH |
2134 | /** |
2135 | * shmem_get_folio - find, and lock a shmem folio. | |
2136 | * @inode: inode to search | |
2137 | * @index: the page index. | |
2138 | * @foliop: pointer to the folio if found | |
2139 | * @sgp: SGP_* flags to control behavior | |
2140 | * | |
2141 | * Looks up the page cache entry at @inode & @index. If a folio is | |
2142 | * present, it is returned locked with an increased refcount. | |
2143 | * | |
9d8b3674 CH |
2144 | * If the caller modifies data in the folio, it must call folio_mark_dirty() |
2145 | * before unlocking the folio to ensure that the folio is not reclaimed. | |
2146 | * There is no need to reserve space before calling folio_mark_dirty(). | |
2147 | * | |
d7468609 | 2148 | * When no folio is found, the behavior depends on @sgp: |
8d4dd9d7 AY |
2149 | * - for SGP_READ, *@foliop is %NULL and 0 is returned |
2150 | * - for SGP_NOALLOC, *@foliop is %NULL and -ENOENT is returned | |
d7468609 CH |
2151 | * - for all other flags a new folio is allocated, inserted into the |
2152 | * page cache and returned locked in @foliop. | |
2153 | * | |
2154 | * Context: May sleep. | |
2155 | * Return: 0 if successful, else a negative error code. | |
2156 | */ | |
4e1fc793 MWO |
2157 | int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop, |
2158 | enum sgp_type sgp) | |
2159 | { | |
2160 | return shmem_get_folio_gfp(inode, index, foliop, sgp, | |
e3e1a506 | 2161 | mapping_gfp_mask(inode->i_mapping), NULL, NULL); |
4e1fc793 | 2162 | } |
d7468609 | 2163 | EXPORT_SYMBOL_GPL(shmem_get_folio); |
4e1fc793 | 2164 | |
10d20bd2 LT |
2165 | /* |
2166 | * This is like autoremove_wake_function, but it removes the wait queue | |
2167 | * entry unconditionally - even if something else had already woken the | |
2168 | * target. | |
2169 | */ | |
f0a9ad1d HD |
2170 | static int synchronous_wake_function(wait_queue_entry_t *wait, |
2171 | unsigned int mode, int sync, void *key) | |
10d20bd2 LT |
2172 | { |
2173 | int ret = default_wake_function(wait, mode, sync, key); | |
2055da97 | 2174 | list_del_init(&wait->entry); |
10d20bd2 LT |
2175 | return ret; |
2176 | } | |
2177 | ||
f0a9ad1d HD |
2178 | /* |
2179 | * Trinity finds that probing a hole which tmpfs is punching can | |
2180 | * prevent the hole-punch from ever completing: which in turn | |
2181 | * locks writers out with its hold on i_rwsem. So refrain from | |
2182 | * faulting pages into the hole while it's being punched. Although | |
2183 | * shmem_undo_range() does remove the additions, it may be unable to | |
2184 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
2185 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
2186 | * | |
2187 | * It does not matter if we sometimes reach this check just before the | |
2188 | * hole-punch begins, so that one fault then races with the punch: | |
2189 | * we just need to make racing faults a rare case. | |
2190 | * | |
2191 | * The implementation below would be much simpler if we just used a | |
2192 | * standard mutex or completion: but we cannot take i_rwsem in fault, | |
2193 | * and bloating every shmem inode for this unlikely case would be sad. | |
2194 | */ | |
2195 | static vm_fault_t shmem_falloc_wait(struct vm_fault *vmf, struct inode *inode) | |
2196 | { | |
2197 | struct shmem_falloc *shmem_falloc; | |
2198 | struct file *fpin = NULL; | |
2199 | vm_fault_t ret = 0; | |
2200 | ||
2201 | spin_lock(&inode->i_lock); | |
2202 | shmem_falloc = inode->i_private; | |
2203 | if (shmem_falloc && | |
2204 | shmem_falloc->waitq && | |
2205 | vmf->pgoff >= shmem_falloc->start && | |
2206 | vmf->pgoff < shmem_falloc->next) { | |
2207 | wait_queue_head_t *shmem_falloc_waitq; | |
2208 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); | |
2209 | ||
2210 | ret = VM_FAULT_NOPAGE; | |
2211 | fpin = maybe_unlock_mmap_for_io(vmf, NULL); | |
2212 | shmem_falloc_waitq = shmem_falloc->waitq; | |
2213 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
2214 | TASK_UNINTERRUPTIBLE); | |
2215 | spin_unlock(&inode->i_lock); | |
2216 | schedule(); | |
2217 | ||
2218 | /* | |
2219 | * shmem_falloc_waitq points into the shmem_fallocate() | |
2220 | * stack of the hole-punching task: shmem_falloc_waitq | |
2221 | * is usually invalid by the time we reach here, but | |
2222 | * finish_wait() does not dereference it in that case; | |
2223 | * though i_lock needed lest racing with wake_up_all(). | |
2224 | */ | |
2225 | spin_lock(&inode->i_lock); | |
2226 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
2227 | } | |
2228 | spin_unlock(&inode->i_lock); | |
2229 | if (fpin) { | |
2230 | fput(fpin); | |
2231 | ret = VM_FAULT_RETRY; | |
2232 | } | |
2233 | return ret; | |
2234 | } | |
2235 | ||
20acce67 | 2236 | static vm_fault_t shmem_fault(struct vm_fault *vmf) |
1da177e4 | 2237 | { |
f0a9ad1d | 2238 | struct inode *inode = file_inode(vmf->vma->vm_file); |
9e18eb29 | 2239 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
68a54100 | 2240 | struct folio *folio = NULL; |
f0a9ad1d | 2241 | vm_fault_t ret = 0; |
20acce67 | 2242 | int err; |
1da177e4 | 2243 | |
f00cdc6d HD |
2244 | /* |
2245 | * Trinity finds that probing a hole which tmpfs is punching can | |
f0a9ad1d | 2246 | * prevent the hole-punch from ever completing: noted in i_private. |
f00cdc6d HD |
2247 | */ |
2248 | if (unlikely(inode->i_private)) { | |
f0a9ad1d HD |
2249 | ret = shmem_falloc_wait(vmf, inode); |
2250 | if (ret) | |
8e205f77 | 2251 | return ret; |
f00cdc6d HD |
2252 | } |
2253 | ||
f0a9ad1d | 2254 | WARN_ON_ONCE(vmf->page != NULL); |
68a54100 | 2255 | err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE, |
e3e1a506 | 2256 | gfp, vmf, &ret); |
20acce67 SJ |
2257 | if (err) |
2258 | return vmf_error(err); | |
f0a9ad1d | 2259 | if (folio) { |
68a54100 | 2260 | vmf->page = folio_file_page(folio, vmf->pgoff); |
f0a9ad1d HD |
2261 | ret |= VM_FAULT_LOCKED; |
2262 | } | |
68da9f05 | 2263 | return ret; |
1da177e4 LT |
2264 | } |
2265 | ||
c01d5b30 HD |
2266 | unsigned long shmem_get_unmapped_area(struct file *file, |
2267 | unsigned long uaddr, unsigned long len, | |
2268 | unsigned long pgoff, unsigned long flags) | |
2269 | { | |
2270 | unsigned long (*get_area)(struct file *, | |
2271 | unsigned long, unsigned long, unsigned long, unsigned long); | |
2272 | unsigned long addr; | |
2273 | unsigned long offset; | |
2274 | unsigned long inflated_len; | |
2275 | unsigned long inflated_addr; | |
2276 | unsigned long inflated_offset; | |
2277 | ||
2278 | if (len > TASK_SIZE) | |
2279 | return -ENOMEM; | |
2280 | ||
2281 | get_area = current->mm->get_unmapped_area; | |
2282 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2283 | ||
396bcc52 | 2284 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) |
c01d5b30 HD |
2285 | return addr; |
2286 | if (IS_ERR_VALUE(addr)) | |
2287 | return addr; | |
2288 | if (addr & ~PAGE_MASK) | |
2289 | return addr; | |
2290 | if (addr > TASK_SIZE - len) | |
2291 | return addr; | |
2292 | ||
2293 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2294 | return addr; | |
2295 | if (len < HPAGE_PMD_SIZE) | |
2296 | return addr; | |
2297 | if (flags & MAP_FIXED) | |
2298 | return addr; | |
2299 | /* | |
2300 | * Our priority is to support MAP_SHARED mapped hugely; | |
2301 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
99158997 KS |
2302 | * But if caller specified an address hint and we allocated area there |
2303 | * successfully, respect that as before. | |
c01d5b30 | 2304 | */ |
99158997 | 2305 | if (uaddr == addr) |
c01d5b30 HD |
2306 | return addr; |
2307 | ||
2308 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2309 | struct super_block *sb; | |
2310 | ||
2311 | if (file) { | |
2312 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2313 | sb = file_inode(file)->i_sb; | |
2314 | } else { | |
2315 | /* | |
2316 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2317 | * for "/dev/zero", to create a shared anonymous object. | |
2318 | */ | |
2319 | if (IS_ERR(shm_mnt)) | |
2320 | return addr; | |
2321 | sb = shm_mnt->mnt_sb; | |
2322 | } | |
3089bf61 | 2323 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2324 | return addr; |
2325 | } | |
2326 | ||
2327 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2328 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2329 | return addr; | |
2330 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2331 | return addr; | |
2332 | ||
2333 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2334 | if (inflated_len > TASK_SIZE) | |
2335 | return addr; | |
2336 | if (inflated_len < len) | |
2337 | return addr; | |
2338 | ||
99158997 | 2339 | inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags); |
c01d5b30 HD |
2340 | if (IS_ERR_VALUE(inflated_addr)) |
2341 | return addr; | |
2342 | if (inflated_addr & ~PAGE_MASK) | |
2343 | return addr; | |
2344 | ||
2345 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2346 | inflated_addr += offset - inflated_offset; | |
2347 | if (inflated_offset > offset) | |
2348 | inflated_addr += HPAGE_PMD_SIZE; | |
2349 | ||
2350 | if (inflated_addr > TASK_SIZE - len) | |
2351 | return addr; | |
2352 | return inflated_addr; | |
2353 | } | |
2354 | ||
1da177e4 | 2355 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2356 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2357 | { |
496ad9aa | 2358 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2359 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2360 | } |
2361 | ||
d8dc74f2 | 2362 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
ddc1a5cb | 2363 | unsigned long addr, pgoff_t *ilx) |
1da177e4 | 2364 | { |
496ad9aa | 2365 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2366 | pgoff_t index; |
1da177e4 | 2367 | |
ddc1a5cb HD |
2368 | /* |
2369 | * Bias interleave by inode number to distribute better across nodes; | |
2370 | * but this interface is independent of which page order is used, so | |
2371 | * supplies only that bias, letting caller apply the offset (adjusted | |
2372 | * by page order, as in shmem_get_pgoff_policy() and get_vma_policy()). | |
2373 | */ | |
2374 | *ilx = inode->i_ino; | |
41ffe5d5 HD |
2375 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2376 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 | 2377 | } |
ddc1a5cb HD |
2378 | |
2379 | static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info, | |
2380 | pgoff_t index, unsigned int order, pgoff_t *ilx) | |
2381 | { | |
2382 | struct mempolicy *mpol; | |
2383 | ||
2384 | /* Bias interleave by inode number to distribute better across nodes */ | |
2385 | *ilx = info->vfs_inode.i_ino + (index >> order); | |
2386 | ||
2387 | mpol = mpol_shared_policy_lookup(&info->policy, index); | |
2388 | return mpol ? mpol : get_task_policy(current); | |
2389 | } | |
2390 | #else | |
2391 | static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info, | |
2392 | pgoff_t index, unsigned int order, pgoff_t *ilx) | |
2393 | { | |
2394 | *ilx = 0; | |
2395 | return NULL; | |
2396 | } | |
2397 | #endif /* CONFIG_NUMA */ | |
1da177e4 | 2398 | |
d7c9e99a | 2399 | int shmem_lock(struct file *file, int lock, struct ucounts *ucounts) |
1da177e4 | 2400 | { |
496ad9aa | 2401 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2402 | struct shmem_inode_info *info = SHMEM_I(inode); |
2403 | int retval = -ENOMEM; | |
2404 | ||
ea0dfeb4 HD |
2405 | /* |
2406 | * What serializes the accesses to info->flags? | |
2407 | * ipc_lock_object() when called from shmctl_do_lock(), | |
2408 | * no serialization needed when called from shm_destroy(). | |
2409 | */ | |
1da177e4 | 2410 | if (lock && !(info->flags & VM_LOCKED)) { |
d7c9e99a | 2411 | if (!user_shm_lock(inode->i_size, ucounts)) |
1da177e4 LT |
2412 | goto out_nomem; |
2413 | info->flags |= VM_LOCKED; | |
89e004ea | 2414 | mapping_set_unevictable(file->f_mapping); |
1da177e4 | 2415 | } |
d7c9e99a AG |
2416 | if (!lock && (info->flags & VM_LOCKED) && ucounts) { |
2417 | user_shm_unlock(inode->i_size, ucounts); | |
1da177e4 | 2418 | info->flags &= ~VM_LOCKED; |
89e004ea | 2419 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2420 | } |
2421 | retval = 0; | |
89e004ea | 2422 | |
1da177e4 | 2423 | out_nomem: |
1da177e4 LT |
2424 | return retval; |
2425 | } | |
2426 | ||
9b83a6a8 | 2427 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 | 2428 | { |
d09e8ca6 PT |
2429 | struct inode *inode = file_inode(file); |
2430 | struct shmem_inode_info *info = SHMEM_I(inode); | |
22247efd | 2431 | int ret; |
ab3948f5 | 2432 | |
28464bbb | 2433 | ret = seal_check_write(info->seals, vma); |
22247efd PX |
2434 | if (ret) |
2435 | return ret; | |
ab3948f5 | 2436 | |
51b0bff2 | 2437 | /* arm64 - allow memory tagging on RAM-based files */ |
1c71222e | 2438 | vm_flags_set(vma, VM_MTE_ALLOWED); |
51b0bff2 | 2439 | |
1da177e4 | 2440 | file_accessed(file); |
d09e8ca6 PT |
2441 | /* This is anonymous shared memory if it is unlinked at the time of mmap */ |
2442 | if (inode->i_nlink) | |
2443 | vma->vm_ops = &shmem_vm_ops; | |
2444 | else | |
2445 | vma->vm_ops = &shmem_anon_vm_ops; | |
1da177e4 LT |
2446 | return 0; |
2447 | } | |
2448 | ||
e88e0d36 HD |
2449 | static int shmem_file_open(struct inode *inode, struct file *file) |
2450 | { | |
2451 | file->f_mode |= FMODE_CAN_ODIRECT; | |
2452 | return generic_file_open(inode, file); | |
2453 | } | |
2454 | ||
cb241339 HD |
2455 | #ifdef CONFIG_TMPFS_XATTR |
2456 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2457 | ||
2458 | /* | |
2459 | * chattr's fsflags are unrelated to extended attributes, | |
2460 | * but tmpfs has chosen to enable them under the same config option. | |
2461 | */ | |
2462 | static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags) | |
2463 | { | |
2464 | unsigned int i_flags = 0; | |
2465 | ||
2466 | if (fsflags & FS_NOATIME_FL) | |
2467 | i_flags |= S_NOATIME; | |
2468 | if (fsflags & FS_APPEND_FL) | |
2469 | i_flags |= S_APPEND; | |
2470 | if (fsflags & FS_IMMUTABLE_FL) | |
2471 | i_flags |= S_IMMUTABLE; | |
2472 | /* | |
2473 | * But FS_NODUMP_FL does not require any action in i_flags. | |
2474 | */ | |
2475 | inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE); | |
2476 | } | |
2477 | #else | |
2478 | static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags) | |
e408e695 | 2479 | { |
e408e695 | 2480 | } |
cb241339 HD |
2481 | #define shmem_initxattrs NULL |
2482 | #endif | |
e408e695 | 2483 | |
a2e45955 CL |
2484 | static struct offset_ctx *shmem_get_offset_ctx(struct inode *inode) |
2485 | { | |
2486 | return &SHMEM_I(inode)->dir_offsets; | |
2487 | } | |
2488 | ||
e09764cf CM |
2489 | static struct inode *__shmem_get_inode(struct mnt_idmap *idmap, |
2490 | struct super_block *sb, | |
2491 | struct inode *dir, umode_t mode, | |
2492 | dev_t dev, unsigned long flags) | |
1da177e4 LT |
2493 | { |
2494 | struct inode *inode; | |
2495 | struct shmem_inode_info *info; | |
2496 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
e809d5f0 | 2497 | ino_t ino; |
71480663 CM |
2498 | int err; |
2499 | ||
2500 | err = shmem_reserve_inode(sb, &ino); | |
2501 | if (err) | |
2502 | return ERR_PTR(err); | |
1da177e4 | 2503 | |
1da177e4 | 2504 | inode = new_inode(sb); |
71480663 | 2505 | if (!inode) { |
2daf18a7 | 2506 | shmem_free_inode(sb, 0); |
71480663 CM |
2507 | return ERR_PTR(-ENOSPC); |
2508 | } | |
b45d71fb | 2509 | |
71480663 CM |
2510 | inode->i_ino = ino; |
2511 | inode_init_owner(idmap, inode, dir, mode); | |
2512 | inode->i_blocks = 0; | |
cf2766bb | 2513 | simple_inode_init_ts(inode); |
71480663 CM |
2514 | inode->i_generation = get_random_u32(); |
2515 | info = SHMEM_I(inode); | |
2516 | memset(info, 0, (char *)inode - (char *)info); | |
2517 | spin_lock_init(&info->lock); | |
2518 | atomic_set(&info->stop_eviction, 0); | |
2519 | info->seals = F_SEAL_SEAL; | |
2520 | info->flags = flags & VM_NORESERVE; | |
cf2766bb | 2521 | info->i_crtime = inode_get_mtime(inode); |
71480663 CM |
2522 | info->fsflags = (dir == NULL) ? 0 : |
2523 | SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED; | |
2524 | if (info->fsflags) | |
2525 | shmem_set_inode_flags(inode, info->fsflags); | |
2526 | INIT_LIST_HEAD(&info->shrinklist); | |
2527 | INIT_LIST_HEAD(&info->swaplist); | |
71480663 CM |
2528 | simple_xattrs_init(&info->xattrs); |
2529 | cache_no_acl(inode); | |
9be7d5b0 HD |
2530 | if (sbinfo->noswap) |
2531 | mapping_set_unevictable(inode->i_mapping); | |
71480663 CM |
2532 | mapping_set_large_folios(inode->i_mapping); |
2533 | ||
2534 | switch (mode & S_IFMT) { | |
2535 | default: | |
2536 | inode->i_op = &shmem_special_inode_operations; | |
2537 | init_special_inode(inode, mode, dev); | |
2538 | break; | |
2539 | case S_IFREG: | |
2540 | inode->i_mapping->a_ops = &shmem_aops; | |
2541 | inode->i_op = &shmem_inode_operations; | |
2542 | inode->i_fop = &shmem_file_operations; | |
2543 | mpol_shared_policy_init(&info->policy, | |
2544 | shmem_get_sbmpol(sbinfo)); | |
2545 | break; | |
2546 | case S_IFDIR: | |
2547 | inc_nlink(inode); | |
2548 | /* Some things misbehave if size == 0 on a directory */ | |
2549 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2550 | inode->i_op = &shmem_dir_inode_operations; | |
a2e45955 CL |
2551 | inode->i_fop = &simple_offset_dir_operations; |
2552 | simple_offset_init(shmem_get_offset_ctx(inode)); | |
71480663 CM |
2553 | break; |
2554 | case S_IFLNK: | |
2555 | /* | |
2556 | * Must not load anything in the rbtree, | |
2557 | * mpol_free_shared_policy will not be called. | |
2558 | */ | |
2559 | mpol_shared_policy_init(&info->policy, NULL); | |
2560 | break; | |
2561 | } | |
2562 | ||
2563 | lockdep_annotate_inode_mutex_key(inode); | |
1da177e4 LT |
2564 | return inode; |
2565 | } | |
2566 | ||
e09764cf CM |
2567 | #ifdef CONFIG_TMPFS_QUOTA |
2568 | static struct inode *shmem_get_inode(struct mnt_idmap *idmap, | |
2569 | struct super_block *sb, struct inode *dir, | |
2570 | umode_t mode, dev_t dev, unsigned long flags) | |
2571 | { | |
2572 | int err; | |
2573 | struct inode *inode; | |
2574 | ||
2575 | inode = __shmem_get_inode(idmap, sb, dir, mode, dev, flags); | |
2576 | if (IS_ERR(inode)) | |
2577 | return inode; | |
2578 | ||
2579 | err = dquot_initialize(inode); | |
2580 | if (err) | |
2581 | goto errout; | |
2582 | ||
2583 | err = dquot_alloc_inode(inode); | |
2584 | if (err) { | |
2585 | dquot_drop(inode); | |
2586 | goto errout; | |
2587 | } | |
2588 | return inode; | |
2589 | ||
2590 | errout: | |
2591 | inode->i_flags |= S_NOQUOTA; | |
2592 | iput(inode); | |
2593 | return ERR_PTR(err); | |
2594 | } | |
2595 | #else | |
2596 | static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap, | |
2597 | struct super_block *sb, struct inode *dir, | |
2598 | umode_t mode, dev_t dev, unsigned long flags) | |
2599 | { | |
2600 | return __shmem_get_inode(idmap, sb, dir, mode, dev, flags); | |
2601 | } | |
2602 | #endif /* CONFIG_TMPFS_QUOTA */ | |
2603 | ||
3460f6e5 | 2604 | #ifdef CONFIG_USERFAULTFD |
61c50040 | 2605 | int shmem_mfill_atomic_pte(pmd_t *dst_pmd, |
3460f6e5 AR |
2606 | struct vm_area_struct *dst_vma, |
2607 | unsigned long dst_addr, | |
2608 | unsigned long src_addr, | |
d9712937 | 2609 | uffd_flags_t flags, |
d7be6d7e | 2610 | struct folio **foliop) |
4c27fe4c MR |
2611 | { |
2612 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2613 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4c27fe4c MR |
2614 | struct address_space *mapping = inode->i_mapping; |
2615 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2616 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
4c27fe4c | 2617 | void *page_kaddr; |
b7dd44a1 | 2618 | struct folio *folio; |
4c27fe4c | 2619 | int ret; |
3460f6e5 | 2620 | pgoff_t max_off; |
4c27fe4c | 2621 | |
4199f51a | 2622 | if (shmem_inode_acct_blocks(inode, 1)) { |
7ed9d238 AR |
2623 | /* |
2624 | * We may have got a page, returned -ENOENT triggering a retry, | |
2625 | * and now we find ourselves with -ENOMEM. Release the page, to | |
2626 | * avoid a BUG_ON in our caller. | |
2627 | */ | |
d7be6d7e Z |
2628 | if (unlikely(*foliop)) { |
2629 | folio_put(*foliop); | |
2630 | *foliop = NULL; | |
7ed9d238 | 2631 | } |
7d64ae3a | 2632 | return -ENOMEM; |
7ed9d238 | 2633 | } |
4c27fe4c | 2634 | |
d7be6d7e | 2635 | if (!*foliop) { |
7d64ae3a | 2636 | ret = -ENOMEM; |
7a7256d5 MWO |
2637 | folio = shmem_alloc_folio(gfp, info, pgoff); |
2638 | if (!folio) | |
0f079694 | 2639 | goto out_unacct_blocks; |
4c27fe4c | 2640 | |
d9712937 | 2641 | if (uffd_flags_mode_is(flags, MFILL_ATOMIC_COPY)) { |
7a7256d5 | 2642 | page_kaddr = kmap_local_folio(folio, 0); |
5dc21f0c IW |
2643 | /* |
2644 | * The read mmap_lock is held here. Despite the | |
2645 | * mmap_lock being read recursive a deadlock is still | |
2646 | * possible if a writer has taken a lock. For example: | |
2647 | * | |
2648 | * process A thread 1 takes read lock on own mmap_lock | |
2649 | * process A thread 2 calls mmap, blocks taking write lock | |
2650 | * process B thread 1 takes page fault, read lock on own mmap lock | |
2651 | * process B thread 2 calls mmap, blocks taking write lock | |
2652 | * process A thread 1 blocks taking read lock on process B | |
2653 | * process B thread 1 blocks taking read lock on process A | |
2654 | * | |
2655 | * Disable page faults to prevent potential deadlock | |
2656 | * and retry the copy outside the mmap_lock. | |
2657 | */ | |
2658 | pagefault_disable(); | |
8d103963 MR |
2659 | ret = copy_from_user(page_kaddr, |
2660 | (const void __user *)src_addr, | |
2661 | PAGE_SIZE); | |
5dc21f0c | 2662 | pagefault_enable(); |
7a7256d5 | 2663 | kunmap_local(page_kaddr); |
8d103963 | 2664 | |
c1e8d7c6 | 2665 | /* fallback to copy_from_user outside mmap_lock */ |
8d103963 | 2666 | if (unlikely(ret)) { |
d7be6d7e | 2667 | *foliop = folio; |
7d64ae3a | 2668 | ret = -ENOENT; |
8d103963 | 2669 | /* don't free the page */ |
7d64ae3a | 2670 | goto out_unacct_blocks; |
8d103963 | 2671 | } |
19b482c2 | 2672 | |
7a7256d5 | 2673 | flush_dcache_folio(folio); |
3460f6e5 | 2674 | } else { /* ZEROPAGE */ |
7a7256d5 | 2675 | clear_user_highpage(&folio->page, dst_addr); |
4c27fe4c MR |
2676 | } |
2677 | } else { | |
d7be6d7e | 2678 | folio = *foliop; |
7a7256d5 | 2679 | VM_BUG_ON_FOLIO(folio_test_large(folio), folio); |
d7be6d7e | 2680 | *foliop = NULL; |
4c27fe4c MR |
2681 | } |
2682 | ||
7a7256d5 MWO |
2683 | VM_BUG_ON(folio_test_locked(folio)); |
2684 | VM_BUG_ON(folio_test_swapbacked(folio)); | |
2685 | __folio_set_locked(folio); | |
2686 | __folio_set_swapbacked(folio); | |
2687 | __folio_mark_uptodate(folio); | |
9cc90c66 | 2688 | |
e2a50c1f | 2689 | ret = -EFAULT; |
e2a50c1f | 2690 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3460f6e5 | 2691 | if (unlikely(pgoff >= max_off)) |
e2a50c1f AA |
2692 | goto out_release; |
2693 | ||
054a9f7c HD |
2694 | ret = mem_cgroup_charge(folio, dst_vma->vm_mm, gfp); |
2695 | if (ret) | |
2696 | goto out_release; | |
2697 | ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL, gfp); | |
4c27fe4c | 2698 | if (ret) |
3fea5a49 | 2699 | goto out_release; |
4c27fe4c | 2700 | |
61c50040 | 2701 | ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr, |
d9712937 | 2702 | &folio->page, true, flags); |
7d64ae3a AR |
2703 | if (ret) |
2704 | goto out_delete_from_cache; | |
4c27fe4c | 2705 | |
3c1b7528 | 2706 | shmem_recalc_inode(inode, 1, 0); |
7a7256d5 | 2707 | folio_unlock(folio); |
7d64ae3a AR |
2708 | return 0; |
2709 | out_delete_from_cache: | |
7a7256d5 | 2710 | filemap_remove_folio(folio); |
4c27fe4c | 2711 | out_release: |
7a7256d5 MWO |
2712 | folio_unlock(folio); |
2713 | folio_put(folio); | |
4c27fe4c | 2714 | out_unacct_blocks: |
0f079694 | 2715 | shmem_inode_unacct_blocks(inode, 1); |
7d64ae3a | 2716 | return ret; |
8d103963 | 2717 | } |
3460f6e5 | 2718 | #endif /* CONFIG_USERFAULTFD */ |
8d103963 | 2719 | |
1da177e4 | 2720 | #ifdef CONFIG_TMPFS |
92e1d5be | 2721 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2722 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2723 | |
1da177e4 | 2724 | static int |
800d15a5 | 2725 | shmem_write_begin(struct file *file, struct address_space *mapping, |
9d6b0cd7 | 2726 | loff_t pos, unsigned len, |
800d15a5 | 2727 | struct page **pagep, void **fsdata) |
1da177e4 | 2728 | { |
800d15a5 | 2729 | struct inode *inode = mapping->host; |
40e041a2 | 2730 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2731 | pgoff_t index = pos >> PAGE_SHIFT; |
eff1f906 | 2732 | struct folio *folio; |
a7605426 | 2733 | int ret = 0; |
40e041a2 | 2734 | |
9608703e | 2735 | /* i_rwsem is held by caller */ |
ab3948f5 JFG |
2736 | if (unlikely(info->seals & (F_SEAL_GROW | |
2737 | F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) { | |
2738 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) | |
40e041a2 DR |
2739 | return -EPERM; |
2740 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2741 | return -EPERM; | |
2742 | } | |
2743 | ||
eff1f906 | 2744 | ret = shmem_get_folio(inode, index, &folio, SGP_WRITE); |
a7605426 YS |
2745 | if (ret) |
2746 | return ret; | |
2747 | ||
eff1f906 | 2748 | *pagep = folio_file_page(folio, index); |
a7605426 | 2749 | if (PageHWPoison(*pagep)) { |
eff1f906 MWO |
2750 | folio_unlock(folio); |
2751 | folio_put(folio); | |
a7605426 YS |
2752 | *pagep = NULL; |
2753 | return -EIO; | |
2754 | } | |
2755 | ||
2756 | return 0; | |
800d15a5 NP |
2757 | } |
2758 | ||
2759 | static int | |
2760 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2761 | loff_t pos, unsigned len, unsigned copied, | |
2762 | struct page *page, void *fsdata) | |
2763 | { | |
69bbb87b | 2764 | struct folio *folio = page_folio(page); |
800d15a5 NP |
2765 | struct inode *inode = mapping->host; |
2766 | ||
d3602444 HD |
2767 | if (pos + copied > inode->i_size) |
2768 | i_size_write(inode, pos + copied); | |
2769 | ||
69bbb87b MWO |
2770 | if (!folio_test_uptodate(folio)) { |
2771 | if (copied < folio_size(folio)) { | |
2772 | size_t from = offset_in_folio(folio, pos); | |
2773 | folio_zero_segments(folio, 0, from, | |
2774 | from + copied, folio_size(folio)); | |
ec9516fb | 2775 | } |
69bbb87b | 2776 | folio_mark_uptodate(folio); |
ec9516fb | 2777 | } |
69bbb87b MWO |
2778 | folio_mark_dirty(folio); |
2779 | folio_unlock(folio); | |
2780 | folio_put(folio); | |
800d15a5 | 2781 | |
800d15a5 | 2782 | return copied; |
1da177e4 LT |
2783 | } |
2784 | ||
2ba5bbed | 2785 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2786 | { |
6e58e79d AV |
2787 | struct file *file = iocb->ki_filp; |
2788 | struct inode *inode = file_inode(file); | |
1da177e4 | 2789 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2790 | pgoff_t index; |
2791 | unsigned long offset; | |
f7c1d074 | 2792 | int error = 0; |
cb66a7a1 | 2793 | ssize_t retval = 0; |
6e58e79d | 2794 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 | 2795 | |
09cbfeaf KS |
2796 | index = *ppos >> PAGE_SHIFT; |
2797 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2798 | |
2799 | for (;;) { | |
4601e2fc | 2800 | struct folio *folio = NULL; |
1da177e4 | 2801 | struct page *page = NULL; |
41ffe5d5 HD |
2802 | pgoff_t end_index; |
2803 | unsigned long nr, ret; | |
1da177e4 LT |
2804 | loff_t i_size = i_size_read(inode); |
2805 | ||
09cbfeaf | 2806 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2807 | if (index > end_index) |
2808 | break; | |
2809 | if (index == end_index) { | |
09cbfeaf | 2810 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2811 | if (nr <= offset) |
2812 | break; | |
2813 | } | |
2814 | ||
4601e2fc | 2815 | error = shmem_get_folio(inode, index, &folio, SGP_READ); |
6e58e79d AV |
2816 | if (error) { |
2817 | if (error == -EINVAL) | |
2818 | error = 0; | |
1da177e4 LT |
2819 | break; |
2820 | } | |
4601e2fc MWO |
2821 | if (folio) { |
2822 | folio_unlock(folio); | |
a7605426 | 2823 | |
4601e2fc | 2824 | page = folio_file_page(folio, index); |
a7605426 | 2825 | if (PageHWPoison(page)) { |
4601e2fc | 2826 | folio_put(folio); |
a7605426 YS |
2827 | error = -EIO; |
2828 | break; | |
2829 | } | |
75edd345 | 2830 | } |
1da177e4 LT |
2831 | |
2832 | /* | |
2833 | * We must evaluate after, since reads (unlike writes) | |
9608703e | 2834 | * are called without i_rwsem protection against truncate |
1da177e4 | 2835 | */ |
09cbfeaf | 2836 | nr = PAGE_SIZE; |
1da177e4 | 2837 | i_size = i_size_read(inode); |
09cbfeaf | 2838 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2839 | if (index == end_index) { |
09cbfeaf | 2840 | nr = i_size & ~PAGE_MASK; |
1da177e4 | 2841 | if (nr <= offset) { |
4601e2fc MWO |
2842 | if (folio) |
2843 | folio_put(folio); | |
1da177e4 LT |
2844 | break; |
2845 | } | |
2846 | } | |
2847 | nr -= offset; | |
2848 | ||
4601e2fc | 2849 | if (folio) { |
1da177e4 LT |
2850 | /* |
2851 | * If users can be writing to this page using arbitrary | |
2852 | * virtual addresses, take care about potential aliasing | |
2853 | * before reading the page on the kernel side. | |
2854 | */ | |
2855 | if (mapping_writably_mapped(mapping)) | |
2856 | flush_dcache_page(page); | |
2857 | /* | |
2858 | * Mark the page accessed if we read the beginning. | |
2859 | */ | |
2860 | if (!offset) | |
4601e2fc | 2861 | folio_mark_accessed(folio); |
1bdec44b HD |
2862 | /* |
2863 | * Ok, we have the page, and it's up-to-date, so | |
2864 | * now we can copy it to user space... | |
2865 | */ | |
2866 | ret = copy_page_to_iter(page, offset, nr, to); | |
4601e2fc | 2867 | folio_put(folio); |
1bdec44b | 2868 | |
fcb14cb1 | 2869 | } else if (user_backed_iter(to)) { |
1bdec44b HD |
2870 | /* |
2871 | * Copy to user tends to be so well optimized, but | |
2872 | * clear_user() not so much, that it is noticeably | |
2873 | * faster to copy the zero page instead of clearing. | |
2874 | */ | |
2875 | ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to); | |
b5810039 | 2876 | } else { |
1bdec44b HD |
2877 | /* |
2878 | * But submitting the same page twice in a row to | |
2879 | * splice() - or others? - can result in confusion: | |
2880 | * so don't attempt that optimization on pipes etc. | |
2881 | */ | |
2882 | ret = iov_iter_zero(nr, to); | |
b5810039 | 2883 | } |
1da177e4 | 2884 | |
6e58e79d | 2885 | retval += ret; |
1da177e4 | 2886 | offset += ret; |
09cbfeaf KS |
2887 | index += offset >> PAGE_SHIFT; |
2888 | offset &= ~PAGE_MASK; | |
1da177e4 | 2889 | |
2ba5bbed | 2890 | if (!iov_iter_count(to)) |
1da177e4 | 2891 | break; |
6e58e79d AV |
2892 | if (ret < nr) { |
2893 | error = -EFAULT; | |
2894 | break; | |
2895 | } | |
1da177e4 LT |
2896 | cond_resched(); |
2897 | } | |
2898 | ||
09cbfeaf | 2899 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2900 | file_accessed(file); |
2901 | return retval ? retval : error; | |
1da177e4 LT |
2902 | } |
2903 | ||
e88e0d36 HD |
2904 | static ssize_t shmem_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
2905 | { | |
2906 | struct file *file = iocb->ki_filp; | |
2907 | struct inode *inode = file->f_mapping->host; | |
2908 | ssize_t ret; | |
2909 | ||
2910 | inode_lock(inode); | |
2911 | ret = generic_write_checks(iocb, from); | |
2912 | if (ret <= 0) | |
2913 | goto unlock; | |
2914 | ret = file_remove_privs(file); | |
2915 | if (ret) | |
2916 | goto unlock; | |
2917 | ret = file_update_time(file); | |
2918 | if (ret) | |
2919 | goto unlock; | |
2920 | ret = generic_perform_write(iocb, from); | |
2921 | unlock: | |
2922 | inode_unlock(inode); | |
2923 | return ret; | |
2924 | } | |
2925 | ||
bd194b18 DH |
2926 | static bool zero_pipe_buf_get(struct pipe_inode_info *pipe, |
2927 | struct pipe_buffer *buf) | |
2928 | { | |
2929 | return true; | |
2930 | } | |
2931 | ||
2932 | static void zero_pipe_buf_release(struct pipe_inode_info *pipe, | |
2933 | struct pipe_buffer *buf) | |
2934 | { | |
2935 | } | |
2936 | ||
2937 | static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe, | |
2938 | struct pipe_buffer *buf) | |
2939 | { | |
2940 | return false; | |
2941 | } | |
2942 | ||
2943 | static const struct pipe_buf_operations zero_pipe_buf_ops = { | |
2944 | .release = zero_pipe_buf_release, | |
2945 | .try_steal = zero_pipe_buf_try_steal, | |
2946 | .get = zero_pipe_buf_get, | |
2947 | }; | |
2948 | ||
2949 | static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe, | |
2950 | loff_t fpos, size_t size) | |
2951 | { | |
2952 | size_t offset = fpos & ~PAGE_MASK; | |
2953 | ||
2954 | size = min_t(size_t, size, PAGE_SIZE - offset); | |
2955 | ||
2956 | if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) { | |
2957 | struct pipe_buffer *buf = pipe_head_buf(pipe); | |
2958 | ||
2959 | *buf = (struct pipe_buffer) { | |
2960 | .ops = &zero_pipe_buf_ops, | |
2961 | .page = ZERO_PAGE(0), | |
2962 | .offset = offset, | |
2963 | .len = size, | |
2964 | }; | |
2965 | pipe->head++; | |
2966 | } | |
2967 | ||
2968 | return size; | |
2969 | } | |
2970 | ||
2971 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, | |
2972 | struct pipe_inode_info *pipe, | |
2973 | size_t len, unsigned int flags) | |
2974 | { | |
2975 | struct inode *inode = file_inode(in); | |
2976 | struct address_space *mapping = inode->i_mapping; | |
2977 | struct folio *folio = NULL; | |
2978 | size_t total_spliced = 0, used, npages, n, part; | |
2979 | loff_t isize; | |
2980 | int error = 0; | |
2981 | ||
2982 | /* Work out how much data we can actually add into the pipe */ | |
2983 | used = pipe_occupancy(pipe->head, pipe->tail); | |
2984 | npages = max_t(ssize_t, pipe->max_usage - used, 0); | |
2985 | len = min_t(size_t, len, npages * PAGE_SIZE); | |
2986 | ||
2987 | do { | |
2988 | if (*ppos >= i_size_read(inode)) | |
2989 | break; | |
2990 | ||
fa598952 HD |
2991 | error = shmem_get_folio(inode, *ppos / PAGE_SIZE, &folio, |
2992 | SGP_READ); | |
bd194b18 DH |
2993 | if (error) { |
2994 | if (error == -EINVAL) | |
2995 | error = 0; | |
2996 | break; | |
2997 | } | |
2998 | if (folio) { | |
2999 | folio_unlock(folio); | |
3000 | ||
fa598952 HD |
3001 | if (folio_test_hwpoison(folio) || |
3002 | (folio_test_large(folio) && | |
3003 | folio_test_has_hwpoisoned(folio))) { | |
bd194b18 DH |
3004 | error = -EIO; |
3005 | break; | |
3006 | } | |
3007 | } | |
3008 | ||
3009 | /* | |
3010 | * i_size must be checked after we know the pages are Uptodate. | |
3011 | * | |
3012 | * Checking i_size after the check allows us to calculate | |
3013 | * the correct value for "nr", which means the zero-filled | |
3014 | * part of the page is not copied back to userspace (unless | |
3015 | * another truncate extends the file - this is desired though). | |
3016 | */ | |
3017 | isize = i_size_read(inode); | |
3018 | if (unlikely(*ppos >= isize)) | |
3019 | break; | |
3020 | part = min_t(loff_t, isize - *ppos, len); | |
3021 | ||
3022 | if (folio) { | |
3023 | /* | |
3024 | * If users can be writing to this page using arbitrary | |
3025 | * virtual addresses, take care about potential aliasing | |
3026 | * before reading the page on the kernel side. | |
3027 | */ | |
3028 | if (mapping_writably_mapped(mapping)) | |
3029 | flush_dcache_folio(folio); | |
3030 | folio_mark_accessed(folio); | |
3031 | /* | |
3032 | * Ok, we have the page, and it's up-to-date, so we can | |
3033 | * now splice it into the pipe. | |
3034 | */ | |
3035 | n = splice_folio_into_pipe(pipe, folio, *ppos, part); | |
3036 | folio_put(folio); | |
3037 | folio = NULL; | |
3038 | } else { | |
fa598952 | 3039 | n = splice_zeropage_into_pipe(pipe, *ppos, part); |
bd194b18 DH |
3040 | } |
3041 | ||
3042 | if (!n) | |
3043 | break; | |
3044 | len -= n; | |
3045 | total_spliced += n; | |
3046 | *ppos += n; | |
3047 | in->f_ra.prev_pos = *ppos; | |
3048 | if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) | |
3049 | break; | |
3050 | ||
3051 | cond_resched(); | |
3052 | } while (len); | |
3053 | ||
3054 | if (folio) | |
3055 | folio_put(folio); | |
3056 | ||
3057 | file_accessed(in); | |
3058 | return total_spliced ? total_spliced : error; | |
3059 | } | |
3060 | ||
965c8e59 | 3061 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
3062 | { |
3063 | struct address_space *mapping = file->f_mapping; | |
3064 | struct inode *inode = mapping->host; | |
220f2ac9 | 3065 | |
965c8e59 AM |
3066 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
3067 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 3068 | MAX_LFS_FILESIZE, i_size_read(inode)); |
41139aa4 MWO |
3069 | if (offset < 0) |
3070 | return -ENXIO; | |
3071 | ||
5955102c | 3072 | inode_lock(inode); |
9608703e | 3073 | /* We're holding i_rwsem so we can access i_size directly */ |
41139aa4 | 3074 | offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence); |
387aae6f HD |
3075 | if (offset >= 0) |
3076 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 3077 | inode_unlock(inode); |
220f2ac9 HD |
3078 | return offset; |
3079 | } | |
3080 | ||
83e4fa9c HD |
3081 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
3082 | loff_t len) | |
3083 | { | |
496ad9aa | 3084 | struct inode *inode = file_inode(file); |
e2d12e22 | 3085 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 3086 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 3087 | struct shmem_falloc shmem_falloc; |
d144bf62 | 3088 | pgoff_t start, index, end, undo_fallocend; |
e2d12e22 | 3089 | int error; |
83e4fa9c | 3090 | |
13ace4d0 HD |
3091 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
3092 | return -EOPNOTSUPP; | |
3093 | ||
5955102c | 3094 | inode_lock(inode); |
83e4fa9c HD |
3095 | |
3096 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
3097 | struct address_space *mapping = file->f_mapping; | |
3098 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
3099 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 3100 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 3101 | |
9608703e | 3102 | /* protected by i_rwsem */ |
ab3948f5 | 3103 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { |
40e041a2 DR |
3104 | error = -EPERM; |
3105 | goto out; | |
3106 | } | |
3107 | ||
8e205f77 | 3108 | shmem_falloc.waitq = &shmem_falloc_waitq; |
aa71ecd8 | 3109 | shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT; |
f00cdc6d HD |
3110 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; |
3111 | spin_lock(&inode->i_lock); | |
3112 | inode->i_private = &shmem_falloc; | |
3113 | spin_unlock(&inode->i_lock); | |
3114 | ||
83e4fa9c HD |
3115 | if ((u64)unmap_end > (u64)unmap_start) |
3116 | unmap_mapping_range(mapping, unmap_start, | |
3117 | 1 + unmap_end - unmap_start, 0); | |
3118 | shmem_truncate_range(inode, offset, offset + len - 1); | |
3119 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
3120 | |
3121 | spin_lock(&inode->i_lock); | |
3122 | inode->i_private = NULL; | |
3123 | wake_up_all(&shmem_falloc_waitq); | |
2055da97 | 3124 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head)); |
8e205f77 | 3125 | spin_unlock(&inode->i_lock); |
83e4fa9c | 3126 | error = 0; |
8e205f77 | 3127 | goto out; |
e2d12e22 HD |
3128 | } |
3129 | ||
3130 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
3131 | error = inode_newsize_ok(inode, offset + len); | |
3132 | if (error) | |
3133 | goto out; | |
3134 | ||
40e041a2 DR |
3135 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
3136 | error = -EPERM; | |
3137 | goto out; | |
3138 | } | |
3139 | ||
09cbfeaf KS |
3140 | start = offset >> PAGE_SHIFT; |
3141 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
3142 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
3143 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
3144 | error = -ENOSPC; | |
3145 | goto out; | |
83e4fa9c HD |
3146 | } |
3147 | ||
8e205f77 | 3148 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
3149 | shmem_falloc.start = start; |
3150 | shmem_falloc.next = start; | |
3151 | shmem_falloc.nr_falloced = 0; | |
3152 | shmem_falloc.nr_unswapped = 0; | |
3153 | spin_lock(&inode->i_lock); | |
3154 | inode->i_private = &shmem_falloc; | |
3155 | spin_unlock(&inode->i_lock); | |
3156 | ||
d144bf62 HD |
3157 | /* |
3158 | * info->fallocend is only relevant when huge pages might be | |
3159 | * involved: to prevent split_huge_page() freeing fallocated | |
3160 | * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size. | |
3161 | */ | |
3162 | undo_fallocend = info->fallocend; | |
3163 | if (info->fallocend < end) | |
3164 | info->fallocend = end; | |
3165 | ||
050dcb5c | 3166 | for (index = start; index < end; ) { |
b0802b22 | 3167 | struct folio *folio; |
e2d12e22 HD |
3168 | |
3169 | /* | |
3170 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
3171 | * been interrupted because we are using up too much memory. | |
3172 | */ | |
3173 | if (signal_pending(current)) | |
3174 | error = -EINTR; | |
1aac1400 HD |
3175 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
3176 | error = -ENOMEM; | |
e2d12e22 | 3177 | else |
b0802b22 MWO |
3178 | error = shmem_get_folio(inode, index, &folio, |
3179 | SGP_FALLOC); | |
e2d12e22 | 3180 | if (error) { |
d144bf62 | 3181 | info->fallocend = undo_fallocend; |
b0802b22 | 3182 | /* Remove the !uptodate folios we added */ |
7f556567 HD |
3183 | if (index > start) { |
3184 | shmem_undo_range(inode, | |
3185 | (loff_t)start << PAGE_SHIFT, | |
3186 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
3187 | } | |
1aac1400 | 3188 | goto undone; |
e2d12e22 HD |
3189 | } |
3190 | ||
050dcb5c HD |
3191 | /* |
3192 | * Here is a more important optimization than it appears: | |
b0802b22 MWO |
3193 | * a second SGP_FALLOC on the same large folio will clear it, |
3194 | * making it uptodate and un-undoable if we fail later. | |
050dcb5c | 3195 | */ |
b0802b22 MWO |
3196 | index = folio_next_index(folio); |
3197 | /* Beware 32-bit wraparound */ | |
3198 | if (!index) | |
3199 | index--; | |
050dcb5c | 3200 | |
1aac1400 HD |
3201 | /* |
3202 | * Inform shmem_writepage() how far we have reached. | |
3203 | * No need for lock or barrier: we have the page lock. | |
3204 | */ | |
b0802b22 | 3205 | if (!folio_test_uptodate(folio)) |
050dcb5c HD |
3206 | shmem_falloc.nr_falloced += index - shmem_falloc.next; |
3207 | shmem_falloc.next = index; | |
1aac1400 | 3208 | |
e2d12e22 | 3209 | /* |
b0802b22 | 3210 | * If !uptodate, leave it that way so that freeable folios |
1635f6a7 | 3211 | * can be recognized if we need to rollback on error later. |
b0802b22 MWO |
3212 | * But mark it dirty so that memory pressure will swap rather |
3213 | * than free the folios we are allocating (and SGP_CACHE folios | |
e2d12e22 HD |
3214 | * might still be clean: we now need to mark those dirty too). |
3215 | */ | |
b0802b22 MWO |
3216 | folio_mark_dirty(folio); |
3217 | folio_unlock(folio); | |
3218 | folio_put(folio); | |
e2d12e22 HD |
3219 | cond_resched(); |
3220 | } | |
3221 | ||
3222 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
3223 | i_size_write(inode, offset + len); | |
1aac1400 HD |
3224 | undone: |
3225 | spin_lock(&inode->i_lock); | |
3226 | inode->i_private = NULL; | |
3227 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 3228 | out: |
15f242bb HD |
3229 | if (!error) |
3230 | file_modified(file); | |
5955102c | 3231 | inode_unlock(inode); |
83e4fa9c HD |
3232 | return error; |
3233 | } | |
3234 | ||
726c3342 | 3235 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 3236 | { |
726c3342 | 3237 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
3238 | |
3239 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 3240 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 3241 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 3242 | if (sbinfo->max_blocks) { |
1da177e4 | 3243 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
3244 | buf->f_bavail = |
3245 | buf->f_bfree = sbinfo->max_blocks - | |
3246 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
3247 | } |
3248 | if (sbinfo->max_inodes) { | |
1da177e4 | 3249 | buf->f_files = sbinfo->max_inodes; |
e07c469e | 3250 | buf->f_ffree = sbinfo->free_ispace / BOGO_INODE_SIZE; |
1da177e4 LT |
3251 | } |
3252 | /* else leave those fields 0 like simple_statfs */ | |
59cda49e AG |
3253 | |
3254 | buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b); | |
3255 | ||
1da177e4 LT |
3256 | return 0; |
3257 | } | |
3258 | ||
3259 | /* | |
3260 | * File creation. Allocate an inode, and we're done.. | |
3261 | */ | |
3262 | static int | |
5ebb29be | 3263 | shmem_mknod(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 3264 | struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3265 | { |
0b0a0806 | 3266 | struct inode *inode; |
71480663 | 3267 | int error; |
1da177e4 | 3268 | |
7a80e5b8 | 3269 | inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE); |
71480663 CM |
3270 | if (IS_ERR(inode)) |
3271 | return PTR_ERR(inode); | |
37ec43cd | 3272 | |
71480663 CM |
3273 | error = simple_acl_create(dir, inode); |
3274 | if (error) | |
3275 | goto out_iput; | |
9be7d5b0 | 3276 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
71480663 CM |
3277 | shmem_initxattrs, NULL); |
3278 | if (error && error != -EOPNOTSUPP) | |
3279 | goto out_iput; | |
3280 | ||
a2e45955 CL |
3281 | error = simple_offset_add(shmem_get_offset_ctx(dir), dentry); |
3282 | if (error) | |
3283 | goto out_iput; | |
3284 | ||
71480663 | 3285 | dir->i_size += BOGO_DIRENT_SIZE; |
cf2766bb | 3286 | inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir)); |
71480663 CM |
3287 | inode_inc_iversion(dir); |
3288 | d_instantiate(dentry, inode); | |
3289 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 | 3290 | return error; |
71480663 | 3291 | |
feda821e CH |
3292 | out_iput: |
3293 | iput(inode); | |
3294 | return error; | |
1da177e4 LT |
3295 | } |
3296 | ||
60545d0d | 3297 | static int |
011e2b71 | 3298 | shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir, |
863f144f | 3299 | struct file *file, umode_t mode) |
60545d0d AV |
3300 | { |
3301 | struct inode *inode; | |
71480663 | 3302 | int error; |
60545d0d | 3303 | |
7a80e5b8 | 3304 | inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE); |
71480663 CM |
3305 | if (IS_ERR(inode)) { |
3306 | error = PTR_ERR(inode); | |
3307 | goto err_out; | |
60545d0d | 3308 | } |
9be7d5b0 | 3309 | error = security_inode_init_security(inode, dir, NULL, |
71480663 CM |
3310 | shmem_initxattrs, NULL); |
3311 | if (error && error != -EOPNOTSUPP) | |
3312 | goto out_iput; | |
3313 | error = simple_acl_create(dir, inode); | |
3314 | if (error) | |
3315 | goto out_iput; | |
3316 | d_tmpfile(file, inode); | |
3317 | ||
3318 | err_out: | |
863f144f | 3319 | return finish_open_simple(file, error); |
feda821e CH |
3320 | out_iput: |
3321 | iput(inode); | |
3322 | return error; | |
60545d0d AV |
3323 | } |
3324 | ||
c54bd91e | 3325 | static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 3326 | struct dentry *dentry, umode_t mode) |
1da177e4 LT |
3327 | { |
3328 | int error; | |
3329 | ||
7a80e5b8 GS |
3330 | error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0); |
3331 | if (error) | |
1da177e4 | 3332 | return error; |
d8c76e6f | 3333 | inc_nlink(dir); |
1da177e4 LT |
3334 | return 0; |
3335 | } | |
3336 | ||
6c960e68 | 3337 | static int shmem_create(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 3338 | struct dentry *dentry, umode_t mode, bool excl) |
1da177e4 | 3339 | { |
7a80e5b8 | 3340 | return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0); |
1da177e4 LT |
3341 | } |
3342 | ||
3343 | /* | |
3344 | * Link a file.. | |
3345 | */ | |
9be7d5b0 HD |
3346 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, |
3347 | struct dentry *dentry) | |
1da177e4 | 3348 | { |
75c3cfa8 | 3349 | struct inode *inode = d_inode(old_dentry); |
29b00e60 | 3350 | int ret = 0; |
1da177e4 LT |
3351 | |
3352 | /* | |
3353 | * No ordinary (disk based) filesystem counts links as inodes; | |
3354 | * but each new link needs a new dentry, pinning lowmem, and | |
3355 | * tmpfs dentries cannot be pruned until they are unlinked. | |
1062af92 DW |
3356 | * But if an O_TMPFILE file is linked into the tmpfs, the |
3357 | * first link must skip that, to get the accounting right. | |
1da177e4 | 3358 | */ |
1062af92 | 3359 | if (inode->i_nlink) { |
e809d5f0 | 3360 | ret = shmem_reserve_inode(inode->i_sb, NULL); |
1062af92 DW |
3361 | if (ret) |
3362 | goto out; | |
3363 | } | |
1da177e4 | 3364 | |
a2e45955 CL |
3365 | ret = simple_offset_add(shmem_get_offset_ctx(dir), dentry); |
3366 | if (ret) { | |
3367 | if (inode->i_nlink) | |
2daf18a7 | 3368 | shmem_free_inode(inode->i_sb, 0); |
a2e45955 CL |
3369 | goto out; |
3370 | } | |
3371 | ||
1da177e4 | 3372 | dir->i_size += BOGO_DIRENT_SIZE; |
cf2766bb JL |
3373 | inode_set_mtime_to_ts(dir, |
3374 | inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode))); | |
36f05cab | 3375 | inode_inc_iversion(dir); |
d8c76e6f | 3376 | inc_nlink(inode); |
7de9c6ee | 3377 | ihold(inode); /* New dentry reference */ |
9be7d5b0 | 3378 | dget(dentry); /* Extra pinning count for the created dentry */ |
1da177e4 | 3379 | d_instantiate(dentry, inode); |
5b04c689 PE |
3380 | out: |
3381 | return ret; | |
1da177e4 LT |
3382 | } |
3383 | ||
3384 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
3385 | { | |
75c3cfa8 | 3386 | struct inode *inode = d_inode(dentry); |
1da177e4 | 3387 | |
5b04c689 | 3388 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2daf18a7 | 3389 | shmem_free_inode(inode->i_sb, 0); |
1da177e4 | 3390 | |
a2e45955 | 3391 | simple_offset_remove(shmem_get_offset_ctx(dir), dentry); |
1da177e4 LT |
3392 | |
3393 | dir->i_size -= BOGO_DIRENT_SIZE; | |
cf2766bb JL |
3394 | inode_set_mtime_to_ts(dir, |
3395 | inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode))); | |
36f05cab | 3396 | inode_inc_iversion(dir); |
9a53c3a7 | 3397 | drop_nlink(inode); |
9be7d5b0 | 3398 | dput(dentry); /* Undo the count from "create" - does all the work */ |
1da177e4 LT |
3399 | return 0; |
3400 | } | |
3401 | ||
3402 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
3403 | { | |
ecba88a3 | 3404 | if (!simple_offset_empty(dentry)) |
1da177e4 LT |
3405 | return -ENOTEMPTY; |
3406 | ||
75c3cfa8 | 3407 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 3408 | drop_nlink(dir); |
1da177e4 LT |
3409 | return shmem_unlink(dir, dentry); |
3410 | } | |
3411 | ||
e18275ae | 3412 | static int shmem_whiteout(struct mnt_idmap *idmap, |
549c7297 | 3413 | struct inode *old_dir, struct dentry *old_dentry) |
46fdb794 MS |
3414 | { |
3415 | struct dentry *whiteout; | |
3416 | int error; | |
3417 | ||
3418 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
3419 | if (!whiteout) | |
3420 | return -ENOMEM; | |
3421 | ||
7a80e5b8 | 3422 | error = shmem_mknod(idmap, old_dir, whiteout, |
46fdb794 MS |
3423 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); |
3424 | dput(whiteout); | |
3425 | if (error) | |
3426 | return error; | |
3427 | ||
3428 | /* | |
3429 | * Cheat and hash the whiteout while the old dentry is still in | |
3430 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
3431 | * | |
3432 | * d_lookup() will consistently find one of them at this point, | |
3433 | * not sure which one, but that isn't even important. | |
3434 | */ | |
3435 | d_rehash(whiteout); | |
3436 | return 0; | |
3437 | } | |
3438 | ||
1da177e4 LT |
3439 | /* |
3440 | * The VFS layer already does all the dentry stuff for rename, | |
3441 | * we just have to decrement the usage count for the target if | |
3442 | * it exists so that the VFS layer correctly free's it when it | |
3443 | * gets overwritten. | |
3444 | */ | |
e18275ae | 3445 | static int shmem_rename2(struct mnt_idmap *idmap, |
549c7297 CB |
3446 | struct inode *old_dir, struct dentry *old_dentry, |
3447 | struct inode *new_dir, struct dentry *new_dentry, | |
3448 | unsigned int flags) | |
1da177e4 | 3449 | { |
75c3cfa8 | 3450 | struct inode *inode = d_inode(old_dentry); |
1da177e4 | 3451 | int they_are_dirs = S_ISDIR(inode->i_mode); |
a2e45955 | 3452 | int error; |
1da177e4 | 3453 | |
46fdb794 | 3454 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
3455 | return -EINVAL; |
3456 | ||
37456771 | 3457 | if (flags & RENAME_EXCHANGE) |
a2e45955 CL |
3458 | return simple_offset_rename_exchange(old_dir, old_dentry, |
3459 | new_dir, new_dentry); | |
37456771 | 3460 | |
ecba88a3 | 3461 | if (!simple_offset_empty(new_dentry)) |
1da177e4 LT |
3462 | return -ENOTEMPTY; |
3463 | ||
46fdb794 | 3464 | if (flags & RENAME_WHITEOUT) { |
7a80e5b8 | 3465 | error = shmem_whiteout(idmap, old_dir, old_dentry); |
46fdb794 MS |
3466 | if (error) |
3467 | return error; | |
3468 | } | |
3469 | ||
5a1a25be | 3470 | error = simple_offset_rename(old_dir, old_dentry, new_dir, new_dentry); |
a2e45955 CL |
3471 | if (error) |
3472 | return error; | |
3473 | ||
75c3cfa8 | 3474 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 3475 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 3476 | if (they_are_dirs) { |
75c3cfa8 | 3477 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 3478 | drop_nlink(old_dir); |
b928095b | 3479 | } |
1da177e4 | 3480 | } else if (they_are_dirs) { |
9a53c3a7 | 3481 | drop_nlink(old_dir); |
d8c76e6f | 3482 | inc_nlink(new_dir); |
1da177e4 LT |
3483 | } |
3484 | ||
3485 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
3486 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
944d0d9d | 3487 | simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry); |
36f05cab JL |
3488 | inode_inc_iversion(old_dir); |
3489 | inode_inc_iversion(new_dir); | |
1da177e4 LT |
3490 | return 0; |
3491 | } | |
3492 | ||
7a77db95 | 3493 | static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 3494 | struct dentry *dentry, const char *symname) |
1da177e4 LT |
3495 | { |
3496 | int error; | |
3497 | int len; | |
3498 | struct inode *inode; | |
7ad0414b | 3499 | struct folio *folio; |
1da177e4 LT |
3500 | |
3501 | len = strlen(symname) + 1; | |
09cbfeaf | 3502 | if (len > PAGE_SIZE) |
1da177e4 LT |
3503 | return -ENAMETOOLONG; |
3504 | ||
7a80e5b8 | 3505 | inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0, |
0825a6f9 | 3506 | VM_NORESERVE); |
71480663 CM |
3507 | if (IS_ERR(inode)) |
3508 | return PTR_ERR(inode); | |
1da177e4 | 3509 | |
9d8f13ba | 3510 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3511 | shmem_initxattrs, NULL); |
23a31d87 CL |
3512 | if (error && error != -EOPNOTSUPP) |
3513 | goto out_iput; | |
570bc1c2 | 3514 | |
a2e45955 CL |
3515 | error = simple_offset_add(shmem_get_offset_ctx(dir), dentry); |
3516 | if (error) | |
3517 | goto out_iput; | |
570bc1c2 | 3518 | |
1da177e4 | 3519 | inode->i_size = len-1; |
69f07ec9 | 3520 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3521 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3522 | if (!inode->i_link) { | |
23a31d87 | 3523 | error = -ENOMEM; |
a2e45955 | 3524 | goto out_remove_offset; |
69f07ec9 HD |
3525 | } |
3526 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3527 | } else { |
e8ecde25 | 3528 | inode_nohighmem(inode); |
e11381d8 | 3529 | inode->i_mapping->a_ops = &shmem_aops; |
7ad0414b | 3530 | error = shmem_get_folio(inode, 0, &folio, SGP_WRITE); |
23a31d87 | 3531 | if (error) |
a2e45955 | 3532 | goto out_remove_offset; |
1da177e4 | 3533 | inode->i_op = &shmem_symlink_inode_operations; |
7ad0414b MWO |
3534 | memcpy(folio_address(folio), symname, len); |
3535 | folio_mark_uptodate(folio); | |
3536 | folio_mark_dirty(folio); | |
3537 | folio_unlock(folio); | |
3538 | folio_put(folio); | |
1da177e4 | 3539 | } |
1da177e4 | 3540 | dir->i_size += BOGO_DIRENT_SIZE; |
cf2766bb | 3541 | inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir)); |
36f05cab | 3542 | inode_inc_iversion(dir); |
1da177e4 LT |
3543 | d_instantiate(dentry, inode); |
3544 | dget(dentry); | |
3545 | return 0; | |
a2e45955 CL |
3546 | |
3547 | out_remove_offset: | |
3548 | simple_offset_remove(shmem_get_offset_ctx(dir), dentry); | |
23a31d87 CL |
3549 | out_iput: |
3550 | iput(inode); | |
3551 | return error; | |
1da177e4 LT |
3552 | } |
3553 | ||
fceef393 | 3554 | static void shmem_put_link(void *arg) |
1da177e4 | 3555 | { |
e4b57722 MWO |
3556 | folio_mark_accessed(arg); |
3557 | folio_put(arg); | |
1da177e4 LT |
3558 | } |
3559 | ||
9be7d5b0 | 3560 | static const char *shmem_get_link(struct dentry *dentry, struct inode *inode, |
fceef393 | 3561 | struct delayed_call *done) |
1da177e4 | 3562 | { |
e4b57722 | 3563 | struct folio *folio = NULL; |
6b255391 | 3564 | int error; |
e4b57722 | 3565 | |
6a6c9904 | 3566 | if (!dentry) { |
e4b57722 | 3567 | folio = filemap_get_folio(inode->i_mapping, 0); |
66dabbb6 | 3568 | if (IS_ERR(folio)) |
6a6c9904 | 3569 | return ERR_PTR(-ECHILD); |
7459c149 | 3570 | if (PageHWPoison(folio_page(folio, 0)) || |
e4b57722 MWO |
3571 | !folio_test_uptodate(folio)) { |
3572 | folio_put(folio); | |
6a6c9904 AV |
3573 | return ERR_PTR(-ECHILD); |
3574 | } | |
3575 | } else { | |
e4b57722 | 3576 | error = shmem_get_folio(inode, 0, &folio, SGP_READ); |
6a6c9904 AV |
3577 | if (error) |
3578 | return ERR_PTR(error); | |
e4b57722 | 3579 | if (!folio) |
a7605426 | 3580 | return ERR_PTR(-ECHILD); |
7459c149 | 3581 | if (PageHWPoison(folio_page(folio, 0))) { |
e4b57722 MWO |
3582 | folio_unlock(folio); |
3583 | folio_put(folio); | |
a7605426 YS |
3584 | return ERR_PTR(-ECHILD); |
3585 | } | |
e4b57722 | 3586 | folio_unlock(folio); |
6a6c9904 | 3587 | } |
e4b57722 MWO |
3588 | set_delayed_call(done, shmem_put_link, folio); |
3589 | return folio_address(folio); | |
1da177e4 LT |
3590 | } |
3591 | ||
b09e0fa4 | 3592 | #ifdef CONFIG_TMPFS_XATTR |
e408e695 TT |
3593 | |
3594 | static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa) | |
3595 | { | |
3596 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); | |
3597 | ||
3598 | fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE); | |
3599 | ||
3600 | return 0; | |
3601 | } | |
3602 | ||
8782a9ae | 3603 | static int shmem_fileattr_set(struct mnt_idmap *idmap, |
e408e695 TT |
3604 | struct dentry *dentry, struct fileattr *fa) |
3605 | { | |
3606 | struct inode *inode = d_inode(dentry); | |
3607 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3608 | ||
3609 | if (fileattr_has_fsx(fa)) | |
3610 | return -EOPNOTSUPP; | |
cb241339 HD |
3611 | if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE) |
3612 | return -EOPNOTSUPP; | |
e408e695 TT |
3613 | |
3614 | info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) | | |
3615 | (fa->flags & SHMEM_FL_USER_MODIFIABLE); | |
3616 | ||
cb241339 | 3617 | shmem_set_inode_flags(inode, info->fsflags); |
65287334 | 3618 | inode_set_ctime_current(inode); |
36f05cab | 3619 | inode_inc_iversion(inode); |
e408e695 TT |
3620 | return 0; |
3621 | } | |
3622 | ||
46711810 | 3623 | /* |
b09e0fa4 EP |
3624 | * Superblocks without xattr inode operations may get some security.* xattr |
3625 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3626 | * like ACLs, we also need to implement the security.* handlers at |
3627 | * filesystem level, though. | |
3628 | */ | |
3629 | ||
6d9d88d0 JS |
3630 | /* |
3631 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3632 | */ | |
3633 | static int shmem_initxattrs(struct inode *inode, | |
9be7d5b0 | 3634 | const struct xattr *xattr_array, void *fs_info) |
6d9d88d0 JS |
3635 | { |
3636 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2daf18a7 | 3637 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
6d9d88d0 | 3638 | const struct xattr *xattr; |
38f38657 | 3639 | struct simple_xattr *new_xattr; |
2daf18a7 | 3640 | size_t ispace = 0; |
6d9d88d0 JS |
3641 | size_t len; |
3642 | ||
2daf18a7 HD |
3643 | if (sbinfo->max_inodes) { |
3644 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
3645 | ispace += simple_xattr_space(xattr->name, | |
3646 | xattr->value_len + XATTR_SECURITY_PREFIX_LEN); | |
3647 | } | |
3648 | if (ispace) { | |
3649 | raw_spin_lock(&sbinfo->stat_lock); | |
3650 | if (sbinfo->free_ispace < ispace) | |
3651 | ispace = 0; | |
3652 | else | |
3653 | sbinfo->free_ispace -= ispace; | |
3654 | raw_spin_unlock(&sbinfo->stat_lock); | |
3655 | if (!ispace) | |
3656 | return -ENOSPC; | |
3657 | } | |
3658 | } | |
3659 | ||
6d9d88d0 | 3660 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { |
38f38657 | 3661 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 | 3662 | if (!new_xattr) |
2daf18a7 | 3663 | break; |
6d9d88d0 JS |
3664 | |
3665 | len = strlen(xattr->name) + 1; | |
3666 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
572a3d1e | 3667 | GFP_KERNEL_ACCOUNT); |
6d9d88d0 | 3668 | if (!new_xattr->name) { |
3bef735a | 3669 | kvfree(new_xattr); |
2daf18a7 | 3670 | break; |
6d9d88d0 JS |
3671 | } |
3672 | ||
3673 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3674 | XATTR_SECURITY_PREFIX_LEN); | |
3675 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3676 | xattr->name, len); | |
3677 | ||
3b4c7bc0 | 3678 | simple_xattr_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3679 | } |
3680 | ||
2daf18a7 HD |
3681 | if (xattr->name != NULL) { |
3682 | if (ispace) { | |
3683 | raw_spin_lock(&sbinfo->stat_lock); | |
3684 | sbinfo->free_ispace += ispace; | |
3685 | raw_spin_unlock(&sbinfo->stat_lock); | |
3686 | } | |
3687 | simple_xattrs_free(&info->xattrs, NULL); | |
3688 | return -ENOMEM; | |
3689 | } | |
3690 | ||
6d9d88d0 JS |
3691 | return 0; |
3692 | } | |
3693 | ||
aa7c5241 | 3694 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3695 | struct dentry *unused, struct inode *inode, |
3696 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3697 | { |
b296821a | 3698 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3699 | |
aa7c5241 | 3700 | name = xattr_full_name(handler, name); |
38f38657 | 3701 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3702 | } |
3703 | ||
aa7c5241 | 3704 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
39f60c1c | 3705 | struct mnt_idmap *idmap, |
59301226 AV |
3706 | struct dentry *unused, struct inode *inode, |
3707 | const char *name, const void *value, | |
3708 | size_t size, int flags) | |
b09e0fa4 | 3709 | { |
59301226 | 3710 | struct shmem_inode_info *info = SHMEM_I(inode); |
2daf18a7 | 3711 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
5de75970 | 3712 | struct simple_xattr *old_xattr; |
2daf18a7 | 3713 | size_t ispace = 0; |
b09e0fa4 | 3714 | |
aa7c5241 | 3715 | name = xattr_full_name(handler, name); |
2daf18a7 HD |
3716 | if (value && sbinfo->max_inodes) { |
3717 | ispace = simple_xattr_space(name, size); | |
3718 | raw_spin_lock(&sbinfo->stat_lock); | |
3719 | if (sbinfo->free_ispace < ispace) | |
3720 | ispace = 0; | |
3721 | else | |
3722 | sbinfo->free_ispace -= ispace; | |
3723 | raw_spin_unlock(&sbinfo->stat_lock); | |
3724 | if (!ispace) | |
3725 | return -ENOSPC; | |
3726 | } | |
3727 | ||
5de75970 HD |
3728 | old_xattr = simple_xattr_set(&info->xattrs, name, value, size, flags); |
3729 | if (!IS_ERR(old_xattr)) { | |
2daf18a7 HD |
3730 | ispace = 0; |
3731 | if (old_xattr && sbinfo->max_inodes) | |
3732 | ispace = simple_xattr_space(old_xattr->name, | |
3733 | old_xattr->size); | |
5de75970 HD |
3734 | simple_xattr_free(old_xattr); |
3735 | old_xattr = NULL; | |
65287334 | 3736 | inode_set_ctime_current(inode); |
36f05cab JL |
3737 | inode_inc_iversion(inode); |
3738 | } | |
2daf18a7 HD |
3739 | if (ispace) { |
3740 | raw_spin_lock(&sbinfo->stat_lock); | |
3741 | sbinfo->free_ispace += ispace; | |
3742 | raw_spin_unlock(&sbinfo->stat_lock); | |
3743 | } | |
5de75970 | 3744 | return PTR_ERR(old_xattr); |
b09e0fa4 EP |
3745 | } |
3746 | ||
aa7c5241 AG |
3747 | static const struct xattr_handler shmem_security_xattr_handler = { |
3748 | .prefix = XATTR_SECURITY_PREFIX, | |
3749 | .get = shmem_xattr_handler_get, | |
3750 | .set = shmem_xattr_handler_set, | |
3751 | }; | |
b09e0fa4 | 3752 | |
aa7c5241 AG |
3753 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3754 | .prefix = XATTR_TRUSTED_PREFIX, | |
3755 | .get = shmem_xattr_handler_get, | |
3756 | .set = shmem_xattr_handler_set, | |
3757 | }; | |
b09e0fa4 | 3758 | |
2daf18a7 HD |
3759 | static const struct xattr_handler shmem_user_xattr_handler = { |
3760 | .prefix = XATTR_USER_PREFIX, | |
3761 | .get = shmem_xattr_handler_get, | |
3762 | .set = shmem_xattr_handler_set, | |
3763 | }; | |
3764 | ||
2f502860 | 3765 | static const struct xattr_handler * const shmem_xattr_handlers[] = { |
aa7c5241 AG |
3766 | &shmem_security_xattr_handler, |
3767 | &shmem_trusted_xattr_handler, | |
2daf18a7 | 3768 | &shmem_user_xattr_handler, |
aa7c5241 AG |
3769 | NULL |
3770 | }; | |
b09e0fa4 EP |
3771 | |
3772 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3773 | { | |
75c3cfa8 | 3774 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3775 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3776 | } |
3777 | #endif /* CONFIG_TMPFS_XATTR */ | |
3778 | ||
69f07ec9 | 3779 | static const struct inode_operations shmem_short_symlink_operations = { |
f7cd16a5 | 3780 | .getattr = shmem_getattr, |
e09764cf | 3781 | .setattr = shmem_setattr, |
6b255391 | 3782 | .get_link = simple_get_link, |
b09e0fa4 | 3783 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3784 | .listxattr = shmem_listxattr, |
b09e0fa4 EP |
3785 | #endif |
3786 | }; | |
3787 | ||
3788 | static const struct inode_operations shmem_symlink_inode_operations = { | |
f7cd16a5 | 3789 | .getattr = shmem_getattr, |
e09764cf | 3790 | .setattr = shmem_setattr, |
6b255391 | 3791 | .get_link = shmem_get_link, |
b09e0fa4 | 3792 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3793 | .listxattr = shmem_listxattr, |
39f0247d | 3794 | #endif |
b09e0fa4 | 3795 | }; |
39f0247d | 3796 | |
91828a40 DG |
3797 | static struct dentry *shmem_get_parent(struct dentry *child) |
3798 | { | |
3799 | return ERR_PTR(-ESTALE); | |
3800 | } | |
3801 | ||
3802 | static int shmem_match(struct inode *ino, void *vfh) | |
3803 | { | |
3804 | __u32 *fh = vfh; | |
3805 | __u64 inum = fh[2]; | |
3806 | inum = (inum << 32) | fh[1]; | |
3807 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
3808 | } | |
3809 | ||
12ba780d AG |
3810 | /* Find any alias of inode, but prefer a hashed alias */ |
3811 | static struct dentry *shmem_find_alias(struct inode *inode) | |
3812 | { | |
3813 | struct dentry *alias = d_find_alias(inode); | |
3814 | ||
3815 | return alias ?: d_find_any_alias(inode); | |
3816 | } | |
3817 | ||
480b116c CH |
3818 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3819 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3820 | { |
91828a40 | 3821 | struct inode *inode; |
480b116c | 3822 | struct dentry *dentry = NULL; |
35c2a7f4 | 3823 | u64 inum; |
480b116c CH |
3824 | |
3825 | if (fh_len < 3) | |
3826 | return NULL; | |
91828a40 | 3827 | |
35c2a7f4 HD |
3828 | inum = fid->raw[2]; |
3829 | inum = (inum << 32) | fid->raw[1]; | |
3830 | ||
480b116c CH |
3831 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
3832 | shmem_match, fid->raw); | |
91828a40 | 3833 | if (inode) { |
12ba780d | 3834 | dentry = shmem_find_alias(inode); |
91828a40 DG |
3835 | iput(inode); |
3836 | } | |
3837 | ||
480b116c | 3838 | return dentry; |
91828a40 DG |
3839 | } |
3840 | ||
b0b0382b AV |
3841 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3842 | struct inode *parent) | |
91828a40 | 3843 | { |
5fe0c237 AK |
3844 | if (*len < 3) { |
3845 | *len = 3; | |
94e07a75 | 3846 | return FILEID_INVALID; |
5fe0c237 | 3847 | } |
91828a40 | 3848 | |
1d3382cb | 3849 | if (inode_unhashed(inode)) { |
91828a40 DG |
3850 | /* Unfortunately insert_inode_hash is not idempotent, |
3851 | * so as we hash inodes here rather than at creation | |
3852 | * time, we need a lock to ensure we only try | |
3853 | * to do it once | |
3854 | */ | |
3855 | static DEFINE_SPINLOCK(lock); | |
3856 | spin_lock(&lock); | |
1d3382cb | 3857 | if (inode_unhashed(inode)) |
91828a40 DG |
3858 | __insert_inode_hash(inode, |
3859 | inode->i_ino + inode->i_generation); | |
3860 | spin_unlock(&lock); | |
3861 | } | |
3862 | ||
3863 | fh[0] = inode->i_generation; | |
3864 | fh[1] = inode->i_ino; | |
3865 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
3866 | ||
3867 | *len = 3; | |
3868 | return 1; | |
3869 | } | |
3870 | ||
39655164 | 3871 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3872 | .get_parent = shmem_get_parent, |
91828a40 | 3873 | .encode_fh = shmem_encode_fh, |
480b116c | 3874 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3875 | }; |
3876 | ||
626c3920 AV |
3877 | enum shmem_param { |
3878 | Opt_gid, | |
3879 | Opt_huge, | |
3880 | Opt_mode, | |
3881 | Opt_mpol, | |
3882 | Opt_nr_blocks, | |
3883 | Opt_nr_inodes, | |
3884 | Opt_size, | |
3885 | Opt_uid, | |
ea3271f7 CD |
3886 | Opt_inode32, |
3887 | Opt_inode64, | |
2c6efe9c | 3888 | Opt_noswap, |
e09764cf CM |
3889 | Opt_quota, |
3890 | Opt_usrquota, | |
3891 | Opt_grpquota, | |
de4c0e7c LC |
3892 | Opt_usrquota_block_hardlimit, |
3893 | Opt_usrquota_inode_hardlimit, | |
3894 | Opt_grpquota_block_hardlimit, | |
3895 | Opt_grpquota_inode_hardlimit, | |
626c3920 AV |
3896 | }; |
3897 | ||
5eede625 | 3898 | static const struct constant_table shmem_param_enums_huge[] = { |
2710c957 AV |
3899 | {"never", SHMEM_HUGE_NEVER }, |
3900 | {"always", SHMEM_HUGE_ALWAYS }, | |
3901 | {"within_size", SHMEM_HUGE_WITHIN_SIZE }, | |
3902 | {"advise", SHMEM_HUGE_ADVISE }, | |
2710c957 AV |
3903 | {} |
3904 | }; | |
3905 | ||
d7167b14 | 3906 | const struct fs_parameter_spec shmem_fs_parameters[] = { |
626c3920 | 3907 | fsparam_u32 ("gid", Opt_gid), |
2710c957 | 3908 | fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge), |
626c3920 AV |
3909 | fsparam_u32oct("mode", Opt_mode), |
3910 | fsparam_string("mpol", Opt_mpol), | |
3911 | fsparam_string("nr_blocks", Opt_nr_blocks), | |
3912 | fsparam_string("nr_inodes", Opt_nr_inodes), | |
3913 | fsparam_string("size", Opt_size), | |
3914 | fsparam_u32 ("uid", Opt_uid), | |
ea3271f7 CD |
3915 | fsparam_flag ("inode32", Opt_inode32), |
3916 | fsparam_flag ("inode64", Opt_inode64), | |
2c6efe9c | 3917 | fsparam_flag ("noswap", Opt_noswap), |
e09764cf CM |
3918 | #ifdef CONFIG_TMPFS_QUOTA |
3919 | fsparam_flag ("quota", Opt_quota), | |
3920 | fsparam_flag ("usrquota", Opt_usrquota), | |
3921 | fsparam_flag ("grpquota", Opt_grpquota), | |
de4c0e7c LC |
3922 | fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit), |
3923 | fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit), | |
3924 | fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit), | |
3925 | fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit), | |
e09764cf | 3926 | #endif |
626c3920 AV |
3927 | {} |
3928 | }; | |
3929 | ||
f3235626 | 3930 | static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param) |
1da177e4 | 3931 | { |
f3235626 | 3932 | struct shmem_options *ctx = fc->fs_private; |
626c3920 AV |
3933 | struct fs_parse_result result; |
3934 | unsigned long long size; | |
e04dc423 | 3935 | char *rest; |
626c3920 | 3936 | int opt; |
0200679f CB |
3937 | kuid_t kuid; |
3938 | kgid_t kgid; | |
626c3920 | 3939 | |
d7167b14 | 3940 | opt = fs_parse(fc, shmem_fs_parameters, param, &result); |
f3235626 | 3941 | if (opt < 0) |
626c3920 | 3942 | return opt; |
1da177e4 | 3943 | |
626c3920 AV |
3944 | switch (opt) { |
3945 | case Opt_size: | |
3946 | size = memparse(param->string, &rest); | |
e04dc423 AV |
3947 | if (*rest == '%') { |
3948 | size <<= PAGE_SHIFT; | |
3949 | size *= totalram_pages(); | |
3950 | do_div(size, 100); | |
3951 | rest++; | |
3952 | } | |
3953 | if (*rest) | |
626c3920 | 3954 | goto bad_value; |
e04dc423 AV |
3955 | ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE); |
3956 | ctx->seen |= SHMEM_SEEN_BLOCKS; | |
626c3920 AV |
3957 | break; |
3958 | case Opt_nr_blocks: | |
3959 | ctx->blocks = memparse(param->string, &rest); | |
e07c469e | 3960 | if (*rest || ctx->blocks > LONG_MAX) |
626c3920 | 3961 | goto bad_value; |
e04dc423 | 3962 | ctx->seen |= SHMEM_SEEN_BLOCKS; |
626c3920 AV |
3963 | break; |
3964 | case Opt_nr_inodes: | |
3965 | ctx->inodes = memparse(param->string, &rest); | |
e07c469e | 3966 | if (*rest || ctx->inodes > ULONG_MAX / BOGO_INODE_SIZE) |
626c3920 | 3967 | goto bad_value; |
e04dc423 | 3968 | ctx->seen |= SHMEM_SEEN_INODES; |
626c3920 AV |
3969 | break; |
3970 | case Opt_mode: | |
3971 | ctx->mode = result.uint_32 & 07777; | |
3972 | break; | |
3973 | case Opt_uid: | |
0200679f CB |
3974 | kuid = make_kuid(current_user_ns(), result.uint_32); |
3975 | if (!uid_valid(kuid)) | |
626c3920 | 3976 | goto bad_value; |
0200679f CB |
3977 | |
3978 | /* | |
3979 | * The requested uid must be representable in the | |
3980 | * filesystem's idmapping. | |
3981 | */ | |
3982 | if (!kuid_has_mapping(fc->user_ns, kuid)) | |
626c3920 | 3983 | goto bad_value; |
0200679f CB |
3984 | |
3985 | ctx->uid = kuid; | |
626c3920 AV |
3986 | break; |
3987 | case Opt_gid: | |
0200679f CB |
3988 | kgid = make_kgid(current_user_ns(), result.uint_32); |
3989 | if (!gid_valid(kgid)) | |
626c3920 | 3990 | goto bad_value; |
0200679f CB |
3991 | |
3992 | /* | |
3993 | * The requested gid must be representable in the | |
3994 | * filesystem's idmapping. | |
3995 | */ | |
3996 | if (!kgid_has_mapping(fc->user_ns, kgid)) | |
626c3920 | 3997 | goto bad_value; |
0200679f CB |
3998 | |
3999 | ctx->gid = kgid; | |
626c3920 AV |
4000 | break; |
4001 | case Opt_huge: | |
4002 | ctx->huge = result.uint_32; | |
4003 | if (ctx->huge != SHMEM_HUGE_NEVER && | |
396bcc52 | 4004 | !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && |
626c3920 AV |
4005 | has_transparent_hugepage())) |
4006 | goto unsupported_parameter; | |
e04dc423 | 4007 | ctx->seen |= SHMEM_SEEN_HUGE; |
626c3920 AV |
4008 | break; |
4009 | case Opt_mpol: | |
4010 | if (IS_ENABLED(CONFIG_NUMA)) { | |
4011 | mpol_put(ctx->mpol); | |
4012 | ctx->mpol = NULL; | |
4013 | if (mpol_parse_str(param->string, &ctx->mpol)) | |
4014 | goto bad_value; | |
4015 | break; | |
4016 | } | |
4017 | goto unsupported_parameter; | |
ea3271f7 CD |
4018 | case Opt_inode32: |
4019 | ctx->full_inums = false; | |
4020 | ctx->seen |= SHMEM_SEEN_INUMS; | |
4021 | break; | |
4022 | case Opt_inode64: | |
4023 | if (sizeof(ino_t) < 8) { | |
4024 | return invalfc(fc, | |
4025 | "Cannot use inode64 with <64bit inums in kernel\n"); | |
4026 | } | |
4027 | ctx->full_inums = true; | |
4028 | ctx->seen |= SHMEM_SEEN_INUMS; | |
4029 | break; | |
2c6efe9c | 4030 | case Opt_noswap: |
01106e14 CB |
4031 | if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) { |
4032 | return invalfc(fc, | |
4033 | "Turning off swap in unprivileged tmpfs mounts unsupported"); | |
4034 | } | |
2c6efe9c LC |
4035 | ctx->noswap = true; |
4036 | ctx->seen |= SHMEM_SEEN_NOSWAP; | |
4037 | break; | |
e09764cf CM |
4038 | case Opt_quota: |
4039 | if (fc->user_ns != &init_user_ns) | |
4040 | return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported"); | |
4041 | ctx->seen |= SHMEM_SEEN_QUOTA; | |
4042 | ctx->quota_types |= (QTYPE_MASK_USR | QTYPE_MASK_GRP); | |
4043 | break; | |
4044 | case Opt_usrquota: | |
4045 | if (fc->user_ns != &init_user_ns) | |
4046 | return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported"); | |
4047 | ctx->seen |= SHMEM_SEEN_QUOTA; | |
4048 | ctx->quota_types |= QTYPE_MASK_USR; | |
4049 | break; | |
4050 | case Opt_grpquota: | |
4051 | if (fc->user_ns != &init_user_ns) | |
4052 | return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported"); | |
4053 | ctx->seen |= SHMEM_SEEN_QUOTA; | |
4054 | ctx->quota_types |= QTYPE_MASK_GRP; | |
4055 | break; | |
de4c0e7c LC |
4056 | case Opt_usrquota_block_hardlimit: |
4057 | size = memparse(param->string, &rest); | |
4058 | if (*rest || !size) | |
4059 | goto bad_value; | |
4060 | if (size > SHMEM_QUOTA_MAX_SPC_LIMIT) | |
4061 | return invalfc(fc, | |
4062 | "User quota block hardlimit too large."); | |
4063 | ctx->qlimits.usrquota_bhardlimit = size; | |
4064 | break; | |
4065 | case Opt_grpquota_block_hardlimit: | |
4066 | size = memparse(param->string, &rest); | |
4067 | if (*rest || !size) | |
4068 | goto bad_value; | |
4069 | if (size > SHMEM_QUOTA_MAX_SPC_LIMIT) | |
4070 | return invalfc(fc, | |
4071 | "Group quota block hardlimit too large."); | |
4072 | ctx->qlimits.grpquota_bhardlimit = size; | |
4073 | break; | |
4074 | case Opt_usrquota_inode_hardlimit: | |
4075 | size = memparse(param->string, &rest); | |
4076 | if (*rest || !size) | |
4077 | goto bad_value; | |
4078 | if (size > SHMEM_QUOTA_MAX_INO_LIMIT) | |
4079 | return invalfc(fc, | |
4080 | "User quota inode hardlimit too large."); | |
4081 | ctx->qlimits.usrquota_ihardlimit = size; | |
4082 | break; | |
4083 | case Opt_grpquota_inode_hardlimit: | |
4084 | size = memparse(param->string, &rest); | |
4085 | if (*rest || !size) | |
4086 | goto bad_value; | |
4087 | if (size > SHMEM_QUOTA_MAX_INO_LIMIT) | |
4088 | return invalfc(fc, | |
4089 | "Group quota inode hardlimit too large."); | |
4090 | ctx->qlimits.grpquota_ihardlimit = size; | |
4091 | break; | |
e04dc423 AV |
4092 | } |
4093 | return 0; | |
4094 | ||
626c3920 | 4095 | unsupported_parameter: |
f35aa2bc | 4096 | return invalfc(fc, "Unsupported parameter '%s'", param->key); |
626c3920 | 4097 | bad_value: |
f35aa2bc | 4098 | return invalfc(fc, "Bad value for '%s'", param->key); |
e04dc423 AV |
4099 | } |
4100 | ||
f3235626 | 4101 | static int shmem_parse_options(struct fs_context *fc, void *data) |
e04dc423 | 4102 | { |
f3235626 DH |
4103 | char *options = data; |
4104 | ||
33f37c64 AV |
4105 | if (options) { |
4106 | int err = security_sb_eat_lsm_opts(options, &fc->security); | |
4107 | if (err) | |
4108 | return err; | |
4109 | } | |
4110 | ||
b00dc3ad | 4111 | while (options != NULL) { |
626c3920 | 4112 | char *this_char = options; |
b00dc3ad HD |
4113 | for (;;) { |
4114 | /* | |
4115 | * NUL-terminate this option: unfortunately, | |
4116 | * mount options form a comma-separated list, | |
4117 | * but mpol's nodelist may also contain commas. | |
4118 | */ | |
4119 | options = strchr(options, ','); | |
4120 | if (options == NULL) | |
4121 | break; | |
4122 | options++; | |
4123 | if (!isdigit(*options)) { | |
4124 | options[-1] = '\0'; | |
4125 | break; | |
4126 | } | |
4127 | } | |
626c3920 | 4128 | if (*this_char) { |
68d68ff6 | 4129 | char *value = strchr(this_char, '='); |
f3235626 | 4130 | size_t len = 0; |
626c3920 AV |
4131 | int err; |
4132 | ||
4133 | if (value) { | |
4134 | *value++ = '\0'; | |
f3235626 | 4135 | len = strlen(value); |
626c3920 | 4136 | } |
f3235626 DH |
4137 | err = vfs_parse_fs_string(fc, this_char, value, len); |
4138 | if (err < 0) | |
4139 | return err; | |
1da177e4 | 4140 | } |
1da177e4 LT |
4141 | } |
4142 | return 0; | |
1da177e4 LT |
4143 | } |
4144 | ||
f3235626 DH |
4145 | /* |
4146 | * Reconfigure a shmem filesystem. | |
f3235626 DH |
4147 | */ |
4148 | static int shmem_reconfigure(struct fs_context *fc) | |
1da177e4 | 4149 | { |
f3235626 DH |
4150 | struct shmem_options *ctx = fc->fs_private; |
4151 | struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb); | |
e07c469e | 4152 | unsigned long used_isp; |
bf11b9a8 | 4153 | struct mempolicy *mpol = NULL; |
f3235626 | 4154 | const char *err; |
1da177e4 | 4155 | |
bf11b9a8 | 4156 | raw_spin_lock(&sbinfo->stat_lock); |
e07c469e | 4157 | used_isp = sbinfo->max_inodes * BOGO_INODE_SIZE - sbinfo->free_ispace; |
0c98c8e1 | 4158 | |
f3235626 DH |
4159 | if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) { |
4160 | if (!sbinfo->max_blocks) { | |
4161 | err = "Cannot retroactively limit size"; | |
0b5071dd | 4162 | goto out; |
f3235626 | 4163 | } |
0b5071dd | 4164 | if (percpu_counter_compare(&sbinfo->used_blocks, |
f3235626 DH |
4165 | ctx->blocks) > 0) { |
4166 | err = "Too small a size for current use"; | |
0b5071dd | 4167 | goto out; |
f3235626 | 4168 | } |
0b5071dd | 4169 | } |
f3235626 DH |
4170 | if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) { |
4171 | if (!sbinfo->max_inodes) { | |
4172 | err = "Cannot retroactively limit inodes"; | |
0b5071dd | 4173 | goto out; |
f3235626 | 4174 | } |
e07c469e | 4175 | if (ctx->inodes * BOGO_INODE_SIZE < used_isp) { |
f3235626 | 4176 | err = "Too few inodes for current use"; |
0b5071dd | 4177 | goto out; |
f3235626 | 4178 | } |
0b5071dd | 4179 | } |
0edd73b3 | 4180 | |
ea3271f7 CD |
4181 | if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums && |
4182 | sbinfo->next_ino > UINT_MAX) { | |
4183 | err = "Current inum too high to switch to 32-bit inums"; | |
4184 | goto out; | |
4185 | } | |
2c6efe9c LC |
4186 | if ((ctx->seen & SHMEM_SEEN_NOSWAP) && ctx->noswap && !sbinfo->noswap) { |
4187 | err = "Cannot disable swap on remount"; | |
4188 | goto out; | |
4189 | } | |
4190 | if (!(ctx->seen & SHMEM_SEEN_NOSWAP) && !ctx->noswap && sbinfo->noswap) { | |
4191 | err = "Cannot enable swap on remount if it was disabled on first mount"; | |
4192 | goto out; | |
4193 | } | |
ea3271f7 | 4194 | |
e09764cf CM |
4195 | if (ctx->seen & SHMEM_SEEN_QUOTA && |
4196 | !sb_any_quota_loaded(fc->root->d_sb)) { | |
4197 | err = "Cannot enable quota on remount"; | |
4198 | goto out; | |
4199 | } | |
4200 | ||
de4c0e7c LC |
4201 | #ifdef CONFIG_TMPFS_QUOTA |
4202 | #define CHANGED_LIMIT(name) \ | |
4203 | (ctx->qlimits.name## hardlimit && \ | |
4204 | (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit)) | |
4205 | ||
4206 | if (CHANGED_LIMIT(usrquota_b) || CHANGED_LIMIT(usrquota_i) || | |
4207 | CHANGED_LIMIT(grpquota_b) || CHANGED_LIMIT(grpquota_i)) { | |
4208 | err = "Cannot change global quota limit on remount"; | |
4209 | goto out; | |
4210 | } | |
4211 | #endif /* CONFIG_TMPFS_QUOTA */ | |
4212 | ||
f3235626 DH |
4213 | if (ctx->seen & SHMEM_SEEN_HUGE) |
4214 | sbinfo->huge = ctx->huge; | |
ea3271f7 CD |
4215 | if (ctx->seen & SHMEM_SEEN_INUMS) |
4216 | sbinfo->full_inums = ctx->full_inums; | |
f3235626 DH |
4217 | if (ctx->seen & SHMEM_SEEN_BLOCKS) |
4218 | sbinfo->max_blocks = ctx->blocks; | |
4219 | if (ctx->seen & SHMEM_SEEN_INODES) { | |
4220 | sbinfo->max_inodes = ctx->inodes; | |
e07c469e | 4221 | sbinfo->free_ispace = ctx->inodes * BOGO_INODE_SIZE - used_isp; |
0b5071dd | 4222 | } |
71fe804b | 4223 | |
5f00110f GT |
4224 | /* |
4225 | * Preserve previous mempolicy unless mpol remount option was specified. | |
4226 | */ | |
f3235626 | 4227 | if (ctx->mpol) { |
bf11b9a8 | 4228 | mpol = sbinfo->mpol; |
f3235626 DH |
4229 | sbinfo->mpol = ctx->mpol; /* transfers initial ref */ |
4230 | ctx->mpol = NULL; | |
5f00110f | 4231 | } |
2c6efe9c LC |
4232 | |
4233 | if (ctx->noswap) | |
4234 | sbinfo->noswap = true; | |
4235 | ||
bf11b9a8 SAS |
4236 | raw_spin_unlock(&sbinfo->stat_lock); |
4237 | mpol_put(mpol); | |
f3235626 | 4238 | return 0; |
0edd73b3 | 4239 | out: |
bf11b9a8 | 4240 | raw_spin_unlock(&sbinfo->stat_lock); |
f35aa2bc | 4241 | return invalfc(fc, "%s", err); |
1da177e4 | 4242 | } |
680d794b | 4243 | |
34c80b1d | 4244 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 4245 | { |
34c80b1d | 4246 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
283ebdee | 4247 | struct mempolicy *mpol; |
680d794b AM |
4248 | |
4249 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
b91742d8 | 4250 | seq_printf(seq, ",size=%luk", K(sbinfo->max_blocks)); |
680d794b AM |
4251 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
4252 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
0825a6f9 | 4253 | if (sbinfo->mode != (0777 | S_ISVTX)) |
09208d15 | 4254 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
4255 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
4256 | seq_printf(seq, ",uid=%u", | |
4257 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
4258 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
4259 | seq_printf(seq, ",gid=%u", | |
4260 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
ea3271f7 CD |
4261 | |
4262 | /* | |
4263 | * Showing inode{64,32} might be useful even if it's the system default, | |
4264 | * since then people don't have to resort to checking both here and | |
4265 | * /proc/config.gz to confirm 64-bit inums were successfully applied | |
4266 | * (which may not even exist if IKCONFIG_PROC isn't enabled). | |
4267 | * | |
4268 | * We hide it when inode64 isn't the default and we are using 32-bit | |
4269 | * inodes, since that probably just means the feature isn't even under | |
4270 | * consideration. | |
4271 | * | |
4272 | * As such: | |
4273 | * | |
4274 | * +-----------------+-----------------+ | |
4275 | * | TMPFS_INODE64=y | TMPFS_INODE64=n | | |
4276 | * +------------------+-----------------+-----------------+ | |
4277 | * | full_inums=true | show | show | | |
4278 | * | full_inums=false | show | hide | | |
4279 | * +------------------+-----------------+-----------------+ | |
4280 | * | |
4281 | */ | |
4282 | if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums) | |
4283 | seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32)); | |
396bcc52 | 4284 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5a6e75f8 KS |
4285 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
4286 | if (sbinfo->huge) | |
4287 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
4288 | #endif | |
283ebdee TJ |
4289 | mpol = shmem_get_sbmpol(sbinfo); |
4290 | shmem_show_mpol(seq, mpol); | |
4291 | mpol_put(mpol); | |
2c6efe9c LC |
4292 | if (sbinfo->noswap) |
4293 | seq_printf(seq, ",noswap"); | |
b4d3de57 JK |
4294 | #ifdef CONFIG_TMPFS_QUOTA |
4295 | if (sb_has_quota_active(root->d_sb, USRQUOTA)) | |
4296 | seq_printf(seq, ",usrquota"); | |
4297 | if (sb_has_quota_active(root->d_sb, GRPQUOTA)) | |
4298 | seq_printf(seq, ",grpquota"); | |
4299 | if (sbinfo->qlimits.usrquota_bhardlimit) | |
4300 | seq_printf(seq, ",usrquota_block_hardlimit=%lld", | |
4301 | sbinfo->qlimits.usrquota_bhardlimit); | |
4302 | if (sbinfo->qlimits.grpquota_bhardlimit) | |
4303 | seq_printf(seq, ",grpquota_block_hardlimit=%lld", | |
4304 | sbinfo->qlimits.grpquota_bhardlimit); | |
4305 | if (sbinfo->qlimits.usrquota_ihardlimit) | |
4306 | seq_printf(seq, ",usrquota_inode_hardlimit=%lld", | |
4307 | sbinfo->qlimits.usrquota_ihardlimit); | |
4308 | if (sbinfo->qlimits.grpquota_ihardlimit) | |
4309 | seq_printf(seq, ",grpquota_inode_hardlimit=%lld", | |
4310 | sbinfo->qlimits.grpquota_ihardlimit); | |
4311 | #endif | |
680d794b AM |
4312 | return 0; |
4313 | } | |
9183df25 | 4314 | |
680d794b | 4315 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
4316 | |
4317 | static void shmem_put_super(struct super_block *sb) | |
4318 | { | |
602586a8 HD |
4319 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
4320 | ||
e09764cf CM |
4321 | #ifdef CONFIG_TMPFS_QUOTA |
4322 | shmem_disable_quotas(sb); | |
4323 | #endif | |
e809d5f0 | 4324 | free_percpu(sbinfo->ino_batch); |
602586a8 | 4325 | percpu_counter_destroy(&sbinfo->used_blocks); |
49cd0a5c | 4326 | mpol_put(sbinfo->mpol); |
602586a8 | 4327 | kfree(sbinfo); |
1da177e4 LT |
4328 | sb->s_fs_info = NULL; |
4329 | } | |
4330 | ||
f3235626 | 4331 | static int shmem_fill_super(struct super_block *sb, struct fs_context *fc) |
1da177e4 | 4332 | { |
f3235626 | 4333 | struct shmem_options *ctx = fc->fs_private; |
1da177e4 | 4334 | struct inode *inode; |
0edd73b3 | 4335 | struct shmem_sb_info *sbinfo; |
71480663 | 4336 | int error = -ENOMEM; |
680d794b AM |
4337 | |
4338 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 4339 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b AM |
4340 | L1_CACHE_BYTES), GFP_KERNEL); |
4341 | if (!sbinfo) | |
71480663 | 4342 | return error; |
680d794b | 4343 | |
680d794b | 4344 | sb->s_fs_info = sbinfo; |
1da177e4 | 4345 | |
0edd73b3 | 4346 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
4347 | /* |
4348 | * Per default we only allow half of the physical ram per | |
4349 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
4350 | * but the internal instance is left unlimited. | |
4351 | */ | |
1751e8a6 | 4352 | if (!(sb->s_flags & SB_KERNMOUNT)) { |
f3235626 DH |
4353 | if (!(ctx->seen & SHMEM_SEEN_BLOCKS)) |
4354 | ctx->blocks = shmem_default_max_blocks(); | |
4355 | if (!(ctx->seen & SHMEM_SEEN_INODES)) | |
4356 | ctx->inodes = shmem_default_max_inodes(); | |
ea3271f7 CD |
4357 | if (!(ctx->seen & SHMEM_SEEN_INUMS)) |
4358 | ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64); | |
2c6efe9c | 4359 | sbinfo->noswap = ctx->noswap; |
ca4e0519 | 4360 | } else { |
1751e8a6 | 4361 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 4362 | } |
91828a40 | 4363 | sb->s_export_op = &shmem_export_ops; |
36f05cab | 4364 | sb->s_flags |= SB_NOSEC | SB_I_VERSION; |
1da177e4 | 4365 | #else |
1751e8a6 | 4366 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 4367 | #endif |
f3235626 | 4368 | sbinfo->max_blocks = ctx->blocks; |
e07c469e HD |
4369 | sbinfo->max_inodes = ctx->inodes; |
4370 | sbinfo->free_ispace = sbinfo->max_inodes * BOGO_INODE_SIZE; | |
e809d5f0 CD |
4371 | if (sb->s_flags & SB_KERNMOUNT) { |
4372 | sbinfo->ino_batch = alloc_percpu(ino_t); | |
4373 | if (!sbinfo->ino_batch) | |
4374 | goto failed; | |
4375 | } | |
f3235626 DH |
4376 | sbinfo->uid = ctx->uid; |
4377 | sbinfo->gid = ctx->gid; | |
ea3271f7 | 4378 | sbinfo->full_inums = ctx->full_inums; |
f3235626 DH |
4379 | sbinfo->mode = ctx->mode; |
4380 | sbinfo->huge = ctx->huge; | |
4381 | sbinfo->mpol = ctx->mpol; | |
4382 | ctx->mpol = NULL; | |
1da177e4 | 4383 | |
bf11b9a8 | 4384 | raw_spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 4385 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 4386 | goto failed; |
779750d2 KS |
4387 | spin_lock_init(&sbinfo->shrinklist_lock); |
4388 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 4389 | |
285b2c4f | 4390 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
4391 | sb->s_blocksize = PAGE_SIZE; |
4392 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
4393 | sb->s_magic = TMPFS_MAGIC; |
4394 | sb->s_op = &shmem_ops; | |
cfd95a9c | 4395 | sb->s_time_gran = 1; |
b09e0fa4 | 4396 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 4397 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
4398 | #endif |
4399 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
1751e8a6 | 4400 | sb->s_flags |= SB_POSIXACL; |
39f0247d | 4401 | #endif |
a4af51ce KO |
4402 | uuid_t uuid; |
4403 | uuid_gen(&uuid); | |
4404 | super_set_uuid(sb, uuid.b, sizeof(uuid)); | |
0edd73b3 | 4405 | |
e09764cf CM |
4406 | #ifdef CONFIG_TMPFS_QUOTA |
4407 | if (ctx->seen & SHMEM_SEEN_QUOTA) { | |
4408 | sb->dq_op = &shmem_quota_operations; | |
4409 | sb->s_qcop = &dquot_quotactl_sysfile_ops; | |
4410 | sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP; | |
4411 | ||
de4c0e7c LC |
4412 | /* Copy the default limits from ctx into sbinfo */ |
4413 | memcpy(&sbinfo->qlimits, &ctx->qlimits, | |
4414 | sizeof(struct shmem_quota_limits)); | |
4415 | ||
e09764cf CM |
4416 | if (shmem_enable_quotas(sb, ctx->quota_types)) |
4417 | goto failed; | |
4418 | } | |
4419 | #endif /* CONFIG_TMPFS_QUOTA */ | |
4420 | ||
9be7d5b0 HD |
4421 | inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL, |
4422 | S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); | |
71480663 CM |
4423 | if (IS_ERR(inode)) { |
4424 | error = PTR_ERR(inode); | |
1da177e4 | 4425 | goto failed; |
71480663 | 4426 | } |
680d794b AM |
4427 | inode->i_uid = sbinfo->uid; |
4428 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
4429 | sb->s_root = d_make_root(inode); |
4430 | if (!sb->s_root) | |
48fde701 | 4431 | goto failed; |
1da177e4 LT |
4432 | return 0; |
4433 | ||
1da177e4 LT |
4434 | failed: |
4435 | shmem_put_super(sb); | |
71480663 | 4436 | return error; |
1da177e4 LT |
4437 | } |
4438 | ||
f3235626 DH |
4439 | static int shmem_get_tree(struct fs_context *fc) |
4440 | { | |
4441 | return get_tree_nodev(fc, shmem_fill_super); | |
4442 | } | |
4443 | ||
4444 | static void shmem_free_fc(struct fs_context *fc) | |
4445 | { | |
4446 | struct shmem_options *ctx = fc->fs_private; | |
4447 | ||
4448 | if (ctx) { | |
4449 | mpol_put(ctx->mpol); | |
4450 | kfree(ctx); | |
4451 | } | |
4452 | } | |
4453 | ||
4454 | static const struct fs_context_operations shmem_fs_context_ops = { | |
4455 | .free = shmem_free_fc, | |
4456 | .get_tree = shmem_get_tree, | |
4457 | #ifdef CONFIG_TMPFS | |
4458 | .parse_monolithic = shmem_parse_options, | |
4459 | .parse_param = shmem_parse_one, | |
4460 | .reconfigure = shmem_reconfigure, | |
4461 | #endif | |
4462 | }; | |
4463 | ||
68279f9c | 4464 | static struct kmem_cache *shmem_inode_cachep __ro_after_init; |
1da177e4 LT |
4465 | |
4466 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
4467 | { | |
41ffe5d5 | 4468 | struct shmem_inode_info *info; |
fd60b288 | 4469 | info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL); |
41ffe5d5 | 4470 | if (!info) |
1da177e4 | 4471 | return NULL; |
41ffe5d5 | 4472 | return &info->vfs_inode; |
1da177e4 LT |
4473 | } |
4474 | ||
74b1da56 | 4475 | static void shmem_free_in_core_inode(struct inode *inode) |
fa0d7e3d | 4476 | { |
84e710da AV |
4477 | if (S_ISLNK(inode->i_mode)) |
4478 | kfree(inode->i_link); | |
fa0d7e3d NP |
4479 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
4480 | } | |
4481 | ||
1da177e4 LT |
4482 | static void shmem_destroy_inode(struct inode *inode) |
4483 | { | |
09208d15 | 4484 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 4485 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
a2e45955 CL |
4486 | if (S_ISDIR(inode->i_mode)) |
4487 | simple_offset_destroy(shmem_get_offset_ctx(inode)); | |
1da177e4 LT |
4488 | } |
4489 | ||
41ffe5d5 | 4490 | static void shmem_init_inode(void *foo) |
1da177e4 | 4491 | { |
41ffe5d5 HD |
4492 | struct shmem_inode_info *info = foo; |
4493 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
4494 | } |
4495 | ||
68279f9c | 4496 | static void __init shmem_init_inodecache(void) |
1da177e4 LT |
4497 | { |
4498 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
4499 | sizeof(struct shmem_inode_info), | |
5d097056 | 4500 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
4501 | } |
4502 | ||
68279f9c | 4503 | static void __init shmem_destroy_inodecache(void) |
1da177e4 | 4504 | { |
1a1d92c1 | 4505 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
4506 | } |
4507 | ||
a7605426 | 4508 | /* Keep the page in page cache instead of truncating it */ |
af7628d6 MWO |
4509 | static int shmem_error_remove_folio(struct address_space *mapping, |
4510 | struct folio *folio) | |
a7605426 YS |
4511 | { |
4512 | return 0; | |
4513 | } | |
4514 | ||
aefacb20 | 4515 | static const struct address_space_operations shmem_aops = { |
1da177e4 | 4516 | .writepage = shmem_writepage, |
46de8b97 | 4517 | .dirty_folio = noop_dirty_folio, |
1da177e4 | 4518 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
4519 | .write_begin = shmem_write_begin, |
4520 | .write_end = shmem_write_end, | |
1da177e4 | 4521 | #endif |
1c93923c | 4522 | #ifdef CONFIG_MIGRATION |
54184650 | 4523 | .migrate_folio = migrate_folio, |
1c93923c | 4524 | #endif |
af7628d6 | 4525 | .error_remove_folio = shmem_error_remove_folio, |
1da177e4 LT |
4526 | }; |
4527 | ||
15ad7cdc | 4528 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 4529 | .mmap = shmem_mmap, |
e88e0d36 | 4530 | .open = shmem_file_open, |
c01d5b30 | 4531 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 4532 | #ifdef CONFIG_TMPFS |
220f2ac9 | 4533 | .llseek = shmem_file_llseek, |
2ba5bbed | 4534 | .read_iter = shmem_file_read_iter, |
e88e0d36 | 4535 | .write_iter = shmem_file_write_iter, |
1b061d92 | 4536 | .fsync = noop_fsync, |
bd194b18 | 4537 | .splice_read = shmem_file_splice_read, |
f6cb85d0 | 4538 | .splice_write = iter_file_splice_write, |
83e4fa9c | 4539 | .fallocate = shmem_fallocate, |
1da177e4 LT |
4540 | #endif |
4541 | }; | |
4542 | ||
92e1d5be | 4543 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 4544 | .getattr = shmem_getattr, |
94c1e62d | 4545 | .setattr = shmem_setattr, |
b09e0fa4 | 4546 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 4547 | .listxattr = shmem_listxattr, |
feda821e | 4548 | .set_acl = simple_set_acl, |
e408e695 TT |
4549 | .fileattr_get = shmem_fileattr_get, |
4550 | .fileattr_set = shmem_fileattr_set, | |
b09e0fa4 | 4551 | #endif |
1da177e4 LT |
4552 | }; |
4553 | ||
92e1d5be | 4554 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 | 4555 | #ifdef CONFIG_TMPFS |
f7cd16a5 | 4556 | .getattr = shmem_getattr, |
1da177e4 LT |
4557 | .create = shmem_create, |
4558 | .lookup = simple_lookup, | |
4559 | .link = shmem_link, | |
4560 | .unlink = shmem_unlink, | |
4561 | .symlink = shmem_symlink, | |
4562 | .mkdir = shmem_mkdir, | |
4563 | .rmdir = shmem_rmdir, | |
4564 | .mknod = shmem_mknod, | |
2773bf00 | 4565 | .rename = shmem_rename2, |
60545d0d | 4566 | .tmpfile = shmem_tmpfile, |
a2e45955 | 4567 | .get_offset_ctx = shmem_get_offset_ctx, |
1da177e4 | 4568 | #endif |
b09e0fa4 | 4569 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 4570 | .listxattr = shmem_listxattr, |
e408e695 TT |
4571 | .fileattr_get = shmem_fileattr_get, |
4572 | .fileattr_set = shmem_fileattr_set, | |
b09e0fa4 | 4573 | #endif |
39f0247d | 4574 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 4575 | .setattr = shmem_setattr, |
feda821e | 4576 | .set_acl = simple_set_acl, |
39f0247d AG |
4577 | #endif |
4578 | }; | |
4579 | ||
92e1d5be | 4580 | static const struct inode_operations shmem_special_inode_operations = { |
f7cd16a5 | 4581 | .getattr = shmem_getattr, |
b09e0fa4 | 4582 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 4583 | .listxattr = shmem_listxattr, |
b09e0fa4 | 4584 | #endif |
39f0247d | 4585 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 4586 | .setattr = shmem_setattr, |
feda821e | 4587 | .set_acl = simple_set_acl, |
39f0247d | 4588 | #endif |
1da177e4 LT |
4589 | }; |
4590 | ||
759b9775 | 4591 | static const struct super_operations shmem_ops = { |
1da177e4 | 4592 | .alloc_inode = shmem_alloc_inode, |
74b1da56 | 4593 | .free_inode = shmem_free_in_core_inode, |
1da177e4 LT |
4594 | .destroy_inode = shmem_destroy_inode, |
4595 | #ifdef CONFIG_TMPFS | |
4596 | .statfs = shmem_statfs, | |
680d794b | 4597 | .show_options = shmem_show_options, |
e09764cf CM |
4598 | #endif |
4599 | #ifdef CONFIG_TMPFS_QUOTA | |
4600 | .get_dquots = shmem_get_dquots, | |
1da177e4 | 4601 | #endif |
1f895f75 | 4602 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
4603 | .drop_inode = generic_delete_inode, |
4604 | .put_super = shmem_put_super, | |
396bcc52 | 4605 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
779750d2 KS |
4606 | .nr_cached_objects = shmem_unused_huge_count, |
4607 | .free_cached_objects = shmem_unused_huge_scan, | |
4608 | #endif | |
1da177e4 LT |
4609 | }; |
4610 | ||
f0f37e2f | 4611 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 4612 | .fault = shmem_fault, |
d7c17551 | 4613 | .map_pages = filemap_map_pages, |
1da177e4 LT |
4614 | #ifdef CONFIG_NUMA |
4615 | .set_policy = shmem_set_policy, | |
4616 | .get_policy = shmem_get_policy, | |
4617 | #endif | |
4618 | }; | |
4619 | ||
d09e8ca6 PT |
4620 | static const struct vm_operations_struct shmem_anon_vm_ops = { |
4621 | .fault = shmem_fault, | |
4622 | .map_pages = filemap_map_pages, | |
4623 | #ifdef CONFIG_NUMA | |
4624 | .set_policy = shmem_set_policy, | |
4625 | .get_policy = shmem_get_policy, | |
4626 | #endif | |
4627 | }; | |
4628 | ||
f3235626 | 4629 | int shmem_init_fs_context(struct fs_context *fc) |
1da177e4 | 4630 | { |
f3235626 DH |
4631 | struct shmem_options *ctx; |
4632 | ||
4633 | ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL); | |
4634 | if (!ctx) | |
4635 | return -ENOMEM; | |
4636 | ||
4637 | ctx->mode = 0777 | S_ISVTX; | |
4638 | ctx->uid = current_fsuid(); | |
4639 | ctx->gid = current_fsgid(); | |
4640 | ||
4641 | fc->fs_private = ctx; | |
4642 | fc->ops = &shmem_fs_context_ops; | |
4643 | return 0; | |
1da177e4 LT |
4644 | } |
4645 | ||
41ffe5d5 | 4646 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
4647 | .owner = THIS_MODULE, |
4648 | .name = "tmpfs", | |
f3235626 DH |
4649 | .init_fs_context = shmem_init_fs_context, |
4650 | #ifdef CONFIG_TMPFS | |
d7167b14 | 4651 | .parameters = shmem_fs_parameters, |
f3235626 | 4652 | #endif |
1da177e4 | 4653 | .kill_sb = kill_litter_super, |
db58b5ee | 4654 | .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP, |
1da177e4 | 4655 | }; |
1da177e4 | 4656 | |
9096bbe9 | 4657 | void __init shmem_init(void) |
1da177e4 LT |
4658 | { |
4659 | int error; | |
4660 | ||
9a8ec03e | 4661 | shmem_init_inodecache(); |
1da177e4 | 4662 | |
e09764cf CM |
4663 | #ifdef CONFIG_TMPFS_QUOTA |
4664 | error = register_quota_format(&shmem_quota_format); | |
4665 | if (error < 0) { | |
4666 | pr_err("Could not register quota format\n"); | |
4667 | goto out3; | |
4668 | } | |
4669 | #endif | |
4670 | ||
41ffe5d5 | 4671 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 4672 | if (error) { |
1170532b | 4673 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
4674 | goto out2; |
4675 | } | |
95dc112a | 4676 | |
ca4e0519 | 4677 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
4678 | if (IS_ERR(shm_mnt)) { |
4679 | error = PTR_ERR(shm_mnt); | |
1170532b | 4680 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
4681 | goto out1; |
4682 | } | |
5a6e75f8 | 4683 | |
396bcc52 | 4684 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
435c0b87 | 4685 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4686 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4687 | else | |
5e6e5a12 | 4688 | shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */ |
5a6e75f8 | 4689 | #endif |
9096bbe9 | 4690 | return; |
1da177e4 LT |
4691 | |
4692 | out1: | |
41ffe5d5 | 4693 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 4694 | out2: |
e09764cf CM |
4695 | #ifdef CONFIG_TMPFS_QUOTA |
4696 | unregister_quota_format(&shmem_quota_format); | |
4697 | out3: | |
4698 | #endif | |
41ffe5d5 | 4699 | shmem_destroy_inodecache(); |
1da177e4 | 4700 | shm_mnt = ERR_PTR(error); |
1da177e4 | 4701 | } |
853ac43a | 4702 | |
396bcc52 | 4703 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS) |
5a6e75f8 | 4704 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
79d4d38a | 4705 | struct kobj_attribute *attr, char *buf) |
5a6e75f8 | 4706 | { |
26083eb6 | 4707 | static const int values[] = { |
5a6e75f8 KS |
4708 | SHMEM_HUGE_ALWAYS, |
4709 | SHMEM_HUGE_WITHIN_SIZE, | |
4710 | SHMEM_HUGE_ADVISE, | |
4711 | SHMEM_HUGE_NEVER, | |
4712 | SHMEM_HUGE_DENY, | |
4713 | SHMEM_HUGE_FORCE, | |
4714 | }; | |
79d4d38a JP |
4715 | int len = 0; |
4716 | int i; | |
5a6e75f8 | 4717 | |
79d4d38a JP |
4718 | for (i = 0; i < ARRAY_SIZE(values); i++) { |
4719 | len += sysfs_emit_at(buf, len, | |
9be7d5b0 HD |
4720 | shmem_huge == values[i] ? "%s[%s]" : "%s%s", |
4721 | i ? " " : "", shmem_format_huge(values[i])); | |
5a6e75f8 | 4722 | } |
79d4d38a JP |
4723 | len += sysfs_emit_at(buf, len, "\n"); |
4724 | ||
4725 | return len; | |
5a6e75f8 KS |
4726 | } |
4727 | ||
4728 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
4729 | struct kobj_attribute *attr, const char *buf, size_t count) | |
4730 | { | |
4731 | char tmp[16]; | |
4732 | int huge; | |
4733 | ||
4734 | if (count + 1 > sizeof(tmp)) | |
4735 | return -EINVAL; | |
4736 | memcpy(tmp, buf, count); | |
4737 | tmp[count] = '\0'; | |
4738 | if (count && tmp[count - 1] == '\n') | |
4739 | tmp[count - 1] = '\0'; | |
4740 | ||
4741 | huge = shmem_parse_huge(tmp); | |
4742 | if (huge == -EINVAL) | |
4743 | return -EINVAL; | |
4744 | if (!has_transparent_hugepage() && | |
4745 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
4746 | return -EINVAL; | |
4747 | ||
4748 | shmem_huge = huge; | |
435c0b87 | 4749 | if (shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4750 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4751 | return count; | |
4752 | } | |
4753 | ||
4bfa8ada | 4754 | struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled); |
396bcc52 | 4755 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */ |
f3f0e1d2 | 4756 | |
853ac43a MM |
4757 | #else /* !CONFIG_SHMEM */ |
4758 | ||
4759 | /* | |
4760 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
4761 | * | |
4762 | * This is intended for small system where the benefits of the full | |
4763 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4764 | * their complexity. On systems without swap this code should be | |
4765 | * effectively equivalent, but much lighter weight. | |
4766 | */ | |
4767 | ||
41ffe5d5 | 4768 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4769 | .name = "tmpfs", |
f3235626 | 4770 | .init_fs_context = ramfs_init_fs_context, |
d7167b14 | 4771 | .parameters = ramfs_fs_parameters, |
36ce9d76 | 4772 | .kill_sb = ramfs_kill_sb, |
2b8576cb | 4773 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4774 | }; |
4775 | ||
9096bbe9 | 4776 | void __init shmem_init(void) |
853ac43a | 4777 | { |
41ffe5d5 | 4778 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4779 | |
41ffe5d5 | 4780 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a | 4781 | BUG_ON(IS_ERR(shm_mnt)); |
853ac43a MM |
4782 | } |
4783 | ||
10a9c496 | 4784 | int shmem_unuse(unsigned int type) |
853ac43a MM |
4785 | { |
4786 | return 0; | |
4787 | } | |
4788 | ||
d7c9e99a | 4789 | int shmem_lock(struct file *file, int lock, struct ucounts *ucounts) |
3f96b79a HD |
4790 | { |
4791 | return 0; | |
4792 | } | |
4793 | ||
24513264 HD |
4794 | void shmem_unlock_mapping(struct address_space *mapping) |
4795 | { | |
4796 | } | |
4797 | ||
c01d5b30 HD |
4798 | #ifdef CONFIG_MMU |
4799 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4800 | unsigned long addr, unsigned long len, | |
4801 | unsigned long pgoff, unsigned long flags) | |
4802 | { | |
4803 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4804 | } | |
4805 | #endif | |
4806 | ||
41ffe5d5 | 4807 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4808 | { |
41ffe5d5 | 4809 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4810 | } |
4811 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4812 | ||
0b0a0806 | 4813 | #define shmem_vm_ops generic_file_vm_ops |
d09e8ca6 | 4814 | #define shmem_anon_vm_ops generic_file_vm_ops |
0b0a0806 | 4815 | #define shmem_file_operations ramfs_file_operations |
0b0a0806 HD |
4816 | #define shmem_acct_size(flags, size) 0 |
4817 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a | 4818 | |
9be7d5b0 HD |
4819 | static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap, |
4820 | struct super_block *sb, struct inode *dir, | |
4821 | umode_t mode, dev_t dev, unsigned long flags) | |
71480663 CM |
4822 | { |
4823 | struct inode *inode = ramfs_get_inode(sb, dir, mode, dev); | |
4824 | return inode ? inode : ERR_PTR(-ENOSPC); | |
4825 | } | |
4826 | ||
853ac43a MM |
4827 | #endif /* CONFIG_SHMEM */ |
4828 | ||
4829 | /* common code */ | |
1da177e4 | 4830 | |
9be7d5b0 HD |
4831 | static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, |
4832 | loff_t size, unsigned long flags, unsigned int i_flags) | |
1da177e4 | 4833 | { |
1da177e4 | 4834 | struct inode *inode; |
93dec2da | 4835 | struct file *res; |
1da177e4 | 4836 | |
703321b6 MA |
4837 | if (IS_ERR(mnt)) |
4838 | return ERR_CAST(mnt); | |
1da177e4 | 4839 | |
285b2c4f | 4840 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4841 | return ERR_PTR(-EINVAL); |
4842 | ||
4843 | if (shmem_acct_size(flags, size)) | |
4844 | return ERR_PTR(-ENOMEM); | |
4845 | ||
7a80e5b8 GS |
4846 | if (is_idmapped_mnt(mnt)) |
4847 | return ERR_PTR(-EINVAL); | |
4848 | ||
4849 | inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL, | |
4850 | S_IFREG | S_IRWXUGO, 0, flags); | |
71480663 | 4851 | if (IS_ERR(inode)) { |
dac2d1f6 | 4852 | shmem_unacct_size(flags, size); |
71480663 | 4853 | return ERR_CAST(inode); |
dac2d1f6 | 4854 | } |
c7277090 | 4855 | inode->i_flags |= i_flags; |
1da177e4 | 4856 | inode->i_size = size; |
6d6b77f1 | 4857 | clear_nlink(inode); /* It is unlinked */ |
26567cdb | 4858 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
93dec2da AV |
4859 | if (!IS_ERR(res)) |
4860 | res = alloc_file_pseudo(inode, mnt, name, O_RDWR, | |
4861 | &shmem_file_operations); | |
26567cdb | 4862 | if (IS_ERR(res)) |
93dec2da | 4863 | iput(inode); |
6b4d0b27 | 4864 | return res; |
1da177e4 | 4865 | } |
c7277090 EP |
4866 | |
4867 | /** | |
4868 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4869 | * kernel internal. There will be NO LSM permission checks against the | |
4870 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4871 | * higher layer. The users are the big_key and shm implementations. LSM |
4872 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4873 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4874 | * @size: size to be set for the file | |
4875 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4876 | */ | |
4877 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4878 | { | |
703321b6 | 4879 | return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE); |
c7277090 | 4880 | } |
be9d9366 | 4881 | EXPORT_SYMBOL_GPL(shmem_kernel_file_setup); |
c7277090 EP |
4882 | |
4883 | /** | |
4884 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4885 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4886 | * @size: size to be set for the file | |
4887 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4888 | */ | |
4889 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4890 | { | |
703321b6 | 4891 | return __shmem_file_setup(shm_mnt, name, size, flags, 0); |
c7277090 | 4892 | } |
395e0ddc | 4893 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4894 | |
703321b6 MA |
4895 | /** |
4896 | * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs | |
4897 | * @mnt: the tmpfs mount where the file will be created | |
4898 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4899 | * @size: size to be set for the file | |
4900 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4901 | */ | |
4902 | struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name, | |
4903 | loff_t size, unsigned long flags) | |
4904 | { | |
4905 | return __shmem_file_setup(mnt, name, size, flags, 0); | |
4906 | } | |
4907 | EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt); | |
4908 | ||
46711810 | 4909 | /** |
1da177e4 | 4910 | * shmem_zero_setup - setup a shared anonymous mapping |
45e55300 | 4911 | * @vma: the vma to be mmapped is prepared by do_mmap |
1da177e4 LT |
4912 | */ |
4913 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4914 | { | |
4915 | struct file *file; | |
4916 | loff_t size = vma->vm_end - vma->vm_start; | |
4917 | ||
66fc1303 | 4918 | /* |
c1e8d7c6 | 4919 | * Cloning a new file under mmap_lock leads to a lock ordering conflict |
66fc1303 HD |
4920 | * between XFS directory reading and selinux: since this file is only |
4921 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4922 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4923 | */ | |
703321b6 | 4924 | file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags); |
1da177e4 LT |
4925 | if (IS_ERR(file)) |
4926 | return PTR_ERR(file); | |
4927 | ||
4928 | if (vma->vm_file) | |
4929 | fput(vma->vm_file); | |
4930 | vma->vm_file = file; | |
d09e8ca6 | 4931 | vma->vm_ops = &shmem_anon_vm_ops; |
f3f0e1d2 | 4932 | |
1da177e4 LT |
4933 | return 0; |
4934 | } | |
d9d90e5e HD |
4935 | |
4936 | /** | |
f01b2b3e MWO |
4937 | * shmem_read_folio_gfp - read into page cache, using specified page allocation flags. |
4938 | * @mapping: the folio's address_space | |
4939 | * @index: the folio index | |
d9d90e5e HD |
4940 | * @gfp: the page allocator flags to use if allocating |
4941 | * | |
4942 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4943 | * with any new page allocations done using the specified allocation flags. | |
7e0a1265 | 4944 | * But read_cache_page_gfp() uses the ->read_folio() method: which does not |
d9d90e5e HD |
4945 | * suit tmpfs, since it may have pages in swapcache, and needs to find those |
4946 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4947 | * | |
68da9f05 HD |
4948 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4949 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e | 4950 | */ |
f01b2b3e MWO |
4951 | struct folio *shmem_read_folio_gfp(struct address_space *mapping, |
4952 | pgoff_t index, gfp_t gfp) | |
d9d90e5e | 4953 | { |
68da9f05 HD |
4954 | #ifdef CONFIG_SHMEM |
4955 | struct inode *inode = mapping->host; | |
a3a9c397 | 4956 | struct folio *folio; |
68da9f05 HD |
4957 | int error; |
4958 | ||
a3a9c397 | 4959 | error = shmem_get_folio_gfp(inode, index, &folio, SGP_CACHE, |
e3e1a506 | 4960 | gfp, NULL, NULL); |
68da9f05 | 4961 | if (error) |
a7605426 YS |
4962 | return ERR_PTR(error); |
4963 | ||
a3a9c397 | 4964 | folio_unlock(folio); |
f01b2b3e MWO |
4965 | return folio; |
4966 | #else | |
4967 | /* | |
4968 | * The tiny !SHMEM case uses ramfs without swap | |
4969 | */ | |
4970 | return mapping_read_folio_gfp(mapping, index, gfp); | |
4971 | #endif | |
4972 | } | |
4973 | EXPORT_SYMBOL_GPL(shmem_read_folio_gfp); | |
4974 | ||
4975 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4976 | pgoff_t index, gfp_t gfp) | |
4977 | { | |
4978 | struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp); | |
4979 | struct page *page; | |
4980 | ||
4981 | if (IS_ERR(folio)) | |
4982 | return &folio->page; | |
4983 | ||
a3a9c397 | 4984 | page = folio_file_page(folio, index); |
a7605426 | 4985 | if (PageHWPoison(page)) { |
a3a9c397 | 4986 | folio_put(folio); |
a7605426 YS |
4987 | return ERR_PTR(-EIO); |
4988 | } | |
4989 | ||
68da9f05 | 4990 | return page; |
d9d90e5e HD |
4991 | } |
4992 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |