Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Resizable virtual memory filesystem for Linux. | |
3 | * | |
4 | * Copyright (C) 2000 Linus Torvalds. | |
5 | * 2000 Transmeta Corp. | |
6 | * 2000-2001 Christoph Rohland | |
7 | * 2000-2001 SAP AG | |
8 | * 2002 Red Hat Inc. | |
6922c0c7 HD |
9 | * Copyright (C) 2002-2011 Hugh Dickins. |
10 | * Copyright (C) 2011 Google Inc. | |
0edd73b3 | 11 | * Copyright (C) 2002-2005 VERITAS Software Corporation. |
1da177e4 LT |
12 | * Copyright (C) 2004 Andi Kleen, SuSE Labs |
13 | * | |
14 | * Extended attribute support for tmpfs: | |
15 | * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net> | |
16 | * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com> | |
17 | * | |
853ac43a MM |
18 | * tiny-shmem: |
19 | * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com> | |
20 | * | |
1da177e4 LT |
21 | * This file is released under the GPL. |
22 | */ | |
23 | ||
853ac43a MM |
24 | #include <linux/fs.h> |
25 | #include <linux/init.h> | |
26 | #include <linux/vfs.h> | |
27 | #include <linux/mount.h> | |
250297ed | 28 | #include <linux/ramfs.h> |
caefba17 | 29 | #include <linux/pagemap.h> |
853ac43a MM |
30 | #include <linux/file.h> |
31 | #include <linux/mm.h> | |
46c9a946 | 32 | #include <linux/random.h> |
174cd4b1 | 33 | #include <linux/sched/signal.h> |
b95f1b31 | 34 | #include <linux/export.h> |
853ac43a | 35 | #include <linux/swap.h> |
e2e40f2c | 36 | #include <linux/uio.h> |
f3f0e1d2 | 37 | #include <linux/khugepaged.h> |
749df87b | 38 | #include <linux/hugetlb.h> |
b56a2d8a | 39 | #include <linux/frontswap.h> |
626c3920 | 40 | #include <linux/fs_parser.h> |
853ac43a | 41 | |
95cc09d6 AA |
42 | #include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */ |
43 | ||
853ac43a MM |
44 | static struct vfsmount *shm_mnt; |
45 | ||
46 | #ifdef CONFIG_SHMEM | |
1da177e4 LT |
47 | /* |
48 | * This virtual memory filesystem is heavily based on the ramfs. It | |
49 | * extends ramfs by the ability to use swap and honor resource limits | |
50 | * which makes it a completely usable filesystem. | |
51 | */ | |
52 | ||
39f0247d | 53 | #include <linux/xattr.h> |
a5694255 | 54 | #include <linux/exportfs.h> |
1c7c474c | 55 | #include <linux/posix_acl.h> |
feda821e | 56 | #include <linux/posix_acl_xattr.h> |
1da177e4 | 57 | #include <linux/mman.h> |
1da177e4 LT |
58 | #include <linux/string.h> |
59 | #include <linux/slab.h> | |
60 | #include <linux/backing-dev.h> | |
61 | #include <linux/shmem_fs.h> | |
1da177e4 | 62 | #include <linux/writeback.h> |
1da177e4 | 63 | #include <linux/blkdev.h> |
bda97eab | 64 | #include <linux/pagevec.h> |
41ffe5d5 | 65 | #include <linux/percpu_counter.h> |
83e4fa9c | 66 | #include <linux/falloc.h> |
708e3508 | 67 | #include <linux/splice.h> |
1da177e4 LT |
68 | #include <linux/security.h> |
69 | #include <linux/swapops.h> | |
70 | #include <linux/mempolicy.h> | |
71 | #include <linux/namei.h> | |
b00dc3ad | 72 | #include <linux/ctype.h> |
304dbdb7 | 73 | #include <linux/migrate.h> |
c1f60a5a | 74 | #include <linux/highmem.h> |
680d794b | 75 | #include <linux/seq_file.h> |
92562927 | 76 | #include <linux/magic.h> |
9183df25 | 77 | #include <linux/syscalls.h> |
40e041a2 | 78 | #include <linux/fcntl.h> |
9183df25 | 79 | #include <uapi/linux/memfd.h> |
cfda0526 | 80 | #include <linux/userfaultfd_k.h> |
4c27fe4c | 81 | #include <linux/rmap.h> |
2b4db796 | 82 | #include <linux/uuid.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 | ||
69f07ec9 HD |
94 | /* Symlink up to this size is kmalloc'ed instead of using a swappable page */ |
95 | #define SHORT_SYMLINK_LEN 128 | |
96 | ||
1aac1400 | 97 | /* |
f00cdc6d HD |
98 | * shmem_fallocate communicates with shmem_fault or shmem_writepage via |
99 | * inode->i_private (with i_mutex making sure that it has only one user at | |
100 | * a time): we would prefer not to enlarge the shmem inode just for that. | |
1aac1400 HD |
101 | */ |
102 | struct shmem_falloc { | |
8e205f77 | 103 | wait_queue_head_t *waitq; /* faults into hole wait for punch to end */ |
1aac1400 HD |
104 | pgoff_t start; /* start of range currently being fallocated */ |
105 | pgoff_t next; /* the next page offset to be fallocated */ | |
106 | pgoff_t nr_falloced; /* how many new pages have been fallocated */ | |
107 | pgoff_t nr_unswapped; /* how often writepage refused to swap out */ | |
108 | }; | |
109 | ||
0b5071dd AV |
110 | struct shmem_options { |
111 | unsigned long long blocks; | |
112 | unsigned long long inodes; | |
113 | struct mempolicy *mpol; | |
114 | kuid_t uid; | |
115 | kgid_t gid; | |
116 | umode_t mode; | |
ea3271f7 | 117 | bool full_inums; |
0b5071dd AV |
118 | int huge; |
119 | int seen; | |
120 | #define SHMEM_SEEN_BLOCKS 1 | |
121 | #define SHMEM_SEEN_INODES 2 | |
122 | #define SHMEM_SEEN_HUGE 4 | |
ea3271f7 | 123 | #define SHMEM_SEEN_INUMS 8 |
0b5071dd AV |
124 | }; |
125 | ||
b76db735 | 126 | #ifdef CONFIG_TMPFS |
680d794b | 127 | static unsigned long shmem_default_max_blocks(void) |
128 | { | |
ca79b0c2 | 129 | return totalram_pages() / 2; |
680d794b | 130 | } |
131 | ||
132 | static unsigned long shmem_default_max_inodes(void) | |
133 | { | |
ca79b0c2 AK |
134 | unsigned long nr_pages = totalram_pages(); |
135 | ||
136 | return min(nr_pages - totalhigh_pages(), nr_pages / 2); | |
680d794b | 137 | } |
b76db735 | 138 | #endif |
680d794b | 139 | |
bde05d1c HD |
140 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp); |
141 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
142 | struct shmem_inode_info *info, pgoff_t index); | |
c5bf121e VRP |
143 | static int shmem_swapin_page(struct inode *inode, pgoff_t index, |
144 | struct page **pagep, enum sgp_type sgp, | |
145 | gfp_t gfp, struct vm_area_struct *vma, | |
146 | vm_fault_t *fault_type); | |
68da9f05 | 147 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 148 | struct page **pagep, enum sgp_type sgp, |
cfda0526 | 149 | gfp_t gfp, struct vm_area_struct *vma, |
2b740303 | 150 | struct vm_fault *vmf, vm_fault_t *fault_type); |
68da9f05 | 151 | |
f3f0e1d2 | 152 | int shmem_getpage(struct inode *inode, pgoff_t index, |
9e18eb29 | 153 | struct page **pagep, enum sgp_type sgp) |
68da9f05 HD |
154 | { |
155 | return shmem_getpage_gfp(inode, index, pagep, sgp, | |
cfda0526 | 156 | mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL); |
68da9f05 | 157 | } |
1da177e4 | 158 | |
1da177e4 LT |
159 | static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) |
160 | { | |
161 | return sb->s_fs_info; | |
162 | } | |
163 | ||
164 | /* | |
165 | * shmem_file_setup pre-accounts the whole fixed size of a VM object, | |
166 | * for shared memory and for shared anonymous (/dev/zero) mappings | |
167 | * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), | |
168 | * consistent with the pre-accounting of private mappings ... | |
169 | */ | |
170 | static inline int shmem_acct_size(unsigned long flags, loff_t size) | |
171 | { | |
0b0a0806 | 172 | return (flags & VM_NORESERVE) ? |
191c5424 | 173 | 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size)); |
1da177e4 LT |
174 | } |
175 | ||
176 | static inline void shmem_unacct_size(unsigned long flags, loff_t size) | |
177 | { | |
0b0a0806 | 178 | if (!(flags & VM_NORESERVE)) |
1da177e4 LT |
179 | vm_unacct_memory(VM_ACCT(size)); |
180 | } | |
181 | ||
77142517 KK |
182 | static inline int shmem_reacct_size(unsigned long flags, |
183 | loff_t oldsize, loff_t newsize) | |
184 | { | |
185 | if (!(flags & VM_NORESERVE)) { | |
186 | if (VM_ACCT(newsize) > VM_ACCT(oldsize)) | |
187 | return security_vm_enough_memory_mm(current->mm, | |
188 | VM_ACCT(newsize) - VM_ACCT(oldsize)); | |
189 | else if (VM_ACCT(newsize) < VM_ACCT(oldsize)) | |
190 | vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize)); | |
191 | } | |
192 | return 0; | |
193 | } | |
194 | ||
1da177e4 LT |
195 | /* |
196 | * ... whereas tmpfs objects are accounted incrementally as | |
75edd345 | 197 | * pages are allocated, in order to allow large sparse files. |
1da177e4 LT |
198 | * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, |
199 | * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. | |
200 | */ | |
800d8c63 | 201 | static inline int shmem_acct_block(unsigned long flags, long pages) |
1da177e4 | 202 | { |
800d8c63 KS |
203 | if (!(flags & VM_NORESERVE)) |
204 | return 0; | |
205 | ||
206 | return security_vm_enough_memory_mm(current->mm, | |
207 | pages * VM_ACCT(PAGE_SIZE)); | |
1da177e4 LT |
208 | } |
209 | ||
210 | static inline void shmem_unacct_blocks(unsigned long flags, long pages) | |
211 | { | |
0b0a0806 | 212 | if (flags & VM_NORESERVE) |
09cbfeaf | 213 | vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE)); |
1da177e4 LT |
214 | } |
215 | ||
0f079694 MR |
216 | static inline bool shmem_inode_acct_block(struct inode *inode, long pages) |
217 | { | |
218 | struct shmem_inode_info *info = SHMEM_I(inode); | |
219 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
220 | ||
221 | if (shmem_acct_block(info->flags, pages)) | |
222 | return false; | |
223 | ||
224 | if (sbinfo->max_blocks) { | |
225 | if (percpu_counter_compare(&sbinfo->used_blocks, | |
226 | sbinfo->max_blocks - pages) > 0) | |
227 | goto unacct; | |
228 | percpu_counter_add(&sbinfo->used_blocks, pages); | |
229 | } | |
230 | ||
231 | return true; | |
232 | ||
233 | unacct: | |
234 | shmem_unacct_blocks(info->flags, pages); | |
235 | return false; | |
236 | } | |
237 | ||
238 | static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages) | |
239 | { | |
240 | struct shmem_inode_info *info = SHMEM_I(inode); | |
241 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
242 | ||
243 | if (sbinfo->max_blocks) | |
244 | percpu_counter_sub(&sbinfo->used_blocks, pages); | |
245 | shmem_unacct_blocks(info->flags, pages); | |
246 | } | |
247 | ||
759b9775 | 248 | static const struct super_operations shmem_ops; |
30e6a51d | 249 | const struct address_space_operations shmem_aops; |
15ad7cdc | 250 | static const struct file_operations shmem_file_operations; |
92e1d5be AV |
251 | static const struct inode_operations shmem_inode_operations; |
252 | static const struct inode_operations shmem_dir_inode_operations; | |
253 | static const struct inode_operations shmem_special_inode_operations; | |
f0f37e2f | 254 | static const struct vm_operations_struct shmem_vm_ops; |
779750d2 | 255 | static struct file_system_type shmem_fs_type; |
1da177e4 | 256 | |
b0506e48 MR |
257 | bool vma_is_shmem(struct vm_area_struct *vma) |
258 | { | |
259 | return vma->vm_ops == &shmem_vm_ops; | |
260 | } | |
261 | ||
1da177e4 | 262 | static LIST_HEAD(shmem_swaplist); |
cb5f7b9a | 263 | static DEFINE_MUTEX(shmem_swaplist_mutex); |
1da177e4 | 264 | |
e809d5f0 CD |
265 | /* |
266 | * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and | |
267 | * produces a novel ino for the newly allocated inode. | |
268 | * | |
269 | * It may also be called when making a hard link to permit the space needed by | |
270 | * each dentry. However, in that case, no new inode number is needed since that | |
271 | * internally draws from another pool of inode numbers (currently global | |
272 | * get_next_ino()). This case is indicated by passing NULL as inop. | |
273 | */ | |
274 | #define SHMEM_INO_BATCH 1024 | |
275 | static int shmem_reserve_inode(struct super_block *sb, ino_t *inop) | |
5b04c689 PE |
276 | { |
277 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
e809d5f0 CD |
278 | ino_t ino; |
279 | ||
280 | if (!(sb->s_flags & SB_KERNMOUNT)) { | |
5b04c689 | 281 | spin_lock(&sbinfo->stat_lock); |
bb3e96d6 BS |
282 | if (sbinfo->max_inodes) { |
283 | if (!sbinfo->free_inodes) { | |
284 | spin_unlock(&sbinfo->stat_lock); | |
285 | return -ENOSPC; | |
286 | } | |
287 | sbinfo->free_inodes--; | |
5b04c689 | 288 | } |
e809d5f0 CD |
289 | if (inop) { |
290 | ino = sbinfo->next_ino++; | |
291 | if (unlikely(is_zero_ino(ino))) | |
292 | ino = sbinfo->next_ino++; | |
ea3271f7 CD |
293 | if (unlikely(!sbinfo->full_inums && |
294 | ino > UINT_MAX)) { | |
e809d5f0 CD |
295 | /* |
296 | * Emulate get_next_ino uint wraparound for | |
297 | * compatibility | |
298 | */ | |
ea3271f7 CD |
299 | if (IS_ENABLED(CONFIG_64BIT)) |
300 | pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n", | |
301 | __func__, MINOR(sb->s_dev)); | |
302 | sbinfo->next_ino = 1; | |
303 | ino = sbinfo->next_ino++; | |
e809d5f0 CD |
304 | } |
305 | *inop = ino; | |
306 | } | |
5b04c689 | 307 | spin_unlock(&sbinfo->stat_lock); |
e809d5f0 CD |
308 | } else if (inop) { |
309 | /* | |
310 | * __shmem_file_setup, one of our callers, is lock-free: it | |
311 | * doesn't hold stat_lock in shmem_reserve_inode since | |
312 | * max_inodes is always 0, and is called from potentially | |
313 | * unknown contexts. As such, use a per-cpu batched allocator | |
314 | * which doesn't require the per-sb stat_lock unless we are at | |
315 | * the batch boundary. | |
ea3271f7 CD |
316 | * |
317 | * We don't need to worry about inode{32,64} since SB_KERNMOUNT | |
318 | * shmem mounts are not exposed to userspace, so we don't need | |
319 | * to worry about things like glibc compatibility. | |
e809d5f0 CD |
320 | */ |
321 | ino_t *next_ino; | |
322 | next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu()); | |
323 | ino = *next_ino; | |
324 | if (unlikely(ino % SHMEM_INO_BATCH == 0)) { | |
325 | spin_lock(&sbinfo->stat_lock); | |
326 | ino = sbinfo->next_ino; | |
327 | sbinfo->next_ino += SHMEM_INO_BATCH; | |
328 | spin_unlock(&sbinfo->stat_lock); | |
329 | if (unlikely(is_zero_ino(ino))) | |
330 | ino++; | |
331 | } | |
332 | *inop = ino; | |
333 | *next_ino = ++ino; | |
334 | put_cpu(); | |
5b04c689 | 335 | } |
e809d5f0 | 336 | |
5b04c689 PE |
337 | return 0; |
338 | } | |
339 | ||
340 | static void shmem_free_inode(struct super_block *sb) | |
341 | { | |
342 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
343 | if (sbinfo->max_inodes) { | |
344 | spin_lock(&sbinfo->stat_lock); | |
345 | sbinfo->free_inodes++; | |
346 | spin_unlock(&sbinfo->stat_lock); | |
347 | } | |
348 | } | |
349 | ||
46711810 | 350 | /** |
41ffe5d5 | 351 | * shmem_recalc_inode - recalculate the block usage of an inode |
1da177e4 LT |
352 | * @inode: inode to recalc |
353 | * | |
354 | * We have to calculate the free blocks since the mm can drop | |
355 | * undirtied hole pages behind our back. | |
356 | * | |
357 | * But normally info->alloced == inode->i_mapping->nrpages + info->swapped | |
358 | * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) | |
359 | * | |
360 | * It has to be called with the spinlock held. | |
361 | */ | |
362 | static void shmem_recalc_inode(struct inode *inode) | |
363 | { | |
364 | struct shmem_inode_info *info = SHMEM_I(inode); | |
365 | long freed; | |
366 | ||
367 | freed = info->alloced - info->swapped - inode->i_mapping->nrpages; | |
368 | if (freed > 0) { | |
369 | info->alloced -= freed; | |
54af6042 | 370 | inode->i_blocks -= freed * BLOCKS_PER_PAGE; |
0f079694 | 371 | shmem_inode_unacct_blocks(inode, freed); |
1da177e4 LT |
372 | } |
373 | } | |
374 | ||
800d8c63 KS |
375 | bool shmem_charge(struct inode *inode, long pages) |
376 | { | |
377 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 378 | unsigned long flags; |
800d8c63 | 379 | |
0f079694 | 380 | if (!shmem_inode_acct_block(inode, pages)) |
800d8c63 | 381 | return false; |
b1cc94ab | 382 | |
aaa52e34 HD |
383 | /* nrpages adjustment first, then shmem_recalc_inode() when balanced */ |
384 | inode->i_mapping->nrpages += pages; | |
385 | ||
4595ef88 | 386 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
387 | info->alloced += pages; |
388 | inode->i_blocks += pages * BLOCKS_PER_PAGE; | |
389 | shmem_recalc_inode(inode); | |
4595ef88 | 390 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 | 391 | |
800d8c63 KS |
392 | return true; |
393 | } | |
394 | ||
395 | void shmem_uncharge(struct inode *inode, long pages) | |
396 | { | |
397 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4595ef88 | 398 | unsigned long flags; |
800d8c63 | 399 | |
aaa52e34 HD |
400 | /* nrpages adjustment done by __delete_from_page_cache() or caller */ |
401 | ||
4595ef88 | 402 | spin_lock_irqsave(&info->lock, flags); |
800d8c63 KS |
403 | info->alloced -= pages; |
404 | inode->i_blocks -= pages * BLOCKS_PER_PAGE; | |
405 | shmem_recalc_inode(inode); | |
4595ef88 | 406 | spin_unlock_irqrestore(&info->lock, flags); |
800d8c63 | 407 | |
0f079694 | 408 | shmem_inode_unacct_blocks(inode, pages); |
800d8c63 KS |
409 | } |
410 | ||
7a5d0fbb | 411 | /* |
62f945b6 | 412 | * Replace item expected in xarray by a new item, while holding xa_lock. |
7a5d0fbb | 413 | */ |
62f945b6 | 414 | static int shmem_replace_entry(struct address_space *mapping, |
7a5d0fbb HD |
415 | pgoff_t index, void *expected, void *replacement) |
416 | { | |
62f945b6 | 417 | XA_STATE(xas, &mapping->i_pages, index); |
6dbaf22c | 418 | void *item; |
7a5d0fbb HD |
419 | |
420 | VM_BUG_ON(!expected); | |
6dbaf22c | 421 | VM_BUG_ON(!replacement); |
62f945b6 | 422 | item = xas_load(&xas); |
7a5d0fbb HD |
423 | if (item != expected) |
424 | return -ENOENT; | |
62f945b6 | 425 | xas_store(&xas, replacement); |
7a5d0fbb HD |
426 | return 0; |
427 | } | |
428 | ||
d1899228 HD |
429 | /* |
430 | * Sometimes, before we decide whether to proceed or to fail, we must check | |
431 | * that an entry was not already brought back from swap by a racing thread. | |
432 | * | |
433 | * Checking page is not enough: by the time a SwapCache page is locked, it | |
434 | * might be reused, and again be SwapCache, using the same swap as before. | |
435 | */ | |
436 | static bool shmem_confirm_swap(struct address_space *mapping, | |
437 | pgoff_t index, swp_entry_t swap) | |
438 | { | |
a12831bf | 439 | return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap); |
d1899228 HD |
440 | } |
441 | ||
5a6e75f8 KS |
442 | /* |
443 | * Definitions for "huge tmpfs": tmpfs mounted with the huge= option | |
444 | * | |
445 | * SHMEM_HUGE_NEVER: | |
446 | * disables huge pages for the mount; | |
447 | * SHMEM_HUGE_ALWAYS: | |
448 | * enables huge pages for the mount; | |
449 | * SHMEM_HUGE_WITHIN_SIZE: | |
450 | * only allocate huge pages if the page will be fully within i_size, | |
451 | * also respect fadvise()/madvise() hints; | |
452 | * SHMEM_HUGE_ADVISE: | |
453 | * only allocate huge pages if requested with fadvise()/madvise(); | |
454 | */ | |
455 | ||
456 | #define SHMEM_HUGE_NEVER 0 | |
457 | #define SHMEM_HUGE_ALWAYS 1 | |
458 | #define SHMEM_HUGE_WITHIN_SIZE 2 | |
459 | #define SHMEM_HUGE_ADVISE 3 | |
460 | ||
461 | /* | |
462 | * Special values. | |
463 | * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled: | |
464 | * | |
465 | * SHMEM_HUGE_DENY: | |
466 | * disables huge on shm_mnt and all mounts, for emergency use; | |
467 | * SHMEM_HUGE_FORCE: | |
468 | * enables huge on shm_mnt and all mounts, w/o needing option, for testing; | |
469 | * | |
470 | */ | |
471 | #define SHMEM_HUGE_DENY (-1) | |
472 | #define SHMEM_HUGE_FORCE (-2) | |
473 | ||
396bcc52 | 474 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5a6e75f8 KS |
475 | /* ifdef here to avoid bloating shmem.o when not necessary */ |
476 | ||
5b9c98f3 | 477 | static int shmem_huge __read_mostly; |
5a6e75f8 | 478 | |
e5f2249a | 479 | #if defined(CONFIG_SYSFS) |
5a6e75f8 KS |
480 | static int shmem_parse_huge(const char *str) |
481 | { | |
482 | if (!strcmp(str, "never")) | |
483 | return SHMEM_HUGE_NEVER; | |
484 | if (!strcmp(str, "always")) | |
485 | return SHMEM_HUGE_ALWAYS; | |
486 | if (!strcmp(str, "within_size")) | |
487 | return SHMEM_HUGE_WITHIN_SIZE; | |
488 | if (!strcmp(str, "advise")) | |
489 | return SHMEM_HUGE_ADVISE; | |
490 | if (!strcmp(str, "deny")) | |
491 | return SHMEM_HUGE_DENY; | |
492 | if (!strcmp(str, "force")) | |
493 | return SHMEM_HUGE_FORCE; | |
494 | return -EINVAL; | |
495 | } | |
e5f2249a | 496 | #endif |
5a6e75f8 | 497 | |
e5f2249a | 498 | #if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS) |
5a6e75f8 KS |
499 | static const char *shmem_format_huge(int huge) |
500 | { | |
501 | switch (huge) { | |
502 | case SHMEM_HUGE_NEVER: | |
503 | return "never"; | |
504 | case SHMEM_HUGE_ALWAYS: | |
505 | return "always"; | |
506 | case SHMEM_HUGE_WITHIN_SIZE: | |
507 | return "within_size"; | |
508 | case SHMEM_HUGE_ADVISE: | |
509 | return "advise"; | |
510 | case SHMEM_HUGE_DENY: | |
511 | return "deny"; | |
512 | case SHMEM_HUGE_FORCE: | |
513 | return "force"; | |
514 | default: | |
515 | VM_BUG_ON(1); | |
516 | return "bad_val"; | |
517 | } | |
518 | } | |
f1f5929c | 519 | #endif |
5a6e75f8 | 520 | |
779750d2 KS |
521 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
522 | struct shrink_control *sc, unsigned long nr_to_split) | |
523 | { | |
524 | LIST_HEAD(list), *pos, *next; | |
253fd0f0 | 525 | LIST_HEAD(to_remove); |
779750d2 KS |
526 | struct inode *inode; |
527 | struct shmem_inode_info *info; | |
528 | struct page *page; | |
529 | unsigned long batch = sc ? sc->nr_to_scan : 128; | |
530 | int removed = 0, split = 0; | |
531 | ||
532 | if (list_empty(&sbinfo->shrinklist)) | |
533 | return SHRINK_STOP; | |
534 | ||
535 | spin_lock(&sbinfo->shrinklist_lock); | |
536 | list_for_each_safe(pos, next, &sbinfo->shrinklist) { | |
537 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
538 | ||
539 | /* pin the inode */ | |
540 | inode = igrab(&info->vfs_inode); | |
541 | ||
542 | /* inode is about to be evicted */ | |
543 | if (!inode) { | |
544 | list_del_init(&info->shrinklist); | |
545 | removed++; | |
546 | goto next; | |
547 | } | |
548 | ||
549 | /* Check if there's anything to gain */ | |
550 | if (round_up(inode->i_size, PAGE_SIZE) == | |
551 | round_up(inode->i_size, HPAGE_PMD_SIZE)) { | |
253fd0f0 | 552 | list_move(&info->shrinklist, &to_remove); |
779750d2 | 553 | removed++; |
779750d2 KS |
554 | goto next; |
555 | } | |
556 | ||
557 | list_move(&info->shrinklist, &list); | |
558 | next: | |
559 | if (!--batch) | |
560 | break; | |
561 | } | |
562 | spin_unlock(&sbinfo->shrinklist_lock); | |
563 | ||
253fd0f0 KS |
564 | list_for_each_safe(pos, next, &to_remove) { |
565 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
566 | inode = &info->vfs_inode; | |
567 | list_del_init(&info->shrinklist); | |
568 | iput(inode); | |
569 | } | |
570 | ||
779750d2 KS |
571 | list_for_each_safe(pos, next, &list) { |
572 | int ret; | |
573 | ||
574 | info = list_entry(pos, struct shmem_inode_info, shrinklist); | |
575 | inode = &info->vfs_inode; | |
576 | ||
b3cd54b2 KS |
577 | if (nr_to_split && split >= nr_to_split) |
578 | goto leave; | |
779750d2 | 579 | |
b3cd54b2 | 580 | page = find_get_page(inode->i_mapping, |
779750d2 KS |
581 | (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT); |
582 | if (!page) | |
583 | goto drop; | |
584 | ||
b3cd54b2 | 585 | /* No huge page at the end of the file: nothing to split */ |
779750d2 | 586 | if (!PageTransHuge(page)) { |
779750d2 KS |
587 | put_page(page); |
588 | goto drop; | |
589 | } | |
590 | ||
b3cd54b2 KS |
591 | /* |
592 | * Leave the inode on the list if we failed to lock | |
593 | * the page at this time. | |
594 | * | |
595 | * Waiting for the lock may lead to deadlock in the | |
596 | * reclaim path. | |
597 | */ | |
598 | if (!trylock_page(page)) { | |
599 | put_page(page); | |
600 | goto leave; | |
601 | } | |
602 | ||
779750d2 KS |
603 | ret = split_huge_page(page); |
604 | unlock_page(page); | |
605 | put_page(page); | |
606 | ||
b3cd54b2 KS |
607 | /* If split failed leave the inode on the list */ |
608 | if (ret) | |
609 | goto leave; | |
779750d2 KS |
610 | |
611 | split++; | |
612 | drop: | |
613 | list_del_init(&info->shrinklist); | |
614 | removed++; | |
b3cd54b2 | 615 | leave: |
779750d2 KS |
616 | iput(inode); |
617 | } | |
618 | ||
619 | spin_lock(&sbinfo->shrinklist_lock); | |
620 | list_splice_tail(&list, &sbinfo->shrinklist); | |
621 | sbinfo->shrinklist_len -= removed; | |
622 | spin_unlock(&sbinfo->shrinklist_lock); | |
623 | ||
624 | return split; | |
625 | } | |
626 | ||
627 | static long shmem_unused_huge_scan(struct super_block *sb, | |
628 | struct shrink_control *sc) | |
629 | { | |
630 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
631 | ||
632 | if (!READ_ONCE(sbinfo->shrinklist_len)) | |
633 | return SHRINK_STOP; | |
634 | ||
635 | return shmem_unused_huge_shrink(sbinfo, sc, 0); | |
636 | } | |
637 | ||
638 | static long shmem_unused_huge_count(struct super_block *sb, | |
639 | struct shrink_control *sc) | |
640 | { | |
641 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
642 | return READ_ONCE(sbinfo->shrinklist_len); | |
643 | } | |
396bcc52 | 644 | #else /* !CONFIG_TRANSPARENT_HUGEPAGE */ |
5a6e75f8 KS |
645 | |
646 | #define shmem_huge SHMEM_HUGE_DENY | |
647 | ||
779750d2 KS |
648 | static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo, |
649 | struct shrink_control *sc, unsigned long nr_to_split) | |
650 | { | |
651 | return 0; | |
652 | } | |
396bcc52 | 653 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
5a6e75f8 | 654 | |
89fdcd26 YS |
655 | static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo) |
656 | { | |
396bcc52 | 657 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && |
89fdcd26 YS |
658 | (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) && |
659 | shmem_huge != SHMEM_HUGE_DENY) | |
660 | return true; | |
661 | return false; | |
662 | } | |
663 | ||
46f65ec1 HD |
664 | /* |
665 | * Like add_to_page_cache_locked, but error if expected item has gone. | |
666 | */ | |
667 | static int shmem_add_to_page_cache(struct page *page, | |
668 | struct address_space *mapping, | |
3fea5a49 JW |
669 | pgoff_t index, void *expected, gfp_t gfp, |
670 | struct mm_struct *charge_mm) | |
46f65ec1 | 671 | { |
552446a4 MW |
672 | XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page)); |
673 | unsigned long i = 0; | |
d8c6546b | 674 | unsigned long nr = compound_nr(page); |
3fea5a49 | 675 | int error; |
46f65ec1 | 676 | |
800d8c63 KS |
677 | VM_BUG_ON_PAGE(PageTail(page), page); |
678 | VM_BUG_ON_PAGE(index != round_down(index, nr), page); | |
309381fe SL |
679 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
680 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
800d8c63 | 681 | VM_BUG_ON(expected && PageTransHuge(page)); |
46f65ec1 | 682 | |
800d8c63 | 683 | page_ref_add(page, nr); |
b065b432 HD |
684 | page->mapping = mapping; |
685 | page->index = index; | |
686 | ||
4c6355b2 | 687 | if (!PageSwapCache(page)) { |
d9eb1ea2 | 688 | error = mem_cgroup_charge(page, charge_mm, gfp); |
4c6355b2 JW |
689 | if (error) { |
690 | if (PageTransHuge(page)) { | |
691 | count_vm_event(THP_FILE_FALLBACK); | |
692 | count_vm_event(THP_FILE_FALLBACK_CHARGE); | |
693 | } | |
694 | goto error; | |
3fea5a49 | 695 | } |
3fea5a49 JW |
696 | } |
697 | cgroup_throttle_swaprate(page, gfp); | |
698 | ||
552446a4 MW |
699 | do { |
700 | void *entry; | |
701 | xas_lock_irq(&xas); | |
702 | entry = xas_find_conflict(&xas); | |
703 | if (entry != expected) | |
704 | xas_set_err(&xas, -EEXIST); | |
705 | xas_create_range(&xas); | |
706 | if (xas_error(&xas)) | |
707 | goto unlock; | |
708 | next: | |
4101196b | 709 | xas_store(&xas, page); |
552446a4 MW |
710 | if (++i < nr) { |
711 | xas_next(&xas); | |
712 | goto next; | |
800d8c63 | 713 | } |
552446a4 | 714 | if (PageTransHuge(page)) { |
800d8c63 | 715 | count_vm_event(THP_FILE_ALLOC); |
b8eddff8 | 716 | __inc_lruvec_page_state(page, NR_SHMEM_THPS); |
800d8c63 | 717 | } |
800d8c63 | 718 | mapping->nrpages += nr; |
0d1c2072 JW |
719 | __mod_lruvec_page_state(page, NR_FILE_PAGES, nr); |
720 | __mod_lruvec_page_state(page, NR_SHMEM, nr); | |
552446a4 MW |
721 | unlock: |
722 | xas_unlock_irq(&xas); | |
723 | } while (xas_nomem(&xas, gfp)); | |
724 | ||
725 | if (xas_error(&xas)) { | |
3fea5a49 JW |
726 | error = xas_error(&xas); |
727 | goto error; | |
46f65ec1 | 728 | } |
552446a4 MW |
729 | |
730 | return 0; | |
3fea5a49 JW |
731 | error: |
732 | page->mapping = NULL; | |
733 | page_ref_sub(page, nr); | |
734 | return error; | |
46f65ec1 HD |
735 | } |
736 | ||
6922c0c7 HD |
737 | /* |
738 | * Like delete_from_page_cache, but substitutes swap for page. | |
739 | */ | |
740 | static void shmem_delete_from_page_cache(struct page *page, void *radswap) | |
741 | { | |
742 | struct address_space *mapping = page->mapping; | |
743 | int error; | |
744 | ||
800d8c63 KS |
745 | VM_BUG_ON_PAGE(PageCompound(page), page); |
746 | ||
b93b0163 | 747 | xa_lock_irq(&mapping->i_pages); |
62f945b6 | 748 | error = shmem_replace_entry(mapping, page->index, page, radswap); |
6922c0c7 HD |
749 | page->mapping = NULL; |
750 | mapping->nrpages--; | |
0d1c2072 JW |
751 | __dec_lruvec_page_state(page, NR_FILE_PAGES); |
752 | __dec_lruvec_page_state(page, NR_SHMEM); | |
b93b0163 | 753 | xa_unlock_irq(&mapping->i_pages); |
09cbfeaf | 754 | put_page(page); |
6922c0c7 HD |
755 | BUG_ON(error); |
756 | } | |
757 | ||
7a5d0fbb | 758 | /* |
c121d3bb | 759 | * Remove swap entry from page cache, free the swap and its page cache. |
7a5d0fbb HD |
760 | */ |
761 | static int shmem_free_swap(struct address_space *mapping, | |
762 | pgoff_t index, void *radswap) | |
763 | { | |
6dbaf22c | 764 | void *old; |
7a5d0fbb | 765 | |
55f3f7ea | 766 | old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0); |
6dbaf22c JW |
767 | if (old != radswap) |
768 | return -ENOENT; | |
769 | free_swap_and_cache(radix_to_swp_entry(radswap)); | |
770 | return 0; | |
7a5d0fbb HD |
771 | } |
772 | ||
6a15a370 VB |
773 | /* |
774 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
48131e03 | 775 | * given offsets are swapped out. |
6a15a370 | 776 | * |
b93b0163 | 777 | * This is safe to call without i_mutex or the i_pages lock thanks to RCU, |
6a15a370 VB |
778 | * as long as the inode doesn't go away and racy results are not a problem. |
779 | */ | |
48131e03 VB |
780 | unsigned long shmem_partial_swap_usage(struct address_space *mapping, |
781 | pgoff_t start, pgoff_t end) | |
6a15a370 | 782 | { |
7ae3424f | 783 | XA_STATE(xas, &mapping->i_pages, start); |
6a15a370 | 784 | struct page *page; |
48131e03 | 785 | unsigned long swapped = 0; |
6a15a370 VB |
786 | |
787 | rcu_read_lock(); | |
7ae3424f MW |
788 | xas_for_each(&xas, page, end - 1) { |
789 | if (xas_retry(&xas, page)) | |
2cf938aa | 790 | continue; |
3159f943 | 791 | if (xa_is_value(page)) |
6a15a370 VB |
792 | swapped++; |
793 | ||
794 | if (need_resched()) { | |
7ae3424f | 795 | xas_pause(&xas); |
6a15a370 | 796 | cond_resched_rcu(); |
6a15a370 VB |
797 | } |
798 | } | |
799 | ||
800 | rcu_read_unlock(); | |
801 | ||
802 | return swapped << PAGE_SHIFT; | |
803 | } | |
804 | ||
48131e03 VB |
805 | /* |
806 | * Determine (in bytes) how many of the shmem object's pages mapped by the | |
807 | * given vma is swapped out. | |
808 | * | |
b93b0163 | 809 | * This is safe to call without i_mutex or the i_pages lock thanks to RCU, |
48131e03 VB |
810 | * as long as the inode doesn't go away and racy results are not a problem. |
811 | */ | |
812 | unsigned long shmem_swap_usage(struct vm_area_struct *vma) | |
813 | { | |
814 | struct inode *inode = file_inode(vma->vm_file); | |
815 | struct shmem_inode_info *info = SHMEM_I(inode); | |
816 | struct address_space *mapping = inode->i_mapping; | |
817 | unsigned long swapped; | |
818 | ||
819 | /* Be careful as we don't hold info->lock */ | |
820 | swapped = READ_ONCE(info->swapped); | |
821 | ||
822 | /* | |
823 | * The easier cases are when the shmem object has nothing in swap, or | |
824 | * the vma maps it whole. Then we can simply use the stats that we | |
825 | * already track. | |
826 | */ | |
827 | if (!swapped) | |
828 | return 0; | |
829 | ||
830 | if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size) | |
831 | return swapped << PAGE_SHIFT; | |
832 | ||
833 | /* Here comes the more involved part */ | |
834 | return shmem_partial_swap_usage(mapping, | |
835 | linear_page_index(vma, vma->vm_start), | |
836 | linear_page_index(vma, vma->vm_end)); | |
837 | } | |
838 | ||
24513264 HD |
839 | /* |
840 | * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists. | |
841 | */ | |
842 | void shmem_unlock_mapping(struct address_space *mapping) | |
843 | { | |
844 | struct pagevec pvec; | |
845 | pgoff_t indices[PAGEVEC_SIZE]; | |
846 | pgoff_t index = 0; | |
847 | ||
86679820 | 848 | pagevec_init(&pvec); |
24513264 HD |
849 | /* |
850 | * Minor point, but we might as well stop if someone else SHM_LOCKs it. | |
851 | */ | |
852 | while (!mapping_unevictable(mapping)) { | |
853 | /* | |
854 | * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it | |
855 | * has finished, if it hits a row of PAGEVEC_SIZE swap entries. | |
856 | */ | |
0cd6144a JW |
857 | pvec.nr = find_get_entries(mapping, index, |
858 | PAGEVEC_SIZE, pvec.pages, indices); | |
24513264 HD |
859 | if (!pvec.nr) |
860 | break; | |
861 | index = indices[pvec.nr - 1] + 1; | |
0cd6144a | 862 | pagevec_remove_exceptionals(&pvec); |
64e3d12f | 863 | check_move_unevictable_pages(&pvec); |
24513264 HD |
864 | pagevec_release(&pvec); |
865 | cond_resched(); | |
866 | } | |
7a5d0fbb HD |
867 | } |
868 | ||
71725ed1 HD |
869 | /* |
870 | * Check whether a hole-punch or truncation needs to split a huge page, | |
871 | * returning true if no split was required, or the split has been successful. | |
872 | * | |
873 | * Eviction (or truncation to 0 size) should never need to split a huge page; | |
874 | * but in rare cases might do so, if shmem_undo_range() failed to trylock on | |
875 | * head, and then succeeded to trylock on tail. | |
876 | * | |
877 | * A split can only succeed when there are no additional references on the | |
878 | * huge page: so the split below relies upon find_get_entries() having stopped | |
879 | * when it found a subpage of the huge page, without getting further references. | |
880 | */ | |
881 | static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end) | |
882 | { | |
883 | if (!PageTransCompound(page)) | |
884 | return true; | |
885 | ||
886 | /* Just proceed to delete a huge page wholly within the range punched */ | |
887 | if (PageHead(page) && | |
888 | page->index >= start && page->index + HPAGE_PMD_NR <= end) | |
889 | return true; | |
890 | ||
891 | /* Try to split huge page, so we can truly punch the hole or truncate */ | |
892 | return split_huge_page(page) >= 0; | |
893 | } | |
894 | ||
7a5d0fbb | 895 | /* |
7f4446ee | 896 | * Remove range of pages and swap entries from page cache, and free them. |
1635f6a7 | 897 | * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate. |
7a5d0fbb | 898 | */ |
1635f6a7 HD |
899 | static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend, |
900 | bool unfalloc) | |
1da177e4 | 901 | { |
285b2c4f | 902 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 903 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf KS |
904 | pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
905 | pgoff_t end = (lend + 1) >> PAGE_SHIFT; | |
906 | unsigned int partial_start = lstart & (PAGE_SIZE - 1); | |
907 | unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
bda97eab | 908 | struct pagevec pvec; |
7a5d0fbb HD |
909 | pgoff_t indices[PAGEVEC_SIZE]; |
910 | long nr_swaps_freed = 0; | |
285b2c4f | 911 | pgoff_t index; |
bda97eab HD |
912 | int i; |
913 | ||
83e4fa9c HD |
914 | if (lend == -1) |
915 | end = -1; /* unsigned, so actually very big */ | |
bda97eab | 916 | |
86679820 | 917 | pagevec_init(&pvec); |
bda97eab | 918 | index = start; |
83e4fa9c | 919 | while (index < end) { |
0cd6144a JW |
920 | pvec.nr = find_get_entries(mapping, index, |
921 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
922 | pvec.pages, indices); | |
7a5d0fbb HD |
923 | if (!pvec.nr) |
924 | break; | |
bda97eab HD |
925 | for (i = 0; i < pagevec_count(&pvec); i++) { |
926 | struct page *page = pvec.pages[i]; | |
927 | ||
7a5d0fbb | 928 | index = indices[i]; |
83e4fa9c | 929 | if (index >= end) |
bda97eab HD |
930 | break; |
931 | ||
3159f943 | 932 | if (xa_is_value(page)) { |
1635f6a7 HD |
933 | if (unfalloc) |
934 | continue; | |
7a5d0fbb HD |
935 | nr_swaps_freed += !shmem_free_swap(mapping, |
936 | index, page); | |
bda97eab | 937 | continue; |
7a5d0fbb HD |
938 | } |
939 | ||
800d8c63 KS |
940 | VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page); |
941 | ||
7a5d0fbb | 942 | if (!trylock_page(page)) |
bda97eab | 943 | continue; |
800d8c63 | 944 | |
71725ed1 HD |
945 | if ((!unfalloc || !PageUptodate(page)) && |
946 | page_mapping(page) == mapping) { | |
947 | VM_BUG_ON_PAGE(PageWriteback(page), page); | |
948 | if (shmem_punch_compound(page, start, end)) | |
1635f6a7 | 949 | truncate_inode_page(mapping, page); |
bda97eab | 950 | } |
bda97eab HD |
951 | unlock_page(page); |
952 | } | |
0cd6144a | 953 | pagevec_remove_exceptionals(&pvec); |
24513264 | 954 | pagevec_release(&pvec); |
bda97eab HD |
955 | cond_resched(); |
956 | index++; | |
957 | } | |
1da177e4 | 958 | |
83e4fa9c | 959 | if (partial_start) { |
bda97eab | 960 | struct page *page = NULL; |
9e18eb29 | 961 | shmem_getpage(inode, start - 1, &page, SGP_READ); |
bda97eab | 962 | if (page) { |
09cbfeaf | 963 | unsigned int top = PAGE_SIZE; |
83e4fa9c HD |
964 | if (start > end) { |
965 | top = partial_end; | |
966 | partial_end = 0; | |
967 | } | |
968 | zero_user_segment(page, partial_start, top); | |
969 | set_page_dirty(page); | |
970 | unlock_page(page); | |
09cbfeaf | 971 | put_page(page); |
83e4fa9c HD |
972 | } |
973 | } | |
974 | if (partial_end) { | |
975 | struct page *page = NULL; | |
9e18eb29 | 976 | shmem_getpage(inode, end, &page, SGP_READ); |
83e4fa9c HD |
977 | if (page) { |
978 | zero_user_segment(page, 0, partial_end); | |
bda97eab HD |
979 | set_page_dirty(page); |
980 | unlock_page(page); | |
09cbfeaf | 981 | put_page(page); |
bda97eab HD |
982 | } |
983 | } | |
83e4fa9c HD |
984 | if (start >= end) |
985 | return; | |
bda97eab HD |
986 | |
987 | index = start; | |
b1a36650 | 988 | while (index < end) { |
bda97eab | 989 | cond_resched(); |
0cd6144a JW |
990 | |
991 | pvec.nr = find_get_entries(mapping, index, | |
83e4fa9c | 992 | min(end - index, (pgoff_t)PAGEVEC_SIZE), |
0cd6144a | 993 | pvec.pages, indices); |
7a5d0fbb | 994 | if (!pvec.nr) { |
b1a36650 HD |
995 | /* If all gone or hole-punch or unfalloc, we're done */ |
996 | if (index == start || end != -1) | |
bda97eab | 997 | break; |
b1a36650 | 998 | /* But if truncating, restart to make sure all gone */ |
bda97eab HD |
999 | index = start; |
1000 | continue; | |
1001 | } | |
bda97eab HD |
1002 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1003 | struct page *page = pvec.pages[i]; | |
1004 | ||
7a5d0fbb | 1005 | index = indices[i]; |
83e4fa9c | 1006 | if (index >= end) |
bda97eab HD |
1007 | break; |
1008 | ||
3159f943 | 1009 | if (xa_is_value(page)) { |
1635f6a7 HD |
1010 | if (unfalloc) |
1011 | continue; | |
b1a36650 HD |
1012 | if (shmem_free_swap(mapping, index, page)) { |
1013 | /* Swap was replaced by page: retry */ | |
1014 | index--; | |
1015 | break; | |
1016 | } | |
1017 | nr_swaps_freed++; | |
7a5d0fbb HD |
1018 | continue; |
1019 | } | |
1020 | ||
bda97eab | 1021 | lock_page(page); |
800d8c63 | 1022 | |
1635f6a7 | 1023 | if (!unfalloc || !PageUptodate(page)) { |
71725ed1 | 1024 | if (page_mapping(page) != mapping) { |
b1a36650 HD |
1025 | /* Page was replaced by swap: retry */ |
1026 | unlock_page(page); | |
1027 | index--; | |
1028 | break; | |
1635f6a7 | 1029 | } |
71725ed1 HD |
1030 | VM_BUG_ON_PAGE(PageWriteback(page), page); |
1031 | if (shmem_punch_compound(page, start, end)) | |
1032 | truncate_inode_page(mapping, page); | |
0783ac95 | 1033 | else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { |
71725ed1 HD |
1034 | /* Wipe the page and don't get stuck */ |
1035 | clear_highpage(page); | |
1036 | flush_dcache_page(page); | |
1037 | set_page_dirty(page); | |
1038 | if (index < | |
1039 | round_up(start, HPAGE_PMD_NR)) | |
1040 | start = index + 1; | |
1041 | } | |
7a5d0fbb | 1042 | } |
bda97eab HD |
1043 | unlock_page(page); |
1044 | } | |
0cd6144a | 1045 | pagevec_remove_exceptionals(&pvec); |
24513264 | 1046 | pagevec_release(&pvec); |
bda97eab HD |
1047 | index++; |
1048 | } | |
94c1e62d | 1049 | |
4595ef88 | 1050 | spin_lock_irq(&info->lock); |
7a5d0fbb | 1051 | info->swapped -= nr_swaps_freed; |
1da177e4 | 1052 | shmem_recalc_inode(inode); |
4595ef88 | 1053 | spin_unlock_irq(&info->lock); |
1635f6a7 | 1054 | } |
1da177e4 | 1055 | |
1635f6a7 HD |
1056 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
1057 | { | |
1058 | shmem_undo_range(inode, lstart, lend, false); | |
078cd827 | 1059 | inode->i_ctime = inode->i_mtime = current_time(inode); |
1da177e4 | 1060 | } |
94c1e62d | 1061 | EXPORT_SYMBOL_GPL(shmem_truncate_range); |
1da177e4 | 1062 | |
a528d35e DH |
1063 | static int shmem_getattr(const struct path *path, struct kstat *stat, |
1064 | u32 request_mask, unsigned int query_flags) | |
44a30220 | 1065 | { |
a528d35e | 1066 | struct inode *inode = path->dentry->d_inode; |
44a30220 | 1067 | struct shmem_inode_info *info = SHMEM_I(inode); |
89fdcd26 | 1068 | struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb); |
44a30220 | 1069 | |
d0424c42 | 1070 | if (info->alloced - info->swapped != inode->i_mapping->nrpages) { |
4595ef88 | 1071 | spin_lock_irq(&info->lock); |
d0424c42 | 1072 | shmem_recalc_inode(inode); |
4595ef88 | 1073 | spin_unlock_irq(&info->lock); |
d0424c42 | 1074 | } |
44a30220 | 1075 | generic_fillattr(inode, stat); |
89fdcd26 YS |
1076 | |
1077 | if (is_huge_enabled(sb_info)) | |
1078 | stat->blksize = HPAGE_PMD_SIZE; | |
1079 | ||
44a30220 YZ |
1080 | return 0; |
1081 | } | |
1082 | ||
94c1e62d | 1083 | static int shmem_setattr(struct dentry *dentry, struct iattr *attr) |
1da177e4 | 1084 | { |
75c3cfa8 | 1085 | struct inode *inode = d_inode(dentry); |
40e041a2 | 1086 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1087 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 LT |
1088 | int error; |
1089 | ||
2f221d6f | 1090 | error = setattr_prepare(&init_user_ns, dentry, attr); |
db78b877 CH |
1091 | if (error) |
1092 | return error; | |
1093 | ||
94c1e62d HD |
1094 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
1095 | loff_t oldsize = inode->i_size; | |
1096 | loff_t newsize = attr->ia_size; | |
3889e6e7 | 1097 | |
40e041a2 DH |
1098 | /* protected by i_mutex */ |
1099 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
1100 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
1101 | return -EPERM; | |
1102 | ||
94c1e62d | 1103 | if (newsize != oldsize) { |
77142517 KK |
1104 | error = shmem_reacct_size(SHMEM_I(inode)->flags, |
1105 | oldsize, newsize); | |
1106 | if (error) | |
1107 | return error; | |
94c1e62d | 1108 | i_size_write(inode, newsize); |
078cd827 | 1109 | inode->i_ctime = inode->i_mtime = current_time(inode); |
94c1e62d | 1110 | } |
afa2db2f | 1111 | if (newsize <= oldsize) { |
94c1e62d | 1112 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
d0424c42 HD |
1113 | if (oldsize > holebegin) |
1114 | unmap_mapping_range(inode->i_mapping, | |
1115 | holebegin, 0, 1); | |
1116 | if (info->alloced) | |
1117 | shmem_truncate_range(inode, | |
1118 | newsize, (loff_t)-1); | |
94c1e62d | 1119 | /* unmap again to remove racily COWed private pages */ |
d0424c42 HD |
1120 | if (oldsize > holebegin) |
1121 | unmap_mapping_range(inode->i_mapping, | |
1122 | holebegin, 0, 1); | |
779750d2 KS |
1123 | |
1124 | /* | |
1125 | * Part of the huge page can be beyond i_size: subject | |
1126 | * to shrink under memory pressure. | |
1127 | */ | |
396bcc52 | 1128 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { |
779750d2 | 1129 | spin_lock(&sbinfo->shrinklist_lock); |
d041353d CW |
1130 | /* |
1131 | * _careful to defend against unlocked access to | |
1132 | * ->shrink_list in shmem_unused_huge_shrink() | |
1133 | */ | |
1134 | if (list_empty_careful(&info->shrinklist)) { | |
779750d2 KS |
1135 | list_add_tail(&info->shrinklist, |
1136 | &sbinfo->shrinklist); | |
1137 | sbinfo->shrinklist_len++; | |
1138 | } | |
1139 | spin_unlock(&sbinfo->shrinklist_lock); | |
1140 | } | |
94c1e62d | 1141 | } |
1da177e4 LT |
1142 | } |
1143 | ||
2f221d6f | 1144 | setattr_copy(&init_user_ns, inode, attr); |
db78b877 | 1145 | if (attr->ia_valid & ATTR_MODE) |
e65ce2a5 | 1146 | error = posix_acl_chmod(&init_user_ns, inode, inode->i_mode); |
1da177e4 LT |
1147 | return error; |
1148 | } | |
1149 | ||
1f895f75 | 1150 | static void shmem_evict_inode(struct inode *inode) |
1da177e4 | 1151 | { |
1da177e4 | 1152 | struct shmem_inode_info *info = SHMEM_I(inode); |
779750d2 | 1153 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
1da177e4 | 1154 | |
30e6a51d | 1155 | if (shmem_mapping(inode->i_mapping)) { |
1da177e4 LT |
1156 | shmem_unacct_size(info->flags, inode->i_size); |
1157 | inode->i_size = 0; | |
3889e6e7 | 1158 | shmem_truncate_range(inode, 0, (loff_t)-1); |
779750d2 KS |
1159 | if (!list_empty(&info->shrinklist)) { |
1160 | spin_lock(&sbinfo->shrinklist_lock); | |
1161 | if (!list_empty(&info->shrinklist)) { | |
1162 | list_del_init(&info->shrinklist); | |
1163 | sbinfo->shrinklist_len--; | |
1164 | } | |
1165 | spin_unlock(&sbinfo->shrinklist_lock); | |
1166 | } | |
af53d3e9 HD |
1167 | while (!list_empty(&info->swaplist)) { |
1168 | /* Wait while shmem_unuse() is scanning this inode... */ | |
1169 | wait_var_event(&info->stop_eviction, | |
1170 | !atomic_read(&info->stop_eviction)); | |
cb5f7b9a | 1171 | mutex_lock(&shmem_swaplist_mutex); |
af53d3e9 HD |
1172 | /* ...but beware of the race if we peeked too early */ |
1173 | if (!atomic_read(&info->stop_eviction)) | |
1174 | list_del_init(&info->swaplist); | |
cb5f7b9a | 1175 | mutex_unlock(&shmem_swaplist_mutex); |
1da177e4 | 1176 | } |
3ed47db3 | 1177 | } |
b09e0fa4 | 1178 | |
38f38657 | 1179 | simple_xattrs_free(&info->xattrs); |
0f3c42f5 | 1180 | WARN_ON(inode->i_blocks); |
5b04c689 | 1181 | shmem_free_inode(inode->i_sb); |
dbd5768f | 1182 | clear_inode(inode); |
1da177e4 LT |
1183 | } |
1184 | ||
b56a2d8a VRP |
1185 | extern struct swap_info_struct *swap_info[]; |
1186 | ||
1187 | static int shmem_find_swap_entries(struct address_space *mapping, | |
1188 | pgoff_t start, unsigned int nr_entries, | |
1189 | struct page **entries, pgoff_t *indices, | |
87039546 | 1190 | unsigned int type, bool frontswap) |
478922e2 | 1191 | { |
b56a2d8a VRP |
1192 | XA_STATE(xas, &mapping->i_pages, start); |
1193 | struct page *page; | |
87039546 | 1194 | swp_entry_t entry; |
b56a2d8a VRP |
1195 | unsigned int ret = 0; |
1196 | ||
1197 | if (!nr_entries) | |
1198 | return 0; | |
478922e2 MW |
1199 | |
1200 | rcu_read_lock(); | |
b56a2d8a VRP |
1201 | xas_for_each(&xas, page, ULONG_MAX) { |
1202 | if (xas_retry(&xas, page)) | |
5b9c98f3 | 1203 | continue; |
b56a2d8a VRP |
1204 | |
1205 | if (!xa_is_value(page)) | |
478922e2 | 1206 | continue; |
b56a2d8a | 1207 | |
87039546 HD |
1208 | entry = radix_to_swp_entry(page); |
1209 | if (swp_type(entry) != type) | |
1210 | continue; | |
1211 | if (frontswap && | |
1212 | !frontswap_test(swap_info[type], swp_offset(entry))) | |
1213 | continue; | |
b56a2d8a VRP |
1214 | |
1215 | indices[ret] = xas.xa_index; | |
1216 | entries[ret] = page; | |
1217 | ||
1218 | if (need_resched()) { | |
1219 | xas_pause(&xas); | |
1220 | cond_resched_rcu(); | |
1221 | } | |
1222 | if (++ret == nr_entries) | |
1223 | break; | |
478922e2 | 1224 | } |
478922e2 | 1225 | rcu_read_unlock(); |
e21a2955 | 1226 | |
b56a2d8a | 1227 | return ret; |
478922e2 MW |
1228 | } |
1229 | ||
46f65ec1 | 1230 | /* |
b56a2d8a VRP |
1231 | * Move the swapped pages for an inode to page cache. Returns the count |
1232 | * of pages swapped in, or the error in case of failure. | |
46f65ec1 | 1233 | */ |
b56a2d8a VRP |
1234 | static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec, |
1235 | pgoff_t *indices) | |
1da177e4 | 1236 | { |
b56a2d8a VRP |
1237 | int i = 0; |
1238 | int ret = 0; | |
bde05d1c | 1239 | int error = 0; |
b56a2d8a | 1240 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 1241 | |
b56a2d8a VRP |
1242 | for (i = 0; i < pvec.nr; i++) { |
1243 | struct page *page = pvec.pages[i]; | |
2e0e26c7 | 1244 | |
b56a2d8a VRP |
1245 | if (!xa_is_value(page)) |
1246 | continue; | |
1247 | error = shmem_swapin_page(inode, indices[i], | |
1248 | &page, SGP_CACHE, | |
1249 | mapping_gfp_mask(mapping), | |
1250 | NULL, NULL); | |
1251 | if (error == 0) { | |
1252 | unlock_page(page); | |
1253 | put_page(page); | |
1254 | ret++; | |
1255 | } | |
1256 | if (error == -ENOMEM) | |
1257 | break; | |
1258 | error = 0; | |
bde05d1c | 1259 | } |
b56a2d8a VRP |
1260 | return error ? error : ret; |
1261 | } | |
bde05d1c | 1262 | |
b56a2d8a VRP |
1263 | /* |
1264 | * If swap found in inode, free it and move page from swapcache to filecache. | |
1265 | */ | |
1266 | static int shmem_unuse_inode(struct inode *inode, unsigned int type, | |
1267 | bool frontswap, unsigned long *fs_pages_to_unuse) | |
1268 | { | |
1269 | struct address_space *mapping = inode->i_mapping; | |
1270 | pgoff_t start = 0; | |
1271 | struct pagevec pvec; | |
1272 | pgoff_t indices[PAGEVEC_SIZE]; | |
1273 | bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0); | |
1274 | int ret = 0; | |
1275 | ||
1276 | pagevec_init(&pvec); | |
1277 | do { | |
1278 | unsigned int nr_entries = PAGEVEC_SIZE; | |
1279 | ||
1280 | if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE) | |
1281 | nr_entries = *fs_pages_to_unuse; | |
1282 | ||
1283 | pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries, | |
1284 | pvec.pages, indices, | |
87039546 | 1285 | type, frontswap); |
b56a2d8a VRP |
1286 | if (pvec.nr == 0) { |
1287 | ret = 0; | |
1288 | break; | |
46f65ec1 | 1289 | } |
b56a2d8a VRP |
1290 | |
1291 | ret = shmem_unuse_swap_entries(inode, pvec, indices); | |
1292 | if (ret < 0) | |
1293 | break; | |
1294 | ||
1295 | if (frontswap_partial) { | |
1296 | *fs_pages_to_unuse -= ret; | |
1297 | if (*fs_pages_to_unuse == 0) { | |
1298 | ret = FRONTSWAP_PAGES_UNUSED; | |
1299 | break; | |
1300 | } | |
1301 | } | |
1302 | ||
1303 | start = indices[pvec.nr - 1]; | |
1304 | } while (true); | |
1305 | ||
1306 | return ret; | |
1da177e4 LT |
1307 | } |
1308 | ||
1309 | /* | |
b56a2d8a VRP |
1310 | * Read all the shared memory data that resides in the swap |
1311 | * device 'type' back into memory, so the swap device can be | |
1312 | * unused. | |
1da177e4 | 1313 | */ |
b56a2d8a VRP |
1314 | int shmem_unuse(unsigned int type, bool frontswap, |
1315 | unsigned long *fs_pages_to_unuse) | |
1da177e4 | 1316 | { |
b56a2d8a | 1317 | struct shmem_inode_info *info, *next; |
bde05d1c HD |
1318 | int error = 0; |
1319 | ||
b56a2d8a VRP |
1320 | if (list_empty(&shmem_swaplist)) |
1321 | return 0; | |
1322 | ||
1323 | mutex_lock(&shmem_swaplist_mutex); | |
b56a2d8a VRP |
1324 | list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) { |
1325 | if (!info->swapped) { | |
6922c0c7 | 1326 | list_del_init(&info->swaplist); |
b56a2d8a VRP |
1327 | continue; |
1328 | } | |
af53d3e9 HD |
1329 | /* |
1330 | * Drop the swaplist mutex while searching the inode for swap; | |
1331 | * but before doing so, make sure shmem_evict_inode() will not | |
1332 | * remove placeholder inode from swaplist, nor let it be freed | |
1333 | * (igrab() would protect from unlink, but not from unmount). | |
1334 | */ | |
1335 | atomic_inc(&info->stop_eviction); | |
b56a2d8a | 1336 | mutex_unlock(&shmem_swaplist_mutex); |
b56a2d8a | 1337 | |
af53d3e9 | 1338 | error = shmem_unuse_inode(&info->vfs_inode, type, frontswap, |
b56a2d8a | 1339 | fs_pages_to_unuse); |
cb5f7b9a | 1340 | cond_resched(); |
b56a2d8a VRP |
1341 | |
1342 | mutex_lock(&shmem_swaplist_mutex); | |
1343 | next = list_next_entry(info, swaplist); | |
1344 | if (!info->swapped) | |
1345 | list_del_init(&info->swaplist); | |
af53d3e9 HD |
1346 | if (atomic_dec_and_test(&info->stop_eviction)) |
1347 | wake_up_var(&info->stop_eviction); | |
b56a2d8a | 1348 | if (error) |
778dd893 | 1349 | break; |
1da177e4 | 1350 | } |
cb5f7b9a | 1351 | mutex_unlock(&shmem_swaplist_mutex); |
778dd893 | 1352 | |
778dd893 | 1353 | return error; |
1da177e4 LT |
1354 | } |
1355 | ||
1356 | /* | |
1357 | * Move the page from the page cache to the swap cache. | |
1358 | */ | |
1359 | static int shmem_writepage(struct page *page, struct writeback_control *wbc) | |
1360 | { | |
1361 | struct shmem_inode_info *info; | |
1da177e4 | 1362 | struct address_space *mapping; |
1da177e4 | 1363 | struct inode *inode; |
6922c0c7 HD |
1364 | swp_entry_t swap; |
1365 | pgoff_t index; | |
1da177e4 | 1366 | |
800d8c63 | 1367 | VM_BUG_ON_PAGE(PageCompound(page), page); |
1da177e4 | 1368 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1369 | mapping = page->mapping; |
1370 | index = page->index; | |
1371 | inode = mapping->host; | |
1372 | info = SHMEM_I(inode); | |
1373 | if (info->flags & VM_LOCKED) | |
1374 | goto redirty; | |
d9fe526a | 1375 | if (!total_swap_pages) |
1da177e4 LT |
1376 | goto redirty; |
1377 | ||
d9fe526a | 1378 | /* |
97b713ba CH |
1379 | * Our capabilities prevent regular writeback or sync from ever calling |
1380 | * shmem_writepage; but a stacking filesystem might use ->writepage of | |
1381 | * its underlying filesystem, in which case tmpfs should write out to | |
1382 | * swap only in response to memory pressure, and not for the writeback | |
1383 | * threads or sync. | |
d9fe526a | 1384 | */ |
48f170fb HD |
1385 | if (!wbc->for_reclaim) { |
1386 | WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ | |
1387 | goto redirty; | |
1388 | } | |
1635f6a7 HD |
1389 | |
1390 | /* | |
1391 | * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC | |
1392 | * value into swapfile.c, the only way we can correctly account for a | |
1393 | * fallocated page arriving here is now to initialize it and write it. | |
1aac1400 HD |
1394 | * |
1395 | * That's okay for a page already fallocated earlier, but if we have | |
1396 | * not yet completed the fallocation, then (a) we want to keep track | |
1397 | * of this page in case we have to undo it, and (b) it may not be a | |
1398 | * good idea to continue anyway, once we're pushing into swap. So | |
1399 | * reactivate the page, and let shmem_fallocate() quit when too many. | |
1635f6a7 HD |
1400 | */ |
1401 | if (!PageUptodate(page)) { | |
1aac1400 HD |
1402 | if (inode->i_private) { |
1403 | struct shmem_falloc *shmem_falloc; | |
1404 | spin_lock(&inode->i_lock); | |
1405 | shmem_falloc = inode->i_private; | |
1406 | if (shmem_falloc && | |
8e205f77 | 1407 | !shmem_falloc->waitq && |
1aac1400 HD |
1408 | index >= shmem_falloc->start && |
1409 | index < shmem_falloc->next) | |
1410 | shmem_falloc->nr_unswapped++; | |
1411 | else | |
1412 | shmem_falloc = NULL; | |
1413 | spin_unlock(&inode->i_lock); | |
1414 | if (shmem_falloc) | |
1415 | goto redirty; | |
1416 | } | |
1635f6a7 HD |
1417 | clear_highpage(page); |
1418 | flush_dcache_page(page); | |
1419 | SetPageUptodate(page); | |
1420 | } | |
1421 | ||
38d8b4e6 | 1422 | swap = get_swap_page(page); |
48f170fb HD |
1423 | if (!swap.val) |
1424 | goto redirty; | |
d9fe526a | 1425 | |
b1dea800 HD |
1426 | /* |
1427 | * Add inode to shmem_unuse()'s list of swapped-out inodes, | |
6922c0c7 HD |
1428 | * if it's not already there. Do it now before the page is |
1429 | * moved to swap cache, when its pagelock no longer protects | |
b1dea800 | 1430 | * the inode from eviction. But don't unlock the mutex until |
6922c0c7 HD |
1431 | * we've incremented swapped, because shmem_unuse_inode() will |
1432 | * prune a !swapped inode from the swaplist under this mutex. | |
b1dea800 | 1433 | */ |
48f170fb HD |
1434 | mutex_lock(&shmem_swaplist_mutex); |
1435 | if (list_empty(&info->swaplist)) | |
b56a2d8a | 1436 | list_add(&info->swaplist, &shmem_swaplist); |
b1dea800 | 1437 | |
4afab1cd | 1438 | if (add_to_swap_cache(page, swap, |
3852f676 JK |
1439 | __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN, |
1440 | NULL) == 0) { | |
4595ef88 | 1441 | spin_lock_irq(&info->lock); |
6922c0c7 | 1442 | shmem_recalc_inode(inode); |
267a4c76 | 1443 | info->swapped++; |
4595ef88 | 1444 | spin_unlock_irq(&info->lock); |
6922c0c7 | 1445 | |
267a4c76 HD |
1446 | swap_shmem_alloc(swap); |
1447 | shmem_delete_from_page_cache(page, swp_to_radix_entry(swap)); | |
1448 | ||
6922c0c7 | 1449 | mutex_unlock(&shmem_swaplist_mutex); |
d9fe526a | 1450 | BUG_ON(page_mapped(page)); |
9fab5619 | 1451 | swap_writepage(page, wbc); |
1da177e4 LT |
1452 | return 0; |
1453 | } | |
1454 | ||
6922c0c7 | 1455 | mutex_unlock(&shmem_swaplist_mutex); |
75f6d6d2 | 1456 | put_swap_page(page, swap); |
1da177e4 LT |
1457 | redirty: |
1458 | set_page_dirty(page); | |
d9fe526a HD |
1459 | if (wbc->for_reclaim) |
1460 | return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ | |
1461 | unlock_page(page); | |
1462 | return 0; | |
1da177e4 LT |
1463 | } |
1464 | ||
75edd345 | 1465 | #if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS) |
71fe804b | 1466 | static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) |
680d794b | 1467 | { |
095f1fc4 | 1468 | char buffer[64]; |
680d794b | 1469 | |
71fe804b | 1470 | if (!mpol || mpol->mode == MPOL_DEFAULT) |
095f1fc4 | 1471 | return; /* show nothing */ |
680d794b | 1472 | |
a7a88b23 | 1473 | mpol_to_str(buffer, sizeof(buffer), mpol); |
095f1fc4 LS |
1474 | |
1475 | seq_printf(seq, ",mpol=%s", buffer); | |
680d794b | 1476 | } |
71fe804b LS |
1477 | |
1478 | static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1479 | { | |
1480 | struct mempolicy *mpol = NULL; | |
1481 | if (sbinfo->mpol) { | |
1482 | spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ | |
1483 | mpol = sbinfo->mpol; | |
1484 | mpol_get(mpol); | |
1485 | spin_unlock(&sbinfo->stat_lock); | |
1486 | } | |
1487 | return mpol; | |
1488 | } | |
75edd345 HD |
1489 | #else /* !CONFIG_NUMA || !CONFIG_TMPFS */ |
1490 | static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) | |
1491 | { | |
1492 | } | |
1493 | static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) | |
1494 | { | |
1495 | return NULL; | |
1496 | } | |
1497 | #endif /* CONFIG_NUMA && CONFIG_TMPFS */ | |
1498 | #ifndef CONFIG_NUMA | |
1499 | #define vm_policy vm_private_data | |
1500 | #endif | |
680d794b | 1501 | |
800d8c63 KS |
1502 | static void shmem_pseudo_vma_init(struct vm_area_struct *vma, |
1503 | struct shmem_inode_info *info, pgoff_t index) | |
1504 | { | |
1505 | /* Create a pseudo vma that just contains the policy */ | |
2c4541e2 | 1506 | vma_init(vma, NULL); |
800d8c63 KS |
1507 | /* Bias interleave by inode number to distribute better across nodes */ |
1508 | vma->vm_pgoff = index + info->vfs_inode.i_ino; | |
800d8c63 KS |
1509 | vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index); |
1510 | } | |
1511 | ||
1512 | static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma) | |
1513 | { | |
1514 | /* Drop reference taken by mpol_shared_policy_lookup() */ | |
1515 | mpol_cond_put(vma->vm_policy); | |
1516 | } | |
1517 | ||
41ffe5d5 HD |
1518 | static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, |
1519 | struct shmem_inode_info *info, pgoff_t index) | |
1da177e4 | 1520 | { |
1da177e4 | 1521 | struct vm_area_struct pvma; |
18a2f371 | 1522 | struct page *page; |
e9e9b7ec | 1523 | struct vm_fault vmf; |
52cd3b07 | 1524 | |
800d8c63 | 1525 | shmem_pseudo_vma_init(&pvma, info, index); |
e9e9b7ec MK |
1526 | vmf.vma = &pvma; |
1527 | vmf.address = 0; | |
1528 | page = swap_cluster_readahead(swap, gfp, &vmf); | |
800d8c63 | 1529 | shmem_pseudo_vma_destroy(&pvma); |
18a2f371 | 1530 | |
800d8c63 KS |
1531 | return page; |
1532 | } | |
1533 | ||
1534 | static struct page *shmem_alloc_hugepage(gfp_t gfp, | |
1535 | struct shmem_inode_info *info, pgoff_t index) | |
1536 | { | |
1537 | struct vm_area_struct pvma; | |
7b8d046f MW |
1538 | struct address_space *mapping = info->vfs_inode.i_mapping; |
1539 | pgoff_t hindex; | |
800d8c63 KS |
1540 | struct page *page; |
1541 | ||
4620a06e | 1542 | hindex = round_down(index, HPAGE_PMD_NR); |
7b8d046f MW |
1543 | if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1, |
1544 | XA_PRESENT)) | |
800d8c63 | 1545 | return NULL; |
18a2f371 | 1546 | |
800d8c63 KS |
1547 | shmem_pseudo_vma_init(&pvma, info, hindex); |
1548 | page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN, | |
19deb769 | 1549 | HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true); |
800d8c63 KS |
1550 | shmem_pseudo_vma_destroy(&pvma); |
1551 | if (page) | |
1552 | prep_transhuge_page(page); | |
dcdf11ee DR |
1553 | else |
1554 | count_vm_event(THP_FILE_FALLBACK); | |
18a2f371 | 1555 | return page; |
1da177e4 LT |
1556 | } |
1557 | ||
02098fea | 1558 | static struct page *shmem_alloc_page(gfp_t gfp, |
41ffe5d5 | 1559 | struct shmem_inode_info *info, pgoff_t index) |
1da177e4 LT |
1560 | { |
1561 | struct vm_area_struct pvma; | |
18a2f371 | 1562 | struct page *page; |
1da177e4 | 1563 | |
800d8c63 KS |
1564 | shmem_pseudo_vma_init(&pvma, info, index); |
1565 | page = alloc_page_vma(gfp, &pvma, 0); | |
1566 | shmem_pseudo_vma_destroy(&pvma); | |
1567 | ||
1568 | return page; | |
1569 | } | |
1570 | ||
1571 | static struct page *shmem_alloc_and_acct_page(gfp_t gfp, | |
0f079694 | 1572 | struct inode *inode, |
800d8c63 KS |
1573 | pgoff_t index, bool huge) |
1574 | { | |
0f079694 | 1575 | struct shmem_inode_info *info = SHMEM_I(inode); |
800d8c63 KS |
1576 | struct page *page; |
1577 | int nr; | |
1578 | int err = -ENOSPC; | |
52cd3b07 | 1579 | |
396bcc52 | 1580 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) |
800d8c63 KS |
1581 | huge = false; |
1582 | nr = huge ? HPAGE_PMD_NR : 1; | |
1583 | ||
0f079694 | 1584 | if (!shmem_inode_acct_block(inode, nr)) |
800d8c63 | 1585 | goto failed; |
800d8c63 KS |
1586 | |
1587 | if (huge) | |
1588 | page = shmem_alloc_hugepage(gfp, info, index); | |
1589 | else | |
1590 | page = shmem_alloc_page(gfp, info, index); | |
75edd345 HD |
1591 | if (page) { |
1592 | __SetPageLocked(page); | |
1593 | __SetPageSwapBacked(page); | |
800d8c63 | 1594 | return page; |
75edd345 | 1595 | } |
18a2f371 | 1596 | |
800d8c63 | 1597 | err = -ENOMEM; |
0f079694 | 1598 | shmem_inode_unacct_blocks(inode, nr); |
800d8c63 KS |
1599 | failed: |
1600 | return ERR_PTR(err); | |
1da177e4 | 1601 | } |
71fe804b | 1602 | |
bde05d1c HD |
1603 | /* |
1604 | * When a page is moved from swapcache to shmem filecache (either by the | |
1605 | * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of | |
1606 | * shmem_unuse_inode()), it may have been read in earlier from swap, in | |
1607 | * ignorance of the mapping it belongs to. If that mapping has special | |
1608 | * constraints (like the gma500 GEM driver, which requires RAM below 4GB), | |
1609 | * we may need to copy to a suitable page before moving to filecache. | |
1610 | * | |
1611 | * In a future release, this may well be extended to respect cpuset and | |
1612 | * NUMA mempolicy, and applied also to anonymous pages in do_swap_page(); | |
1613 | * but for now it is a simple matter of zone. | |
1614 | */ | |
1615 | static bool shmem_should_replace_page(struct page *page, gfp_t gfp) | |
1616 | { | |
1617 | return page_zonenum(page) > gfp_zone(gfp); | |
1618 | } | |
1619 | ||
1620 | static int shmem_replace_page(struct page **pagep, gfp_t gfp, | |
1621 | struct shmem_inode_info *info, pgoff_t index) | |
1622 | { | |
1623 | struct page *oldpage, *newpage; | |
1624 | struct address_space *swap_mapping; | |
c1cb20d4 | 1625 | swp_entry_t entry; |
bde05d1c HD |
1626 | pgoff_t swap_index; |
1627 | int error; | |
1628 | ||
1629 | oldpage = *pagep; | |
c1cb20d4 YZ |
1630 | entry.val = page_private(oldpage); |
1631 | swap_index = swp_offset(entry); | |
bde05d1c HD |
1632 | swap_mapping = page_mapping(oldpage); |
1633 | ||
1634 | /* | |
1635 | * We have arrived here because our zones are constrained, so don't | |
1636 | * limit chance of success by further cpuset and node constraints. | |
1637 | */ | |
1638 | gfp &= ~GFP_CONSTRAINT_MASK; | |
1639 | newpage = shmem_alloc_page(gfp, info, index); | |
1640 | if (!newpage) | |
1641 | return -ENOMEM; | |
bde05d1c | 1642 | |
09cbfeaf | 1643 | get_page(newpage); |
bde05d1c | 1644 | copy_highpage(newpage, oldpage); |
0142ef6c | 1645 | flush_dcache_page(newpage); |
bde05d1c | 1646 | |
9956edf3 HD |
1647 | __SetPageLocked(newpage); |
1648 | __SetPageSwapBacked(newpage); | |
bde05d1c | 1649 | SetPageUptodate(newpage); |
c1cb20d4 | 1650 | set_page_private(newpage, entry.val); |
bde05d1c HD |
1651 | SetPageSwapCache(newpage); |
1652 | ||
1653 | /* | |
1654 | * Our caller will very soon move newpage out of swapcache, but it's | |
1655 | * a nice clean interface for us to replace oldpage by newpage there. | |
1656 | */ | |
b93b0163 | 1657 | xa_lock_irq(&swap_mapping->i_pages); |
62f945b6 | 1658 | error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage); |
0142ef6c | 1659 | if (!error) { |
0d1c2072 JW |
1660 | mem_cgroup_migrate(oldpage, newpage); |
1661 | __inc_lruvec_page_state(newpage, NR_FILE_PAGES); | |
1662 | __dec_lruvec_page_state(oldpage, NR_FILE_PAGES); | |
0142ef6c | 1663 | } |
b93b0163 | 1664 | xa_unlock_irq(&swap_mapping->i_pages); |
bde05d1c | 1665 | |
0142ef6c HD |
1666 | if (unlikely(error)) { |
1667 | /* | |
1668 | * Is this possible? I think not, now that our callers check | |
1669 | * both PageSwapCache and page_private after getting page lock; | |
1670 | * but be defensive. Reverse old to newpage for clear and free. | |
1671 | */ | |
1672 | oldpage = newpage; | |
1673 | } else { | |
6058eaec | 1674 | lru_cache_add(newpage); |
0142ef6c HD |
1675 | *pagep = newpage; |
1676 | } | |
bde05d1c HD |
1677 | |
1678 | ClearPageSwapCache(oldpage); | |
1679 | set_page_private(oldpage, 0); | |
1680 | ||
1681 | unlock_page(oldpage); | |
09cbfeaf KS |
1682 | put_page(oldpage); |
1683 | put_page(oldpage); | |
0142ef6c | 1684 | return error; |
bde05d1c HD |
1685 | } |
1686 | ||
c5bf121e VRP |
1687 | /* |
1688 | * Swap in the page pointed to by *pagep. | |
1689 | * Caller has to make sure that *pagep contains a valid swapped page. | |
1690 | * Returns 0 and the page in pagep if success. On failure, returns the | |
af44c12f | 1691 | * error code and NULL in *pagep. |
c5bf121e VRP |
1692 | */ |
1693 | static int shmem_swapin_page(struct inode *inode, pgoff_t index, | |
1694 | struct page **pagep, enum sgp_type sgp, | |
1695 | gfp_t gfp, struct vm_area_struct *vma, | |
1696 | vm_fault_t *fault_type) | |
1697 | { | |
1698 | struct address_space *mapping = inode->i_mapping; | |
1699 | struct shmem_inode_info *info = SHMEM_I(inode); | |
1700 | struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm; | |
c5bf121e VRP |
1701 | struct page *page; |
1702 | swp_entry_t swap; | |
1703 | int error; | |
1704 | ||
1705 | VM_BUG_ON(!*pagep || !xa_is_value(*pagep)); | |
1706 | swap = radix_to_swp_entry(*pagep); | |
1707 | *pagep = NULL; | |
1708 | ||
1709 | /* Look it up and read it in.. */ | |
1710 | page = lookup_swap_cache(swap, NULL, 0); | |
1711 | if (!page) { | |
1712 | /* Or update major stats only when swapin succeeds?? */ | |
1713 | if (fault_type) { | |
1714 | *fault_type |= VM_FAULT_MAJOR; | |
1715 | count_vm_event(PGMAJFAULT); | |
1716 | count_memcg_event_mm(charge_mm, PGMAJFAULT); | |
1717 | } | |
1718 | /* Here we actually start the io */ | |
1719 | page = shmem_swapin(swap, gfp, info, index); | |
1720 | if (!page) { | |
1721 | error = -ENOMEM; | |
1722 | goto failed; | |
1723 | } | |
1724 | } | |
1725 | ||
1726 | /* We have to do this with page locked to prevent races */ | |
1727 | lock_page(page); | |
1728 | if (!PageSwapCache(page) || page_private(page) != swap.val || | |
1729 | !shmem_confirm_swap(mapping, index, swap)) { | |
1730 | error = -EEXIST; | |
1731 | goto unlock; | |
1732 | } | |
1733 | if (!PageUptodate(page)) { | |
1734 | error = -EIO; | |
1735 | goto failed; | |
1736 | } | |
1737 | wait_on_page_writeback(page); | |
1738 | ||
8a84802e SP |
1739 | /* |
1740 | * Some architectures may have to restore extra metadata to the | |
1741 | * physical page after reading from swap. | |
1742 | */ | |
1743 | arch_swap_restore(swap, page); | |
1744 | ||
c5bf121e VRP |
1745 | if (shmem_should_replace_page(page, gfp)) { |
1746 | error = shmem_replace_page(&page, gfp, info, index); | |
1747 | if (error) | |
1748 | goto failed; | |
1749 | } | |
1750 | ||
14235ab3 | 1751 | error = shmem_add_to_page_cache(page, mapping, index, |
3fea5a49 JW |
1752 | swp_to_radix_entry(swap), gfp, |
1753 | charge_mm); | |
1754 | if (error) | |
14235ab3 | 1755 | goto failed; |
c5bf121e VRP |
1756 | |
1757 | spin_lock_irq(&info->lock); | |
1758 | info->swapped--; | |
1759 | shmem_recalc_inode(inode); | |
1760 | spin_unlock_irq(&info->lock); | |
1761 | ||
1762 | if (sgp == SGP_WRITE) | |
1763 | mark_page_accessed(page); | |
1764 | ||
1765 | delete_from_swap_cache(page); | |
1766 | set_page_dirty(page); | |
1767 | swap_free(swap); | |
1768 | ||
1769 | *pagep = page; | |
1770 | return 0; | |
1771 | failed: | |
1772 | if (!shmem_confirm_swap(mapping, index, swap)) | |
1773 | error = -EEXIST; | |
1774 | unlock: | |
1775 | if (page) { | |
1776 | unlock_page(page); | |
1777 | put_page(page); | |
1778 | } | |
1779 | ||
1780 | return error; | |
1781 | } | |
1782 | ||
1da177e4 | 1783 | /* |
68da9f05 | 1784 | * shmem_getpage_gfp - find page in cache, or get from swap, or allocate |
1da177e4 LT |
1785 | * |
1786 | * If we allocate a new one we do not mark it dirty. That's up to the | |
1787 | * vm. If we swap it in we mark it dirty since we also free the swap | |
9e18eb29 ALC |
1788 | * entry since a page cannot live in both the swap and page cache. |
1789 | * | |
28eb3c80 | 1790 | * vmf and fault_type are only supplied by shmem_fault: |
9e18eb29 | 1791 | * otherwise they are NULL. |
1da177e4 | 1792 | */ |
41ffe5d5 | 1793 | static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, |
9e18eb29 | 1794 | struct page **pagep, enum sgp_type sgp, gfp_t gfp, |
2b740303 SJ |
1795 | struct vm_area_struct *vma, struct vm_fault *vmf, |
1796 | vm_fault_t *fault_type) | |
1da177e4 LT |
1797 | { |
1798 | struct address_space *mapping = inode->i_mapping; | |
23f919d4 | 1799 | struct shmem_inode_info *info = SHMEM_I(inode); |
1da177e4 | 1800 | struct shmem_sb_info *sbinfo; |
9e18eb29 | 1801 | struct mm_struct *charge_mm; |
27ab7006 | 1802 | struct page *page; |
657e3038 | 1803 | enum sgp_type sgp_huge = sgp; |
800d8c63 | 1804 | pgoff_t hindex = index; |
1da177e4 | 1805 | int error; |
54af6042 | 1806 | int once = 0; |
1635f6a7 | 1807 | int alloced = 0; |
1da177e4 | 1808 | |
09cbfeaf | 1809 | if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT)) |
1da177e4 | 1810 | return -EFBIG; |
657e3038 KS |
1811 | if (sgp == SGP_NOHUGE || sgp == SGP_HUGE) |
1812 | sgp = SGP_CACHE; | |
1da177e4 | 1813 | repeat: |
c5bf121e VRP |
1814 | if (sgp <= SGP_CACHE && |
1815 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { | |
1816 | return -EINVAL; | |
1817 | } | |
1818 | ||
1819 | sbinfo = SHMEM_SB(inode->i_sb); | |
1820 | charge_mm = vma ? vma->vm_mm : current->mm; | |
1821 | ||
0cd6144a | 1822 | page = find_lock_entry(mapping, index); |
3159f943 | 1823 | if (xa_is_value(page)) { |
c5bf121e VRP |
1824 | error = shmem_swapin_page(inode, index, &page, |
1825 | sgp, gfp, vma, fault_type); | |
1826 | if (error == -EEXIST) | |
1827 | goto repeat; | |
54af6042 | 1828 | |
c5bf121e VRP |
1829 | *pagep = page; |
1830 | return error; | |
54af6042 HD |
1831 | } |
1832 | ||
63ec1973 MWO |
1833 | if (page) |
1834 | hindex = page->index; | |
66d2f4d2 HD |
1835 | if (page && sgp == SGP_WRITE) |
1836 | mark_page_accessed(page); | |
1837 | ||
1635f6a7 HD |
1838 | /* fallocated page? */ |
1839 | if (page && !PageUptodate(page)) { | |
1840 | if (sgp != SGP_READ) | |
1841 | goto clear; | |
1842 | unlock_page(page); | |
09cbfeaf | 1843 | put_page(page); |
1635f6a7 | 1844 | page = NULL; |
63ec1973 | 1845 | hindex = index; |
1635f6a7 | 1846 | } |
63ec1973 MWO |
1847 | if (page || sgp == SGP_READ) |
1848 | goto out; | |
27ab7006 HD |
1849 | |
1850 | /* | |
54af6042 HD |
1851 | * Fast cache lookup did not find it: |
1852 | * bring it back from swap or allocate. | |
27ab7006 | 1853 | */ |
54af6042 | 1854 | |
c5bf121e VRP |
1855 | if (vma && userfaultfd_missing(vma)) { |
1856 | *fault_type = handle_userfault(vmf, VM_UFFD_MISSING); | |
1857 | return 0; | |
1858 | } | |
cfda0526 | 1859 | |
c5bf121e | 1860 | /* shmem_symlink() */ |
30e6a51d | 1861 | if (!shmem_mapping(mapping)) |
c5bf121e VRP |
1862 | goto alloc_nohuge; |
1863 | if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE) | |
1864 | goto alloc_nohuge; | |
1865 | if (shmem_huge == SHMEM_HUGE_FORCE) | |
1866 | goto alloc_huge; | |
1867 | switch (sbinfo->huge) { | |
c5bf121e VRP |
1868 | case SHMEM_HUGE_NEVER: |
1869 | goto alloc_nohuge; | |
27d80fa2 KC |
1870 | case SHMEM_HUGE_WITHIN_SIZE: { |
1871 | loff_t i_size; | |
1872 | pgoff_t off; | |
1873 | ||
c5bf121e VRP |
1874 | off = round_up(index, HPAGE_PMD_NR); |
1875 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
1876 | if (i_size >= HPAGE_PMD_SIZE && | |
1877 | i_size >> PAGE_SHIFT >= off) | |
800d8c63 | 1878 | goto alloc_huge; |
27d80fa2 KC |
1879 | |
1880 | fallthrough; | |
1881 | } | |
c5bf121e VRP |
1882 | case SHMEM_HUGE_ADVISE: |
1883 | if (sgp_huge == SGP_HUGE) | |
1884 | goto alloc_huge; | |
1885 | /* TODO: implement fadvise() hints */ | |
1886 | goto alloc_nohuge; | |
1887 | } | |
1da177e4 | 1888 | |
800d8c63 | 1889 | alloc_huge: |
c5bf121e VRP |
1890 | page = shmem_alloc_and_acct_page(gfp, inode, index, true); |
1891 | if (IS_ERR(page)) { | |
1892 | alloc_nohuge: | |
1893 | page = shmem_alloc_and_acct_page(gfp, inode, | |
1894 | index, false); | |
1895 | } | |
1896 | if (IS_ERR(page)) { | |
1897 | int retry = 5; | |
800d8c63 | 1898 | |
c5bf121e VRP |
1899 | error = PTR_ERR(page); |
1900 | page = NULL; | |
1901 | if (error != -ENOSPC) | |
1902 | goto unlock; | |
1903 | /* | |
1904 | * Try to reclaim some space by splitting a huge page | |
1905 | * beyond i_size on the filesystem. | |
1906 | */ | |
1907 | while (retry--) { | |
1908 | int ret; | |
66d2f4d2 | 1909 | |
c5bf121e VRP |
1910 | ret = shmem_unused_huge_shrink(sbinfo, NULL, 1); |
1911 | if (ret == SHRINK_STOP) | |
1912 | break; | |
1913 | if (ret) | |
1914 | goto alloc_nohuge; | |
b065b432 | 1915 | } |
c5bf121e VRP |
1916 | goto unlock; |
1917 | } | |
54af6042 | 1918 | |
c5bf121e VRP |
1919 | if (PageTransHuge(page)) |
1920 | hindex = round_down(index, HPAGE_PMD_NR); | |
1921 | else | |
1922 | hindex = index; | |
54af6042 | 1923 | |
c5bf121e VRP |
1924 | if (sgp == SGP_WRITE) |
1925 | __SetPageReferenced(page); | |
1926 | ||
c5bf121e | 1927 | error = shmem_add_to_page_cache(page, mapping, hindex, |
3fea5a49 JW |
1928 | NULL, gfp & GFP_RECLAIM_MASK, |
1929 | charge_mm); | |
1930 | if (error) | |
c5bf121e | 1931 | goto unacct; |
6058eaec | 1932 | lru_cache_add(page); |
779750d2 | 1933 | |
c5bf121e | 1934 | spin_lock_irq(&info->lock); |
d8c6546b | 1935 | info->alloced += compound_nr(page); |
c5bf121e VRP |
1936 | inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page); |
1937 | shmem_recalc_inode(inode); | |
1938 | spin_unlock_irq(&info->lock); | |
1939 | alloced = true; | |
1940 | ||
1941 | if (PageTransHuge(page) && | |
1942 | DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) < | |
1943 | hindex + HPAGE_PMD_NR - 1) { | |
ec9516fb | 1944 | /* |
c5bf121e VRP |
1945 | * Part of the huge page is beyond i_size: subject |
1946 | * to shrink under memory pressure. | |
1635f6a7 | 1947 | */ |
c5bf121e | 1948 | spin_lock(&sbinfo->shrinklist_lock); |
1635f6a7 | 1949 | /* |
c5bf121e VRP |
1950 | * _careful to defend against unlocked access to |
1951 | * ->shrink_list in shmem_unused_huge_shrink() | |
ec9516fb | 1952 | */ |
c5bf121e VRP |
1953 | if (list_empty_careful(&info->shrinklist)) { |
1954 | list_add_tail(&info->shrinklist, | |
1955 | &sbinfo->shrinklist); | |
1956 | sbinfo->shrinklist_len++; | |
1957 | } | |
1958 | spin_unlock(&sbinfo->shrinklist_lock); | |
1959 | } | |
800d8c63 | 1960 | |
c5bf121e VRP |
1961 | /* |
1962 | * Let SGP_FALLOC use the SGP_WRITE optimization on a new page. | |
1963 | */ | |
1964 | if (sgp == SGP_FALLOC) | |
1965 | sgp = SGP_WRITE; | |
1966 | clear: | |
1967 | /* | |
1968 | * Let SGP_WRITE caller clear ends if write does not fill page; | |
1969 | * but SGP_FALLOC on a page fallocated earlier must initialize | |
1970 | * it now, lest undo on failure cancel our earlier guarantee. | |
1971 | */ | |
1972 | if (sgp != SGP_WRITE && !PageUptodate(page)) { | |
c5bf121e VRP |
1973 | int i; |
1974 | ||
63ec1973 MWO |
1975 | for (i = 0; i < compound_nr(page); i++) { |
1976 | clear_highpage(page + i); | |
1977 | flush_dcache_page(page + i); | |
ec9516fb | 1978 | } |
63ec1973 | 1979 | SetPageUptodate(page); |
1da177e4 | 1980 | } |
bde05d1c | 1981 | |
54af6042 | 1982 | /* Perhaps the file has been truncated since we checked */ |
75edd345 | 1983 | if (sgp <= SGP_CACHE && |
09cbfeaf | 1984 | ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) { |
267a4c76 HD |
1985 | if (alloced) { |
1986 | ClearPageDirty(page); | |
1987 | delete_from_page_cache(page); | |
4595ef88 | 1988 | spin_lock_irq(&info->lock); |
267a4c76 | 1989 | shmem_recalc_inode(inode); |
4595ef88 | 1990 | spin_unlock_irq(&info->lock); |
267a4c76 | 1991 | } |
54af6042 | 1992 | error = -EINVAL; |
267a4c76 | 1993 | goto unlock; |
e83c32e8 | 1994 | } |
63ec1973 | 1995 | out: |
800d8c63 | 1996 | *pagep = page + index - hindex; |
54af6042 | 1997 | return 0; |
1da177e4 | 1998 | |
59a16ead | 1999 | /* |
54af6042 | 2000 | * Error recovery. |
59a16ead | 2001 | */ |
54af6042 | 2002 | unacct: |
d8c6546b | 2003 | shmem_inode_unacct_blocks(inode, compound_nr(page)); |
800d8c63 KS |
2004 | |
2005 | if (PageTransHuge(page)) { | |
2006 | unlock_page(page); | |
2007 | put_page(page); | |
2008 | goto alloc_nohuge; | |
2009 | } | |
d1899228 | 2010 | unlock: |
27ab7006 | 2011 | if (page) { |
54af6042 | 2012 | unlock_page(page); |
09cbfeaf | 2013 | put_page(page); |
54af6042 HD |
2014 | } |
2015 | if (error == -ENOSPC && !once++) { | |
4595ef88 | 2016 | spin_lock_irq(&info->lock); |
54af6042 | 2017 | shmem_recalc_inode(inode); |
4595ef88 | 2018 | spin_unlock_irq(&info->lock); |
27ab7006 | 2019 | goto repeat; |
ff36b801 | 2020 | } |
7f4446ee | 2021 | if (error == -EEXIST) |
54af6042 HD |
2022 | goto repeat; |
2023 | return error; | |
1da177e4 LT |
2024 | } |
2025 | ||
10d20bd2 LT |
2026 | /* |
2027 | * This is like autoremove_wake_function, but it removes the wait queue | |
2028 | * entry unconditionally - even if something else had already woken the | |
2029 | * target. | |
2030 | */ | |
ac6424b9 | 2031 | static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
10d20bd2 LT |
2032 | { |
2033 | int ret = default_wake_function(wait, mode, sync, key); | |
2055da97 | 2034 | list_del_init(&wait->entry); |
10d20bd2 LT |
2035 | return ret; |
2036 | } | |
2037 | ||
20acce67 | 2038 | static vm_fault_t shmem_fault(struct vm_fault *vmf) |
1da177e4 | 2039 | { |
11bac800 | 2040 | struct vm_area_struct *vma = vmf->vma; |
496ad9aa | 2041 | struct inode *inode = file_inode(vma->vm_file); |
9e18eb29 | 2042 | gfp_t gfp = mapping_gfp_mask(inode->i_mapping); |
657e3038 | 2043 | enum sgp_type sgp; |
20acce67 SJ |
2044 | int err; |
2045 | vm_fault_t ret = VM_FAULT_LOCKED; | |
1da177e4 | 2046 | |
f00cdc6d HD |
2047 | /* |
2048 | * Trinity finds that probing a hole which tmpfs is punching can | |
2049 | * prevent the hole-punch from ever completing: which in turn | |
2050 | * locks writers out with its hold on i_mutex. So refrain from | |
8e205f77 HD |
2051 | * faulting pages into the hole while it's being punched. Although |
2052 | * shmem_undo_range() does remove the additions, it may be unable to | |
2053 | * keep up, as each new page needs its own unmap_mapping_range() call, | |
2054 | * and the i_mmap tree grows ever slower to scan if new vmas are added. | |
2055 | * | |
2056 | * It does not matter if we sometimes reach this check just before the | |
2057 | * hole-punch begins, so that one fault then races with the punch: | |
2058 | * we just need to make racing faults a rare case. | |
2059 | * | |
2060 | * The implementation below would be much simpler if we just used a | |
2061 | * standard mutex or completion: but we cannot take i_mutex in fault, | |
2062 | * and bloating every shmem inode for this unlikely case would be sad. | |
f00cdc6d HD |
2063 | */ |
2064 | if (unlikely(inode->i_private)) { | |
2065 | struct shmem_falloc *shmem_falloc; | |
2066 | ||
2067 | spin_lock(&inode->i_lock); | |
2068 | shmem_falloc = inode->i_private; | |
8e205f77 HD |
2069 | if (shmem_falloc && |
2070 | shmem_falloc->waitq && | |
2071 | vmf->pgoff >= shmem_falloc->start && | |
2072 | vmf->pgoff < shmem_falloc->next) { | |
8897c1b1 | 2073 | struct file *fpin; |
8e205f77 | 2074 | wait_queue_head_t *shmem_falloc_waitq; |
10d20bd2 | 2075 | DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function); |
8e205f77 HD |
2076 | |
2077 | ret = VM_FAULT_NOPAGE; | |
8897c1b1 KS |
2078 | fpin = maybe_unlock_mmap_for_io(vmf, NULL); |
2079 | if (fpin) | |
8e205f77 | 2080 | ret = VM_FAULT_RETRY; |
8e205f77 HD |
2081 | |
2082 | shmem_falloc_waitq = shmem_falloc->waitq; | |
2083 | prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait, | |
2084 | TASK_UNINTERRUPTIBLE); | |
2085 | spin_unlock(&inode->i_lock); | |
2086 | schedule(); | |
2087 | ||
2088 | /* | |
2089 | * shmem_falloc_waitq points into the shmem_fallocate() | |
2090 | * stack of the hole-punching task: shmem_falloc_waitq | |
2091 | * is usually invalid by the time we reach here, but | |
2092 | * finish_wait() does not dereference it in that case; | |
2093 | * though i_lock needed lest racing with wake_up_all(). | |
2094 | */ | |
2095 | spin_lock(&inode->i_lock); | |
2096 | finish_wait(shmem_falloc_waitq, &shmem_fault_wait); | |
2097 | spin_unlock(&inode->i_lock); | |
8897c1b1 KS |
2098 | |
2099 | if (fpin) | |
2100 | fput(fpin); | |
8e205f77 | 2101 | return ret; |
f00cdc6d | 2102 | } |
8e205f77 | 2103 | spin_unlock(&inode->i_lock); |
f00cdc6d HD |
2104 | } |
2105 | ||
657e3038 | 2106 | sgp = SGP_CACHE; |
18600332 MH |
2107 | |
2108 | if ((vma->vm_flags & VM_NOHUGEPAGE) || | |
2109 | test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) | |
657e3038 | 2110 | sgp = SGP_NOHUGE; |
18600332 MH |
2111 | else if (vma->vm_flags & VM_HUGEPAGE) |
2112 | sgp = SGP_HUGE; | |
657e3038 | 2113 | |
20acce67 | 2114 | err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp, |
cfda0526 | 2115 | gfp, vma, vmf, &ret); |
20acce67 SJ |
2116 | if (err) |
2117 | return vmf_error(err); | |
68da9f05 | 2118 | return ret; |
1da177e4 LT |
2119 | } |
2120 | ||
c01d5b30 HD |
2121 | unsigned long shmem_get_unmapped_area(struct file *file, |
2122 | unsigned long uaddr, unsigned long len, | |
2123 | unsigned long pgoff, unsigned long flags) | |
2124 | { | |
2125 | unsigned long (*get_area)(struct file *, | |
2126 | unsigned long, unsigned long, unsigned long, unsigned long); | |
2127 | unsigned long addr; | |
2128 | unsigned long offset; | |
2129 | unsigned long inflated_len; | |
2130 | unsigned long inflated_addr; | |
2131 | unsigned long inflated_offset; | |
2132 | ||
2133 | if (len > TASK_SIZE) | |
2134 | return -ENOMEM; | |
2135 | ||
2136 | get_area = current->mm->get_unmapped_area; | |
2137 | addr = get_area(file, uaddr, len, pgoff, flags); | |
2138 | ||
396bcc52 | 2139 | if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) |
c01d5b30 HD |
2140 | return addr; |
2141 | if (IS_ERR_VALUE(addr)) | |
2142 | return addr; | |
2143 | if (addr & ~PAGE_MASK) | |
2144 | return addr; | |
2145 | if (addr > TASK_SIZE - len) | |
2146 | return addr; | |
2147 | ||
2148 | if (shmem_huge == SHMEM_HUGE_DENY) | |
2149 | return addr; | |
2150 | if (len < HPAGE_PMD_SIZE) | |
2151 | return addr; | |
2152 | if (flags & MAP_FIXED) | |
2153 | return addr; | |
2154 | /* | |
2155 | * Our priority is to support MAP_SHARED mapped hugely; | |
2156 | * and support MAP_PRIVATE mapped hugely too, until it is COWed. | |
99158997 KS |
2157 | * But if caller specified an address hint and we allocated area there |
2158 | * successfully, respect that as before. | |
c01d5b30 | 2159 | */ |
99158997 | 2160 | if (uaddr == addr) |
c01d5b30 HD |
2161 | return addr; |
2162 | ||
2163 | if (shmem_huge != SHMEM_HUGE_FORCE) { | |
2164 | struct super_block *sb; | |
2165 | ||
2166 | if (file) { | |
2167 | VM_BUG_ON(file->f_op != &shmem_file_operations); | |
2168 | sb = file_inode(file)->i_sb; | |
2169 | } else { | |
2170 | /* | |
2171 | * Called directly from mm/mmap.c, or drivers/char/mem.c | |
2172 | * for "/dev/zero", to create a shared anonymous object. | |
2173 | */ | |
2174 | if (IS_ERR(shm_mnt)) | |
2175 | return addr; | |
2176 | sb = shm_mnt->mnt_sb; | |
2177 | } | |
3089bf61 | 2178 | if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER) |
c01d5b30 HD |
2179 | return addr; |
2180 | } | |
2181 | ||
2182 | offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1); | |
2183 | if (offset && offset + len < 2 * HPAGE_PMD_SIZE) | |
2184 | return addr; | |
2185 | if ((addr & (HPAGE_PMD_SIZE-1)) == offset) | |
2186 | return addr; | |
2187 | ||
2188 | inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE; | |
2189 | if (inflated_len > TASK_SIZE) | |
2190 | return addr; | |
2191 | if (inflated_len < len) | |
2192 | return addr; | |
2193 | ||
99158997 | 2194 | inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags); |
c01d5b30 HD |
2195 | if (IS_ERR_VALUE(inflated_addr)) |
2196 | return addr; | |
2197 | if (inflated_addr & ~PAGE_MASK) | |
2198 | return addr; | |
2199 | ||
2200 | inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1); | |
2201 | inflated_addr += offset - inflated_offset; | |
2202 | if (inflated_offset > offset) | |
2203 | inflated_addr += HPAGE_PMD_SIZE; | |
2204 | ||
2205 | if (inflated_addr > TASK_SIZE - len) | |
2206 | return addr; | |
2207 | return inflated_addr; | |
2208 | } | |
2209 | ||
1da177e4 | 2210 | #ifdef CONFIG_NUMA |
41ffe5d5 | 2211 | static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) |
1da177e4 | 2212 | { |
496ad9aa | 2213 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2214 | return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); |
1da177e4 LT |
2215 | } |
2216 | ||
d8dc74f2 AB |
2217 | static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, |
2218 | unsigned long addr) | |
1da177e4 | 2219 | { |
496ad9aa | 2220 | struct inode *inode = file_inode(vma->vm_file); |
41ffe5d5 | 2221 | pgoff_t index; |
1da177e4 | 2222 | |
41ffe5d5 HD |
2223 | index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2224 | return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); | |
1da177e4 LT |
2225 | } |
2226 | #endif | |
2227 | ||
2228 | int shmem_lock(struct file *file, int lock, struct user_struct *user) | |
2229 | { | |
496ad9aa | 2230 | struct inode *inode = file_inode(file); |
1da177e4 LT |
2231 | struct shmem_inode_info *info = SHMEM_I(inode); |
2232 | int retval = -ENOMEM; | |
2233 | ||
ea0dfeb4 HD |
2234 | /* |
2235 | * What serializes the accesses to info->flags? | |
2236 | * ipc_lock_object() when called from shmctl_do_lock(), | |
2237 | * no serialization needed when called from shm_destroy(). | |
2238 | */ | |
1da177e4 LT |
2239 | if (lock && !(info->flags & VM_LOCKED)) { |
2240 | if (!user_shm_lock(inode->i_size, user)) | |
2241 | goto out_nomem; | |
2242 | info->flags |= VM_LOCKED; | |
89e004ea | 2243 | mapping_set_unevictable(file->f_mapping); |
1da177e4 LT |
2244 | } |
2245 | if (!lock && (info->flags & VM_LOCKED) && user) { | |
2246 | user_shm_unlock(inode->i_size, user); | |
2247 | info->flags &= ~VM_LOCKED; | |
89e004ea | 2248 | mapping_clear_unevictable(file->f_mapping); |
1da177e4 LT |
2249 | } |
2250 | retval = 0; | |
89e004ea | 2251 | |
1da177e4 | 2252 | out_nomem: |
1da177e4 LT |
2253 | return retval; |
2254 | } | |
2255 | ||
9b83a6a8 | 2256 | static int shmem_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 | 2257 | { |
ab3948f5 JFG |
2258 | struct shmem_inode_info *info = SHMEM_I(file_inode(file)); |
2259 | ||
2260 | if (info->seals & F_SEAL_FUTURE_WRITE) { | |
2261 | /* | |
2262 | * New PROT_WRITE and MAP_SHARED mmaps are not allowed when | |
2263 | * "future write" seal active. | |
2264 | */ | |
2265 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE)) | |
2266 | return -EPERM; | |
2267 | ||
2268 | /* | |
05d35110 NG |
2269 | * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as |
2270 | * MAP_SHARED and read-only, take care to not allow mprotect to | |
2271 | * revert protections on such mappings. Do this only for shared | |
2272 | * mappings. For private mappings, don't need to mask | |
2273 | * VM_MAYWRITE as we still want them to be COW-writable. | |
ab3948f5 | 2274 | */ |
05d35110 NG |
2275 | if (vma->vm_flags & VM_SHARED) |
2276 | vma->vm_flags &= ~(VM_MAYWRITE); | |
ab3948f5 JFG |
2277 | } |
2278 | ||
51b0bff2 CM |
2279 | /* arm64 - allow memory tagging on RAM-based files */ |
2280 | vma->vm_flags |= VM_MTE_ALLOWED; | |
2281 | ||
1da177e4 LT |
2282 | file_accessed(file); |
2283 | vma->vm_ops = &shmem_vm_ops; | |
396bcc52 | 2284 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && |
f3f0e1d2 KS |
2285 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
2286 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
2287 | khugepaged_enter(vma, vma->vm_flags); | |
2288 | } | |
1da177e4 LT |
2289 | return 0; |
2290 | } | |
2291 | ||
454abafe | 2292 | static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, |
09208d15 | 2293 | umode_t mode, dev_t dev, unsigned long flags) |
1da177e4 LT |
2294 | { |
2295 | struct inode *inode; | |
2296 | struct shmem_inode_info *info; | |
2297 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); | |
e809d5f0 | 2298 | ino_t ino; |
1da177e4 | 2299 | |
e809d5f0 | 2300 | if (shmem_reserve_inode(sb, &ino)) |
5b04c689 | 2301 | return NULL; |
1da177e4 LT |
2302 | |
2303 | inode = new_inode(sb); | |
2304 | if (inode) { | |
e809d5f0 | 2305 | inode->i_ino = ino; |
21cb47be | 2306 | inode_init_owner(&init_user_ns, inode, dir, mode); |
1da177e4 | 2307 | inode->i_blocks = 0; |
078cd827 | 2308 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
46c9a946 | 2309 | inode->i_generation = prandom_u32(); |
1da177e4 LT |
2310 | info = SHMEM_I(inode); |
2311 | memset(info, 0, (char *)inode - (char *)info); | |
2312 | spin_lock_init(&info->lock); | |
af53d3e9 | 2313 | atomic_set(&info->stop_eviction, 0); |
40e041a2 | 2314 | info->seals = F_SEAL_SEAL; |
0b0a0806 | 2315 | info->flags = flags & VM_NORESERVE; |
779750d2 | 2316 | INIT_LIST_HEAD(&info->shrinklist); |
1da177e4 | 2317 | INIT_LIST_HEAD(&info->swaplist); |
38f38657 | 2318 | simple_xattrs_init(&info->xattrs); |
72c04902 | 2319 | cache_no_acl(inode); |
1da177e4 LT |
2320 | |
2321 | switch (mode & S_IFMT) { | |
2322 | default: | |
39f0247d | 2323 | inode->i_op = &shmem_special_inode_operations; |
1da177e4 LT |
2324 | init_special_inode(inode, mode, dev); |
2325 | break; | |
2326 | case S_IFREG: | |
14fcc23f | 2327 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 LT |
2328 | inode->i_op = &shmem_inode_operations; |
2329 | inode->i_fop = &shmem_file_operations; | |
71fe804b LS |
2330 | mpol_shared_policy_init(&info->policy, |
2331 | shmem_get_sbmpol(sbinfo)); | |
1da177e4 LT |
2332 | break; |
2333 | case S_IFDIR: | |
d8c76e6f | 2334 | inc_nlink(inode); |
1da177e4 LT |
2335 | /* Some things misbehave if size == 0 on a directory */ |
2336 | inode->i_size = 2 * BOGO_DIRENT_SIZE; | |
2337 | inode->i_op = &shmem_dir_inode_operations; | |
2338 | inode->i_fop = &simple_dir_operations; | |
2339 | break; | |
2340 | case S_IFLNK: | |
2341 | /* | |
2342 | * Must not load anything in the rbtree, | |
2343 | * mpol_free_shared_policy will not be called. | |
2344 | */ | |
71fe804b | 2345 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
2346 | break; |
2347 | } | |
b45d71fb JFG |
2348 | |
2349 | lockdep_annotate_inode_mutex_key(inode); | |
5b04c689 PE |
2350 | } else |
2351 | shmem_free_inode(sb); | |
1da177e4 LT |
2352 | return inode; |
2353 | } | |
2354 | ||
8d103963 MR |
2355 | static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm, |
2356 | pmd_t *dst_pmd, | |
2357 | struct vm_area_struct *dst_vma, | |
2358 | unsigned long dst_addr, | |
2359 | unsigned long src_addr, | |
2360 | bool zeropage, | |
2361 | struct page **pagep) | |
4c27fe4c MR |
2362 | { |
2363 | struct inode *inode = file_inode(dst_vma->vm_file); | |
2364 | struct shmem_inode_info *info = SHMEM_I(inode); | |
4c27fe4c MR |
2365 | struct address_space *mapping = inode->i_mapping; |
2366 | gfp_t gfp = mapping_gfp_mask(mapping); | |
2367 | pgoff_t pgoff = linear_page_index(dst_vma, dst_addr); | |
4c27fe4c MR |
2368 | spinlock_t *ptl; |
2369 | void *page_kaddr; | |
2370 | struct page *page; | |
2371 | pte_t _dst_pte, *dst_pte; | |
2372 | int ret; | |
e2a50c1f | 2373 | pgoff_t offset, max_off; |
4c27fe4c | 2374 | |
cb658a45 | 2375 | ret = -ENOMEM; |
0f079694 | 2376 | if (!shmem_inode_acct_block(inode, 1)) |
cb658a45 | 2377 | goto out; |
4c27fe4c | 2378 | |
cb658a45 | 2379 | if (!*pagep) { |
4c27fe4c MR |
2380 | page = shmem_alloc_page(gfp, info, pgoff); |
2381 | if (!page) | |
0f079694 | 2382 | goto out_unacct_blocks; |
4c27fe4c | 2383 | |
8d103963 MR |
2384 | if (!zeropage) { /* mcopy_atomic */ |
2385 | page_kaddr = kmap_atomic(page); | |
2386 | ret = copy_from_user(page_kaddr, | |
2387 | (const void __user *)src_addr, | |
2388 | PAGE_SIZE); | |
2389 | kunmap_atomic(page_kaddr); | |
2390 | ||
c1e8d7c6 | 2391 | /* fallback to copy_from_user outside mmap_lock */ |
8d103963 MR |
2392 | if (unlikely(ret)) { |
2393 | *pagep = page; | |
2394 | shmem_inode_unacct_blocks(inode, 1); | |
2395 | /* don't free the page */ | |
9e368259 | 2396 | return -ENOENT; |
8d103963 MR |
2397 | } |
2398 | } else { /* mfill_zeropage_atomic */ | |
2399 | clear_highpage(page); | |
4c27fe4c MR |
2400 | } |
2401 | } else { | |
2402 | page = *pagep; | |
2403 | *pagep = NULL; | |
2404 | } | |
2405 | ||
9cc90c66 AA |
2406 | VM_BUG_ON(PageLocked(page) || PageSwapBacked(page)); |
2407 | __SetPageLocked(page); | |
2408 | __SetPageSwapBacked(page); | |
a425d358 | 2409 | __SetPageUptodate(page); |
9cc90c66 | 2410 | |
e2a50c1f AA |
2411 | ret = -EFAULT; |
2412 | offset = linear_page_index(dst_vma, dst_addr); | |
2413 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); | |
2414 | if (unlikely(offset >= max_off)) | |
2415 | goto out_release; | |
2416 | ||
552446a4 | 2417 | ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL, |
3fea5a49 | 2418 | gfp & GFP_RECLAIM_MASK, dst_mm); |
4c27fe4c | 2419 | if (ret) |
3fea5a49 | 2420 | goto out_release; |
4c27fe4c MR |
2421 | |
2422 | _dst_pte = mk_pte(page, dst_vma->vm_page_prot); | |
2423 | if (dst_vma->vm_flags & VM_WRITE) | |
2424 | _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte)); | |
dcf7fe9d AA |
2425 | else { |
2426 | /* | |
2427 | * We don't set the pte dirty if the vma has no | |
2428 | * VM_WRITE permission, so mark the page dirty or it | |
2429 | * could be freed from under us. We could do it | |
2430 | * unconditionally before unlock_page(), but doing it | |
2431 | * only if VM_WRITE is not set is faster. | |
2432 | */ | |
2433 | set_page_dirty(page); | |
2434 | } | |
4c27fe4c | 2435 | |
4c27fe4c | 2436 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); |
e2a50c1f AA |
2437 | |
2438 | ret = -EFAULT; | |
2439 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); | |
2440 | if (unlikely(offset >= max_off)) | |
3fea5a49 | 2441 | goto out_release_unlock; |
e2a50c1f AA |
2442 | |
2443 | ret = -EEXIST; | |
4c27fe4c | 2444 | if (!pte_none(*dst_pte)) |
3fea5a49 | 2445 | goto out_release_unlock; |
4c27fe4c | 2446 | |
6058eaec | 2447 | lru_cache_add(page); |
4c27fe4c | 2448 | |
94b7cc01 | 2449 | spin_lock_irq(&info->lock); |
4c27fe4c MR |
2450 | info->alloced++; |
2451 | inode->i_blocks += BLOCKS_PER_PAGE; | |
2452 | shmem_recalc_inode(inode); | |
94b7cc01 | 2453 | spin_unlock_irq(&info->lock); |
4c27fe4c MR |
2454 | |
2455 | inc_mm_counter(dst_mm, mm_counter_file(page)); | |
2456 | page_add_file_rmap(page, false); | |
2457 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
2458 | ||
2459 | /* No need to invalidate - it was non-present before */ | |
2460 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
4c27fe4c | 2461 | pte_unmap_unlock(dst_pte, ptl); |
e2a50c1f | 2462 | unlock_page(page); |
4c27fe4c MR |
2463 | ret = 0; |
2464 | out: | |
2465 | return ret; | |
3fea5a49 | 2466 | out_release_unlock: |
4c27fe4c | 2467 | pte_unmap_unlock(dst_pte, ptl); |
dcf7fe9d | 2468 | ClearPageDirty(page); |
e2a50c1f | 2469 | delete_from_page_cache(page); |
4c27fe4c | 2470 | out_release: |
9cc90c66 | 2471 | unlock_page(page); |
4c27fe4c | 2472 | put_page(page); |
4c27fe4c | 2473 | out_unacct_blocks: |
0f079694 | 2474 | shmem_inode_unacct_blocks(inode, 1); |
4c27fe4c MR |
2475 | goto out; |
2476 | } | |
2477 | ||
8d103963 MR |
2478 | int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm, |
2479 | pmd_t *dst_pmd, | |
2480 | struct vm_area_struct *dst_vma, | |
2481 | unsigned long dst_addr, | |
2482 | unsigned long src_addr, | |
2483 | struct page **pagep) | |
2484 | { | |
2485 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2486 | dst_addr, src_addr, false, pagep); | |
2487 | } | |
2488 | ||
2489 | int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm, | |
2490 | pmd_t *dst_pmd, | |
2491 | struct vm_area_struct *dst_vma, | |
2492 | unsigned long dst_addr) | |
2493 | { | |
2494 | struct page *page = NULL; | |
2495 | ||
2496 | return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
2497 | dst_addr, 0, true, &page); | |
2498 | } | |
2499 | ||
1da177e4 | 2500 | #ifdef CONFIG_TMPFS |
92e1d5be | 2501 | static const struct inode_operations shmem_symlink_inode_operations; |
69f07ec9 | 2502 | static const struct inode_operations shmem_short_symlink_operations; |
1da177e4 | 2503 | |
6d9d88d0 JS |
2504 | #ifdef CONFIG_TMPFS_XATTR |
2505 | static int shmem_initxattrs(struct inode *, const struct xattr *, void *); | |
2506 | #else | |
2507 | #define shmem_initxattrs NULL | |
2508 | #endif | |
2509 | ||
1da177e4 | 2510 | static int |
800d15a5 NP |
2511 | shmem_write_begin(struct file *file, struct address_space *mapping, |
2512 | loff_t pos, unsigned len, unsigned flags, | |
2513 | struct page **pagep, void **fsdata) | |
1da177e4 | 2514 | { |
800d15a5 | 2515 | struct inode *inode = mapping->host; |
40e041a2 | 2516 | struct shmem_inode_info *info = SHMEM_I(inode); |
09cbfeaf | 2517 | pgoff_t index = pos >> PAGE_SHIFT; |
40e041a2 DH |
2518 | |
2519 | /* i_mutex is held by caller */ | |
ab3948f5 JFG |
2520 | if (unlikely(info->seals & (F_SEAL_GROW | |
2521 | F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) { | |
2522 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) | |
40e041a2 DH |
2523 | return -EPERM; |
2524 | if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size) | |
2525 | return -EPERM; | |
2526 | } | |
2527 | ||
9e18eb29 | 2528 | return shmem_getpage(inode, index, pagep, SGP_WRITE); |
800d15a5 NP |
2529 | } |
2530 | ||
2531 | static int | |
2532 | shmem_write_end(struct file *file, struct address_space *mapping, | |
2533 | loff_t pos, unsigned len, unsigned copied, | |
2534 | struct page *page, void *fsdata) | |
2535 | { | |
2536 | struct inode *inode = mapping->host; | |
2537 | ||
d3602444 HD |
2538 | if (pos + copied > inode->i_size) |
2539 | i_size_write(inode, pos + copied); | |
2540 | ||
ec9516fb | 2541 | if (!PageUptodate(page)) { |
800d8c63 KS |
2542 | struct page *head = compound_head(page); |
2543 | if (PageTransCompound(page)) { | |
2544 | int i; | |
2545 | ||
2546 | for (i = 0; i < HPAGE_PMD_NR; i++) { | |
2547 | if (head + i == page) | |
2548 | continue; | |
2549 | clear_highpage(head + i); | |
2550 | flush_dcache_page(head + i); | |
2551 | } | |
2552 | } | |
09cbfeaf KS |
2553 | if (copied < PAGE_SIZE) { |
2554 | unsigned from = pos & (PAGE_SIZE - 1); | |
ec9516fb | 2555 | zero_user_segments(page, 0, from, |
09cbfeaf | 2556 | from + copied, PAGE_SIZE); |
ec9516fb | 2557 | } |
800d8c63 | 2558 | SetPageUptodate(head); |
ec9516fb | 2559 | } |
800d15a5 | 2560 | set_page_dirty(page); |
6746aff7 | 2561 | unlock_page(page); |
09cbfeaf | 2562 | put_page(page); |
800d15a5 | 2563 | |
800d15a5 | 2564 | return copied; |
1da177e4 LT |
2565 | } |
2566 | ||
2ba5bbed | 2567 | static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 2568 | { |
6e58e79d AV |
2569 | struct file *file = iocb->ki_filp; |
2570 | struct inode *inode = file_inode(file); | |
1da177e4 | 2571 | struct address_space *mapping = inode->i_mapping; |
41ffe5d5 HD |
2572 | pgoff_t index; |
2573 | unsigned long offset; | |
a0ee5ec5 | 2574 | enum sgp_type sgp = SGP_READ; |
f7c1d074 | 2575 | int error = 0; |
cb66a7a1 | 2576 | ssize_t retval = 0; |
6e58e79d | 2577 | loff_t *ppos = &iocb->ki_pos; |
a0ee5ec5 HD |
2578 | |
2579 | /* | |
2580 | * Might this read be for a stacking filesystem? Then when reading | |
2581 | * holes of a sparse file, we actually need to allocate those pages, | |
2582 | * and even mark them dirty, so it cannot exceed the max_blocks limit. | |
2583 | */ | |
777eda2c | 2584 | if (!iter_is_iovec(to)) |
75edd345 | 2585 | sgp = SGP_CACHE; |
1da177e4 | 2586 | |
09cbfeaf KS |
2587 | index = *ppos >> PAGE_SHIFT; |
2588 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 LT |
2589 | |
2590 | for (;;) { | |
2591 | struct page *page = NULL; | |
41ffe5d5 HD |
2592 | pgoff_t end_index; |
2593 | unsigned long nr, ret; | |
1da177e4 LT |
2594 | loff_t i_size = i_size_read(inode); |
2595 | ||
09cbfeaf | 2596 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
2597 | if (index > end_index) |
2598 | break; | |
2599 | if (index == end_index) { | |
09cbfeaf | 2600 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2601 | if (nr <= offset) |
2602 | break; | |
2603 | } | |
2604 | ||
9e18eb29 | 2605 | error = shmem_getpage(inode, index, &page, sgp); |
6e58e79d AV |
2606 | if (error) { |
2607 | if (error == -EINVAL) | |
2608 | error = 0; | |
1da177e4 LT |
2609 | break; |
2610 | } | |
75edd345 HD |
2611 | if (page) { |
2612 | if (sgp == SGP_CACHE) | |
2613 | set_page_dirty(page); | |
d3602444 | 2614 | unlock_page(page); |
75edd345 | 2615 | } |
1da177e4 LT |
2616 | |
2617 | /* | |
2618 | * We must evaluate after, since reads (unlike writes) | |
1b1dcc1b | 2619 | * are called without i_mutex protection against truncate |
1da177e4 | 2620 | */ |
09cbfeaf | 2621 | nr = PAGE_SIZE; |
1da177e4 | 2622 | i_size = i_size_read(inode); |
09cbfeaf | 2623 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2624 | if (index == end_index) { |
09cbfeaf | 2625 | nr = i_size & ~PAGE_MASK; |
1da177e4 LT |
2626 | if (nr <= offset) { |
2627 | if (page) | |
09cbfeaf | 2628 | put_page(page); |
1da177e4 LT |
2629 | break; |
2630 | } | |
2631 | } | |
2632 | nr -= offset; | |
2633 | ||
2634 | if (page) { | |
2635 | /* | |
2636 | * If users can be writing to this page using arbitrary | |
2637 | * virtual addresses, take care about potential aliasing | |
2638 | * before reading the page on the kernel side. | |
2639 | */ | |
2640 | if (mapping_writably_mapped(mapping)) | |
2641 | flush_dcache_page(page); | |
2642 | /* | |
2643 | * Mark the page accessed if we read the beginning. | |
2644 | */ | |
2645 | if (!offset) | |
2646 | mark_page_accessed(page); | |
b5810039 | 2647 | } else { |
1da177e4 | 2648 | page = ZERO_PAGE(0); |
09cbfeaf | 2649 | get_page(page); |
b5810039 | 2650 | } |
1da177e4 LT |
2651 | |
2652 | /* | |
2653 | * Ok, we have the page, and it's up-to-date, so | |
2654 | * now we can copy it to user space... | |
1da177e4 | 2655 | */ |
2ba5bbed | 2656 | ret = copy_page_to_iter(page, offset, nr, to); |
6e58e79d | 2657 | retval += ret; |
1da177e4 | 2658 | offset += ret; |
09cbfeaf KS |
2659 | index += offset >> PAGE_SHIFT; |
2660 | offset &= ~PAGE_MASK; | |
1da177e4 | 2661 | |
09cbfeaf | 2662 | put_page(page); |
2ba5bbed | 2663 | if (!iov_iter_count(to)) |
1da177e4 | 2664 | break; |
6e58e79d AV |
2665 | if (ret < nr) { |
2666 | error = -EFAULT; | |
2667 | break; | |
2668 | } | |
1da177e4 LT |
2669 | cond_resched(); |
2670 | } | |
2671 | ||
09cbfeaf | 2672 | *ppos = ((loff_t) index << PAGE_SHIFT) + offset; |
6e58e79d AV |
2673 | file_accessed(file); |
2674 | return retval ? retval : error; | |
1da177e4 LT |
2675 | } |
2676 | ||
220f2ac9 | 2677 | /* |
7f4446ee | 2678 | * llseek SEEK_DATA or SEEK_HOLE through the page cache. |
220f2ac9 HD |
2679 | */ |
2680 | static pgoff_t shmem_seek_hole_data(struct address_space *mapping, | |
965c8e59 | 2681 | pgoff_t index, pgoff_t end, int whence) |
220f2ac9 HD |
2682 | { |
2683 | struct page *page; | |
2684 | struct pagevec pvec; | |
2685 | pgoff_t indices[PAGEVEC_SIZE]; | |
2686 | bool done = false; | |
2687 | int i; | |
2688 | ||
86679820 | 2689 | pagevec_init(&pvec); |
220f2ac9 HD |
2690 | pvec.nr = 1; /* start small: we may be there already */ |
2691 | while (!done) { | |
0cd6144a | 2692 | pvec.nr = find_get_entries(mapping, index, |
220f2ac9 HD |
2693 | pvec.nr, pvec.pages, indices); |
2694 | if (!pvec.nr) { | |
965c8e59 | 2695 | if (whence == SEEK_DATA) |
220f2ac9 HD |
2696 | index = end; |
2697 | break; | |
2698 | } | |
2699 | for (i = 0; i < pvec.nr; i++, index++) { | |
2700 | if (index < indices[i]) { | |
965c8e59 | 2701 | if (whence == SEEK_HOLE) { |
220f2ac9 HD |
2702 | done = true; |
2703 | break; | |
2704 | } | |
2705 | index = indices[i]; | |
2706 | } | |
2707 | page = pvec.pages[i]; | |
3159f943 | 2708 | if (page && !xa_is_value(page)) { |
220f2ac9 HD |
2709 | if (!PageUptodate(page)) |
2710 | page = NULL; | |
2711 | } | |
2712 | if (index >= end || | |
965c8e59 AM |
2713 | (page && whence == SEEK_DATA) || |
2714 | (!page && whence == SEEK_HOLE)) { | |
220f2ac9 HD |
2715 | done = true; |
2716 | break; | |
2717 | } | |
2718 | } | |
0cd6144a | 2719 | pagevec_remove_exceptionals(&pvec); |
220f2ac9 HD |
2720 | pagevec_release(&pvec); |
2721 | pvec.nr = PAGEVEC_SIZE; | |
2722 | cond_resched(); | |
2723 | } | |
2724 | return index; | |
2725 | } | |
2726 | ||
965c8e59 | 2727 | static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence) |
220f2ac9 HD |
2728 | { |
2729 | struct address_space *mapping = file->f_mapping; | |
2730 | struct inode *inode = mapping->host; | |
2731 | pgoff_t start, end; | |
2732 | loff_t new_offset; | |
2733 | ||
965c8e59 AM |
2734 | if (whence != SEEK_DATA && whence != SEEK_HOLE) |
2735 | return generic_file_llseek_size(file, offset, whence, | |
220f2ac9 | 2736 | MAX_LFS_FILESIZE, i_size_read(inode)); |
5955102c | 2737 | inode_lock(inode); |
220f2ac9 HD |
2738 | /* We're holding i_mutex so we can access i_size directly */ |
2739 | ||
1a413646 | 2740 | if (offset < 0 || offset >= inode->i_size) |
220f2ac9 HD |
2741 | offset = -ENXIO; |
2742 | else { | |
09cbfeaf KS |
2743 | start = offset >> PAGE_SHIFT; |
2744 | end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
965c8e59 | 2745 | new_offset = shmem_seek_hole_data(mapping, start, end, whence); |
09cbfeaf | 2746 | new_offset <<= PAGE_SHIFT; |
220f2ac9 HD |
2747 | if (new_offset > offset) { |
2748 | if (new_offset < inode->i_size) | |
2749 | offset = new_offset; | |
965c8e59 | 2750 | else if (whence == SEEK_DATA) |
220f2ac9 HD |
2751 | offset = -ENXIO; |
2752 | else | |
2753 | offset = inode->i_size; | |
2754 | } | |
2755 | } | |
2756 | ||
387aae6f HD |
2757 | if (offset >= 0) |
2758 | offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE); | |
5955102c | 2759 | inode_unlock(inode); |
220f2ac9 HD |
2760 | return offset; |
2761 | } | |
2762 | ||
83e4fa9c HD |
2763 | static long shmem_fallocate(struct file *file, int mode, loff_t offset, |
2764 | loff_t len) | |
2765 | { | |
496ad9aa | 2766 | struct inode *inode = file_inode(file); |
e2d12e22 | 2767 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); |
40e041a2 | 2768 | struct shmem_inode_info *info = SHMEM_I(inode); |
1aac1400 | 2769 | struct shmem_falloc shmem_falloc; |
e2d12e22 HD |
2770 | pgoff_t start, index, end; |
2771 | int error; | |
83e4fa9c | 2772 | |
13ace4d0 HD |
2773 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2774 | return -EOPNOTSUPP; | |
2775 | ||
5955102c | 2776 | inode_lock(inode); |
83e4fa9c HD |
2777 | |
2778 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
2779 | struct address_space *mapping = file->f_mapping; | |
2780 | loff_t unmap_start = round_up(offset, PAGE_SIZE); | |
2781 | loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1; | |
8e205f77 | 2782 | DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq); |
83e4fa9c | 2783 | |
40e041a2 | 2784 | /* protected by i_mutex */ |
ab3948f5 | 2785 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { |
40e041a2 DH |
2786 | error = -EPERM; |
2787 | goto out; | |
2788 | } | |
2789 | ||
8e205f77 | 2790 | shmem_falloc.waitq = &shmem_falloc_waitq; |
aa71ecd8 | 2791 | shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT; |
f00cdc6d HD |
2792 | shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT; |
2793 | spin_lock(&inode->i_lock); | |
2794 | inode->i_private = &shmem_falloc; | |
2795 | spin_unlock(&inode->i_lock); | |
2796 | ||
83e4fa9c HD |
2797 | if ((u64)unmap_end > (u64)unmap_start) |
2798 | unmap_mapping_range(mapping, unmap_start, | |
2799 | 1 + unmap_end - unmap_start, 0); | |
2800 | shmem_truncate_range(inode, offset, offset + len - 1); | |
2801 | /* No need to unmap again: hole-punching leaves COWed pages */ | |
8e205f77 HD |
2802 | |
2803 | spin_lock(&inode->i_lock); | |
2804 | inode->i_private = NULL; | |
2805 | wake_up_all(&shmem_falloc_waitq); | |
2055da97 | 2806 | WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head)); |
8e205f77 | 2807 | spin_unlock(&inode->i_lock); |
83e4fa9c | 2808 | error = 0; |
8e205f77 | 2809 | goto out; |
e2d12e22 HD |
2810 | } |
2811 | ||
2812 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
2813 | error = inode_newsize_ok(inode, offset + len); | |
2814 | if (error) | |
2815 | goto out; | |
2816 | ||
40e041a2 DH |
2817 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
2818 | error = -EPERM; | |
2819 | goto out; | |
2820 | } | |
2821 | ||
09cbfeaf KS |
2822 | start = offset >> PAGE_SHIFT; |
2823 | end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
e2d12e22 HD |
2824 | /* Try to avoid a swapstorm if len is impossible to satisfy */ |
2825 | if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) { | |
2826 | error = -ENOSPC; | |
2827 | goto out; | |
83e4fa9c HD |
2828 | } |
2829 | ||
8e205f77 | 2830 | shmem_falloc.waitq = NULL; |
1aac1400 HD |
2831 | shmem_falloc.start = start; |
2832 | shmem_falloc.next = start; | |
2833 | shmem_falloc.nr_falloced = 0; | |
2834 | shmem_falloc.nr_unswapped = 0; | |
2835 | spin_lock(&inode->i_lock); | |
2836 | inode->i_private = &shmem_falloc; | |
2837 | spin_unlock(&inode->i_lock); | |
2838 | ||
e2d12e22 HD |
2839 | for (index = start; index < end; index++) { |
2840 | struct page *page; | |
2841 | ||
2842 | /* | |
2843 | * Good, the fallocate(2) manpage permits EINTR: we may have | |
2844 | * been interrupted because we are using up too much memory. | |
2845 | */ | |
2846 | if (signal_pending(current)) | |
2847 | error = -EINTR; | |
1aac1400 HD |
2848 | else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced) |
2849 | error = -ENOMEM; | |
e2d12e22 | 2850 | else |
9e18eb29 | 2851 | error = shmem_getpage(inode, index, &page, SGP_FALLOC); |
e2d12e22 | 2852 | if (error) { |
1635f6a7 | 2853 | /* Remove the !PageUptodate pages we added */ |
7f556567 HD |
2854 | if (index > start) { |
2855 | shmem_undo_range(inode, | |
2856 | (loff_t)start << PAGE_SHIFT, | |
2857 | ((loff_t)index << PAGE_SHIFT) - 1, true); | |
2858 | } | |
1aac1400 | 2859 | goto undone; |
e2d12e22 HD |
2860 | } |
2861 | ||
1aac1400 HD |
2862 | /* |
2863 | * Inform shmem_writepage() how far we have reached. | |
2864 | * No need for lock or barrier: we have the page lock. | |
2865 | */ | |
2866 | shmem_falloc.next++; | |
2867 | if (!PageUptodate(page)) | |
2868 | shmem_falloc.nr_falloced++; | |
2869 | ||
e2d12e22 | 2870 | /* |
1635f6a7 HD |
2871 | * If !PageUptodate, leave it that way so that freeable pages |
2872 | * can be recognized if we need to rollback on error later. | |
2873 | * But set_page_dirty so that memory pressure will swap rather | |
e2d12e22 HD |
2874 | * than free the pages we are allocating (and SGP_CACHE pages |
2875 | * might still be clean: we now need to mark those dirty too). | |
2876 | */ | |
2877 | set_page_dirty(page); | |
2878 | unlock_page(page); | |
09cbfeaf | 2879 | put_page(page); |
e2d12e22 HD |
2880 | cond_resched(); |
2881 | } | |
2882 | ||
2883 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
2884 | i_size_write(inode, offset + len); | |
078cd827 | 2885 | inode->i_ctime = current_time(inode); |
1aac1400 HD |
2886 | undone: |
2887 | spin_lock(&inode->i_lock); | |
2888 | inode->i_private = NULL; | |
2889 | spin_unlock(&inode->i_lock); | |
e2d12e22 | 2890 | out: |
5955102c | 2891 | inode_unlock(inode); |
83e4fa9c HD |
2892 | return error; |
2893 | } | |
2894 | ||
726c3342 | 2895 | static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 2896 | { |
726c3342 | 2897 | struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); |
1da177e4 LT |
2898 | |
2899 | buf->f_type = TMPFS_MAGIC; | |
09cbfeaf | 2900 | buf->f_bsize = PAGE_SIZE; |
1da177e4 | 2901 | buf->f_namelen = NAME_MAX; |
0edd73b3 | 2902 | if (sbinfo->max_blocks) { |
1da177e4 | 2903 | buf->f_blocks = sbinfo->max_blocks; |
41ffe5d5 HD |
2904 | buf->f_bavail = |
2905 | buf->f_bfree = sbinfo->max_blocks - | |
2906 | percpu_counter_sum(&sbinfo->used_blocks); | |
0edd73b3 HD |
2907 | } |
2908 | if (sbinfo->max_inodes) { | |
1da177e4 LT |
2909 | buf->f_files = sbinfo->max_inodes; |
2910 | buf->f_ffree = sbinfo->free_inodes; | |
1da177e4 LT |
2911 | } |
2912 | /* else leave those fields 0 like simple_statfs */ | |
2913 | return 0; | |
2914 | } | |
2915 | ||
2916 | /* | |
2917 | * File creation. Allocate an inode, and we're done.. | |
2918 | */ | |
2919 | static int | |
1a67aafb | 2920 | shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 | 2921 | { |
0b0a0806 | 2922 | struct inode *inode; |
1da177e4 LT |
2923 | int error = -ENOSPC; |
2924 | ||
454abafe | 2925 | inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); |
1da177e4 | 2926 | if (inode) { |
feda821e CH |
2927 | error = simple_acl_create(dir, inode); |
2928 | if (error) | |
2929 | goto out_iput; | |
2a7dba39 | 2930 | error = security_inode_init_security(inode, dir, |
9d8f13ba | 2931 | &dentry->d_name, |
6d9d88d0 | 2932 | shmem_initxattrs, NULL); |
feda821e CH |
2933 | if (error && error != -EOPNOTSUPP) |
2934 | goto out_iput; | |
37ec43cd | 2935 | |
718deb6b | 2936 | error = 0; |
1da177e4 | 2937 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 2938 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
2939 | d_instantiate(dentry, inode); |
2940 | dget(dentry); /* Extra count - pin the dentry in core */ | |
1da177e4 LT |
2941 | } |
2942 | return error; | |
feda821e CH |
2943 | out_iput: |
2944 | iput(inode); | |
2945 | return error; | |
1da177e4 LT |
2946 | } |
2947 | ||
60545d0d AV |
2948 | static int |
2949 | shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) | |
2950 | { | |
2951 | struct inode *inode; | |
2952 | int error = -ENOSPC; | |
2953 | ||
2954 | inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE); | |
2955 | if (inode) { | |
2956 | error = security_inode_init_security(inode, dir, | |
2957 | NULL, | |
2958 | shmem_initxattrs, NULL); | |
feda821e CH |
2959 | if (error && error != -EOPNOTSUPP) |
2960 | goto out_iput; | |
2961 | error = simple_acl_create(dir, inode); | |
2962 | if (error) | |
2963 | goto out_iput; | |
60545d0d AV |
2964 | d_tmpfile(dentry, inode); |
2965 | } | |
2966 | return error; | |
feda821e CH |
2967 | out_iput: |
2968 | iput(inode); | |
2969 | return error; | |
60545d0d AV |
2970 | } |
2971 | ||
18bb1db3 | 2972 | static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
2973 | { |
2974 | int error; | |
2975 | ||
2976 | if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) | |
2977 | return error; | |
d8c76e6f | 2978 | inc_nlink(dir); |
1da177e4 LT |
2979 | return 0; |
2980 | } | |
2981 | ||
4acdaf27 | 2982 | static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 2983 | bool excl) |
1da177e4 LT |
2984 | { |
2985 | return shmem_mknod(dir, dentry, mode | S_IFREG, 0); | |
2986 | } | |
2987 | ||
2988 | /* | |
2989 | * Link a file.. | |
2990 | */ | |
2991 | static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
2992 | { | |
75c3cfa8 | 2993 | struct inode *inode = d_inode(old_dentry); |
29b00e60 | 2994 | int ret = 0; |
1da177e4 LT |
2995 | |
2996 | /* | |
2997 | * No ordinary (disk based) filesystem counts links as inodes; | |
2998 | * but each new link needs a new dentry, pinning lowmem, and | |
2999 | * tmpfs dentries cannot be pruned until they are unlinked. | |
1062af92 DW |
3000 | * But if an O_TMPFILE file is linked into the tmpfs, the |
3001 | * first link must skip that, to get the accounting right. | |
1da177e4 | 3002 | */ |
1062af92 | 3003 | if (inode->i_nlink) { |
e809d5f0 | 3004 | ret = shmem_reserve_inode(inode->i_sb, NULL); |
1062af92 DW |
3005 | if (ret) |
3006 | goto out; | |
3007 | } | |
1da177e4 LT |
3008 | |
3009 | dir->i_size += BOGO_DIRENT_SIZE; | |
078cd827 | 3010 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
d8c76e6f | 3011 | inc_nlink(inode); |
7de9c6ee | 3012 | ihold(inode); /* New dentry reference */ |
1da177e4 LT |
3013 | dget(dentry); /* Extra pinning count for the created dentry */ |
3014 | d_instantiate(dentry, inode); | |
5b04c689 PE |
3015 | out: |
3016 | return ret; | |
1da177e4 LT |
3017 | } |
3018 | ||
3019 | static int shmem_unlink(struct inode *dir, struct dentry *dentry) | |
3020 | { | |
75c3cfa8 | 3021 | struct inode *inode = d_inode(dentry); |
1da177e4 | 3022 | |
5b04c689 PE |
3023 | if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) |
3024 | shmem_free_inode(inode->i_sb); | |
1da177e4 LT |
3025 | |
3026 | dir->i_size -= BOGO_DIRENT_SIZE; | |
078cd827 | 3027 | inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode); |
9a53c3a7 | 3028 | drop_nlink(inode); |
1da177e4 LT |
3029 | dput(dentry); /* Undo the count from "create" - this does all the work */ |
3030 | return 0; | |
3031 | } | |
3032 | ||
3033 | static int shmem_rmdir(struct inode *dir, struct dentry *dentry) | |
3034 | { | |
3035 | if (!simple_empty(dentry)) | |
3036 | return -ENOTEMPTY; | |
3037 | ||
75c3cfa8 | 3038 | drop_nlink(d_inode(dentry)); |
9a53c3a7 | 3039 | drop_nlink(dir); |
1da177e4 LT |
3040 | return shmem_unlink(dir, dentry); |
3041 | } | |
3042 | ||
37456771 MS |
3043 | static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) |
3044 | { | |
e36cb0b8 DH |
3045 | bool old_is_dir = d_is_dir(old_dentry); |
3046 | bool new_is_dir = d_is_dir(new_dentry); | |
37456771 MS |
3047 | |
3048 | if (old_dir != new_dir && old_is_dir != new_is_dir) { | |
3049 | if (old_is_dir) { | |
3050 | drop_nlink(old_dir); | |
3051 | inc_nlink(new_dir); | |
3052 | } else { | |
3053 | drop_nlink(new_dir); | |
3054 | inc_nlink(old_dir); | |
3055 | } | |
3056 | } | |
3057 | old_dir->i_ctime = old_dir->i_mtime = | |
3058 | new_dir->i_ctime = new_dir->i_mtime = | |
75c3cfa8 | 3059 | d_inode(old_dentry)->i_ctime = |
078cd827 | 3060 | d_inode(new_dentry)->i_ctime = current_time(old_dir); |
37456771 MS |
3061 | |
3062 | return 0; | |
3063 | } | |
3064 | ||
46fdb794 MS |
3065 | static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry) |
3066 | { | |
3067 | struct dentry *whiteout; | |
3068 | int error; | |
3069 | ||
3070 | whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name); | |
3071 | if (!whiteout) | |
3072 | return -ENOMEM; | |
3073 | ||
3074 | error = shmem_mknod(old_dir, whiteout, | |
3075 | S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); | |
3076 | dput(whiteout); | |
3077 | if (error) | |
3078 | return error; | |
3079 | ||
3080 | /* | |
3081 | * Cheat and hash the whiteout while the old dentry is still in | |
3082 | * place, instead of playing games with FS_RENAME_DOES_D_MOVE. | |
3083 | * | |
3084 | * d_lookup() will consistently find one of them at this point, | |
3085 | * not sure which one, but that isn't even important. | |
3086 | */ | |
3087 | d_rehash(whiteout); | |
3088 | return 0; | |
3089 | } | |
3090 | ||
1da177e4 LT |
3091 | /* |
3092 | * The VFS layer already does all the dentry stuff for rename, | |
3093 | * we just have to decrement the usage count for the target if | |
3094 | * it exists so that the VFS layer correctly free's it when it | |
3095 | * gets overwritten. | |
3096 | */ | |
3b69ff51 | 3097 | static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags) |
1da177e4 | 3098 | { |
75c3cfa8 | 3099 | struct inode *inode = d_inode(old_dentry); |
1da177e4 LT |
3100 | int they_are_dirs = S_ISDIR(inode->i_mode); |
3101 | ||
46fdb794 | 3102 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
3b69ff51 MS |
3103 | return -EINVAL; |
3104 | ||
37456771 MS |
3105 | if (flags & RENAME_EXCHANGE) |
3106 | return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry); | |
3107 | ||
1da177e4 LT |
3108 | if (!simple_empty(new_dentry)) |
3109 | return -ENOTEMPTY; | |
3110 | ||
46fdb794 MS |
3111 | if (flags & RENAME_WHITEOUT) { |
3112 | int error; | |
3113 | ||
3114 | error = shmem_whiteout(old_dir, old_dentry); | |
3115 | if (error) | |
3116 | return error; | |
3117 | } | |
3118 | ||
75c3cfa8 | 3119 | if (d_really_is_positive(new_dentry)) { |
1da177e4 | 3120 | (void) shmem_unlink(new_dir, new_dentry); |
b928095b | 3121 | if (they_are_dirs) { |
75c3cfa8 | 3122 | drop_nlink(d_inode(new_dentry)); |
9a53c3a7 | 3123 | drop_nlink(old_dir); |
b928095b | 3124 | } |
1da177e4 | 3125 | } else if (they_are_dirs) { |
9a53c3a7 | 3126 | drop_nlink(old_dir); |
d8c76e6f | 3127 | inc_nlink(new_dir); |
1da177e4 LT |
3128 | } |
3129 | ||
3130 | old_dir->i_size -= BOGO_DIRENT_SIZE; | |
3131 | new_dir->i_size += BOGO_DIRENT_SIZE; | |
3132 | old_dir->i_ctime = old_dir->i_mtime = | |
3133 | new_dir->i_ctime = new_dir->i_mtime = | |
078cd827 | 3134 | inode->i_ctime = current_time(old_dir); |
1da177e4 LT |
3135 | return 0; |
3136 | } | |
3137 | ||
3138 | static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
3139 | { | |
3140 | int error; | |
3141 | int len; | |
3142 | struct inode *inode; | |
9276aad6 | 3143 | struct page *page; |
1da177e4 LT |
3144 | |
3145 | len = strlen(symname) + 1; | |
09cbfeaf | 3146 | if (len > PAGE_SIZE) |
1da177e4 LT |
3147 | return -ENAMETOOLONG; |
3148 | ||
0825a6f9 JP |
3149 | inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0, |
3150 | VM_NORESERVE); | |
1da177e4 LT |
3151 | if (!inode) |
3152 | return -ENOSPC; | |
3153 | ||
9d8f13ba | 3154 | error = security_inode_init_security(inode, dir, &dentry->d_name, |
6d9d88d0 | 3155 | shmem_initxattrs, NULL); |
343c3d7f MN |
3156 | if (error && error != -EOPNOTSUPP) { |
3157 | iput(inode); | |
3158 | return error; | |
570bc1c2 SS |
3159 | } |
3160 | ||
1da177e4 | 3161 | inode->i_size = len-1; |
69f07ec9 | 3162 | if (len <= SHORT_SYMLINK_LEN) { |
3ed47db3 AV |
3163 | inode->i_link = kmemdup(symname, len, GFP_KERNEL); |
3164 | if (!inode->i_link) { | |
69f07ec9 HD |
3165 | iput(inode); |
3166 | return -ENOMEM; | |
3167 | } | |
3168 | inode->i_op = &shmem_short_symlink_operations; | |
1da177e4 | 3169 | } else { |
e8ecde25 | 3170 | inode_nohighmem(inode); |
9e18eb29 | 3171 | error = shmem_getpage(inode, 0, &page, SGP_WRITE); |
1da177e4 LT |
3172 | if (error) { |
3173 | iput(inode); | |
3174 | return error; | |
3175 | } | |
14fcc23f | 3176 | inode->i_mapping->a_ops = &shmem_aops; |
1da177e4 | 3177 | inode->i_op = &shmem_symlink_inode_operations; |
21fc61c7 | 3178 | memcpy(page_address(page), symname, len); |
ec9516fb | 3179 | SetPageUptodate(page); |
1da177e4 | 3180 | set_page_dirty(page); |
6746aff7 | 3181 | unlock_page(page); |
09cbfeaf | 3182 | put_page(page); |
1da177e4 | 3183 | } |
1da177e4 | 3184 | dir->i_size += BOGO_DIRENT_SIZE; |
078cd827 | 3185 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
3186 | d_instantiate(dentry, inode); |
3187 | dget(dentry); | |
3188 | return 0; | |
3189 | } | |
3190 | ||
fceef393 | 3191 | static void shmem_put_link(void *arg) |
1da177e4 | 3192 | { |
fceef393 AV |
3193 | mark_page_accessed(arg); |
3194 | put_page(arg); | |
1da177e4 LT |
3195 | } |
3196 | ||
6b255391 | 3197 | static const char *shmem_get_link(struct dentry *dentry, |
fceef393 AV |
3198 | struct inode *inode, |
3199 | struct delayed_call *done) | |
1da177e4 | 3200 | { |
1da177e4 | 3201 | struct page *page = NULL; |
6b255391 | 3202 | int error; |
6a6c9904 AV |
3203 | if (!dentry) { |
3204 | page = find_get_page(inode->i_mapping, 0); | |
3205 | if (!page) | |
3206 | return ERR_PTR(-ECHILD); | |
3207 | if (!PageUptodate(page)) { | |
3208 | put_page(page); | |
3209 | return ERR_PTR(-ECHILD); | |
3210 | } | |
3211 | } else { | |
9e18eb29 | 3212 | error = shmem_getpage(inode, 0, &page, SGP_READ); |
6a6c9904 AV |
3213 | if (error) |
3214 | return ERR_PTR(error); | |
3215 | unlock_page(page); | |
3216 | } | |
fceef393 | 3217 | set_delayed_call(done, shmem_put_link, page); |
21fc61c7 | 3218 | return page_address(page); |
1da177e4 LT |
3219 | } |
3220 | ||
b09e0fa4 | 3221 | #ifdef CONFIG_TMPFS_XATTR |
46711810 | 3222 | /* |
b09e0fa4 EP |
3223 | * Superblocks without xattr inode operations may get some security.* xattr |
3224 | * support from the LSM "for free". As soon as we have any other xattrs | |
39f0247d AG |
3225 | * like ACLs, we also need to implement the security.* handlers at |
3226 | * filesystem level, though. | |
3227 | */ | |
3228 | ||
6d9d88d0 JS |
3229 | /* |
3230 | * Callback for security_inode_init_security() for acquiring xattrs. | |
3231 | */ | |
3232 | static int shmem_initxattrs(struct inode *inode, | |
3233 | const struct xattr *xattr_array, | |
3234 | void *fs_info) | |
3235 | { | |
3236 | struct shmem_inode_info *info = SHMEM_I(inode); | |
3237 | const struct xattr *xattr; | |
38f38657 | 3238 | struct simple_xattr *new_xattr; |
6d9d88d0 JS |
3239 | size_t len; |
3240 | ||
3241 | for (xattr = xattr_array; xattr->name != NULL; xattr++) { | |
38f38657 | 3242 | new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len); |
6d9d88d0 JS |
3243 | if (!new_xattr) |
3244 | return -ENOMEM; | |
3245 | ||
3246 | len = strlen(xattr->name) + 1; | |
3247 | new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len, | |
3248 | GFP_KERNEL); | |
3249 | if (!new_xattr->name) { | |
3bef735a | 3250 | kvfree(new_xattr); |
6d9d88d0 JS |
3251 | return -ENOMEM; |
3252 | } | |
3253 | ||
3254 | memcpy(new_xattr->name, XATTR_SECURITY_PREFIX, | |
3255 | XATTR_SECURITY_PREFIX_LEN); | |
3256 | memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN, | |
3257 | xattr->name, len); | |
3258 | ||
38f38657 | 3259 | simple_xattr_list_add(&info->xattrs, new_xattr); |
6d9d88d0 JS |
3260 | } |
3261 | ||
3262 | return 0; | |
3263 | } | |
3264 | ||
aa7c5241 | 3265 | static int shmem_xattr_handler_get(const struct xattr_handler *handler, |
b296821a AV |
3266 | struct dentry *unused, struct inode *inode, |
3267 | const char *name, void *buffer, size_t size) | |
b09e0fa4 | 3268 | { |
b296821a | 3269 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3270 | |
aa7c5241 | 3271 | name = xattr_full_name(handler, name); |
38f38657 | 3272 | return simple_xattr_get(&info->xattrs, name, buffer, size); |
b09e0fa4 EP |
3273 | } |
3274 | ||
aa7c5241 | 3275 | static int shmem_xattr_handler_set(const struct xattr_handler *handler, |
e65ce2a5 | 3276 | struct user_namespace *mnt_userns, |
59301226 AV |
3277 | struct dentry *unused, struct inode *inode, |
3278 | const char *name, const void *value, | |
3279 | size_t size, int flags) | |
b09e0fa4 | 3280 | { |
59301226 | 3281 | struct shmem_inode_info *info = SHMEM_I(inode); |
b09e0fa4 | 3282 | |
aa7c5241 | 3283 | name = xattr_full_name(handler, name); |
a46a2295 | 3284 | return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL); |
b09e0fa4 EP |
3285 | } |
3286 | ||
aa7c5241 AG |
3287 | static const struct xattr_handler shmem_security_xattr_handler = { |
3288 | .prefix = XATTR_SECURITY_PREFIX, | |
3289 | .get = shmem_xattr_handler_get, | |
3290 | .set = shmem_xattr_handler_set, | |
3291 | }; | |
b09e0fa4 | 3292 | |
aa7c5241 AG |
3293 | static const struct xattr_handler shmem_trusted_xattr_handler = { |
3294 | .prefix = XATTR_TRUSTED_PREFIX, | |
3295 | .get = shmem_xattr_handler_get, | |
3296 | .set = shmem_xattr_handler_set, | |
3297 | }; | |
b09e0fa4 | 3298 | |
aa7c5241 AG |
3299 | static const struct xattr_handler *shmem_xattr_handlers[] = { |
3300 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
3301 | &posix_acl_access_xattr_handler, | |
3302 | &posix_acl_default_xattr_handler, | |
3303 | #endif | |
3304 | &shmem_security_xattr_handler, | |
3305 | &shmem_trusted_xattr_handler, | |
3306 | NULL | |
3307 | }; | |
b09e0fa4 EP |
3308 | |
3309 | static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) | |
3310 | { | |
75c3cfa8 | 3311 | struct shmem_inode_info *info = SHMEM_I(d_inode(dentry)); |
786534b9 | 3312 | return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size); |
b09e0fa4 EP |
3313 | } |
3314 | #endif /* CONFIG_TMPFS_XATTR */ | |
3315 | ||
69f07ec9 | 3316 | static const struct inode_operations shmem_short_symlink_operations = { |
6b255391 | 3317 | .get_link = simple_get_link, |
b09e0fa4 | 3318 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3319 | .listxattr = shmem_listxattr, |
b09e0fa4 EP |
3320 | #endif |
3321 | }; | |
3322 | ||
3323 | static const struct inode_operations shmem_symlink_inode_operations = { | |
6b255391 | 3324 | .get_link = shmem_get_link, |
b09e0fa4 | 3325 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3326 | .listxattr = shmem_listxattr, |
39f0247d | 3327 | #endif |
b09e0fa4 | 3328 | }; |
39f0247d | 3329 | |
91828a40 DG |
3330 | static struct dentry *shmem_get_parent(struct dentry *child) |
3331 | { | |
3332 | return ERR_PTR(-ESTALE); | |
3333 | } | |
3334 | ||
3335 | static int shmem_match(struct inode *ino, void *vfh) | |
3336 | { | |
3337 | __u32 *fh = vfh; | |
3338 | __u64 inum = fh[2]; | |
3339 | inum = (inum << 32) | fh[1]; | |
3340 | return ino->i_ino == inum && fh[0] == ino->i_generation; | |
3341 | } | |
3342 | ||
12ba780d AG |
3343 | /* Find any alias of inode, but prefer a hashed alias */ |
3344 | static struct dentry *shmem_find_alias(struct inode *inode) | |
3345 | { | |
3346 | struct dentry *alias = d_find_alias(inode); | |
3347 | ||
3348 | return alias ?: d_find_any_alias(inode); | |
3349 | } | |
3350 | ||
3351 | ||
480b116c CH |
3352 | static struct dentry *shmem_fh_to_dentry(struct super_block *sb, |
3353 | struct fid *fid, int fh_len, int fh_type) | |
91828a40 | 3354 | { |
91828a40 | 3355 | struct inode *inode; |
480b116c | 3356 | struct dentry *dentry = NULL; |
35c2a7f4 | 3357 | u64 inum; |
480b116c CH |
3358 | |
3359 | if (fh_len < 3) | |
3360 | return NULL; | |
91828a40 | 3361 | |
35c2a7f4 HD |
3362 | inum = fid->raw[2]; |
3363 | inum = (inum << 32) | fid->raw[1]; | |
3364 | ||
480b116c CH |
3365 | inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), |
3366 | shmem_match, fid->raw); | |
91828a40 | 3367 | if (inode) { |
12ba780d | 3368 | dentry = shmem_find_alias(inode); |
91828a40 DG |
3369 | iput(inode); |
3370 | } | |
3371 | ||
480b116c | 3372 | return dentry; |
91828a40 DG |
3373 | } |
3374 | ||
b0b0382b AV |
3375 | static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len, |
3376 | struct inode *parent) | |
91828a40 | 3377 | { |
5fe0c237 AK |
3378 | if (*len < 3) { |
3379 | *len = 3; | |
94e07a75 | 3380 | return FILEID_INVALID; |
5fe0c237 | 3381 | } |
91828a40 | 3382 | |
1d3382cb | 3383 | if (inode_unhashed(inode)) { |
91828a40 DG |
3384 | /* Unfortunately insert_inode_hash is not idempotent, |
3385 | * so as we hash inodes here rather than at creation | |
3386 | * time, we need a lock to ensure we only try | |
3387 | * to do it once | |
3388 | */ | |
3389 | static DEFINE_SPINLOCK(lock); | |
3390 | spin_lock(&lock); | |
1d3382cb | 3391 | if (inode_unhashed(inode)) |
91828a40 DG |
3392 | __insert_inode_hash(inode, |
3393 | inode->i_ino + inode->i_generation); | |
3394 | spin_unlock(&lock); | |
3395 | } | |
3396 | ||
3397 | fh[0] = inode->i_generation; | |
3398 | fh[1] = inode->i_ino; | |
3399 | fh[2] = ((__u64)inode->i_ino) >> 32; | |
3400 | ||
3401 | *len = 3; | |
3402 | return 1; | |
3403 | } | |
3404 | ||
39655164 | 3405 | static const struct export_operations shmem_export_ops = { |
91828a40 | 3406 | .get_parent = shmem_get_parent, |
91828a40 | 3407 | .encode_fh = shmem_encode_fh, |
480b116c | 3408 | .fh_to_dentry = shmem_fh_to_dentry, |
91828a40 DG |
3409 | }; |
3410 | ||
626c3920 AV |
3411 | enum shmem_param { |
3412 | Opt_gid, | |
3413 | Opt_huge, | |
3414 | Opt_mode, | |
3415 | Opt_mpol, | |
3416 | Opt_nr_blocks, | |
3417 | Opt_nr_inodes, | |
3418 | Opt_size, | |
3419 | Opt_uid, | |
ea3271f7 CD |
3420 | Opt_inode32, |
3421 | Opt_inode64, | |
626c3920 AV |
3422 | }; |
3423 | ||
5eede625 | 3424 | static const struct constant_table shmem_param_enums_huge[] = { |
2710c957 AV |
3425 | {"never", SHMEM_HUGE_NEVER }, |
3426 | {"always", SHMEM_HUGE_ALWAYS }, | |
3427 | {"within_size", SHMEM_HUGE_WITHIN_SIZE }, | |
3428 | {"advise", SHMEM_HUGE_ADVISE }, | |
2710c957 AV |
3429 | {} |
3430 | }; | |
3431 | ||
d7167b14 | 3432 | const struct fs_parameter_spec shmem_fs_parameters[] = { |
626c3920 | 3433 | fsparam_u32 ("gid", Opt_gid), |
2710c957 | 3434 | fsparam_enum ("huge", Opt_huge, shmem_param_enums_huge), |
626c3920 AV |
3435 | fsparam_u32oct("mode", Opt_mode), |
3436 | fsparam_string("mpol", Opt_mpol), | |
3437 | fsparam_string("nr_blocks", Opt_nr_blocks), | |
3438 | fsparam_string("nr_inodes", Opt_nr_inodes), | |
3439 | fsparam_string("size", Opt_size), | |
3440 | fsparam_u32 ("uid", Opt_uid), | |
ea3271f7 CD |
3441 | fsparam_flag ("inode32", Opt_inode32), |
3442 | fsparam_flag ("inode64", Opt_inode64), | |
626c3920 AV |
3443 | {} |
3444 | }; | |
3445 | ||
f3235626 | 3446 | static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param) |
1da177e4 | 3447 | { |
f3235626 | 3448 | struct shmem_options *ctx = fc->fs_private; |
626c3920 AV |
3449 | struct fs_parse_result result; |
3450 | unsigned long long size; | |
e04dc423 | 3451 | char *rest; |
626c3920 AV |
3452 | int opt; |
3453 | ||
d7167b14 | 3454 | opt = fs_parse(fc, shmem_fs_parameters, param, &result); |
f3235626 | 3455 | if (opt < 0) |
626c3920 | 3456 | return opt; |
1da177e4 | 3457 | |
626c3920 AV |
3458 | switch (opt) { |
3459 | case Opt_size: | |
3460 | size = memparse(param->string, &rest); | |
e04dc423 AV |
3461 | if (*rest == '%') { |
3462 | size <<= PAGE_SHIFT; | |
3463 | size *= totalram_pages(); | |
3464 | do_div(size, 100); | |
3465 | rest++; | |
3466 | } | |
3467 | if (*rest) | |
626c3920 | 3468 | goto bad_value; |
e04dc423 AV |
3469 | ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE); |
3470 | ctx->seen |= SHMEM_SEEN_BLOCKS; | |
626c3920 AV |
3471 | break; |
3472 | case Opt_nr_blocks: | |
3473 | ctx->blocks = memparse(param->string, &rest); | |
e04dc423 | 3474 | if (*rest) |
626c3920 | 3475 | goto bad_value; |
e04dc423 | 3476 | ctx->seen |= SHMEM_SEEN_BLOCKS; |
626c3920 AV |
3477 | break; |
3478 | case Opt_nr_inodes: | |
3479 | ctx->inodes = memparse(param->string, &rest); | |
e04dc423 | 3480 | if (*rest) |
626c3920 | 3481 | goto bad_value; |
e04dc423 | 3482 | ctx->seen |= SHMEM_SEEN_INODES; |
626c3920 AV |
3483 | break; |
3484 | case Opt_mode: | |
3485 | ctx->mode = result.uint_32 & 07777; | |
3486 | break; | |
3487 | case Opt_uid: | |
3488 | ctx->uid = make_kuid(current_user_ns(), result.uint_32); | |
e04dc423 | 3489 | if (!uid_valid(ctx->uid)) |
626c3920 AV |
3490 | goto bad_value; |
3491 | break; | |
3492 | case Opt_gid: | |
3493 | ctx->gid = make_kgid(current_user_ns(), result.uint_32); | |
e04dc423 | 3494 | if (!gid_valid(ctx->gid)) |
626c3920 AV |
3495 | goto bad_value; |
3496 | break; | |
3497 | case Opt_huge: | |
3498 | ctx->huge = result.uint_32; | |
3499 | if (ctx->huge != SHMEM_HUGE_NEVER && | |
396bcc52 | 3500 | !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && |
626c3920 AV |
3501 | has_transparent_hugepage())) |
3502 | goto unsupported_parameter; | |
e04dc423 | 3503 | ctx->seen |= SHMEM_SEEN_HUGE; |
626c3920 AV |
3504 | break; |
3505 | case Opt_mpol: | |
3506 | if (IS_ENABLED(CONFIG_NUMA)) { | |
3507 | mpol_put(ctx->mpol); | |
3508 | ctx->mpol = NULL; | |
3509 | if (mpol_parse_str(param->string, &ctx->mpol)) | |
3510 | goto bad_value; | |
3511 | break; | |
3512 | } | |
3513 | goto unsupported_parameter; | |
ea3271f7 CD |
3514 | case Opt_inode32: |
3515 | ctx->full_inums = false; | |
3516 | ctx->seen |= SHMEM_SEEN_INUMS; | |
3517 | break; | |
3518 | case Opt_inode64: | |
3519 | if (sizeof(ino_t) < 8) { | |
3520 | return invalfc(fc, | |
3521 | "Cannot use inode64 with <64bit inums in kernel\n"); | |
3522 | } | |
3523 | ctx->full_inums = true; | |
3524 | ctx->seen |= SHMEM_SEEN_INUMS; | |
3525 | break; | |
e04dc423 AV |
3526 | } |
3527 | return 0; | |
3528 | ||
626c3920 | 3529 | unsupported_parameter: |
f35aa2bc | 3530 | return invalfc(fc, "Unsupported parameter '%s'", param->key); |
626c3920 | 3531 | bad_value: |
f35aa2bc | 3532 | return invalfc(fc, "Bad value for '%s'", param->key); |
e04dc423 AV |
3533 | } |
3534 | ||
f3235626 | 3535 | static int shmem_parse_options(struct fs_context *fc, void *data) |
e04dc423 | 3536 | { |
f3235626 DH |
3537 | char *options = data; |
3538 | ||
33f37c64 AV |
3539 | if (options) { |
3540 | int err = security_sb_eat_lsm_opts(options, &fc->security); | |
3541 | if (err) | |
3542 | return err; | |
3543 | } | |
3544 | ||
b00dc3ad | 3545 | while (options != NULL) { |
626c3920 | 3546 | char *this_char = options; |
b00dc3ad HD |
3547 | for (;;) { |
3548 | /* | |
3549 | * NUL-terminate this option: unfortunately, | |
3550 | * mount options form a comma-separated list, | |
3551 | * but mpol's nodelist may also contain commas. | |
3552 | */ | |
3553 | options = strchr(options, ','); | |
3554 | if (options == NULL) | |
3555 | break; | |
3556 | options++; | |
3557 | if (!isdigit(*options)) { | |
3558 | options[-1] = '\0'; | |
3559 | break; | |
3560 | } | |
3561 | } | |
626c3920 AV |
3562 | if (*this_char) { |
3563 | char *value = strchr(this_char,'='); | |
f3235626 | 3564 | size_t len = 0; |
626c3920 AV |
3565 | int err; |
3566 | ||
3567 | if (value) { | |
3568 | *value++ = '\0'; | |
f3235626 | 3569 | len = strlen(value); |
626c3920 | 3570 | } |
f3235626 DH |
3571 | err = vfs_parse_fs_string(fc, this_char, value, len); |
3572 | if (err < 0) | |
3573 | return err; | |
1da177e4 | 3574 | } |
1da177e4 LT |
3575 | } |
3576 | return 0; | |
1da177e4 LT |
3577 | } |
3578 | ||
f3235626 DH |
3579 | /* |
3580 | * Reconfigure a shmem filesystem. | |
3581 | * | |
3582 | * Note that we disallow change from limited->unlimited blocks/inodes while any | |
3583 | * are in use; but we must separately disallow unlimited->limited, because in | |
3584 | * that case we have no record of how much is already in use. | |
3585 | */ | |
3586 | static int shmem_reconfigure(struct fs_context *fc) | |
1da177e4 | 3587 | { |
f3235626 DH |
3588 | struct shmem_options *ctx = fc->fs_private; |
3589 | struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb); | |
0edd73b3 | 3590 | unsigned long inodes; |
f3235626 | 3591 | const char *err; |
1da177e4 | 3592 | |
0edd73b3 | 3593 | spin_lock(&sbinfo->stat_lock); |
0edd73b3 | 3594 | inodes = sbinfo->max_inodes - sbinfo->free_inodes; |
f3235626 DH |
3595 | if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) { |
3596 | if (!sbinfo->max_blocks) { | |
3597 | err = "Cannot retroactively limit size"; | |
0b5071dd | 3598 | goto out; |
f3235626 | 3599 | } |
0b5071dd | 3600 | if (percpu_counter_compare(&sbinfo->used_blocks, |
f3235626 DH |
3601 | ctx->blocks) > 0) { |
3602 | err = "Too small a size for current use"; | |
0b5071dd | 3603 | goto out; |
f3235626 | 3604 | } |
0b5071dd | 3605 | } |
f3235626 DH |
3606 | if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) { |
3607 | if (!sbinfo->max_inodes) { | |
3608 | err = "Cannot retroactively limit inodes"; | |
0b5071dd | 3609 | goto out; |
f3235626 DH |
3610 | } |
3611 | if (ctx->inodes < inodes) { | |
3612 | err = "Too few inodes for current use"; | |
0b5071dd | 3613 | goto out; |
f3235626 | 3614 | } |
0b5071dd | 3615 | } |
0edd73b3 | 3616 | |
ea3271f7 CD |
3617 | if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums && |
3618 | sbinfo->next_ino > UINT_MAX) { | |
3619 | err = "Current inum too high to switch to 32-bit inums"; | |
3620 | goto out; | |
3621 | } | |
3622 | ||
f3235626 DH |
3623 | if (ctx->seen & SHMEM_SEEN_HUGE) |
3624 | sbinfo->huge = ctx->huge; | |
ea3271f7 CD |
3625 | if (ctx->seen & SHMEM_SEEN_INUMS) |
3626 | sbinfo->full_inums = ctx->full_inums; | |
f3235626 DH |
3627 | if (ctx->seen & SHMEM_SEEN_BLOCKS) |
3628 | sbinfo->max_blocks = ctx->blocks; | |
3629 | if (ctx->seen & SHMEM_SEEN_INODES) { | |
3630 | sbinfo->max_inodes = ctx->inodes; | |
3631 | sbinfo->free_inodes = ctx->inodes - inodes; | |
0b5071dd | 3632 | } |
71fe804b | 3633 | |
5f00110f GT |
3634 | /* |
3635 | * Preserve previous mempolicy unless mpol remount option was specified. | |
3636 | */ | |
f3235626 | 3637 | if (ctx->mpol) { |
5f00110f | 3638 | mpol_put(sbinfo->mpol); |
f3235626 DH |
3639 | sbinfo->mpol = ctx->mpol; /* transfers initial ref */ |
3640 | ctx->mpol = NULL; | |
5f00110f | 3641 | } |
f3235626 DH |
3642 | spin_unlock(&sbinfo->stat_lock); |
3643 | return 0; | |
0edd73b3 HD |
3644 | out: |
3645 | spin_unlock(&sbinfo->stat_lock); | |
f35aa2bc | 3646 | return invalfc(fc, "%s", err); |
1da177e4 | 3647 | } |
680d794b | 3648 | |
34c80b1d | 3649 | static int shmem_show_options(struct seq_file *seq, struct dentry *root) |
680d794b | 3650 | { |
34c80b1d | 3651 | struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb); |
680d794b | 3652 | |
3653 | if (sbinfo->max_blocks != shmem_default_max_blocks()) | |
3654 | seq_printf(seq, ",size=%luk", | |
09cbfeaf | 3655 | sbinfo->max_blocks << (PAGE_SHIFT - 10)); |
680d794b | 3656 | if (sbinfo->max_inodes != shmem_default_max_inodes()) |
3657 | seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); | |
0825a6f9 | 3658 | if (sbinfo->mode != (0777 | S_ISVTX)) |
09208d15 | 3659 | seq_printf(seq, ",mode=%03ho", sbinfo->mode); |
8751e039 EB |
3660 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
3661 | seq_printf(seq, ",uid=%u", | |
3662 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
3663 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
3664 | seq_printf(seq, ",gid=%u", | |
3665 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
ea3271f7 CD |
3666 | |
3667 | /* | |
3668 | * Showing inode{64,32} might be useful even if it's the system default, | |
3669 | * since then people don't have to resort to checking both here and | |
3670 | * /proc/config.gz to confirm 64-bit inums were successfully applied | |
3671 | * (which may not even exist if IKCONFIG_PROC isn't enabled). | |
3672 | * | |
3673 | * We hide it when inode64 isn't the default and we are using 32-bit | |
3674 | * inodes, since that probably just means the feature isn't even under | |
3675 | * consideration. | |
3676 | * | |
3677 | * As such: | |
3678 | * | |
3679 | * +-----------------+-----------------+ | |
3680 | * | TMPFS_INODE64=y | TMPFS_INODE64=n | | |
3681 | * +------------------+-----------------+-----------------+ | |
3682 | * | full_inums=true | show | show | | |
3683 | * | full_inums=false | show | hide | | |
3684 | * +------------------+-----------------+-----------------+ | |
3685 | * | |
3686 | */ | |
3687 | if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums) | |
3688 | seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32)); | |
396bcc52 | 3689 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5a6e75f8 KS |
3690 | /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */ |
3691 | if (sbinfo->huge) | |
3692 | seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge)); | |
3693 | #endif | |
71fe804b | 3694 | shmem_show_mpol(seq, sbinfo->mpol); |
680d794b | 3695 | return 0; |
3696 | } | |
9183df25 | 3697 | |
680d794b | 3698 | #endif /* CONFIG_TMPFS */ |
1da177e4 LT |
3699 | |
3700 | static void shmem_put_super(struct super_block *sb) | |
3701 | { | |
602586a8 HD |
3702 | struct shmem_sb_info *sbinfo = SHMEM_SB(sb); |
3703 | ||
e809d5f0 | 3704 | free_percpu(sbinfo->ino_batch); |
602586a8 | 3705 | percpu_counter_destroy(&sbinfo->used_blocks); |
49cd0a5c | 3706 | mpol_put(sbinfo->mpol); |
602586a8 | 3707 | kfree(sbinfo); |
1da177e4 LT |
3708 | sb->s_fs_info = NULL; |
3709 | } | |
3710 | ||
f3235626 | 3711 | static int shmem_fill_super(struct super_block *sb, struct fs_context *fc) |
1da177e4 | 3712 | { |
f3235626 | 3713 | struct shmem_options *ctx = fc->fs_private; |
1da177e4 | 3714 | struct inode *inode; |
0edd73b3 | 3715 | struct shmem_sb_info *sbinfo; |
680d794b | 3716 | int err = -ENOMEM; |
3717 | ||
3718 | /* Round up to L1_CACHE_BYTES to resist false sharing */ | |
425fbf04 | 3719 | sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), |
680d794b | 3720 | L1_CACHE_BYTES), GFP_KERNEL); |
3721 | if (!sbinfo) | |
3722 | return -ENOMEM; | |
3723 | ||
680d794b | 3724 | sb->s_fs_info = sbinfo; |
1da177e4 | 3725 | |
0edd73b3 | 3726 | #ifdef CONFIG_TMPFS |
1da177e4 LT |
3727 | /* |
3728 | * Per default we only allow half of the physical ram per | |
3729 | * tmpfs instance, limiting inodes to one per page of lowmem; | |
3730 | * but the internal instance is left unlimited. | |
3731 | */ | |
1751e8a6 | 3732 | if (!(sb->s_flags & SB_KERNMOUNT)) { |
f3235626 DH |
3733 | if (!(ctx->seen & SHMEM_SEEN_BLOCKS)) |
3734 | ctx->blocks = shmem_default_max_blocks(); | |
3735 | if (!(ctx->seen & SHMEM_SEEN_INODES)) | |
3736 | ctx->inodes = shmem_default_max_inodes(); | |
ea3271f7 CD |
3737 | if (!(ctx->seen & SHMEM_SEEN_INUMS)) |
3738 | ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64); | |
ca4e0519 | 3739 | } else { |
1751e8a6 | 3740 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 3741 | } |
91828a40 | 3742 | sb->s_export_op = &shmem_export_ops; |
1751e8a6 | 3743 | sb->s_flags |= SB_NOSEC; |
1da177e4 | 3744 | #else |
1751e8a6 | 3745 | sb->s_flags |= SB_NOUSER; |
1da177e4 | 3746 | #endif |
f3235626 DH |
3747 | sbinfo->max_blocks = ctx->blocks; |
3748 | sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes; | |
e809d5f0 CD |
3749 | if (sb->s_flags & SB_KERNMOUNT) { |
3750 | sbinfo->ino_batch = alloc_percpu(ino_t); | |
3751 | if (!sbinfo->ino_batch) | |
3752 | goto failed; | |
3753 | } | |
f3235626 DH |
3754 | sbinfo->uid = ctx->uid; |
3755 | sbinfo->gid = ctx->gid; | |
ea3271f7 | 3756 | sbinfo->full_inums = ctx->full_inums; |
f3235626 DH |
3757 | sbinfo->mode = ctx->mode; |
3758 | sbinfo->huge = ctx->huge; | |
3759 | sbinfo->mpol = ctx->mpol; | |
3760 | ctx->mpol = NULL; | |
1da177e4 | 3761 | |
0edd73b3 | 3762 | spin_lock_init(&sbinfo->stat_lock); |
908c7f19 | 3763 | if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL)) |
602586a8 | 3764 | goto failed; |
779750d2 KS |
3765 | spin_lock_init(&sbinfo->shrinklist_lock); |
3766 | INIT_LIST_HEAD(&sbinfo->shrinklist); | |
0edd73b3 | 3767 | |
285b2c4f | 3768 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
09cbfeaf KS |
3769 | sb->s_blocksize = PAGE_SIZE; |
3770 | sb->s_blocksize_bits = PAGE_SHIFT; | |
1da177e4 LT |
3771 | sb->s_magic = TMPFS_MAGIC; |
3772 | sb->s_op = &shmem_ops; | |
cfd95a9c | 3773 | sb->s_time_gran = 1; |
b09e0fa4 | 3774 | #ifdef CONFIG_TMPFS_XATTR |
39f0247d | 3775 | sb->s_xattr = shmem_xattr_handlers; |
b09e0fa4 EP |
3776 | #endif |
3777 | #ifdef CONFIG_TMPFS_POSIX_ACL | |
1751e8a6 | 3778 | sb->s_flags |= SB_POSIXACL; |
39f0247d | 3779 | #endif |
2b4db796 | 3780 | uuid_gen(&sb->s_uuid); |
0edd73b3 | 3781 | |
454abafe | 3782 | inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); |
1da177e4 LT |
3783 | if (!inode) |
3784 | goto failed; | |
680d794b | 3785 | inode->i_uid = sbinfo->uid; |
3786 | inode->i_gid = sbinfo->gid; | |
318ceed0 AV |
3787 | sb->s_root = d_make_root(inode); |
3788 | if (!sb->s_root) | |
48fde701 | 3789 | goto failed; |
1da177e4 LT |
3790 | return 0; |
3791 | ||
1da177e4 LT |
3792 | failed: |
3793 | shmem_put_super(sb); | |
3794 | return err; | |
3795 | } | |
3796 | ||
f3235626 DH |
3797 | static int shmem_get_tree(struct fs_context *fc) |
3798 | { | |
3799 | return get_tree_nodev(fc, shmem_fill_super); | |
3800 | } | |
3801 | ||
3802 | static void shmem_free_fc(struct fs_context *fc) | |
3803 | { | |
3804 | struct shmem_options *ctx = fc->fs_private; | |
3805 | ||
3806 | if (ctx) { | |
3807 | mpol_put(ctx->mpol); | |
3808 | kfree(ctx); | |
3809 | } | |
3810 | } | |
3811 | ||
3812 | static const struct fs_context_operations shmem_fs_context_ops = { | |
3813 | .free = shmem_free_fc, | |
3814 | .get_tree = shmem_get_tree, | |
3815 | #ifdef CONFIG_TMPFS | |
3816 | .parse_monolithic = shmem_parse_options, | |
3817 | .parse_param = shmem_parse_one, | |
3818 | .reconfigure = shmem_reconfigure, | |
3819 | #endif | |
3820 | }; | |
3821 | ||
fcc234f8 | 3822 | static struct kmem_cache *shmem_inode_cachep; |
1da177e4 LT |
3823 | |
3824 | static struct inode *shmem_alloc_inode(struct super_block *sb) | |
3825 | { | |
41ffe5d5 HD |
3826 | struct shmem_inode_info *info; |
3827 | info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); | |
3828 | if (!info) | |
1da177e4 | 3829 | return NULL; |
41ffe5d5 | 3830 | return &info->vfs_inode; |
1da177e4 LT |
3831 | } |
3832 | ||
74b1da56 | 3833 | static void shmem_free_in_core_inode(struct inode *inode) |
fa0d7e3d | 3834 | { |
84e710da AV |
3835 | if (S_ISLNK(inode->i_mode)) |
3836 | kfree(inode->i_link); | |
fa0d7e3d NP |
3837 | kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); |
3838 | } | |
3839 | ||
1da177e4 LT |
3840 | static void shmem_destroy_inode(struct inode *inode) |
3841 | { | |
09208d15 | 3842 | if (S_ISREG(inode->i_mode)) |
1da177e4 | 3843 | mpol_free_shared_policy(&SHMEM_I(inode)->policy); |
1da177e4 LT |
3844 | } |
3845 | ||
41ffe5d5 | 3846 | static void shmem_init_inode(void *foo) |
1da177e4 | 3847 | { |
41ffe5d5 HD |
3848 | struct shmem_inode_info *info = foo; |
3849 | inode_init_once(&info->vfs_inode); | |
1da177e4 LT |
3850 | } |
3851 | ||
9a8ec03e | 3852 | static void shmem_init_inodecache(void) |
1da177e4 LT |
3853 | { |
3854 | shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", | |
3855 | sizeof(struct shmem_inode_info), | |
5d097056 | 3856 | 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode); |
1da177e4 LT |
3857 | } |
3858 | ||
41ffe5d5 | 3859 | static void shmem_destroy_inodecache(void) |
1da177e4 | 3860 | { |
1a1d92c1 | 3861 | kmem_cache_destroy(shmem_inode_cachep); |
1da177e4 LT |
3862 | } |
3863 | ||
30e6a51d | 3864 | const struct address_space_operations shmem_aops = { |
1da177e4 | 3865 | .writepage = shmem_writepage, |
76719325 | 3866 | .set_page_dirty = __set_page_dirty_no_writeback, |
1da177e4 | 3867 | #ifdef CONFIG_TMPFS |
800d15a5 NP |
3868 | .write_begin = shmem_write_begin, |
3869 | .write_end = shmem_write_end, | |
1da177e4 | 3870 | #endif |
1c93923c | 3871 | #ifdef CONFIG_MIGRATION |
304dbdb7 | 3872 | .migratepage = migrate_page, |
1c93923c | 3873 | #endif |
aa261f54 | 3874 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 3875 | }; |
30e6a51d | 3876 | EXPORT_SYMBOL(shmem_aops); |
1da177e4 | 3877 | |
15ad7cdc | 3878 | static const struct file_operations shmem_file_operations = { |
1da177e4 | 3879 | .mmap = shmem_mmap, |
c01d5b30 | 3880 | .get_unmapped_area = shmem_get_unmapped_area, |
1da177e4 | 3881 | #ifdef CONFIG_TMPFS |
220f2ac9 | 3882 | .llseek = shmem_file_llseek, |
2ba5bbed | 3883 | .read_iter = shmem_file_read_iter, |
8174202b | 3884 | .write_iter = generic_file_write_iter, |
1b061d92 | 3885 | .fsync = noop_fsync, |
82c156f8 | 3886 | .splice_read = generic_file_splice_read, |
f6cb85d0 | 3887 | .splice_write = iter_file_splice_write, |
83e4fa9c | 3888 | .fallocate = shmem_fallocate, |
1da177e4 LT |
3889 | #endif |
3890 | }; | |
3891 | ||
92e1d5be | 3892 | static const struct inode_operations shmem_inode_operations = { |
44a30220 | 3893 | .getattr = shmem_getattr, |
94c1e62d | 3894 | .setattr = shmem_setattr, |
b09e0fa4 | 3895 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3896 | .listxattr = shmem_listxattr, |
feda821e | 3897 | .set_acl = simple_set_acl, |
b09e0fa4 | 3898 | #endif |
1da177e4 LT |
3899 | }; |
3900 | ||
92e1d5be | 3901 | static const struct inode_operations shmem_dir_inode_operations = { |
1da177e4 LT |
3902 | #ifdef CONFIG_TMPFS |
3903 | .create = shmem_create, | |
3904 | .lookup = simple_lookup, | |
3905 | .link = shmem_link, | |
3906 | .unlink = shmem_unlink, | |
3907 | .symlink = shmem_symlink, | |
3908 | .mkdir = shmem_mkdir, | |
3909 | .rmdir = shmem_rmdir, | |
3910 | .mknod = shmem_mknod, | |
2773bf00 | 3911 | .rename = shmem_rename2, |
60545d0d | 3912 | .tmpfile = shmem_tmpfile, |
1da177e4 | 3913 | #endif |
b09e0fa4 | 3914 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3915 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3916 | #endif |
39f0247d | 3917 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3918 | .setattr = shmem_setattr, |
feda821e | 3919 | .set_acl = simple_set_acl, |
39f0247d AG |
3920 | #endif |
3921 | }; | |
3922 | ||
92e1d5be | 3923 | static const struct inode_operations shmem_special_inode_operations = { |
b09e0fa4 | 3924 | #ifdef CONFIG_TMPFS_XATTR |
b09e0fa4 | 3925 | .listxattr = shmem_listxattr, |
b09e0fa4 | 3926 | #endif |
39f0247d | 3927 | #ifdef CONFIG_TMPFS_POSIX_ACL |
94c1e62d | 3928 | .setattr = shmem_setattr, |
feda821e | 3929 | .set_acl = simple_set_acl, |
39f0247d | 3930 | #endif |
1da177e4 LT |
3931 | }; |
3932 | ||
759b9775 | 3933 | static const struct super_operations shmem_ops = { |
1da177e4 | 3934 | .alloc_inode = shmem_alloc_inode, |
74b1da56 | 3935 | .free_inode = shmem_free_in_core_inode, |
1da177e4 LT |
3936 | .destroy_inode = shmem_destroy_inode, |
3937 | #ifdef CONFIG_TMPFS | |
3938 | .statfs = shmem_statfs, | |
680d794b | 3939 | .show_options = shmem_show_options, |
1da177e4 | 3940 | #endif |
1f895f75 | 3941 | .evict_inode = shmem_evict_inode, |
1da177e4 LT |
3942 | .drop_inode = generic_delete_inode, |
3943 | .put_super = shmem_put_super, | |
396bcc52 | 3944 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
779750d2 KS |
3945 | .nr_cached_objects = shmem_unused_huge_count, |
3946 | .free_cached_objects = shmem_unused_huge_scan, | |
3947 | #endif | |
1da177e4 LT |
3948 | }; |
3949 | ||
f0f37e2f | 3950 | static const struct vm_operations_struct shmem_vm_ops = { |
54cb8821 | 3951 | .fault = shmem_fault, |
d7c17551 | 3952 | .map_pages = filemap_map_pages, |
1da177e4 LT |
3953 | #ifdef CONFIG_NUMA |
3954 | .set_policy = shmem_set_policy, | |
3955 | .get_policy = shmem_get_policy, | |
3956 | #endif | |
3957 | }; | |
3958 | ||
f3235626 | 3959 | int shmem_init_fs_context(struct fs_context *fc) |
1da177e4 | 3960 | { |
f3235626 DH |
3961 | struct shmem_options *ctx; |
3962 | ||
3963 | ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL); | |
3964 | if (!ctx) | |
3965 | return -ENOMEM; | |
3966 | ||
3967 | ctx->mode = 0777 | S_ISVTX; | |
3968 | ctx->uid = current_fsuid(); | |
3969 | ctx->gid = current_fsgid(); | |
3970 | ||
3971 | fc->fs_private = ctx; | |
3972 | fc->ops = &shmem_fs_context_ops; | |
3973 | return 0; | |
1da177e4 LT |
3974 | } |
3975 | ||
41ffe5d5 | 3976 | static struct file_system_type shmem_fs_type = { |
1da177e4 LT |
3977 | .owner = THIS_MODULE, |
3978 | .name = "tmpfs", | |
f3235626 DH |
3979 | .init_fs_context = shmem_init_fs_context, |
3980 | #ifdef CONFIG_TMPFS | |
d7167b14 | 3981 | .parameters = shmem_fs_parameters, |
f3235626 | 3982 | #endif |
1da177e4 | 3983 | .kill_sb = kill_litter_super, |
01c70267 | 3984 | .fs_flags = FS_USERNS_MOUNT | FS_THP_SUPPORT, |
1da177e4 | 3985 | }; |
1da177e4 | 3986 | |
41ffe5d5 | 3987 | int __init shmem_init(void) |
1da177e4 LT |
3988 | { |
3989 | int error; | |
3990 | ||
9a8ec03e | 3991 | shmem_init_inodecache(); |
1da177e4 | 3992 | |
41ffe5d5 | 3993 | error = register_filesystem(&shmem_fs_type); |
1da177e4 | 3994 | if (error) { |
1170532b | 3995 | pr_err("Could not register tmpfs\n"); |
1da177e4 LT |
3996 | goto out2; |
3997 | } | |
95dc112a | 3998 | |
ca4e0519 | 3999 | shm_mnt = kern_mount(&shmem_fs_type); |
1da177e4 LT |
4000 | if (IS_ERR(shm_mnt)) { |
4001 | error = PTR_ERR(shm_mnt); | |
1170532b | 4002 | pr_err("Could not kern_mount tmpfs\n"); |
1da177e4 LT |
4003 | goto out1; |
4004 | } | |
5a6e75f8 | 4005 | |
396bcc52 | 4006 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
435c0b87 | 4007 | if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4008 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4009 | else | |
4010 | shmem_huge = 0; /* just in case it was patched */ | |
4011 | #endif | |
1da177e4 LT |
4012 | return 0; |
4013 | ||
4014 | out1: | |
41ffe5d5 | 4015 | unregister_filesystem(&shmem_fs_type); |
1da177e4 | 4016 | out2: |
41ffe5d5 | 4017 | shmem_destroy_inodecache(); |
1da177e4 LT |
4018 | shm_mnt = ERR_PTR(error); |
4019 | return error; | |
4020 | } | |
853ac43a | 4021 | |
396bcc52 | 4022 | #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS) |
5a6e75f8 | 4023 | static ssize_t shmem_enabled_show(struct kobject *kobj, |
79d4d38a | 4024 | struct kobj_attribute *attr, char *buf) |
5a6e75f8 | 4025 | { |
26083eb6 | 4026 | static const int values[] = { |
5a6e75f8 KS |
4027 | SHMEM_HUGE_ALWAYS, |
4028 | SHMEM_HUGE_WITHIN_SIZE, | |
4029 | SHMEM_HUGE_ADVISE, | |
4030 | SHMEM_HUGE_NEVER, | |
4031 | SHMEM_HUGE_DENY, | |
4032 | SHMEM_HUGE_FORCE, | |
4033 | }; | |
79d4d38a JP |
4034 | int len = 0; |
4035 | int i; | |
5a6e75f8 | 4036 | |
79d4d38a JP |
4037 | for (i = 0; i < ARRAY_SIZE(values); i++) { |
4038 | len += sysfs_emit_at(buf, len, | |
4039 | shmem_huge == values[i] ? "%s[%s]" : "%s%s", | |
4040 | i ? " " : "", | |
4041 | shmem_format_huge(values[i])); | |
5a6e75f8 | 4042 | } |
79d4d38a JP |
4043 | |
4044 | len += sysfs_emit_at(buf, len, "\n"); | |
4045 | ||
4046 | return len; | |
5a6e75f8 KS |
4047 | } |
4048 | ||
4049 | static ssize_t shmem_enabled_store(struct kobject *kobj, | |
4050 | struct kobj_attribute *attr, const char *buf, size_t count) | |
4051 | { | |
4052 | char tmp[16]; | |
4053 | int huge; | |
4054 | ||
4055 | if (count + 1 > sizeof(tmp)) | |
4056 | return -EINVAL; | |
4057 | memcpy(tmp, buf, count); | |
4058 | tmp[count] = '\0'; | |
4059 | if (count && tmp[count - 1] == '\n') | |
4060 | tmp[count - 1] = '\0'; | |
4061 | ||
4062 | huge = shmem_parse_huge(tmp); | |
4063 | if (huge == -EINVAL) | |
4064 | return -EINVAL; | |
4065 | if (!has_transparent_hugepage() && | |
4066 | huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY) | |
4067 | return -EINVAL; | |
4068 | ||
4069 | shmem_huge = huge; | |
435c0b87 | 4070 | if (shmem_huge > SHMEM_HUGE_DENY) |
5a6e75f8 KS |
4071 | SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge; |
4072 | return count; | |
4073 | } | |
4074 | ||
4075 | struct kobj_attribute shmem_enabled_attr = | |
4076 | __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store); | |
396bcc52 | 4077 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */ |
f3f0e1d2 | 4078 | |
396bcc52 | 4079 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
f3f0e1d2 KS |
4080 | bool shmem_huge_enabled(struct vm_area_struct *vma) |
4081 | { | |
4082 | struct inode *inode = file_inode(vma->vm_file); | |
4083 | struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); | |
4084 | loff_t i_size; | |
4085 | pgoff_t off; | |
4086 | ||
c0630669 YS |
4087 | if ((vma->vm_flags & VM_NOHUGEPAGE) || |
4088 | test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) | |
4089 | return false; | |
f3f0e1d2 KS |
4090 | if (shmem_huge == SHMEM_HUGE_FORCE) |
4091 | return true; | |
4092 | if (shmem_huge == SHMEM_HUGE_DENY) | |
4093 | return false; | |
4094 | switch (sbinfo->huge) { | |
4095 | case SHMEM_HUGE_NEVER: | |
4096 | return false; | |
4097 | case SHMEM_HUGE_ALWAYS: | |
4098 | return true; | |
4099 | case SHMEM_HUGE_WITHIN_SIZE: | |
4100 | off = round_up(vma->vm_pgoff, HPAGE_PMD_NR); | |
4101 | i_size = round_up(i_size_read(inode), PAGE_SIZE); | |
4102 | if (i_size >= HPAGE_PMD_SIZE && | |
4103 | i_size >> PAGE_SHIFT >= off) | |
4104 | return true; | |
e4a9bc58 | 4105 | fallthrough; |
f3f0e1d2 KS |
4106 | case SHMEM_HUGE_ADVISE: |
4107 | /* TODO: implement fadvise() hints */ | |
4108 | return (vma->vm_flags & VM_HUGEPAGE); | |
4109 | default: | |
4110 | VM_BUG_ON(1); | |
4111 | return false; | |
4112 | } | |
4113 | } | |
396bcc52 | 4114 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
5a6e75f8 | 4115 | |
853ac43a MM |
4116 | #else /* !CONFIG_SHMEM */ |
4117 | ||
4118 | /* | |
4119 | * tiny-shmem: simple shmemfs and tmpfs using ramfs code | |
4120 | * | |
4121 | * This is intended for small system where the benefits of the full | |
4122 | * shmem code (swap-backed and resource-limited) are outweighed by | |
4123 | * their complexity. On systems without swap this code should be | |
4124 | * effectively equivalent, but much lighter weight. | |
4125 | */ | |
4126 | ||
41ffe5d5 | 4127 | static struct file_system_type shmem_fs_type = { |
853ac43a | 4128 | .name = "tmpfs", |
f3235626 | 4129 | .init_fs_context = ramfs_init_fs_context, |
d7167b14 | 4130 | .parameters = ramfs_fs_parameters, |
853ac43a | 4131 | .kill_sb = kill_litter_super, |
2b8576cb | 4132 | .fs_flags = FS_USERNS_MOUNT, |
853ac43a MM |
4133 | }; |
4134 | ||
41ffe5d5 | 4135 | int __init shmem_init(void) |
853ac43a | 4136 | { |
41ffe5d5 | 4137 | BUG_ON(register_filesystem(&shmem_fs_type) != 0); |
853ac43a | 4138 | |
41ffe5d5 | 4139 | shm_mnt = kern_mount(&shmem_fs_type); |
853ac43a MM |
4140 | BUG_ON(IS_ERR(shm_mnt)); |
4141 | ||
4142 | return 0; | |
4143 | } | |
4144 | ||
b56a2d8a VRP |
4145 | int shmem_unuse(unsigned int type, bool frontswap, |
4146 | unsigned long *fs_pages_to_unuse) | |
853ac43a MM |
4147 | { |
4148 | return 0; | |
4149 | } | |
4150 | ||
3f96b79a HD |
4151 | int shmem_lock(struct file *file, int lock, struct user_struct *user) |
4152 | { | |
4153 | return 0; | |
4154 | } | |
4155 | ||
24513264 HD |
4156 | void shmem_unlock_mapping(struct address_space *mapping) |
4157 | { | |
4158 | } | |
4159 | ||
c01d5b30 HD |
4160 | #ifdef CONFIG_MMU |
4161 | unsigned long shmem_get_unmapped_area(struct file *file, | |
4162 | unsigned long addr, unsigned long len, | |
4163 | unsigned long pgoff, unsigned long flags) | |
4164 | { | |
4165 | return current->mm->get_unmapped_area(file, addr, len, pgoff, flags); | |
4166 | } | |
4167 | #endif | |
4168 | ||
41ffe5d5 | 4169 | void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) |
94c1e62d | 4170 | { |
41ffe5d5 | 4171 | truncate_inode_pages_range(inode->i_mapping, lstart, lend); |
94c1e62d HD |
4172 | } |
4173 | EXPORT_SYMBOL_GPL(shmem_truncate_range); | |
4174 | ||
0b0a0806 HD |
4175 | #define shmem_vm_ops generic_file_vm_ops |
4176 | #define shmem_file_operations ramfs_file_operations | |
454abafe | 4177 | #define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) |
0b0a0806 HD |
4178 | #define shmem_acct_size(flags, size) 0 |
4179 | #define shmem_unacct_size(flags, size) do {} while (0) | |
853ac43a MM |
4180 | |
4181 | #endif /* CONFIG_SHMEM */ | |
4182 | ||
4183 | /* common code */ | |
1da177e4 | 4184 | |
703321b6 | 4185 | static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size, |
c7277090 | 4186 | unsigned long flags, unsigned int i_flags) |
1da177e4 | 4187 | { |
1da177e4 | 4188 | struct inode *inode; |
93dec2da | 4189 | struct file *res; |
1da177e4 | 4190 | |
703321b6 MA |
4191 | if (IS_ERR(mnt)) |
4192 | return ERR_CAST(mnt); | |
1da177e4 | 4193 | |
285b2c4f | 4194 | if (size < 0 || size > MAX_LFS_FILESIZE) |
1da177e4 LT |
4195 | return ERR_PTR(-EINVAL); |
4196 | ||
4197 | if (shmem_acct_size(flags, size)) | |
4198 | return ERR_PTR(-ENOMEM); | |
4199 | ||
93dec2da AV |
4200 | inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0, |
4201 | flags); | |
dac2d1f6 AV |
4202 | if (unlikely(!inode)) { |
4203 | shmem_unacct_size(flags, size); | |
4204 | return ERR_PTR(-ENOSPC); | |
4205 | } | |
c7277090 | 4206 | inode->i_flags |= i_flags; |
1da177e4 | 4207 | inode->i_size = size; |
6d6b77f1 | 4208 | clear_nlink(inode); /* It is unlinked */ |
26567cdb | 4209 | res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size)); |
93dec2da AV |
4210 | if (!IS_ERR(res)) |
4211 | res = alloc_file_pseudo(inode, mnt, name, O_RDWR, | |
4212 | &shmem_file_operations); | |
26567cdb | 4213 | if (IS_ERR(res)) |
93dec2da | 4214 | iput(inode); |
6b4d0b27 | 4215 | return res; |
1da177e4 | 4216 | } |
c7277090 EP |
4217 | |
4218 | /** | |
4219 | * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be | |
4220 | * kernel internal. There will be NO LSM permission checks against the | |
4221 | * underlying inode. So users of this interface must do LSM checks at a | |
e1832f29 SS |
4222 | * higher layer. The users are the big_key and shm implementations. LSM |
4223 | * checks are provided at the key or shm level rather than the inode. | |
c7277090 EP |
4224 | * @name: name for dentry (to be seen in /proc/<pid>/maps |
4225 | * @size: size to be set for the file | |
4226 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4227 | */ | |
4228 | struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags) | |
4229 | { | |
703321b6 | 4230 | return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE); |
c7277090 EP |
4231 | } |
4232 | ||
4233 | /** | |
4234 | * shmem_file_setup - get an unlinked file living in tmpfs | |
4235 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4236 | * @size: size to be set for the file | |
4237 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4238 | */ | |
4239 | struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) | |
4240 | { | |
703321b6 | 4241 | return __shmem_file_setup(shm_mnt, name, size, flags, 0); |
c7277090 | 4242 | } |
395e0ddc | 4243 | EXPORT_SYMBOL_GPL(shmem_file_setup); |
1da177e4 | 4244 | |
703321b6 MA |
4245 | /** |
4246 | * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs | |
4247 | * @mnt: the tmpfs mount where the file will be created | |
4248 | * @name: name for dentry (to be seen in /proc/<pid>/maps | |
4249 | * @size: size to be set for the file | |
4250 | * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size | |
4251 | */ | |
4252 | struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name, | |
4253 | loff_t size, unsigned long flags) | |
4254 | { | |
4255 | return __shmem_file_setup(mnt, name, size, flags, 0); | |
4256 | } | |
4257 | EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt); | |
4258 | ||
46711810 | 4259 | /** |
1da177e4 | 4260 | * shmem_zero_setup - setup a shared anonymous mapping |
45e55300 | 4261 | * @vma: the vma to be mmapped is prepared by do_mmap |
1da177e4 LT |
4262 | */ |
4263 | int shmem_zero_setup(struct vm_area_struct *vma) | |
4264 | { | |
4265 | struct file *file; | |
4266 | loff_t size = vma->vm_end - vma->vm_start; | |
4267 | ||
66fc1303 | 4268 | /* |
c1e8d7c6 | 4269 | * Cloning a new file under mmap_lock leads to a lock ordering conflict |
66fc1303 HD |
4270 | * between XFS directory reading and selinux: since this file is only |
4271 | * accessible to the user through its mapping, use S_PRIVATE flag to | |
4272 | * bypass file security, in the same way as shmem_kernel_file_setup(). | |
4273 | */ | |
703321b6 | 4274 | file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags); |
1da177e4 LT |
4275 | if (IS_ERR(file)) |
4276 | return PTR_ERR(file); | |
4277 | ||
4278 | if (vma->vm_file) | |
4279 | fput(vma->vm_file); | |
4280 | vma->vm_file = file; | |
4281 | vma->vm_ops = &shmem_vm_ops; | |
f3f0e1d2 | 4282 | |
396bcc52 | 4283 | if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && |
f3f0e1d2 KS |
4284 | ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) < |
4285 | (vma->vm_end & HPAGE_PMD_MASK)) { | |
4286 | khugepaged_enter(vma, vma->vm_flags); | |
4287 | } | |
4288 | ||
1da177e4 LT |
4289 | return 0; |
4290 | } | |
d9d90e5e HD |
4291 | |
4292 | /** | |
4293 | * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. | |
4294 | * @mapping: the page's address_space | |
4295 | * @index: the page index | |
4296 | * @gfp: the page allocator flags to use if allocating | |
4297 | * | |
4298 | * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", | |
4299 | * with any new page allocations done using the specified allocation flags. | |
4300 | * But read_cache_page_gfp() uses the ->readpage() method: which does not | |
4301 | * suit tmpfs, since it may have pages in swapcache, and needs to find those | |
4302 | * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. | |
4303 | * | |
68da9f05 HD |
4304 | * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in |
4305 | * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. | |
d9d90e5e HD |
4306 | */ |
4307 | struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, | |
4308 | pgoff_t index, gfp_t gfp) | |
4309 | { | |
68da9f05 HD |
4310 | #ifdef CONFIG_SHMEM |
4311 | struct inode *inode = mapping->host; | |
9276aad6 | 4312 | struct page *page; |
68da9f05 HD |
4313 | int error; |
4314 | ||
30e6a51d | 4315 | BUG_ON(!shmem_mapping(mapping)); |
9e18eb29 | 4316 | error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, |
cfda0526 | 4317 | gfp, NULL, NULL, NULL); |
68da9f05 HD |
4318 | if (error) |
4319 | page = ERR_PTR(error); | |
4320 | else | |
4321 | unlock_page(page); | |
4322 | return page; | |
4323 | #else | |
4324 | /* | |
4325 | * The tiny !SHMEM case uses ramfs without swap | |
4326 | */ | |
d9d90e5e | 4327 | return read_cache_page_gfp(mapping, index, gfp); |
68da9f05 | 4328 | #endif |
d9d90e5e HD |
4329 | } |
4330 | EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); |