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 if (unlikely(nfs_current_task_exiting()))
78 return -EINTR;
79 schedule();
80 if (signal_pending_state(mode, current))
81 return -ERESTARTSYS;
82 return 0;
83 }
84 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
85
86 /**
87 * nfs_compat_user_ino64 - returns the user-visible inode number
88 * @fileid: 64-bit fileid
89 *
90 * This function returns a 32-bit inode number if the boot parameter
91 * nfs.enable_ino64 is zero.
92 */
nfs_compat_user_ino64(u64 fileid)93 u64 nfs_compat_user_ino64(u64 fileid)
94 {
95 #ifdef CONFIG_COMPAT
96 compat_ulong_t ino;
97 #else
98 unsigned long ino;
99 #endif
100
101 if (enable_ino64)
102 return fileid;
103 ino = fileid;
104 if (sizeof(ino) < sizeof(fileid))
105 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
106 return ino;
107 }
108
nfs_drop_inode(struct inode * inode)109 int nfs_drop_inode(struct inode *inode)
110 {
111 return NFS_STALE(inode) || generic_drop_inode(inode);
112 }
113 EXPORT_SYMBOL_GPL(nfs_drop_inode);
114
nfs_clear_inode(struct inode * inode)115 void nfs_clear_inode(struct inode *inode)
116 {
117 /*
118 * The following should never happen...
119 */
120 WARN_ON_ONCE(nfs_have_writebacks(inode));
121 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
122 nfs_zap_acl_cache(inode);
123 nfs_access_zap_cache(inode);
124 nfs_fscache_clear_inode(inode);
125 }
126 EXPORT_SYMBOL_GPL(nfs_clear_inode);
127
nfs_evict_inode(struct inode * inode)128 void nfs_evict_inode(struct inode *inode)
129 {
130 truncate_inode_pages_final(&inode->i_data);
131 clear_inode(inode);
132 nfs_clear_inode(inode);
133 }
134
nfs_sync_inode(struct inode * inode)135 int nfs_sync_inode(struct inode *inode)
136 {
137 inode_dio_wait(inode);
138 return nfs_wb_all(inode);
139 }
140 EXPORT_SYMBOL_GPL(nfs_sync_inode);
141
142 /**
143 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
144 * @mapping: pointer to struct address_space
145 */
nfs_sync_mapping(struct address_space * mapping)146 int nfs_sync_mapping(struct address_space *mapping)
147 {
148 int ret = 0;
149
150 if (mapping->nrpages != 0) {
151 unmap_mapping_range(mapping, 0, 0, 0);
152 ret = nfs_wb_all(mapping->host);
153 }
154 return ret;
155 }
156
nfs_attribute_timeout(struct inode * inode)157 static int nfs_attribute_timeout(struct inode *inode)
158 {
159 struct nfs_inode *nfsi = NFS_I(inode);
160
161 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
162 }
163
nfs_check_cache_flags_invalid(struct inode * inode,unsigned long flags)164 static bool nfs_check_cache_flags_invalid(struct inode *inode,
165 unsigned long flags)
166 {
167 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
168
169 return (cache_validity & flags) != 0;
170 }
171
nfs_check_cache_invalid(struct inode * inode,unsigned long flags)172 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
173 {
174 if (nfs_check_cache_flags_invalid(inode, flags))
175 return true;
176 return nfs_attribute_cache_expired(inode);
177 }
178 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
179
180 #ifdef CONFIG_NFS_V4_2
nfs_has_xattr_cache(const struct nfs_inode * nfsi)181 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
182 {
183 return nfsi->xattr_cache != NULL;
184 }
185 #else
nfs_has_xattr_cache(const struct nfs_inode * nfsi)186 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
187 {
188 return false;
189 }
190 #endif
191
nfs_set_cache_invalid(struct inode * inode,unsigned long flags)192 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
193 {
194 struct nfs_inode *nfsi = NFS_I(inode);
195 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
196
197 if (have_delegation) {
198 if (!(flags & NFS_INO_REVAL_FORCED))
199 flags &= ~(NFS_INO_INVALID_MODE |
200 NFS_INO_INVALID_OTHER |
201 NFS_INO_INVALID_XATTR);
202 flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
203 }
204
205 if (!nfs_has_xattr_cache(nfsi))
206 flags &= ~NFS_INO_INVALID_XATTR;
207 if (flags & NFS_INO_INVALID_DATA)
208 nfs_fscache_invalidate(inode, 0);
209 flags &= ~NFS_INO_REVAL_FORCED;
210
211 flags |= nfsi->cache_validity;
212 if (inode->i_mapping->nrpages == 0)
213 flags &= ~NFS_INO_INVALID_DATA;
214
215 /* pairs with nfs_clear_invalid_mapping()'s smp_load_acquire() */
216 smp_store_release(&nfsi->cache_validity, flags);
217
218 if (inode->i_mapping->nrpages == 0 ||
219 nfsi->cache_validity & NFS_INO_INVALID_DATA) {
220 nfs_ooo_clear(nfsi);
221 }
222 trace_nfs_set_cache_invalid(inode, 0);
223 }
224 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
225
226 /*
227 * Invalidate the local caches
228 */
nfs_zap_caches_locked(struct inode * inode)229 static void nfs_zap_caches_locked(struct inode *inode)
230 {
231 struct nfs_inode *nfsi = NFS_I(inode);
232 int mode = inode->i_mode;
233
234 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
235
236 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
237 nfsi->attrtimeo_timestamp = jiffies;
238
239 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
240 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
241 NFS_INO_INVALID_DATA |
242 NFS_INO_INVALID_ACCESS |
243 NFS_INO_INVALID_ACL |
244 NFS_INO_INVALID_XATTR);
245 else
246 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
247 NFS_INO_INVALID_ACCESS |
248 NFS_INO_INVALID_ACL |
249 NFS_INO_INVALID_XATTR);
250 nfs_zap_label_cache_locked(nfsi);
251 }
252
nfs_zap_caches(struct inode * inode)253 void nfs_zap_caches(struct inode *inode)
254 {
255 spin_lock(&inode->i_lock);
256 nfs_zap_caches_locked(inode);
257 spin_unlock(&inode->i_lock);
258 }
259
nfs_zap_mapping(struct inode * inode,struct address_space * mapping)260 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
261 {
262 if (mapping->nrpages != 0) {
263 spin_lock(&inode->i_lock);
264 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
265 spin_unlock(&inode->i_lock);
266 }
267 }
268
nfs_zap_acl_cache(struct inode * inode)269 void nfs_zap_acl_cache(struct inode *inode)
270 {
271 void (*clear_acl_cache)(struct inode *);
272
273 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
274 if (clear_acl_cache != NULL)
275 clear_acl_cache(inode);
276 spin_lock(&inode->i_lock);
277 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
278 spin_unlock(&inode->i_lock);
279 }
280 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
281
nfs_invalidate_atime(struct inode * inode)282 void nfs_invalidate_atime(struct inode *inode)
283 {
284 spin_lock(&inode->i_lock);
285 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
286 spin_unlock(&inode->i_lock);
287 }
288 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
289
290 /*
291 * Invalidate, but do not unhash, the inode.
292 * NB: must be called with inode->i_lock held!
293 */
nfs_set_inode_stale_locked(struct inode * inode)294 static void nfs_set_inode_stale_locked(struct inode *inode)
295 {
296 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
297 nfs_zap_caches_locked(inode);
298 trace_nfs_set_inode_stale(inode);
299 }
300
nfs_set_inode_stale(struct inode * inode)301 void nfs_set_inode_stale(struct inode *inode)
302 {
303 spin_lock(&inode->i_lock);
304 nfs_set_inode_stale_locked(inode);
305 spin_unlock(&inode->i_lock);
306 }
307
308 struct nfs_find_desc {
309 struct nfs_fh *fh;
310 struct nfs_fattr *fattr;
311 };
312
313 /*
314 * In NFSv3 we can have 64bit inode numbers. In order to support
315 * this, and re-exported directories (also seen in NFSv2)
316 * we are forced to allow 2 different inodes to have the same
317 * i_ino.
318 */
319 static int
nfs_find_actor(struct inode * inode,void * opaque)320 nfs_find_actor(struct inode *inode, void *opaque)
321 {
322 struct nfs_find_desc *desc = opaque;
323 struct nfs_fh *fh = desc->fh;
324 struct nfs_fattr *fattr = desc->fattr;
325
326 if (NFS_FILEID(inode) != fattr->fileid)
327 return 0;
328 if (inode_wrong_type(inode, fattr->mode))
329 return 0;
330 if (nfs_compare_fh(NFS_FH(inode), fh))
331 return 0;
332 if (is_bad_inode(inode) || NFS_STALE(inode))
333 return 0;
334 return 1;
335 }
336
337 static int
nfs_init_locked(struct inode * inode,void * opaque)338 nfs_init_locked(struct inode *inode, void *opaque)
339 {
340 struct nfs_find_desc *desc = opaque;
341 struct nfs_fattr *fattr = desc->fattr;
342
343 set_nfs_fileid(inode, fattr->fileid);
344 inode->i_mode = fattr->mode;
345 nfs_copy_fh(NFS_FH(inode), desc->fh);
346 return 0;
347 }
348
349 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
nfs_clear_label_invalid(struct inode * inode)350 static void nfs_clear_label_invalid(struct inode *inode)
351 {
352 spin_lock(&inode->i_lock);
353 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
354 spin_unlock(&inode->i_lock);
355 }
356
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)357 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
358 {
359 int error;
360
361 if (fattr->label == NULL)
362 return;
363
364 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
365 error = security_inode_notifysecctx(inode, fattr->label->label,
366 fattr->label->len);
367 if (error)
368 printk(KERN_ERR "%s() %s %d "
369 "security_inode_notifysecctx() %d\n",
370 __func__,
371 (char *)fattr->label->label,
372 fattr->label->len, error);
373 nfs_clear_label_invalid(inode);
374 }
375 }
376
nfs4_label_alloc(struct nfs_server * server,gfp_t flags)377 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
378 {
379 struct nfs4_label *label;
380
381 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
382 return NULL;
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
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)399 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
400 {
401 }
402 #endif
403 EXPORT_SYMBOL_GPL(nfs_setsecurity);
404
405 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
406 struct inode *
nfs_ilookup(struct super_block * sb,struct nfs_fattr * fattr,struct nfs_fh * fh)407 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
408 {
409 struct nfs_find_desc desc = {
410 .fh = fh,
411 .fattr = fattr,
412 };
413 struct inode *inode;
414 unsigned long hash;
415
416 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
417 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
418 return NULL;
419
420 hash = nfs_fattr_to_ino_t(fattr);
421 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
422
423 dprintk("%s: returning %p\n", __func__, inode);
424 return inode;
425 }
426
nfs_inode_init_regular(struct nfs_inode * nfsi)427 static void nfs_inode_init_regular(struct nfs_inode *nfsi)
428 {
429 atomic_long_set(&nfsi->nrequests, 0);
430 atomic_long_set(&nfsi->redirtied_pages, 0);
431 INIT_LIST_HEAD(&nfsi->commit_info.list);
432 atomic_long_set(&nfsi->commit_info.ncommit, 0);
433 atomic_set(&nfsi->commit_info.rpcs_out, 0);
434 mutex_init(&nfsi->commit_mutex);
435 }
436
nfs_inode_init_dir(struct nfs_inode * nfsi)437 static void nfs_inode_init_dir(struct nfs_inode *nfsi)
438 {
439 nfsi->cache_change_attribute = 0;
440 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
441 init_rwsem(&nfsi->rmdir_sem);
442 }
443
444 /*
445 * This is our front-end to iget that looks up inodes by file handle
446 * instead of inode number.
447 */
448 struct inode *
nfs_fhget(struct super_block * sb,struct nfs_fh * fh,struct nfs_fattr * fattr)449 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
450 {
451 struct nfs_find_desc desc = {
452 .fh = fh,
453 .fattr = fattr
454 };
455 struct inode *inode = ERR_PTR(-ENOENT);
456 u64 fattr_supported = NFS_SB(sb)->fattr_valid;
457 unsigned long hash;
458
459 nfs_attr_check_mountpoint(sb, fattr);
460
461 if (nfs_attr_use_mounted_on_fileid(fattr))
462 fattr->fileid = fattr->mounted_on_fileid;
463 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
464 goto out_no_inode;
465 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
466 goto out_no_inode;
467
468 hash = nfs_fattr_to_ino_t(fattr);
469
470 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
471 if (inode == NULL) {
472 inode = ERR_PTR(-ENOMEM);
473 goto out_no_inode;
474 }
475
476 if (inode->i_state & I_NEW) {
477 struct nfs_inode *nfsi = NFS_I(inode);
478 unsigned long now = jiffies;
479
480 /* We set i_ino for the few things that still rely on it,
481 * such as stat(2) */
482 inode->i_ino = hash;
483
484 /* We can't support update_atime(), since the server will reset it */
485 inode->i_flags |= S_NOATIME|S_NOCMTIME;
486 inode->i_mode = fattr->mode;
487 nfsi->cache_validity = 0;
488 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
489 && (fattr_supported & NFS_ATTR_FATTR_MODE))
490 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
491 /* Why so? Because we want revalidate for devices/FIFOs, and
492 * that's precisely what we have in nfs_file_inode_operations.
493 */
494 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
495 if (S_ISREG(inode->i_mode)) {
496 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
497 inode->i_data.a_ops = &nfs_file_aops;
498 nfs_inode_init_regular(nfsi);
499 } else if (S_ISDIR(inode->i_mode)) {
500 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
501 inode->i_fop = &nfs_dir_operations;
502 inode->i_data.a_ops = &nfs_dir_aops;
503 nfs_inode_init_dir(nfsi);
504 /* Deal with crossing mountpoints */
505 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
506 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
507 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
508 inode->i_op = &nfs_referral_inode_operations;
509 else
510 inode->i_op = &nfs_mountpoint_inode_operations;
511 inode->i_fop = NULL;
512 inode->i_flags |= S_AUTOMOUNT;
513 }
514 } else if (S_ISLNK(inode->i_mode)) {
515 inode->i_op = &nfs_symlink_inode_operations;
516 inode_nohighmem(inode);
517 } else
518 init_special_inode(inode, inode->i_mode, fattr->rdev);
519
520 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
521 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
522 inode_set_ctime(inode, 0, 0);
523 inode_set_iversion_raw(inode, 0);
524 inode->i_size = 0;
525 clear_nlink(inode);
526 inode->i_uid = make_kuid(&init_user_ns, -2);
527 inode->i_gid = make_kgid(&init_user_ns, -2);
528 inode->i_blocks = 0;
529 nfsi->write_io = 0;
530 nfsi->read_io = 0;
531
532 nfsi->read_cache_jiffies = fattr->time_start;
533 nfsi->attr_gencount = fattr->gencount;
534 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
535 inode->i_atime = fattr->atime;
536 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
537 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
538 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
539 inode->i_mtime = fattr->mtime;
540 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
541 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
542 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
543 inode_set_ctime_to_ts(inode, fattr->ctime);
544 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
545 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
546 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
547 inode_set_iversion_raw(inode, fattr->change_attr);
548 else
549 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
550 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
551 inode->i_size = nfs_size_to_loff_t(fattr->size);
552 else
553 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
554 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
555 set_nlink(inode, fattr->nlink);
556 else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
557 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
558 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
559 inode->i_uid = fattr->uid;
560 else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
561 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
562 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
563 inode->i_gid = fattr->gid;
564 else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
565 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
566 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
567 inode->i_blocks = fattr->du.nfs2.blocks;
568 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
569 fattr->size != 0)
570 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
571 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
572 /*
573 * report the blocks in 512byte units
574 */
575 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
576 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
577 fattr->size != 0)
578 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
579
580 nfs_setsecurity(inode, fattr);
581
582 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
583 nfsi->attrtimeo_timestamp = now;
584 nfsi->access_cache = RB_ROOT;
585
586 nfs_fscache_init_inode(inode);
587
588 unlock_new_inode(inode);
589 } else {
590 int err = nfs_refresh_inode(inode, fattr);
591 if (err < 0) {
592 iput(inode);
593 inode = ERR_PTR(err);
594 goto out_no_inode;
595 }
596 }
597 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
598 inode->i_sb->s_id,
599 (unsigned long long)NFS_FILEID(inode),
600 nfs_display_fhandle_hash(fh),
601 atomic_read(&inode->i_count));
602
603 out:
604 return inode;
605
606 out_no_inode:
607 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
608 goto out;
609 }
610 EXPORT_SYMBOL_GPL(nfs_fhget);
611
612 #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)
613
614 int
nfs_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)615 nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
616 struct iattr *attr)
617 {
618 struct inode *inode = d_inode(dentry);
619 struct nfs_fattr *fattr;
620 int error = 0;
621
622 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
623
624 /* skip mode change if it's just for clearing setuid/setgid */
625 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
626 attr->ia_valid &= ~ATTR_MODE;
627
628 if (attr->ia_valid & ATTR_SIZE) {
629 BUG_ON(!S_ISREG(inode->i_mode));
630
631 error = inode_newsize_ok(inode, attr->ia_size);
632 if (error)
633 return error;
634
635 if (attr->ia_size == i_size_read(inode))
636 attr->ia_valid &= ~ATTR_SIZE;
637 }
638
639 /* Optimization: if the end result is no change, don't RPC */
640 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
641 return 0;
642
643 trace_nfs_setattr_enter(inode);
644
645 /* Write all dirty data */
646 if (S_ISREG(inode->i_mode))
647 nfs_sync_inode(inode);
648
649 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
650 if (fattr == NULL) {
651 error = -ENOMEM;
652 goto out;
653 }
654
655 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
656 if (error == 0)
657 error = nfs_refresh_inode(inode, fattr);
658 nfs_free_fattr(fattr);
659 out:
660 trace_nfs_setattr_exit(inode, error);
661 return error;
662 }
663 EXPORT_SYMBOL_GPL(nfs_setattr);
664
665 /**
666 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
667 * @inode: inode of the file used
668 * @offset: file offset to start truncating
669 *
670 * This is a copy of the common vmtruncate, but with the locking
671 * corrected to take into account the fact that NFS requires
672 * inode->i_size to be updated under the inode->i_lock.
673 * Note: must be called with inode->i_lock held!
674 */
nfs_vmtruncate(struct inode * inode,loff_t offset)675 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
676 {
677 int err;
678
679 err = inode_newsize_ok(inode, offset);
680 if (err)
681 goto out;
682
683 trace_nfs_size_truncate(inode, offset);
684 i_size_write(inode, offset);
685 /* Optimisation */
686 if (offset == 0) {
687 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
688 nfs_ooo_clear(NFS_I(inode));
689 }
690 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
691
692 spin_unlock(&inode->i_lock);
693 truncate_pagecache(inode, offset);
694 spin_lock(&inode->i_lock);
695 out:
696 return err;
697 }
698
699 /**
700 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
701 * @inode: pointer to struct inode
702 * @attr: pointer to struct iattr
703 * @fattr: pointer to struct nfs_fattr
704 *
705 * Note: we do this in the *proc.c in order to ensure that
706 * it works for things like exclusive creates too.
707 */
nfs_setattr_update_inode(struct inode * inode,struct iattr * attr,struct nfs_fattr * fattr)708 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
709 struct nfs_fattr *fattr)
710 {
711 /* Barrier: bump the attribute generation count. */
712 nfs_fattr_set_barrier(fattr);
713
714 spin_lock(&inode->i_lock);
715 NFS_I(inode)->attr_gencount = fattr->gencount;
716 if ((attr->ia_valid & ATTR_SIZE) != 0) {
717 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME |
718 NFS_INO_INVALID_BLOCKS);
719 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
720 nfs_vmtruncate(inode, attr->ia_size);
721 }
722 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
723 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
724 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
725 inode->i_mode & S_ISUID)
726 inode->i_mode &= ~S_ISUID;
727 if (setattr_should_drop_sgid(&nop_mnt_idmap, inode))
728 inode->i_mode &= ~S_ISGID;
729 if ((attr->ia_valid & ATTR_MODE) != 0) {
730 int mode = attr->ia_mode & S_IALLUGO;
731 mode |= inode->i_mode & ~S_IALLUGO;
732 inode->i_mode = mode;
733 }
734 if ((attr->ia_valid & ATTR_UID) != 0)
735 inode->i_uid = attr->ia_uid;
736 if ((attr->ia_valid & ATTR_GID) != 0)
737 inode->i_gid = attr->ia_gid;
738 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
739 inode_set_ctime_to_ts(inode, fattr->ctime);
740 else
741 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
742 | NFS_INO_INVALID_CTIME);
743 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
744 | NFS_INO_INVALID_ACL);
745 }
746 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
747 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
748 | NFS_INO_INVALID_CTIME);
749 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
750 inode->i_atime = fattr->atime;
751 else if (attr->ia_valid & ATTR_ATIME_SET)
752 inode->i_atime = attr->ia_atime;
753 else
754 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
755
756 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
757 inode_set_ctime_to_ts(inode, fattr->ctime);
758 else
759 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
760 | NFS_INO_INVALID_CTIME);
761 }
762 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
763 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
764 | NFS_INO_INVALID_CTIME);
765 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
766 inode->i_mtime = fattr->mtime;
767 else if (attr->ia_valid & ATTR_MTIME_SET)
768 inode->i_mtime = attr->ia_mtime;
769 else
770 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
771
772 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
773 inode_set_ctime_to_ts(inode, fattr->ctime);
774 else
775 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
776 | NFS_INO_INVALID_CTIME);
777 }
778 if (fattr->valid)
779 nfs_update_inode(inode, fattr);
780 spin_unlock(&inode->i_lock);
781 }
782 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
783
784 /*
785 * Don't request help from readdirplus if the file is being written to,
786 * or if attribute caching is turned off
787 */
nfs_getattr_readdirplus_enable(const struct inode * inode)788 static bool nfs_getattr_readdirplus_enable(const struct inode *inode)
789 {
790 return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) &&
791 !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ;
792 }
793
nfs_readdirplus_parent_cache_miss(struct dentry * dentry)794 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
795 {
796 if (!IS_ROOT(dentry)) {
797 struct dentry *parent = dget_parent(dentry);
798 nfs_readdir_record_entry_cache_miss(d_inode(parent));
799 dput(parent);
800 }
801 }
802
nfs_readdirplus_parent_cache_hit(struct dentry * dentry)803 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
804 {
805 if (!IS_ROOT(dentry)) {
806 struct dentry *parent = dget_parent(dentry);
807 nfs_readdir_record_entry_cache_hit(d_inode(parent));
808 dput(parent);
809 }
810 }
811
nfs_get_valid_attrmask(struct inode * inode)812 static u32 nfs_get_valid_attrmask(struct inode *inode)
813 {
814 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
815 u32 reply_mask = STATX_INO | STATX_TYPE;
816
817 if (!(cache_validity & NFS_INO_INVALID_ATIME))
818 reply_mask |= STATX_ATIME;
819 if (!(cache_validity & NFS_INO_INVALID_CTIME))
820 reply_mask |= STATX_CTIME;
821 if (!(cache_validity & NFS_INO_INVALID_MTIME))
822 reply_mask |= STATX_MTIME;
823 if (!(cache_validity & NFS_INO_INVALID_SIZE))
824 reply_mask |= STATX_SIZE;
825 if (!(cache_validity & NFS_INO_INVALID_NLINK))
826 reply_mask |= STATX_NLINK;
827 if (!(cache_validity & NFS_INO_INVALID_MODE))
828 reply_mask |= STATX_MODE;
829 if (!(cache_validity & NFS_INO_INVALID_OTHER))
830 reply_mask |= STATX_UID | STATX_GID;
831 if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
832 reply_mask |= STATX_BLOCKS;
833 if (!(cache_validity & NFS_INO_INVALID_CHANGE))
834 reply_mask |= STATX_CHANGE_COOKIE;
835 return reply_mask;
836 }
837
nfs_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)838 int nfs_getattr(struct mnt_idmap *idmap, const struct path *path,
839 struct kstat *stat, u32 request_mask, unsigned int query_flags)
840 {
841 struct inode *inode = d_inode(path->dentry);
842 struct nfs_server *server = NFS_SERVER(inode);
843 unsigned long cache_validity;
844 int err = 0;
845 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
846 bool do_update = false;
847 bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
848
849 trace_nfs_getattr_enter(inode);
850
851 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
852 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
853 STATX_INO | STATX_SIZE | STATX_BLOCKS |
854 STATX_CHANGE_COOKIE;
855
856 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
857 if (readdirplus_enabled)
858 nfs_readdirplus_parent_cache_hit(path->dentry);
859 goto out_no_revalidate;
860 }
861
862 /* Flush out writes to the server in order to update c/mtime/version. */
863 if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) &&
864 S_ISREG(inode->i_mode))
865 filemap_write_and_wait(inode->i_mapping);
866
867 /*
868 * We may force a getattr if the user cares about atime.
869 *
870 * Note that we only have to check the vfsmount flags here:
871 * - NFS always sets S_NOATIME by so checking it would give a
872 * bogus result
873 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
874 * no point in checking those.
875 */
876 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
877 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
878 request_mask &= ~STATX_ATIME;
879
880 /* Is the user requesting attributes that might need revalidation? */
881 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
882 STATX_MTIME|STATX_UID|STATX_GID|
883 STATX_SIZE|STATX_BLOCKS|
884 STATX_CHANGE_COOKIE)))
885 goto out_no_revalidate;
886
887 /* Check whether the cached attributes are stale */
888 do_update |= force_sync || nfs_attribute_cache_expired(inode);
889 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
890 do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
891 if (request_mask & STATX_ATIME)
892 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
893 if (request_mask & STATX_CTIME)
894 do_update |= cache_validity & NFS_INO_INVALID_CTIME;
895 if (request_mask & STATX_MTIME)
896 do_update |= cache_validity & NFS_INO_INVALID_MTIME;
897 if (request_mask & STATX_SIZE)
898 do_update |= cache_validity & NFS_INO_INVALID_SIZE;
899 if (request_mask & STATX_NLINK)
900 do_update |= cache_validity & NFS_INO_INVALID_NLINK;
901 if (request_mask & STATX_MODE)
902 do_update |= cache_validity & NFS_INO_INVALID_MODE;
903 if (request_mask & (STATX_UID | STATX_GID))
904 do_update |= cache_validity & NFS_INO_INVALID_OTHER;
905 if (request_mask & STATX_BLOCKS)
906 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
907
908 if (do_update) {
909 if (readdirplus_enabled)
910 nfs_readdirplus_parent_cache_miss(path->dentry);
911 err = __nfs_revalidate_inode(server, inode);
912 if (err)
913 goto out;
914 } else if (readdirplus_enabled)
915 nfs_readdirplus_parent_cache_hit(path->dentry);
916 out_no_revalidate:
917 /* Only return attributes that were revalidated. */
918 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
919
920 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
921 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
922 stat->change_cookie = inode_peek_iversion_raw(inode);
923 stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC;
924 if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED)
925 stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
926 if (S_ISDIR(inode->i_mode))
927 stat->blksize = NFS_SERVER(inode)->dtsize;
928 out:
929 trace_nfs_getattr_exit(inode, err);
930 return err;
931 }
932 EXPORT_SYMBOL_GPL(nfs_getattr);
933
nfs_init_lock_context(struct nfs_lock_context * l_ctx)934 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
935 {
936 refcount_set(&l_ctx->count, 1);
937 l_ctx->lockowner = current->files;
938 INIT_LIST_HEAD(&l_ctx->list);
939 atomic_set(&l_ctx->io_count, 0);
940 }
941
__nfs_find_lock_context(struct nfs_open_context * ctx)942 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
943 {
944 struct nfs_lock_context *pos;
945
946 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
947 if (pos->lockowner != current->files)
948 continue;
949 if (refcount_inc_not_zero(&pos->count))
950 return pos;
951 }
952 return NULL;
953 }
954
nfs_get_lock_context(struct nfs_open_context * ctx)955 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
956 {
957 struct nfs_lock_context *res, *new = NULL;
958 struct inode *inode = d_inode(ctx->dentry);
959
960 rcu_read_lock();
961 res = __nfs_find_lock_context(ctx);
962 rcu_read_unlock();
963 if (res == NULL) {
964 new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
965 if (new == NULL)
966 return ERR_PTR(-ENOMEM);
967 nfs_init_lock_context(new);
968 spin_lock(&inode->i_lock);
969 res = __nfs_find_lock_context(ctx);
970 if (res == NULL) {
971 new->open_context = get_nfs_open_context(ctx);
972 if (new->open_context) {
973 list_add_tail_rcu(&new->list,
974 &ctx->lock_context.list);
975 res = new;
976 new = NULL;
977 } else
978 res = ERR_PTR(-EBADF);
979 }
980 spin_unlock(&inode->i_lock);
981 kfree(new);
982 }
983 return res;
984 }
985 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
986
nfs_put_lock_context(struct nfs_lock_context * l_ctx)987 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
988 {
989 struct nfs_open_context *ctx = l_ctx->open_context;
990 struct inode *inode = d_inode(ctx->dentry);
991
992 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
993 return;
994 list_del_rcu(&l_ctx->list);
995 spin_unlock(&inode->i_lock);
996 put_nfs_open_context(ctx);
997 kfree_rcu(l_ctx, rcu_head);
998 }
999 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
1000
1001 /**
1002 * nfs_close_context - Common close_context() routine NFSv2/v3
1003 * @ctx: pointer to context
1004 * @is_sync: is this a synchronous close
1005 *
1006 * Ensure that the attributes are up to date if we're mounted
1007 * with close-to-open semantics and we have cached data that will
1008 * need to be revalidated on open.
1009 */
nfs_close_context(struct nfs_open_context * ctx,int is_sync)1010 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1011 {
1012 struct nfs_inode *nfsi;
1013 struct inode *inode;
1014
1015 if (!(ctx->mode & FMODE_WRITE))
1016 return;
1017 if (!is_sync)
1018 return;
1019 inode = d_inode(ctx->dentry);
1020 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1021 return;
1022 nfsi = NFS_I(inode);
1023 if (inode->i_mapping->nrpages == 0)
1024 return;
1025 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1026 return;
1027 if (!list_empty(&nfsi->open_files))
1028 return;
1029 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1030 return;
1031 nfs_revalidate_inode(inode,
1032 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1033 }
1034 EXPORT_SYMBOL_GPL(nfs_close_context);
1035
alloc_nfs_open_context(struct dentry * dentry,fmode_t f_mode,struct file * filp)1036 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1037 fmode_t f_mode,
1038 struct file *filp)
1039 {
1040 struct nfs_open_context *ctx;
1041
1042 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1043 if (!ctx)
1044 return ERR_PTR(-ENOMEM);
1045 nfs_sb_active(dentry->d_sb);
1046 ctx->dentry = dget(dentry);
1047 if (filp)
1048 ctx->cred = get_cred(filp->f_cred);
1049 else
1050 ctx->cred = get_current_cred();
1051 rcu_assign_pointer(ctx->ll_cred, NULL);
1052 ctx->state = NULL;
1053 ctx->mode = f_mode;
1054 ctx->flags = 0;
1055 ctx->error = 0;
1056 ctx->flock_owner = (fl_owner_t)filp;
1057 nfs_init_lock_context(&ctx->lock_context);
1058 ctx->lock_context.open_context = ctx;
1059 INIT_LIST_HEAD(&ctx->list);
1060 ctx->mdsthreshold = NULL;
1061 return ctx;
1062 }
1063 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1064
get_nfs_open_context(struct nfs_open_context * ctx)1065 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1066 {
1067 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1068 return ctx;
1069 return NULL;
1070 }
1071 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1072
__put_nfs_open_context(struct nfs_open_context * ctx,int is_sync)1073 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1074 {
1075 struct inode *inode = d_inode(ctx->dentry);
1076 struct super_block *sb = ctx->dentry->d_sb;
1077
1078 if (!refcount_dec_and_test(&ctx->lock_context.count))
1079 return;
1080 if (!list_empty(&ctx->list)) {
1081 spin_lock(&inode->i_lock);
1082 list_del_rcu(&ctx->list);
1083 spin_unlock(&inode->i_lock);
1084 }
1085 if (inode != NULL)
1086 NFS_PROTO(inode)->close_context(ctx, is_sync);
1087 put_cred(ctx->cred);
1088 dput(ctx->dentry);
1089 nfs_sb_deactive(sb);
1090 put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1091 kfree(ctx->mdsthreshold);
1092 kfree_rcu(ctx, rcu_head);
1093 }
1094
put_nfs_open_context(struct nfs_open_context * ctx)1095 void put_nfs_open_context(struct nfs_open_context *ctx)
1096 {
1097 __put_nfs_open_context(ctx, 0);
1098 }
1099 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1100
put_nfs_open_context_sync(struct nfs_open_context * ctx)1101 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1102 {
1103 __put_nfs_open_context(ctx, 1);
1104 }
1105
1106 /*
1107 * Ensure that mmap has a recent RPC credential for use when writing out
1108 * shared pages
1109 */
nfs_inode_attach_open_context(struct nfs_open_context * ctx)1110 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1111 {
1112 struct inode *inode = d_inode(ctx->dentry);
1113 struct nfs_inode *nfsi = NFS_I(inode);
1114
1115 spin_lock(&inode->i_lock);
1116 if (list_empty(&nfsi->open_files) &&
1117 nfs_ooo_test(nfsi))
1118 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1119 NFS_INO_REVAL_FORCED);
1120 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1121 spin_unlock(&inode->i_lock);
1122 }
1123 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1124
nfs_file_set_open_context(struct file * filp,struct nfs_open_context * ctx)1125 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1126 {
1127 filp->private_data = get_nfs_open_context(ctx);
1128 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1129 if (list_empty(&ctx->list))
1130 nfs_inode_attach_open_context(ctx);
1131 }
1132 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1133
1134 /*
1135 * Given an inode, search for an open context with the desired characteristics
1136 */
nfs_find_open_context(struct inode * inode,const struct cred * cred,fmode_t mode)1137 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1138 {
1139 struct nfs_inode *nfsi = NFS_I(inode);
1140 struct nfs_open_context *pos, *ctx = NULL;
1141
1142 rcu_read_lock();
1143 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1144 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1145 continue;
1146 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1147 continue;
1148 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1149 continue;
1150 ctx = get_nfs_open_context(pos);
1151 if (ctx)
1152 break;
1153 }
1154 rcu_read_unlock();
1155 return ctx;
1156 }
1157
nfs_file_clear_open_context(struct file * filp)1158 void nfs_file_clear_open_context(struct file *filp)
1159 {
1160 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1161
1162 if (ctx) {
1163 struct inode *inode = d_inode(ctx->dentry);
1164
1165 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1166 /*
1167 * We fatal error on write before. Try to writeback
1168 * every page again.
1169 */
1170 if (ctx->error < 0)
1171 invalidate_inode_pages2(inode->i_mapping);
1172 filp->private_data = NULL;
1173 put_nfs_open_context_sync(ctx);
1174 }
1175 }
1176
1177 /*
1178 * These allocate and release file read/write context information.
1179 */
nfs_open(struct inode * inode,struct file * filp)1180 int nfs_open(struct inode *inode, struct file *filp)
1181 {
1182 struct nfs_open_context *ctx;
1183
1184 ctx = alloc_nfs_open_context(file_dentry(filp),
1185 flags_to_mode(filp->f_flags), filp);
1186 if (IS_ERR(ctx))
1187 return PTR_ERR(ctx);
1188 nfs_file_set_open_context(filp, ctx);
1189 put_nfs_open_context(ctx);
1190 nfs_fscache_open_file(inode, filp);
1191 return 0;
1192 }
1193
1194 /*
1195 * This function is called whenever some part of NFS notices that
1196 * the cached attributes have to be refreshed.
1197 */
1198 int
__nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1199 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1200 {
1201 int status = -ESTALE;
1202 struct nfs_fattr *fattr = NULL;
1203 struct nfs_inode *nfsi = NFS_I(inode);
1204
1205 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1206 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1207
1208 trace_nfs_revalidate_inode_enter(inode);
1209
1210 if (is_bad_inode(inode))
1211 goto out;
1212 if (NFS_STALE(inode))
1213 goto out;
1214
1215 /* pNFS: Attributes aren't updated until we layoutcommit */
1216 if (S_ISREG(inode->i_mode)) {
1217 status = pnfs_sync_inode(inode, false);
1218 if (status)
1219 goto out;
1220 }
1221
1222 status = -ENOMEM;
1223 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1224 if (fattr == NULL)
1225 goto out;
1226
1227 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1228
1229 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1230 if (status != 0) {
1231 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1232 inode->i_sb->s_id,
1233 (unsigned long long)NFS_FILEID(inode), status);
1234 switch (status) {
1235 case -ETIMEDOUT:
1236 /* A soft timeout occurred. Use cached information? */
1237 if (server->flags & NFS_MOUNT_SOFTREVAL)
1238 status = 0;
1239 break;
1240 case -ESTALE:
1241 if (!S_ISDIR(inode->i_mode))
1242 nfs_set_inode_stale(inode);
1243 else
1244 nfs_zap_caches(inode);
1245 }
1246 goto out;
1247 }
1248
1249 status = nfs_refresh_inode(inode, fattr);
1250 if (status) {
1251 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1252 inode->i_sb->s_id,
1253 (unsigned long long)NFS_FILEID(inode), status);
1254 goto out;
1255 }
1256
1257 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1258 nfs_zap_acl_cache(inode);
1259
1260 nfs_setsecurity(inode, fattr);
1261
1262 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1263 inode->i_sb->s_id,
1264 (unsigned long long)NFS_FILEID(inode));
1265
1266 out:
1267 nfs_free_fattr(fattr);
1268 trace_nfs_revalidate_inode_exit(inode, status);
1269 return status;
1270 }
1271
nfs_attribute_cache_expired(struct inode * inode)1272 int nfs_attribute_cache_expired(struct inode *inode)
1273 {
1274 if (nfs_have_delegated_attributes(inode))
1275 return 0;
1276 return nfs_attribute_timeout(inode);
1277 }
1278
1279 /**
1280 * nfs_revalidate_inode - Revalidate the inode attributes
1281 * @inode: pointer to inode struct
1282 * @flags: cache flags to check
1283 *
1284 * Updates inode attribute information by retrieving the data from the server.
1285 */
nfs_revalidate_inode(struct inode * inode,unsigned long flags)1286 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1287 {
1288 if (!nfs_check_cache_invalid(inode, flags))
1289 return NFS_STALE(inode) ? -ESTALE : 0;
1290 return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1291 }
1292 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1293
nfs_invalidate_mapping(struct inode * inode,struct address_space * mapping)1294 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1295 {
1296 int ret;
1297
1298 nfs_fscache_invalidate(inode, 0);
1299 if (mapping->nrpages != 0) {
1300 if (S_ISREG(inode->i_mode)) {
1301 ret = nfs_sync_mapping(mapping);
1302 if (ret < 0)
1303 return ret;
1304 }
1305 ret = invalidate_inode_pages2(mapping);
1306 if (ret < 0)
1307 return ret;
1308 }
1309 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1310
1311 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1312 inode->i_sb->s_id,
1313 (unsigned long long)NFS_FILEID(inode));
1314 return 0;
1315 }
1316
1317 /**
1318 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1319 * @mapping: pointer to mapping
1320 *
1321 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1322 */
nfs_clear_invalid_mapping(struct address_space * mapping)1323 int nfs_clear_invalid_mapping(struct address_space *mapping)
1324 {
1325 struct inode *inode = mapping->host;
1326 struct nfs_inode *nfsi = NFS_I(inode);
1327 unsigned long *bitlock = &nfsi->flags;
1328 int ret = 0;
1329
1330 /*
1331 * We must clear NFS_INO_INVALID_DATA first to ensure that
1332 * invalidations that come in while we're shooting down the mappings
1333 * are respected. But, that leaves a race window where one revalidator
1334 * can clear the flag, and then another checks it before the mapping
1335 * gets invalidated. Fix that by serializing access to this part of
1336 * the function.
1337 *
1338 * At the same time, we need to allow other tasks to see whether we
1339 * might be in the middle of invalidating the pages, so we only set
1340 * the bit lock here if it looks like we're going to be doing that.
1341 */
1342 for (;;) {
1343 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1344 nfs_wait_bit_killable,
1345 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1346 if (ret)
1347 goto out;
1348 smp_rmb(); /* pairs with smp_wmb() below */
1349 if (test_bit(NFS_INO_INVALIDATING, bitlock))
1350 continue;
1351 /* pairs with nfs_set_cache_invalid()'s smp_store_release() */
1352 if (!(smp_load_acquire(&nfsi->cache_validity) & NFS_INO_INVALID_DATA))
1353 goto out;
1354 /* Slow-path that double-checks with spinlock held */
1355 spin_lock(&inode->i_lock);
1356 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1357 spin_unlock(&inode->i_lock);
1358 continue;
1359 }
1360 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1361 break;
1362 spin_unlock(&inode->i_lock);
1363 goto out;
1364 }
1365
1366 set_bit(NFS_INO_INVALIDATING, bitlock);
1367 smp_wmb();
1368 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1369 nfs_ooo_clear(nfsi);
1370 spin_unlock(&inode->i_lock);
1371 trace_nfs_invalidate_mapping_enter(inode);
1372 ret = nfs_invalidate_mapping(inode, mapping);
1373 trace_nfs_invalidate_mapping_exit(inode, ret);
1374
1375 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1376 smp_mb__after_atomic();
1377 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1378 out:
1379 return ret;
1380 }
1381
nfs_mapping_need_revalidate_inode(struct inode * inode)1382 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1383 {
1384 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1385 NFS_STALE(inode);
1386 }
1387
nfs_revalidate_mapping_rcu(struct inode * inode)1388 int nfs_revalidate_mapping_rcu(struct inode *inode)
1389 {
1390 struct nfs_inode *nfsi = NFS_I(inode);
1391 unsigned long *bitlock = &nfsi->flags;
1392 int ret = 0;
1393
1394 if (IS_SWAPFILE(inode))
1395 goto out;
1396 if (nfs_mapping_need_revalidate_inode(inode)) {
1397 ret = -ECHILD;
1398 goto out;
1399 }
1400 spin_lock(&inode->i_lock);
1401 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1402 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1403 ret = -ECHILD;
1404 spin_unlock(&inode->i_lock);
1405 out:
1406 return ret;
1407 }
1408
1409 /**
1410 * nfs_revalidate_mapping - Revalidate the pagecache
1411 * @inode: pointer to host inode
1412 * @mapping: pointer to mapping
1413 */
nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping)1414 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1415 {
1416 /* swapfiles are not supposed to be shared. */
1417 if (IS_SWAPFILE(inode))
1418 return 0;
1419
1420 if (nfs_mapping_need_revalidate_inode(inode)) {
1421 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1422 if (ret < 0)
1423 return ret;
1424 }
1425
1426 return nfs_clear_invalid_mapping(mapping);
1427 }
1428
nfs_file_has_writers(struct nfs_inode * nfsi)1429 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1430 {
1431 struct inode *inode = &nfsi->vfs_inode;
1432
1433 if (!S_ISREG(inode->i_mode))
1434 return false;
1435 if (list_empty(&nfsi->open_files))
1436 return false;
1437 return inode_is_open_for_write(inode);
1438 }
1439
nfs_file_has_buffered_writers(struct nfs_inode * nfsi)1440 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1441 {
1442 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1443 }
1444
nfs_wcc_update_inode(struct inode * inode,struct nfs_fattr * fattr)1445 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1446 {
1447 struct timespec64 ts;
1448
1449 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1450 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1451 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1452 inode_set_iversion_raw(inode, fattr->change_attr);
1453 if (S_ISDIR(inode->i_mode))
1454 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1455 else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1456 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1457 }
1458 /* If we have atomic WCC data, we may update some attributes */
1459 ts = inode_get_ctime(inode);
1460 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1461 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1462 && timespec64_equal(&ts, &fattr->pre_ctime)) {
1463 inode_set_ctime_to_ts(inode, fattr->ctime);
1464 }
1465
1466 ts = inode->i_mtime;
1467 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1468 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1469 && timespec64_equal(&ts, &fattr->pre_mtime)) {
1470 inode->i_mtime = fattr->mtime;
1471 }
1472 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1473 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1474 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1475 && !nfs_have_writebacks(inode)) {
1476 trace_nfs_size_wcc(inode, fattr->size);
1477 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1478 }
1479 }
1480
1481 /**
1482 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1483 * @inode: pointer to inode
1484 * @fattr: updated attributes
1485 *
1486 * Verifies the attribute cache. If we have just changed the attributes,
1487 * so that fattr carries weak cache consistency data, then it may
1488 * also update the ctime/mtime/change_attribute.
1489 */
nfs_check_inode_attributes(struct inode * inode,struct nfs_fattr * fattr)1490 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1491 {
1492 struct nfs_inode *nfsi = NFS_I(inode);
1493 loff_t cur_size, new_isize;
1494 unsigned long invalid = 0;
1495 struct timespec64 ts;
1496
1497 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1498 return 0;
1499
1500 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1501 /* Only a mounted-on-fileid? Just exit */
1502 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1503 return 0;
1504 /* Has the inode gone and changed behind our back? */
1505 } else if (nfsi->fileid != fattr->fileid) {
1506 /* Is this perhaps the mounted-on fileid? */
1507 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1508 nfsi->fileid == fattr->mounted_on_fileid)
1509 return 0;
1510 return -ESTALE;
1511 }
1512 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1513 return -ESTALE;
1514
1515
1516 if (!nfs_file_has_buffered_writers(nfsi)) {
1517 /* Verify a few of the more important attributes */
1518 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1519 invalid |= NFS_INO_INVALID_CHANGE;
1520
1521 ts = inode->i_mtime;
1522 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1523 invalid |= NFS_INO_INVALID_MTIME;
1524
1525 ts = inode_get_ctime(inode);
1526 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1527 invalid |= NFS_INO_INVALID_CTIME;
1528
1529 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1530 cur_size = i_size_read(inode);
1531 new_isize = nfs_size_to_loff_t(fattr->size);
1532 if (cur_size != new_isize)
1533 invalid |= NFS_INO_INVALID_SIZE;
1534 }
1535 }
1536
1537 /* Have any file permissions changed? */
1538 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1539 invalid |= NFS_INO_INVALID_MODE;
1540 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1541 invalid |= NFS_INO_INVALID_OTHER;
1542 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1543 invalid |= NFS_INO_INVALID_OTHER;
1544
1545 /* Has the link count changed? */
1546 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1547 invalid |= NFS_INO_INVALID_NLINK;
1548
1549 ts = inode->i_atime;
1550 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1551 invalid |= NFS_INO_INVALID_ATIME;
1552
1553 if (invalid != 0)
1554 nfs_set_cache_invalid(inode, invalid);
1555
1556 nfsi->read_cache_jiffies = fattr->time_start;
1557 return 0;
1558 }
1559
1560 static atomic_long_t nfs_attr_generation_counter;
1561
nfs_read_attr_generation_counter(void)1562 static unsigned long nfs_read_attr_generation_counter(void)
1563 {
1564 return atomic_long_read(&nfs_attr_generation_counter);
1565 }
1566
nfs_inc_attr_generation_counter(void)1567 unsigned long nfs_inc_attr_generation_counter(void)
1568 {
1569 return atomic_long_inc_return(&nfs_attr_generation_counter);
1570 }
1571 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1572
nfs_fattr_init(struct nfs_fattr * fattr)1573 void nfs_fattr_init(struct nfs_fattr *fattr)
1574 {
1575 fattr->valid = 0;
1576 fattr->time_start = jiffies;
1577 fattr->gencount = nfs_inc_attr_generation_counter();
1578 fattr->owner_name = NULL;
1579 fattr->group_name = NULL;
1580 fattr->mdsthreshold = NULL;
1581 }
1582 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1583
1584 /**
1585 * nfs_fattr_set_barrier
1586 * @fattr: attributes
1587 *
1588 * Used to set a barrier after an attribute was updated. This
1589 * barrier ensures that older attributes from RPC calls that may
1590 * have raced with our update cannot clobber these new values.
1591 * Note that you are still responsible for ensuring that other
1592 * operations which change the attribute on the server do not
1593 * collide.
1594 */
nfs_fattr_set_barrier(struct nfs_fattr * fattr)1595 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1596 {
1597 fattr->gencount = nfs_inc_attr_generation_counter();
1598 }
1599
nfs_alloc_fattr(void)1600 struct nfs_fattr *nfs_alloc_fattr(void)
1601 {
1602 struct nfs_fattr *fattr;
1603
1604 fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1605 if (fattr != NULL) {
1606 nfs_fattr_init(fattr);
1607 fattr->label = NULL;
1608 }
1609 return fattr;
1610 }
1611 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1612
nfs_alloc_fattr_with_label(struct nfs_server * server)1613 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1614 {
1615 struct nfs_fattr *fattr = nfs_alloc_fattr();
1616
1617 if (!fattr)
1618 return NULL;
1619
1620 fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1621 if (IS_ERR(fattr->label)) {
1622 kfree(fattr);
1623 return NULL;
1624 }
1625
1626 return fattr;
1627 }
1628 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1629
nfs_alloc_fhandle(void)1630 struct nfs_fh *nfs_alloc_fhandle(void)
1631 {
1632 struct nfs_fh *fh;
1633
1634 fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1635 if (fh != NULL)
1636 fh->size = 0;
1637 return fh;
1638 }
1639 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1640
1641 #ifdef NFS_DEBUG
1642 /*
1643 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1644 * in the same way that wireshark does
1645 *
1646 * @fh: file handle
1647 *
1648 * For debugging only.
1649 */
_nfs_display_fhandle_hash(const struct nfs_fh * fh)1650 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1651 {
1652 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1653 * not on the result */
1654 return nfs_fhandle_hash(fh);
1655 }
1656 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1657
1658 /*
1659 * _nfs_display_fhandle - display an NFS file handle on the console
1660 *
1661 * @fh: file handle to display
1662 * @caption: display caption
1663 *
1664 * For debugging only.
1665 */
_nfs_display_fhandle(const struct nfs_fh * fh,const char * caption)1666 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1667 {
1668 unsigned short i;
1669
1670 if (fh == NULL || fh->size == 0) {
1671 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1672 return;
1673 }
1674
1675 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1676 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1677 for (i = 0; i < fh->size; i += 16) {
1678 __be32 *pos = (__be32 *)&fh->data[i];
1679
1680 switch ((fh->size - i - 1) >> 2) {
1681 case 0:
1682 printk(KERN_DEFAULT " %08x\n",
1683 be32_to_cpup(pos));
1684 break;
1685 case 1:
1686 printk(KERN_DEFAULT " %08x %08x\n",
1687 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1688 break;
1689 case 2:
1690 printk(KERN_DEFAULT " %08x %08x %08x\n",
1691 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1692 be32_to_cpup(pos + 2));
1693 break;
1694 default:
1695 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1696 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1697 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1698 }
1699 }
1700 }
1701 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1702 #endif
1703
1704 /**
1705 * nfs_inode_attrs_cmp_generic - compare attributes
1706 * @fattr: attributes
1707 * @inode: pointer to inode
1708 *
1709 * Attempt to divine whether or not an RPC call reply carrying stale
1710 * attributes got scheduled after another call carrying updated ones.
1711 * Note also the check for wraparound of 'attr_gencount'
1712 *
1713 * The function returns '1' if it thinks the attributes in @fattr are
1714 * more recent than the ones cached in @inode. Otherwise it returns
1715 * the value '0'.
1716 */
nfs_inode_attrs_cmp_generic(const struct nfs_fattr * fattr,const struct inode * inode)1717 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1718 const struct inode *inode)
1719 {
1720 unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1721
1722 return (long)(fattr->gencount - attr_gencount) > 0 ||
1723 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1724 }
1725
1726 /**
1727 * nfs_inode_attrs_cmp_monotonic - compare attributes
1728 * @fattr: attributes
1729 * @inode: pointer to inode
1730 *
1731 * Attempt to divine whether or not an RPC call reply carrying stale
1732 * attributes got scheduled after another call carrying updated ones.
1733 *
1734 * We assume that the server observes monotonic semantics for
1735 * the change attribute, so a larger value means that the attributes in
1736 * @fattr are more recent, in which case the function returns the
1737 * value '1'.
1738 * A return value of '0' indicates no measurable change
1739 * A return value of '-1' means that the attributes in @inode are
1740 * more recent.
1741 */
nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1742 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1743 const struct inode *inode)
1744 {
1745 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1746 if (diff > 0)
1747 return 1;
1748 return diff == 0 ? 0 : -1;
1749 }
1750
1751 /**
1752 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1753 * @fattr: attributes
1754 * @inode: pointer to inode
1755 *
1756 * Attempt to divine whether or not an RPC call reply carrying stale
1757 * attributes got scheduled after another call carrying updated ones.
1758 *
1759 * We assume that the server observes strictly monotonic semantics for
1760 * the change attribute, so a larger value means that the attributes in
1761 * @fattr are more recent, in which case the function returns the
1762 * value '1'.
1763 * A return value of '-1' means that the attributes in @inode are
1764 * more recent or unchanged.
1765 */
nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1766 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1767 const struct inode *inode)
1768 {
1769 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1770 }
1771
1772 /**
1773 * nfs_inode_attrs_cmp - compare attributes
1774 * @fattr: attributes
1775 * @inode: pointer to inode
1776 *
1777 * This function returns '1' if it thinks the attributes in @fattr are
1778 * more recent than the ones cached in @inode. It returns '-1' if
1779 * the attributes in @inode are more recent than the ones in @fattr,
1780 * and it returns 0 if not sure.
1781 */
nfs_inode_attrs_cmp(const struct nfs_fattr * fattr,const struct inode * inode)1782 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1783 const struct inode *inode)
1784 {
1785 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1786 return 1;
1787 switch (NFS_SERVER(inode)->change_attr_type) {
1788 case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1789 break;
1790 case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1791 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1792 break;
1793 return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1794 default:
1795 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1796 break;
1797 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1798 }
1799 return 0;
1800 }
1801
1802 /**
1803 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1804 * @fattr: attributes
1805 * @inode: pointer to inode
1806 *
1807 * Returns '1' if the last attribute update left the inode cached
1808 * attributes in a partially unrevalidated state, and @fattr
1809 * matches the change attribute of that partial update.
1810 * Otherwise returns '0'.
1811 */
nfs_inode_finish_partial_attr_update(const struct nfs_fattr * fattr,const struct inode * inode)1812 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1813 const struct inode *inode)
1814 {
1815 const unsigned long check_valid =
1816 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1817 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1818 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1819 NFS_INO_INVALID_NLINK;
1820 unsigned long cache_validity = NFS_I(inode)->cache_validity;
1821 enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1822
1823 if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1824 !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1825 (cache_validity & check_valid) != 0 &&
1826 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1827 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1828 return 1;
1829 return 0;
1830 }
1831
nfs_ooo_merge(struct nfs_inode * nfsi,u64 start,u64 end)1832 static void nfs_ooo_merge(struct nfs_inode *nfsi,
1833 u64 start, u64 end)
1834 {
1835 int i, cnt;
1836
1837 if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)
1838 /* No point merging anything */
1839 return;
1840
1841 if (!nfsi->ooo) {
1842 nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC);
1843 if (!nfsi->ooo) {
1844 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1845 return;
1846 }
1847 nfsi->ooo->cnt = 0;
1848 }
1849
1850 /* add this range, merging if possible */
1851 cnt = nfsi->ooo->cnt;
1852 for (i = 0; i < cnt; i++) {
1853 if (end == nfsi->ooo->gap[i].start)
1854 end = nfsi->ooo->gap[i].end;
1855 else if (start == nfsi->ooo->gap[i].end)
1856 start = nfsi->ooo->gap[i].start;
1857 else
1858 continue;
1859 /* Remove 'i' from table and loop to insert the new range */
1860 cnt -= 1;
1861 nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt];
1862 i = -1;
1863 }
1864 if (start != end) {
1865 if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) {
1866 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1867 kfree(nfsi->ooo);
1868 nfsi->ooo = NULL;
1869 return;
1870 }
1871 nfsi->ooo->gap[cnt].start = start;
1872 nfsi->ooo->gap[cnt].end = end;
1873 cnt += 1;
1874 }
1875 nfsi->ooo->cnt = cnt;
1876 }
1877
nfs_ooo_record(struct nfs_inode * nfsi,struct nfs_fattr * fattr)1878 static void nfs_ooo_record(struct nfs_inode *nfsi,
1879 struct nfs_fattr *fattr)
1880 {
1881 /* This reply was out-of-order, so record in the
1882 * pre/post change id, possibly cancelling
1883 * gaps created when iversion was jumpped forward.
1884 */
1885 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) &&
1886 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE))
1887 nfs_ooo_merge(nfsi,
1888 fattr->change_attr,
1889 fattr->pre_change_attr);
1890 }
1891
nfs_refresh_inode_locked(struct inode * inode,struct nfs_fattr * fattr)1892 static int nfs_refresh_inode_locked(struct inode *inode,
1893 struct nfs_fattr *fattr)
1894 {
1895 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1896 int ret = 0;
1897
1898 trace_nfs_refresh_inode_enter(inode);
1899
1900 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1901 ret = nfs_update_inode(inode, fattr);
1902 else {
1903 nfs_ooo_record(NFS_I(inode), fattr);
1904
1905 if (attr_cmp == 0)
1906 ret = nfs_check_inode_attributes(inode, fattr);
1907 }
1908
1909 trace_nfs_refresh_inode_exit(inode, ret);
1910 return ret;
1911 }
1912
1913 /**
1914 * nfs_refresh_inode - try to update the inode attribute cache
1915 * @inode: pointer to inode
1916 * @fattr: updated attributes
1917 *
1918 * Check that an RPC call that returned attributes has not overlapped with
1919 * other recent updates of the inode metadata, then decide whether it is
1920 * safe to do a full update of the inode attributes, or whether just to
1921 * call nfs_check_inode_attributes.
1922 */
nfs_refresh_inode(struct inode * inode,struct nfs_fattr * fattr)1923 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1924 {
1925 int status;
1926
1927 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1928 return 0;
1929 spin_lock(&inode->i_lock);
1930 status = nfs_refresh_inode_locked(inode, fattr);
1931 spin_unlock(&inode->i_lock);
1932
1933 return status;
1934 }
1935 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1936
nfs_post_op_update_inode_locked(struct inode * inode,struct nfs_fattr * fattr,unsigned int invalid)1937 static int nfs_post_op_update_inode_locked(struct inode *inode,
1938 struct nfs_fattr *fattr, unsigned int invalid)
1939 {
1940 if (S_ISDIR(inode->i_mode))
1941 invalid |= NFS_INO_INVALID_DATA;
1942 nfs_set_cache_invalid(inode, invalid);
1943 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1944 return 0;
1945 return nfs_refresh_inode_locked(inode, fattr);
1946 }
1947
1948 /**
1949 * nfs_post_op_update_inode - try to update the inode attribute cache
1950 * @inode: pointer to inode
1951 * @fattr: updated attributes
1952 *
1953 * After an operation that has changed the inode metadata, mark the
1954 * attribute cache as being invalid, then try to update it.
1955 *
1956 * NB: if the server didn't return any post op attributes, this
1957 * function will force the retrieval of attributes before the next
1958 * NFS request. Thus it should be used only for operations that
1959 * are expected to change one or more attributes, to avoid
1960 * unnecessary NFS requests and trips through nfs_update_inode().
1961 */
nfs_post_op_update_inode(struct inode * inode,struct nfs_fattr * fattr)1962 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1963 {
1964 int status;
1965
1966 spin_lock(&inode->i_lock);
1967 nfs_fattr_set_barrier(fattr);
1968 status = nfs_post_op_update_inode_locked(inode, fattr,
1969 NFS_INO_INVALID_CHANGE
1970 | NFS_INO_INVALID_CTIME
1971 | NFS_INO_REVAL_FORCED);
1972 spin_unlock(&inode->i_lock);
1973
1974 return status;
1975 }
1976 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1977
1978 /**
1979 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1980 * @inode: pointer to inode
1981 * @fattr: updated attributes
1982 *
1983 * After an operation that has changed the inode metadata, mark the
1984 * attribute cache as being invalid, then try to update it. Fake up
1985 * weak cache consistency data, if none exist.
1986 *
1987 * This function is mainly designed to be used by the ->write_done() functions.
1988 */
nfs_post_op_update_inode_force_wcc_locked(struct inode * inode,struct nfs_fattr * fattr)1989 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1990 {
1991 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1992 int status;
1993
1994 /* Don't do a WCC update if these attributes are already stale */
1995 if (attr_cmp < 0)
1996 return 0;
1997 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
1998 /* Record the pre/post change info before clearing PRECHANGE */
1999 nfs_ooo_record(NFS_I(inode), fattr);
2000 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
2001 | NFS_ATTR_FATTR_PRESIZE
2002 | NFS_ATTR_FATTR_PREMTIME
2003 | NFS_ATTR_FATTR_PRECTIME);
2004 goto out_noforce;
2005 }
2006 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
2007 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
2008 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
2009 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
2010 }
2011 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
2012 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
2013 fattr->pre_ctime = inode_get_ctime(inode);
2014 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
2015 }
2016 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
2017 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
2018 fattr->pre_mtime = inode->i_mtime;
2019 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
2020 }
2021 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
2022 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
2023 fattr->pre_size = i_size_read(inode);
2024 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
2025 }
2026 out_noforce:
2027 status = nfs_post_op_update_inode_locked(inode, fattr,
2028 NFS_INO_INVALID_CHANGE
2029 | NFS_INO_INVALID_CTIME
2030 | NFS_INO_INVALID_MTIME
2031 | NFS_INO_INVALID_BLOCKS);
2032 return status;
2033 }
2034
2035 /**
2036 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
2037 * @inode: pointer to inode
2038 * @fattr: updated attributes
2039 *
2040 * After an operation that has changed the inode metadata, mark the
2041 * attribute cache as being invalid, then try to update it. Fake up
2042 * weak cache consistency data, if none exist.
2043 *
2044 * This function is mainly designed to be used by the ->write_done() functions.
2045 */
nfs_post_op_update_inode_force_wcc(struct inode * inode,struct nfs_fattr * fattr)2046 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
2047 {
2048 int status;
2049
2050 spin_lock(&inode->i_lock);
2051 nfs_fattr_set_barrier(fattr);
2052 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
2053 spin_unlock(&inode->i_lock);
2054 return status;
2055 }
2056 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
2057
2058
2059 /*
2060 * Many nfs protocol calls return the new file attributes after
2061 * an operation. Here we update the inode to reflect the state
2062 * of the server's inode.
2063 *
2064 * This is a bit tricky because we have to make sure all dirty pages
2065 * have been sent off to the server before calling invalidate_inode_pages.
2066 * To make sure no other process adds more write requests while we try
2067 * our best to flush them, we make them sleep during the attribute refresh.
2068 *
2069 * A very similar scenario holds for the dir cache.
2070 */
nfs_update_inode(struct inode * inode,struct nfs_fattr * fattr)2071 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2072 {
2073 struct nfs_server *server = NFS_SERVER(inode);
2074 struct nfs_inode *nfsi = NFS_I(inode);
2075 loff_t cur_isize, new_isize;
2076 u64 fattr_supported = server->fattr_valid;
2077 unsigned long invalid = 0;
2078 unsigned long now = jiffies;
2079 unsigned long save_cache_validity;
2080 bool have_writers = nfs_file_has_buffered_writers(nfsi);
2081 bool cache_revalidated = true;
2082 bool attr_changed = false;
2083 bool have_delegation;
2084
2085 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
2086 __func__, inode->i_sb->s_id, inode->i_ino,
2087 nfs_display_fhandle_hash(NFS_FH(inode)),
2088 atomic_read(&inode->i_count), fattr->valid);
2089
2090 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2091 /* Only a mounted-on-fileid? Just exit */
2092 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2093 return 0;
2094 /* Has the inode gone and changed behind our back? */
2095 } else if (nfsi->fileid != fattr->fileid) {
2096 /* Is this perhaps the mounted-on fileid? */
2097 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2098 nfsi->fileid == fattr->mounted_on_fileid)
2099 return 0;
2100 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2101 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2102 NFS_SERVER(inode)->nfs_client->cl_hostname,
2103 inode->i_sb->s_id, (long long)nfsi->fileid,
2104 (long long)fattr->fileid);
2105 goto out_err;
2106 }
2107
2108 /*
2109 * Make sure the inode's type hasn't changed.
2110 */
2111 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2112 /*
2113 * Big trouble! The inode has become a different object.
2114 */
2115 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2116 __func__, inode->i_ino, inode->i_mode, fattr->mode);
2117 goto out_err;
2118 }
2119
2120 /* Update the fsid? */
2121 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2122 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2123 !IS_AUTOMOUNT(inode))
2124 server->fsid = fattr->fsid;
2125
2126 /* Save the delegation state before clearing cache_validity */
2127 have_delegation = nfs_have_delegated_attributes(inode);
2128
2129 /*
2130 * Update the read time so we don't revalidate too often.
2131 */
2132 nfsi->read_cache_jiffies = fattr->time_start;
2133
2134 save_cache_validity = nfsi->cache_validity;
2135 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2136 | NFS_INO_INVALID_ATIME
2137 | NFS_INO_REVAL_FORCED
2138 | NFS_INO_INVALID_BLOCKS);
2139
2140 /* Do atomic weak cache consistency updates */
2141 nfs_wcc_update_inode(inode, fattr);
2142
2143 if (pnfs_layoutcommit_outstanding(inode)) {
2144 nfsi->cache_validity |=
2145 save_cache_validity &
2146 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2147 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2148 NFS_INO_INVALID_BLOCKS);
2149 cache_revalidated = false;
2150 }
2151
2152 /* More cache consistency checks */
2153 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2154 if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 &&
2155 nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) {
2156 /* There is one remaining gap that hasn't been
2157 * merged into iversion - do that now.
2158 */
2159 inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start);
2160 kfree(nfsi->ooo);
2161 nfsi->ooo = NULL;
2162 }
2163 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2164 /* Could it be a race with writeback? */
2165 if (!(have_writers || have_delegation)) {
2166 invalid |= NFS_INO_INVALID_DATA
2167 | NFS_INO_INVALID_ACCESS
2168 | NFS_INO_INVALID_ACL
2169 | NFS_INO_INVALID_XATTR;
2170 /* Force revalidate of all attributes */
2171 save_cache_validity |= NFS_INO_INVALID_CTIME
2172 | NFS_INO_INVALID_MTIME
2173 | NFS_INO_INVALID_SIZE
2174 | NFS_INO_INVALID_BLOCKS
2175 | NFS_INO_INVALID_NLINK
2176 | NFS_INO_INVALID_MODE
2177 | NFS_INO_INVALID_OTHER;
2178 if (S_ISDIR(inode->i_mode))
2179 nfs_force_lookup_revalidate(inode);
2180 attr_changed = true;
2181 dprintk("NFS: change_attr change on server for file %s/%ld\n",
2182 inode->i_sb->s_id,
2183 inode->i_ino);
2184 } else if (!have_delegation) {
2185 nfs_ooo_record(nfsi, fattr);
2186 nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode),
2187 fattr->change_attr);
2188 }
2189 inode_set_iversion_raw(inode, fattr->change_attr);
2190 }
2191 } else {
2192 nfsi->cache_validity |=
2193 save_cache_validity & NFS_INO_INVALID_CHANGE;
2194 if (!have_delegation ||
2195 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2196 cache_revalidated = false;
2197 }
2198
2199 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2200 inode->i_mtime = fattr->mtime;
2201 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2202 nfsi->cache_validity |=
2203 save_cache_validity & NFS_INO_INVALID_MTIME;
2204
2205 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2206 inode_set_ctime_to_ts(inode, fattr->ctime);
2207 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2208 nfsi->cache_validity |=
2209 save_cache_validity & NFS_INO_INVALID_CTIME;
2210
2211 /* Check if our cached file size is stale */
2212 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2213 new_isize = nfs_size_to_loff_t(fattr->size);
2214 cur_isize = i_size_read(inode);
2215 if (new_isize != cur_isize && !have_delegation) {
2216 /* Do we perhaps have any outstanding writes, or has
2217 * the file grown beyond our last write? */
2218 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2219 trace_nfs_size_update(inode, new_isize);
2220 i_size_write(inode, new_isize);
2221 if (!have_writers)
2222 invalid |= NFS_INO_INVALID_DATA;
2223 }
2224 }
2225 if (new_isize == 0 &&
2226 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2227 NFS_ATTR_FATTR_BLOCKS_USED))) {
2228 fattr->du.nfs3.used = 0;
2229 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2230 }
2231 } else
2232 nfsi->cache_validity |=
2233 save_cache_validity & NFS_INO_INVALID_SIZE;
2234
2235 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2236 inode->i_atime = fattr->atime;
2237 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2238 nfsi->cache_validity |=
2239 save_cache_validity & NFS_INO_INVALID_ATIME;
2240
2241 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2242 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2243 umode_t newmode = inode->i_mode & S_IFMT;
2244 newmode |= fattr->mode & S_IALLUGO;
2245 inode->i_mode = newmode;
2246 invalid |= NFS_INO_INVALID_ACCESS
2247 | NFS_INO_INVALID_ACL;
2248 }
2249 } else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2250 nfsi->cache_validity |=
2251 save_cache_validity & NFS_INO_INVALID_MODE;
2252
2253 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2254 if (!uid_eq(inode->i_uid, fattr->uid)) {
2255 invalid |= NFS_INO_INVALID_ACCESS
2256 | NFS_INO_INVALID_ACL;
2257 inode->i_uid = fattr->uid;
2258 }
2259 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2260 nfsi->cache_validity |=
2261 save_cache_validity & NFS_INO_INVALID_OTHER;
2262
2263 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2264 if (!gid_eq(inode->i_gid, fattr->gid)) {
2265 invalid |= NFS_INO_INVALID_ACCESS
2266 | NFS_INO_INVALID_ACL;
2267 inode->i_gid = fattr->gid;
2268 }
2269 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2270 nfsi->cache_validity |=
2271 save_cache_validity & NFS_INO_INVALID_OTHER;
2272
2273 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2274 if (inode->i_nlink != fattr->nlink)
2275 set_nlink(inode, fattr->nlink);
2276 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2277 nfsi->cache_validity |=
2278 save_cache_validity & NFS_INO_INVALID_NLINK;
2279
2280 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2281 /*
2282 * report the blocks in 512byte units
2283 */
2284 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2285 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2286 nfsi->cache_validity |=
2287 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2288
2289 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2290 inode->i_blocks = fattr->du.nfs2.blocks;
2291 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2292 nfsi->cache_validity |=
2293 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2294
2295 /* Update attrtimeo value if we're out of the unstable period */
2296 if (attr_changed) {
2297 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2298 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2299 nfsi->attrtimeo_timestamp = now;
2300 /* Set barrier to be more recent than all outstanding updates */
2301 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2302 } else {
2303 if (cache_revalidated) {
2304 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2305 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2306 nfsi->attrtimeo <<= 1;
2307 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2308 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2309 }
2310 nfsi->attrtimeo_timestamp = now;
2311 }
2312 /* Set the barrier to be more recent than this fattr */
2313 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2314 nfsi->attr_gencount = fattr->gencount;
2315 }
2316
2317 /* Don't invalidate the data if we were to blame */
2318 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2319 || S_ISLNK(inode->i_mode)))
2320 invalid &= ~NFS_INO_INVALID_DATA;
2321 nfs_set_cache_invalid(inode, invalid);
2322
2323 return 0;
2324 out_err:
2325 /*
2326 * No need to worry about unhashing the dentry, as the
2327 * lookup validation will know that the inode is bad.
2328 * (But we fall through to invalidate the caches.)
2329 */
2330 nfs_set_inode_stale_locked(inode);
2331 return -ESTALE;
2332 }
2333
nfs_alloc_inode(struct super_block * sb)2334 struct inode *nfs_alloc_inode(struct super_block *sb)
2335 {
2336 struct nfs_inode *nfsi;
2337 nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2338 if (!nfsi)
2339 return NULL;
2340 nfsi->flags = 0UL;
2341 nfsi->cache_validity = 0UL;
2342 nfsi->ooo = NULL;
2343 #if IS_ENABLED(CONFIG_NFS_V4)
2344 nfsi->nfs4_acl = NULL;
2345 #endif /* CONFIG_NFS_V4 */
2346 #ifdef CONFIG_NFS_V4_2
2347 nfsi->xattr_cache = NULL;
2348 #endif
2349 nfs_netfs_inode_init(nfsi);
2350
2351 return &nfsi->vfs_inode;
2352 }
2353 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2354
nfs_free_inode(struct inode * inode)2355 void nfs_free_inode(struct inode *inode)
2356 {
2357 kfree(NFS_I(inode)->ooo);
2358 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2359 }
2360 EXPORT_SYMBOL_GPL(nfs_free_inode);
2361
nfs4_init_once(struct nfs_inode * nfsi)2362 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2363 {
2364 #if IS_ENABLED(CONFIG_NFS_V4)
2365 INIT_LIST_HEAD(&nfsi->open_states);
2366 nfsi->delegation = NULL;
2367 init_rwsem(&nfsi->rwsem);
2368 nfsi->layout = NULL;
2369 #endif
2370 }
2371
init_once(void * foo)2372 static void init_once(void *foo)
2373 {
2374 struct nfs_inode *nfsi = foo;
2375
2376 inode_init_once(&nfsi->vfs_inode);
2377 INIT_LIST_HEAD(&nfsi->open_files);
2378 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2379 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2380 nfs4_init_once(nfsi);
2381 }
2382
nfs_init_inodecache(void)2383 static int __init nfs_init_inodecache(void)
2384 {
2385 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2386 sizeof(struct nfs_inode),
2387 0, (SLAB_RECLAIM_ACCOUNT|
2388 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2389 init_once);
2390 if (nfs_inode_cachep == NULL)
2391 return -ENOMEM;
2392
2393 return 0;
2394 }
2395
nfs_destroy_inodecache(void)2396 static void nfs_destroy_inodecache(void)
2397 {
2398 /*
2399 * Make sure all delayed rcu free inodes are flushed before we
2400 * destroy cache.
2401 */
2402 rcu_barrier();
2403 kmem_cache_destroy(nfs_inode_cachep);
2404 }
2405
2406 struct workqueue_struct *nfsiod_workqueue;
2407 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2408
2409 /*
2410 * start up the nfsiod workqueue
2411 */
nfsiod_start(void)2412 static int nfsiod_start(void)
2413 {
2414 struct workqueue_struct *wq;
2415 dprintk("RPC: creating workqueue nfsiod\n");
2416 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2417 if (wq == NULL)
2418 return -ENOMEM;
2419 nfsiod_workqueue = wq;
2420 return 0;
2421 }
2422
2423 /*
2424 * Destroy the nfsiod workqueue
2425 */
nfsiod_stop(void)2426 static void nfsiod_stop(void)
2427 {
2428 struct workqueue_struct *wq;
2429
2430 wq = nfsiod_workqueue;
2431 if (wq == NULL)
2432 return;
2433 nfsiod_workqueue = NULL;
2434 destroy_workqueue(wq);
2435 }
2436
2437 unsigned int nfs_net_id;
2438 EXPORT_SYMBOL_GPL(nfs_net_id);
2439
nfs_net_init(struct net * net)2440 static int nfs_net_init(struct net *net)
2441 {
2442 struct nfs_net *nn = net_generic(net, nfs_net_id);
2443
2444 nfs_clients_init(net);
2445
2446 if (!rpc_proc_register(net, &nn->rpcstats)) {
2447 nfs_clients_exit(net);
2448 return -ENOMEM;
2449 }
2450
2451 return nfs_fs_proc_net_init(net);
2452 }
2453
nfs_net_exit(struct net * net)2454 static void nfs_net_exit(struct net *net)
2455 {
2456 rpc_proc_unregister(net, "nfs");
2457 nfs_fs_proc_net_exit(net);
2458 nfs_clients_exit(net);
2459 }
2460
2461 static struct pernet_operations nfs_net_ops = {
2462 .init = nfs_net_init,
2463 .exit = nfs_net_exit,
2464 .id = &nfs_net_id,
2465 .size = sizeof(struct nfs_net),
2466 };
2467
2468 /*
2469 * Initialize NFS
2470 */
init_nfs_fs(void)2471 static int __init init_nfs_fs(void)
2472 {
2473 int err;
2474
2475 err = nfs_sysfs_init();
2476 if (err < 0)
2477 goto out10;
2478
2479 err = register_pernet_subsys(&nfs_net_ops);
2480 if (err < 0)
2481 goto out9;
2482
2483 err = nfsiod_start();
2484 if (err)
2485 goto out7;
2486
2487 err = nfs_fs_proc_init();
2488 if (err)
2489 goto out6;
2490
2491 err = nfs_init_nfspagecache();
2492 if (err)
2493 goto out5;
2494
2495 err = nfs_init_inodecache();
2496 if (err)
2497 goto out4;
2498
2499 err = nfs_init_readpagecache();
2500 if (err)
2501 goto out3;
2502
2503 err = nfs_init_writepagecache();
2504 if (err)
2505 goto out2;
2506
2507 err = nfs_init_directcache();
2508 if (err)
2509 goto out1;
2510
2511 err = register_nfs_fs();
2512 if (err)
2513 goto out0;
2514
2515 return 0;
2516 out0:
2517 nfs_destroy_directcache();
2518 out1:
2519 nfs_destroy_writepagecache();
2520 out2:
2521 nfs_destroy_readpagecache();
2522 out3:
2523 nfs_destroy_inodecache();
2524 out4:
2525 nfs_destroy_nfspagecache();
2526 out5:
2527 nfs_fs_proc_exit();
2528 out6:
2529 nfsiod_stop();
2530 out7:
2531 unregister_pernet_subsys(&nfs_net_ops);
2532 out9:
2533 nfs_sysfs_exit();
2534 out10:
2535 return err;
2536 }
2537
exit_nfs_fs(void)2538 static void __exit exit_nfs_fs(void)
2539 {
2540 nfs_destroy_directcache();
2541 nfs_destroy_writepagecache();
2542 nfs_destroy_readpagecache();
2543 nfs_destroy_inodecache();
2544 nfs_destroy_nfspagecache();
2545 unregister_pernet_subsys(&nfs_net_ops);
2546 unregister_nfs_fs();
2547 nfs_fs_proc_exit();
2548 nfsiod_stop();
2549 nfs_sysfs_exit();
2550 }
2551
2552 /* Not quite true; I just maintain it */
2553 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2554 MODULE_LICENSE("GPL");
2555 module_param(enable_ino64, bool, 0644);
2556
2557 module_init(init_nfs_fs)
2558 module_exit(exit_nfs_fs)
2559