Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Resizable virtual memory filesystem for Linux. | |
3 | * | |
4 | * Copyright (C) 2000 Linus Torvalds. | |
5 | * 2000 Transmeta Corp. | |
6 | * 2000-2001 Christoph Rohland | |
7 | * 2000-2001 SAP AG | |
8 | * 2002 Red Hat Inc. | |
6922c0c7 HD |
9 | * Copyright (C) 2002-2011 Hugh Dickins. |
10 | * Copyright (C) 2011 Google Inc. | |
0edd73b3 | 11 | * Copyright (C) 2002-2005 VERITAS Software Corporation. |
1da177e4 LT |
12 | * Copyright (C) 2004 Andi Kleen, SuSE Labs |
13 | * | |
14 | * Extended attribute support for tmpfs: | |
15 | * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net> | |
16 | * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com> | |
17 | * | |
853ac43a MM |
18 | * tiny-shmem: |
19 | * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com> | |
20 | * | |
1da177e4 LT |
21 | * This file is released under the GPL. |
22 | */ | |
23 | ||
853ac43a MM |
24 | #include <linux/fs.h> |
25 | #include <linux/init.h> | |
26 | #include <linux/vfs.h> | |
27 | #include <linux/mount.h> | |
250297ed | 28 | #include <linux/ramfs.h> |
caefba17 | 29 | #include <linux/pagemap.h> |
853ac43a | 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 | 135 | static unsigned long shmem_default_max_blocks(void) |
136 | { | |
ca79b0c2 | 137 | return totalram_pages() / 2; |
680d794b | 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; |
30e6a51d | 257 | 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 | |
d09e8ca6 PT |
266 | bool vma_is_anon_shmem(struct vm_area_struct *vma) |
267 | { | |
268 | return vma->vm_ops == &shmem_anon_vm_ops; | |
269 | } | |
270 | ||
b0506e48 MR |
271 | bool vma_is_shmem(struct vm_area_struct *vma) |
272 | { | |
d09e8ca6 | 273 | return vma_is_anon_shmem(vma) || vma->vm_ops == &shmem_vm_ops; |
b0506e48 MR |
274 | } |
275 | ||
1da177e4 | 276 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 277 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 278 | |
e09764cf CM |
279 | #ifdef CONFIG_TMPFS_QUOTA |
280 | ||
281 | static int shmem_enable_quotas(struct super_block *sb, | |
282 | unsigned short quota_types) | |
283 | { | |
284 | int type, err = 0; | |
285 | ||
286 | sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NOLIST_DIRTY; | |
287 | for (type = 0; type < SHMEM_MAXQUOTAS; type++) { | |
288 | if (!(quota_types & (1 << type))) | |
289 | continue; | |
290 | err = dquot_load_quota_sb(sb, type, QFMT_SHMEM, | |
291 | DQUOT_USAGE_ENABLED | | |
292 | DQUOT_LIMITS_ENABLED); | |
293 | if (err) | |
294 | goto out_err; | |
295 | } | |
296 | return 0; | |
297 | ||
298 | out_err: | |
299 | pr_warn("tmpfs: failed to enable quota tracking (type=%d, err=%d)\n", | |
300 | type, err); | |
301 | for (type--; type >= 0; type--) | |
302 | dquot_quota_off(sb, type); | |
303 | return err; | |
304 | } | |
305 | ||
306 | static void shmem_disable_quotas(struct super_block *sb) | |
307 | { | |
308 | int type; | |
309 | ||
310 | for (type = 0; type < SHMEM_MAXQUOTAS; type++) | |
311 | dquot_quota_off(sb, type); | |
312 | } | |
313 | ||
314 | static struct dquot **shmem_get_dquots(struct inode *inode) | |
315 | { | |
316 | return SHMEM_I(inode)->i_dquot; | |
317 | } | |
318 | #endif /* CONFIG_TMPFS_QUOTA */ | |
319 | ||
e809d5f0 CD |
320 | /* |
321 | * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and | |
322 | * produces a novel ino for the newly allocated inode. | |
323 | * | |
324 | * It may also be called when making a hard link to permit the space needed by | |
325 | * each dentry. However, in that case, no new inode number is needed since that | |
326 | * internally draws from another pool of inode numbers (currently global | |
327 | * get_next_ino()). This case is indicated by passing NULL as inop. | |
328 | */ | |
329 | #define SHMEM_INO_BATCH 1024 | |
330 | static int shmem_reserve_inode(struct super_block *sb, ino_t *inop) | |
5b04c689 PE |
331 | { |
332 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
e809d5f0 CD |
333 | ino_t ino; |
334 | ||
335 | if (!(sb->s_flags & SB_KERNMOUNT)) { | |
bf11b9a8 | 336 | raw_spin_lock(&sbinfo->stat_lock); |
bb3e96d6 | 337 | if (sbinfo->max_inodes) { |
e07c469e | 338 | if (sbinfo->free_ispace < BOGO_INODE_SIZE) { |
bf11b9a8 | 339 | raw_spin_unlock(&sbinfo->stat_lock); |
bb3e96d6 BS |
340 | return -ENOSPC; |
341 | } | |
e07c469e | 342 | sbinfo->free_ispace -= BOGO_INODE_SIZE; |
5b04c689 | 343 | } |
e809d5f0 CD |
344 | if (inop) { |
345 | ino = sbinfo->next_ino++; | |
346 | if (unlikely(is_zero_ino(ino))) | |
347 | ino = sbinfo->next_ino++; | |
ea3271f7 CD |
348 | if (unlikely(!sbinfo->full_inums && |
349 | ino > UINT_MAX)) { | |
e809d5f0 CD |
350 | /* |
351 | * Emulate get_next_ino uint wraparound for | |
352 | * compatibility | |
353 | */ | |
ea3271f7 CD |
354 | if (IS_ENABLED(CONFIG_64BIT)) |
355 | pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n", | |
356 | __func__, MINOR(sb->s_dev)); | |
357 | sbinfo->next_ino = 1; | |
358 | ino = sbinfo->next_ino++; | |
e809d5f0 CD |
359 | } |
360 | *inop = ino; | |
361 | } | |
bf11b9a8 | 362 | raw_spin_unlock(&sbinfo->stat_lock); |
e809d5f0 CD |
363 | } else if (inop) { |
364 | /* | |
365 | * __shmem_file_setup, one of our callers, is lock-free: it | |
366 | * doesn't hold stat_lock in shmem_reserve_inode since | |
367 | * max_inodes is always 0, and is called from potentially | |
368 | * unknown contexts. As such, use a per-cpu batched allocator | |
369 | * which doesn't require the per-sb stat_lock unless we are at | |
370 | * the batch boundary. | |
ea3271f7 CD |
371 | * |
372 | * We don't need to worry about inode{32,64} since SB_KERNMOUNT | |
373 | * shmem mounts are not exposed to userspace, so we don't need | |
374 | * to worry about things like glibc compatibility. | |
e809d5f0 CD |
375 | */ |
376 | ino_t *next_ino; | |
bf11b9a8 | 377 | |
e809d5f0 CD |
378 | next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu()); |
379 | ino = *next_ino; | |
380 | if (unlikely(ino % SHMEM_INO_BATCH == 0)) { | |
bf11b9a8 | 381 | raw_spin_lock(&sbinfo->stat_lock); |
e809d5f0 CD |
382 | ino = sbinfo->next_ino; |
383 | sbinfo->next_ino += SHMEM_INO_BATCH; | |
bf11b9a8 | 384 | raw_spin_unlock(&sbinfo->stat_lock); |
e809d5f0 CD |
385 | if (unlikely(is_zero_ino(ino))) |
386 | ino++; | |
387 | } | |
388 | *inop = ino; | |
389 | *next_ino = ++ino; | |
390 | put_cpu(); | |
5b04c689 | 391 | } |
e809d5f0 | 392 | |
5b04c689 PE |
393 | return 0; |
394 | } | |
395 | ||
2daf18a7 | 396 | static void shmem_free_inode(struct super_block *sb, size_t freed_ispace) |
5b04c689 PE |
397 | { |
398 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
399 | if (sbinfo->max_inodes) { | |
bf11b9a8 | 400 | raw_spin_lock(&sbinfo->stat_lock); |
2daf18a7 | 401 | sbinfo->free_ispace += BOGO_INODE_SIZE + freed_ispace; |
bf11b9a8 | 402 | raw_spin_unlock(&sbinfo->stat_lock); |
5b04c689 PE |
403 | } |
404 | } | |
405 | ||
46711810 | 406 | /** |
41ffe5d5 | 407 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 | 408 | * @inode: inode to recalc |
3c1b7528 HD |
409 | * @alloced: the change in number of pages allocated to inode |
410 | * @swapped: the change in number of pages swapped from inode | |
1da177e4 LT |
411 | * |
412 | * We have to calculate the free blocks since the mm can drop | |
413 | * undirtied hole pages behind our back. | |
414 | * | |
415 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
416 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
1da177e4 | 417 | */ |
3c1b7528 | 418 | static void shmem_recalc_inode(struct inode *inode, long alloced, long swapped) |
1da177e4 LT |
419 | { |
420 | struct shmem_inode_info *info = SHMEM_I(inode); | |
421 | long freed; | |
422 | ||
3c1b7528 HD |
423 | spin_lock(&info->lock); |
424 | info->alloced += alloced; | |
425 | info->swapped += swapped; | |
426 | freed = info->alloced - info->swapped - | |
427 | READ_ONCE(inode->i_mapping->nrpages); | |
428 | /* | |
429 | * Special case: whereas normally shmem_recalc_inode() is called | |
430 | * after i_mapping->nrpages has already been adjusted (up or down), | |
431 | * shmem_writepage() has to raise swapped before nrpages is lowered - | |
432 | * to stop a racing shmem_recalc_inode() from thinking that a page has | |
433 | * been freed. Compensate here, to avoid the need for a followup call. | |
434 | */ | |
435 | if (swapped > 0) | |
436 | freed += swapped; | |
437 | if (freed > 0) | |
1da177e4 | 438 | info->alloced -= freed; |
3c1b7528 HD |
439 | spin_unlock(&info->lock); |
440 | ||
441 | /* The quota case may block */ | |
442 | if (freed > 0) | |
0f079694 | 443 | shmem_inode_unacct_blocks(inode, freed); |
1da177e4 LT |
444 | } |
445 | ||
800d8c63 KS |
446 | bool shmem_charge(struct inode *inode, long pages) |
447 | { | |
509f0069 | 448 | struct address_space *mapping = inode->i_mapping; |
800d8c63 | 449 | |
4199f51a | 450 | if (shmem_inode_acct_blocks(inode, pages)) |
800d8c63 | 451 | return false; |
b1cc94ab | 452 | |
aaa52e34 | 453 | /* nrpages adjustment first, then shmem_recalc_inode() when balanced */ |
509f0069 HD |
454 | xa_lock_irq(&mapping->i_pages); |
455 | mapping->nrpages += pages; | |
456 | xa_unlock_irq(&mapping->i_pages); | |
800d8c63 | 457 | |
3c1b7528 | 458 | shmem_recalc_inode(inode, pages, 0); |
800d8c63 KS |
459 | return true; |
460 | } | |
461 | ||
462 | void shmem_uncharge(struct inode *inode, long pages) | |
463 | { | |
3c1b7528 | 464 | /* pages argument is currently unused: keep it to help debugging */ |
6ffcd825 | 465 | /* nrpages adjustment done by __filemap_remove_folio() or caller */ |
aaa52e34 | 466 | |
3c1b7528 | 467 | shmem_recalc_inode(inode, 0, 0); |
800d8c63 KS |
468 | } |
469 | ||
7a5d0fbb | 470 | /* |
62f945b6 | 471 | * Replace item expected in xarray by a new item, while holding xa_lock. |
7a5d0fbb | 472 | */ |
62f945b6 | 473 | static int shmem_replace_entry(struct address_space *mapping, |
7a5d0fbb HD |
474 | pgoff_t index, void *expected, void *replacement) |
475 | { | |
62f945b6 | 476 | XA_STATE(xas, &mapping->i_pages, index); |
6dbaf22c | 477 | void *item; |
7a5d0fbb HD |
478 | |
479 | VM_BUG_ON(!expected); | |
6dbaf22c | 480 | VM_BUG_ON(!replacement); |
62f945b6 | 481 | item = xas_load(&xas); |
7a5d0fbb HD |
482 | if (item != expected) |
483 | return -ENOENT; | |
62f945b6 | 484 | xas_store(&xas, replacement); |
7a5d0fbb HD |
485 | return 0; |
486 | } | |
487 | ||
d1899228 HD |
488 | /* |
489 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
490 | * that an entry was not already brought back from swap by a racing thread. | |
491 | * | |
492 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
493 | * might be reused, and again be SwapCache, using the same swap as before. | |
494 | */ | |
495 | static bool shmem_confirm_swap(struct address_space *mapping, | |
496 | pgoff_t index, swp_entry_t swap) | |
497 | { | |
a12831bf | 498 | return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap); |
d1899228 HD |
499 | } |
500 | ||
5a6e75f8 KS |
501 | /* |
502 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
503 | * | |
504 | * SHMEM_HUGE_NEVER: | |
505 | * disables huge pages for the mount; | |
506 | * SHMEM_HUGE_ALWAYS: | |
507 | * enables huge pages for the mount; | |
508 | * SHMEM_HUGE_WITHIN_SIZE: | |
509 | * only allocate huge pages if the page will be fully within i_size, | |
510 | * also respect fadvise()/madvise() hints; | |
511 | * SHMEM_HUGE_ADVISE: | |
512 | * only allocate huge pages if requested with fadvise()/madvise(); | |
513 | */ | |
514 | ||
515 | #define SHMEM_HUGE_NEVER 0 | |
516 | #define SHMEM_HUGE_ALWAYS 1 | |
517 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
518 | #define SHMEM_HUGE_ADVISE 3 | |
519 | ||
520 | /* | |
521 | * Special values. | |
522 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
523 | * | |
524 | * SHMEM_HUGE_DENY: | |
525 | * disables huge on shm_mnt and all mounts, for emergency use; | |
526 | * SHMEM_HUGE_FORCE: | |
527 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
528 | * | |
529 | */ | |
530 | #define SHMEM_HUGE_DENY (-1) | |
531 | #define SHMEM_HUGE_FORCE (-2) | |
532 | ||
396bcc52 | 533 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5a6e75f8 KS |
534 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
535 | ||
5e6e5a12 | 536 | static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER; |
5a6e75f8 | 537 | |
2cf13384 DS |
538 | bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force, |
539 | struct mm_struct *mm, unsigned long vm_flags) | |
c852023e | 540 | { |
c852023e | 541 | loff_t i_size; |
c852023e | 542 | |
f7cd16a5 XR |
543 | if (!S_ISREG(inode->i_mode)) |
544 | return false; | |
2cf13384 | 545 | if (mm && ((vm_flags & VM_NOHUGEPAGE) || test_bit(MMF_DISABLE_THP, &mm->flags))) |
c852023e | 546 | return false; |
7c6c6cc4 ZK |
547 | if (shmem_huge == SHMEM_HUGE_DENY) |
548 | return false; | |
3de0c269 ZK |
549 | if (shmem_huge_force || shmem_huge == SHMEM_HUGE_FORCE) |
550 | return true; | |
5e6e5a12 HD |
551 | |
552 | switch (SHMEM_SB(inode->i_sb)->huge) { | |
c852023e HD |
553 | case SHMEM_HUGE_ALWAYS: |
554 | return true; | |
555 | case SHMEM_HUGE_WITHIN_SIZE: | |
de6ee659 | 556 | index = round_up(index + 1, HPAGE_PMD_NR); |
c852023e | 557 | i_size = round_up(i_size_read(inode), PAGE_SIZE); |
de6ee659 | 558 | if (i_size >> PAGE_SHIFT >= index) |
c852023e HD |
559 | return true; |
560 | fallthrough; | |
561 | case SHMEM_HUGE_ADVISE: | |
2cf13384 | 562 | if (mm && (vm_flags & VM_HUGEPAGE)) |
5e6e5a12 HD |
563 | return true; |
564 | fallthrough; | |
c852023e | 565 | default: |
c852023e HD |
566 | return false; |
567 | } | |
568 | } | |
5a6e75f8 | 569 | |
e5f2249a | 570 | #if defined(CONFIG_SYSFS) |
5a6e75f8 KS |
571 | static int shmem_parse_huge(const char *str) |
572 | { | |
573 | if (!strcmp(str, "never")) | |
574 | return SHMEM_HUGE_NEVER; | |
575 | if (!strcmp(str, "always")) | |
576 | return SHMEM_HUGE_ALWAYS; | |
577 | if (!strcmp(str, "within_size")) | |
578 | return SHMEM_HUGE_WITHIN_SIZE; | |
579 | if (!strcmp(str, "advise")) | |
580 | return SHMEM_HUGE_ADVISE; | |
581 | if (!strcmp(str, "deny")) | |
582 | return SHMEM_HUGE_DENY; | |
583 | if (!strcmp(str, "force")) | |
584 | return SHMEM_HUGE_FORCE; | |
585 | return -EINVAL; | |
586 | } | |
e5f2249a | 587 | #endif |
5a6e75f8 | 588 | |
e5f2249a | 589 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
590 | static const char *shmem_format_huge(int huge) |
591 | { | |
592 | switch (huge) { | |
593 | case SHMEM_HUGE_NEVER: | |
594 | return "never"; | |
595 | case SHMEM_HUGE_ALWAYS: | |
596 | return "always"; | |
597 | case SHMEM_HUGE_WITHIN_SIZE: | |
598 | return "within_size"; | |
599 | case SHMEM_HUGE_ADVISE: | |
600 | return "advise"; | |
601 | case SHMEM_HUGE_DENY: | |
602 | return "deny"; | |
603 | case SHMEM_HUGE_FORCE: | |
604 | return "force"; | |
605 | default: | |
606 | VM_BUG_ON(1); | |
607 | return "bad_val"; | |
608 | } | |
609 | } | |
f1f5929c | 610 | #endif |
5a6e75f8 | 611 | |
779750d2 KS |
612 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
613 | struct shrink_control *sc, unsigned long nr_to_split) | |
614 | { | |
615 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 616 | LIST_HEAD(to_remove); |
779750d2 KS |
617 | struct inode *inode; |
618 | struct shmem_inode_info *info; | |
05624571 | 619 | struct folio *folio; |
779750d2 | 620 | unsigned long batch = sc ? sc->nr_to_scan : 128; |
62c9827c | 621 | int split = 0; |
779750d2 KS |
622 | |
623 | if (list_empty(&sbinfo->shrinklist)) | |
624 | return SHRINK_STOP; | |
625 | ||
626 | spin_lock(&sbinfo->shrinklist_lock); | |
627 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
628 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
629 | ||
630 | /* pin the inode */ | |
631 | inode = igrab(&info->vfs_inode); | |
632 | ||
633 | /* inode is about to be evicted */ | |
634 | if (!inode) { | |
635 | list_del_init(&info->shrinklist); | |
779750d2 KS |
636 | goto next; |
637 | } | |
638 | ||
639 | /* Check if there's anything to gain */ | |
640 | if (round_up(inode->i_size, PAGE_SIZE) == | |
641 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 642 | list_move(&info->shrinklist, &to_remove); |
779750d2 KS |
643 | goto next; |
644 | } | |
645 | ||
646 | list_move(&info->shrinklist, &list); | |
647 | next: | |
62c9827c | 648 | sbinfo->shrinklist_len--; |
779750d2 KS |
649 | if (!--batch) |
650 | break; | |
651 | } | |
652 | spin_unlock(&sbinfo->shrinklist_lock); | |
653 | ||
253fd0f0 KS |
654 | list_for_each_safe(pos, next, &to_remove) { |
655 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
656 | inode = &info->vfs_inode; | |
657 | list_del_init(&info->shrinklist); | |
658 | iput(inode); | |
659 | } | |
660 | ||
779750d2 KS |
661 | list_for_each_safe(pos, next, &list) { |
662 | int ret; | |
05624571 | 663 | pgoff_t index; |
779750d2 KS |
664 | |
665 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
666 | inode = &info->vfs_inode; | |
667 | ||
b3cd54b2 | 668 | if (nr_to_split && split >= nr_to_split) |
62c9827c | 669 | goto move_back; |
779750d2 | 670 | |
05624571 MWO |
671 | index = (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT; |
672 | folio = filemap_get_folio(inode->i_mapping, index); | |
66dabbb6 | 673 | if (IS_ERR(folio)) |
779750d2 KS |
674 | goto drop; |
675 | ||
b3cd54b2 | 676 | /* No huge page at the end of the file: nothing to split */ |
05624571 MWO |
677 | if (!folio_test_large(folio)) { |
678 | folio_put(folio); | |
779750d2 KS |
679 | goto drop; |
680 | } | |
681 | ||
b3cd54b2 | 682 | /* |
62c9827c GL |
683 | * Move the inode on the list back to shrinklist if we failed |
684 | * to lock the page at this time. | |
b3cd54b2 KS |
685 | * |
686 | * Waiting for the lock may lead to deadlock in the | |
687 | * reclaim path. | |
688 | */ | |
05624571 MWO |
689 | if (!folio_trylock(folio)) { |
690 | folio_put(folio); | |
62c9827c | 691 | goto move_back; |
b3cd54b2 KS |
692 | } |
693 | ||
d788f5b3 | 694 | ret = split_folio(folio); |
05624571 MWO |
695 | folio_unlock(folio); |
696 | folio_put(folio); | |
779750d2 | 697 | |
62c9827c | 698 | /* If split failed move the inode on the list back to shrinklist */ |
b3cd54b2 | 699 | if (ret) |
62c9827c | 700 | goto move_back; |
779750d2 KS |
701 | |
702 | split++; | |
703 | drop: | |
704 | list_del_init(&info->shrinklist); | |
62c9827c GL |
705 | goto put; |
706 | move_back: | |
707 | /* | |
708 | * Make sure the inode is either on the global list or deleted | |
709 | * from any local list before iput() since it could be deleted | |
710 | * in another thread once we put the inode (then the local list | |
711 | * is corrupted). | |
712 | */ | |
713 | spin_lock(&sbinfo->shrinklist_lock); | |
714 | list_move(&info->shrinklist, &sbinfo->shrinklist); | |
715 | sbinfo->shrinklist_len++; | |
716 | spin_unlock(&sbinfo->shrinklist_lock); | |
717 | put: | |
779750d2 KS |
718 | iput(inode); |
719 | } | |
720 | ||
779750d2 KS |
721 | return split; |
722 | } | |
723 | ||
724 | static long shmem_unused_huge_scan(struct super_block *sb, | |
725 | struct shrink_control *sc) | |
726 | { | |
727 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
728 | ||
729 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
730 | return SHRINK_STOP; | |
731 | ||
732 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
733 | } | |
734 | ||
735 | static long shmem_unused_huge_count(struct super_block *sb, | |
736 | struct shrink_control *sc) | |
737 | { | |
738 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
739 | return READ_ONCE(sbinfo->shrinklist_len); | |
740 | } | |
396bcc52 | 741 | #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ |
5a6e75f8 KS |
742 | |
743 | #define shmem_huge SHMEM_HUGE_DENY | |
744 | ||
2cf13384 DS |
745 | bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force, |
746 | struct mm_struct *mm, unsigned long vm_flags) | |
5e6e5a12 HD |
747 | { |
748 | return false; | |
749 | } | |
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 DH |
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 | |
09cbfeaf | 1969 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1970 | return -EFBIG; |
1da177e4 | 1971 | repeat: |
c5bf121e | 1972 | if (sgp <= SGP_CACHE && |
3022fd7a | 1973 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) |
c5bf121e | 1974 | return -EINVAL; |
c5bf121e | 1975 | |
3022fd7a | 1976 | alloced = false; |
054a9f7c | 1977 | fault_mm = vma ? vma->vm_mm : NULL; |
c5bf121e | 1978 | |
3022fd7a | 1979 | folio = filemap_get_entry(inode->i_mapping, index); |
b1d0ec3a | 1980 | if (folio && vma && userfaultfd_minor(vma)) { |
aaeb94eb | 1981 | if (!xa_is_value(folio)) |
b1d0ec3a | 1982 | folio_put(folio); |
c949b097 AR |
1983 | *fault_type = handle_userfault(vmf, VM_UFFD_MINOR); |
1984 | return 0; | |
1985 | } | |
1986 | ||
b1d0ec3a | 1987 | if (xa_is_value(folio)) { |
da08e9b7 | 1988 | error = shmem_swapin_folio(inode, index, &folio, |
054a9f7c | 1989 | sgp, gfp, fault_mm, fault_type); |
c5bf121e VRP |
1990 | if (error == -EEXIST) |
1991 | goto repeat; | |
54af6042 | 1992 | |
fc26babb | 1993 | *foliop = folio; |
c5bf121e | 1994 | return error; |
54af6042 HD |
1995 | } |
1996 | ||
b1d0ec3a | 1997 | if (folio) { |
aaeb94eb CH |
1998 | folio_lock(folio); |
1999 | ||
2000 | /* Has the folio been truncated or swapped out? */ | |
3022fd7a | 2001 | if (unlikely(folio->mapping != inode->i_mapping)) { |
aaeb94eb CH |
2002 | folio_unlock(folio); |
2003 | folio_put(folio); | |
2004 | goto repeat; | |
2005 | } | |
acdd9f8e | 2006 | if (sgp == SGP_WRITE) |
b1d0ec3a MWO |
2007 | folio_mark_accessed(folio); |
2008 | if (folio_test_uptodate(folio)) | |
acdd9f8e | 2009 | goto out; |
fc26babb | 2010 | /* fallocated folio */ |
1635f6a7 HD |
2011 | if (sgp != SGP_READ) |
2012 | goto clear; | |
b1d0ec3a MWO |
2013 | folio_unlock(folio); |
2014 | folio_put(folio); | |
1635f6a7 | 2015 | } |
27ab7006 HD |
2016 | |
2017 | /* | |
fc26babb MWO |
2018 | * SGP_READ: succeed on hole, with NULL folio, letting caller zero. |
2019 | * SGP_NOALLOC: fail on hole, with NULL folio, letting caller fail. | |
acdd9f8e | 2020 | */ |
fc26babb | 2021 | *foliop = NULL; |
acdd9f8e HD |
2022 | if (sgp == SGP_READ) |
2023 | return 0; | |
2024 | if (sgp == SGP_NOALLOC) | |
2025 | return -ENOENT; | |
2026 | ||
2027 | /* | |
2028 | * Fast cache lookup and swap lookup did not find it: allocate. | |
27ab7006 | 2029 | */ |
54af6042 | 2030 | |
c5bf121e VRP |
2031 | if (vma && userfaultfd_missing(vma)) { |
2032 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
2033 | return 0; | |
2034 | } | |
cfda0526 | 2035 | |
3022fd7a HD |
2036 | if (shmem_is_huge(inode, index, false, fault_mm, |
2037 | vma ? vma->vm_flags : 0)) { | |
2038 | gfp_t huge_gfp; | |
1da177e4 | 2039 | |
3022fd7a HD |
2040 | huge_gfp = vma_thp_gfp_mask(vma); |
2041 | huge_gfp = limit_gfp_mask(huge_gfp, gfp); | |
2042 | folio = shmem_alloc_and_add_folio(huge_gfp, | |
2043 | inode, index, fault_mm, true); | |
2044 | if (!IS_ERR(folio)) { | |
2045 | count_vm_event(THP_FILE_ALLOC); | |
2046 | goto alloced; | |
2047 | } | |
2048 | if (PTR_ERR(folio) == -EEXIST) | |
2049 | goto repeat; | |
c5bf121e | 2050 | } |
800d8c63 | 2051 | |
3022fd7a HD |
2052 | folio = shmem_alloc_and_add_folio(gfp, inode, index, fault_mm, false); |
2053 | if (IS_ERR(folio)) { | |
b1d0ec3a | 2054 | error = PTR_ERR(folio); |
3022fd7a HD |
2055 | if (error == -EEXIST) |
2056 | goto repeat; | |
b1d0ec3a | 2057 | folio = NULL; |
c5bf121e VRP |
2058 | goto unlock; |
2059 | } | |
54af6042 | 2060 | |
3022fd7a | 2061 | alloced: |
c5bf121e | 2062 | alloced = true; |
b1d0ec3a | 2063 | if (folio_test_pmd_mappable(folio) && |
c5bf121e | 2064 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < |
fc26babb | 2065 | folio_next_index(folio) - 1) { |
3022fd7a HD |
2066 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
2067 | struct shmem_inode_info *info = SHMEM_I(inode); | |
ec9516fb | 2068 | /* |
fc26babb | 2069 | * Part of the large folio is beyond i_size: subject |
c5bf121e | 2070 | * to shrink under memory pressure. |
1635f6a7 | 2071 | */ |
c5bf121e | 2072 | spin_lock(&sbinfo->shrinklist_lock); |
1635f6a7 | 2073 | /* |
c5bf121e VRP |
2074 | * _careful to defend against unlocked access to |
2075 | * ->shrink_list in shmem_unused_huge_shrink() | |
ec9516fb | 2076 | */ |
c5bf121e VRP |
2077 | if (list_empty_careful(&info->shrinklist)) { |
2078 | list_add_tail(&info->shrinklist, | |
2079 | &sbinfo->shrinklist); | |
2080 | sbinfo->shrinklist_len++; | |
2081 | } | |
2082 | spin_unlock(&sbinfo->shrinklist_lock); | |
2083 | } | |
800d8c63 | 2084 | |
3022fd7a HD |
2085 | if (sgp == SGP_WRITE) |
2086 | folio_set_referenced(folio); | |
c5bf121e | 2087 | /* |
fc26babb | 2088 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new folio. |
c5bf121e VRP |
2089 | */ |
2090 | if (sgp == SGP_FALLOC) | |
2091 | sgp = SGP_WRITE; | |
2092 | clear: | |
2093 | /* | |
fc26babb MWO |
2094 | * Let SGP_WRITE caller clear ends if write does not fill folio; |
2095 | * but SGP_FALLOC on a folio fallocated earlier must initialize | |
c5bf121e VRP |
2096 | * it now, lest undo on failure cancel our earlier guarantee. |
2097 | */ | |
b1d0ec3a MWO |
2098 | if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { |
2099 | long i, n = folio_nr_pages(folio); | |
c5bf121e | 2100 | |
b1d0ec3a MWO |
2101 | for (i = 0; i < n; i++) |
2102 | clear_highpage(folio_page(folio, i)); | |
2103 | flush_dcache_folio(folio); | |
2104 | folio_mark_uptodate(folio); | |
1da177e4 | 2105 | } |
bde05d1c | 2106 | |
54af6042 | 2107 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 2108 | if (sgp <= SGP_CACHE && |
09cbfeaf | 2109 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
54af6042 | 2110 | error = -EINVAL; |
267a4c76 | 2111 | goto unlock; |
e83c32e8 | 2112 | } |
63ec1973 | 2113 | out: |
fc26babb | 2114 | *foliop = folio; |
54af6042 | 2115 | return 0; |
1da177e4 | 2116 | |
59a16ead | 2117 | /* |
54af6042 | 2118 | * Error recovery. |
59a16ead | 2119 | */ |
d1899228 | 2120 | unlock: |
3022fd7a HD |
2121 | if (alloced) |
2122 | filemap_remove_folio(folio); | |
2123 | shmem_recalc_inode(inode, 0, 0); | |
b1d0ec3a MWO |
2124 | if (folio) { |
2125 | folio_unlock(folio); | |
2126 | folio_put(folio); | |
54af6042 | 2127 | } |
54af6042 | 2128 | return error; |
1da177e4 LT |
2129 | } |
2130 | ||
4e1fc793 MWO |
2131 | int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop, |
2132 | enum sgp_type sgp) | |
2133 | { | |
2134 | return shmem_get_folio_gfp(inode, index, foliop, sgp, | |
e3e1a506 | 2135 | mapping_gfp_mask(inode->i_mapping), NULL, NULL); |
4e1fc793 MWO |
2136 | } |
2137 | ||
10d20bd2 LT |
2138 | /* |
2139 | * This is like autoremove_wake_function, but it removes the wait queue | |
2140 | * entry unconditionally - even if something else had already woken the | |
2141 | * target. | |
2142 | */ | |
f0a9ad1d HD |
2143 | static int synchronous_wake_function(wait_queue_entry_t *wait, |
2144 | unsigned int mode, int sync, void *key) | |
10d20bd2 LT |
2145 | { |
2146 | int ret = default_wake_function(wait, mode, sync, key); | |
2055da97 | 2147 | list_del_init(&wait->entry); |
10d20bd2 LT |
2148 | return ret; |
2149 | } | |
2150 | ||
f0a9ad1d HD |
2151 | /* |
2152 | * Trinity finds that probing a hole which tmpfs is punching can | |
2153 | * prevent the hole-punch from ever completing: which in turn | |
2154 | * locks writers out with its hold on i_rwsem. So refrain from | |
2155 | * faulting pages into the hole while it's being punched. Although | |
2156 | * shmem_undo_range() does remove the additions, it may be unable to | |
2157 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
2158 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
2159 | * | |
2160 | * It does not matter if we sometimes reach this check just before the | |
2161 | * hole-punch begins, so that one fault then races with the punch: | |
2162 | * we just need to make racing faults a rare case. | |
2163 | * | |
2164 | * The implementation below would be much simpler if we just used a | |
2165 | * standard mutex or completion: but we cannot take i_rwsem in fault, | |
2166 | * and bloating every shmem inode for this unlikely case would be sad. | |
2167 | */ | |
2168 | static vm_fault_t shmem_falloc_wait(struct vm_fault *vmf, struct inode *inode) | |
2169 | { | |
2170 | struct shmem_falloc *shmem_falloc; | |
2171 | struct file *fpin = NULL; | |
2172 | vm_fault_t ret = 0; | |
2173 | ||
2174 | spin_lock(&inode->i_lock); | |
2175 | shmem_falloc = inode->i_private; | |
2176 | if (shmem_falloc && | |
2177 | shmem_falloc->waitq && | |
2178 | vmf->pgoff >= shmem_falloc->start && | |
2179 | vmf->pgoff < shmem_falloc->next) { | |
2180 | wait_queue_head_t *shmem_falloc_waitq; | |
2181 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); | |
2182 | ||
2183 | ret = VM_FAULT_NOPAGE; | |
2184 | fpin = maybe_unlock_mmap_for_io(vmf, NULL); | |
2185 | shmem_falloc_waitq = shmem_falloc->waitq; | |
2186 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
2187 | TASK_UNINTERRUPTIBLE); | |
2188 | spin_unlock(&inode->i_lock); | |
2189 | schedule(); | |
2190 | ||
2191 | /* | |
2192 | * shmem_falloc_waitq points into the shmem_fallocate() | |
2193 | * stack of the hole-punching task: shmem_falloc_waitq | |
2194 | * is usually invalid by the time we reach here, but | |
2195 | * finish_wait() does not dereference it in that case; | |
2196 | * though i_lock needed lest racing with wake_up_all(). | |
2197 | */ | |
2198 | spin_lock(&inode->i_lock); | |
2199 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
2200 | } | |
2201 | spin_unlock(&inode->i_lock); | |
2202 | if (fpin) { | |
2203 | fput(fpin); | |
2204 | ret = VM_FAULT_RETRY; | |
2205 | } | |
2206 | return ret; | |
2207 | } | |
2208 | ||
20acce67 | 2209 | static vm_fault_t shmem_fault(struct vm_fault *vmf) |
1da177e4 | 2210 | { |
f0a9ad1d | 2211 | struct inode *inode = file_inode(vmf->vma->vm_file); |
9e18eb29 | 2212 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
68a54100 | 2213 | struct folio *folio = NULL; |
f0a9ad1d | 2214 | vm_fault_t ret = 0; |
20acce67 | 2215 | int err; |
1da177e4 | 2216 | |
f00cdc6d HD |
2217 | /* |
2218 | * Trinity finds that probing a hole which tmpfs is punching can | |
f0a9ad1d | 2219 | * prevent the hole-punch from ever completing: noted in i_private. |
f00cdc6d HD |
2220 | */ |
2221 | if (unlikely(inode->i_private)) { | |
f0a9ad1d HD |
2222 | ret = shmem_falloc_wait(vmf, inode); |
2223 | if (ret) | |
8e205f77 | 2224 | return ret; |
f00cdc6d HD |
2225 | } |
2226 | ||
f0a9ad1d | 2227 | WARN_ON_ONCE(vmf->page != NULL); |
68a54100 | 2228 | err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE, |
e3e1a506 | 2229 | gfp, vmf, &ret); |
20acce67 SJ |
2230 | if (err) |
2231 | return vmf_error(err); | |
f0a9ad1d | 2232 | if (folio) { |
68a54100 | 2233 | vmf->page = folio_file_page(folio, vmf->pgoff); |
f0a9ad1d HD |
2234 | ret |= VM_FAULT_LOCKED; |
2235 | } | |
68da9f05 | 2236 | return ret; |
1da177e4 LT |
2237 | } |
2238 | ||
c01d5b30 HD |
2239 | unsigned long shmem_get_unmapped_area(struct file *file, |
2240 | unsigned long uaddr, unsigned long len, | |
2241 | unsigned long pgoff, unsigned long flags) | |
2242 | { | |
2243 | unsigned long (*get_area)(struct file *, | |
2244 | unsigned long, unsigned long, unsigned long, unsigned long); | |
2245 | unsigned long addr; | |
2246 | unsigned long offset; | |
2247 | unsigned long inflated_len; | |
2248 | unsigned long inflated_addr; | |
2249 | unsigned long inflated_offset; | |
2250 | ||
2251 | if (len > TASK_SIZE) | |
2252 | return -ENOMEM; | |
2253 | ||
2254 | get_area = current->mm->get_unmapped_area; | |
2255 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2256 | ||
396bcc52 | 2257 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) |
c01d5b30 HD |
2258 | return addr; |
2259 | if (IS_ERR_VALUE(addr)) | |
2260 | return addr; | |
2261 | if (addr & ~PAGE_MASK) | |
2262 | return addr; | |
2263 | if (addr > TASK_SIZE - len) | |
2264 | return addr; | |
2265 | ||
2266 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2267 | return addr; | |
2268 | if (len < HPAGE_PMD_SIZE) | |
2269 | return addr; | |
2270 | if (flags & MAP_FIXED) | |
2271 | return addr; | |
2272 | /* | |
2273 | * Our priority is to support MAP_SHARED mapped hugely; | |
2274 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
99158997 KS |
2275 | * But if caller specified an address hint and we allocated area there |
2276 | * successfully, respect that as before. | |
c01d5b30 | 2277 | */ |
99158997 | 2278 | if (uaddr == addr) |
c01d5b30 HD |
2279 | return addr; |
2280 | ||
2281 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2282 | struct super_block *sb; | |
2283 | ||
2284 | if (file) { | |
2285 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2286 | sb = file_inode(file)->i_sb; | |
2287 | } else { | |
2288 | /* | |
2289 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2290 | * for "/dev/zero", to create a shared anonymous object. | |
2291 | */ | |
2292 | if (IS_ERR(shm_mnt)) | |
2293 | return addr; | |
2294 | sb = shm_mnt->mnt_sb; | |
2295 | } | |
3089bf61 | 2296 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2297 | return addr; |
2298 | } | |
2299 | ||
2300 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2301 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2302 | return addr; | |
2303 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2304 | return addr; | |
2305 | ||
2306 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2307 | if (inflated_len > TASK_SIZE) | |
2308 | return addr; | |
2309 | if (inflated_len < len) | |
2310 | return addr; | |
2311 | ||
99158997 | 2312 | inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags); |
c01d5b30 HD |
2313 | if (IS_ERR_VALUE(inflated_addr)) |
2314 | return addr; | |
2315 | if (inflated_addr & ~PAGE_MASK) | |
2316 | return addr; | |
2317 | ||
2318 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2319 | inflated_addr += offset - inflated_offset; | |
2320 | if (inflated_offset > offset) | |
2321 | inflated_addr += HPAGE_PMD_SIZE; | |
2322 | ||
2323 | if (inflated_addr > TASK_SIZE - len) | |
2324 | return addr; | |
2325 | return inflated_addr; | |
2326 | } | |
2327 | ||
1da177e4 | 2328 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2329 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2330 | { |
496ad9aa | 2331 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2332 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2333 | } |
2334 | ||
d8dc74f2 | 2335 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
ddc1a5cb | 2336 | unsigned long addr, pgoff_t *ilx) |
1da177e4 | 2337 | { |
496ad9aa | 2338 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2339 | pgoff_t index; |
1da177e4 | 2340 | |
ddc1a5cb HD |
2341 | /* |
2342 | * Bias interleave by inode number to distribute better across nodes; | |
2343 | * but this interface is independent of which page order is used, so | |
2344 | * supplies only that bias, letting caller apply the offset (adjusted | |
2345 | * by page order, as in shmem_get_pgoff_policy() and get_vma_policy()). | |
2346 | */ | |
2347 | *ilx = inode->i_ino; | |
41ffe5d5 HD |
2348 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2349 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 | 2350 | } |
ddc1a5cb HD |
2351 | |
2352 | static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info, | |
2353 | pgoff_t index, unsigned int order, pgoff_t *ilx) | |
2354 | { | |
2355 | struct mempolicy *mpol; | |
2356 | ||
2357 | /* Bias interleave by inode number to distribute better across nodes */ | |
2358 | *ilx = info->vfs_inode.i_ino + (index >> order); | |
2359 | ||
2360 | mpol = mpol_shared_policy_lookup(&info->policy, index); | |
2361 | return mpol ? mpol : get_task_policy(current); | |
2362 | } | |
2363 | #else | |
2364 | static struct mempolicy *shmem_get_pgoff_policy(struct shmem_inode_info *info, | |
2365 | pgoff_t index, unsigned int order, pgoff_t *ilx) | |
2366 | { | |
2367 | *ilx = 0; | |
2368 | return NULL; | |
2369 | } | |
2370 | #endif /* CONFIG_NUMA */ | |
1da177e4 | 2371 | |
d7c9e99a | 2372 | int shmem_lock(struct file *file, int lock, struct ucounts *ucounts) |
1da177e4 | 2373 | { |
496ad9aa | 2374 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2375 | struct shmem_inode_info *info = SHMEM_I(inode); |
2376 | int retval = -ENOMEM; | |
2377 | ||
ea0dfeb4 HD |
2378 | /* |
2379 | * What serializes the accesses to info->flags? | |
2380 | * ipc_lock_object() when called from shmctl_do_lock(), | |
2381 | * no serialization needed when called from shm_destroy(). | |
2382 | */ | |
1da177e4 | 2383 | if (lock && !(info->flags & VM_LOCKED)) { |
d7c9e99a | 2384 | if (!user_shm_lock(inode->i_size, ucounts)) |
1da177e4 LT |
2385 | goto out_nomem; |
2386 | info->flags |= VM_LOCKED; | |
89e004ea | 2387 | mapping_set_unevictable(file->f_mapping); |
1da177e4 | 2388 | } |
d7c9e99a AG |
2389 | if (!lock && (info->flags & VM_LOCKED) && ucounts) { |
2390 | user_shm_unlock(inode->i_size, ucounts); | |
1da177e4 | 2391 | info->flags &= ~VM_LOCKED; |
89e004ea | 2392 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2393 | } |
2394 | retval = 0; | |
89e004ea | 2395 | |
1da177e4 | 2396 | out_nomem: |
1da177e4 LT |
2397 | return retval; |
2398 | } | |
2399 | ||
9b83a6a8 | 2400 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 | 2401 | { |
d09e8ca6 PT |
2402 | struct inode *inode = file_inode(file); |
2403 | struct shmem_inode_info *info = SHMEM_I(inode); | |
22247efd | 2404 | int ret; |
ab3948f5 | 2405 | |
28464bbb | 2406 | ret = seal_check_write(info->seals, vma); |
22247efd PX |
2407 | if (ret) |
2408 | return ret; | |
ab3948f5 | 2409 | |
51b0bff2 | 2410 | /* arm64 - allow memory tagging on RAM-based files */ |
1c71222e | 2411 | vm_flags_set(vma, VM_MTE_ALLOWED); |
51b0bff2 | 2412 | |
1da177e4 | 2413 | file_accessed(file); |
d09e8ca6 PT |
2414 | /* This is anonymous shared memory if it is unlinked at the time of mmap */ |
2415 | if (inode->i_nlink) | |
2416 | vma->vm_ops = &shmem_vm_ops; | |
2417 | else | |
2418 | vma->vm_ops = &shmem_anon_vm_ops; | |
1da177e4 LT |
2419 | return 0; |
2420 | } | |
2421 | ||
e88e0d36 HD |
2422 | static int shmem_file_open(struct inode *inode, struct file *file) |
2423 | { | |
2424 | file->f_mode |= FMODE_CAN_ODIRECT; | |
2425 | return generic_file_open(inode, file); | |
2426 | } | |
2427 | ||
cb241339 HD |
2428 | #ifdef CONFIG_TMPFS_XATTR |
2429 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2430 | ||
2431 | /* | |
2432 | * chattr's fsflags are unrelated to extended attributes, | |
2433 | * but tmpfs has chosen to enable them under the same config option. | |
2434 | */ | |
2435 | static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags) | |
2436 | { | |
2437 | unsigned int i_flags = 0; | |
2438 | ||
2439 | if (fsflags & FS_NOATIME_FL) | |
2440 | i_flags |= S_NOATIME; | |
2441 | if (fsflags & FS_APPEND_FL) | |
2442 | i_flags |= S_APPEND; | |
2443 | if (fsflags & FS_IMMUTABLE_FL) | |
2444 | i_flags |= S_IMMUTABLE; | |
2445 | /* | |
2446 | * But FS_NODUMP_FL does not require any action in i_flags. | |
2447 | */ | |
2448 | inode_set_flags(inode, i_flags, S_NOATIME | S_APPEND | S_IMMUTABLE); | |
2449 | } | |
2450 | #else | |
2451 | static void shmem_set_inode_flags(struct inode *inode, unsigned int fsflags) | |
e408e695 | 2452 | { |
e408e695 | 2453 | } |
cb241339 HD |
2454 | #define shmem_initxattrs NULL |
2455 | #endif | |
e408e695 | 2456 | |
a2e45955 CL |
2457 | static struct offset_ctx *shmem_get_offset_ctx(struct inode *inode) |
2458 | { | |
2459 | return &SHMEM_I(inode)->dir_offsets; | |
2460 | } | |
2461 | ||
e09764cf CM |
2462 | static struct inode *__shmem_get_inode(struct mnt_idmap *idmap, |
2463 | struct super_block *sb, | |
2464 | struct inode *dir, umode_t mode, | |
2465 | dev_t dev, unsigned long flags) | |
1da177e4 LT |
2466 | { |
2467 | struct inode *inode; | |
2468 | struct shmem_inode_info *info; | |
2469 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
e809d5f0 | 2470 | ino_t ino; |
71480663 CM |
2471 | int err; |
2472 | ||
2473 | err = shmem_reserve_inode(sb, &ino); | |
2474 | if (err) | |
2475 | return ERR_PTR(err); | |
1da177e4 | 2476 | |
1da177e4 | 2477 | inode = new_inode(sb); |
71480663 | 2478 | if (!inode) { |
2daf18a7 | 2479 | shmem_free_inode(sb, 0); |
71480663 CM |
2480 | return ERR_PTR(-ENOSPC); |
2481 | } | |
b45d71fb | 2482 | |
71480663 CM |
2483 | inode->i_ino = ino; |
2484 | inode_init_owner(idmap, inode, dir, mode); | |
2485 | inode->i_blocks = 0; | |
cf2766bb | 2486 | simple_inode_init_ts(inode); |
71480663 CM |
2487 | inode->i_generation = get_random_u32(); |
2488 | info = SHMEM_I(inode); | |
2489 | memset(info, 0, (char *)inode - (char *)info); | |
2490 | spin_lock_init(&info->lock); | |
2491 | atomic_set(&info->stop_eviction, 0); | |
2492 | info->seals = F_SEAL_SEAL; | |
2493 | info->flags = flags & VM_NORESERVE; | |
cf2766bb | 2494 | info->i_crtime = inode_get_mtime(inode); |
71480663 CM |
2495 | info->fsflags = (dir == NULL) ? 0 : |
2496 | SHMEM_I(dir)->fsflags & SHMEM_FL_INHERITED; | |
2497 | if (info->fsflags) | |
2498 | shmem_set_inode_flags(inode, info->fsflags); | |
2499 | INIT_LIST_HEAD(&info->shrinklist); | |
2500 | INIT_LIST_HEAD(&info->swaplist); | |
71480663 CM |
2501 | simple_xattrs_init(&info->xattrs); |
2502 | cache_no_acl(inode); | |
9be7d5b0 HD |
2503 | if (sbinfo->noswap) |
2504 | mapping_set_unevictable(inode->i_mapping); | |
71480663 CM |
2505 | mapping_set_large_folios(inode->i_mapping); |
2506 | ||
2507 | switch (mode & S_IFMT) { | |
2508 | default: | |
2509 | inode->i_op = &shmem_special_inode_operations; | |
2510 | init_special_inode(inode, mode, dev); | |
2511 | break; | |
2512 | case S_IFREG: | |
2513 | inode->i_mapping->a_ops = &shmem_aops; | |
2514 | inode->i_op = &shmem_inode_operations; | |
2515 | inode->i_fop = &shmem_file_operations; | |
2516 | mpol_shared_policy_init(&info->policy, | |
2517 | shmem_get_sbmpol(sbinfo)); | |
2518 | break; | |
2519 | case S_IFDIR: | |
2520 | inc_nlink(inode); | |
2521 | /* Some things misbehave if size == 0 on a directory */ | |
2522 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2523 | inode->i_op = &shmem_dir_inode_operations; | |
a2e45955 CL |
2524 | inode->i_fop = &simple_offset_dir_operations; |
2525 | simple_offset_init(shmem_get_offset_ctx(inode)); | |
71480663 CM |
2526 | break; |
2527 | case S_IFLNK: | |
2528 | /* | |
2529 | * Must not load anything in the rbtree, | |
2530 | * mpol_free_shared_policy will not be called. | |
2531 | */ | |
2532 | mpol_shared_policy_init(&info->policy, NULL); | |
2533 | break; | |
2534 | } | |
2535 | ||
2536 | lockdep_annotate_inode_mutex_key(inode); | |
1da177e4 LT |
2537 | return inode; |
2538 | } | |
2539 | ||
e09764cf CM |
2540 | #ifdef CONFIG_TMPFS_QUOTA |
2541 | static struct inode *shmem_get_inode(struct mnt_idmap *idmap, | |
2542 | struct super_block *sb, struct inode *dir, | |
2543 | umode_t mode, dev_t dev, unsigned long flags) | |
2544 | { | |
2545 | int err; | |
2546 | struct inode *inode; | |
2547 | ||
2548 | inode = __shmem_get_inode(idmap, sb, dir, mode, dev, flags); | |
2549 | if (IS_ERR(inode)) | |
2550 | return inode; | |
2551 | ||
2552 | err = dquot_initialize(inode); | |
2553 | if (err) | |
2554 | goto errout; | |
2555 | ||
2556 | err = dquot_alloc_inode(inode); | |
2557 | if (err) { | |
2558 | dquot_drop(inode); | |
2559 | goto errout; | |
2560 | } | |
2561 | return inode; | |
2562 | ||
2563 | errout: | |
2564 | inode->i_flags |= S_NOQUOTA; | |
2565 | iput(inode); | |
2566 | return ERR_PTR(err); | |
2567 | } | |
2568 | #else | |
2569 | static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap, | |
2570 | struct super_block *sb, struct inode *dir, | |
2571 | umode_t mode, dev_t dev, unsigned long flags) | |
2572 | { | |
2573 | return __shmem_get_inode(idmap, sb, dir, mode, dev, flags); | |
2574 | } | |
2575 | #endif /* CONFIG_TMPFS_QUOTA */ | |
2576 | ||
3460f6e5 | 2577 | #ifdef CONFIG_USERFAULTFD |
61c50040 | 2578 | int shmem_mfill_atomic_pte(pmd_t *dst_pmd, |
3460f6e5 AR |
2579 | struct vm_area_struct *dst_vma, |
2580 | unsigned long dst_addr, | |
2581 | unsigned long src_addr, | |
d9712937 | 2582 | uffd_flags_t flags, |
d7be6d7e | 2583 | struct folio **foliop) |
4c27fe4c MR |
2584 | { |
2585 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2586 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4c27fe4c MR |
2587 | struct address_space *mapping = inode->i_mapping; |
2588 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2589 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
4c27fe4c | 2590 | void *page_kaddr; |
b7dd44a1 | 2591 | struct folio *folio; |
4c27fe4c | 2592 | int ret; |
3460f6e5 | 2593 | pgoff_t max_off; |
4c27fe4c | 2594 | |
4199f51a | 2595 | if (shmem_inode_acct_blocks(inode, 1)) { |
7ed9d238 AR |
2596 | /* |
2597 | * We may have got a page, returned -ENOENT triggering a retry, | |
2598 | * and now we find ourselves with -ENOMEM. Release the page, to | |
2599 | * avoid a BUG_ON in our caller. | |
2600 | */ | |
d7be6d7e Z |
2601 | if (unlikely(*foliop)) { |
2602 | folio_put(*foliop); | |
2603 | *foliop = NULL; | |
7ed9d238 | 2604 | } |
7d64ae3a | 2605 | return -ENOMEM; |
7ed9d238 | 2606 | } |
4c27fe4c | 2607 | |
d7be6d7e | 2608 | if (!*foliop) { |
7d64ae3a | 2609 | ret = -ENOMEM; |
7a7256d5 MWO |
2610 | folio = shmem_alloc_folio(gfp, info, pgoff); |
2611 | if (!folio) | |
0f079694 | 2612 | goto out_unacct_blocks; |
4c27fe4c | 2613 | |
d9712937 | 2614 | if (uffd_flags_mode_is(flags, MFILL_ATOMIC_COPY)) { |
7a7256d5 | 2615 | page_kaddr = kmap_local_folio(folio, 0); |
5dc21f0c IW |
2616 | /* |
2617 | * The read mmap_lock is held here. Despite the | |
2618 | * mmap_lock being read recursive a deadlock is still | |
2619 | * possible if a writer has taken a lock. For example: | |
2620 | * | |
2621 | * process A thread 1 takes read lock on own mmap_lock | |
2622 | * process A thread 2 calls mmap, blocks taking write lock | |
2623 | * process B thread 1 takes page fault, read lock on own mmap lock | |
2624 | * process B thread 2 calls mmap, blocks taking write lock | |
2625 | * process A thread 1 blocks taking read lock on process B | |
2626 | * process B thread 1 blocks taking read lock on process A | |
2627 | * | |
2628 | * Disable page faults to prevent potential deadlock | |
2629 | * and retry the copy outside the mmap_lock. | |
2630 | */ | |
2631 | pagefault_disable(); | |
8d103963 MR |
2632 | ret = copy_from_user(page_kaddr, |
2633 | (const void __user *)src_addr, | |
2634 | PAGE_SIZE); | |
5dc21f0c | 2635 | pagefault_enable(); |
7a7256d5 | 2636 | kunmap_local(page_kaddr); |
8d103963 | 2637 | |
c1e8d7c6 | 2638 | /* fallback to copy_from_user outside mmap_lock */ |
8d103963 | 2639 | if (unlikely(ret)) { |
d7be6d7e | 2640 | *foliop = folio; |
7d64ae3a | 2641 | ret = -ENOENT; |
8d103963 | 2642 | /* don't free the page */ |
7d64ae3a | 2643 | goto out_unacct_blocks; |
8d103963 | 2644 | } |
19b482c2 | 2645 | |
7a7256d5 | 2646 | flush_dcache_folio(folio); |
3460f6e5 | 2647 | } else { /* ZEROPAGE */ |
7a7256d5 | 2648 | clear_user_highpage(&folio->page, dst_addr); |
4c27fe4c MR |
2649 | } |
2650 | } else { | |
d7be6d7e | 2651 | folio = *foliop; |
7a7256d5 | 2652 | VM_BUG_ON_FOLIO(folio_test_large(folio), folio); |
d7be6d7e | 2653 | *foliop = NULL; |
4c27fe4c MR |
2654 | } |
2655 | ||
7a7256d5 MWO |
2656 | VM_BUG_ON(folio_test_locked(folio)); |
2657 | VM_BUG_ON(folio_test_swapbacked(folio)); | |
2658 | __folio_set_locked(folio); | |
2659 | __folio_set_swapbacked(folio); | |
2660 | __folio_mark_uptodate(folio); | |
9cc90c66 | 2661 | |
e2a50c1f | 2662 | ret = -EFAULT; |
e2a50c1f | 2663 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3460f6e5 | 2664 | if (unlikely(pgoff >= max_off)) |
e2a50c1f AA |
2665 | goto out_release; |
2666 | ||
054a9f7c HD |
2667 | ret = mem_cgroup_charge(folio, dst_vma->vm_mm, gfp); |
2668 | if (ret) | |
2669 | goto out_release; | |
2670 | ret = shmem_add_to_page_cache(folio, mapping, pgoff, NULL, gfp); | |
4c27fe4c | 2671 | if (ret) |
3fea5a49 | 2672 | goto out_release; |
4c27fe4c | 2673 | |
61c50040 | 2674 | ret = mfill_atomic_install_pte(dst_pmd, dst_vma, dst_addr, |
d9712937 | 2675 | &folio->page, true, flags); |
7d64ae3a AR |
2676 | if (ret) |
2677 | goto out_delete_from_cache; | |
4c27fe4c | 2678 | |
3c1b7528 | 2679 | shmem_recalc_inode(inode, 1, 0); |
7a7256d5 | 2680 | folio_unlock(folio); |
7d64ae3a AR |
2681 | return 0; |
2682 | out_delete_from_cache: | |
7a7256d5 | 2683 | filemap_remove_folio(folio); |
4c27fe4c | 2684 | out_release: |
7a7256d5 MWO |
2685 | folio_unlock(folio); |
2686 | folio_put(folio); | |
4c27fe4c | 2687 | out_unacct_blocks: |
0f079694 | 2688 | shmem_inode_unacct_blocks(inode, 1); |
7d64ae3a | 2689 | return ret; |
8d103963 | 2690 | } |
3460f6e5 | 2691 | #endif /* CONFIG_USERFAULTFD */ |
8d103963 | 2692 | |
1da177e4 | 2693 | #ifdef CONFIG_TMPFS |
92e1d5be | 2694 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2695 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2696 | |
1da177e4 | 2697 | static int |
800d15a5 | 2698 | shmem_write_begin(struct file *file, struct address_space *mapping, |
9d6b0cd7 | 2699 | loff_t pos, unsigned len, |
800d15a5 | 2700 | struct page **pagep, void **fsdata) |
1da177e4 | 2701 | { |
800d15a5 | 2702 | struct inode *inode = mapping->host; |
40e041a2 | 2703 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2704 | pgoff_t index = pos >> PAGE_SHIFT; |
eff1f906 | 2705 | struct folio *folio; |
a7605426 | 2706 | int ret = 0; |
40e041a2 | 2707 | |
9608703e | 2708 | /* i_rwsem is held by caller */ |
ab3948f5 JFG |
2709 | if (unlikely(info->seals & (F_SEAL_GROW | |
2710 | F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) { | |
2711 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) | |
40e041a2 DH |
2712 | return -EPERM; |
2713 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2714 | return -EPERM; | |
2715 | } | |
2716 | ||
eff1f906 | 2717 | ret = shmem_get_folio(inode, index, &folio, SGP_WRITE); |
a7605426 YS |
2718 | if (ret) |
2719 | return ret; | |
2720 | ||
eff1f906 | 2721 | *pagep = folio_file_page(folio, index); |
a7605426 | 2722 | if (PageHWPoison(*pagep)) { |
eff1f906 MWO |
2723 | folio_unlock(folio); |
2724 | folio_put(folio); | |
a7605426 YS |
2725 | *pagep = NULL; |
2726 | return -EIO; | |
2727 | } | |
2728 | ||
2729 | return 0; | |
800d15a5 NP |
2730 | } |
2731 | ||
2732 | static int | |
2733 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2734 | loff_t pos, unsigned len, unsigned copied, | |
2735 | struct page *page, void *fsdata) | |
2736 | { | |
69bbb87b | 2737 | struct folio *folio = page_folio(page); |
800d15a5 NP |
2738 | struct inode *inode = mapping->host; |
2739 | ||
d3602444 HD |
2740 | if (pos + copied > inode->i_size) |
2741 | i_size_write(inode, pos + copied); | |
2742 | ||
69bbb87b MWO |
2743 | if (!folio_test_uptodate(folio)) { |
2744 | if (copied < folio_size(folio)) { | |
2745 | size_t from = offset_in_folio(folio, pos); | |
2746 | folio_zero_segments(folio, 0, from, | |
2747 | from + copied, folio_size(folio)); | |
ec9516fb | 2748 | } |
69bbb87b | 2749 | folio_mark_uptodate(folio); |
ec9516fb | 2750 | } |
69bbb87b MWO |
2751 | folio_mark_dirty(folio); |
2752 | folio_unlock(folio); | |
2753 | folio_put(folio); | |
800d15a5 | 2754 | |
800d15a5 | 2755 | return copied; |
1da177e4 LT |
2756 | } |
2757 | ||
2ba5bbed | 2758 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2759 | { |
6e58e79d AV |
2760 | struct file *file = iocb->ki_filp; |
2761 | struct inode *inode = file_inode(file); | |
1da177e4 | 2762 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2763 | pgoff_t index; |
2764 | unsigned long offset; | |
f7c1d074 | 2765 | int error = 0; |
cb66a7a1 | 2766 | ssize_t retval = 0; |
6e58e79d | 2767 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 | 2768 | |
09cbfeaf KS |
2769 | index = *ppos >> PAGE_SHIFT; |
2770 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2771 | |
2772 | for (;;) { | |
4601e2fc | 2773 | struct folio *folio = NULL; |
1da177e4 | 2774 | struct page *page = NULL; |
41ffe5d5 HD |
2775 | pgoff_t end_index; |
2776 | unsigned long nr, ret; | |
1da177e4 LT |
2777 | loff_t i_size = i_size_read(inode); |
2778 | ||
09cbfeaf | 2779 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2780 | if (index > end_index) |
2781 | break; | |
2782 | if (index == end_index) { | |
09cbfeaf | 2783 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2784 | if (nr <= offset) |
2785 | break; | |
2786 | } | |
2787 | ||
4601e2fc | 2788 | error = shmem_get_folio(inode, index, &folio, SGP_READ); |
6e58e79d AV |
2789 | if (error) { |
2790 | if (error == -EINVAL) | |
2791 | error = 0; | |
1da177e4 LT |
2792 | break; |
2793 | } | |
4601e2fc MWO |
2794 | if (folio) { |
2795 | folio_unlock(folio); | |
a7605426 | 2796 | |
4601e2fc | 2797 | page = folio_file_page(folio, index); |
a7605426 | 2798 | if (PageHWPoison(page)) { |
4601e2fc | 2799 | folio_put(folio); |
a7605426 YS |
2800 | error = -EIO; |
2801 | break; | |
2802 | } | |
75edd345 | 2803 | } |
1da177e4 LT |
2804 | |
2805 | /* | |
2806 | * We must evaluate after, since reads (unlike writes) | |
9608703e | 2807 | * are called without i_rwsem protection against truncate |
1da177e4 | 2808 | */ |
09cbfeaf | 2809 | nr = PAGE_SIZE; |
1da177e4 | 2810 | i_size = i_size_read(inode); |
09cbfeaf | 2811 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2812 | if (index == end_index) { |
09cbfeaf | 2813 | nr = i_size & ~PAGE_MASK; |
1da177e4 | 2814 | if (nr <= offset) { |
4601e2fc MWO |
2815 | if (folio) |
2816 | folio_put(folio); | |
1da177e4 LT |
2817 | break; |
2818 | } | |
2819 | } | |
2820 | nr -= offset; | |
2821 | ||
4601e2fc | 2822 | if (folio) { |
1da177e4 LT |
2823 | /* |
2824 | * If users can be writing to this page using arbitrary | |
2825 | * virtual addresses, take care about potential aliasing | |
2826 | * before reading the page on the kernel side. | |
2827 | */ | |
2828 | if (mapping_writably_mapped(mapping)) | |
2829 | flush_dcache_page(page); | |
2830 | /* | |
2831 | * Mark the page accessed if we read the beginning. | |
2832 | */ | |
2833 | if (!offset) | |
4601e2fc | 2834 | folio_mark_accessed(folio); |
1bdec44b HD |
2835 | /* |
2836 | * Ok, we have the page, and it's up-to-date, so | |
2837 | * now we can copy it to user space... | |
2838 | */ | |
2839 | ret = copy_page_to_iter(page, offset, nr, to); | |
4601e2fc | 2840 | folio_put(folio); |
1bdec44b | 2841 | |
fcb14cb1 | 2842 | } else if (user_backed_iter(to)) { |
1bdec44b HD |
2843 | /* |
2844 | * Copy to user tends to be so well optimized, but | |
2845 | * clear_user() not so much, that it is noticeably | |
2846 | * faster to copy the zero page instead of clearing. | |
2847 | */ | |
2848 | ret = copy_page_to_iter(ZERO_PAGE(0), offset, nr, to); | |
b5810039 | 2849 | } else { |
1bdec44b HD |
2850 | /* |
2851 | * But submitting the same page twice in a row to | |
2852 | * splice() - or others? - can result in confusion: | |
2853 | * so don't attempt that optimization on pipes etc. | |
2854 | */ | |
2855 | ret = iov_iter_zero(nr, to); | |
b5810039 | 2856 | } |
1da177e4 | 2857 | |
6e58e79d | 2858 | retval += ret; |
1da177e4 | 2859 | offset += ret; |
09cbfeaf KS |
2860 | index += offset >> PAGE_SHIFT; |
2861 | offset &= ~PAGE_MASK; | |
1da177e4 | 2862 | |
2ba5bbed | 2863 | if (!iov_iter_count(to)) |
1da177e4 | 2864 | break; |
6e58e79d AV |
2865 | if (ret < nr) { |
2866 | error = -EFAULT; | |
2867 | break; | |
2868 | } | |
1da177e4 LT |
2869 | cond_resched(); |
2870 | } | |
2871 | ||
09cbfeaf | 2872 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2873 | file_accessed(file); |
2874 | return retval ? retval : error; | |
1da177e4 LT |
2875 | } |
2876 | ||
e88e0d36 HD |
2877 | static ssize_t shmem_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
2878 | { | |
2879 | struct file *file = iocb->ki_filp; | |
2880 | struct inode *inode = file->f_mapping->host; | |
2881 | ssize_t ret; | |
2882 | ||
2883 | inode_lock(inode); | |
2884 | ret = generic_write_checks(iocb, from); | |
2885 | if (ret <= 0) | |
2886 | goto unlock; | |
2887 | ret = file_remove_privs(file); | |
2888 | if (ret) | |
2889 | goto unlock; | |
2890 | ret = file_update_time(file); | |
2891 | if (ret) | |
2892 | goto unlock; | |
2893 | ret = generic_perform_write(iocb, from); | |
2894 | unlock: | |
2895 | inode_unlock(inode); | |
2896 | return ret; | |
2897 | } | |
2898 | ||
bd194b18 DH |
2899 | static bool zero_pipe_buf_get(struct pipe_inode_info *pipe, |
2900 | struct pipe_buffer *buf) | |
2901 | { | |
2902 | return true; | |
2903 | } | |
2904 | ||
2905 | static void zero_pipe_buf_release(struct pipe_inode_info *pipe, | |
2906 | struct pipe_buffer *buf) | |
2907 | { | |
2908 | } | |
2909 | ||
2910 | static bool zero_pipe_buf_try_steal(struct pipe_inode_info *pipe, | |
2911 | struct pipe_buffer *buf) | |
2912 | { | |
2913 | return false; | |
2914 | } | |
2915 | ||
2916 | static const struct pipe_buf_operations zero_pipe_buf_ops = { | |
2917 | .release = zero_pipe_buf_release, | |
2918 | .try_steal = zero_pipe_buf_try_steal, | |
2919 | .get = zero_pipe_buf_get, | |
2920 | }; | |
2921 | ||
2922 | static size_t splice_zeropage_into_pipe(struct pipe_inode_info *pipe, | |
2923 | loff_t fpos, size_t size) | |
2924 | { | |
2925 | size_t offset = fpos & ~PAGE_MASK; | |
2926 | ||
2927 | size = min_t(size_t, size, PAGE_SIZE - offset); | |
2928 | ||
2929 | if (!pipe_full(pipe->head, pipe->tail, pipe->max_usage)) { | |
2930 | struct pipe_buffer *buf = pipe_head_buf(pipe); | |
2931 | ||
2932 | *buf = (struct pipe_buffer) { | |
2933 | .ops = &zero_pipe_buf_ops, | |
2934 | .page = ZERO_PAGE(0), | |
2935 | .offset = offset, | |
2936 | .len = size, | |
2937 | }; | |
2938 | pipe->head++; | |
2939 | } | |
2940 | ||
2941 | return size; | |
2942 | } | |
2943 | ||
2944 | static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, | |
2945 | struct pipe_inode_info *pipe, | |
2946 | size_t len, unsigned int flags) | |
2947 | { | |
2948 | struct inode *inode = file_inode(in); | |
2949 | struct address_space *mapping = inode->i_mapping; | |
2950 | struct folio *folio = NULL; | |
2951 | size_t total_spliced = 0, used, npages, n, part; | |
2952 | loff_t isize; | |
2953 | int error = 0; | |
2954 | ||
2955 | /* Work out how much data we can actually add into the pipe */ | |
2956 | used = pipe_occupancy(pipe->head, pipe->tail); | |
2957 | npages = max_t(ssize_t, pipe->max_usage - used, 0); | |
2958 | len = min_t(size_t, len, npages * PAGE_SIZE); | |
2959 | ||
2960 | do { | |
2961 | if (*ppos >= i_size_read(inode)) | |
2962 | break; | |
2963 | ||
fa598952 HD |
2964 | error = shmem_get_folio(inode, *ppos / PAGE_SIZE, &folio, |
2965 | SGP_READ); | |
bd194b18 DH |
2966 | if (error) { |
2967 | if (error == -EINVAL) | |
2968 | error = 0; | |
2969 | break; | |
2970 | } | |
2971 | if (folio) { | |
2972 | folio_unlock(folio); | |
2973 | ||
fa598952 HD |
2974 | if (folio_test_hwpoison(folio) || |
2975 | (folio_test_large(folio) && | |
2976 | folio_test_has_hwpoisoned(folio))) { | |
bd194b18 DH |
2977 | error = -EIO; |
2978 | break; | |
2979 | } | |
2980 | } | |
2981 | ||
2982 | /* | |
2983 | * i_size must be checked after we know the pages are Uptodate. | |
2984 | * | |
2985 | * Checking i_size after the check allows us to calculate | |
2986 | * the correct value for "nr", which means the zero-filled | |
2987 | * part of the page is not copied back to userspace (unless | |
2988 | * another truncate extends the file - this is desired though). | |
2989 | */ | |
2990 | isize = i_size_read(inode); | |
2991 | if (unlikely(*ppos >= isize)) | |
2992 | break; | |
2993 | part = min_t(loff_t, isize - *ppos, len); | |
2994 | ||
2995 | if (folio) { | |
2996 | /* | |
2997 | * If users can be writing to this page using arbitrary | |
2998 | * virtual addresses, take care about potential aliasing | |
2999 | * before reading the page on the kernel side. | |
3000 | */ | |
3001 | if (mapping_writably_mapped(mapping)) | |
3002 | flush_dcache_folio(folio); | |
3003 | folio_mark_accessed(folio); | |
3004 | /* | |
3005 | * Ok, we have the page, and it's up-to-date, so we can | |
3006 | * now splice it into the pipe. | |
3007 | */ | |
3008 | n = splice_folio_into_pipe(pipe, folio, *ppos, part); | |
3009 | folio_put(folio); | |
3010 | folio = NULL; | |
3011 | } else { | |
fa598952 | 3012 | n = splice_zeropage_into_pipe(pipe, *ppos, part); |
bd194b18 DH |
3013 | } |
3014 | ||
3015 | if (!n) | |
3016 | break; | |
3017 | len -= n; | |
3018 | total_spliced += n; | |
3019 | *ppos += n; | |
3020 | in->f_ra.prev_pos = *ppos; | |
3021 | if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) | |
3022 | break; | |
3023 | ||
3024 | cond_resched(); | |
3025 | } while (len); | |
3026 | ||
3027 | if (folio) | |
3028 | folio_put(folio); | |
3029 | ||
3030 | file_accessed(in); | |
3031 | return total_spliced ? total_spliced : error; | |
3032 | } | |
3033 | ||
965c8e59 | 3034 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
3035 | { |
3036 | struct address_space *mapping = file->f_mapping; | |
3037 | struct inode *inode = mapping->host; | |
220f2ac9 | 3038 | |
965c8e59 AM |
3039 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
3040 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 3041 | MAX_LFS_FILESIZE, i_size_read(inode)); |
41139aa4 MWO |
3042 | if (offset < 0) |
3043 | return -ENXIO; | |
3044 | ||
5955102c | 3045 | inode_lock(inode); |
9608703e | 3046 | /* We're holding i_rwsem so we can access i_size directly */ |
41139aa4 | 3047 | offset = mapping_seek_hole_data(mapping, offset, inode->i_size, whence); |
387aae6f HD |
3048 | if (offset >= 0) |
3049 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 3050 | inode_unlock(inode); |
220f2ac9 HD |
3051 | return offset; |
3052 | } | |
3053 | ||
83e4fa9c HD |
3054 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
3055 | loff_t len) | |
3056 | { | |
496ad9aa | 3057 | struct inode *inode = file_inode(file); |
e2d12e22 | 3058 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 3059 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 3060 | struct shmem_falloc shmem_falloc; |
d144bf62 | 3061 | pgoff_t start, index, end, undo_fallocend; |
e2d12e22 | 3062 | int error; |
83e4fa9c | 3063 | |
13ace4d0 HD |
3064 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
3065 | return -EOPNOTSUPP; | |
3066 | ||
5955102c | 3067 | inode_lock(inode); |
83e4fa9c HD |
3068 | |
3069 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
3070 | struct address_space *mapping = file->f_mapping; | |
3071 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
3072 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 3073 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 3074 | |
9608703e | 3075 | /* protected by i_rwsem */ |
ab3948f5 | 3076 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { |
40e041a2 DH |
3077 | error = -EPERM; |
3078 | goto out; | |
3079 | } | |
3080 | ||
8e205f77 | 3081 | shmem_falloc.waitq = &shmem_falloc_waitq; |
aa71ecd8 | 3082 | shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT; |
f00cdc6d HD |
3083 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; |
3084 | spin_lock(&inode->i_lock); | |
3085 | inode->i_private = &shmem_falloc; | |
3086 | spin_unlock(&inode->i_lock); | |
3087 | ||
83e4fa9c HD |
3088 | if ((u64)unmap_end > (u64)unmap_start) |
3089 | unmap_mapping_range(mapping, unmap_start, | |
3090 | 1 + unmap_end - unmap_start, 0); | |
3091 | shmem_truncate_range(inode, offset, offset + len - 1); | |
3092 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
3093 | |
3094 | spin_lock(&inode->i_lock); | |
3095 | inode->i_private = NULL; | |
3096 | wake_up_all(&shmem_falloc_waitq); | |
2055da97 | 3097 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head)); |
8e205f77 | 3098 | spin_unlock(&inode->i_lock); |
83e4fa9c | 3099 | error = 0; |
8e205f77 | 3100 | goto out; |
e2d12e22 HD |
3101 | } |
3102 | ||
3103 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
3104 | error = inode_newsize_ok(inode, offset + len); | |
3105 | if (error) | |
3106 | goto out; | |
3107 | ||
40e041a2 DH |
3108 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
3109 | error = -EPERM; | |
3110 | goto out; | |
3111 | } | |
3112 | ||
09cbfeaf KS |
3113 | start = offset >> PAGE_SHIFT; |
3114 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
3115 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
3116 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
3117 | error = -ENOSPC; | |
3118 | goto out; | |
83e4fa9c HD |
3119 | } |
3120 | ||
8e205f77 | 3121 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
3122 | shmem_falloc.start = start; |
3123 | shmem_falloc.next = start; | |
3124 | shmem_falloc.nr_falloced = 0; | |
3125 | shmem_falloc.nr_unswapped = 0; | |
3126 | spin_lock(&inode->i_lock); | |
3127 | inode->i_private = &shmem_falloc; | |
3128 | spin_unlock(&inode->i_lock); | |
3129 | ||
d144bf62 HD |
3130 | /* |
3131 | * info->fallocend is only relevant when huge pages might be | |
3132 | * involved: to prevent split_huge_page() freeing fallocated | |
3133 | * pages when FALLOC_FL_KEEP_SIZE committed beyond i_size. | |
3134 | */ | |
3135 | undo_fallocend = info->fallocend; | |
3136 | if (info->fallocend < end) | |
3137 | info->fallocend = end; | |
3138 | ||
050dcb5c | 3139 | for (index = start; index < end; ) { |
b0802b22 | 3140 | struct folio *folio; |
e2d12e22 HD |
3141 | |
3142 | /* | |
3143 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
3144 | * been interrupted because we are using up too much memory. | |
3145 | */ | |
3146 | if (signal_pending(current)) | |
3147 | error = -EINTR; | |
1aac1400 HD |
3148 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
3149 | error = -ENOMEM; | |
e2d12e22 | 3150 | else |
b0802b22 MWO |
3151 | error = shmem_get_folio(inode, index, &folio, |
3152 | SGP_FALLOC); | |
e2d12e22 | 3153 | if (error) { |
d144bf62 | 3154 | info->fallocend = undo_fallocend; |
b0802b22 | 3155 | /* Remove the !uptodate folios we added */ |
7f556567 HD |
3156 | if (index > start) { |
3157 | shmem_undo_range(inode, | |
3158 | (loff_t)start << PAGE_SHIFT, | |
3159 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
3160 | } | |
1aac1400 | 3161 | goto undone; |
e2d12e22 HD |
3162 | } |
3163 | ||
050dcb5c HD |
3164 | /* |
3165 | * Here is a more important optimization than it appears: | |
b0802b22 MWO |
3166 | * a second SGP_FALLOC on the same large folio will clear it, |
3167 | * making it uptodate and un-undoable if we fail later. | |
050dcb5c | 3168 | */ |
b0802b22 MWO |
3169 | index = folio_next_index(folio); |
3170 | /* Beware 32-bit wraparound */ | |
3171 | if (!index) | |
3172 | index--; | |
050dcb5c | 3173 | |
1aac1400 HD |
3174 | /* |
3175 | * Inform shmem_writepage() how far we have reached. | |
3176 | * No need for lock or barrier: we have the page lock. | |
3177 | */ | |
b0802b22 | 3178 | if (!folio_test_uptodate(folio)) |
050dcb5c HD |
3179 | shmem_falloc.nr_falloced += index - shmem_falloc.next; |
3180 | shmem_falloc.next = index; | |
1aac1400 | 3181 | |
e2d12e22 | 3182 | /* |
b0802b22 | 3183 | * If !uptodate, leave it that way so that freeable folios |
1635f6a7 | 3184 | * can be recognized if we need to rollback on error later. |
b0802b22 MWO |
3185 | * But mark it dirty so that memory pressure will swap rather |
3186 | * than free the folios we are allocating (and SGP_CACHE folios | |
e2d12e22 HD |
3187 | * might still be clean: we now need to mark those dirty too). |
3188 | */ | |
b0802b22 MWO |
3189 | folio_mark_dirty(folio); |
3190 | folio_unlock(folio); | |
3191 | folio_put(folio); | |
e2d12e22 HD |
3192 | cond_resched(); |
3193 | } | |
3194 | ||
3195 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
3196 | i_size_write(inode, offset + len); | |
1aac1400 HD |
3197 | undone: |
3198 | spin_lock(&inode->i_lock); | |
3199 | inode->i_private = NULL; | |
3200 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 3201 | out: |
15f242bb HD |
3202 | if (!error) |
3203 | file_modified(file); | |
5955102c | 3204 | inode_unlock(inode); |
83e4fa9c HD |
3205 | return error; |
3206 | } | |
3207 | ||
726c3342 | 3208 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 3209 | { |
726c3342 | 3210 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
3211 | |
3212 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 3213 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 3214 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 3215 | if (sbinfo->max_blocks) { |
1da177e4 | 3216 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
3217 | buf->f_bavail = |
3218 | buf->f_bfree = sbinfo->max_blocks - | |
3219 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
3220 | } |
3221 | if (sbinfo->max_inodes) { | |
1da177e4 | 3222 | buf->f_files = sbinfo->max_inodes; |
e07c469e | 3223 | buf->f_ffree = sbinfo->free_ispace / BOGO_INODE_SIZE; |
1da177e4 LT |
3224 | } |
3225 | /* else leave those fields 0 like simple_statfs */ | |
59cda49e AG |
3226 | |
3227 | buf->f_fsid = uuid_to_fsid(dentry->d_sb->s_uuid.b); | |
3228 | ||
1da177e4 LT |
3229 | return 0; |
3230 | } | |
3231 | ||
3232 | /* | |
3233 | * File creation. Allocate an inode, and we're done.. | |
3234 | */ | |
3235 | static int | |
5ebb29be | 3236 | shmem_mknod(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 3237 | struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 3238 | { |
0b0a0806 | 3239 | struct inode *inode; |
71480663 | 3240 | int error; |
1da177e4 | 3241 | |
7a80e5b8 | 3242 | inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, dev, VM_NORESERVE); |
71480663 CM |
3243 | if (IS_ERR(inode)) |
3244 | return PTR_ERR(inode); | |
37ec43cd | 3245 | |
71480663 CM |
3246 | error = simple_acl_create(dir, inode); |
3247 | if (error) | |
3248 | goto out_iput; | |
9be7d5b0 | 3249 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
71480663 CM |
3250 | shmem_initxattrs, NULL); |
3251 | if (error && error != -EOPNOTSUPP) | |
3252 | goto out_iput; | |
3253 | ||
a2e45955 CL |
3254 | error = simple_offset_add(shmem_get_offset_ctx(dir), dentry); |
3255 | if (error) | |
3256 | goto out_iput; | |
3257 | ||
71480663 | 3258 | dir->i_size += BOGO_DIRENT_SIZE; |
cf2766bb | 3259 | inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir)); |
71480663 CM |
3260 | inode_inc_iversion(dir); |
3261 | d_instantiate(dentry, inode); | |
3262 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 | 3263 | return error; |
71480663 | 3264 | |
feda821e CH |
3265 | out_iput: |
3266 | iput(inode); | |
3267 | return error; | |
1da177e4 LT |
3268 | } |
3269 | ||
60545d0d | 3270 | static int |
011e2b71 | 3271 | shmem_tmpfile(struct mnt_idmap *idmap, struct inode *dir, |
863f144f | 3272 | struct file *file, umode_t mode) |
60545d0d AV |
3273 | { |
3274 | struct inode *inode; | |
71480663 | 3275 | int error; |
60545d0d | 3276 | |
7a80e5b8 | 3277 | inode = shmem_get_inode(idmap, dir->i_sb, dir, mode, 0, VM_NORESERVE); |
71480663 CM |
3278 | if (IS_ERR(inode)) { |
3279 | error = PTR_ERR(inode); | |
3280 | goto err_out; | |
60545d0d | 3281 | } |
9be7d5b0 | 3282 | error = security_inode_init_security(inode, dir, NULL, |
71480663 CM |
3283 | shmem_initxattrs, NULL); |
3284 | if (error && error != -EOPNOTSUPP) | |
3285 | goto out_iput; | |
3286 | error = simple_acl_create(dir, inode); | |
3287 | if (error) | |
3288 | goto out_iput; | |
3289 | d_tmpfile(file, inode); | |
3290 | ||
3291 | err_out: | |
863f144f | 3292 | return finish_open_simple(file, error); |
feda821e CH |
3293 | out_iput: |
3294 | iput(inode); | |
3295 | return error; | |
60545d0d AV |
3296 | } |
3297 | ||
c54bd91e | 3298 | static int shmem_mkdir(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 3299 | struct dentry *dentry, umode_t mode) |
1da177e4 LT |
3300 | { |
3301 | int error; | |
3302 | ||
7a80e5b8 GS |
3303 | error = shmem_mknod(idmap, dir, dentry, mode | S_IFDIR, 0); |
3304 | if (error) | |
1da177e4 | 3305 | return error; |
d8c76e6f | 3306 | inc_nlink(dir); |
1da177e4 LT |
3307 | return 0; |
3308 | } | |
3309 | ||
6c960e68 | 3310 | static int shmem_create(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 3311 | struct dentry *dentry, umode_t mode, bool excl) |
1da177e4 | 3312 | { |
7a80e5b8 | 3313 | return shmem_mknod(idmap, dir, dentry, mode | S_IFREG, 0); |
1da177e4 LT |
3314 | } |
3315 | ||
3316 | /* | |
3317 | * Link a file.. | |
3318 | */ | |
9be7d5b0 HD |
3319 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, |
3320 | struct dentry *dentry) | |
1da177e4 | 3321 | { |
75c3cfa8 | 3322 | struct inode *inode = d_inode(old_dentry); |
29b00e60 | 3323 | int ret = 0; |
1da177e4 LT |
3324 | |
3325 | /* | |
3326 | * No ordinary (disk based) filesystem counts links as inodes; | |
3327 | * but each new link needs a new dentry, pinning lowmem, and | |
3328 | * tmpfs dentries cannot be pruned until they are unlinked. | |
1062af92 DW |
3329 | * But if an O_TMPFILE file is linked into the tmpfs, the |
3330 | * first link must skip that, to get the accounting right. | |
1da177e4 | 3331 | */ |
1062af92 | 3332 | if (inode->i_nlink) { |
e809d5f0 | 3333 | ret = shmem_reserve_inode(inode->i_sb, NULL); |
1062af92 DW |
3334 | if (ret) |
3335 | goto out; | |
3336 | } | |
1da177e4 | 3337 | |
a2e45955 CL |
3338 | ret = simple_offset_add(shmem_get_offset_ctx(dir), dentry); |
3339 | if (ret) { | |
3340 | if (inode->i_nlink) | |
2daf18a7 | 3341 | shmem_free_inode(inode->i_sb, 0); |
a2e45955 CL |
3342 | goto out; |
3343 | } | |
3344 | ||
1da177e4 | 3345 | dir->i_size += BOGO_DIRENT_SIZE; |
cf2766bb JL |
3346 | inode_set_mtime_to_ts(dir, |
3347 | inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode))); | |
36f05cab | 3348 | inode_inc_iversion(dir); |
d8c76e6f | 3349 | inc_nlink(inode); |
7de9c6ee | 3350 | ihold(inode); /* New dentry reference */ |
9be7d5b0 | 3351 | dget(dentry); /* Extra pinning count for the created dentry */ |
1da177e4 | 3352 | d_instantiate(dentry, inode); |
5b04c689 PE |
3353 | out: |
3354 | return ret; | |
1da177e4 LT |
3355 | } |
3356 | ||
3357 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
3358 | { | |
75c3cfa8 | 3359 | struct inode *inode = d_inode(dentry); |
1da177e4 | 3360 | |
5b04c689 | 3361 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
2daf18a7 | 3362 | shmem_free_inode(inode->i_sb, 0); |
1da177e4 | 3363 | |
a2e45955 | 3364 | simple_offset_remove(shmem_get_offset_ctx(dir), dentry); |
1da177e4 LT |
3365 | |
3366 | dir->i_size -= BOGO_DIRENT_SIZE; | |
cf2766bb JL |
3367 | inode_set_mtime_to_ts(dir, |
3368 | inode_set_ctime_to_ts(dir, inode_set_ctime_current(inode))); | |
36f05cab | 3369 | inode_inc_iversion(dir); |
9a53c3a7 | 3370 | drop_nlink(inode); |
9be7d5b0 | 3371 | dput(dentry); /* Undo the count from "create" - does all the work */ |
1da177e4 LT |
3372 | return 0; |
3373 | } | |
3374 | ||
3375 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
3376 | { | |
3377 | if (!simple_empty(dentry)) | |
3378 | return -ENOTEMPTY; | |
3379 | ||
75c3cfa8 | 3380 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 3381 | drop_nlink(dir); |
1da177e4 LT |
3382 | return shmem_unlink(dir, dentry); |
3383 | } | |
3384 | ||
e18275ae | 3385 | static int shmem_whiteout(struct mnt_idmap *idmap, |
549c7297 | 3386 | struct inode *old_dir, struct dentry *old_dentry) |
46fdb794 MS |
3387 | { |
3388 | struct dentry *whiteout; | |
3389 | int error; | |
3390 | ||
3391 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
3392 | if (!whiteout) | |
3393 | return -ENOMEM; | |
3394 | ||
7a80e5b8 | 3395 | error = shmem_mknod(idmap, old_dir, whiteout, |
46fdb794 MS |
3396 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); |
3397 | dput(whiteout); | |
3398 | if (error) | |
3399 | return error; | |
3400 | ||
3401 | /* | |
3402 | * Cheat and hash the whiteout while the old dentry is still in | |
3403 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
3404 | * | |
3405 | * d_lookup() will consistently find one of them at this point, | |
3406 | * not sure which one, but that isn't even important. | |
3407 | */ | |
3408 | d_rehash(whiteout); | |
3409 | return 0; | |
3410 | } | |
3411 | ||
1da177e4 LT |
3412 | /* |
3413 | * The VFS layer already does all the dentry stuff for rename, | |
3414 | * we just have to decrement the usage count for the target if | |
3415 | * it exists so that the VFS layer correctly free's it when it | |
3416 | * gets overwritten. | |
3417 | */ | |
e18275ae | 3418 | static int shmem_rename2(struct mnt_idmap *idmap, |
549c7297 CB |
3419 | struct inode *old_dir, struct dentry *old_dentry, |
3420 | struct inode *new_dir, struct dentry *new_dentry, | |
3421 | unsigned int flags) | |
1da177e4 | 3422 | { |
75c3cfa8 | 3423 | struct inode *inode = d_inode(old_dentry); |
1da177e4 | 3424 | int they_are_dirs = S_ISDIR(inode->i_mode); |
a2e45955 | 3425 | int error; |
1da177e4 | 3426 | |
46fdb794 | 3427 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
3428 | return -EINVAL; |
3429 | ||
37456771 | 3430 | if (flags & RENAME_EXCHANGE) |
a2e45955 CL |
3431 | return simple_offset_rename_exchange(old_dir, old_dentry, |
3432 | new_dir, new_dentry); | |
37456771 | 3433 | |
1da177e4 LT |
3434 | if (!simple_empty(new_dentry)) |
3435 | return -ENOTEMPTY; | |
3436 | ||
46fdb794 | 3437 | if (flags & RENAME_WHITEOUT) { |
7a80e5b8 | 3438 | error = shmem_whiteout(idmap, old_dir, old_dentry); |
46fdb794 MS |
3439 | if (error) |
3440 | return error; | |
3441 | } | |
3442 | ||
a2e45955 CL |
3443 | simple_offset_remove(shmem_get_offset_ctx(old_dir), old_dentry); |
3444 | error = simple_offset_add(shmem_get_offset_ctx(new_dir), old_dentry); | |
3445 | if (error) | |
3446 | return error; | |
3447 | ||
75c3cfa8 | 3448 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 3449 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 3450 | if (they_are_dirs) { |
75c3cfa8 | 3451 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 3452 | drop_nlink(old_dir); |
b928095b | 3453 | } |
1da177e4 | 3454 | } else if (they_are_dirs) { |
9a53c3a7 | 3455 | drop_nlink(old_dir); |
d8c76e6f | 3456 | inc_nlink(new_dir); |
1da177e4 LT |
3457 | } |
3458 | ||
3459 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
3460 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
944d0d9d | 3461 | simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry); |
36f05cab JL |
3462 | inode_inc_iversion(old_dir); |
3463 | inode_inc_iversion(new_dir); | |
1da177e4 LT |
3464 | return 0; |
3465 | } | |
3466 | ||
7a77db95 | 3467 | static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir, |
549c7297 | 3468 | struct dentry *dentry, const char *symname) |
1da177e4 LT |
3469 | { |
3470 | int error; | |
3471 | int len; | |
3472 | struct inode *inode; | |
7ad0414b | 3473 | struct folio *folio; |
1da177e4 LT |
3474 | |
3475 | len = strlen(symname) + 1; | |
09cbfeaf | 3476 | if (len > PAGE_SIZE) |
1da177e4 LT |
3477 | return -ENAMETOOLONG; |
3478 | ||
7a80e5b8 | 3479 | inode = shmem_get_inode(idmap, dir->i_sb, dir, S_IFLNK | 0777, 0, |
0825a6f9 | 3480 | VM_NORESERVE); |
71480663 CM |
3481 | if (IS_ERR(inode)) |
3482 | return PTR_ERR(inode); | |
1da177e4 | 3483 | |
9d8f13ba | 3484 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3485 | shmem_initxattrs, NULL); |
23a31d87 CL |
3486 | if (error && error != -EOPNOTSUPP) |
3487 | goto out_iput; | |
570bc1c2 | 3488 | |
a2e45955 CL |
3489 | error = simple_offset_add(shmem_get_offset_ctx(dir), dentry); |
3490 | if (error) | |
3491 | goto out_iput; | |
570bc1c2 | 3492 | |
1da177e4 | 3493 | inode->i_size = len-1; |
69f07ec9 | 3494 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3495 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3496 | if (!inode->i_link) { | |
23a31d87 | 3497 | error = -ENOMEM; |
a2e45955 | 3498 | goto out_remove_offset; |
69f07ec9 HD |
3499 | } |
3500 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3501 | } else { |
e8ecde25 | 3502 | inode_nohighmem(inode); |
7ad0414b | 3503 | error = shmem_get_folio(inode, 0, &folio, SGP_WRITE); |
23a31d87 | 3504 | if (error) |
a2e45955 | 3505 | goto out_remove_offset; |
14fcc23f | 3506 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 3507 | inode->i_op = &shmem_symlink_inode_operations; |
7ad0414b MWO |
3508 | memcpy(folio_address(folio), symname, len); |
3509 | folio_mark_uptodate(folio); | |
3510 | folio_mark_dirty(folio); | |
3511 | folio_unlock(folio); | |
3512 | folio_put(folio); | |
1da177e4 | 3513 | } |
1da177e4 | 3514 | dir->i_size += BOGO_DIRENT_SIZE; |
cf2766bb | 3515 | inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir)); |
36f05cab | 3516 | inode_inc_iversion(dir); |
1da177e4 LT |
3517 | d_instantiate(dentry, inode); |
3518 | dget(dentry); | |
3519 | return 0; | |
a2e45955 CL |
3520 | |
3521 | out_remove_offset: | |
3522 | simple_offset_remove(shmem_get_offset_ctx(dir), dentry); | |
23a31d87 CL |
3523 | out_iput: |
3524 | iput(inode); | |
3525 | return error; | |
1da177e4 LT |
3526 | } |
3527 | ||
fceef393 | 3528 | static void shmem_put_link(void *arg) |
1da177e4 | 3529 | { |
e4b57722 MWO |
3530 | folio_mark_accessed(arg); |
3531 | folio_put(arg); | |
1da177e4 LT |
3532 | } |
3533 | ||
9be7d5b0 | 3534 | static const char *shmem_get_link(struct dentry *dentry, struct inode *inode, |
fceef393 | 3535 | struct delayed_call *done) |
1da177e4 | 3536 | { |
e4b57722 | 3537 | struct folio *folio = NULL; |
6b255391 | 3538 | int error; |
e4b57722 | 3539 | |
6a6c9904 | 3540 | if (!dentry) { |
e4b57722 | 3541 | folio = filemap_get_folio(inode->i_mapping, 0); |
66dabbb6 | 3542 | if (IS_ERR(folio)) |
6a6c9904 | 3543 | return ERR_PTR(-ECHILD); |
7459c149 | 3544 | if (PageHWPoison(folio_page(folio, 0)) || |
e4b57722 MWO |
3545 | !folio_test_uptodate(folio)) { |
3546 | folio_put(folio); | |
6a6c9904 AV |
3547 | return ERR_PTR(-ECHILD); |
3548 | } | |
3549 | } else { | |
e4b57722 | 3550 | error = shmem_get_folio(inode, 0, &folio, SGP_READ); |
6a6c9904 AV |
3551 | if (error) |
3552 | return ERR_PTR(error); | |
e4b57722 | 3553 | if (!folio) |
a7605426 | 3554 | return ERR_PTR(-ECHILD); |
7459c149 | 3555 | if (PageHWPoison(folio_page(folio, 0))) { |
e4b57722 MWO |
3556 | folio_unlock(folio); |
3557 | folio_put(folio); | |
a7605426 YS |
3558 | return ERR_PTR(-ECHILD); |
3559 | } | |
e4b57722 | 3560 | folio_unlock(folio); |
6a6c9904 | 3561 | } |
e4b57722 MWO |
3562 | set_delayed_call(done, shmem_put_link, folio); |
3563 | return folio_address(folio); | |
1da177e4 LT |
3564 | } |
3565 | ||
b09e0fa4 | 3566 | #ifdef CONFIG_TMPFS_XATTR |
e408e695 TT |
3567 | |
3568 | static int shmem_fileattr_get(struct dentry *dentry, struct fileattr *fa) | |
3569 | { | |
3570 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); | |
3571 | ||
3572 | fileattr_fill_flags(fa, info->fsflags & SHMEM_FL_USER_VISIBLE); | |
3573 | ||
3574 | return 0; | |
3575 | } | |
3576 | ||
8782a9ae | 3577 | static int shmem_fileattr_set(struct mnt_idmap *idmap, |
e408e695 TT |
3578 | struct dentry *dentry, struct fileattr *fa) |
3579 | { | |
3580 | struct inode *inode = d_inode(dentry); | |
3581 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3582 | ||
3583 | if (fileattr_has_fsx(fa)) | |
3584 | return -EOPNOTSUPP; | |
cb241339 HD |
3585 | if (fa->flags & ~SHMEM_FL_USER_MODIFIABLE) |
3586 | return -EOPNOTSUPP; | |
e408e695 TT |
3587 | |
3588 | info->fsflags = (info->fsflags & ~SHMEM_FL_USER_MODIFIABLE) | | |
3589 | (fa->flags & SHMEM_FL_USER_MODIFIABLE); | |
3590 | ||
cb241339 | 3591 | shmem_set_inode_flags(inode, info->fsflags); |
65287334 | 3592 | inode_set_ctime_current(inode); |
36f05cab | 3593 | inode_inc_iversion(inode); |
e408e695 TT |
3594 | return 0; |
3595 | } | |
3596 | ||
46711810 | 3597 | /* |
b09e0fa4 EP |
3598 | * Superblocks without xattr inode operations may get some security.* xattr |
3599 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3600 | * like ACLs, we also need to implement the security.* handlers at |
3601 | * filesystem level, though. | |
3602 | */ | |
3603 | ||
6d9d88d0 JS |
3604 | /* |
3605 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3606 | */ | |
3607 | static int shmem_initxattrs(struct inode *inode, | |
9be7d5b0 | 3608 | const struct xattr *xattr_array, void *fs_info) |
6d9d88d0 JS |
3609 | { |
3610 | struct shmem_inode_info *info = SHMEM_I(inode); | |
2daf18a7 | 3611 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
6d9d88d0 | 3612 | const struct xattr *xattr; |
38f38657 | 3613 | struct simple_xattr *new_xattr; |
2daf18a7 | 3614 | size_t ispace = 0; |
6d9d88d0 JS |
3615 | size_t len; |
3616 | ||
2daf18a7 HD |
3617 | if (sbinfo->max_inodes) { |
3618 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
3619 | ispace += simple_xattr_space(xattr->name, | |
3620 | xattr->value_len + XATTR_SECURITY_PREFIX_LEN); | |
3621 | } | |
3622 | if (ispace) { | |
3623 | raw_spin_lock(&sbinfo->stat_lock); | |
3624 | if (sbinfo->free_ispace < ispace) | |
3625 | ispace = 0; | |
3626 | else | |
3627 | sbinfo->free_ispace -= ispace; | |
3628 | raw_spin_unlock(&sbinfo->stat_lock); | |
3629 | if (!ispace) | |
3630 | return -ENOSPC; | |
3631 | } | |
3632 | } | |
3633 | ||
6d9d88d0 | 3634 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { |
38f38657 | 3635 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 | 3636 | if (!new_xattr) |
2daf18a7 | 3637 | break; |
6d9d88d0 JS |
3638 | |
3639 | len = strlen(xattr->name) + 1; | |
3640 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
572a3d1e | 3641 | GFP_KERNEL_ACCOUNT); |
6d9d88d0 | 3642 | if (!new_xattr->name) { |
3bef735a | 3643 | kvfree(new_xattr); |
2daf18a7 | 3644 | break; |
6d9d88d0 JS |
3645 | } |
3646 | ||
3647 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3648 | XATTR_SECURITY_PREFIX_LEN); | |
3649 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3650 | xattr->name, len); | |
3651 | ||
3b4c7bc0 | 3652 | simple_xattr_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3653 | } |
3654 | ||
2daf18a7 HD |
3655 | if (xattr->name != NULL) { |
3656 | if (ispace) { | |
3657 | raw_spin_lock(&sbinfo->stat_lock); | |
3658 | sbinfo->free_ispace += ispace; | |
3659 | raw_spin_unlock(&sbinfo->stat_lock); | |
3660 | } | |
3661 | simple_xattrs_free(&info->xattrs, NULL); | |
3662 | return -ENOMEM; | |
3663 | } | |
3664 | ||
6d9d88d0 JS |
3665 | return 0; |
3666 | } | |
3667 | ||
aa7c5241 | 3668 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3669 | struct dentry *unused, struct inode *inode, |
3670 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3671 | { |
b296821a | 3672 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3673 | |
aa7c5241 | 3674 | name = xattr_full_name(handler, name); |
38f38657 | 3675 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3676 | } |
3677 | ||
aa7c5241 | 3678 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
39f60c1c | 3679 | struct mnt_idmap *idmap, |
59301226 AV |
3680 | struct dentry *unused, struct inode *inode, |
3681 | const char *name, const void *value, | |
3682 | size_t size, int flags) | |
b09e0fa4 | 3683 | { |
59301226 | 3684 | struct shmem_inode_info *info = SHMEM_I(inode); |
2daf18a7 | 3685 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
5de75970 | 3686 | struct simple_xattr *old_xattr; |
2daf18a7 | 3687 | size_t ispace = 0; |
b09e0fa4 | 3688 | |
aa7c5241 | 3689 | name = xattr_full_name(handler, name); |
2daf18a7 HD |
3690 | if (value && sbinfo->max_inodes) { |
3691 | ispace = simple_xattr_space(name, size); | |
3692 | raw_spin_lock(&sbinfo->stat_lock); | |
3693 | if (sbinfo->free_ispace < ispace) | |
3694 | ispace = 0; | |
3695 | else | |
3696 | sbinfo->free_ispace -= ispace; | |
3697 | raw_spin_unlock(&sbinfo->stat_lock); | |
3698 | if (!ispace) | |
3699 | return -ENOSPC; | |
3700 | } | |
3701 | ||
5de75970 HD |
3702 | old_xattr = simple_xattr_set(&info->xattrs, name, value, size, flags); |
3703 | if (!IS_ERR(old_xattr)) { | |
2daf18a7 HD |
3704 | ispace = 0; |
3705 | if (old_xattr && sbinfo->max_inodes) | |
3706 | ispace = simple_xattr_space(old_xattr->name, | |
3707 | old_xattr->size); | |
5de75970 HD |
3708 | simple_xattr_free(old_xattr); |
3709 | old_xattr = NULL; | |
65287334 | 3710 | inode_set_ctime_current(inode); |
36f05cab JL |
3711 | inode_inc_iversion(inode); |
3712 | } | |
2daf18a7 HD |
3713 | if (ispace) { |
3714 | raw_spin_lock(&sbinfo->stat_lock); | |
3715 | sbinfo->free_ispace += ispace; | |
3716 | raw_spin_unlock(&sbinfo->stat_lock); | |
3717 | } | |
5de75970 | 3718 | return PTR_ERR(old_xattr); |
b09e0fa4 EP |
3719 | } |
3720 | ||
aa7c5241 AG |
3721 | static const struct xattr_handler shmem_security_xattr_handler = { |
3722 | .prefix = XATTR_SECURITY_PREFIX, | |
3723 | .get = shmem_xattr_handler_get, | |
3724 | .set = shmem_xattr_handler_set, | |
3725 | }; | |
b09e0fa4 | 3726 | |
aa7c5241 AG |
3727 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3728 | .prefix = XATTR_TRUSTED_PREFIX, | |
3729 | .get = shmem_xattr_handler_get, | |
3730 | .set = shmem_xattr_handler_set, | |
3731 | }; | |
b09e0fa4 | 3732 | |
2daf18a7 HD |
3733 | static const struct xattr_handler shmem_user_xattr_handler = { |
3734 | .prefix = XATTR_USER_PREFIX, | |
3735 | .get = shmem_xattr_handler_get, | |
3736 | .set = shmem_xattr_handler_set, | |
3737 | }; | |
3738 | ||
2f502860 | 3739 | static const struct xattr_handler * const shmem_xattr_handlers[] = { |
aa7c5241 AG |
3740 | &shmem_security_xattr_handler, |
3741 | &shmem_trusted_xattr_handler, | |
2daf18a7 | 3742 | &shmem_user_xattr_handler, |
aa7c5241 AG |
3743 | NULL |
3744 | }; | |
b09e0fa4 EP |
3745 | |
3746 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3747 | { | |
75c3cfa8 | 3748 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3749 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3750 | } |
3751 | #endif /* CONFIG_TMPFS_XATTR */ | |
3752 | ||
69f07ec9 | 3753 | static const struct inode_operations shmem_short_symlink_operations = { |
f7cd16a5 | 3754 | .getattr = shmem_getattr, |
e09764cf | 3755 | .setattr = shmem_setattr, |
6b255391 | 3756 | .get_link = simple_get_link, |
b09e0fa4 | 3757 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3758 | .listxattr = shmem_listxattr, |
b09e0fa4 EP |
3759 | #endif |
3760 | }; | |
3761 | ||
3762 | static const struct inode_operations shmem_symlink_inode_operations = { | |
f7cd16a5 | 3763 | .getattr = shmem_getattr, |
e09764cf | 3764 | .setattr = shmem_setattr, |
6b255391 | 3765 | .get_link = shmem_get_link, |
b09e0fa4 | 3766 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3767 | .listxattr = shmem_listxattr, |
39f0247d | 3768 | #endif |
b09e0fa4 | 3769 | }; |
39f0247d | 3770 | |
91828a40 DG |
3771 | static struct dentry *shmem_get_parent(struct dentry *child) |
3772 | { | |
3773 | return ERR_PTR(-ESTALE); | |
3774 | } | |
3775 | ||
3776 | static int shmem_match(struct inode *ino, void *vfh) | |
3777 | { | |
3778 | __u32 *fh = vfh; | |
3779 | __u64 inum = fh[2]; | |
3780 | inum = (inum << 32) | fh[1]; | |
3781 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
3782 | } | |
3783 | ||
12ba780d AG |
3784 | /* Find any alias of inode, but prefer a hashed alias */ |
3785 | static struct dentry *shmem_find_alias(struct inode *inode) | |
3786 | { | |
3787 | struct dentry *alias = d_find_alias(inode); | |
3788 | ||
3789 | return alias ?: d_find_any_alias(inode); | |
3790 | } | |
3791 | ||
480b116c CH |
3792 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3793 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3794 | { |
91828a40 | 3795 | struct inode *inode; |
480b116c | 3796 | struct dentry *dentry = NULL; |
35c2a7f4 | 3797 | u64 inum; |
480b116c CH |
3798 | |
3799 | if (fh_len < 3) | |
3800 | return NULL; | |
91828a40 | 3801 | |
35c2a7f4 HD |
3802 | inum = fid->raw[2]; |
3803 | inum = (inum << 32) | fid->raw[1]; | |
3804 | ||
480b116c CH |
3805 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
3806 | shmem_match, fid->raw); | |
91828a40 | 3807 | if (inode) { |
12ba780d | 3808 | dentry = shmem_find_alias(inode); |
91828a40 DG |
3809 | iput(inode); |
3810 | } | |
3811 | ||
480b116c | 3812 | return dentry; |
91828a40 DG |
3813 | } |
3814 | ||
b0b0382b AV |
3815 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3816 | struct inode *parent) | |
91828a40 | 3817 | { |
5fe0c237 AK |
3818 | if (*len < 3) { |
3819 | *len = 3; | |
94e07a75 | 3820 | return FILEID_INVALID; |
5fe0c237 | 3821 | } |
91828a40 | 3822 | |
1d3382cb | 3823 | if (inode_unhashed(inode)) { |
91828a40 DG |
3824 | /* Unfortunately insert_inode_hash is not idempotent, |
3825 | * so as we hash inodes here rather than at creation | |
3826 | * time, we need a lock to ensure we only try | |
3827 | * to do it once | |
3828 | */ | |
3829 | static DEFINE_SPINLOCK(lock); | |
3830 | spin_lock(&lock); | |
1d3382cb | 3831 | if (inode_unhashed(inode)) |
91828a40 DG |
3832 | __insert_inode_hash(inode, |
3833 | inode->i_ino + inode->i_generation); | |
3834 | spin_unlock(&lock); | |
3835 | } | |
3836 | ||
3837 | fh[0] = inode->i_generation; | |
3838 | fh[1] = inode->i_ino; | |
3839 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
3840 | ||
3841 | *len = 3; | |
3842 | return 1; | |
3843 | } | |
3844 | ||
39655164 | 3845 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3846 | .get_parent = shmem_get_parent, |
91828a40 | 3847 | .encode_fh = shmem_encode_fh, |
480b116c | 3848 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3849 | }; |
3850 | ||
626c3920 AV |
3851 | enum shmem_param { |
3852 | Opt_gid, | |
3853 | Opt_huge, | |
3854 | Opt_mode, | |
3855 | Opt_mpol, | |
3856 | Opt_nr_blocks, | |
3857 | Opt_nr_inodes, | |
3858 | Opt_size, | |
3859 | Opt_uid, | |
ea3271f7 CD |
3860 | Opt_inode32, |
3861 | Opt_inode64, | |
2c6efe9c | 3862 | Opt_noswap, |
e09764cf CM |
3863 | Opt_quota, |
3864 | Opt_usrquota, | |
3865 | Opt_grpquota, | |
de4c0e7c LC |
3866 | Opt_usrquota_block_hardlimit, |
3867 | Opt_usrquota_inode_hardlimit, | |
3868 | Opt_grpquota_block_hardlimit, | |
3869 | Opt_grpquota_inode_hardlimit, | |
626c3920 AV |
3870 | }; |
3871 | ||
5eede625 | 3872 | static const struct constant_table shmem_param_enums_huge[] = { |
2710c957 AV |
3873 | {"never", SHMEM_HUGE_NEVER }, |
3874 | {"always", SHMEM_HUGE_ALWAYS }, | |
3875 | {"within_size", SHMEM_HUGE_WITHIN_SIZE }, | |
3876 | {"advise", SHMEM_HUGE_ADVISE }, | |
2710c957 AV |
3877 | {} |
3878 | }; | |
3879 | ||
d7167b14 | 3880 | const struct fs_parameter_spec shmem_fs_parameters[] = { |
626c3920 | 3881 | fsparam_u32 ("gid", Opt_gid), |
2710c957 | 3882 | fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge), |
626c3920 AV |
3883 | fsparam_u32oct("mode", Opt_mode), |
3884 | fsparam_string("mpol", Opt_mpol), | |
3885 | fsparam_string("nr_blocks", Opt_nr_blocks), | |
3886 | fsparam_string("nr_inodes", Opt_nr_inodes), | |
3887 | fsparam_string("size", Opt_size), | |
3888 | fsparam_u32 ("uid", Opt_uid), | |
ea3271f7 CD |
3889 | fsparam_flag ("inode32", Opt_inode32), |
3890 | fsparam_flag ("inode64", Opt_inode64), | |
2c6efe9c | 3891 | fsparam_flag ("noswap", Opt_noswap), |
e09764cf CM |
3892 | #ifdef CONFIG_TMPFS_QUOTA |
3893 | fsparam_flag ("quota", Opt_quota), | |
3894 | fsparam_flag ("usrquota", Opt_usrquota), | |
3895 | fsparam_flag ("grpquota", Opt_grpquota), | |
de4c0e7c LC |
3896 | fsparam_string("usrquota_block_hardlimit", Opt_usrquota_block_hardlimit), |
3897 | fsparam_string("usrquota_inode_hardlimit", Opt_usrquota_inode_hardlimit), | |
3898 | fsparam_string("grpquota_block_hardlimit", Opt_grpquota_block_hardlimit), | |
3899 | fsparam_string("grpquota_inode_hardlimit", Opt_grpquota_inode_hardlimit), | |
e09764cf | 3900 | #endif |
626c3920 AV |
3901 | {} |
3902 | }; | |
3903 | ||
f3235626 | 3904 | static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param) |
1da177e4 | 3905 | { |
f3235626 | 3906 | struct shmem_options *ctx = fc->fs_private; |
626c3920 AV |
3907 | struct fs_parse_result result; |
3908 | unsigned long long size; | |
e04dc423 | 3909 | char *rest; |
626c3920 | 3910 | int opt; |
0200679f CB |
3911 | kuid_t kuid; |
3912 | kgid_t kgid; | |
626c3920 | 3913 | |
d7167b14 | 3914 | opt = fs_parse(fc, shmem_fs_parameters, param, &result); |
f3235626 | 3915 | if (opt < 0) |
626c3920 | 3916 | return opt; |
1da177e4 | 3917 | |
626c3920 AV |
3918 | switch (opt) { |
3919 | case Opt_size: | |
3920 | size = memparse(param->string, &rest); | |
e04dc423 AV |
3921 | if (*rest == '%') { |
3922 | size <<= PAGE_SHIFT; | |
3923 | size *= totalram_pages(); | |
3924 | do_div(size, 100); | |
3925 | rest++; | |
3926 | } | |
3927 | if (*rest) | |
626c3920 | 3928 | goto bad_value; |
e04dc423 AV |
3929 | ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE); |
3930 | ctx->seen |= SHMEM_SEEN_BLOCKS; | |
626c3920 AV |
3931 | break; |
3932 | case Opt_nr_blocks: | |
3933 | ctx->blocks = memparse(param->string, &rest); | |
e07c469e | 3934 | if (*rest || ctx->blocks > LONG_MAX) |
626c3920 | 3935 | goto bad_value; |
e04dc423 | 3936 | ctx->seen |= SHMEM_SEEN_BLOCKS; |
626c3920 AV |
3937 | break; |
3938 | case Opt_nr_inodes: | |
3939 | ctx->inodes = memparse(param->string, &rest); | |
e07c469e | 3940 | if (*rest || ctx->inodes > ULONG_MAX / BOGO_INODE_SIZE) |
626c3920 | 3941 | goto bad_value; |
e04dc423 | 3942 | ctx->seen |= SHMEM_SEEN_INODES; |
626c3920 AV |
3943 | break; |
3944 | case Opt_mode: | |
3945 | ctx->mode = result.uint_32 & 07777; | |
3946 | break; | |
3947 | case Opt_uid: | |
0200679f CB |
3948 | kuid = make_kuid(current_user_ns(), result.uint_32); |
3949 | if (!uid_valid(kuid)) | |
626c3920 | 3950 | goto bad_value; |
0200679f CB |
3951 | |
3952 | /* | |
3953 | * The requested uid must be representable in the | |
3954 | * filesystem's idmapping. | |
3955 | */ | |
3956 | if (!kuid_has_mapping(fc->user_ns, kuid)) | |
626c3920 | 3957 | goto bad_value; |
0200679f CB |
3958 | |
3959 | ctx->uid = kuid; | |
626c3920 AV |
3960 | break; |
3961 | case Opt_gid: | |
0200679f CB |
3962 | kgid = make_kgid(current_user_ns(), result.uint_32); |
3963 | if (!gid_valid(kgid)) | |
626c3920 | 3964 | goto bad_value; |
0200679f CB |
3965 | |
3966 | /* | |
3967 | * The requested gid must be representable in the | |
3968 | * filesystem's idmapping. | |
3969 | */ | |
3970 | if (!kgid_has_mapping(fc->user_ns, kgid)) | |
626c3920 | 3971 | goto bad_value; |
0200679f CB |
3972 | |
3973 | ctx->gid = kgid; | |
626c3920 AV |
3974 | break; |
3975 | case Opt_huge: | |
3976 | ctx->huge = result.uint_32; | |
3977 | if (ctx->huge != SHMEM_HUGE_NEVER && | |
396bcc52 | 3978 | !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && |
626c3920 AV |
3979 | has_transparent_hugepage())) |
3980 | goto unsupported_parameter; | |
e04dc423 | 3981 | ctx->seen |= SHMEM_SEEN_HUGE; |
626c3920 AV |
3982 | break; |
3983 | case Opt_mpol: | |
3984 | if (IS_ENABLED(CONFIG_NUMA)) { | |
3985 | mpol_put(ctx->mpol); | |
3986 | ctx->mpol = NULL; | |
3987 | if (mpol_parse_str(param->string, &ctx->mpol)) | |
3988 | goto bad_value; | |
3989 | break; | |
3990 | } | |
3991 | goto unsupported_parameter; | |
ea3271f7 CD |
3992 | case Opt_inode32: |
3993 | ctx->full_inums = false; | |
3994 | ctx->seen |= SHMEM_SEEN_INUMS; | |
3995 | break; | |
3996 | case Opt_inode64: | |
3997 | if (sizeof(ino_t) < 8) { | |
3998 | return invalfc(fc, | |
3999 | "Cannot use inode64 with <64bit inums in kernel\n"); | |
4000 | } | |
4001 | ctx->full_inums = true; | |
4002 | ctx->seen |= SHMEM_SEEN_INUMS; | |
4003 | break; | |
2c6efe9c | 4004 | case Opt_noswap: |
01106e14 CB |
4005 | if ((fc->user_ns != &init_user_ns) || !capable(CAP_SYS_ADMIN)) { |
4006 | return invalfc(fc, | |
4007 | "Turning off swap in unprivileged tmpfs mounts unsupported"); | |
4008 | } | |
2c6efe9c LC |
4009 | ctx->noswap = true; |
4010 | ctx->seen |= SHMEM_SEEN_NOSWAP; | |
4011 | break; | |
e09764cf CM |
4012 | case Opt_quota: |
4013 | if (fc->user_ns != &init_user_ns) | |
4014 | return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported"); | |
4015 | ctx->seen |= SHMEM_SEEN_QUOTA; | |
4016 | ctx->quota_types |= (QTYPE_MASK_USR | QTYPE_MASK_GRP); | |
4017 | break; | |
4018 | case Opt_usrquota: | |
4019 | if (fc->user_ns != &init_user_ns) | |
4020 | return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported"); | |
4021 | ctx->seen |= SHMEM_SEEN_QUOTA; | |
4022 | ctx->quota_types |= QTYPE_MASK_USR; | |
4023 | break; | |
4024 | case Opt_grpquota: | |
4025 | if (fc->user_ns != &init_user_ns) | |
4026 | return invalfc(fc, "Quotas in unprivileged tmpfs mounts are unsupported"); | |
4027 | ctx->seen |= SHMEM_SEEN_QUOTA; | |
4028 | ctx->quota_types |= QTYPE_MASK_GRP; | |
4029 | break; | |
de4c0e7c LC |
4030 | case Opt_usrquota_block_hardlimit: |
4031 | size = memparse(param->string, &rest); | |
4032 | if (*rest || !size) | |
4033 | goto bad_value; | |
4034 | if (size > SHMEM_QUOTA_MAX_SPC_LIMIT) | |
4035 | return invalfc(fc, | |
4036 | "User quota block hardlimit too large."); | |
4037 | ctx->qlimits.usrquota_bhardlimit = size; | |
4038 | break; | |
4039 | case Opt_grpquota_block_hardlimit: | |
4040 | size = memparse(param->string, &rest); | |
4041 | if (*rest || !size) | |
4042 | goto bad_value; | |
4043 | if (size > SHMEM_QUOTA_MAX_SPC_LIMIT) | |
4044 | return invalfc(fc, | |
4045 | "Group quota block hardlimit too large."); | |
4046 | ctx->qlimits.grpquota_bhardlimit = size; | |
4047 | break; | |
4048 | case Opt_usrquota_inode_hardlimit: | |
4049 | size = memparse(param->string, &rest); | |
4050 | if (*rest || !size) | |
4051 | goto bad_value; | |
4052 | if (size > SHMEM_QUOTA_MAX_INO_LIMIT) | |
4053 | return invalfc(fc, | |
4054 | "User quota inode hardlimit too large."); | |
4055 | ctx->qlimits.usrquota_ihardlimit = size; | |
4056 | break; | |
4057 | case Opt_grpquota_inode_hardlimit: | |
4058 | size = memparse(param->string, &rest); | |
4059 | if (*rest || !size) | |
4060 | goto bad_value; | |
4061 | if (size > SHMEM_QUOTA_MAX_INO_LIMIT) | |
4062 | return invalfc(fc, | |
4063 | "Group quota inode hardlimit too large."); | |
4064 | ctx->qlimits.grpquota_ihardlimit = size; | |
4065 | break; | |
e04dc423 AV |
4066 | } |
4067 | return 0; | |
4068 | ||
626c3920 | 4069 | unsupported_parameter: |
f35aa2bc | 4070 | return invalfc(fc, "Unsupported parameter '%s'", param->key); |
626c3920 | 4071 | bad_value: |
f35aa2bc | 4072 | return invalfc(fc, "Bad value for '%s'", param->key); |
e04dc423 AV |
4073 | } |
4074 | ||
f3235626 | 4075 | static int shmem_parse_options(struct fs_context *fc, void *data) |
e04dc423 | 4076 | { |
f3235626 DH |
4077 | char *options = data; |
4078 | ||
33f37c64 AV |
4079 | if (options) { |
4080 | int err = security_sb_eat_lsm_opts(options, &fc->security); | |
4081 | if (err) | |
4082 | return err; | |
4083 | } | |
4084 | ||
b00dc3ad | 4085 | while (options != NULL) { |
626c3920 | 4086 | char *this_char = options; |
b00dc3ad HD |
4087 | for (;;) { |
4088 | /* | |
4089 | * NUL-terminate this option: unfortunately, | |
4090 | * mount options form a comma-separated list, | |
4091 | * but mpol's nodelist may also contain commas. | |
4092 | */ | |
4093 | options = strchr(options, ','); | |
4094 | if (options == NULL) | |
4095 | break; | |
4096 | options++; | |
4097 | if (!isdigit(*options)) { | |
4098 | options[-1] = '\0'; | |
4099 | break; | |
4100 | } | |
4101 | } | |
626c3920 | 4102 | if (*this_char) { |
68d68ff6 | 4103 | char *value = strchr(this_char, '='); |
f3235626 | 4104 | size_t len = 0; |
626c3920 AV |
4105 | int err; |
4106 | ||
4107 | if (value) { | |
4108 | *value++ = '\0'; | |
f3235626 | 4109 | len = strlen(value); |
626c3920 | 4110 | } |
f3235626 DH |
4111 | err = vfs_parse_fs_string(fc, this_char, value, len); |
4112 | if (err < 0) | |
4113 | return err; | |
1da177e4 | 4114 | } |
1da177e4 LT |
4115 | } |
4116 | return 0; | |
1da177e4 LT |
4117 | } |
4118 | ||
f3235626 DH |
4119 | /* |
4120 | * Reconfigure a shmem filesystem. | |
f3235626 DH |
4121 | */ |
4122 | static int shmem_reconfigure(struct fs_context *fc) | |
1da177e4 | 4123 | { |
f3235626 DH |
4124 | struct shmem_options *ctx = fc->fs_private; |
4125 | struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb); | |
e07c469e | 4126 | unsigned long used_isp; |
bf11b9a8 | 4127 | struct mempolicy *mpol = NULL; |
f3235626 | 4128 | const char *err; |
1da177e4 | 4129 | |
bf11b9a8 | 4130 | raw_spin_lock(&sbinfo->stat_lock); |
e07c469e | 4131 | used_isp = sbinfo->max_inodes * BOGO_INODE_SIZE - sbinfo->free_ispace; |
0c98c8e1 | 4132 | |
f3235626 DH |
4133 | if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) { |
4134 | if (!sbinfo->max_blocks) { | |
4135 | err = "Cannot retroactively limit size"; | |
0b5071dd | 4136 | goto out; |
f3235626 | 4137 | } |
0b5071dd | 4138 | if (percpu_counter_compare(&sbinfo->used_blocks, |
f3235626 DH |
4139 | ctx->blocks) > 0) { |
4140 | err = "Too small a size for current use"; | |
0b5071dd | 4141 | goto out; |
f3235626 | 4142 | } |
0b5071dd | 4143 | } |
f3235626 DH |
4144 | if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) { |
4145 | if (!sbinfo->max_inodes) { | |
4146 | err = "Cannot retroactively limit inodes"; | |
0b5071dd | 4147 | goto out; |
f3235626 | 4148 | } |
e07c469e | 4149 | if (ctx->inodes * BOGO_INODE_SIZE < used_isp) { |
f3235626 | 4150 | err = "Too few inodes for current use"; |
0b5071dd | 4151 | goto out; |
f3235626 | 4152 | } |
0b5071dd | 4153 | } |
0edd73b3 | 4154 | |
ea3271f7 CD |
4155 | if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums && |
4156 | sbinfo->next_ino > UINT_MAX) { | |
4157 | err = "Current inum too high to switch to 32-bit inums"; | |
4158 | goto out; | |
4159 | } | |
2c6efe9c LC |
4160 | if ((ctx->seen & SHMEM_SEEN_NOSWAP) && ctx->noswap && !sbinfo->noswap) { |
4161 | err = "Cannot disable swap on remount"; | |
4162 | goto out; | |
4163 | } | |
4164 | if (!(ctx->seen & SHMEM_SEEN_NOSWAP) && !ctx->noswap && sbinfo->noswap) { | |
4165 | err = "Cannot enable swap on remount if it was disabled on first mount"; | |
4166 | goto out; | |
4167 | } | |
ea3271f7 | 4168 | |
e09764cf CM |
4169 | if (ctx->seen & SHMEM_SEEN_QUOTA && |
4170 | !sb_any_quota_loaded(fc->root->d_sb)) { | |
4171 | err = "Cannot enable quota on remount"; | |
4172 | goto out; | |
4173 | } | |
4174 | ||
de4c0e7c LC |
4175 | #ifdef CONFIG_TMPFS_QUOTA |
4176 | #define CHANGED_LIMIT(name) \ | |
4177 | (ctx->qlimits.name## hardlimit && \ | |
4178 | (ctx->qlimits.name## hardlimit != sbinfo->qlimits.name## hardlimit)) | |
4179 | ||
4180 | if (CHANGED_LIMIT(usrquota_b) || CHANGED_LIMIT(usrquota_i) || | |
4181 | CHANGED_LIMIT(grpquota_b) || CHANGED_LIMIT(grpquota_i)) { | |
4182 | err = "Cannot change global quota limit on remount"; | |
4183 | goto out; | |
4184 | } | |
4185 | #endif /* CONFIG_TMPFS_QUOTA */ | |
4186 | ||
f3235626 DH |
4187 | if (ctx->seen & SHMEM_SEEN_HUGE) |
4188 | sbinfo->huge = ctx->huge; | |
ea3271f7 CD |
4189 | if (ctx->seen & SHMEM_SEEN_INUMS) |
4190 | sbinfo->full_inums = ctx->full_inums; | |
f3235626 DH |
4191 | if (ctx->seen & SHMEM_SEEN_BLOCKS) |
4192 | sbinfo->max_blocks = ctx->blocks; | |
4193 | if (ctx->seen & SHMEM_SEEN_INODES) { | |
4194 | sbinfo->max_inodes = ctx->inodes; | |
e07c469e | 4195 | sbinfo->free_ispace = ctx->inodes * BOGO_INODE_SIZE - used_isp; |
0b5071dd | 4196 | } |
71fe804b | 4197 | |
5f00110f GT |
4198 | /* |
4199 | * Preserve previous mempolicy unless mpol remount option was specified. | |
4200 | */ | |
f3235626 | 4201 | if (ctx->mpol) { |
bf11b9a8 | 4202 | mpol = sbinfo->mpol; |
f3235626 DH |
4203 | sbinfo->mpol = ctx->mpol; /* transfers initial ref */ |
4204 | ctx->mpol = NULL; | |
5f00110f | 4205 | } |
2c6efe9c LC |
4206 | |
4207 | if (ctx->noswap) | |
4208 | sbinfo->noswap = true; | |
4209 | ||
bf11b9a8 SAS |
4210 | raw_spin_unlock(&sbinfo->stat_lock); |
4211 | mpol_put(mpol); | |
f3235626 | 4212 | return 0; |
0edd73b3 | 4213 | out: |
bf11b9a8 | 4214 | raw_spin_unlock(&sbinfo->stat_lock); |
f35aa2bc | 4215 | return invalfc(fc, "%s", err); |
1da177e4 | 4216 | } |
680d794b | 4217 | |
34c80b1d | 4218 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 4219 | { |
34c80b1d | 4220 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
283ebdee | 4221 | struct mempolicy *mpol; |
680d794b | 4222 | |
4223 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
b91742d8 | 4224 | seq_printf(seq, ",size=%luk", K(sbinfo->max_blocks)); |
680d794b | 4225 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
4226 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
0825a6f9 | 4227 | if (sbinfo->mode != (0777 | S_ISVTX)) |
09208d15 | 4228 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
4229 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
4230 | seq_printf(seq, ",uid=%u", | |
4231 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
4232 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
4233 | seq_printf(seq, ",gid=%u", | |
4234 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
ea3271f7 CD |
4235 | |
4236 | /* | |
4237 | * Showing inode{64,32} might be useful even if it's the system default, | |
4238 | * since then people don't have to resort to checking both here and | |
4239 | * /proc/config.gz to confirm 64-bit inums were successfully applied | |
4240 | * (which may not even exist if IKCONFIG_PROC isn't enabled). | |
4241 | * | |
4242 | * We hide it when inode64 isn't the default and we are using 32-bit | |
4243 | * inodes, since that probably just means the feature isn't even under | |
4244 | * consideration. | |
4245 | * | |
4246 | * As such: | |
4247 | * | |
4248 | * +-----------------+-----------------+ | |
4249 | * | TMPFS_INODE64=y | TMPFS_INODE64=n | | |
4250 | * +------------------+-----------------+-----------------+ | |
4251 | * | full_inums=true | show | show | | |
4252 | * | full_inums=false | show | hide | | |
4253 | * +------------------+-----------------+-----------------+ | |
4254 | * | |
4255 | */ | |
4256 | if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums) | |
4257 | seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32)); | |
396bcc52 | 4258 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5a6e75f8 KS |
4259 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
4260 | if (sbinfo->huge) | |
4261 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
4262 | #endif | |
283ebdee TJ |
4263 | mpol = shmem_get_sbmpol(sbinfo); |
4264 | shmem_show_mpol(seq, mpol); | |
4265 | mpol_put(mpol); | |
2c6efe9c LC |
4266 | if (sbinfo->noswap) |
4267 | seq_printf(seq, ",noswap"); | |
b4d3de57 JK |
4268 | #ifdef CONFIG_TMPFS_QUOTA |
4269 | if (sb_has_quota_active(root->d_sb, USRQUOTA)) | |
4270 | seq_printf(seq, ",usrquota"); | |
4271 | if (sb_has_quota_active(root->d_sb, GRPQUOTA)) | |
4272 | seq_printf(seq, ",grpquota"); | |
4273 | if (sbinfo->qlimits.usrquota_bhardlimit) | |
4274 | seq_printf(seq, ",usrquota_block_hardlimit=%lld", | |
4275 | sbinfo->qlimits.usrquota_bhardlimit); | |
4276 | if (sbinfo->qlimits.grpquota_bhardlimit) | |
4277 | seq_printf(seq, ",grpquota_block_hardlimit=%lld", | |
4278 | sbinfo->qlimits.grpquota_bhardlimit); | |
4279 | if (sbinfo->qlimits.usrquota_ihardlimit) | |
4280 | seq_printf(seq, ",usrquota_inode_hardlimit=%lld", | |
4281 | sbinfo->qlimits.usrquota_ihardlimit); | |
4282 | if (sbinfo->qlimits.grpquota_ihardlimit) | |
4283 | seq_printf(seq, ",grpquota_inode_hardlimit=%lld", | |
4284 | sbinfo->qlimits.grpquota_ihardlimit); | |
4285 | #endif | |
680d794b | 4286 | return 0; |
4287 | } | |
9183df25 | 4288 | |
680d794b | 4289 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
4290 | |
4291 | static void shmem_put_super(struct super_block *sb) | |
4292 | { | |
602586a8 HD |
4293 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
4294 | ||
e09764cf CM |
4295 | #ifdef CONFIG_TMPFS_QUOTA |
4296 | shmem_disable_quotas(sb); | |
4297 | #endif | |
e809d5f0 | 4298 | free_percpu(sbinfo->ino_batch); |
602586a8 | 4299 | percpu_counter_destroy(&sbinfo->used_blocks); |
49cd0a5c | 4300 | mpol_put(sbinfo->mpol); |
602586a8 | 4301 | kfree(sbinfo); |
1da177e4 LT |
4302 | sb->s_fs_info = NULL; |
4303 | } | |
4304 | ||
f3235626 | 4305 | static int shmem_fill_super(struct super_block *sb, struct fs_context *fc) |
1da177e4 | 4306 | { |
f3235626 | 4307 | struct shmem_options *ctx = fc->fs_private; |
1da177e4 | 4308 | struct inode *inode; |
0edd73b3 | 4309 | struct shmem_sb_info *sbinfo; |
71480663 | 4310 | int error = -ENOMEM; |
680d794b | 4311 | |
4312 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 4313 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 4314 | L1_CACHE_BYTES), GFP_KERNEL); |
4315 | if (!sbinfo) | |
71480663 | 4316 | return error; |
680d794b | 4317 | |
680d794b | 4318 | sb->s_fs_info = sbinfo; |
1da177e4 | 4319 | |
0edd73b3 | 4320 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
4321 | /* |
4322 | * Per default we only allow half of the physical ram per | |
4323 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
4324 | * but the internal instance is left unlimited. | |
4325 | */ | |
1751e8a6 | 4326 | if (!(sb->s_flags & SB_KERNMOUNT)) { |
f3235626 DH |
4327 | if (!(ctx->seen & SHMEM_SEEN_BLOCKS)) |
4328 | ctx->blocks = shmem_default_max_blocks(); | |
4329 | if (!(ctx->seen & SHMEM_SEEN_INODES)) | |
4330 | ctx->inodes = shmem_default_max_inodes(); | |
ea3271f7 CD |
4331 | if (!(ctx->seen & SHMEM_SEEN_INUMS)) |
4332 | ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64); | |
2c6efe9c | 4333 | sbinfo->noswap = ctx->noswap; |
ca4e0519 | 4334 | } else { |
1751e8a6 | 4335 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 4336 | } |
91828a40 | 4337 | sb->s_export_op = &shmem_export_ops; |
36f05cab | 4338 | sb->s_flags |= SB_NOSEC | SB_I_VERSION; |
1da177e4 | 4339 | #else |
1751e8a6 | 4340 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 4341 | #endif |
f3235626 | 4342 | sbinfo->max_blocks = ctx->blocks; |
e07c469e HD |
4343 | sbinfo->max_inodes = ctx->inodes; |
4344 | sbinfo->free_ispace = sbinfo->max_inodes * BOGO_INODE_SIZE; | |
e809d5f0 CD |
4345 | if (sb->s_flags & SB_KERNMOUNT) { |
4346 | sbinfo->ino_batch = alloc_percpu(ino_t); | |
4347 | if (!sbinfo->ino_batch) | |
4348 | goto failed; | |
4349 | } | |
f3235626 DH |
4350 | sbinfo->uid = ctx->uid; |
4351 | sbinfo->gid = ctx->gid; | |
ea3271f7 | 4352 | sbinfo->full_inums = ctx->full_inums; |
f3235626 DH |
4353 | sbinfo->mode = ctx->mode; |
4354 | sbinfo->huge = ctx->huge; | |
4355 | sbinfo->mpol = ctx->mpol; | |
4356 | ctx->mpol = NULL; | |
1da177e4 | 4357 | |
bf11b9a8 | 4358 | raw_spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 4359 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 4360 | goto failed; |
779750d2 KS |
4361 | spin_lock_init(&sbinfo->shrinklist_lock); |
4362 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 4363 | |
285b2c4f | 4364 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
4365 | sb->s_blocksize = PAGE_SIZE; |
4366 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
4367 | sb->s_magic = TMPFS_MAGIC; |
4368 | sb->s_op = &shmem_ops; | |
cfd95a9c | 4369 | sb->s_time_gran = 1; |
b09e0fa4 | 4370 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 4371 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
4372 | #endif |
4373 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
1751e8a6 | 4374 | sb->s_flags |= SB_POSIXACL; |
39f0247d | 4375 | #endif |
2b4db796 | 4376 | uuid_gen(&sb->s_uuid); |
0edd73b3 | 4377 | |
e09764cf CM |
4378 | #ifdef CONFIG_TMPFS_QUOTA |
4379 | if (ctx->seen & SHMEM_SEEN_QUOTA) { | |
4380 | sb->dq_op = &shmem_quota_operations; | |
4381 | sb->s_qcop = &dquot_quotactl_sysfile_ops; | |
4382 | sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP; | |
4383 | ||
de4c0e7c LC |
4384 | /* Copy the default limits from ctx into sbinfo */ |
4385 | memcpy(&sbinfo->qlimits, &ctx->qlimits, | |
4386 | sizeof(struct shmem_quota_limits)); | |
4387 | ||
e09764cf CM |
4388 | if (shmem_enable_quotas(sb, ctx->quota_types)) |
4389 | goto failed; | |
4390 | } | |
4391 | #endif /* CONFIG_TMPFS_QUOTA */ | |
4392 | ||
9be7d5b0 HD |
4393 | inode = shmem_get_inode(&nop_mnt_idmap, sb, NULL, |
4394 | S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); | |
71480663 CM |
4395 | if (IS_ERR(inode)) { |
4396 | error = PTR_ERR(inode); | |
1da177e4 | 4397 | goto failed; |
71480663 | 4398 | } |
680d794b | 4399 | inode->i_uid = sbinfo->uid; |
4400 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
4401 | sb->s_root = d_make_root(inode); |
4402 | if (!sb->s_root) | |
48fde701 | 4403 | goto failed; |
1da177e4 LT |
4404 | return 0; |
4405 | ||
1da177e4 LT |
4406 | failed: |
4407 | shmem_put_super(sb); | |
71480663 | 4408 | return error; |
1da177e4 LT |
4409 | } |
4410 | ||
f3235626 DH |
4411 | static int shmem_get_tree(struct fs_context *fc) |
4412 | { | |
4413 | return get_tree_nodev(fc, shmem_fill_super); | |
4414 | } | |
4415 | ||
4416 | static void shmem_free_fc(struct fs_context *fc) | |
4417 | { | |
4418 | struct shmem_options *ctx = fc->fs_private; | |
4419 | ||
4420 | if (ctx) { | |
4421 | mpol_put(ctx->mpol); | |
4422 | kfree(ctx); | |
4423 | } | |
4424 | } | |
4425 | ||
4426 | static const struct fs_context_operations shmem_fs_context_ops = { | |
4427 | .free = shmem_free_fc, | |
4428 | .get_tree = shmem_get_tree, | |
4429 | #ifdef CONFIG_TMPFS | |
4430 | .parse_monolithic = shmem_parse_options, | |
4431 | .parse_param = shmem_parse_one, | |
4432 | .reconfigure = shmem_reconfigure, | |
4433 | #endif | |
4434 | }; | |
4435 | ||
68279f9c | 4436 | static struct kmem_cache *shmem_inode_cachep __ro_after_init; |
1da177e4 LT |
4437 | |
4438 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
4439 | { | |
41ffe5d5 | 4440 | struct shmem_inode_info *info; |
fd60b288 | 4441 | info = alloc_inode_sb(sb, shmem_inode_cachep, GFP_KERNEL); |
41ffe5d5 | 4442 | if (!info) |
1da177e4 | 4443 | return NULL; |
41ffe5d5 | 4444 | return &info->vfs_inode; |
1da177e4 LT |
4445 | } |
4446 | ||
74b1da56 | 4447 | static void shmem_free_in_core_inode(struct inode *inode) |
fa0d7e3d | 4448 | { |
84e710da AV |
4449 | if (S_ISLNK(inode->i_mode)) |
4450 | kfree(inode->i_link); | |
fa0d7e3d NP |
4451 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
4452 | } | |
4453 | ||
1da177e4 LT |
4454 | static void shmem_destroy_inode(struct inode *inode) |
4455 | { | |
09208d15 | 4456 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 4457 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
a2e45955 CL |
4458 | if (S_ISDIR(inode->i_mode)) |
4459 | simple_offset_destroy(shmem_get_offset_ctx(inode)); | |
1da177e4 LT |
4460 | } |
4461 | ||
41ffe5d5 | 4462 | static void shmem_init_inode(void *foo) |
1da177e4 | 4463 | { |
41ffe5d5 HD |
4464 | struct shmem_inode_info *info = foo; |
4465 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
4466 | } |
4467 | ||
68279f9c | 4468 | static void __init shmem_init_inodecache(void) |
1da177e4 LT |
4469 | { |
4470 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
4471 | sizeof(struct shmem_inode_info), | |
5d097056 | 4472 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
4473 | } |
4474 | ||
68279f9c | 4475 | static void __init shmem_destroy_inodecache(void) |
1da177e4 | 4476 | { |
1a1d92c1 | 4477 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
4478 | } |
4479 | ||
a7605426 | 4480 | /* Keep the page in page cache instead of truncating it */ |
af7628d6 MWO |
4481 | static int shmem_error_remove_folio(struct address_space *mapping, |
4482 | struct folio *folio) | |
a7605426 YS |
4483 | { |
4484 | return 0; | |
4485 | } | |
4486 | ||
30e6a51d | 4487 | const struct address_space_operations shmem_aops = { |
1da177e4 | 4488 | .writepage = shmem_writepage, |
46de8b97 | 4489 | .dirty_folio = noop_dirty_folio, |
1da177e4 | 4490 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
4491 | .write_begin = shmem_write_begin, |
4492 | .write_end = shmem_write_end, | |
1da177e4 | 4493 | #endif |
1c93923c | 4494 | #ifdef CONFIG_MIGRATION |
54184650 | 4495 | .migrate_folio = migrate_folio, |
1c93923c | 4496 | #endif |
af7628d6 | 4497 | .error_remove_folio = shmem_error_remove_folio, |
1da177e4 | 4498 | }; |
30e6a51d | 4499 | EXPORT_SYMBOL(shmem_aops); |
1da177e4 | 4500 | |
15ad7cdc | 4501 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 4502 | .mmap = shmem_mmap, |
e88e0d36 | 4503 | .open = shmem_file_open, |
c01d5b30 | 4504 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 4505 | #ifdef CONFIG_TMPFS |
220f2ac9 | 4506 | .llseek = shmem_file_llseek, |
2ba5bbed | 4507 | .read_iter = shmem_file_read_iter, |
e88e0d36 | 4508 | .write_iter = shmem_file_write_iter, |
1b061d92 | 4509 | .fsync = noop_fsync, |
bd194b18 | 4510 | .splice_read = shmem_file_splice_read, |
f6cb85d0 | 4511 | .splice_write = iter_file_splice_write, |
83e4fa9c | 4512 | .fallocate = shmem_fallocate, |
1da177e4 LT |
4513 | #endif |
4514 | }; | |
4515 | ||
92e1d5be | 4516 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 4517 | .getattr = shmem_getattr, |
94c1e62d | 4518 | .setattr = shmem_setattr, |
b09e0fa4 | 4519 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 4520 | .listxattr = shmem_listxattr, |
feda821e | 4521 | .set_acl = simple_set_acl, |
e408e695 TT |
4522 | .fileattr_get = shmem_fileattr_get, |
4523 | .fileattr_set = shmem_fileattr_set, | |
b09e0fa4 | 4524 | #endif |
1da177e4 LT |
4525 | }; |
4526 | ||
92e1d5be | 4527 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 | 4528 | #ifdef CONFIG_TMPFS |
f7cd16a5 | 4529 | .getattr = shmem_getattr, |
1da177e4 LT |
4530 | .create = shmem_create, |
4531 | .lookup = simple_lookup, | |
4532 | .link = shmem_link, | |
4533 | .unlink = shmem_unlink, | |
4534 | .symlink = shmem_symlink, | |
4535 | .mkdir = shmem_mkdir, | |
4536 | .rmdir = shmem_rmdir, | |
4537 | .mknod = shmem_mknod, | |
2773bf00 | 4538 | .rename = shmem_rename2, |
60545d0d | 4539 | .tmpfile = shmem_tmpfile, |
a2e45955 | 4540 | .get_offset_ctx = shmem_get_offset_ctx, |
1da177e4 | 4541 | #endif |
b09e0fa4 | 4542 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 4543 | .listxattr = shmem_listxattr, |
e408e695 TT |
4544 | .fileattr_get = shmem_fileattr_get, |
4545 | .fileattr_set = shmem_fileattr_set, | |
b09e0fa4 | 4546 | #endif |
39f0247d | 4547 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 4548 | .setattr = shmem_setattr, |
feda821e | 4549 | .set_acl = simple_set_acl, |
39f0247d AG |
4550 | #endif |
4551 | }; | |
4552 | ||
92e1d5be | 4553 | static const struct inode_operations shmem_special_inode_operations = { |
f7cd16a5 | 4554 | .getattr = shmem_getattr, |
b09e0fa4 | 4555 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 4556 | .listxattr = shmem_listxattr, |
b09e0fa4 | 4557 | #endif |
39f0247d | 4558 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 4559 | .setattr = shmem_setattr, |
feda821e | 4560 | .set_acl = simple_set_acl, |
39f0247d | 4561 | #endif |
1da177e4 LT |
4562 | }; |
4563 | ||
759b9775 | 4564 | static const struct super_operations shmem_ops = { |
1da177e4 | 4565 | .alloc_inode = shmem_alloc_inode, |
74b1da56 | 4566 | .free_inode = shmem_free_in_core_inode, |
1da177e4 LT |
4567 | .destroy_inode = shmem_destroy_inode, |
4568 | #ifdef CONFIG_TMPFS | |
4569 | .statfs = shmem_statfs, | |
680d794b | 4570 | .show_options = shmem_show_options, |
e09764cf CM |
4571 | #endif |
4572 | #ifdef CONFIG_TMPFS_QUOTA | |
4573 | .get_dquots = shmem_get_dquots, | |
1da177e4 | 4574 | #endif |
1f895f75 | 4575 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
4576 | .drop_inode = generic_delete_inode, |
4577 | .put_super = shmem_put_super, | |
396bcc52 | 4578 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
779750d2 KS |
4579 | .nr_cached_objects = shmem_unused_huge_count, |
4580 | .free_cached_objects = shmem_unused_huge_scan, | |
4581 | #endif | |
1da177e4 LT |
4582 | }; |
4583 | ||
f0f37e2f | 4584 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 4585 | .fault = shmem_fault, |
d7c17551 | 4586 | .map_pages = filemap_map_pages, |
1da177e4 LT |
4587 | #ifdef CONFIG_NUMA |
4588 | .set_policy = shmem_set_policy, | |
4589 | .get_policy = shmem_get_policy, | |
4590 | #endif | |
4591 | }; | |
4592 | ||
d09e8ca6 PT |
4593 | static const struct vm_operations_struct shmem_anon_vm_ops = { |
4594 | .fault = shmem_fault, | |
4595 | .map_pages = filemap_map_pages, | |
4596 | #ifdef CONFIG_NUMA | |
4597 | .set_policy = shmem_set_policy, | |
4598 | .get_policy = shmem_get_policy, | |
4599 | #endif | |
4600 | }; | |
4601 | ||
f3235626 | 4602 | int shmem_init_fs_context(struct fs_context *fc) |
1da177e4 | 4603 | { |
f3235626 DH |
4604 | struct shmem_options *ctx; |
4605 | ||
4606 | ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL); | |
4607 | if (!ctx) | |
4608 | return -ENOMEM; | |
4609 | ||
4610 | ctx->mode = 0777 | S_ISVTX; | |
4611 | ctx->uid = current_fsuid(); | |
4612 | ctx->gid = current_fsgid(); | |
4613 | ||
4614 | fc->fs_private = ctx; | |
4615 | fc->ops = &shmem_fs_context_ops; | |
4616 | return 0; | |
1da177e4 LT |
4617 | } |
4618 | ||
41ffe5d5 | 4619 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
4620 | .owner = THIS_MODULE, |
4621 | .name = "tmpfs", | |
f3235626 DH |
4622 | .init_fs_context = shmem_init_fs_context, |
4623 | #ifdef CONFIG_TMPFS | |
d7167b14 | 4624 | .parameters = shmem_fs_parameters, |
f3235626 | 4625 | #endif |
1da177e4 | 4626 | .kill_sb = kill_litter_super, |
db58b5ee | 4627 | .fs_flags = FS_USERNS_MOUNT | FS_ALLOW_IDMAP, |
1da177e4 | 4628 | }; |
1da177e4 | 4629 | |
9096bbe9 | 4630 | void __init shmem_init(void) |
1da177e4 LT |
4631 | { |
4632 | int error; | |
4633 | ||
9a8ec03e | 4634 | shmem_init_inodecache(); |
1da177e4 | 4635 | |
e09764cf CM |
4636 | #ifdef CONFIG_TMPFS_QUOTA |
4637 | error = register_quota_format(&shmem_quota_format); | |
4638 | if (error < 0) { | |
4639 | pr_err("Could not register quota format\n"); | |
4640 | goto out3; | |
4641 | } | |
4642 | #endif | |
4643 | ||
41ffe5d5 | 4644 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 4645 | if (error) { |
1170532b | 4646 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
4647 | goto out2; |
4648 | } | |
95dc112a | 4649 | |
ca4e0519 | 4650 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
4651 | if (IS_ERR(shm_mnt)) { |
4652 | error = PTR_ERR(shm_mnt); | |
1170532b | 4653 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
4654 | goto out1; |
4655 | } | |
5a6e75f8 | 4656 | |
396bcc52 | 4657 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
435c0b87 | 4658 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4659 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4660 | else | |
5e6e5a12 | 4661 | shmem_huge = SHMEM_HUGE_NEVER; /* just in case it was patched */ |
5a6e75f8 | 4662 | #endif |
9096bbe9 | 4663 | return; |
1da177e4 LT |
4664 | |
4665 | out1: | |
41ffe5d5 | 4666 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 4667 | out2: |
e09764cf CM |
4668 | #ifdef CONFIG_TMPFS_QUOTA |
4669 | unregister_quota_format(&shmem_quota_format); | |
4670 | out3: | |
4671 | #endif | |
41ffe5d5 | 4672 | shmem_destroy_inodecache(); |
1da177e4 | 4673 | shm_mnt = ERR_PTR(error); |
1da177e4 | 4674 | } |
853ac43a | 4675 | |
396bcc52 | 4676 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS) |
5a6e75f8 | 4677 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
79d4d38a | 4678 | struct kobj_attribute *attr, char *buf) |
5a6e75f8 | 4679 | { |
26083eb6 | 4680 | static const int values[] = { |
5a6e75f8 KS |
4681 | SHMEM_HUGE_ALWAYS, |
4682 | SHMEM_HUGE_WITHIN_SIZE, | |
4683 | SHMEM_HUGE_ADVISE, | |
4684 | SHMEM_HUGE_NEVER, | |
4685 | SHMEM_HUGE_DENY, | |
4686 | SHMEM_HUGE_FORCE, | |
4687 | }; | |
79d4d38a JP |
4688 | int len = 0; |
4689 | int i; | |
5a6e75f8 | 4690 | |
79d4d38a JP |
4691 | for (i = 0; i < ARRAY_SIZE(values); i++) { |
4692 | len += sysfs_emit_at(buf, len, | |
9be7d5b0 HD |
4693 | shmem_huge == values[i] ? "%s[%s]" : "%s%s", |
4694 | i ? " " : "", shmem_format_huge(values[i])); | |
5a6e75f8 | 4695 | } |
79d4d38a JP |
4696 | len += sysfs_emit_at(buf, len, "\n"); |
4697 | ||
4698 | return len; | |
5a6e75f8 KS |
4699 | } |
4700 | ||
4701 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
4702 | struct kobj_attribute *attr, const char *buf, size_t count) | |
4703 | { | |
4704 | char tmp[16]; | |
4705 | int huge; | |
4706 | ||
4707 | if (count + 1 > sizeof(tmp)) | |
4708 | return -EINVAL; | |
4709 | memcpy(tmp, buf, count); | |
4710 | tmp[count] = '\0'; | |
4711 | if (count && tmp[count - 1] == '\n') | |
4712 | tmp[count - 1] = '\0'; | |
4713 | ||
4714 | huge = shmem_parse_huge(tmp); | |
4715 | if (huge == -EINVAL) | |
4716 | return -EINVAL; | |
4717 | if (!has_transparent_hugepage() && | |
4718 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
4719 | return -EINVAL; | |
4720 | ||
4721 | shmem_huge = huge; | |
435c0b87 | 4722 | if (shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4723 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4724 | return count; | |
4725 | } | |
4726 | ||
4bfa8ada | 4727 | struct kobj_attribute shmem_enabled_attr = __ATTR_RW(shmem_enabled); |
396bcc52 | 4728 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */ |
f3f0e1d2 | 4729 | |
853ac43a MM |
4730 | #else /* !CONFIG_SHMEM */ |
4731 | ||
4732 | /* | |
4733 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
4734 | * | |
4735 | * This is intended for small system where the benefits of the full | |
4736 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4737 | * their complexity. On systems without swap this code should be | |
4738 | * effectively equivalent, but much lighter weight. | |
4739 | */ | |
4740 | ||
41ffe5d5 | 4741 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4742 | .name = "tmpfs", |
f3235626 | 4743 | .init_fs_context = ramfs_init_fs_context, |
d7167b14 | 4744 | .parameters = ramfs_fs_parameters, |
36ce9d76 | 4745 | .kill_sb = ramfs_kill_sb, |
2b8576cb | 4746 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4747 | }; |
4748 | ||
9096bbe9 | 4749 | void __init shmem_init(void) |
853ac43a | 4750 | { |
41ffe5d5 | 4751 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4752 | |
41ffe5d5 | 4753 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a | 4754 | BUG_ON(IS_ERR(shm_mnt)); |
853ac43a MM |
4755 | } |
4756 | ||
10a9c496 | 4757 | int shmem_unuse(unsigned int type) |
853ac43a MM |
4758 | { |
4759 | return 0; | |
4760 | } | |
4761 | ||
d7c9e99a | 4762 | int shmem_lock(struct file *file, int lock, struct ucounts *ucounts) |
3f96b79a HD |
4763 | { |
4764 | return 0; | |
4765 | } | |
4766 | ||
24513264 HD |
4767 | void shmem_unlock_mapping(struct address_space *mapping) |
4768 | { | |
4769 | } | |
4770 | ||
c01d5b30 HD |
4771 | #ifdef CONFIG_MMU |
4772 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4773 | unsigned long addr, unsigned long len, | |
4774 | unsigned long pgoff, unsigned long flags) | |
4775 | { | |
4776 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4777 | } | |
4778 | #endif | |
4779 | ||
41ffe5d5 | 4780 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4781 | { |
41ffe5d5 | 4782 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4783 | } |
4784 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4785 | ||
0b0a0806 | 4786 | #define shmem_vm_ops generic_file_vm_ops |
d09e8ca6 | 4787 | #define shmem_anon_vm_ops generic_file_vm_ops |
0b0a0806 | 4788 | #define shmem_file_operations ramfs_file_operations |
0b0a0806 HD |
4789 | #define shmem_acct_size(flags, size) 0 |
4790 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a | 4791 | |
9be7d5b0 HD |
4792 | static inline struct inode *shmem_get_inode(struct mnt_idmap *idmap, |
4793 | struct super_block *sb, struct inode *dir, | |
4794 | umode_t mode, dev_t dev, unsigned long flags) | |
71480663 CM |
4795 | { |
4796 | struct inode *inode = ramfs_get_inode(sb, dir, mode, dev); | |
4797 | return inode ? inode : ERR_PTR(-ENOSPC); | |
4798 | } | |
4799 | ||
853ac43a MM |
4800 | #endif /* CONFIG_SHMEM */ |
4801 | ||
4802 | /* common code */ | |
1da177e4 | 4803 | |
9be7d5b0 HD |
4804 | static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, |
4805 | loff_t size, unsigned long flags, unsigned int i_flags) | |
1da177e4 | 4806 | { |
1da177e4 | 4807 | struct inode *inode; |
93dec2da | 4808 | struct file *res; |
1da177e4 | 4809 | |
703321b6 MA |
4810 | if (IS_ERR(mnt)) |
4811 | return ERR_CAST(mnt); | |
1da177e4 | 4812 | |
285b2c4f | 4813 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4814 | return ERR_PTR(-EINVAL); |
4815 | ||
4816 | if (shmem_acct_size(flags, size)) | |
4817 | return ERR_PTR(-ENOMEM); | |
4818 | ||
7a80e5b8 GS |
4819 | if (is_idmapped_mnt(mnt)) |
4820 | return ERR_PTR(-EINVAL); | |
4821 | ||
4822 | inode = shmem_get_inode(&nop_mnt_idmap, mnt->mnt_sb, NULL, | |
4823 | S_IFREG | S_IRWXUGO, 0, flags); | |
71480663 | 4824 | if (IS_ERR(inode)) { |
dac2d1f6 | 4825 | shmem_unacct_size(flags, size); |
71480663 | 4826 | return ERR_CAST(inode); |
dac2d1f6 | 4827 | } |
c7277090 | 4828 | inode->i_flags |= i_flags; |
1da177e4 | 4829 | inode->i_size = size; |
6d6b77f1 | 4830 | clear_nlink(inode); /* It is unlinked */ |
26567cdb | 4831 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
93dec2da AV |
4832 | if (!IS_ERR(res)) |
4833 | res = alloc_file_pseudo(inode, mnt, name, O_RDWR, | |
4834 | &shmem_file_operations); | |
26567cdb | 4835 | if (IS_ERR(res)) |
93dec2da | 4836 | iput(inode); |
6b4d0b27 | 4837 | return res; |
1da177e4 | 4838 | } |
c7277090 EP |
4839 | |
4840 | /** | |
4841 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4842 | * kernel internal. There will be NO LSM permission checks against the | |
4843 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4844 | * higher layer. The users are the big_key and shm implementations. LSM |
4845 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4846 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4847 | * @size: size to be set for the file | |
4848 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4849 | */ | |
4850 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4851 | { | |
703321b6 | 4852 | return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE); |
c7277090 EP |
4853 | } |
4854 | ||
4855 | /** | |
4856 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4857 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4858 | * @size: size to be set for the file | |
4859 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4860 | */ | |
4861 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4862 | { | |
703321b6 | 4863 | return __shmem_file_setup(shm_mnt, name, size, flags, 0); |
c7277090 | 4864 | } |
395e0ddc | 4865 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4866 | |
703321b6 MA |
4867 | /** |
4868 | * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs | |
4869 | * @mnt: the tmpfs mount where the file will be created | |
4870 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4871 | * @size: size to be set for the file | |
4872 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4873 | */ | |
4874 | struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name, | |
4875 | loff_t size, unsigned long flags) | |
4876 | { | |
4877 | return __shmem_file_setup(mnt, name, size, flags, 0); | |
4878 | } | |
4879 | EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt); | |
4880 | ||
46711810 | 4881 | /** |
1da177e4 | 4882 | * shmem_zero_setup - setup a shared anonymous mapping |
45e55300 | 4883 | * @vma: the vma to be mmapped is prepared by do_mmap |
1da177e4 LT |
4884 | */ |
4885 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4886 | { | |
4887 | struct file *file; | |
4888 | loff_t size = vma->vm_end - vma->vm_start; | |
4889 | ||
66fc1303 | 4890 | /* |
c1e8d7c6 | 4891 | * Cloning a new file under mmap_lock leads to a lock ordering conflict |
66fc1303 HD |
4892 | * between XFS directory reading and selinux: since this file is only |
4893 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4894 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4895 | */ | |
703321b6 | 4896 | file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags); |
1da177e4 LT |
4897 | if (IS_ERR(file)) |
4898 | return PTR_ERR(file); | |
4899 | ||
4900 | if (vma->vm_file) | |
4901 | fput(vma->vm_file); | |
4902 | vma->vm_file = file; | |
d09e8ca6 | 4903 | vma->vm_ops = &shmem_anon_vm_ops; |
f3f0e1d2 | 4904 | |
1da177e4 LT |
4905 | return 0; |
4906 | } | |
d9d90e5e HD |
4907 | |
4908 | /** | |
f01b2b3e MWO |
4909 | * shmem_read_folio_gfp - read into page cache, using specified page allocation flags. |
4910 | * @mapping: the folio's address_space | |
4911 | * @index: the folio index | |
d9d90e5e HD |
4912 | * @gfp: the page allocator flags to use if allocating |
4913 | * | |
4914 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4915 | * with any new page allocations done using the specified allocation flags. | |
7e0a1265 | 4916 | * But read_cache_page_gfp() uses the ->read_folio() method: which does not |
d9d90e5e HD |
4917 | * suit tmpfs, since it may have pages in swapcache, and needs to find those |
4918 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4919 | * | |
68da9f05 HD |
4920 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4921 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e | 4922 | */ |
f01b2b3e MWO |
4923 | struct folio *shmem_read_folio_gfp(struct address_space *mapping, |
4924 | pgoff_t index, gfp_t gfp) | |
d9d90e5e | 4925 | { |
68da9f05 HD |
4926 | #ifdef CONFIG_SHMEM |
4927 | struct inode *inode = mapping->host; | |
a3a9c397 | 4928 | struct folio *folio; |
68da9f05 HD |
4929 | int error; |
4930 | ||
30e6a51d | 4931 | BUG_ON(!shmem_mapping(mapping)); |
a3a9c397 | 4932 | error = shmem_get_folio_gfp(inode, index, &folio, SGP_CACHE, |
e3e1a506 | 4933 | gfp, NULL, NULL); |
68da9f05 | 4934 | if (error) |
a7605426 YS |
4935 | return ERR_PTR(error); |
4936 | ||
a3a9c397 | 4937 | folio_unlock(folio); |
f01b2b3e MWO |
4938 | return folio; |
4939 | #else | |
4940 | /* | |
4941 | * The tiny !SHMEM case uses ramfs without swap | |
4942 | */ | |
4943 | return mapping_read_folio_gfp(mapping, index, gfp); | |
4944 | #endif | |
4945 | } | |
4946 | EXPORT_SYMBOL_GPL(shmem_read_folio_gfp); | |
4947 | ||
4948 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4949 | pgoff_t index, gfp_t gfp) | |
4950 | { | |
4951 | struct folio *folio = shmem_read_folio_gfp(mapping, index, gfp); | |
4952 | struct page *page; | |
4953 | ||
4954 | if (IS_ERR(folio)) | |
4955 | return &folio->page; | |
4956 | ||
a3a9c397 | 4957 | page = folio_file_page(folio, index); |
a7605426 | 4958 | if (PageHWPoison(page)) { |
a3a9c397 | 4959 | folio_put(folio); |
a7605426 YS |
4960 | return ERR_PTR(-EIO); |
4961 | } | |
4962 | ||
68da9f05 | 4963 | return page; |
d9d90e5e HD |
4964 | } |
4965 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |