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