?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'regmap-fix-v5.11-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux/fpc-iii.git] / fs / nfs / inode.c
blob522aa10a1a3e7cfe64f07f79786c55bd6a35d84a
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/fs/nfs/inode.c
4 *
5 * Copyright (C) 1992 Rick Sladkey
6 *
7 * nfs inode and superblock handling functions
8 *
9 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
10 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 *
12 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
13 * J.S.Peatfield@damtp.cam.ac.uk
14 *
15 */
16
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/sched/signal.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/sunrpc/metrics.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/nfs4_mount.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 #include <linux/compat.h>
41 #include <linux/freezer.h>
42 #include <linux/uaccess.h>
43 #include <linux/iversion.h>
44
45 #include "nfs4_fs.h"
46 #include "callback.h"
47 #include "delegation.h"
48 #include "iostat.h"
49 #include "internal.h"
50 #include "fscache.h"
51 #include "pnfs.h"
52 #include "nfs.h"
53 #include "netns.h"
54 #include "sysfs.h"
55
56 #include "nfstrace.h"
57
58 #define NFSDBG_FACILITY NFSDBG_VFS
59
60 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
61
62 /* Default is to see 64-bit inode numbers */
63 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
64
65 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
66
67 static struct kmem_cache * nfs_inode_cachep;
68
69 static inline unsigned long
70 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71 {
72 return nfs_fileid_to_ino_t(fattr->fileid);
73 }
74
75 static int nfs_wait_killable(int mode)
76 {
77 freezable_schedule_unsafe();
78 if (signal_pending_state(mode, current))
79 return -ERESTARTSYS;
80 return 0;
81 }
82
83 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
84 {
85 return nfs_wait_killable(mode);
86 }
87 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
88
89 /**
90 * nfs_compat_user_ino64 - returns the user-visible inode number
91 * @fileid: 64-bit fileid
92 *
93 * This function returns a 32-bit inode number if the boot parameter
94 * nfs.enable_ino64 is zero.
95 */
96 u64 nfs_compat_user_ino64(u64 fileid)
97 {
98 #ifdef CONFIG_COMPAT
99 compat_ulong_t ino;
100 #else
101 unsigned long ino;
102 #endif
103
104 if (enable_ino64)
105 return fileid;
106 ino = fileid;
107 if (sizeof(ino) < sizeof(fileid))
108 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
109 return ino;
110 }
111
112 int nfs_drop_inode(struct inode *inode)
113 {
114 return NFS_STALE(inode) || generic_drop_inode(inode);
115 }
116 EXPORT_SYMBOL_GPL(nfs_drop_inode);
117
118 void nfs_clear_inode(struct inode *inode)
119 {
120 /*
121 * The following should never happen...
122 */
123 WARN_ON_ONCE(nfs_have_writebacks(inode));
124 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
125 nfs_zap_acl_cache(inode);
126 nfs_access_zap_cache(inode);
127 nfs_fscache_clear_inode(inode);
128 }
129 EXPORT_SYMBOL_GPL(nfs_clear_inode);
130
131 void nfs_evict_inode(struct inode *inode)
132 {
133 truncate_inode_pages_final(&inode->i_data);
134 clear_inode(inode);
135 nfs_clear_inode(inode);
136 }
137
138 int nfs_sync_inode(struct inode *inode)
139 {
140 inode_dio_wait(inode);
141 return nfs_wb_all(inode);
142 }
143 EXPORT_SYMBOL_GPL(nfs_sync_inode);
144
145 /**
146 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
147 * @mapping: pointer to struct address_space
148 */
149 int nfs_sync_mapping(struct address_space *mapping)
150 {
151 int ret = 0;
152
153 if (mapping->nrpages != 0) {
154 unmap_mapping_range(mapping, 0, 0, 0);
155 ret = nfs_wb_all(mapping->host);
156 }
157 return ret;
158 }
159
160 static int nfs_attribute_timeout(struct inode *inode)
161 {
162 struct nfs_inode *nfsi = NFS_I(inode);
163
164 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
165 }
166
167 static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
168 {
169 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
170
171 /* Special case for the pagecache or access cache */
172 if (flags == NFS_INO_REVAL_PAGECACHE &&
173 !(cache_validity & NFS_INO_REVAL_FORCED))
174 return false;
175 return (cache_validity & flags) != 0;
176 }
177
178 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
179 {
180 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
181
182 if ((cache_validity & flags) != 0)
183 return true;
184 if (nfs_attribute_timeout(inode))
185 return true;
186 return false;
187 }
188
189 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
190 {
191 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
192 return nfs_check_cache_invalid_delegated(inode, flags);
193
194 return nfs_check_cache_invalid_not_delegated(inode, flags);
195 }
196 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
197
198 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
199 {
200 struct nfs_inode *nfsi = NFS_I(inode);
201 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
202
203 if (have_delegation) {
204 if (!(flags & NFS_INO_REVAL_FORCED))
205 flags &= ~NFS_INO_INVALID_OTHER;
206 flags &= ~(NFS_INO_INVALID_CHANGE
207 | NFS_INO_INVALID_SIZE
208 | NFS_INO_REVAL_PAGECACHE
209 | NFS_INO_INVALID_XATTR);
210 }
211
212 if (inode->i_mapping->nrpages == 0)
213 flags &= ~(NFS_INO_INVALID_DATA|NFS_INO_DATA_INVAL_DEFER);
214 nfsi->cache_validity |= flags;
215 if (flags & NFS_INO_INVALID_DATA)
216 nfs_fscache_invalidate(inode);
217 }
218
219 /*
220 * Invalidate the local caches
221 */
222 static void nfs_zap_caches_locked(struct inode *inode)
223 {
224 struct nfs_inode *nfsi = NFS_I(inode);
225 int mode = inode->i_mode;
226
227 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
228
229 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
230 nfsi->attrtimeo_timestamp = jiffies;
231
232 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
233 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
234 | NFS_INO_INVALID_DATA
235 | NFS_INO_INVALID_ACCESS
236 | NFS_INO_INVALID_ACL
237 | NFS_INO_INVALID_XATTR
238 | NFS_INO_REVAL_PAGECACHE);
239 } else
240 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
241 | NFS_INO_INVALID_ACCESS
242 | NFS_INO_INVALID_ACL
243 | NFS_INO_INVALID_XATTR
244 | NFS_INO_REVAL_PAGECACHE);
245 nfs_zap_label_cache_locked(nfsi);
246 }
247
248 void nfs_zap_caches(struct inode *inode)
249 {
250 spin_lock(&inode->i_lock);
251 nfs_zap_caches_locked(inode);
252 spin_unlock(&inode->i_lock);
253 }
254
255 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
256 {
257 if (mapping->nrpages != 0) {
258 spin_lock(&inode->i_lock);
259 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
260 spin_unlock(&inode->i_lock);
261 }
262 }
263
264 void nfs_zap_acl_cache(struct inode *inode)
265 {
266 void (*clear_acl_cache)(struct inode *);
267
268 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
269 if (clear_acl_cache != NULL)
270 clear_acl_cache(inode);
271 spin_lock(&inode->i_lock);
272 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
273 spin_unlock(&inode->i_lock);
274 }
275 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
276
277 void nfs_invalidate_atime(struct inode *inode)
278 {
279 spin_lock(&inode->i_lock);
280 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
281 spin_unlock(&inode->i_lock);
282 }
283 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
284
285 /*
286 * Invalidate, but do not unhash, the inode.
287 * NB: must be called with inode->i_lock held!
288 */
289 static void nfs_set_inode_stale_locked(struct inode *inode)
290 {
291 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
292 nfs_zap_caches_locked(inode);
293 trace_nfs_set_inode_stale(inode);
294 }
295
296 void nfs_set_inode_stale(struct inode *inode)
297 {
298 spin_lock(&inode->i_lock);
299 nfs_set_inode_stale_locked(inode);
300 spin_unlock(&inode->i_lock);
301 }
302
303 struct nfs_find_desc {
304 struct nfs_fh *fh;
305 struct nfs_fattr *fattr;
306 };
307
308 /*
309 * In NFSv3 we can have 64bit inode numbers. In order to support
310 * this, and re-exported directories (also seen in NFSv2)
311 * we are forced to allow 2 different inodes to have the same
312 * i_ino.
313 */
314 static int
315 nfs_find_actor(struct inode *inode, void *opaque)
316 {
317 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
318 struct nfs_fh *fh = desc->fh;
319 struct nfs_fattr *fattr = desc->fattr;
320
321 if (NFS_FILEID(inode) != fattr->fileid)
322 return 0;
323 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
324 return 0;
325 if (nfs_compare_fh(NFS_FH(inode), fh))
326 return 0;
327 if (is_bad_inode(inode) || NFS_STALE(inode))
328 return 0;
329 return 1;
330 }
331
332 static int
333 nfs_init_locked(struct inode *inode, void *opaque)
334 {
335 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
336 struct nfs_fattr *fattr = desc->fattr;
337
338 set_nfs_fileid(inode, fattr->fileid);
339 inode->i_mode = fattr->mode;
340 nfs_copy_fh(NFS_FH(inode), desc->fh);
341 return 0;
342 }
343
344 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
345 static void nfs_clear_label_invalid(struct inode *inode)
346 {
347 spin_lock(&inode->i_lock);
348 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
349 spin_unlock(&inode->i_lock);
350 }
351
352 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
353 struct nfs4_label *label)
354 {
355 int error;
356
357 if (label == NULL)
358 return;
359
360 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
361 error = security_inode_notifysecctx(inode, label->label,
362 label->len);
363 if (error)
364 printk(KERN_ERR "%s() %s %d "
365 "security_inode_notifysecctx() %d\n",
366 __func__,
367 (char *)label->label,
368 label->len, error);
369 nfs_clear_label_invalid(inode);
370 }
371 }
372
373 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
374 {
375 struct nfs4_label *label = NULL;
376 int minor_version = server->nfs_client->cl_minorversion;
377
378 if (minor_version < 2)
379 return label;
380
381 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
382 return label;
383
384 label = kzalloc(sizeof(struct nfs4_label), flags);
385 if (label == NULL)
386 return ERR_PTR(-ENOMEM);
387
388 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
389 if (label->label == NULL) {
390 kfree(label);
391 return ERR_PTR(-ENOMEM);
392 }
393 label->len = NFS4_MAXLABELLEN;
394
395 return label;
396 }
397 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
398 #else
399 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
400 struct nfs4_label *label)
401 {
402 }
403 #endif
404 EXPORT_SYMBOL_GPL(nfs_setsecurity);
405
406 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
407 struct inode *
408 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
409 {
410 struct nfs_find_desc desc = {
411 .fh = fh,
412 .fattr = fattr,
413 };
414 struct inode *inode;
415 unsigned long hash;
416
417 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
418 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
419 return NULL;
420
421 hash = nfs_fattr_to_ino_t(fattr);
422 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
423
424 dprintk("%s: returning %p\n", __func__, inode);
425 return inode;
426 }
427
428 /*
429 * This is our front-end to iget that looks up inodes by file handle
430 * instead of inode number.
431 */
432 struct inode *
433 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
434 {
435 struct nfs_find_desc desc = {
436 .fh = fh,
437 .fattr = fattr
438 };
439 struct inode *inode = ERR_PTR(-ENOENT);
440 unsigned long hash;
441
442 nfs_attr_check_mountpoint(sb, fattr);
443
444 if (nfs_attr_use_mounted_on_fileid(fattr))
445 fattr->fileid = fattr->mounted_on_fileid;
446 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
447 goto out_no_inode;
448 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
449 goto out_no_inode;
450
451 hash = nfs_fattr_to_ino_t(fattr);
452
453 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
454 if (inode == NULL) {
455 inode = ERR_PTR(-ENOMEM);
456 goto out_no_inode;
457 }
458
459 if (inode->i_state & I_NEW) {
460 struct nfs_inode *nfsi = NFS_I(inode);
461 unsigned long now = jiffies;
462
463 /* We set i_ino for the few things that still rely on it,
464 * such as stat(2) */
465 inode->i_ino = hash;
466
467 /* We can't support update_atime(), since the server will reset it */
468 inode->i_flags |= S_NOATIME|S_NOCMTIME;
469 inode->i_mode = fattr->mode;
470 nfsi->cache_validity = 0;
471 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
472 && nfs_server_capable(inode, NFS_CAP_MODE))
473 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
474 /* Why so? Because we want revalidate for devices/FIFOs, and
475 * that's precisely what we have in nfs_file_inode_operations.
476 */
477 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
478 if (S_ISREG(inode->i_mode)) {
479 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
480 inode->i_data.a_ops = &nfs_file_aops;
481 } else if (S_ISDIR(inode->i_mode)) {
482 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
483 inode->i_fop = &nfs_dir_operations;
484 inode->i_data.a_ops = &nfs_dir_aops;
485 /* Deal with crossing mountpoints */
486 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
487 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
488 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
489 inode->i_op = &nfs_referral_inode_operations;
490 else
491 inode->i_op = &nfs_mountpoint_inode_operations;
492 inode->i_fop = NULL;
493 inode->i_flags |= S_AUTOMOUNT;
494 }
495 } else if (S_ISLNK(inode->i_mode)) {
496 inode->i_op = &nfs_symlink_inode_operations;
497 inode_nohighmem(inode);
498 } else
499 init_special_inode(inode, inode->i_mode, fattr->rdev);
500
501 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
502 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
503 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
504 inode_set_iversion_raw(inode, 0);
505 inode->i_size = 0;
506 clear_nlink(inode);
507 inode->i_uid = make_kuid(&init_user_ns, -2);
508 inode->i_gid = make_kgid(&init_user_ns, -2);
509 inode->i_blocks = 0;
510 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
511 nfsi->write_io = 0;
512 nfsi->read_io = 0;
513
514 nfsi->read_cache_jiffies = fattr->time_start;
515 nfsi->attr_gencount = fattr->gencount;
516 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
517 inode->i_atime = fattr->atime;
518 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
519 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
520 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
521 inode->i_mtime = fattr->mtime;
522 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
523 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
524 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
525 inode->i_ctime = fattr->ctime;
526 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
527 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
528 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
529 inode_set_iversion_raw(inode, fattr->change_attr);
530 else
531 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
532 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
533 inode->i_size = nfs_size_to_loff_t(fattr->size);
534 else
535 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
536 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
537 set_nlink(inode, fattr->nlink);
538 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
539 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
540 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
541 inode->i_uid = fattr->uid;
542 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
543 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
544 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
545 inode->i_gid = fattr->gid;
546 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
547 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
548 if (nfs_server_capable(inode, NFS_CAP_XATTR))
549 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
550 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
551 inode->i_blocks = fattr->du.nfs2.blocks;
552 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
553 /*
554 * report the blocks in 512byte units
555 */
556 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
557 }
558
559 if (nfsi->cache_validity != 0)
560 nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
561
562 nfs_setsecurity(inode, fattr, label);
563
564 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
565 nfsi->attrtimeo_timestamp = now;
566 nfsi->access_cache = RB_ROOT;
567
568 nfs_fscache_init_inode(inode);
569
570 unlock_new_inode(inode);
571 } else {
572 int err = nfs_refresh_inode(inode, fattr);
573 if (err < 0) {
574 iput(inode);
575 inode = ERR_PTR(err);
576 goto out_no_inode;
577 }
578 }
579 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
580 inode->i_sb->s_id,
581 (unsigned long long)NFS_FILEID(inode),
582 nfs_display_fhandle_hash(fh),
583 atomic_read(&inode->i_count));
584
585 out:
586 return inode;
587
588 out_no_inode:
589 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
590 goto out;
591 }
592 EXPORT_SYMBOL_GPL(nfs_fhget);
593
594 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN)
595
596 int
597 nfs_setattr(struct dentry *dentry, struct iattr *attr)
598 {
599 struct inode *inode = d_inode(dentry);
600 struct nfs_fattr *fattr;
601 int error = 0;
602
603 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
604
605 /* skip mode change if it's just for clearing setuid/setgid */
606 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
607 attr->ia_valid &= ~ATTR_MODE;
608
609 if (attr->ia_valid & ATTR_SIZE) {
610 BUG_ON(!S_ISREG(inode->i_mode));
611
612 error = inode_newsize_ok(inode, attr->ia_size);
613 if (error)
614 return error;
615
616 if (attr->ia_size == i_size_read(inode))
617 attr->ia_valid &= ~ATTR_SIZE;
618 }
619
620 /* Optimization: if the end result is no change, don't RPC */
621 attr->ia_valid &= NFS_VALID_ATTRS;
622 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
623 return 0;
624
625 trace_nfs_setattr_enter(inode);
626
627 /* Write all dirty data */
628 if (S_ISREG(inode->i_mode))
629 nfs_sync_inode(inode);
630
631 fattr = nfs_alloc_fattr();
632 if (fattr == NULL) {
633 error = -ENOMEM;
634 goto out;
635 }
636
637 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
638 if (error == 0)
639 error = nfs_refresh_inode(inode, fattr);
640 nfs_free_fattr(fattr);
641 out:
642 trace_nfs_setattr_exit(inode, error);
643 return error;
644 }
645 EXPORT_SYMBOL_GPL(nfs_setattr);
646
647 /**
648 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
649 * @inode: inode of the file used
650 * @offset: file offset to start truncating
651 *
652 * This is a copy of the common vmtruncate, but with the locking
653 * corrected to take into account the fact that NFS requires
654 * inode->i_size to be updated under the inode->i_lock.
655 * Note: must be called with inode->i_lock held!
656 */
657 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
658 {
659 int err;
660
661 err = inode_newsize_ok(inode, offset);
662 if (err)
663 goto out;
664
665 i_size_write(inode, offset);
666 /* Optimisation */
667 if (offset == 0)
668 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
669 NFS_INO_DATA_INVAL_DEFER);
670 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
671
672 spin_unlock(&inode->i_lock);
673 truncate_pagecache(inode, offset);
674 spin_lock(&inode->i_lock);
675 out:
676 return err;
677 }
678
679 /**
680 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
681 * @inode: pointer to struct inode
682 * @attr: pointer to struct iattr
683 * @fattr: pointer to struct nfs_fattr
684 *
685 * Note: we do this in the *proc.c in order to ensure that
686 * it works for things like exclusive creates too.
687 */
688 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
689 struct nfs_fattr *fattr)
690 {
691 /* Barrier: bump the attribute generation count. */
692 nfs_fattr_set_barrier(fattr);
693
694 spin_lock(&inode->i_lock);
695 NFS_I(inode)->attr_gencount = fattr->gencount;
696 if ((attr->ia_valid & ATTR_SIZE) != 0) {
697 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
698 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
699 nfs_vmtruncate(inode, attr->ia_size);
700 }
701 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
702 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
703 if ((attr->ia_valid & ATTR_MODE) != 0) {
704 int mode = attr->ia_mode & S_IALLUGO;
705 mode |= inode->i_mode & ~S_IALLUGO;
706 inode->i_mode = mode;
707 }
708 if ((attr->ia_valid & ATTR_UID) != 0)
709 inode->i_uid = attr->ia_uid;
710 if ((attr->ia_valid & ATTR_GID) != 0)
711 inode->i_gid = attr->ia_gid;
712 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
713 inode->i_ctime = fattr->ctime;
714 else
715 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
716 | NFS_INO_INVALID_CTIME);
717 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
718 | NFS_INO_INVALID_ACL);
719 }
720 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
721 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
722 | NFS_INO_INVALID_CTIME);
723 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
724 inode->i_atime = fattr->atime;
725 else if (attr->ia_valid & ATTR_ATIME_SET)
726 inode->i_atime = attr->ia_atime;
727 else
728 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
729
730 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
731 inode->i_ctime = fattr->ctime;
732 else
733 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
734 | NFS_INO_INVALID_CTIME);
735 }
736 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
737 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
738 | NFS_INO_INVALID_CTIME);
739 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
740 inode->i_mtime = fattr->mtime;
741 else if (attr->ia_valid & ATTR_MTIME_SET)
742 inode->i_mtime = attr->ia_mtime;
743 else
744 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
745
746 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
747 inode->i_ctime = fattr->ctime;
748 else
749 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
750 | NFS_INO_INVALID_CTIME);
751 }
752 if (fattr->valid)
753 nfs_update_inode(inode, fattr);
754 spin_unlock(&inode->i_lock);
755 }
756 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
757
758 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
759 {
760 struct dentry *parent;
761
762 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
763 return;
764 parent = dget_parent(dentry);
765 nfs_force_use_readdirplus(d_inode(parent));
766 dput(parent);
767 }
768
769 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
770 {
771 struct dentry *parent;
772
773 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
774 return;
775 parent = dget_parent(dentry);
776 nfs_advise_use_readdirplus(d_inode(parent));
777 dput(parent);
778 }
779
780 static bool nfs_need_revalidate_inode(struct inode *inode)
781 {
782 if (NFS_I(inode)->cache_validity &
783 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
784 return true;
785 if (nfs_attribute_cache_expired(inode))
786 return true;
787 return false;
788 }
789
790 int nfs_getattr(const struct path *path, struct kstat *stat,
791 u32 request_mask, unsigned int query_flags)
792 {
793 struct inode *inode = d_inode(path->dentry);
794 struct nfs_server *server = NFS_SERVER(inode);
795 unsigned long cache_validity;
796 int err = 0;
797 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
798 bool do_update = false;
799
800 trace_nfs_getattr_enter(inode);
801
802 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
803 nfs_readdirplus_parent_cache_hit(path->dentry);
804 goto out_no_update;
805 }
806
807 /* Flush out writes to the server in order to update c/mtime. */
808 if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
809 S_ISREG(inode->i_mode)) {
810 err = filemap_write_and_wait(inode->i_mapping);
811 if (err)
812 goto out;
813 }
814
815 /*
816 * We may force a getattr if the user cares about atime.
817 *
818 * Note that we only have to check the vfsmount flags here:
819 * - NFS always sets S_NOATIME by so checking it would give a
820 * bogus result
821 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
822 * no point in checking those.
823 */
824 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
825 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
826 request_mask &= ~STATX_ATIME;
827
828 /* Is the user requesting attributes that might need revalidation? */
829 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
830 STATX_MTIME|STATX_UID|STATX_GID|
831 STATX_SIZE|STATX_BLOCKS)))
832 goto out_no_revalidate;
833
834 /* Check whether the cached attributes are stale */
835 do_update |= force_sync || nfs_attribute_cache_expired(inode);
836 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
837 do_update |= cache_validity &
838 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
839 if (request_mask & STATX_ATIME)
840 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
841 if (request_mask & (STATX_CTIME|STATX_MTIME))
842 do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
843 if (request_mask & STATX_BLOCKS)
844 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
845 if (do_update) {
846 /* Update the attribute cache */
847 if (!(server->flags & NFS_MOUNT_NOAC))
848 nfs_readdirplus_parent_cache_miss(path->dentry);
849 else
850 nfs_readdirplus_parent_cache_hit(path->dentry);
851 err = __nfs_revalidate_inode(server, inode);
852 if (err)
853 goto out;
854 } else
855 nfs_readdirplus_parent_cache_hit(path->dentry);
856 out_no_revalidate:
857 /* Only return attributes that were revalidated. */
858 stat->result_mask &= request_mask;
859 out_no_update:
860 generic_fillattr(inode, stat);
861 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
862 if (S_ISDIR(inode->i_mode))
863 stat->blksize = NFS_SERVER(inode)->dtsize;
864 out:
865 trace_nfs_getattr_exit(inode, err);
866 return err;
867 }
868 EXPORT_SYMBOL_GPL(nfs_getattr);
869
870 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
871 {
872 refcount_set(&l_ctx->count, 1);
873 l_ctx->lockowner = current->files;
874 INIT_LIST_HEAD(&l_ctx->list);
875 atomic_set(&l_ctx->io_count, 0);
876 }
877
878 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
879 {
880 struct nfs_lock_context *pos;
881
882 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
883 if (pos->lockowner != current->files)
884 continue;
885 if (refcount_inc_not_zero(&pos->count))
886 return pos;
887 }
888 return NULL;
889 }
890
891 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
892 {
893 struct nfs_lock_context *res, *new = NULL;
894 struct inode *inode = d_inode(ctx->dentry);
895
896 rcu_read_lock();
897 res = __nfs_find_lock_context(ctx);
898 rcu_read_unlock();
899 if (res == NULL) {
900 new = kmalloc(sizeof(*new), GFP_KERNEL);
901 if (new == NULL)
902 return ERR_PTR(-ENOMEM);
903 nfs_init_lock_context(new);
904 spin_lock(&inode->i_lock);
905 res = __nfs_find_lock_context(ctx);
906 if (res == NULL) {
907 new->open_context = get_nfs_open_context(ctx);
908 if (new->open_context) {
909 list_add_tail_rcu(&new->list,
910 &ctx->lock_context.list);
911 res = new;
912 new = NULL;
913 } else
914 res = ERR_PTR(-EBADF);
915 }
916 spin_unlock(&inode->i_lock);
917 kfree(new);
918 }
919 return res;
920 }
921 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
922
923 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
924 {
925 struct nfs_open_context *ctx = l_ctx->open_context;
926 struct inode *inode = d_inode(ctx->dentry);
927
928 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
929 return;
930 list_del_rcu(&l_ctx->list);
931 spin_unlock(&inode->i_lock);
932 put_nfs_open_context(ctx);
933 kfree_rcu(l_ctx, rcu_head);
934 }
935 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
936
937 /**
938 * nfs_close_context - Common close_context() routine NFSv2/v3
939 * @ctx: pointer to context
940 * @is_sync: is this a synchronous close
941 *
942 * Ensure that the attributes are up to date if we're mounted
943 * with close-to-open semantics and we have cached data that will
944 * need to be revalidated on open.
945 */
946 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
947 {
948 struct nfs_inode *nfsi;
949 struct inode *inode;
950 struct nfs_server *server;
951
952 if (!(ctx->mode & FMODE_WRITE))
953 return;
954 if (!is_sync)
955 return;
956 inode = d_inode(ctx->dentry);
957 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
958 return;
959 nfsi = NFS_I(inode);
960 if (inode->i_mapping->nrpages == 0)
961 return;
962 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
963 return;
964 if (!list_empty(&nfsi->open_files))
965 return;
966 server = NFS_SERVER(inode);
967 if (server->flags & NFS_MOUNT_NOCTO)
968 return;
969 nfs_revalidate_inode(server, inode);
970 }
971 EXPORT_SYMBOL_GPL(nfs_close_context);
972
973 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
974 fmode_t f_mode,
975 struct file *filp)
976 {
977 struct nfs_open_context *ctx;
978
979 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
980 if (!ctx)
981 return ERR_PTR(-ENOMEM);
982 nfs_sb_active(dentry->d_sb);
983 ctx->dentry = dget(dentry);
984 if (filp)
985 ctx->cred = get_cred(filp->f_cred);
986 else
987 ctx->cred = get_current_cred();
988 ctx->ll_cred = NULL;
989 ctx->state = NULL;
990 ctx->mode = f_mode;
991 ctx->flags = 0;
992 ctx->error = 0;
993 ctx->flock_owner = (fl_owner_t)filp;
994 nfs_init_lock_context(&ctx->lock_context);
995 ctx->lock_context.open_context = ctx;
996 INIT_LIST_HEAD(&ctx->list);
997 ctx->mdsthreshold = NULL;
998 return ctx;
999 }
1000 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1001
1002 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1003 {
1004 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1005 return ctx;
1006 return NULL;
1007 }
1008 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1009
1010 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1011 {
1012 struct inode *inode = d_inode(ctx->dentry);
1013 struct super_block *sb = ctx->dentry->d_sb;
1014
1015 if (!refcount_dec_and_test(&ctx->lock_context.count))
1016 return;
1017 if (!list_empty(&ctx->list)) {
1018 spin_lock(&inode->i_lock);
1019 list_del_rcu(&ctx->list);
1020 spin_unlock(&inode->i_lock);
1021 }
1022 if (inode != NULL)
1023 NFS_PROTO(inode)->close_context(ctx, is_sync);
1024 put_cred(ctx->cred);
1025 dput(ctx->dentry);
1026 nfs_sb_deactive(sb);
1027 put_rpccred(ctx->ll_cred);
1028 kfree(ctx->mdsthreshold);
1029 kfree_rcu(ctx, rcu_head);
1030 }
1031
1032 void put_nfs_open_context(struct nfs_open_context *ctx)
1033 {
1034 __put_nfs_open_context(ctx, 0);
1035 }
1036 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1037
1038 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1039 {
1040 __put_nfs_open_context(ctx, 1);
1041 }
1042
1043 /*
1044 * Ensure that mmap has a recent RPC credential for use when writing out
1045 * shared pages
1046 */
1047 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1048 {
1049 struct inode *inode = d_inode(ctx->dentry);
1050 struct nfs_inode *nfsi = NFS_I(inode);
1051
1052 spin_lock(&inode->i_lock);
1053 if (list_empty(&nfsi->open_files) &&
1054 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1055 nfsi->cache_validity |= NFS_INO_INVALID_DATA |
1056 NFS_INO_REVAL_FORCED;
1057 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1058 spin_unlock(&inode->i_lock);
1059 }
1060 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1061
1062 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1063 {
1064 filp->private_data = get_nfs_open_context(ctx);
1065 if (list_empty(&ctx->list))
1066 nfs_inode_attach_open_context(ctx);
1067 }
1068 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1069
1070 /*
1071 * Given an inode, search for an open context with the desired characteristics
1072 */
1073 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1074 {
1075 struct nfs_inode *nfsi = NFS_I(inode);
1076 struct nfs_open_context *pos, *ctx = NULL;
1077
1078 rcu_read_lock();
1079 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1080 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1081 continue;
1082 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1083 continue;
1084 ctx = get_nfs_open_context(pos);
1085 if (ctx)
1086 break;
1087 }
1088 rcu_read_unlock();
1089 return ctx;
1090 }
1091
1092 void nfs_file_clear_open_context(struct file *filp)
1093 {
1094 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1095
1096 if (ctx) {
1097 struct inode *inode = d_inode(ctx->dentry);
1098
1099 /*
1100 * We fatal error on write before. Try to writeback
1101 * every page again.
1102 */
1103 if (ctx->error < 0)
1104 invalidate_inode_pages2(inode->i_mapping);
1105 filp->private_data = NULL;
1106 put_nfs_open_context_sync(ctx);
1107 }
1108 }
1109
1110 /*
1111 * These allocate and release file read/write context information.
1112 */
1113 int nfs_open(struct inode *inode, struct file *filp)
1114 {
1115 struct nfs_open_context *ctx;
1116
1117 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1118 if (IS_ERR(ctx))
1119 return PTR_ERR(ctx);
1120 nfs_file_set_open_context(filp, ctx);
1121 put_nfs_open_context(ctx);
1122 nfs_fscache_open_file(inode, filp);
1123 return 0;
1124 }
1125 EXPORT_SYMBOL_GPL(nfs_open);
1126
1127 /*
1128 * This function is called whenever some part of NFS notices that
1129 * the cached attributes have to be refreshed.
1130 */
1131 int
1132 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1133 {
1134 int status = -ESTALE;
1135 struct nfs4_label *label = NULL;
1136 struct nfs_fattr *fattr = NULL;
1137 struct nfs_inode *nfsi = NFS_I(inode);
1138
1139 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1140 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1141
1142 trace_nfs_revalidate_inode_enter(inode);
1143
1144 if (is_bad_inode(inode))
1145 goto out;
1146 if (NFS_STALE(inode))
1147 goto out;
1148
1149 /* pNFS: Attributes aren't updated until we layoutcommit */
1150 if (S_ISREG(inode->i_mode)) {
1151 status = pnfs_sync_inode(inode, false);
1152 if (status)
1153 goto out;
1154 }
1155
1156 status = -ENOMEM;
1157 fattr = nfs_alloc_fattr();
1158 if (fattr == NULL)
1159 goto out;
1160
1161 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1162
1163 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1164 if (IS_ERR(label)) {
1165 status = PTR_ERR(label);
1166 goto out;
1167 }
1168
1169 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1170 label, inode);
1171 if (status != 0) {
1172 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1173 inode->i_sb->s_id,
1174 (unsigned long long)NFS_FILEID(inode), status);
1175 switch (status) {
1176 case -ETIMEDOUT:
1177 /* A soft timeout occurred. Use cached information? */
1178 if (server->flags & NFS_MOUNT_SOFTREVAL)
1179 status = 0;
1180 break;
1181 case -ESTALE:
1182 if (!S_ISDIR(inode->i_mode))
1183 nfs_set_inode_stale(inode);
1184 else
1185 nfs_zap_caches(inode);
1186 }
1187 goto err_out;
1188 }
1189
1190 status = nfs_refresh_inode(inode, fattr);
1191 if (status) {
1192 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1193 inode->i_sb->s_id,
1194 (unsigned long long)NFS_FILEID(inode), status);
1195 goto err_out;
1196 }
1197
1198 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1199 nfs_zap_acl_cache(inode);
1200
1201 nfs_setsecurity(inode, fattr, label);
1202
1203 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1204 inode->i_sb->s_id,
1205 (unsigned long long)NFS_FILEID(inode));
1206
1207 err_out:
1208 nfs4_label_free(label);
1209 out:
1210 nfs_free_fattr(fattr);
1211 trace_nfs_revalidate_inode_exit(inode, status);
1212 return status;
1213 }
1214
1215 int nfs_attribute_cache_expired(struct inode *inode)
1216 {
1217 if (nfs_have_delegated_attributes(inode))
1218 return 0;
1219 return nfs_attribute_timeout(inode);
1220 }
1221
1222 /**
1223 * nfs_revalidate_inode - Revalidate the inode attributes
1224 * @server: pointer to nfs_server struct
1225 * @inode: pointer to inode struct
1226 *
1227 * Updates inode attribute information by retrieving the data from the server.
1228 */
1229 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1230 {
1231 if (!nfs_need_revalidate_inode(inode))
1232 return NFS_STALE(inode) ? -ESTALE : 0;
1233 return __nfs_revalidate_inode(server, inode);
1234 }
1235 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1236
1237 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1238 {
1239 int ret;
1240
1241 if (mapping->nrpages != 0) {
1242 if (S_ISREG(inode->i_mode)) {
1243 ret = nfs_sync_mapping(mapping);
1244 if (ret < 0)
1245 return ret;
1246 }
1247 ret = invalidate_inode_pages2(mapping);
1248 if (ret < 0)
1249 return ret;
1250 }
1251 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1252 nfs_fscache_wait_on_invalidate(inode);
1253
1254 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1255 inode->i_sb->s_id,
1256 (unsigned long long)NFS_FILEID(inode));
1257 return 0;
1258 }
1259
1260 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1261 {
1262 return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1263 NFS_STALE(inode);
1264 }
1265
1266 int nfs_revalidate_mapping_rcu(struct inode *inode)
1267 {
1268 struct nfs_inode *nfsi = NFS_I(inode);
1269 unsigned long *bitlock = &nfsi->flags;
1270 int ret = 0;
1271
1272 if (IS_SWAPFILE(inode))
1273 goto out;
1274 if (nfs_mapping_need_revalidate_inode(inode)) {
1275 ret = -ECHILD;
1276 goto out;
1277 }
1278 spin_lock(&inode->i_lock);
1279 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1280 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1281 ret = -ECHILD;
1282 spin_unlock(&inode->i_lock);
1283 out:
1284 return ret;
1285 }
1286
1287 /**
1288 * nfs_revalidate_mapping - Revalidate the pagecache
1289 * @inode: pointer to host inode
1290 * @mapping: pointer to mapping
1291 */
1292 int nfs_revalidate_mapping(struct inode *inode,
1293 struct address_space *mapping)
1294 {
1295 struct nfs_inode *nfsi = NFS_I(inode);
1296 unsigned long *bitlock = &nfsi->flags;
1297 int ret = 0;
1298
1299 /* swapfiles are not supposed to be shared. */
1300 if (IS_SWAPFILE(inode))
1301 goto out;
1302
1303 if (nfs_mapping_need_revalidate_inode(inode)) {
1304 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1305 if (ret < 0)
1306 goto out;
1307 }
1308
1309 /*
1310 * We must clear NFS_INO_INVALID_DATA first to ensure that
1311 * invalidations that come in while we're shooting down the mappings
1312 * are respected. But, that leaves a race window where one revalidator
1313 * can clear the flag, and then another checks it before the mapping
1314 * gets invalidated. Fix that by serializing access to this part of
1315 * the function.
1316 *
1317 * At the same time, we need to allow other tasks to see whether we
1318 * might be in the middle of invalidating the pages, so we only set
1319 * the bit lock here if it looks like we're going to be doing that.
1320 */
1321 for (;;) {
1322 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1323 nfs_wait_bit_killable, TASK_KILLABLE);
1324 if (ret)
1325 goto out;
1326 spin_lock(&inode->i_lock);
1327 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1328 spin_unlock(&inode->i_lock);
1329 continue;
1330 }
1331 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1332 break;
1333 spin_unlock(&inode->i_lock);
1334 goto out;
1335 }
1336
1337 set_bit(NFS_INO_INVALIDATING, bitlock);
1338 smp_wmb();
1339 nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA|
1340 NFS_INO_DATA_INVAL_DEFER);
1341 spin_unlock(&inode->i_lock);
1342 trace_nfs_invalidate_mapping_enter(inode);
1343 ret = nfs_invalidate_mapping(inode, mapping);
1344 trace_nfs_invalidate_mapping_exit(inode, ret);
1345
1346 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1347 smp_mb__after_atomic();
1348 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1349 out:
1350 return ret;
1351 }
1352
1353 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1354 {
1355 struct inode *inode = &nfsi->vfs_inode;
1356
1357 if (!S_ISREG(inode->i_mode))
1358 return false;
1359 if (list_empty(&nfsi->open_files))
1360 return false;
1361 return inode_is_open_for_write(inode);
1362 }
1363
1364 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1365 {
1366 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1367 }
1368
1369 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1370 {
1371 struct timespec64 ts;
1372
1373 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1374 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1375 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1376 inode_set_iversion_raw(inode, fattr->change_attr);
1377 if (S_ISDIR(inode->i_mode))
1378 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1379 else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1380 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1381 }
1382 /* If we have atomic WCC data, we may update some attributes */
1383 ts = inode->i_ctime;
1384 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1385 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1386 && timespec64_equal(&ts, &fattr->pre_ctime)) {
1387 inode->i_ctime = fattr->ctime;
1388 }
1389
1390 ts = inode->i_mtime;
1391 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1392 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1393 && timespec64_equal(&ts, &fattr->pre_mtime)) {
1394 inode->i_mtime = fattr->mtime;
1395 if (S_ISDIR(inode->i_mode))
1396 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1397 }
1398 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1399 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1400 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1401 && !nfs_have_writebacks(inode)) {
1402 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1403 }
1404 }
1405
1406 /**
1407 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1408 * @inode: pointer to inode
1409 * @fattr: updated attributes
1410 *
1411 * Verifies the attribute cache. If we have just changed the attributes,
1412 * so that fattr carries weak cache consistency data, then it may
1413 * also update the ctime/mtime/change_attribute.
1414 */
1415 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1416 {
1417 struct nfs_inode *nfsi = NFS_I(inode);
1418 loff_t cur_size, new_isize;
1419 unsigned long invalid = 0;
1420 struct timespec64 ts;
1421
1422 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1423 return 0;
1424
1425 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1426 /* Only a mounted-on-fileid? Just exit */
1427 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1428 return 0;
1429 /* Has the inode gone and changed behind our back? */
1430 } else if (nfsi->fileid != fattr->fileid) {
1431 /* Is this perhaps the mounted-on fileid? */
1432 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1433 nfsi->fileid == fattr->mounted_on_fileid)
1434 return 0;
1435 return -ESTALE;
1436 }
1437 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1438 return -ESTALE;
1439
1440
1441 if (!nfs_file_has_buffered_writers(nfsi)) {
1442 /* Verify a few of the more important attributes */
1443 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1444 invalid |= NFS_INO_INVALID_CHANGE
1445 | NFS_INO_REVAL_PAGECACHE;
1446
1447 ts = inode->i_mtime;
1448 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1449 invalid |= NFS_INO_INVALID_MTIME;
1450
1451 ts = inode->i_ctime;
1452 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1453 invalid |= NFS_INO_INVALID_CTIME;
1454
1455 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1456 cur_size = i_size_read(inode);
1457 new_isize = nfs_size_to_loff_t(fattr->size);
1458 if (cur_size != new_isize)
1459 invalid |= NFS_INO_INVALID_SIZE
1460 | NFS_INO_REVAL_PAGECACHE;
1461 }
1462 }
1463
1464 /* Have any file permissions changed? */
1465 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1466 invalid |= NFS_INO_INVALID_ACCESS
1467 | NFS_INO_INVALID_ACL
1468 | NFS_INO_INVALID_OTHER;
1469 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1470 invalid |= NFS_INO_INVALID_ACCESS
1471 | NFS_INO_INVALID_ACL
1472 | NFS_INO_INVALID_OTHER;
1473 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1474 invalid |= NFS_INO_INVALID_ACCESS
1475 | NFS_INO_INVALID_ACL
1476 | NFS_INO_INVALID_OTHER;
1477
1478 /* Has the link count changed? */
1479 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1480 invalid |= NFS_INO_INVALID_OTHER;
1481
1482 ts = inode->i_atime;
1483 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1484 invalid |= NFS_INO_INVALID_ATIME;
1485
1486 if (invalid != 0)
1487 nfs_set_cache_invalid(inode, invalid);
1488
1489 nfsi->read_cache_jiffies = fattr->time_start;
1490 return 0;
1491 }
1492
1493 static atomic_long_t nfs_attr_generation_counter;
1494
1495 static unsigned long nfs_read_attr_generation_counter(void)
1496 {
1497 return atomic_long_read(&nfs_attr_generation_counter);
1498 }
1499
1500 unsigned long nfs_inc_attr_generation_counter(void)
1501 {
1502 return atomic_long_inc_return(&nfs_attr_generation_counter);
1503 }
1504 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1505
1506 void nfs_fattr_init(struct nfs_fattr *fattr)
1507 {
1508 fattr->valid = 0;
1509 fattr->time_start = jiffies;
1510 fattr->gencount = nfs_inc_attr_generation_counter();
1511 fattr->owner_name = NULL;
1512 fattr->group_name = NULL;
1513 }
1514 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1515
1516 /**
1517 * nfs_fattr_set_barrier
1518 * @fattr: attributes
1519 *
1520 * Used to set a barrier after an attribute was updated. This
1521 * barrier ensures that older attributes from RPC calls that may
1522 * have raced with our update cannot clobber these new values.
1523 * Note that you are still responsible for ensuring that other
1524 * operations which change the attribute on the server do not
1525 * collide.
1526 */
1527 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1528 {
1529 fattr->gencount = nfs_inc_attr_generation_counter();
1530 }
1531
1532 struct nfs_fattr *nfs_alloc_fattr(void)
1533 {
1534 struct nfs_fattr *fattr;
1535
1536 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1537 if (fattr != NULL)
1538 nfs_fattr_init(fattr);
1539 return fattr;
1540 }
1541 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1542
1543 struct nfs_fh *nfs_alloc_fhandle(void)
1544 {
1545 struct nfs_fh *fh;
1546
1547 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1548 if (fh != NULL)
1549 fh->size = 0;
1550 return fh;
1551 }
1552 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1553
1554 #ifdef NFS_DEBUG
1555 /*
1556 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1557 * in the same way that wireshark does
1558 *
1559 * @fh: file handle
1560 *
1561 * For debugging only.
1562 */
1563 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1564 {
1565 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1566 * not on the result */
1567 return nfs_fhandle_hash(fh);
1568 }
1569 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1570
1571 /*
1572 * _nfs_display_fhandle - display an NFS file handle on the console
1573 *
1574 * @fh: file handle to display
1575 * @caption: display caption
1576 *
1577 * For debugging only.
1578 */
1579 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1580 {
1581 unsigned short i;
1582
1583 if (fh == NULL || fh->size == 0) {
1584 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1585 return;
1586 }
1587
1588 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1589 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1590 for (i = 0; i < fh->size; i += 16) {
1591 __be32 *pos = (__be32 *)&fh->data[i];
1592
1593 switch ((fh->size - i - 1) >> 2) {
1594 case 0:
1595 printk(KERN_DEFAULT " %08x\n",
1596 be32_to_cpup(pos));
1597 break;
1598 case 1:
1599 printk(KERN_DEFAULT " %08x %08x\n",
1600 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1601 break;
1602 case 2:
1603 printk(KERN_DEFAULT " %08x %08x %08x\n",
1604 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1605 be32_to_cpup(pos + 2));
1606 break;
1607 default:
1608 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1609 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1610 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1611 }
1612 }
1613 }
1614 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1615 #endif
1616
1617 /**
1618 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1619 * @inode: pointer to inode
1620 * @fattr: attributes
1621 *
1622 * Attempt to divine whether or not an RPC call reply carrying stale
1623 * attributes got scheduled after another call carrying updated ones.
1624 *
1625 * To do so, the function first assumes that a more recent ctime means
1626 * that the attributes in fattr are newer, however it also attempt to
1627 * catch the case where ctime either didn't change, or went backwards
1628 * (if someone reset the clock on the server) by looking at whether
1629 * or not this RPC call was started after the inode was last updated.
1630 * Note also the check for wraparound of 'attr_gencount'
1631 *
1632 * The function returns 'true' if it thinks the attributes in 'fattr' are
1633 * more recent than the ones cached in the inode.
1634 *
1635 */
1636 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1637 {
1638 const struct nfs_inode *nfsi = NFS_I(inode);
1639
1640 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1641 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1642 }
1643
1644 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1645 {
1646 int ret;
1647
1648 trace_nfs_refresh_inode_enter(inode);
1649
1650 if (nfs_inode_attrs_need_update(inode, fattr))
1651 ret = nfs_update_inode(inode, fattr);
1652 else
1653 ret = nfs_check_inode_attributes(inode, fattr);
1654
1655 trace_nfs_refresh_inode_exit(inode, ret);
1656 return ret;
1657 }
1658
1659 /**
1660 * nfs_refresh_inode - try to update the inode attribute cache
1661 * @inode: pointer to inode
1662 * @fattr: updated attributes
1663 *
1664 * Check that an RPC call that returned attributes has not overlapped with
1665 * other recent updates of the inode metadata, then decide whether it is
1666 * safe to do a full update of the inode attributes, or whether just to
1667 * call nfs_check_inode_attributes.
1668 */
1669 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1670 {
1671 int status;
1672
1673 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1674 return 0;
1675 spin_lock(&inode->i_lock);
1676 status = nfs_refresh_inode_locked(inode, fattr);
1677 spin_unlock(&inode->i_lock);
1678
1679 return status;
1680 }
1681 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1682
1683 static int nfs_post_op_update_inode_locked(struct inode *inode,
1684 struct nfs_fattr *fattr, unsigned int invalid)
1685 {
1686 if (S_ISDIR(inode->i_mode))
1687 invalid |= NFS_INO_INVALID_DATA;
1688 nfs_set_cache_invalid(inode, invalid);
1689 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1690 return 0;
1691 return nfs_refresh_inode_locked(inode, fattr);
1692 }
1693
1694 /**
1695 * nfs_post_op_update_inode - try to update the inode attribute cache
1696 * @inode: pointer to inode
1697 * @fattr: updated attributes
1698 *
1699 * After an operation that has changed the inode metadata, mark the
1700 * attribute cache as being invalid, then try to update it.
1701 *
1702 * NB: if the server didn't return any post op attributes, this
1703 * function will force the retrieval of attributes before the next
1704 * NFS request. Thus it should be used only for operations that
1705 * are expected to change one or more attributes, to avoid
1706 * unnecessary NFS requests and trips through nfs_update_inode().
1707 */
1708 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1709 {
1710 int status;
1711
1712 spin_lock(&inode->i_lock);
1713 nfs_fattr_set_barrier(fattr);
1714 status = nfs_post_op_update_inode_locked(inode, fattr,
1715 NFS_INO_INVALID_CHANGE
1716 | NFS_INO_INVALID_CTIME
1717 | NFS_INO_REVAL_FORCED);
1718 spin_unlock(&inode->i_lock);
1719
1720 return status;
1721 }
1722 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1723
1724 /**
1725 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1726 * @inode: pointer to inode
1727 * @fattr: updated attributes
1728 *
1729 * After an operation that has changed the inode metadata, mark the
1730 * attribute cache as being invalid, then try to update it. Fake up
1731 * weak cache consistency data, if none exist.
1732 *
1733 * This function is mainly designed to be used by the ->write_done() functions.
1734 */
1735 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1736 {
1737 int status;
1738
1739 /* Don't do a WCC update if these attributes are already stale */
1740 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1741 !nfs_inode_attrs_need_update(inode, fattr)) {
1742 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1743 | NFS_ATTR_FATTR_PRESIZE
1744 | NFS_ATTR_FATTR_PREMTIME
1745 | NFS_ATTR_FATTR_PRECTIME);
1746 goto out_noforce;
1747 }
1748 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1749 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1750 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1751 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1752 }
1753 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1754 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1755 fattr->pre_ctime = inode->i_ctime;
1756 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1757 }
1758 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1759 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1760 fattr->pre_mtime = inode->i_mtime;
1761 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1762 }
1763 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1764 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1765 fattr->pre_size = i_size_read(inode);
1766 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1767 }
1768 out_noforce:
1769 status = nfs_post_op_update_inode_locked(inode, fattr,
1770 NFS_INO_INVALID_CHANGE
1771 | NFS_INO_INVALID_CTIME
1772 | NFS_INO_INVALID_MTIME
1773 | NFS_INO_INVALID_BLOCKS);
1774 return status;
1775 }
1776
1777 /**
1778 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1779 * @inode: pointer to inode
1780 * @fattr: updated attributes
1781 *
1782 * After an operation that has changed the inode metadata, mark the
1783 * attribute cache as being invalid, then try to update it. Fake up
1784 * weak cache consistency data, if none exist.
1785 *
1786 * This function is mainly designed to be used by the ->write_done() functions.
1787 */
1788 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1789 {
1790 int status;
1791
1792 spin_lock(&inode->i_lock);
1793 nfs_fattr_set_barrier(fattr);
1794 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1795 spin_unlock(&inode->i_lock);
1796 return status;
1797 }
1798 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1799
1800
1801 /*
1802 * Many nfs protocol calls return the new file attributes after
1803 * an operation. Here we update the inode to reflect the state
1804 * of the server's inode.
1805 *
1806 * This is a bit tricky because we have to make sure all dirty pages
1807 * have been sent off to the server before calling invalidate_inode_pages.
1808 * To make sure no other process adds more write requests while we try
1809 * our best to flush them, we make them sleep during the attribute refresh.
1810 *
1811 * A very similar scenario holds for the dir cache.
1812 */
1813 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1814 {
1815 struct nfs_server *server;
1816 struct nfs_inode *nfsi = NFS_I(inode);
1817 loff_t cur_isize, new_isize;
1818 unsigned long invalid = 0;
1819 unsigned long now = jiffies;
1820 unsigned long save_cache_validity;
1821 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1822 bool cache_revalidated = true;
1823 bool attr_changed = false;
1824 bool have_delegation;
1825
1826 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1827 __func__, inode->i_sb->s_id, inode->i_ino,
1828 nfs_display_fhandle_hash(NFS_FH(inode)),
1829 atomic_read(&inode->i_count), fattr->valid);
1830
1831 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1832 /* Only a mounted-on-fileid? Just exit */
1833 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1834 return 0;
1835 /* Has the inode gone and changed behind our back? */
1836 } else if (nfsi->fileid != fattr->fileid) {
1837 /* Is this perhaps the mounted-on fileid? */
1838 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1839 nfsi->fileid == fattr->mounted_on_fileid)
1840 return 0;
1841 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1842 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1843 NFS_SERVER(inode)->nfs_client->cl_hostname,
1844 inode->i_sb->s_id, (long long)nfsi->fileid,
1845 (long long)fattr->fileid);
1846 goto out_err;
1847 }
1848
1849 /*
1850 * Make sure the inode's type hasn't changed.
1851 */
1852 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1853 /*
1854 * Big trouble! The inode has become a different object.
1855 */
1856 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1857 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1858 goto out_err;
1859 }
1860
1861 server = NFS_SERVER(inode);
1862 /* Update the fsid? */
1863 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1864 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1865 !IS_AUTOMOUNT(inode))
1866 server->fsid = fattr->fsid;
1867
1868 /* Save the delegation state before clearing cache_validity */
1869 have_delegation = nfs_have_delegated_attributes(inode);
1870
1871 /*
1872 * Update the read time so we don't revalidate too often.
1873 */
1874 nfsi->read_cache_jiffies = fattr->time_start;
1875
1876 save_cache_validity = nfsi->cache_validity;
1877 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1878 | NFS_INO_INVALID_ATIME
1879 | NFS_INO_REVAL_FORCED
1880 | NFS_INO_REVAL_PAGECACHE
1881 | NFS_INO_INVALID_BLOCKS);
1882
1883 /* Do atomic weak cache consistency updates */
1884 nfs_wcc_update_inode(inode, fattr);
1885
1886 if (pnfs_layoutcommit_outstanding(inode)) {
1887 nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1888 cache_revalidated = false;
1889 }
1890
1891 /* More cache consistency checks */
1892 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1893 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1894 /* Could it be a race with writeback? */
1895 if (!(have_writers || have_delegation)) {
1896 invalid |= NFS_INO_INVALID_DATA
1897 | NFS_INO_INVALID_ACCESS
1898 | NFS_INO_INVALID_ACL
1899 | NFS_INO_INVALID_XATTR;
1900 /* Force revalidate of all attributes */
1901 save_cache_validity |= NFS_INO_INVALID_CTIME
1902 | NFS_INO_INVALID_MTIME
1903 | NFS_INO_INVALID_SIZE
1904 | NFS_INO_INVALID_OTHER;
1905 if (S_ISDIR(inode->i_mode))
1906 nfs_force_lookup_revalidate(inode);
1907 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1908 inode->i_sb->s_id,
1909 inode->i_ino);
1910 } else if (!have_delegation)
1911 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1912 inode_set_iversion_raw(inode, fattr->change_attr);
1913 attr_changed = true;
1914 }
1915 } else {
1916 nfsi->cache_validity |= save_cache_validity &
1917 (NFS_INO_INVALID_CHANGE
1918 | NFS_INO_REVAL_PAGECACHE
1919 | NFS_INO_REVAL_FORCED);
1920 cache_revalidated = false;
1921 }
1922
1923 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1924 inode->i_mtime = fattr->mtime;
1925 } else if (server->caps & NFS_CAP_MTIME) {
1926 nfsi->cache_validity |= save_cache_validity &
1927 (NFS_INO_INVALID_MTIME
1928 | NFS_INO_REVAL_FORCED);
1929 cache_revalidated = false;
1930 }
1931
1932 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1933 inode->i_ctime = fattr->ctime;
1934 } else if (server->caps & NFS_CAP_CTIME) {
1935 nfsi->cache_validity |= save_cache_validity &
1936 (NFS_INO_INVALID_CTIME
1937 | NFS_INO_REVAL_FORCED);
1938 cache_revalidated = false;
1939 }
1940
1941 /* Check if our cached file size is stale */
1942 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1943 new_isize = nfs_size_to_loff_t(fattr->size);
1944 cur_isize = i_size_read(inode);
1945 if (new_isize != cur_isize && !have_delegation) {
1946 /* Do we perhaps have any outstanding writes, or has
1947 * the file grown beyond our last write? */
1948 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1949 i_size_write(inode, new_isize);
1950 if (!have_writers)
1951 invalid |= NFS_INO_INVALID_DATA;
1952 attr_changed = true;
1953 }
1954 dprintk("NFS: isize change on server for file %s/%ld "
1955 "(%Ld to %Ld)\n",
1956 inode->i_sb->s_id,
1957 inode->i_ino,
1958 (long long)cur_isize,
1959 (long long)new_isize);
1960 }
1961 } else {
1962 nfsi->cache_validity |= save_cache_validity &
1963 (NFS_INO_INVALID_SIZE
1964 | NFS_INO_REVAL_PAGECACHE
1965 | NFS_INO_REVAL_FORCED);
1966 cache_revalidated = false;
1967 }
1968
1969
1970 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1971 inode->i_atime = fattr->atime;
1972 else if (server->caps & NFS_CAP_ATIME) {
1973 nfsi->cache_validity |= save_cache_validity &
1974 (NFS_INO_INVALID_ATIME
1975 | NFS_INO_REVAL_FORCED);
1976 cache_revalidated = false;
1977 }
1978
1979 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1980 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1981 umode_t newmode = inode->i_mode & S_IFMT;
1982 newmode |= fattr->mode & S_IALLUGO;
1983 inode->i_mode = newmode;
1984 invalid |= NFS_INO_INVALID_ACCESS
1985 | NFS_INO_INVALID_ACL;
1986 attr_changed = true;
1987 }
1988 } else if (server->caps & NFS_CAP_MODE) {
1989 nfsi->cache_validity |= save_cache_validity &
1990 (NFS_INO_INVALID_OTHER
1991 | NFS_INO_REVAL_FORCED);
1992 cache_revalidated = false;
1993 }
1994
1995 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1996 if (!uid_eq(inode->i_uid, fattr->uid)) {
1997 invalid |= NFS_INO_INVALID_ACCESS
1998 | NFS_INO_INVALID_ACL;
1999 inode->i_uid = fattr->uid;
2000 attr_changed = true;
2001 }
2002 } else if (server->caps & NFS_CAP_OWNER) {
2003 nfsi->cache_validity |= save_cache_validity &
2004 (NFS_INO_INVALID_OTHER
2005 | NFS_INO_REVAL_FORCED);
2006 cache_revalidated = false;
2007 }
2008
2009 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2010 if (!gid_eq(inode->i_gid, fattr->gid)) {
2011 invalid |= NFS_INO_INVALID_ACCESS
2012 | NFS_INO_INVALID_ACL;
2013 inode->i_gid = fattr->gid;
2014 attr_changed = true;
2015 }
2016 } else if (server->caps & NFS_CAP_OWNER_GROUP) {
2017 nfsi->cache_validity |= save_cache_validity &
2018 (NFS_INO_INVALID_OTHER
2019 | NFS_INO_REVAL_FORCED);
2020 cache_revalidated = false;
2021 }
2022
2023 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2024 if (inode->i_nlink != fattr->nlink) {
2025 if (S_ISDIR(inode->i_mode))
2026 invalid |= NFS_INO_INVALID_DATA;
2027 set_nlink(inode, fattr->nlink);
2028 attr_changed = true;
2029 }
2030 } else if (server->caps & NFS_CAP_NLINK) {
2031 nfsi->cache_validity |= save_cache_validity &
2032 (NFS_INO_INVALID_OTHER
2033 | NFS_INO_REVAL_FORCED);
2034 cache_revalidated = false;
2035 }
2036
2037 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2038 /*
2039 * report the blocks in 512byte units
2040 */
2041 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2042 } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2043 inode->i_blocks = fattr->du.nfs2.blocks;
2044 else {
2045 nfsi->cache_validity |= save_cache_validity &
2046 (NFS_INO_INVALID_BLOCKS
2047 | NFS_INO_REVAL_FORCED);
2048 cache_revalidated = false;
2049 }
2050
2051 /* Update attrtimeo value if we're out of the unstable period */
2052 if (attr_changed) {
2053 invalid &= ~NFS_INO_INVALID_ATTR;
2054 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2055 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2056 nfsi->attrtimeo_timestamp = now;
2057 /* Set barrier to be more recent than all outstanding updates */
2058 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2059 } else {
2060 if (cache_revalidated) {
2061 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2062 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2063 nfsi->attrtimeo <<= 1;
2064 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2065 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2066 }
2067 nfsi->attrtimeo_timestamp = now;
2068 }
2069 /* Set the barrier to be more recent than this fattr */
2070 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
2071 nfsi->attr_gencount = fattr->gencount;
2072 }
2073
2074 /* Don't invalidate the data if we were to blame */
2075 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2076 || S_ISLNK(inode->i_mode)))
2077 invalid &= ~NFS_INO_INVALID_DATA;
2078 nfs_set_cache_invalid(inode, invalid);
2079
2080 return 0;
2081 out_err:
2082 /*
2083 * No need to worry about unhashing the dentry, as the
2084 * lookup validation will know that the inode is bad.
2085 * (But we fall through to invalidate the caches.)
2086 */
2087 nfs_set_inode_stale_locked(inode);
2088 return -ESTALE;
2089 }
2090
2091 struct inode *nfs_alloc_inode(struct super_block *sb)
2092 {
2093 struct nfs_inode *nfsi;
2094 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2095 if (!nfsi)
2096 return NULL;
2097 nfsi->flags = 0UL;
2098 nfsi->cache_validity = 0UL;
2099 #if IS_ENABLED(CONFIG_NFS_V4)
2100 nfsi->nfs4_acl = NULL;
2101 #endif /* CONFIG_NFS_V4 */
2102 #ifdef CONFIG_NFS_V4_2
2103 nfsi->xattr_cache = NULL;
2104 #endif
2105 return &nfsi->vfs_inode;
2106 }
2107 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2108
2109 void nfs_free_inode(struct inode *inode)
2110 {
2111 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2112 }
2113 EXPORT_SYMBOL_GPL(nfs_free_inode);
2114
2115 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2116 {
2117 #if IS_ENABLED(CONFIG_NFS_V4)
2118 INIT_LIST_HEAD(&nfsi->open_states);
2119 nfsi->delegation = NULL;
2120 init_rwsem(&nfsi->rwsem);
2121 nfsi->layout = NULL;
2122 #endif
2123 }
2124
2125 static void init_once(void *foo)
2126 {
2127 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2128
2129 inode_init_once(&nfsi->vfs_inode);
2130 INIT_LIST_HEAD(&nfsi->open_files);
2131 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2132 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2133 INIT_LIST_HEAD(&nfsi->commit_info.list);
2134 atomic_long_set(&nfsi->nrequests, 0);
2135 atomic_long_set(&nfsi->commit_info.ncommit, 0);
2136 atomic_set(&nfsi->commit_info.rpcs_out, 0);
2137 init_rwsem(&nfsi->rmdir_sem);
2138 mutex_init(&nfsi->commit_mutex);
2139 nfs4_init_once(nfsi);
2140 nfsi->cache_change_attribute = 0;
2141 }
2142
2143 static int __init nfs_init_inodecache(void)
2144 {
2145 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2146 sizeof(struct nfs_inode),
2147 0, (SLAB_RECLAIM_ACCOUNT|
2148 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2149 init_once);
2150 if (nfs_inode_cachep == NULL)
2151 return -ENOMEM;
2152
2153 return 0;
2154 }
2155
2156 static void nfs_destroy_inodecache(void)
2157 {
2158 /*
2159 * Make sure all delayed rcu free inodes are flushed before we
2160 * destroy cache.
2161 */
2162 rcu_barrier();
2163 kmem_cache_destroy(nfs_inode_cachep);
2164 }
2165
2166 struct workqueue_struct *nfsiod_workqueue;
2167 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2168
2169 /*
2170 * start up the nfsiod workqueue
2171 */
2172 static int nfsiod_start(void)
2173 {
2174 struct workqueue_struct *wq;
2175 dprintk("RPC: creating workqueue nfsiod\n");
2176 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2177 if (wq == NULL)
2178 return -ENOMEM;
2179 nfsiod_workqueue = wq;
2180 return 0;
2181 }
2182
2183 /*
2184 * Destroy the nfsiod workqueue
2185 */
2186 static void nfsiod_stop(void)
2187 {
2188 struct workqueue_struct *wq;
2189
2190 wq = nfsiod_workqueue;
2191 if (wq == NULL)
2192 return;
2193 nfsiod_workqueue = NULL;
2194 destroy_workqueue(wq);
2195 }
2196
2197 unsigned int nfs_net_id;
2198 EXPORT_SYMBOL_GPL(nfs_net_id);
2199
2200 static int nfs_net_init(struct net *net)
2201 {
2202 nfs_clients_init(net);
2203 return nfs_fs_proc_net_init(net);
2204 }
2205
2206 static void nfs_net_exit(struct net *net)
2207 {
2208 nfs_fs_proc_net_exit(net);
2209 nfs_clients_exit(net);
2210 }
2211
2212 static struct pernet_operations nfs_net_ops = {
2213 .init = nfs_net_init,
2214 .exit = nfs_net_exit,
2215 .id = &nfs_net_id,
2216 .size = sizeof(struct nfs_net),
2217 };
2218
2219 /*
2220 * Initialize NFS
2221 */
2222 static int __init init_nfs_fs(void)
2223 {
2224 int err;
2225
2226 err = nfs_sysfs_init();
2227 if (err < 0)
2228 goto out10;
2229
2230 err = register_pernet_subsys(&nfs_net_ops);
2231 if (err < 0)
2232 goto out9;
2233
2234 err = nfs_fscache_register();
2235 if (err < 0)
2236 goto out8;
2237
2238 err = nfsiod_start();
2239 if (err)
2240 goto out7;
2241
2242 err = nfs_fs_proc_init();
2243 if (err)
2244 goto out6;
2245
2246 err = nfs_init_nfspagecache();
2247 if (err)
2248 goto out5;
2249
2250 err = nfs_init_inodecache();
2251 if (err)
2252 goto out4;
2253
2254 err = nfs_init_readpagecache();
2255 if (err)
2256 goto out3;
2257
2258 err = nfs_init_writepagecache();
2259 if (err)
2260 goto out2;
2261
2262 err = nfs_init_directcache();
2263 if (err)
2264 goto out1;
2265
2266 rpc_proc_register(&init_net, &nfs_rpcstat);
2267
2268 err = register_nfs_fs();
2269 if (err)
2270 goto out0;
2271
2272 return 0;
2273 out0:
2274 rpc_proc_unregister(&init_net, "nfs");
2275 nfs_destroy_directcache();
2276 out1:
2277 nfs_destroy_writepagecache();
2278 out2:
2279 nfs_destroy_readpagecache();
2280 out3:
2281 nfs_destroy_inodecache();
2282 out4:
2283 nfs_destroy_nfspagecache();
2284 out5:
2285 nfs_fs_proc_exit();
2286 out6:
2287 nfsiod_stop();
2288 out7:
2289 nfs_fscache_unregister();
2290 out8:
2291 unregister_pernet_subsys(&nfs_net_ops);
2292 out9:
2293 nfs_sysfs_exit();
2294 out10:
2295 return err;
2296 }
2297
2298 static void __exit exit_nfs_fs(void)
2299 {
2300 nfs_destroy_directcache();
2301 nfs_destroy_writepagecache();
2302 nfs_destroy_readpagecache();
2303 nfs_destroy_inodecache();
2304 nfs_destroy_nfspagecache();
2305 nfs_fscache_unregister();
2306 unregister_pernet_subsys(&nfs_net_ops);
2307 rpc_proc_unregister(&init_net, "nfs");
2308 unregister_nfs_fs();
2309 nfs_fs_proc_exit();
2310 nfsiod_stop();
2311 nfs_sysfs_exit();
2312 }
2313
2314 /* Not quite true; I just maintain it */
2315 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2316 MODULE_LICENSE("GPL");
2317 module_param(enable_ino64, bool, 0644);
2318
2319 module_init(init_nfs_fs)
2320 module_exit(exit_nfs_fs)
Linux with added FPC-III support