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
nfs_fattr_to_ino_t(struct nfs_fattr * fattr)70 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71 {
72 return nfs_fileid_to_ino_t(fattr->fileid);
73 }
74
nfs_wait_bit_killable(struct wait_bit_key * key,int mode)75 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
76 {
77 schedule();
78 if (signal_pending_state(mode, current))
79 return -ERESTARTSYS;
80 return 0;
81 }
82 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
83
84 /**
85 * nfs_compat_user_ino64 - returns the user-visible inode number
86 * @fileid: 64-bit fileid
87 *
88 * This function returns a 32-bit inode number if the boot parameter
89 * nfs.enable_ino64 is zero.
90 */
nfs_compat_user_ino64(u64 fileid)91 u64 nfs_compat_user_ino64(u64 fileid)
92 {
93 #ifdef CONFIG_COMPAT
94 compat_ulong_t ino;
95 #else
96 unsigned long ino;
97 #endif
98
99 if (enable_ino64)
100 return fileid;
101 ino = fileid;
102 if (sizeof(ino) < sizeof(fileid))
103 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
104 return ino;
105 }
106
nfs_drop_inode(struct inode * inode)107 int nfs_drop_inode(struct inode *inode)
108 {
109 return NFS_STALE(inode) || generic_drop_inode(inode);
110 }
111 EXPORT_SYMBOL_GPL(nfs_drop_inode);
112
nfs_clear_inode(struct inode * inode)113 void nfs_clear_inode(struct inode *inode)
114 {
115 /*
116 * The following should never happen...
117 */
118 WARN_ON_ONCE(nfs_have_writebacks(inode));
119 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
120 nfs_zap_acl_cache(inode);
121 nfs_access_zap_cache(inode);
122 nfs_fscache_clear_inode(inode);
123 }
124 EXPORT_SYMBOL_GPL(nfs_clear_inode);
125
nfs_evict_inode(struct inode * inode)126 void nfs_evict_inode(struct inode *inode)
127 {
128 truncate_inode_pages_final(&inode->i_data);
129 clear_inode(inode);
130 nfs_clear_inode(inode);
131 }
132
nfs_sync_inode(struct inode * inode)133 int nfs_sync_inode(struct inode *inode)
134 {
135 inode_dio_wait(inode);
136 return nfs_wb_all(inode);
137 }
138 EXPORT_SYMBOL_GPL(nfs_sync_inode);
139
140 /**
141 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
142 * @mapping: pointer to struct address_space
143 */
nfs_sync_mapping(struct address_space * mapping)144 int nfs_sync_mapping(struct address_space *mapping)
145 {
146 int ret = 0;
147
148 if (mapping->nrpages != 0) {
149 unmap_mapping_range(mapping, 0, 0, 0);
150 ret = nfs_wb_all(mapping->host);
151 }
152 return ret;
153 }
154
nfs_attribute_timeout(struct inode * inode)155 static int nfs_attribute_timeout(struct inode *inode)
156 {
157 struct nfs_inode *nfsi = NFS_I(inode);
158
159 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
160 }
161
nfs_check_cache_flags_invalid(struct inode * inode,unsigned long flags)162 static bool nfs_check_cache_flags_invalid(struct inode *inode,
163 unsigned long flags)
164 {
165 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
166
167 return (cache_validity & flags) != 0;
168 }
169
nfs_check_cache_invalid(struct inode * inode,unsigned long flags)170 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
171 {
172 if (nfs_check_cache_flags_invalid(inode, flags))
173 return true;
174 return nfs_attribute_cache_expired(inode);
175 }
176 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
177
178 #ifdef CONFIG_NFS_V4_2
nfs_has_xattr_cache(const struct nfs_inode * nfsi)179 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
180 {
181 return nfsi->xattr_cache != NULL;
182 }
183 #else
nfs_has_xattr_cache(const struct nfs_inode * nfsi)184 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
185 {
186 return false;
187 }
188 #endif
189
nfs_set_cache_invalid(struct inode * inode,unsigned long flags)190 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
191 {
192 struct nfs_inode *nfsi = NFS_I(inode);
193 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
194
195 if (have_delegation) {
196 if (!(flags & NFS_INO_REVAL_FORCED))
197 flags &= ~(NFS_INO_INVALID_MODE |
198 NFS_INO_INVALID_OTHER |
199 NFS_INO_INVALID_XATTR);
200 flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
201 }
202
203 if (!nfs_has_xattr_cache(nfsi))
204 flags &= ~NFS_INO_INVALID_XATTR;
205 if (flags & NFS_INO_INVALID_DATA)
206 nfs_fscache_invalidate(inode, 0);
207 flags &= ~NFS_INO_REVAL_FORCED;
208
209 nfsi->cache_validity |= flags;
210
211 if (inode->i_mapping->nrpages == 0)
212 nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA |
213 NFS_INO_DATA_INVAL_DEFER);
214 else if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
215 nfsi->cache_validity &= ~NFS_INO_DATA_INVAL_DEFER;
216 trace_nfs_set_cache_invalid(inode, 0);
217 }
218 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
219
220 /*
221 * Invalidate the local caches
222 */
nfs_zap_caches_locked(struct inode * inode)223 static void nfs_zap_caches_locked(struct inode *inode)
224 {
225 struct nfs_inode *nfsi = NFS_I(inode);
226 int mode = inode->i_mode;
227
228 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
229
230 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
231 nfsi->attrtimeo_timestamp = jiffies;
232
233 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
234 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
235 NFS_INO_INVALID_DATA |
236 NFS_INO_INVALID_ACCESS |
237 NFS_INO_INVALID_ACL |
238 NFS_INO_INVALID_XATTR);
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_zap_label_cache_locked(nfsi);
245 }
246
nfs_zap_caches(struct inode * inode)247 void nfs_zap_caches(struct inode *inode)
248 {
249 spin_lock(&inode->i_lock);
250 nfs_zap_caches_locked(inode);
251 spin_unlock(&inode->i_lock);
252 }
253
nfs_zap_mapping(struct inode * inode,struct address_space * mapping)254 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
255 {
256 if (mapping->nrpages != 0) {
257 spin_lock(&inode->i_lock);
258 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
259 spin_unlock(&inode->i_lock);
260 }
261 }
262
nfs_zap_acl_cache(struct inode * inode)263 void nfs_zap_acl_cache(struct inode *inode)
264 {
265 void (*clear_acl_cache)(struct inode *);
266
267 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
268 if (clear_acl_cache != NULL)
269 clear_acl_cache(inode);
270 spin_lock(&inode->i_lock);
271 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
272 spin_unlock(&inode->i_lock);
273 }
274 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
275
nfs_invalidate_atime(struct inode * inode)276 void nfs_invalidate_atime(struct inode *inode)
277 {
278 spin_lock(&inode->i_lock);
279 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
280 spin_unlock(&inode->i_lock);
281 }
282 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
283
284 /*
285 * Invalidate, but do not unhash, the inode.
286 * NB: must be called with inode->i_lock held!
287 */
nfs_set_inode_stale_locked(struct inode * inode)288 static void nfs_set_inode_stale_locked(struct inode *inode)
289 {
290 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
291 nfs_zap_caches_locked(inode);
292 trace_nfs_set_inode_stale(inode);
293 }
294
nfs_set_inode_stale(struct inode * inode)295 void nfs_set_inode_stale(struct inode *inode)
296 {
297 spin_lock(&inode->i_lock);
298 nfs_set_inode_stale_locked(inode);
299 spin_unlock(&inode->i_lock);
300 }
301
302 struct nfs_find_desc {
303 struct nfs_fh *fh;
304 struct nfs_fattr *fattr;
305 };
306
307 /*
308 * In NFSv3 we can have 64bit inode numbers. In order to support
309 * this, and re-exported directories (also seen in NFSv2)
310 * we are forced to allow 2 different inodes to have the same
311 * i_ino.
312 */
313 static int
nfs_find_actor(struct inode * inode,void * opaque)314 nfs_find_actor(struct inode *inode, void *opaque)
315 {
316 struct nfs_find_desc *desc = opaque;
317 struct nfs_fh *fh = desc->fh;
318 struct nfs_fattr *fattr = desc->fattr;
319
320 if (NFS_FILEID(inode) != fattr->fileid)
321 return 0;
322 if (inode_wrong_type(inode, fattr->mode))
323 return 0;
324 if (nfs_compare_fh(NFS_FH(inode), fh))
325 return 0;
326 if (is_bad_inode(inode) || NFS_STALE(inode))
327 return 0;
328 return 1;
329 }
330
331 static int
nfs_init_locked(struct inode * inode,void * opaque)332 nfs_init_locked(struct inode *inode, void *opaque)
333 {
334 struct nfs_find_desc *desc = opaque;
335 struct nfs_fattr *fattr = desc->fattr;
336
337 set_nfs_fileid(inode, fattr->fileid);
338 inode->i_mode = fattr->mode;
339 nfs_copy_fh(NFS_FH(inode), desc->fh);
340 return 0;
341 }
342
343 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
nfs_clear_label_invalid(struct inode * inode)344 static void nfs_clear_label_invalid(struct inode *inode)
345 {
346 spin_lock(&inode->i_lock);
347 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
348 spin_unlock(&inode->i_lock);
349 }
350
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)351 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
352 {
353 int error;
354
355 if (fattr->label == NULL)
356 return;
357
358 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
359 error = security_inode_notifysecctx(inode, fattr->label->label,
360 fattr->label->len);
361 if (error)
362 printk(KERN_ERR "%s() %s %d "
363 "security_inode_notifysecctx() %d\n",
364 __func__,
365 (char *)fattr->label->label,
366 fattr->label->len, error);
367 nfs_clear_label_invalid(inode);
368 }
369 }
370
nfs4_label_alloc(struct nfs_server * server,gfp_t flags)371 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
372 {
373 struct nfs4_label *label;
374
375 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
376 return NULL;
377
378 label = kzalloc(sizeof(struct nfs4_label), flags);
379 if (label == NULL)
380 return ERR_PTR(-ENOMEM);
381
382 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
383 if (label->label == NULL) {
384 kfree(label);
385 return ERR_PTR(-ENOMEM);
386 }
387 label->len = NFS4_MAXLABELLEN;
388
389 return label;
390 }
391 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
392 #else
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)393 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
394 {
395 }
396 #endif
397 EXPORT_SYMBOL_GPL(nfs_setsecurity);
398
399 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
400 struct inode *
nfs_ilookup(struct super_block * sb,struct nfs_fattr * fattr,struct nfs_fh * fh)401 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
402 {
403 struct nfs_find_desc desc = {
404 .fh = fh,
405 .fattr = fattr,
406 };
407 struct inode *inode;
408 unsigned long hash;
409
410 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
411 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
412 return NULL;
413
414 hash = nfs_fattr_to_ino_t(fattr);
415 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
416
417 dprintk("%s: returning %p\n", __func__, inode);
418 return inode;
419 }
420
nfs_inode_init_regular(struct nfs_inode * nfsi)421 static void nfs_inode_init_regular(struct nfs_inode *nfsi)
422 {
423 atomic_long_set(&nfsi->nrequests, 0);
424 atomic_long_set(&nfsi->redirtied_pages, 0);
425 INIT_LIST_HEAD(&nfsi->commit_info.list);
426 atomic_long_set(&nfsi->commit_info.ncommit, 0);
427 atomic_set(&nfsi->commit_info.rpcs_out, 0);
428 mutex_init(&nfsi->commit_mutex);
429 }
430
nfs_inode_init_dir(struct nfs_inode * nfsi)431 static void nfs_inode_init_dir(struct nfs_inode *nfsi)
432 {
433 nfsi->cache_change_attribute = 0;
434 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
435 init_rwsem(&nfsi->rmdir_sem);
436 }
437
438 /*
439 * This is our front-end to iget that looks up inodes by file handle
440 * instead of inode number.
441 */
442 struct inode *
nfs_fhget(struct super_block * sb,struct nfs_fh * fh,struct nfs_fattr * fattr)443 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
444 {
445 struct nfs_find_desc desc = {
446 .fh = fh,
447 .fattr = fattr
448 };
449 struct inode *inode = ERR_PTR(-ENOENT);
450 u64 fattr_supported = NFS_SB(sb)->fattr_valid;
451 unsigned long hash;
452
453 nfs_attr_check_mountpoint(sb, fattr);
454
455 if (nfs_attr_use_mounted_on_fileid(fattr))
456 fattr->fileid = fattr->mounted_on_fileid;
457 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
458 goto out_no_inode;
459 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
460 goto out_no_inode;
461
462 hash = nfs_fattr_to_ino_t(fattr);
463
464 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
465 if (inode == NULL) {
466 inode = ERR_PTR(-ENOMEM);
467 goto out_no_inode;
468 }
469
470 if (inode->i_state & I_NEW) {
471 struct nfs_inode *nfsi = NFS_I(inode);
472 unsigned long now = jiffies;
473
474 /* We set i_ino for the few things that still rely on it,
475 * such as stat(2) */
476 inode->i_ino = hash;
477
478 /* We can't support update_atime(), since the server will reset it */
479 inode->i_flags |= S_NOATIME|S_NOCMTIME;
480 inode->i_mode = fattr->mode;
481 nfsi->cache_validity = 0;
482 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
483 && (fattr_supported & NFS_ATTR_FATTR_MODE))
484 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
485 /* Why so? Because we want revalidate for devices/FIFOs, and
486 * that's precisely what we have in nfs_file_inode_operations.
487 */
488 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
489 if (S_ISREG(inode->i_mode)) {
490 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
491 inode->i_data.a_ops = &nfs_file_aops;
492 nfs_inode_init_regular(nfsi);
493 } else if (S_ISDIR(inode->i_mode)) {
494 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
495 inode->i_fop = &nfs_dir_operations;
496 inode->i_data.a_ops = &nfs_dir_aops;
497 nfs_inode_init_dir(nfsi);
498 /* Deal with crossing mountpoints */
499 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
500 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
501 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
502 inode->i_op = &nfs_referral_inode_operations;
503 else
504 inode->i_op = &nfs_mountpoint_inode_operations;
505 inode->i_fop = NULL;
506 inode->i_flags |= S_AUTOMOUNT;
507 }
508 } else if (S_ISLNK(inode->i_mode)) {
509 inode->i_op = &nfs_symlink_inode_operations;
510 inode_nohighmem(inode);
511 } else
512 init_special_inode(inode, inode->i_mode, fattr->rdev);
513
514 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
515 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
516 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
517 inode_set_iversion_raw(inode, 0);
518 inode->i_size = 0;
519 clear_nlink(inode);
520 inode->i_uid = make_kuid(&init_user_ns, -2);
521 inode->i_gid = make_kgid(&init_user_ns, -2);
522 inode->i_blocks = 0;
523 nfsi->write_io = 0;
524 nfsi->read_io = 0;
525
526 nfsi->read_cache_jiffies = fattr->time_start;
527 nfsi->attr_gencount = fattr->gencount;
528 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
529 inode->i_atime = fattr->atime;
530 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
531 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
532 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
533 inode->i_mtime = fattr->mtime;
534 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
535 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
536 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
537 inode->i_ctime = fattr->ctime;
538 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
539 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
540 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
541 inode_set_iversion_raw(inode, fattr->change_attr);
542 else
543 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
544 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
545 inode->i_size = nfs_size_to_loff_t(fattr->size);
546 else
547 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
548 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
549 set_nlink(inode, fattr->nlink);
550 else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
551 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
552 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
553 inode->i_uid = fattr->uid;
554 else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
555 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
556 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
557 inode->i_gid = fattr->gid;
558 else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
559 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
560 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
561 inode->i_blocks = fattr->du.nfs2.blocks;
562 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
563 fattr->size != 0)
564 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
565 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
566 /*
567 * report the blocks in 512byte units
568 */
569 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
570 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
571 fattr->size != 0)
572 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
573
574 nfs_setsecurity(inode, fattr);
575
576 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
577 nfsi->attrtimeo_timestamp = now;
578 nfsi->access_cache = RB_ROOT;
579
580 nfs_fscache_init_inode(inode);
581
582 unlock_new_inode(inode);
583 } else {
584 int err = nfs_refresh_inode(inode, fattr);
585 if (err < 0) {
586 iput(inode);
587 inode = ERR_PTR(err);
588 goto out_no_inode;
589 }
590 }
591 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
592 inode->i_sb->s_id,
593 (unsigned long long)NFS_FILEID(inode),
594 nfs_display_fhandle_hash(fh),
595 atomic_read(&inode->i_count));
596
597 out:
598 return inode;
599
600 out_no_inode:
601 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
602 goto out;
603 }
604 EXPORT_SYMBOL_GPL(nfs_fhget);
605
606 #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)
607
608 int
nfs_setattr(struct user_namespace * mnt_userns,struct dentry * dentry,struct iattr * attr)609 nfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
610 struct iattr *attr)
611 {
612 struct inode *inode = d_inode(dentry);
613 struct nfs_fattr *fattr;
614 int error = 0;
615
616 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
617
618 /* skip mode change if it's just for clearing setuid/setgid */
619 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
620 attr->ia_valid &= ~ATTR_MODE;
621
622 if (attr->ia_valid & ATTR_SIZE) {
623 BUG_ON(!S_ISREG(inode->i_mode));
624
625 error = inode_newsize_ok(inode, attr->ia_size);
626 if (error)
627 return error;
628
629 if (attr->ia_size == i_size_read(inode))
630 attr->ia_valid &= ~ATTR_SIZE;
631 }
632
633 /* Optimization: if the end result is no change, don't RPC */
634 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
635 return 0;
636
637 trace_nfs_setattr_enter(inode);
638
639 /* Write all dirty data */
640 if (S_ISREG(inode->i_mode))
641 nfs_sync_inode(inode);
642
643 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
644 if (fattr == NULL) {
645 error = -ENOMEM;
646 goto out;
647 }
648
649 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
650 if (error == 0)
651 error = nfs_refresh_inode(inode, fattr);
652 nfs_free_fattr(fattr);
653 out:
654 trace_nfs_setattr_exit(inode, error);
655 return error;
656 }
657 EXPORT_SYMBOL_GPL(nfs_setattr);
658
659 /**
660 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
661 * @inode: inode of the file used
662 * @offset: file offset to start truncating
663 *
664 * This is a copy of the common vmtruncate, but with the locking
665 * corrected to take into account the fact that NFS requires
666 * inode->i_size to be updated under the inode->i_lock.
667 * Note: must be called with inode->i_lock held!
668 */
nfs_vmtruncate(struct inode * inode,loff_t offset)669 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
670 {
671 int err;
672
673 err = inode_newsize_ok(inode, offset);
674 if (err)
675 goto out;
676
677 trace_nfs_size_truncate(inode, offset);
678 i_size_write(inode, offset);
679 /* Optimisation */
680 if (offset == 0)
681 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
682 NFS_INO_DATA_INVAL_DEFER);
683 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
684
685 spin_unlock(&inode->i_lock);
686 truncate_pagecache(inode, offset);
687 spin_lock(&inode->i_lock);
688 out:
689 return err;
690 }
691
692 /**
693 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
694 * @inode: pointer to struct inode
695 * @attr: pointer to struct iattr
696 * @fattr: pointer to struct nfs_fattr
697 *
698 * Note: we do this in the *proc.c in order to ensure that
699 * it works for things like exclusive creates too.
700 */
nfs_setattr_update_inode(struct inode * inode,struct iattr * attr,struct nfs_fattr * fattr)701 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
702 struct nfs_fattr *fattr)
703 {
704 /* Barrier: bump the attribute generation count. */
705 nfs_fattr_set_barrier(fattr);
706
707 spin_lock(&inode->i_lock);
708 NFS_I(inode)->attr_gencount = fattr->gencount;
709 if ((attr->ia_valid & ATTR_SIZE) != 0) {
710 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME |
711 NFS_INO_INVALID_BLOCKS);
712 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
713 nfs_vmtruncate(inode, attr->ia_size);
714 }
715 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
716 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
717 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
718 inode->i_mode & S_ISUID)
719 inode->i_mode &= ~S_ISUID;
720 if (setattr_should_drop_sgid(&init_user_ns, inode))
721 inode->i_mode &= ~S_ISGID;
722 if ((attr->ia_valid & ATTR_MODE) != 0) {
723 int mode = attr->ia_mode & S_IALLUGO;
724 mode |= inode->i_mode & ~S_IALLUGO;
725 inode->i_mode = mode;
726 }
727 if ((attr->ia_valid & ATTR_UID) != 0)
728 inode->i_uid = attr->ia_uid;
729 if ((attr->ia_valid & ATTR_GID) != 0)
730 inode->i_gid = attr->ia_gid;
731 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
732 inode->i_ctime = fattr->ctime;
733 else
734 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
735 | NFS_INO_INVALID_CTIME);
736 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
737 | NFS_INO_INVALID_ACL);
738 }
739 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
740 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
741 | NFS_INO_INVALID_CTIME);
742 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
743 inode->i_atime = fattr->atime;
744 else if (attr->ia_valid & ATTR_ATIME_SET)
745 inode->i_atime = attr->ia_atime;
746 else
747 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
748
749 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
750 inode->i_ctime = fattr->ctime;
751 else
752 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
753 | NFS_INO_INVALID_CTIME);
754 }
755 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
756 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
757 | NFS_INO_INVALID_CTIME);
758 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
759 inode->i_mtime = fattr->mtime;
760 else if (attr->ia_valid & ATTR_MTIME_SET)
761 inode->i_mtime = attr->ia_mtime;
762 else
763 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
764
765 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
766 inode->i_ctime = fattr->ctime;
767 else
768 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
769 | NFS_INO_INVALID_CTIME);
770 }
771 if (fattr->valid)
772 nfs_update_inode(inode, fattr);
773 spin_unlock(&inode->i_lock);
774 }
775 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
776
777 /*
778 * Don't request help from readdirplus if the file is being written to,
779 * or if attribute caching is turned off
780 */
nfs_getattr_readdirplus_enable(const struct inode * inode)781 static bool nfs_getattr_readdirplus_enable(const struct inode *inode)
782 {
783 return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) &&
784 !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ;
785 }
786
nfs_readdirplus_parent_cache_miss(struct dentry * dentry)787 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
788 {
789 if (!IS_ROOT(dentry)) {
790 struct dentry *parent = dget_parent(dentry);
791 nfs_readdir_record_entry_cache_miss(d_inode(parent));
792 dput(parent);
793 }
794 }
795
nfs_readdirplus_parent_cache_hit(struct dentry * dentry)796 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
797 {
798 if (!IS_ROOT(dentry)) {
799 struct dentry *parent = dget_parent(dentry);
800 nfs_readdir_record_entry_cache_hit(d_inode(parent));
801 dput(parent);
802 }
803 }
804
nfs_get_valid_attrmask(struct inode * inode)805 static u32 nfs_get_valid_attrmask(struct inode *inode)
806 {
807 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
808 u32 reply_mask = STATX_INO | STATX_TYPE;
809
810 if (!(cache_validity & NFS_INO_INVALID_ATIME))
811 reply_mask |= STATX_ATIME;
812 if (!(cache_validity & NFS_INO_INVALID_CTIME))
813 reply_mask |= STATX_CTIME;
814 if (!(cache_validity & NFS_INO_INVALID_MTIME))
815 reply_mask |= STATX_MTIME;
816 if (!(cache_validity & NFS_INO_INVALID_SIZE))
817 reply_mask |= STATX_SIZE;
818 if (!(cache_validity & NFS_INO_INVALID_NLINK))
819 reply_mask |= STATX_NLINK;
820 if (!(cache_validity & NFS_INO_INVALID_MODE))
821 reply_mask |= STATX_MODE;
822 if (!(cache_validity & NFS_INO_INVALID_OTHER))
823 reply_mask |= STATX_UID | STATX_GID;
824 if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
825 reply_mask |= STATX_BLOCKS;
826 return reply_mask;
827 }
828
nfs_getattr(struct user_namespace * mnt_userns,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)829 int nfs_getattr(struct user_namespace *mnt_userns, const struct path *path,
830 struct kstat *stat, u32 request_mask, unsigned int query_flags)
831 {
832 struct inode *inode = d_inode(path->dentry);
833 struct nfs_server *server = NFS_SERVER(inode);
834 unsigned long cache_validity;
835 int err = 0;
836 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
837 bool do_update = false;
838 bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
839
840 trace_nfs_getattr_enter(inode);
841
842 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
843 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
844 STATX_INO | STATX_SIZE | STATX_BLOCKS;
845
846 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
847 if (readdirplus_enabled)
848 nfs_readdirplus_parent_cache_hit(path->dentry);
849 goto out_no_revalidate;
850 }
851
852 /* Flush out writes to the server in order to update c/mtime. */
853 if ((request_mask & (STATX_CTIME | STATX_MTIME)) &&
854 S_ISREG(inode->i_mode))
855 filemap_write_and_wait(inode->i_mapping);
856
857 /*
858 * We may force a getattr if the user cares about atime.
859 *
860 * Note that we only have to check the vfsmount flags here:
861 * - NFS always sets S_NOATIME by so checking it would give a
862 * bogus result
863 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
864 * no point in checking those.
865 */
866 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
867 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
868 request_mask &= ~STATX_ATIME;
869
870 /* Is the user requesting attributes that might need revalidation? */
871 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
872 STATX_MTIME|STATX_UID|STATX_GID|
873 STATX_SIZE|STATX_BLOCKS)))
874 goto out_no_revalidate;
875
876 /* Check whether the cached attributes are stale */
877 do_update |= force_sync || nfs_attribute_cache_expired(inode);
878 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
879 do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
880 if (request_mask & STATX_ATIME)
881 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
882 if (request_mask & STATX_CTIME)
883 do_update |= cache_validity & NFS_INO_INVALID_CTIME;
884 if (request_mask & STATX_MTIME)
885 do_update |= cache_validity & NFS_INO_INVALID_MTIME;
886 if (request_mask & STATX_SIZE)
887 do_update |= cache_validity & NFS_INO_INVALID_SIZE;
888 if (request_mask & STATX_NLINK)
889 do_update |= cache_validity & NFS_INO_INVALID_NLINK;
890 if (request_mask & STATX_MODE)
891 do_update |= cache_validity & NFS_INO_INVALID_MODE;
892 if (request_mask & (STATX_UID | STATX_GID))
893 do_update |= cache_validity & NFS_INO_INVALID_OTHER;
894 if (request_mask & STATX_BLOCKS)
895 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
896
897 if (do_update) {
898 if (readdirplus_enabled)
899 nfs_readdirplus_parent_cache_miss(path->dentry);
900 err = __nfs_revalidate_inode(server, inode);
901 if (err)
902 goto out;
903 } else if (readdirplus_enabled)
904 nfs_readdirplus_parent_cache_hit(path->dentry);
905 out_no_revalidate:
906 /* Only return attributes that were revalidated. */
907 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
908
909 generic_fillattr(&init_user_ns, inode, stat);
910 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
911 if (S_ISDIR(inode->i_mode))
912 stat->blksize = NFS_SERVER(inode)->dtsize;
913 out:
914 trace_nfs_getattr_exit(inode, err);
915 return err;
916 }
917 EXPORT_SYMBOL_GPL(nfs_getattr);
918
nfs_init_lock_context(struct nfs_lock_context * l_ctx)919 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
920 {
921 refcount_set(&l_ctx->count, 1);
922 l_ctx->lockowner = current->files;
923 INIT_LIST_HEAD(&l_ctx->list);
924 atomic_set(&l_ctx->io_count, 0);
925 }
926
__nfs_find_lock_context(struct nfs_open_context * ctx)927 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
928 {
929 struct nfs_lock_context *pos;
930
931 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
932 if (pos->lockowner != current->files)
933 continue;
934 if (refcount_inc_not_zero(&pos->count))
935 return pos;
936 }
937 return NULL;
938 }
939
nfs_get_lock_context(struct nfs_open_context * ctx)940 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
941 {
942 struct nfs_lock_context *res, *new = NULL;
943 struct inode *inode = d_inode(ctx->dentry);
944
945 rcu_read_lock();
946 res = __nfs_find_lock_context(ctx);
947 rcu_read_unlock();
948 if (res == NULL) {
949 new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
950 if (new == NULL)
951 return ERR_PTR(-ENOMEM);
952 nfs_init_lock_context(new);
953 spin_lock(&inode->i_lock);
954 res = __nfs_find_lock_context(ctx);
955 if (res == NULL) {
956 new->open_context = get_nfs_open_context(ctx);
957 if (new->open_context) {
958 list_add_tail_rcu(&new->list,
959 &ctx->lock_context.list);
960 res = new;
961 new = NULL;
962 } else
963 res = ERR_PTR(-EBADF);
964 }
965 spin_unlock(&inode->i_lock);
966 kfree(new);
967 }
968 return res;
969 }
970 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
971
nfs_put_lock_context(struct nfs_lock_context * l_ctx)972 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
973 {
974 struct nfs_open_context *ctx = l_ctx->open_context;
975 struct inode *inode = d_inode(ctx->dentry);
976
977 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
978 return;
979 list_del_rcu(&l_ctx->list);
980 spin_unlock(&inode->i_lock);
981 put_nfs_open_context(ctx);
982 kfree_rcu(l_ctx, rcu_head);
983 }
984 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
985
986 /**
987 * nfs_close_context - Common close_context() routine NFSv2/v3
988 * @ctx: pointer to context
989 * @is_sync: is this a synchronous close
990 *
991 * Ensure that the attributes are up to date if we're mounted
992 * with close-to-open semantics and we have cached data that will
993 * need to be revalidated on open.
994 */
nfs_close_context(struct nfs_open_context * ctx,int is_sync)995 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
996 {
997 struct nfs_inode *nfsi;
998 struct inode *inode;
999
1000 if (!(ctx->mode & FMODE_WRITE))
1001 return;
1002 if (!is_sync)
1003 return;
1004 inode = d_inode(ctx->dentry);
1005 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1006 return;
1007 nfsi = NFS_I(inode);
1008 if (inode->i_mapping->nrpages == 0)
1009 return;
1010 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1011 return;
1012 if (!list_empty(&nfsi->open_files))
1013 return;
1014 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1015 return;
1016 nfs_revalidate_inode(inode,
1017 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1018 }
1019 EXPORT_SYMBOL_GPL(nfs_close_context);
1020
alloc_nfs_open_context(struct dentry * dentry,fmode_t f_mode,struct file * filp)1021 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1022 fmode_t f_mode,
1023 struct file *filp)
1024 {
1025 struct nfs_open_context *ctx;
1026
1027 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1028 if (!ctx)
1029 return ERR_PTR(-ENOMEM);
1030 nfs_sb_active(dentry->d_sb);
1031 ctx->dentry = dget(dentry);
1032 if (filp)
1033 ctx->cred = get_cred(filp->f_cred);
1034 else
1035 ctx->cred = get_current_cred();
1036 rcu_assign_pointer(ctx->ll_cred, NULL);
1037 ctx->state = NULL;
1038 ctx->mode = f_mode;
1039 ctx->flags = 0;
1040 ctx->error = 0;
1041 ctx->flock_owner = (fl_owner_t)filp;
1042 nfs_init_lock_context(&ctx->lock_context);
1043 ctx->lock_context.open_context = ctx;
1044 INIT_LIST_HEAD(&ctx->list);
1045 ctx->mdsthreshold = NULL;
1046 return ctx;
1047 }
1048 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1049
get_nfs_open_context(struct nfs_open_context * ctx)1050 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1051 {
1052 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1053 return ctx;
1054 return NULL;
1055 }
1056 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1057
__put_nfs_open_context(struct nfs_open_context * ctx,int is_sync)1058 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1059 {
1060 struct inode *inode = d_inode(ctx->dentry);
1061 struct super_block *sb = ctx->dentry->d_sb;
1062
1063 if (!refcount_dec_and_test(&ctx->lock_context.count))
1064 return;
1065 if (!list_empty(&ctx->list)) {
1066 spin_lock(&inode->i_lock);
1067 list_del_rcu(&ctx->list);
1068 spin_unlock(&inode->i_lock);
1069 }
1070 if (inode != NULL)
1071 NFS_PROTO(inode)->close_context(ctx, is_sync);
1072 put_cred(ctx->cred);
1073 dput(ctx->dentry);
1074 nfs_sb_deactive(sb);
1075 put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1076 kfree(ctx->mdsthreshold);
1077 kfree_rcu(ctx, rcu_head);
1078 }
1079
put_nfs_open_context(struct nfs_open_context * ctx)1080 void put_nfs_open_context(struct nfs_open_context *ctx)
1081 {
1082 __put_nfs_open_context(ctx, 0);
1083 }
1084 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1085
put_nfs_open_context_sync(struct nfs_open_context * ctx)1086 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1087 {
1088 __put_nfs_open_context(ctx, 1);
1089 }
1090
1091 /*
1092 * Ensure that mmap has a recent RPC credential for use when writing out
1093 * shared pages
1094 */
nfs_inode_attach_open_context(struct nfs_open_context * ctx)1095 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1096 {
1097 struct inode *inode = d_inode(ctx->dentry);
1098 struct nfs_inode *nfsi = NFS_I(inode);
1099
1100 spin_lock(&inode->i_lock);
1101 if (list_empty(&nfsi->open_files) &&
1102 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1103 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1104 NFS_INO_REVAL_FORCED);
1105 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1106 spin_unlock(&inode->i_lock);
1107 }
1108 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1109
nfs_file_set_open_context(struct file * filp,struct nfs_open_context * ctx)1110 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1111 {
1112 filp->private_data = get_nfs_open_context(ctx);
1113 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1114 if (list_empty(&ctx->list))
1115 nfs_inode_attach_open_context(ctx);
1116 }
1117 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1118
1119 /*
1120 * Given an inode, search for an open context with the desired characteristics
1121 */
nfs_find_open_context(struct inode * inode,const struct cred * cred,fmode_t mode)1122 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1123 {
1124 struct nfs_inode *nfsi = NFS_I(inode);
1125 struct nfs_open_context *pos, *ctx = NULL;
1126
1127 rcu_read_lock();
1128 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1129 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1130 continue;
1131 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1132 continue;
1133 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1134 continue;
1135 ctx = get_nfs_open_context(pos);
1136 if (ctx)
1137 break;
1138 }
1139 rcu_read_unlock();
1140 return ctx;
1141 }
1142
nfs_file_clear_open_context(struct file * filp)1143 void nfs_file_clear_open_context(struct file *filp)
1144 {
1145 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1146
1147 if (ctx) {
1148 struct inode *inode = d_inode(ctx->dentry);
1149
1150 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1151 /*
1152 * We fatal error on write before. Try to writeback
1153 * every page again.
1154 */
1155 if (ctx->error < 0)
1156 invalidate_inode_pages2(inode->i_mapping);
1157 filp->private_data = NULL;
1158 put_nfs_open_context_sync(ctx);
1159 }
1160 }
1161
1162 /*
1163 * These allocate and release file read/write context information.
1164 */
nfs_open(struct inode * inode,struct file * filp)1165 int nfs_open(struct inode *inode, struct file *filp)
1166 {
1167 struct nfs_open_context *ctx;
1168
1169 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1170 if (IS_ERR(ctx))
1171 return PTR_ERR(ctx);
1172 nfs_file_set_open_context(filp, ctx);
1173 put_nfs_open_context(ctx);
1174 nfs_fscache_open_file(inode, filp);
1175 return 0;
1176 }
1177
1178 /*
1179 * This function is called whenever some part of NFS notices that
1180 * the cached attributes have to be refreshed.
1181 */
1182 int
__nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1183 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1184 {
1185 int status = -ESTALE;
1186 struct nfs_fattr *fattr = NULL;
1187 struct nfs_inode *nfsi = NFS_I(inode);
1188
1189 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1190 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1191
1192 trace_nfs_revalidate_inode_enter(inode);
1193
1194 if (is_bad_inode(inode))
1195 goto out;
1196 if (NFS_STALE(inode))
1197 goto out;
1198
1199 /* pNFS: Attributes aren't updated until we layoutcommit */
1200 if (S_ISREG(inode->i_mode)) {
1201 status = pnfs_sync_inode(inode, false);
1202 if (status)
1203 goto out;
1204 }
1205
1206 status = -ENOMEM;
1207 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1208 if (fattr == NULL)
1209 goto out;
1210
1211 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1212
1213 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1214 if (status != 0) {
1215 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1216 inode->i_sb->s_id,
1217 (unsigned long long)NFS_FILEID(inode), status);
1218 switch (status) {
1219 case -ETIMEDOUT:
1220 /* A soft timeout occurred. Use cached information? */
1221 if (server->flags & NFS_MOUNT_SOFTREVAL)
1222 status = 0;
1223 break;
1224 case -ESTALE:
1225 if (!S_ISDIR(inode->i_mode))
1226 nfs_set_inode_stale(inode);
1227 else
1228 nfs_zap_caches(inode);
1229 }
1230 goto out;
1231 }
1232
1233 status = nfs_refresh_inode(inode, fattr);
1234 if (status) {
1235 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1236 inode->i_sb->s_id,
1237 (unsigned long long)NFS_FILEID(inode), status);
1238 goto out;
1239 }
1240
1241 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1242 nfs_zap_acl_cache(inode);
1243
1244 nfs_setsecurity(inode, fattr);
1245
1246 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1247 inode->i_sb->s_id,
1248 (unsigned long long)NFS_FILEID(inode));
1249
1250 out:
1251 nfs_free_fattr(fattr);
1252 trace_nfs_revalidate_inode_exit(inode, status);
1253 return status;
1254 }
1255
nfs_attribute_cache_expired(struct inode * inode)1256 int nfs_attribute_cache_expired(struct inode *inode)
1257 {
1258 if (nfs_have_delegated_attributes(inode))
1259 return 0;
1260 return nfs_attribute_timeout(inode);
1261 }
1262
1263 /**
1264 * nfs_revalidate_inode - Revalidate the inode attributes
1265 * @inode: pointer to inode struct
1266 * @flags: cache flags to check
1267 *
1268 * Updates inode attribute information by retrieving the data from the server.
1269 */
nfs_revalidate_inode(struct inode * inode,unsigned long flags)1270 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1271 {
1272 if (!nfs_check_cache_invalid(inode, flags))
1273 return NFS_STALE(inode) ? -ESTALE : 0;
1274 return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1275 }
1276 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1277
nfs_invalidate_mapping(struct inode * inode,struct address_space * mapping)1278 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1279 {
1280 int ret;
1281
1282 nfs_fscache_invalidate(inode, 0);
1283 if (mapping->nrpages != 0) {
1284 if (S_ISREG(inode->i_mode)) {
1285 ret = nfs_sync_mapping(mapping);
1286 if (ret < 0)
1287 return ret;
1288 }
1289 ret = invalidate_inode_pages2(mapping);
1290 if (ret < 0)
1291 return ret;
1292 }
1293 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1294
1295 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1296 inode->i_sb->s_id,
1297 (unsigned long long)NFS_FILEID(inode));
1298 return 0;
1299 }
1300
1301 /**
1302 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1303 * @mapping: pointer to mapping
1304 *
1305 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1306 */
nfs_clear_invalid_mapping(struct address_space * mapping)1307 int nfs_clear_invalid_mapping(struct address_space *mapping)
1308 {
1309 struct inode *inode = mapping->host;
1310 struct nfs_inode *nfsi = NFS_I(inode);
1311 unsigned long *bitlock = &nfsi->flags;
1312 int ret = 0;
1313
1314 /*
1315 * We must clear NFS_INO_INVALID_DATA first to ensure that
1316 * invalidations that come in while we're shooting down the mappings
1317 * are respected. But, that leaves a race window where one revalidator
1318 * can clear the flag, and then another checks it before the mapping
1319 * gets invalidated. Fix that by serializing access to this part of
1320 * the function.
1321 *
1322 * At the same time, we need to allow other tasks to see whether we
1323 * might be in the middle of invalidating the pages, so we only set
1324 * the bit lock here if it looks like we're going to be doing that.
1325 */
1326 for (;;) {
1327 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1328 nfs_wait_bit_killable,
1329 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1330 if (ret)
1331 goto out;
1332 spin_lock(&inode->i_lock);
1333 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1334 spin_unlock(&inode->i_lock);
1335 continue;
1336 }
1337 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1338 break;
1339 spin_unlock(&inode->i_lock);
1340 goto out;
1341 }
1342
1343 set_bit(NFS_INO_INVALIDATING, bitlock);
1344 smp_wmb();
1345 nfsi->cache_validity &=
1346 ~(NFS_INO_INVALID_DATA | NFS_INO_DATA_INVAL_DEFER);
1347 spin_unlock(&inode->i_lock);
1348 trace_nfs_invalidate_mapping_enter(inode);
1349 ret = nfs_invalidate_mapping(inode, mapping);
1350 trace_nfs_invalidate_mapping_exit(inode, ret);
1351
1352 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1353 smp_mb__after_atomic();
1354 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1355 out:
1356 return ret;
1357 }
1358
nfs_mapping_need_revalidate_inode(struct inode * inode)1359 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1360 {
1361 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1362 NFS_STALE(inode);
1363 }
1364
nfs_revalidate_mapping_rcu(struct inode * inode)1365 int nfs_revalidate_mapping_rcu(struct inode *inode)
1366 {
1367 struct nfs_inode *nfsi = NFS_I(inode);
1368 unsigned long *bitlock = &nfsi->flags;
1369 int ret = 0;
1370
1371 if (IS_SWAPFILE(inode))
1372 goto out;
1373 if (nfs_mapping_need_revalidate_inode(inode)) {
1374 ret = -ECHILD;
1375 goto out;
1376 }
1377 spin_lock(&inode->i_lock);
1378 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1379 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1380 ret = -ECHILD;
1381 spin_unlock(&inode->i_lock);
1382 out:
1383 return ret;
1384 }
1385
1386 /**
1387 * nfs_revalidate_mapping - Revalidate the pagecache
1388 * @inode: pointer to host inode
1389 * @mapping: pointer to mapping
1390 */
nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping)1391 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1392 {
1393 /* swapfiles are not supposed to be shared. */
1394 if (IS_SWAPFILE(inode))
1395 return 0;
1396
1397 if (nfs_mapping_need_revalidate_inode(inode)) {
1398 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1399 if (ret < 0)
1400 return ret;
1401 }
1402
1403 return nfs_clear_invalid_mapping(mapping);
1404 }
1405
nfs_file_has_writers(struct nfs_inode * nfsi)1406 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1407 {
1408 struct inode *inode = &nfsi->vfs_inode;
1409
1410 if (!S_ISREG(inode->i_mode))
1411 return false;
1412 if (list_empty(&nfsi->open_files))
1413 return false;
1414 return inode_is_open_for_write(inode);
1415 }
1416
nfs_file_has_buffered_writers(struct nfs_inode * nfsi)1417 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1418 {
1419 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1420 }
1421
nfs_wcc_update_inode(struct inode * inode,struct nfs_fattr * fattr)1422 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1423 {
1424 struct timespec64 ts;
1425
1426 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1427 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1428 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1429 inode_set_iversion_raw(inode, fattr->change_attr);
1430 if (S_ISDIR(inode->i_mode))
1431 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1432 else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1433 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1434 }
1435 /* If we have atomic WCC data, we may update some attributes */
1436 ts = inode->i_ctime;
1437 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1438 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1439 && timespec64_equal(&ts, &fattr->pre_ctime)) {
1440 inode->i_ctime = fattr->ctime;
1441 }
1442
1443 ts = inode->i_mtime;
1444 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1445 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1446 && timespec64_equal(&ts, &fattr->pre_mtime)) {
1447 inode->i_mtime = fattr->mtime;
1448 }
1449 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1450 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1451 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1452 && !nfs_have_writebacks(inode)) {
1453 trace_nfs_size_wcc(inode, fattr->size);
1454 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1455 }
1456 }
1457
1458 /**
1459 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1460 * @inode: pointer to inode
1461 * @fattr: updated attributes
1462 *
1463 * Verifies the attribute cache. If we have just changed the attributes,
1464 * so that fattr carries weak cache consistency data, then it may
1465 * also update the ctime/mtime/change_attribute.
1466 */
nfs_check_inode_attributes(struct inode * inode,struct nfs_fattr * fattr)1467 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1468 {
1469 struct nfs_inode *nfsi = NFS_I(inode);
1470 loff_t cur_size, new_isize;
1471 unsigned long invalid = 0;
1472 struct timespec64 ts;
1473
1474 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1475 return 0;
1476
1477 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1478 /* Only a mounted-on-fileid? Just exit */
1479 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1480 return 0;
1481 /* Has the inode gone and changed behind our back? */
1482 } else if (nfsi->fileid != fattr->fileid) {
1483 /* Is this perhaps the mounted-on fileid? */
1484 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1485 nfsi->fileid == fattr->mounted_on_fileid)
1486 return 0;
1487 return -ESTALE;
1488 }
1489 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1490 return -ESTALE;
1491
1492
1493 if (!nfs_file_has_buffered_writers(nfsi)) {
1494 /* Verify a few of the more important attributes */
1495 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1496 invalid |= NFS_INO_INVALID_CHANGE;
1497
1498 ts = inode->i_mtime;
1499 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1500 invalid |= NFS_INO_INVALID_MTIME;
1501
1502 ts = inode->i_ctime;
1503 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1504 invalid |= NFS_INO_INVALID_CTIME;
1505
1506 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1507 cur_size = i_size_read(inode);
1508 new_isize = nfs_size_to_loff_t(fattr->size);
1509 if (cur_size != new_isize)
1510 invalid |= NFS_INO_INVALID_SIZE;
1511 }
1512 }
1513
1514 /* Have any file permissions changed? */
1515 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1516 invalid |= NFS_INO_INVALID_MODE;
1517 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1518 invalid |= NFS_INO_INVALID_OTHER;
1519 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1520 invalid |= NFS_INO_INVALID_OTHER;
1521
1522 /* Has the link count changed? */
1523 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1524 invalid |= NFS_INO_INVALID_NLINK;
1525
1526 ts = inode->i_atime;
1527 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1528 invalid |= NFS_INO_INVALID_ATIME;
1529
1530 if (invalid != 0)
1531 nfs_set_cache_invalid(inode, invalid);
1532
1533 nfsi->read_cache_jiffies = fattr->time_start;
1534 return 0;
1535 }
1536
1537 static atomic_long_t nfs_attr_generation_counter;
1538
nfs_read_attr_generation_counter(void)1539 static unsigned long nfs_read_attr_generation_counter(void)
1540 {
1541 return atomic_long_read(&nfs_attr_generation_counter);
1542 }
1543
nfs_inc_attr_generation_counter(void)1544 unsigned long nfs_inc_attr_generation_counter(void)
1545 {
1546 return atomic_long_inc_return(&nfs_attr_generation_counter);
1547 }
1548 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1549
nfs_fattr_init(struct nfs_fattr * fattr)1550 void nfs_fattr_init(struct nfs_fattr *fattr)
1551 {
1552 fattr->valid = 0;
1553 fattr->time_start = jiffies;
1554 fattr->gencount = nfs_inc_attr_generation_counter();
1555 fattr->owner_name = NULL;
1556 fattr->group_name = NULL;
1557 }
1558 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1559
1560 /**
1561 * nfs_fattr_set_barrier
1562 * @fattr: attributes
1563 *
1564 * Used to set a barrier after an attribute was updated. This
1565 * barrier ensures that older attributes from RPC calls that may
1566 * have raced with our update cannot clobber these new values.
1567 * Note that you are still responsible for ensuring that other
1568 * operations which change the attribute on the server do not
1569 * collide.
1570 */
nfs_fattr_set_barrier(struct nfs_fattr * fattr)1571 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1572 {
1573 fattr->gencount = nfs_inc_attr_generation_counter();
1574 }
1575
nfs_alloc_fattr(void)1576 struct nfs_fattr *nfs_alloc_fattr(void)
1577 {
1578 struct nfs_fattr *fattr;
1579
1580 fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1581 if (fattr != NULL) {
1582 nfs_fattr_init(fattr);
1583 fattr->label = NULL;
1584 }
1585 return fattr;
1586 }
1587 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1588
nfs_alloc_fattr_with_label(struct nfs_server * server)1589 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1590 {
1591 struct nfs_fattr *fattr = nfs_alloc_fattr();
1592
1593 if (!fattr)
1594 return NULL;
1595
1596 fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1597 if (IS_ERR(fattr->label)) {
1598 kfree(fattr);
1599 return NULL;
1600 }
1601
1602 return fattr;
1603 }
1604 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1605
nfs_alloc_fhandle(void)1606 struct nfs_fh *nfs_alloc_fhandle(void)
1607 {
1608 struct nfs_fh *fh;
1609
1610 fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1611 if (fh != NULL)
1612 fh->size = 0;
1613 return fh;
1614 }
1615 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1616
1617 #ifdef NFS_DEBUG
1618 /*
1619 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1620 * in the same way that wireshark does
1621 *
1622 * @fh: file handle
1623 *
1624 * For debugging only.
1625 */
_nfs_display_fhandle_hash(const struct nfs_fh * fh)1626 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1627 {
1628 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1629 * not on the result */
1630 return nfs_fhandle_hash(fh);
1631 }
1632 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1633
1634 /*
1635 * _nfs_display_fhandle - display an NFS file handle on the console
1636 *
1637 * @fh: file handle to display
1638 * @caption: display caption
1639 *
1640 * For debugging only.
1641 */
_nfs_display_fhandle(const struct nfs_fh * fh,const char * caption)1642 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1643 {
1644 unsigned short i;
1645
1646 if (fh == NULL || fh->size == 0) {
1647 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1648 return;
1649 }
1650
1651 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1652 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1653 for (i = 0; i < fh->size; i += 16) {
1654 __be32 *pos = (__be32 *)&fh->data[i];
1655
1656 switch ((fh->size - i - 1) >> 2) {
1657 case 0:
1658 printk(KERN_DEFAULT " %08x\n",
1659 be32_to_cpup(pos));
1660 break;
1661 case 1:
1662 printk(KERN_DEFAULT " %08x %08x\n",
1663 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1664 break;
1665 case 2:
1666 printk(KERN_DEFAULT " %08x %08x %08x\n",
1667 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1668 be32_to_cpup(pos + 2));
1669 break;
1670 default:
1671 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1672 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1673 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1674 }
1675 }
1676 }
1677 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1678 #endif
1679
1680 /**
1681 * nfs_inode_attrs_cmp_generic - compare attributes
1682 * @fattr: attributes
1683 * @inode: pointer to inode
1684 *
1685 * Attempt to divine whether or not an RPC call reply carrying stale
1686 * attributes got scheduled after another call carrying updated ones.
1687 * Note also the check for wraparound of 'attr_gencount'
1688 *
1689 * The function returns '1' if it thinks the attributes in @fattr are
1690 * more recent than the ones cached in @inode. Otherwise it returns
1691 * the value '0'.
1692 */
nfs_inode_attrs_cmp_generic(const struct nfs_fattr * fattr,const struct inode * inode)1693 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1694 const struct inode *inode)
1695 {
1696 unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1697
1698 return (long)(fattr->gencount - attr_gencount) > 0 ||
1699 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1700 }
1701
1702 /**
1703 * nfs_inode_attrs_cmp_monotonic - compare attributes
1704 * @fattr: attributes
1705 * @inode: pointer to inode
1706 *
1707 * Attempt to divine whether or not an RPC call reply carrying stale
1708 * attributes got scheduled after another call carrying updated ones.
1709 *
1710 * We assume that the server observes monotonic semantics for
1711 * the change attribute, so a larger value means that the attributes in
1712 * @fattr are more recent, in which case the function returns the
1713 * value '1'.
1714 * A return value of '0' indicates no measurable change
1715 * A return value of '-1' means that the attributes in @inode are
1716 * more recent.
1717 */
nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1718 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1719 const struct inode *inode)
1720 {
1721 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1722 if (diff > 0)
1723 return 1;
1724 return diff == 0 ? 0 : -1;
1725 }
1726
1727 /**
1728 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1729 * @fattr: attributes
1730 * @inode: pointer to inode
1731 *
1732 * Attempt to divine whether or not an RPC call reply carrying stale
1733 * attributes got scheduled after another call carrying updated ones.
1734 *
1735 * We assume that the server observes strictly monotonic semantics for
1736 * the change attribute, so a larger value means that the attributes in
1737 * @fattr are more recent, in which case the function returns the
1738 * value '1'.
1739 * A return value of '-1' means that the attributes in @inode are
1740 * more recent or unchanged.
1741 */
nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1742 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1743 const struct inode *inode)
1744 {
1745 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1746 }
1747
1748 /**
1749 * nfs_inode_attrs_cmp - compare attributes
1750 * @fattr: attributes
1751 * @inode: pointer to inode
1752 *
1753 * This function returns '1' if it thinks the attributes in @fattr are
1754 * more recent than the ones cached in @inode. It returns '-1' if
1755 * the attributes in @inode are more recent than the ones in @fattr,
1756 * and it returns 0 if not sure.
1757 */
nfs_inode_attrs_cmp(const struct nfs_fattr * fattr,const struct inode * inode)1758 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1759 const struct inode *inode)
1760 {
1761 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1762 return 1;
1763 switch (NFS_SERVER(inode)->change_attr_type) {
1764 case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1765 break;
1766 case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1767 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1768 break;
1769 return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1770 default:
1771 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1772 break;
1773 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1774 }
1775 return 0;
1776 }
1777
1778 /**
1779 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1780 * @fattr: attributes
1781 * @inode: pointer to inode
1782 *
1783 * Returns '1' if the last attribute update left the inode cached
1784 * attributes in a partially unrevalidated state, and @fattr
1785 * matches the change attribute of that partial update.
1786 * Otherwise returns '0'.
1787 */
nfs_inode_finish_partial_attr_update(const struct nfs_fattr * fattr,const struct inode * inode)1788 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1789 const struct inode *inode)
1790 {
1791 const unsigned long check_valid =
1792 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1793 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1794 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1795 NFS_INO_INVALID_NLINK;
1796 unsigned long cache_validity = NFS_I(inode)->cache_validity;
1797 enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1798
1799 if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1800 !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1801 (cache_validity & check_valid) != 0 &&
1802 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1803 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1804 return 1;
1805 return 0;
1806 }
1807
nfs_refresh_inode_locked(struct inode * inode,struct nfs_fattr * fattr)1808 static int nfs_refresh_inode_locked(struct inode *inode,
1809 struct nfs_fattr *fattr)
1810 {
1811 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1812 int ret = 0;
1813
1814 trace_nfs_refresh_inode_enter(inode);
1815
1816 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1817 ret = nfs_update_inode(inode, fattr);
1818 else if (attr_cmp == 0)
1819 ret = nfs_check_inode_attributes(inode, fattr);
1820
1821 trace_nfs_refresh_inode_exit(inode, ret);
1822 return ret;
1823 }
1824
1825 /**
1826 * nfs_refresh_inode - try to update the inode attribute cache
1827 * @inode: pointer to inode
1828 * @fattr: updated attributes
1829 *
1830 * Check that an RPC call that returned attributes has not overlapped with
1831 * other recent updates of the inode metadata, then decide whether it is
1832 * safe to do a full update of the inode attributes, or whether just to
1833 * call nfs_check_inode_attributes.
1834 */
nfs_refresh_inode(struct inode * inode,struct nfs_fattr * fattr)1835 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1836 {
1837 int status;
1838
1839 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1840 return 0;
1841 spin_lock(&inode->i_lock);
1842 status = nfs_refresh_inode_locked(inode, fattr);
1843 spin_unlock(&inode->i_lock);
1844
1845 return status;
1846 }
1847 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1848
nfs_post_op_update_inode_locked(struct inode * inode,struct nfs_fattr * fattr,unsigned int invalid)1849 static int nfs_post_op_update_inode_locked(struct inode *inode,
1850 struct nfs_fattr *fattr, unsigned int invalid)
1851 {
1852 if (S_ISDIR(inode->i_mode))
1853 invalid |= NFS_INO_INVALID_DATA;
1854 nfs_set_cache_invalid(inode, invalid);
1855 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1856 return 0;
1857 return nfs_refresh_inode_locked(inode, fattr);
1858 }
1859
1860 /**
1861 * nfs_post_op_update_inode - try to update the inode attribute cache
1862 * @inode: pointer to inode
1863 * @fattr: updated attributes
1864 *
1865 * After an operation that has changed the inode metadata, mark the
1866 * attribute cache as being invalid, then try to update it.
1867 *
1868 * NB: if the server didn't return any post op attributes, this
1869 * function will force the retrieval of attributes before the next
1870 * NFS request. Thus it should be used only for operations that
1871 * are expected to change one or more attributes, to avoid
1872 * unnecessary NFS requests and trips through nfs_update_inode().
1873 */
nfs_post_op_update_inode(struct inode * inode,struct nfs_fattr * fattr)1874 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1875 {
1876 int status;
1877
1878 spin_lock(&inode->i_lock);
1879 nfs_fattr_set_barrier(fattr);
1880 status = nfs_post_op_update_inode_locked(inode, fattr,
1881 NFS_INO_INVALID_CHANGE
1882 | NFS_INO_INVALID_CTIME
1883 | NFS_INO_REVAL_FORCED);
1884 spin_unlock(&inode->i_lock);
1885
1886 return status;
1887 }
1888 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1889
1890 /**
1891 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1892 * @inode: pointer to inode
1893 * @fattr: updated attributes
1894 *
1895 * After an operation that has changed the inode metadata, mark the
1896 * attribute cache as being invalid, then try to update it. Fake up
1897 * weak cache consistency data, if none exist.
1898 *
1899 * This function is mainly designed to be used by the ->write_done() functions.
1900 */
nfs_post_op_update_inode_force_wcc_locked(struct inode * inode,struct nfs_fattr * fattr)1901 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1902 {
1903 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1904 int status;
1905
1906 /* Don't do a WCC update if these attributes are already stale */
1907 if (attr_cmp < 0)
1908 return 0;
1909 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
1910 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1911 | NFS_ATTR_FATTR_PRESIZE
1912 | NFS_ATTR_FATTR_PREMTIME
1913 | NFS_ATTR_FATTR_PRECTIME);
1914 goto out_noforce;
1915 }
1916 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1917 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1918 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1919 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1920 }
1921 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1922 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1923 fattr->pre_ctime = inode->i_ctime;
1924 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1925 }
1926 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1927 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1928 fattr->pre_mtime = inode->i_mtime;
1929 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1930 }
1931 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1932 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1933 fattr->pre_size = i_size_read(inode);
1934 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1935 }
1936 out_noforce:
1937 status = nfs_post_op_update_inode_locked(inode, fattr,
1938 NFS_INO_INVALID_CHANGE
1939 | NFS_INO_INVALID_CTIME
1940 | NFS_INO_INVALID_MTIME
1941 | NFS_INO_INVALID_BLOCKS);
1942 return status;
1943 }
1944
1945 /**
1946 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1947 * @inode: pointer to inode
1948 * @fattr: updated attributes
1949 *
1950 * After an operation that has changed the inode metadata, mark the
1951 * attribute cache as being invalid, then try to update it. Fake up
1952 * weak cache consistency data, if none exist.
1953 *
1954 * This function is mainly designed to be used by the ->write_done() functions.
1955 */
nfs_post_op_update_inode_force_wcc(struct inode * inode,struct nfs_fattr * fattr)1956 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1957 {
1958 int status;
1959
1960 spin_lock(&inode->i_lock);
1961 nfs_fattr_set_barrier(fattr);
1962 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1963 spin_unlock(&inode->i_lock);
1964 return status;
1965 }
1966 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1967
1968
1969 /*
1970 * Many nfs protocol calls return the new file attributes after
1971 * an operation. Here we update the inode to reflect the state
1972 * of the server's inode.
1973 *
1974 * This is a bit tricky because we have to make sure all dirty pages
1975 * have been sent off to the server before calling invalidate_inode_pages.
1976 * To make sure no other process adds more write requests while we try
1977 * our best to flush them, we make them sleep during the attribute refresh.
1978 *
1979 * A very similar scenario holds for the dir cache.
1980 */
nfs_update_inode(struct inode * inode,struct nfs_fattr * fattr)1981 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1982 {
1983 struct nfs_server *server = NFS_SERVER(inode);
1984 struct nfs_inode *nfsi = NFS_I(inode);
1985 loff_t cur_isize, new_isize;
1986 u64 fattr_supported = server->fattr_valid;
1987 unsigned long invalid = 0;
1988 unsigned long now = jiffies;
1989 unsigned long save_cache_validity;
1990 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1991 bool cache_revalidated = true;
1992 bool attr_changed = false;
1993 bool have_delegation;
1994
1995 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1996 __func__, inode->i_sb->s_id, inode->i_ino,
1997 nfs_display_fhandle_hash(NFS_FH(inode)),
1998 atomic_read(&inode->i_count), fattr->valid);
1999
2000 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2001 /* Only a mounted-on-fileid? Just exit */
2002 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2003 return 0;
2004 /* Has the inode gone and changed behind our back? */
2005 } else if (nfsi->fileid != fattr->fileid) {
2006 /* Is this perhaps the mounted-on fileid? */
2007 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2008 nfsi->fileid == fattr->mounted_on_fileid)
2009 return 0;
2010 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2011 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2012 NFS_SERVER(inode)->nfs_client->cl_hostname,
2013 inode->i_sb->s_id, (long long)nfsi->fileid,
2014 (long long)fattr->fileid);
2015 goto out_err;
2016 }
2017
2018 /*
2019 * Make sure the inode's type hasn't changed.
2020 */
2021 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2022 /*
2023 * Big trouble! The inode has become a different object.
2024 */
2025 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2026 __func__, inode->i_ino, inode->i_mode, fattr->mode);
2027 goto out_err;
2028 }
2029
2030 /* Update the fsid? */
2031 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2032 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2033 !IS_AUTOMOUNT(inode))
2034 server->fsid = fattr->fsid;
2035
2036 /* Save the delegation state before clearing cache_validity */
2037 have_delegation = nfs_have_delegated_attributes(inode);
2038
2039 /*
2040 * Update the read time so we don't revalidate too often.
2041 */
2042 nfsi->read_cache_jiffies = fattr->time_start;
2043
2044 save_cache_validity = nfsi->cache_validity;
2045 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2046 | NFS_INO_INVALID_ATIME
2047 | NFS_INO_REVAL_FORCED
2048 | NFS_INO_INVALID_BLOCKS);
2049
2050 /* Do atomic weak cache consistency updates */
2051 nfs_wcc_update_inode(inode, fattr);
2052
2053 if (pnfs_layoutcommit_outstanding(inode)) {
2054 nfsi->cache_validity |=
2055 save_cache_validity &
2056 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2057 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2058 NFS_INO_INVALID_BLOCKS);
2059 cache_revalidated = false;
2060 }
2061
2062 /* More cache consistency checks */
2063 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2064 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2065 /* Could it be a race with writeback? */
2066 if (!(have_writers || have_delegation)) {
2067 invalid |= NFS_INO_INVALID_DATA
2068 | NFS_INO_INVALID_ACCESS
2069 | NFS_INO_INVALID_ACL
2070 | NFS_INO_INVALID_XATTR;
2071 /* Force revalidate of all attributes */
2072 save_cache_validity |= NFS_INO_INVALID_CTIME
2073 | NFS_INO_INVALID_MTIME
2074 | NFS_INO_INVALID_SIZE
2075 | NFS_INO_INVALID_BLOCKS
2076 | NFS_INO_INVALID_NLINK
2077 | NFS_INO_INVALID_MODE
2078 | NFS_INO_INVALID_OTHER;
2079 if (S_ISDIR(inode->i_mode))
2080 nfs_force_lookup_revalidate(inode);
2081 attr_changed = true;
2082 dprintk("NFS: change_attr change on server for file %s/%ld\n",
2083 inode->i_sb->s_id,
2084 inode->i_ino);
2085 } else if (!have_delegation)
2086 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
2087 inode_set_iversion_raw(inode, fattr->change_attr);
2088 }
2089 } else {
2090 nfsi->cache_validity |=
2091 save_cache_validity & NFS_INO_INVALID_CHANGE;
2092 if (!have_delegation ||
2093 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2094 cache_revalidated = false;
2095 }
2096
2097 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2098 inode->i_mtime = fattr->mtime;
2099 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2100 nfsi->cache_validity |=
2101 save_cache_validity & NFS_INO_INVALID_MTIME;
2102
2103 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2104 inode->i_ctime = fattr->ctime;
2105 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2106 nfsi->cache_validity |=
2107 save_cache_validity & NFS_INO_INVALID_CTIME;
2108
2109 /* Check if our cached file size is stale */
2110 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2111 new_isize = nfs_size_to_loff_t(fattr->size);
2112 cur_isize = i_size_read(inode);
2113 if (new_isize != cur_isize && !have_delegation) {
2114 /* Do we perhaps have any outstanding writes, or has
2115 * the file grown beyond our last write? */
2116 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2117 trace_nfs_size_update(inode, new_isize);
2118 i_size_write(inode, new_isize);
2119 if (!have_writers)
2120 invalid |= NFS_INO_INVALID_DATA;
2121 }
2122 }
2123 if (new_isize == 0 &&
2124 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2125 NFS_ATTR_FATTR_BLOCKS_USED))) {
2126 fattr->du.nfs3.used = 0;
2127 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2128 }
2129 } else
2130 nfsi->cache_validity |=
2131 save_cache_validity & NFS_INO_INVALID_SIZE;
2132
2133 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2134 inode->i_atime = fattr->atime;
2135 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2136 nfsi->cache_validity |=
2137 save_cache_validity & NFS_INO_INVALID_ATIME;
2138
2139 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2140 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2141 umode_t newmode = inode->i_mode & S_IFMT;
2142 newmode |= fattr->mode & S_IALLUGO;
2143 inode->i_mode = newmode;
2144 invalid |= NFS_INO_INVALID_ACCESS
2145 | NFS_INO_INVALID_ACL;
2146 }
2147 } else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2148 nfsi->cache_validity |=
2149 save_cache_validity & NFS_INO_INVALID_MODE;
2150
2151 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2152 if (!uid_eq(inode->i_uid, fattr->uid)) {
2153 invalid |= NFS_INO_INVALID_ACCESS
2154 | NFS_INO_INVALID_ACL;
2155 inode->i_uid = fattr->uid;
2156 }
2157 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2158 nfsi->cache_validity |=
2159 save_cache_validity & NFS_INO_INVALID_OTHER;
2160
2161 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2162 if (!gid_eq(inode->i_gid, fattr->gid)) {
2163 invalid |= NFS_INO_INVALID_ACCESS
2164 | NFS_INO_INVALID_ACL;
2165 inode->i_gid = fattr->gid;
2166 }
2167 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2168 nfsi->cache_validity |=
2169 save_cache_validity & NFS_INO_INVALID_OTHER;
2170
2171 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2172 if (inode->i_nlink != fattr->nlink)
2173 set_nlink(inode, fattr->nlink);
2174 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2175 nfsi->cache_validity |=
2176 save_cache_validity & NFS_INO_INVALID_NLINK;
2177
2178 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2179 /*
2180 * report the blocks in 512byte units
2181 */
2182 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2183 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2184 nfsi->cache_validity |=
2185 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2186
2187 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2188 inode->i_blocks = fattr->du.nfs2.blocks;
2189 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2190 nfsi->cache_validity |=
2191 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2192
2193 /* Update attrtimeo value if we're out of the unstable period */
2194 if (attr_changed) {
2195 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2196 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2197 nfsi->attrtimeo_timestamp = now;
2198 /* Set barrier to be more recent than all outstanding updates */
2199 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2200 } else {
2201 if (cache_revalidated) {
2202 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2203 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2204 nfsi->attrtimeo <<= 1;
2205 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2206 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2207 }
2208 nfsi->attrtimeo_timestamp = now;
2209 }
2210 /* Set the barrier to be more recent than this fattr */
2211 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2212 nfsi->attr_gencount = fattr->gencount;
2213 }
2214
2215 /* Don't invalidate the data if we were to blame */
2216 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2217 || S_ISLNK(inode->i_mode)))
2218 invalid &= ~NFS_INO_INVALID_DATA;
2219 nfs_set_cache_invalid(inode, invalid);
2220
2221 return 0;
2222 out_err:
2223 /*
2224 * No need to worry about unhashing the dentry, as the
2225 * lookup validation will know that the inode is bad.
2226 * (But we fall through to invalidate the caches.)
2227 */
2228 nfs_set_inode_stale_locked(inode);
2229 return -ESTALE;
2230 }
2231
nfs_alloc_inode(struct super_block * sb)2232 struct inode *nfs_alloc_inode(struct super_block *sb)
2233 {
2234 struct nfs_inode *nfsi;
2235 nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2236 if (!nfsi)
2237 return NULL;
2238 nfsi->flags = 0UL;
2239 nfsi->cache_validity = 0UL;
2240 #if IS_ENABLED(CONFIG_NFS_V4)
2241 nfsi->nfs4_acl = NULL;
2242 #endif /* CONFIG_NFS_V4 */
2243 #ifdef CONFIG_NFS_V4_2
2244 nfsi->xattr_cache = NULL;
2245 #endif
2246 return &nfsi->vfs_inode;
2247 }
2248 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2249
nfs_free_inode(struct inode * inode)2250 void nfs_free_inode(struct inode *inode)
2251 {
2252 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2253 }
2254 EXPORT_SYMBOL_GPL(nfs_free_inode);
2255
nfs4_init_once(struct nfs_inode * nfsi)2256 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2257 {
2258 #if IS_ENABLED(CONFIG_NFS_V4)
2259 INIT_LIST_HEAD(&nfsi->open_states);
2260 nfsi->delegation = NULL;
2261 init_rwsem(&nfsi->rwsem);
2262 nfsi->layout = NULL;
2263 #endif
2264 }
2265
init_once(void * foo)2266 static void init_once(void *foo)
2267 {
2268 struct nfs_inode *nfsi = foo;
2269
2270 inode_init_once(&nfsi->vfs_inode);
2271 INIT_LIST_HEAD(&nfsi->open_files);
2272 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2273 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2274 nfs4_init_once(nfsi);
2275 }
2276
nfs_init_inodecache(void)2277 static int __init nfs_init_inodecache(void)
2278 {
2279 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2280 sizeof(struct nfs_inode),
2281 0, (SLAB_RECLAIM_ACCOUNT|
2282 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2283 init_once);
2284 if (nfs_inode_cachep == NULL)
2285 return -ENOMEM;
2286
2287 return 0;
2288 }
2289
nfs_destroy_inodecache(void)2290 static void nfs_destroy_inodecache(void)
2291 {
2292 /*
2293 * Make sure all delayed rcu free inodes are flushed before we
2294 * destroy cache.
2295 */
2296 rcu_barrier();
2297 kmem_cache_destroy(nfs_inode_cachep);
2298 }
2299
2300 struct workqueue_struct *nfsiod_workqueue;
2301 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2302
2303 /*
2304 * start up the nfsiod workqueue
2305 */
nfsiod_start(void)2306 static int nfsiod_start(void)
2307 {
2308 struct workqueue_struct *wq;
2309 dprintk("RPC: creating workqueue nfsiod\n");
2310 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2311 if (wq == NULL)
2312 return -ENOMEM;
2313 nfsiod_workqueue = wq;
2314 return 0;
2315 }
2316
2317 /*
2318 * Destroy the nfsiod workqueue
2319 */
nfsiod_stop(void)2320 static void nfsiod_stop(void)
2321 {
2322 struct workqueue_struct *wq;
2323
2324 wq = nfsiod_workqueue;
2325 if (wq == NULL)
2326 return;
2327 nfsiod_workqueue = NULL;
2328 destroy_workqueue(wq);
2329 }
2330
2331 unsigned int nfs_net_id;
2332 EXPORT_SYMBOL_GPL(nfs_net_id);
2333
nfs_net_init(struct net * net)2334 static int nfs_net_init(struct net *net)
2335 {
2336 nfs_clients_init(net);
2337 return nfs_fs_proc_net_init(net);
2338 }
2339
nfs_net_exit(struct net * net)2340 static void nfs_net_exit(struct net *net)
2341 {
2342 nfs_fs_proc_net_exit(net);
2343 nfs_clients_exit(net);
2344 }
2345
2346 static struct pernet_operations nfs_net_ops = {
2347 .init = nfs_net_init,
2348 .exit = nfs_net_exit,
2349 .id = &nfs_net_id,
2350 .size = sizeof(struct nfs_net),
2351 };
2352
2353 /*
2354 * Initialize NFS
2355 */
init_nfs_fs(void)2356 static int __init init_nfs_fs(void)
2357 {
2358 int err;
2359
2360 err = nfs_sysfs_init();
2361 if (err < 0)
2362 goto out10;
2363
2364 err = register_pernet_subsys(&nfs_net_ops);
2365 if (err < 0)
2366 goto out9;
2367
2368 err = nfsiod_start();
2369 if (err)
2370 goto out7;
2371
2372 err = nfs_fs_proc_init();
2373 if (err)
2374 goto out6;
2375
2376 err = nfs_init_nfspagecache();
2377 if (err)
2378 goto out5;
2379
2380 err = nfs_init_inodecache();
2381 if (err)
2382 goto out4;
2383
2384 err = nfs_init_readpagecache();
2385 if (err)
2386 goto out3;
2387
2388 err = nfs_init_writepagecache();
2389 if (err)
2390 goto out2;
2391
2392 err = nfs_init_directcache();
2393 if (err)
2394 goto out1;
2395
2396 rpc_proc_register(&init_net, &nfs_rpcstat);
2397
2398 err = register_nfs_fs();
2399 if (err)
2400 goto out0;
2401
2402 return 0;
2403 out0:
2404 rpc_proc_unregister(&init_net, "nfs");
2405 nfs_destroy_directcache();
2406 out1:
2407 nfs_destroy_writepagecache();
2408 out2:
2409 nfs_destroy_readpagecache();
2410 out3:
2411 nfs_destroy_inodecache();
2412 out4:
2413 nfs_destroy_nfspagecache();
2414 out5:
2415 nfs_fs_proc_exit();
2416 out6:
2417 nfsiod_stop();
2418 out7:
2419 unregister_pernet_subsys(&nfs_net_ops);
2420 out9:
2421 nfs_sysfs_exit();
2422 out10:
2423 return err;
2424 }
2425
exit_nfs_fs(void)2426 static void __exit exit_nfs_fs(void)
2427 {
2428 nfs_destroy_directcache();
2429 nfs_destroy_writepagecache();
2430 nfs_destroy_readpagecache();
2431 nfs_destroy_inodecache();
2432 nfs_destroy_nfspagecache();
2433 unregister_pernet_subsys(&nfs_net_ops);
2434 rpc_proc_unregister(&init_net, "nfs");
2435 unregister_nfs_fs();
2436 nfs_fs_proc_exit();
2437 nfsiod_stop();
2438 nfs_sysfs_exit();
2439 }
2440
2441 /* Not quite true; I just maintain it */
2442 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2443 MODULE_LICENSE("GPL");
2444 module_param(enable_ino64, bool, 0644);
2445
2446 module_init(init_nfs_fs)
2447 module_exit(exit_nfs_fs)
2448