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 flags |= nfsi->cache_validity;
210 if (inode->i_mapping->nrpages == 0)
211 flags &= ~NFS_INO_INVALID_DATA;
212
213 /* pairs with nfs_clear_invalid_mapping()'s smp_load_acquire() */
214 smp_store_release(&nfsi->cache_validity, flags);
215
216 if (inode->i_mapping->nrpages == 0 ||
217 nfsi->cache_validity & NFS_INO_INVALID_DATA) {
218 nfs_ooo_clear(nfsi);
219 }
220 trace_nfs_set_cache_invalid(inode, 0);
221 }
222 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
223
224 /*
225 * Invalidate the local caches
226 */
nfs_zap_caches_locked(struct inode * inode)227 static void nfs_zap_caches_locked(struct inode *inode)
228 {
229 struct nfs_inode *nfsi = NFS_I(inode);
230 int mode = inode->i_mode;
231
232 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
233
234 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
235 nfsi->attrtimeo_timestamp = jiffies;
236
237 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
238 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
239 NFS_INO_INVALID_DATA |
240 NFS_INO_INVALID_ACCESS |
241 NFS_INO_INVALID_ACL |
242 NFS_INO_INVALID_XATTR);
243 else
244 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
245 NFS_INO_INVALID_ACCESS |
246 NFS_INO_INVALID_ACL |
247 NFS_INO_INVALID_XATTR);
248 nfs_zap_label_cache_locked(nfsi);
249 }
250
nfs_zap_caches(struct inode * inode)251 void nfs_zap_caches(struct inode *inode)
252 {
253 spin_lock(&inode->i_lock);
254 nfs_zap_caches_locked(inode);
255 spin_unlock(&inode->i_lock);
256 }
257
nfs_zap_mapping(struct inode * inode,struct address_space * mapping)258 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
259 {
260 if (mapping->nrpages != 0) {
261 spin_lock(&inode->i_lock);
262 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
263 spin_unlock(&inode->i_lock);
264 }
265 }
266
nfs_zap_acl_cache(struct inode * inode)267 void nfs_zap_acl_cache(struct inode *inode)
268 {
269 void (*clear_acl_cache)(struct inode *);
270
271 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
272 if (clear_acl_cache != NULL)
273 clear_acl_cache(inode);
274 spin_lock(&inode->i_lock);
275 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
276 spin_unlock(&inode->i_lock);
277 }
278 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
279
nfs_invalidate_atime(struct inode * inode)280 void nfs_invalidate_atime(struct inode *inode)
281 {
282 spin_lock(&inode->i_lock);
283 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
284 spin_unlock(&inode->i_lock);
285 }
286 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
287
288 /*
289 * Invalidate, but do not unhash, the inode.
290 * NB: must be called with inode->i_lock held!
291 */
nfs_set_inode_stale_locked(struct inode * inode)292 static void nfs_set_inode_stale_locked(struct inode *inode)
293 {
294 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
295 nfs_zap_caches_locked(inode);
296 trace_nfs_set_inode_stale(inode);
297 }
298
nfs_set_inode_stale(struct inode * inode)299 void nfs_set_inode_stale(struct inode *inode)
300 {
301 spin_lock(&inode->i_lock);
302 nfs_set_inode_stale_locked(inode);
303 spin_unlock(&inode->i_lock);
304 }
305
306 struct nfs_find_desc {
307 struct nfs_fh *fh;
308 struct nfs_fattr *fattr;
309 };
310
311 /*
312 * In NFSv3 we can have 64bit inode numbers. In order to support
313 * this, and re-exported directories (also seen in NFSv2)
314 * we are forced to allow 2 different inodes to have the same
315 * i_ino.
316 */
317 static int
nfs_find_actor(struct inode * inode,void * opaque)318 nfs_find_actor(struct inode *inode, void *opaque)
319 {
320 struct nfs_find_desc *desc = opaque;
321 struct nfs_fh *fh = desc->fh;
322 struct nfs_fattr *fattr = desc->fattr;
323
324 if (NFS_FILEID(inode) != fattr->fileid)
325 return 0;
326 if (inode_wrong_type(inode, fattr->mode))
327 return 0;
328 if (nfs_compare_fh(NFS_FH(inode), fh))
329 return 0;
330 if (is_bad_inode(inode) || NFS_STALE(inode))
331 return 0;
332 return 1;
333 }
334
335 static int
nfs_init_locked(struct inode * inode,void * opaque)336 nfs_init_locked(struct inode *inode, void *opaque)
337 {
338 struct nfs_find_desc *desc = opaque;
339 struct nfs_fattr *fattr = desc->fattr;
340
341 set_nfs_fileid(inode, fattr->fileid);
342 inode->i_mode = fattr->mode;
343 nfs_copy_fh(NFS_FH(inode), desc->fh);
344 return 0;
345 }
346
347 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
nfs_clear_label_invalid(struct inode * inode)348 static void nfs_clear_label_invalid(struct inode *inode)
349 {
350 spin_lock(&inode->i_lock);
351 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
352 spin_unlock(&inode->i_lock);
353 }
354
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)355 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
356 {
357 int error;
358
359 if (fattr->label == NULL)
360 return;
361
362 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
363 error = security_inode_notifysecctx(inode, fattr->label->label,
364 fattr->label->len);
365 if (error)
366 printk(KERN_ERR "%s() %s %d "
367 "security_inode_notifysecctx() %d\n",
368 __func__,
369 (char *)fattr->label->label,
370 fattr->label->len, error);
371 nfs_clear_label_invalid(inode);
372 }
373 }
374
nfs4_label_alloc(struct nfs_server * server,gfp_t flags)375 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
376 {
377 struct nfs4_label *label;
378
379 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
380 return NULL;
381
382 label = kzalloc(sizeof(struct nfs4_label), flags);
383 if (label == NULL)
384 return ERR_PTR(-ENOMEM);
385
386 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
387 if (label->label == NULL) {
388 kfree(label);
389 return ERR_PTR(-ENOMEM);
390 }
391 label->len = NFS4_MAXLABELLEN;
392
393 return label;
394 }
395 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
396 #else
nfs_setsecurity(struct inode * inode,struct nfs_fattr * fattr)397 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
398 {
399 }
400 #endif
401 EXPORT_SYMBOL_GPL(nfs_setsecurity);
402
403 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
404 struct inode *
nfs_ilookup(struct super_block * sb,struct nfs_fattr * fattr,struct nfs_fh * fh)405 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
406 {
407 struct nfs_find_desc desc = {
408 .fh = fh,
409 .fattr = fattr,
410 };
411 struct inode *inode;
412 unsigned long hash;
413
414 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
415 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
416 return NULL;
417
418 hash = nfs_fattr_to_ino_t(fattr);
419 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
420
421 dprintk("%s: returning %p\n", __func__, inode);
422 return inode;
423 }
424
nfs_inode_init_regular(struct nfs_inode * nfsi)425 static void nfs_inode_init_regular(struct nfs_inode *nfsi)
426 {
427 atomic_long_set(&nfsi->nrequests, 0);
428 atomic_long_set(&nfsi->redirtied_pages, 0);
429 INIT_LIST_HEAD(&nfsi->commit_info.list);
430 atomic_long_set(&nfsi->commit_info.ncommit, 0);
431 atomic_set(&nfsi->commit_info.rpcs_out, 0);
432 mutex_init(&nfsi->commit_mutex);
433 }
434
nfs_inode_init_dir(struct nfs_inode * nfsi)435 static void nfs_inode_init_dir(struct nfs_inode *nfsi)
436 {
437 nfsi->cache_change_attribute = 0;
438 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
439 init_rwsem(&nfsi->rmdir_sem);
440 }
441
442 /*
443 * This is our front-end to iget that looks up inodes by file handle
444 * instead of inode number.
445 */
446 struct inode *
nfs_fhget(struct super_block * sb,struct nfs_fh * fh,struct nfs_fattr * fattr)447 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
448 {
449 struct nfs_find_desc desc = {
450 .fh = fh,
451 .fattr = fattr
452 };
453 struct inode *inode = ERR_PTR(-ENOENT);
454 u64 fattr_supported = NFS_SB(sb)->fattr_valid;
455 unsigned long hash;
456
457 nfs_attr_check_mountpoint(sb, fattr);
458
459 if (nfs_attr_use_mounted_on_fileid(fattr))
460 fattr->fileid = fattr->mounted_on_fileid;
461 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
462 goto out_no_inode;
463 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
464 goto out_no_inode;
465
466 hash = nfs_fattr_to_ino_t(fattr);
467
468 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
469 if (inode == NULL) {
470 inode = ERR_PTR(-ENOMEM);
471 goto out_no_inode;
472 }
473
474 if (inode->i_state & I_NEW) {
475 struct nfs_inode *nfsi = NFS_I(inode);
476 unsigned long now = jiffies;
477
478 /* We set i_ino for the few things that still rely on it,
479 * such as stat(2) */
480 inode->i_ino = hash;
481
482 /* We can't support update_atime(), since the server will reset it */
483 inode->i_flags |= S_NOATIME|S_NOCMTIME;
484 inode->i_mode = fattr->mode;
485 nfsi->cache_validity = 0;
486 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
487 && (fattr_supported & NFS_ATTR_FATTR_MODE))
488 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
489 /* Why so? Because we want revalidate for devices/FIFOs, and
490 * that's precisely what we have in nfs_file_inode_operations.
491 */
492 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
493 if (S_ISREG(inode->i_mode)) {
494 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
495 inode->i_data.a_ops = &nfs_file_aops;
496 nfs_inode_init_regular(nfsi);
497 } else if (S_ISDIR(inode->i_mode)) {
498 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
499 inode->i_fop = &nfs_dir_operations;
500 inode->i_data.a_ops = &nfs_dir_aops;
501 nfs_inode_init_dir(nfsi);
502 /* Deal with crossing mountpoints */
503 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
504 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
505 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
506 inode->i_op = &nfs_referral_inode_operations;
507 else
508 inode->i_op = &nfs_mountpoint_inode_operations;
509 inode->i_fop = NULL;
510 inode->i_flags |= S_AUTOMOUNT;
511 }
512 } else if (S_ISLNK(inode->i_mode)) {
513 inode->i_op = &nfs_symlink_inode_operations;
514 inode_nohighmem(inode);
515 } else
516 init_special_inode(inode, inode->i_mode, fattr->rdev);
517
518 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
519 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
520 inode_set_ctime(inode, 0, 0);
521 inode_set_iversion_raw(inode, 0);
522 inode->i_size = 0;
523 clear_nlink(inode);
524 inode->i_uid = make_kuid(&init_user_ns, -2);
525 inode->i_gid = make_kgid(&init_user_ns, -2);
526 inode->i_blocks = 0;
527 nfsi->write_io = 0;
528 nfsi->read_io = 0;
529
530 nfsi->read_cache_jiffies = fattr->time_start;
531 nfsi->attr_gencount = fattr->gencount;
532 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
533 inode->i_atime = fattr->atime;
534 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
535 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
536 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
537 inode->i_mtime = fattr->mtime;
538 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
539 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
540 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
541 inode_set_ctime_to_ts(inode, fattr->ctime);
542 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
543 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
544 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
545 inode_set_iversion_raw(inode, fattr->change_attr);
546 else
547 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
548 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
549 inode->i_size = nfs_size_to_loff_t(fattr->size);
550 else
551 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
552 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
553 set_nlink(inode, fattr->nlink);
554 else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
555 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
556 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
557 inode->i_uid = fattr->uid;
558 else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
559 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
560 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
561 inode->i_gid = fattr->gid;
562 else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
563 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
564 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
565 inode->i_blocks = fattr->du.nfs2.blocks;
566 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
567 fattr->size != 0)
568 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
569 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
570 /*
571 * report the blocks in 512byte units
572 */
573 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
574 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
575 fattr->size != 0)
576 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
577
578 nfs_setsecurity(inode, fattr);
579
580 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
581 nfsi->attrtimeo_timestamp = now;
582 nfsi->access_cache = RB_ROOT;
583
584 nfs_fscache_init_inode(inode);
585
586 unlock_new_inode(inode);
587 } else {
588 int err = nfs_refresh_inode(inode, fattr);
589 if (err < 0) {
590 iput(inode);
591 inode = ERR_PTR(err);
592 goto out_no_inode;
593 }
594 }
595 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
596 inode->i_sb->s_id,
597 (unsigned long long)NFS_FILEID(inode),
598 nfs_display_fhandle_hash(fh),
599 atomic_read(&inode->i_count));
600
601 out:
602 return inode;
603
604 out_no_inode:
605 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
606 goto out;
607 }
608 EXPORT_SYMBOL_GPL(nfs_fhget);
609
610 #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)
611
612 int
nfs_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)613 nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
614 struct iattr *attr)
615 {
616 struct inode *inode = d_inode(dentry);
617 struct nfs_fattr *fattr;
618 int error = 0;
619
620 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
621
622 /* skip mode change if it's just for clearing setuid/setgid */
623 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
624 attr->ia_valid &= ~ATTR_MODE;
625
626 if (attr->ia_valid & ATTR_SIZE) {
627 BUG_ON(!S_ISREG(inode->i_mode));
628
629 error = inode_newsize_ok(inode, attr->ia_size);
630 if (error)
631 return error;
632
633 if (attr->ia_size == i_size_read(inode))
634 attr->ia_valid &= ~ATTR_SIZE;
635 }
636
637 /* Optimization: if the end result is no change, don't RPC */
638 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
639 return 0;
640
641 trace_nfs_setattr_enter(inode);
642
643 /* Write all dirty data */
644 if (S_ISREG(inode->i_mode))
645 nfs_sync_inode(inode);
646
647 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
648 if (fattr == NULL) {
649 error = -ENOMEM;
650 goto out;
651 }
652
653 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
654 if (error == 0)
655 error = nfs_refresh_inode(inode, fattr);
656 nfs_free_fattr(fattr);
657 out:
658 trace_nfs_setattr_exit(inode, error);
659 return error;
660 }
661 EXPORT_SYMBOL_GPL(nfs_setattr);
662
663 /**
664 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
665 * @inode: inode of the file used
666 * @offset: file offset to start truncating
667 *
668 * This is a copy of the common vmtruncate, but with the locking
669 * corrected to take into account the fact that NFS requires
670 * inode->i_size to be updated under the inode->i_lock.
671 * Note: must be called with inode->i_lock held!
672 */
nfs_vmtruncate(struct inode * inode,loff_t offset)673 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
674 {
675 int err;
676
677 err = inode_newsize_ok(inode, offset);
678 if (err)
679 goto out;
680
681 trace_nfs_size_truncate(inode, offset);
682 i_size_write(inode, offset);
683 /* Optimisation */
684 if (offset == 0) {
685 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
686 nfs_ooo_clear(NFS_I(inode));
687 }
688 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
689
690 spin_unlock(&inode->i_lock);
691 truncate_pagecache(inode, offset);
692 spin_lock(&inode->i_lock);
693 out:
694 return err;
695 }
696
697 /**
698 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
699 * @inode: pointer to struct inode
700 * @attr: pointer to struct iattr
701 * @fattr: pointer to struct nfs_fattr
702 *
703 * Note: we do this in the *proc.c in order to ensure that
704 * it works for things like exclusive creates too.
705 */
nfs_setattr_update_inode(struct inode * inode,struct iattr * attr,struct nfs_fattr * fattr)706 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
707 struct nfs_fattr *fattr)
708 {
709 /* Barrier: bump the attribute generation count. */
710 nfs_fattr_set_barrier(fattr);
711
712 spin_lock(&inode->i_lock);
713 NFS_I(inode)->attr_gencount = fattr->gencount;
714 if ((attr->ia_valid & ATTR_SIZE) != 0) {
715 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME |
716 NFS_INO_INVALID_BLOCKS);
717 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
718 nfs_vmtruncate(inode, attr->ia_size);
719 }
720 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
721 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
722 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
723 inode->i_mode & S_ISUID)
724 inode->i_mode &= ~S_ISUID;
725 if (setattr_should_drop_sgid(&nop_mnt_idmap, inode))
726 inode->i_mode &= ~S_ISGID;
727 if ((attr->ia_valid & ATTR_MODE) != 0) {
728 int mode = attr->ia_mode & S_IALLUGO;
729 mode |= inode->i_mode & ~S_IALLUGO;
730 inode->i_mode = mode;
731 }
732 if ((attr->ia_valid & ATTR_UID) != 0)
733 inode->i_uid = attr->ia_uid;
734 if ((attr->ia_valid & ATTR_GID) != 0)
735 inode->i_gid = attr->ia_gid;
736 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
737 inode_set_ctime_to_ts(inode, fattr->ctime);
738 else
739 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
740 | NFS_INO_INVALID_CTIME);
741 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
742 | NFS_INO_INVALID_ACL);
743 }
744 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
745 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
746 | NFS_INO_INVALID_CTIME);
747 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
748 inode->i_atime = fattr->atime;
749 else if (attr->ia_valid & ATTR_ATIME_SET)
750 inode->i_atime = attr->ia_atime;
751 else
752 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
753
754 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
755 inode_set_ctime_to_ts(inode, fattr->ctime);
756 else
757 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
758 | NFS_INO_INVALID_CTIME);
759 }
760 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
761 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
762 | NFS_INO_INVALID_CTIME);
763 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
764 inode->i_mtime = fattr->mtime;
765 else if (attr->ia_valid & ATTR_MTIME_SET)
766 inode->i_mtime = attr->ia_mtime;
767 else
768 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
769
770 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
771 inode_set_ctime_to_ts(inode, fattr->ctime);
772 else
773 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
774 | NFS_INO_INVALID_CTIME);
775 }
776 if (fattr->valid)
777 nfs_update_inode(inode, fattr);
778 spin_unlock(&inode->i_lock);
779 }
780 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
781
782 /*
783 * Don't request help from readdirplus if the file is being written to,
784 * or if attribute caching is turned off
785 */
nfs_getattr_readdirplus_enable(const struct inode * inode)786 static bool nfs_getattr_readdirplus_enable(const struct inode *inode)
787 {
788 return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) &&
789 !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ;
790 }
791
nfs_readdirplus_parent_cache_miss(struct dentry * dentry)792 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
793 {
794 if (!IS_ROOT(dentry)) {
795 struct dentry *parent = dget_parent(dentry);
796 nfs_readdir_record_entry_cache_miss(d_inode(parent));
797 dput(parent);
798 }
799 }
800
nfs_readdirplus_parent_cache_hit(struct dentry * dentry)801 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
802 {
803 if (!IS_ROOT(dentry)) {
804 struct dentry *parent = dget_parent(dentry);
805 nfs_readdir_record_entry_cache_hit(d_inode(parent));
806 dput(parent);
807 }
808 }
809
nfs_get_valid_attrmask(struct inode * inode)810 static u32 nfs_get_valid_attrmask(struct inode *inode)
811 {
812 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
813 u32 reply_mask = STATX_INO | STATX_TYPE;
814
815 if (!(cache_validity & NFS_INO_INVALID_ATIME))
816 reply_mask |= STATX_ATIME;
817 if (!(cache_validity & NFS_INO_INVALID_CTIME))
818 reply_mask |= STATX_CTIME;
819 if (!(cache_validity & NFS_INO_INVALID_MTIME))
820 reply_mask |= STATX_MTIME;
821 if (!(cache_validity & NFS_INO_INVALID_SIZE))
822 reply_mask |= STATX_SIZE;
823 if (!(cache_validity & NFS_INO_INVALID_NLINK))
824 reply_mask |= STATX_NLINK;
825 if (!(cache_validity & NFS_INO_INVALID_MODE))
826 reply_mask |= STATX_MODE;
827 if (!(cache_validity & NFS_INO_INVALID_OTHER))
828 reply_mask |= STATX_UID | STATX_GID;
829 if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
830 reply_mask |= STATX_BLOCKS;
831 if (!(cache_validity & NFS_INO_INVALID_CHANGE))
832 reply_mask |= STATX_CHANGE_COOKIE;
833 return reply_mask;
834 }
835
nfs_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)836 int nfs_getattr(struct mnt_idmap *idmap, const struct path *path,
837 struct kstat *stat, u32 request_mask, unsigned int query_flags)
838 {
839 struct inode *inode = d_inode(path->dentry);
840 struct nfs_server *server = NFS_SERVER(inode);
841 unsigned long cache_validity;
842 int err = 0;
843 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
844 bool do_update = false;
845 bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
846
847 trace_nfs_getattr_enter(inode);
848
849 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
850 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
851 STATX_INO | STATX_SIZE | STATX_BLOCKS |
852 STATX_CHANGE_COOKIE;
853
854 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
855 if (readdirplus_enabled)
856 nfs_readdirplus_parent_cache_hit(path->dentry);
857 goto out_no_revalidate;
858 }
859
860 /* Flush out writes to the server in order to update c/mtime/version. */
861 if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) &&
862 S_ISREG(inode->i_mode))
863 filemap_write_and_wait(inode->i_mapping);
864
865 /*
866 * We may force a getattr if the user cares about atime.
867 *
868 * Note that we only have to check the vfsmount flags here:
869 * - NFS always sets S_NOATIME by so checking it would give a
870 * bogus result
871 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
872 * no point in checking those.
873 */
874 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
875 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
876 request_mask &= ~STATX_ATIME;
877
878 /* Is the user requesting attributes that might need revalidation? */
879 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
880 STATX_MTIME|STATX_UID|STATX_GID|
881 STATX_SIZE|STATX_BLOCKS|
882 STATX_CHANGE_COOKIE)))
883 goto out_no_revalidate;
884
885 /* Check whether the cached attributes are stale */
886 do_update |= force_sync || nfs_attribute_cache_expired(inode);
887 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
888 do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
889 if (request_mask & STATX_ATIME)
890 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
891 if (request_mask & STATX_CTIME)
892 do_update |= cache_validity & NFS_INO_INVALID_CTIME;
893 if (request_mask & STATX_MTIME)
894 do_update |= cache_validity & NFS_INO_INVALID_MTIME;
895 if (request_mask & STATX_SIZE)
896 do_update |= cache_validity & NFS_INO_INVALID_SIZE;
897 if (request_mask & STATX_NLINK)
898 do_update |= cache_validity & NFS_INO_INVALID_NLINK;
899 if (request_mask & STATX_MODE)
900 do_update |= cache_validity & NFS_INO_INVALID_MODE;
901 if (request_mask & (STATX_UID | STATX_GID))
902 do_update |= cache_validity & NFS_INO_INVALID_OTHER;
903 if (request_mask & STATX_BLOCKS)
904 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
905
906 if (do_update) {
907 if (readdirplus_enabled)
908 nfs_readdirplus_parent_cache_miss(path->dentry);
909 err = __nfs_revalidate_inode(server, inode);
910 if (err)
911 goto out;
912 } else if (readdirplus_enabled)
913 nfs_readdirplus_parent_cache_hit(path->dentry);
914 out_no_revalidate:
915 /* Only return attributes that were revalidated. */
916 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
917
918 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
919 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
920 stat->change_cookie = inode_peek_iversion_raw(inode);
921 stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC;
922 if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED)
923 stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
924 if (S_ISDIR(inode->i_mode))
925 stat->blksize = NFS_SERVER(inode)->dtsize;
926 out:
927 trace_nfs_getattr_exit(inode, err);
928 return err;
929 }
930 EXPORT_SYMBOL_GPL(nfs_getattr);
931
nfs_init_lock_context(struct nfs_lock_context * l_ctx)932 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
933 {
934 refcount_set(&l_ctx->count, 1);
935 l_ctx->lockowner = current->files;
936 INIT_LIST_HEAD(&l_ctx->list);
937 atomic_set(&l_ctx->io_count, 0);
938 }
939
__nfs_find_lock_context(struct nfs_open_context * ctx)940 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
941 {
942 struct nfs_lock_context *pos;
943
944 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
945 if (pos->lockowner != current->files)
946 continue;
947 if (refcount_inc_not_zero(&pos->count))
948 return pos;
949 }
950 return NULL;
951 }
952
nfs_get_lock_context(struct nfs_open_context * ctx)953 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
954 {
955 struct nfs_lock_context *res, *new = NULL;
956 struct inode *inode = d_inode(ctx->dentry);
957
958 rcu_read_lock();
959 res = __nfs_find_lock_context(ctx);
960 rcu_read_unlock();
961 if (res == NULL) {
962 new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
963 if (new == NULL)
964 return ERR_PTR(-ENOMEM);
965 nfs_init_lock_context(new);
966 spin_lock(&inode->i_lock);
967 res = __nfs_find_lock_context(ctx);
968 if (res == NULL) {
969 new->open_context = get_nfs_open_context(ctx);
970 if (new->open_context) {
971 list_add_tail_rcu(&new->list,
972 &ctx->lock_context.list);
973 res = new;
974 new = NULL;
975 } else
976 res = ERR_PTR(-EBADF);
977 }
978 spin_unlock(&inode->i_lock);
979 kfree(new);
980 }
981 return res;
982 }
983 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
984
nfs_put_lock_context(struct nfs_lock_context * l_ctx)985 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
986 {
987 struct nfs_open_context *ctx = l_ctx->open_context;
988 struct inode *inode = d_inode(ctx->dentry);
989
990 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
991 return;
992 list_del_rcu(&l_ctx->list);
993 spin_unlock(&inode->i_lock);
994 put_nfs_open_context(ctx);
995 kfree_rcu(l_ctx, rcu_head);
996 }
997 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
998
999 /**
1000 * nfs_close_context - Common close_context() routine NFSv2/v3
1001 * @ctx: pointer to context
1002 * @is_sync: is this a synchronous close
1003 *
1004 * Ensure that the attributes are up to date if we're mounted
1005 * with close-to-open semantics and we have cached data that will
1006 * need to be revalidated on open.
1007 */
nfs_close_context(struct nfs_open_context * ctx,int is_sync)1008 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1009 {
1010 struct nfs_inode *nfsi;
1011 struct inode *inode;
1012
1013 if (!(ctx->mode & FMODE_WRITE))
1014 return;
1015 if (!is_sync)
1016 return;
1017 inode = d_inode(ctx->dentry);
1018 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1019 return;
1020 nfsi = NFS_I(inode);
1021 if (inode->i_mapping->nrpages == 0)
1022 return;
1023 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1024 return;
1025 if (!list_empty(&nfsi->open_files))
1026 return;
1027 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1028 return;
1029 nfs_revalidate_inode(inode,
1030 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1031 }
1032 EXPORT_SYMBOL_GPL(nfs_close_context);
1033
alloc_nfs_open_context(struct dentry * dentry,fmode_t f_mode,struct file * filp)1034 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1035 fmode_t f_mode,
1036 struct file *filp)
1037 {
1038 struct nfs_open_context *ctx;
1039
1040 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1041 if (!ctx)
1042 return ERR_PTR(-ENOMEM);
1043 nfs_sb_active(dentry->d_sb);
1044 ctx->dentry = dget(dentry);
1045 if (filp)
1046 ctx->cred = get_cred(filp->f_cred);
1047 else
1048 ctx->cred = get_current_cred();
1049 rcu_assign_pointer(ctx->ll_cred, NULL);
1050 ctx->state = NULL;
1051 ctx->mode = f_mode;
1052 ctx->flags = 0;
1053 ctx->error = 0;
1054 ctx->flock_owner = (fl_owner_t)filp;
1055 nfs_init_lock_context(&ctx->lock_context);
1056 ctx->lock_context.open_context = ctx;
1057 INIT_LIST_HEAD(&ctx->list);
1058 ctx->mdsthreshold = NULL;
1059 return ctx;
1060 }
1061 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1062
get_nfs_open_context(struct nfs_open_context * ctx)1063 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1064 {
1065 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1066 return ctx;
1067 return NULL;
1068 }
1069 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1070
__put_nfs_open_context(struct nfs_open_context * ctx,int is_sync)1071 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1072 {
1073 struct inode *inode = d_inode(ctx->dentry);
1074 struct super_block *sb = ctx->dentry->d_sb;
1075
1076 if (!refcount_dec_and_test(&ctx->lock_context.count))
1077 return;
1078 if (!list_empty(&ctx->list)) {
1079 spin_lock(&inode->i_lock);
1080 list_del_rcu(&ctx->list);
1081 spin_unlock(&inode->i_lock);
1082 }
1083 if (inode != NULL)
1084 NFS_PROTO(inode)->close_context(ctx, is_sync);
1085 put_cred(ctx->cred);
1086 dput(ctx->dentry);
1087 nfs_sb_deactive(sb);
1088 put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1089 kfree(ctx->mdsthreshold);
1090 kfree_rcu(ctx, rcu_head);
1091 }
1092
put_nfs_open_context(struct nfs_open_context * ctx)1093 void put_nfs_open_context(struct nfs_open_context *ctx)
1094 {
1095 __put_nfs_open_context(ctx, 0);
1096 }
1097 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1098
put_nfs_open_context_sync(struct nfs_open_context * ctx)1099 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1100 {
1101 __put_nfs_open_context(ctx, 1);
1102 }
1103
1104 /*
1105 * Ensure that mmap has a recent RPC credential for use when writing out
1106 * shared pages
1107 */
nfs_inode_attach_open_context(struct nfs_open_context * ctx)1108 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1109 {
1110 struct inode *inode = d_inode(ctx->dentry);
1111 struct nfs_inode *nfsi = NFS_I(inode);
1112
1113 spin_lock(&inode->i_lock);
1114 if (list_empty(&nfsi->open_files) &&
1115 nfs_ooo_test(nfsi))
1116 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1117 NFS_INO_REVAL_FORCED);
1118 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1119 spin_unlock(&inode->i_lock);
1120 }
1121 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1122
nfs_file_set_open_context(struct file * filp,struct nfs_open_context * ctx)1123 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1124 {
1125 filp->private_data = get_nfs_open_context(ctx);
1126 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1127 if (list_empty(&ctx->list))
1128 nfs_inode_attach_open_context(ctx);
1129 }
1130 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1131
1132 /*
1133 * Given an inode, search for an open context with the desired characteristics
1134 */
nfs_find_open_context(struct inode * inode,const struct cred * cred,fmode_t mode)1135 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1136 {
1137 struct nfs_inode *nfsi = NFS_I(inode);
1138 struct nfs_open_context *pos, *ctx = NULL;
1139
1140 rcu_read_lock();
1141 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1142 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1143 continue;
1144 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1145 continue;
1146 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1147 continue;
1148 ctx = get_nfs_open_context(pos);
1149 if (ctx)
1150 break;
1151 }
1152 rcu_read_unlock();
1153 return ctx;
1154 }
1155
nfs_file_clear_open_context(struct file * filp)1156 void nfs_file_clear_open_context(struct file *filp)
1157 {
1158 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1159
1160 if (ctx) {
1161 struct inode *inode = d_inode(ctx->dentry);
1162
1163 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1164 /*
1165 * We fatal error on write before. Try to writeback
1166 * every page again.
1167 */
1168 if (ctx->error < 0)
1169 invalidate_inode_pages2(inode->i_mapping);
1170 filp->private_data = NULL;
1171 put_nfs_open_context_sync(ctx);
1172 }
1173 }
1174
1175 /*
1176 * These allocate and release file read/write context information.
1177 */
nfs_open(struct inode * inode,struct file * filp)1178 int nfs_open(struct inode *inode, struct file *filp)
1179 {
1180 struct nfs_open_context *ctx;
1181
1182 ctx = alloc_nfs_open_context(file_dentry(filp),
1183 flags_to_mode(filp->f_flags), filp);
1184 if (IS_ERR(ctx))
1185 return PTR_ERR(ctx);
1186 nfs_file_set_open_context(filp, ctx);
1187 put_nfs_open_context(ctx);
1188 nfs_fscache_open_file(inode, filp);
1189 return 0;
1190 }
1191
1192 /*
1193 * This function is called whenever some part of NFS notices that
1194 * the cached attributes have to be refreshed.
1195 */
1196 int
__nfs_revalidate_inode(struct nfs_server * server,struct inode * inode)1197 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1198 {
1199 int status = -ESTALE;
1200 struct nfs_fattr *fattr = NULL;
1201 struct nfs_inode *nfsi = NFS_I(inode);
1202
1203 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1204 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1205
1206 trace_nfs_revalidate_inode_enter(inode);
1207
1208 if (is_bad_inode(inode))
1209 goto out;
1210 if (NFS_STALE(inode))
1211 goto out;
1212
1213 /* pNFS: Attributes aren't updated until we layoutcommit */
1214 if (S_ISREG(inode->i_mode)) {
1215 status = pnfs_sync_inode(inode, false);
1216 if (status)
1217 goto out;
1218 }
1219
1220 status = -ENOMEM;
1221 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1222 if (fattr == NULL)
1223 goto out;
1224
1225 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1226
1227 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1228 if (status != 0) {
1229 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1230 inode->i_sb->s_id,
1231 (unsigned long long)NFS_FILEID(inode), status);
1232 switch (status) {
1233 case -ETIMEDOUT:
1234 /* A soft timeout occurred. Use cached information? */
1235 if (server->flags & NFS_MOUNT_SOFTREVAL)
1236 status = 0;
1237 break;
1238 case -ESTALE:
1239 if (!S_ISDIR(inode->i_mode))
1240 nfs_set_inode_stale(inode);
1241 else
1242 nfs_zap_caches(inode);
1243 }
1244 goto out;
1245 }
1246
1247 status = nfs_refresh_inode(inode, fattr);
1248 if (status) {
1249 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1250 inode->i_sb->s_id,
1251 (unsigned long long)NFS_FILEID(inode), status);
1252 goto out;
1253 }
1254
1255 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1256 nfs_zap_acl_cache(inode);
1257
1258 nfs_setsecurity(inode, fattr);
1259
1260 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1261 inode->i_sb->s_id,
1262 (unsigned long long)NFS_FILEID(inode));
1263
1264 out:
1265 nfs_free_fattr(fattr);
1266 trace_nfs_revalidate_inode_exit(inode, status);
1267 return status;
1268 }
1269
nfs_attribute_cache_expired(struct inode * inode)1270 int nfs_attribute_cache_expired(struct inode *inode)
1271 {
1272 if (nfs_have_delegated_attributes(inode))
1273 return 0;
1274 return nfs_attribute_timeout(inode);
1275 }
1276
1277 /**
1278 * nfs_revalidate_inode - Revalidate the inode attributes
1279 * @inode: pointer to inode struct
1280 * @flags: cache flags to check
1281 *
1282 * Updates inode attribute information by retrieving the data from the server.
1283 */
nfs_revalidate_inode(struct inode * inode,unsigned long flags)1284 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1285 {
1286 if (!nfs_check_cache_invalid(inode, flags))
1287 return NFS_STALE(inode) ? -ESTALE : 0;
1288 return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1289 }
1290 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1291
nfs_invalidate_mapping(struct inode * inode,struct address_space * mapping)1292 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1293 {
1294 int ret;
1295
1296 nfs_fscache_invalidate(inode, 0);
1297 if (mapping->nrpages != 0) {
1298 if (S_ISREG(inode->i_mode)) {
1299 ret = nfs_sync_mapping(mapping);
1300 if (ret < 0)
1301 return ret;
1302 }
1303 ret = invalidate_inode_pages2(mapping);
1304 if (ret < 0)
1305 return ret;
1306 }
1307 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1308
1309 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1310 inode->i_sb->s_id,
1311 (unsigned long long)NFS_FILEID(inode));
1312 return 0;
1313 }
1314
1315 /**
1316 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1317 * @mapping: pointer to mapping
1318 *
1319 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1320 */
nfs_clear_invalid_mapping(struct address_space * mapping)1321 int nfs_clear_invalid_mapping(struct address_space *mapping)
1322 {
1323 struct inode *inode = mapping->host;
1324 struct nfs_inode *nfsi = NFS_I(inode);
1325 unsigned long *bitlock = &nfsi->flags;
1326 int ret = 0;
1327
1328 /*
1329 * We must clear NFS_INO_INVALID_DATA first to ensure that
1330 * invalidations that come in while we're shooting down the mappings
1331 * are respected. But, that leaves a race window where one revalidator
1332 * can clear the flag, and then another checks it before the mapping
1333 * gets invalidated. Fix that by serializing access to this part of
1334 * the function.
1335 *
1336 * At the same time, we need to allow other tasks to see whether we
1337 * might be in the middle of invalidating the pages, so we only set
1338 * the bit lock here if it looks like we're going to be doing that.
1339 */
1340 for (;;) {
1341 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1342 nfs_wait_bit_killable,
1343 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1344 if (ret)
1345 goto out;
1346 smp_rmb(); /* pairs with smp_wmb() below */
1347 if (test_bit(NFS_INO_INVALIDATING, bitlock))
1348 continue;
1349 /* pairs with nfs_set_cache_invalid()'s smp_store_release() */
1350 if (!(smp_load_acquire(&nfsi->cache_validity) & NFS_INO_INVALID_DATA))
1351 goto out;
1352 /* Slow-path that double-checks with spinlock held */
1353 spin_lock(&inode->i_lock);
1354 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1355 spin_unlock(&inode->i_lock);
1356 continue;
1357 }
1358 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1359 break;
1360 spin_unlock(&inode->i_lock);
1361 goto out;
1362 }
1363
1364 set_bit(NFS_INO_INVALIDATING, bitlock);
1365 smp_wmb();
1366 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1367 nfs_ooo_clear(nfsi);
1368 spin_unlock(&inode->i_lock);
1369 trace_nfs_invalidate_mapping_enter(inode);
1370 ret = nfs_invalidate_mapping(inode, mapping);
1371 trace_nfs_invalidate_mapping_exit(inode, ret);
1372
1373 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1374 smp_mb__after_atomic();
1375 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1376 out:
1377 return ret;
1378 }
1379
nfs_mapping_need_revalidate_inode(struct inode * inode)1380 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1381 {
1382 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1383 NFS_STALE(inode);
1384 }
1385
nfs_revalidate_mapping_rcu(struct inode * inode)1386 int nfs_revalidate_mapping_rcu(struct inode *inode)
1387 {
1388 struct nfs_inode *nfsi = NFS_I(inode);
1389 unsigned long *bitlock = &nfsi->flags;
1390 int ret = 0;
1391
1392 if (IS_SWAPFILE(inode))
1393 goto out;
1394 if (nfs_mapping_need_revalidate_inode(inode)) {
1395 ret = -ECHILD;
1396 goto out;
1397 }
1398 spin_lock(&inode->i_lock);
1399 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1400 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1401 ret = -ECHILD;
1402 spin_unlock(&inode->i_lock);
1403 out:
1404 return ret;
1405 }
1406
1407 /**
1408 * nfs_revalidate_mapping - Revalidate the pagecache
1409 * @inode: pointer to host inode
1410 * @mapping: pointer to mapping
1411 */
nfs_revalidate_mapping(struct inode * inode,struct address_space * mapping)1412 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1413 {
1414 /* swapfiles are not supposed to be shared. */
1415 if (IS_SWAPFILE(inode))
1416 return 0;
1417
1418 if (nfs_mapping_need_revalidate_inode(inode)) {
1419 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1420 if (ret < 0)
1421 return ret;
1422 }
1423
1424 return nfs_clear_invalid_mapping(mapping);
1425 }
1426
nfs_file_has_writers(struct nfs_inode * nfsi)1427 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1428 {
1429 struct inode *inode = &nfsi->vfs_inode;
1430
1431 if (!S_ISREG(inode->i_mode))
1432 return false;
1433 if (list_empty(&nfsi->open_files))
1434 return false;
1435 return inode_is_open_for_write(inode);
1436 }
1437
nfs_file_has_buffered_writers(struct nfs_inode * nfsi)1438 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1439 {
1440 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1441 }
1442
nfs_wcc_update_inode(struct inode * inode,struct nfs_fattr * fattr)1443 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1444 {
1445 struct timespec64 ts;
1446
1447 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1448 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1449 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1450 inode_set_iversion_raw(inode, fattr->change_attr);
1451 if (S_ISDIR(inode->i_mode))
1452 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1453 else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1454 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1455 }
1456 /* If we have atomic WCC data, we may update some attributes */
1457 ts = inode_get_ctime(inode);
1458 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1459 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1460 && timespec64_equal(&ts, &fattr->pre_ctime)) {
1461 inode_set_ctime_to_ts(inode, fattr->ctime);
1462 }
1463
1464 ts = inode->i_mtime;
1465 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1466 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1467 && timespec64_equal(&ts, &fattr->pre_mtime)) {
1468 inode->i_mtime = fattr->mtime;
1469 }
1470 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1471 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1472 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1473 && !nfs_have_writebacks(inode)) {
1474 trace_nfs_size_wcc(inode, fattr->size);
1475 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1476 }
1477 }
1478
1479 /**
1480 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1481 * @inode: pointer to inode
1482 * @fattr: updated attributes
1483 *
1484 * Verifies the attribute cache. If we have just changed the attributes,
1485 * so that fattr carries weak cache consistency data, then it may
1486 * also update the ctime/mtime/change_attribute.
1487 */
nfs_check_inode_attributes(struct inode * inode,struct nfs_fattr * fattr)1488 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1489 {
1490 struct nfs_inode *nfsi = NFS_I(inode);
1491 loff_t cur_size, new_isize;
1492 unsigned long invalid = 0;
1493 struct timespec64 ts;
1494
1495 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1496 return 0;
1497
1498 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1499 /* Only a mounted-on-fileid? Just exit */
1500 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1501 return 0;
1502 /* Has the inode gone and changed behind our back? */
1503 } else if (nfsi->fileid != fattr->fileid) {
1504 /* Is this perhaps the mounted-on fileid? */
1505 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1506 nfsi->fileid == fattr->mounted_on_fileid)
1507 return 0;
1508 return -ESTALE;
1509 }
1510 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1511 return -ESTALE;
1512
1513
1514 if (!nfs_file_has_buffered_writers(nfsi)) {
1515 /* Verify a few of the more important attributes */
1516 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1517 invalid |= NFS_INO_INVALID_CHANGE;
1518
1519 ts = inode->i_mtime;
1520 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1521 invalid |= NFS_INO_INVALID_MTIME;
1522
1523 ts = inode_get_ctime(inode);
1524 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1525 invalid |= NFS_INO_INVALID_CTIME;
1526
1527 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1528 cur_size = i_size_read(inode);
1529 new_isize = nfs_size_to_loff_t(fattr->size);
1530 if (cur_size != new_isize)
1531 invalid |= NFS_INO_INVALID_SIZE;
1532 }
1533 }
1534
1535 /* Have any file permissions changed? */
1536 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1537 invalid |= NFS_INO_INVALID_MODE;
1538 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1539 invalid |= NFS_INO_INVALID_OTHER;
1540 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1541 invalid |= NFS_INO_INVALID_OTHER;
1542
1543 /* Has the link count changed? */
1544 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1545 invalid |= NFS_INO_INVALID_NLINK;
1546
1547 ts = inode->i_atime;
1548 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1549 invalid |= NFS_INO_INVALID_ATIME;
1550
1551 if (invalid != 0)
1552 nfs_set_cache_invalid(inode, invalid);
1553
1554 nfsi->read_cache_jiffies = fattr->time_start;
1555 return 0;
1556 }
1557
1558 static atomic_long_t nfs_attr_generation_counter;
1559
nfs_read_attr_generation_counter(void)1560 static unsigned long nfs_read_attr_generation_counter(void)
1561 {
1562 return atomic_long_read(&nfs_attr_generation_counter);
1563 }
1564
nfs_inc_attr_generation_counter(void)1565 unsigned long nfs_inc_attr_generation_counter(void)
1566 {
1567 return atomic_long_inc_return(&nfs_attr_generation_counter);
1568 }
1569 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1570
nfs_fattr_init(struct nfs_fattr * fattr)1571 void nfs_fattr_init(struct nfs_fattr *fattr)
1572 {
1573 fattr->valid = 0;
1574 fattr->time_start = jiffies;
1575 fattr->gencount = nfs_inc_attr_generation_counter();
1576 fattr->owner_name = NULL;
1577 fattr->group_name = NULL;
1578 fattr->mdsthreshold = NULL;
1579 }
1580 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1581
1582 /**
1583 * nfs_fattr_set_barrier
1584 * @fattr: attributes
1585 *
1586 * Used to set a barrier after an attribute was updated. This
1587 * barrier ensures that older attributes from RPC calls that may
1588 * have raced with our update cannot clobber these new values.
1589 * Note that you are still responsible for ensuring that other
1590 * operations which change the attribute on the server do not
1591 * collide.
1592 */
nfs_fattr_set_barrier(struct nfs_fattr * fattr)1593 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1594 {
1595 fattr->gencount = nfs_inc_attr_generation_counter();
1596 }
1597
nfs_alloc_fattr(void)1598 struct nfs_fattr *nfs_alloc_fattr(void)
1599 {
1600 struct nfs_fattr *fattr;
1601
1602 fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1603 if (fattr != NULL) {
1604 nfs_fattr_init(fattr);
1605 fattr->label = NULL;
1606 }
1607 return fattr;
1608 }
1609 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1610
nfs_alloc_fattr_with_label(struct nfs_server * server)1611 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1612 {
1613 struct nfs_fattr *fattr = nfs_alloc_fattr();
1614
1615 if (!fattr)
1616 return NULL;
1617
1618 fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1619 if (IS_ERR(fattr->label)) {
1620 kfree(fattr);
1621 return NULL;
1622 }
1623
1624 return fattr;
1625 }
1626 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1627
nfs_alloc_fhandle(void)1628 struct nfs_fh *nfs_alloc_fhandle(void)
1629 {
1630 struct nfs_fh *fh;
1631
1632 fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1633 if (fh != NULL)
1634 fh->size = 0;
1635 return fh;
1636 }
1637 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1638
1639 #ifdef NFS_DEBUG
1640 /*
1641 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1642 * in the same way that wireshark does
1643 *
1644 * @fh: file handle
1645 *
1646 * For debugging only.
1647 */
_nfs_display_fhandle_hash(const struct nfs_fh * fh)1648 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1649 {
1650 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1651 * not on the result */
1652 return nfs_fhandle_hash(fh);
1653 }
1654 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1655
1656 /*
1657 * _nfs_display_fhandle - display an NFS file handle on the console
1658 *
1659 * @fh: file handle to display
1660 * @caption: display caption
1661 *
1662 * For debugging only.
1663 */
_nfs_display_fhandle(const struct nfs_fh * fh,const char * caption)1664 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1665 {
1666 unsigned short i;
1667
1668 if (fh == NULL || fh->size == 0) {
1669 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1670 return;
1671 }
1672
1673 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1674 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1675 for (i = 0; i < fh->size; i += 16) {
1676 __be32 *pos = (__be32 *)&fh->data[i];
1677
1678 switch ((fh->size - i - 1) >> 2) {
1679 case 0:
1680 printk(KERN_DEFAULT " %08x\n",
1681 be32_to_cpup(pos));
1682 break;
1683 case 1:
1684 printk(KERN_DEFAULT " %08x %08x\n",
1685 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1686 break;
1687 case 2:
1688 printk(KERN_DEFAULT " %08x %08x %08x\n",
1689 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1690 be32_to_cpup(pos + 2));
1691 break;
1692 default:
1693 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1694 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1695 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1696 }
1697 }
1698 }
1699 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1700 #endif
1701
1702 /**
1703 * nfs_inode_attrs_cmp_generic - 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 * Note also the check for wraparound of 'attr_gencount'
1710 *
1711 * The function returns '1' if it thinks the attributes in @fattr are
1712 * more recent than the ones cached in @inode. Otherwise it returns
1713 * the value '0'.
1714 */
nfs_inode_attrs_cmp_generic(const struct nfs_fattr * fattr,const struct inode * inode)1715 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1716 const struct inode *inode)
1717 {
1718 unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1719
1720 return (long)(fattr->gencount - attr_gencount) > 0 ||
1721 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1722 }
1723
1724 /**
1725 * nfs_inode_attrs_cmp_monotonic - compare attributes
1726 * @fattr: attributes
1727 * @inode: pointer to inode
1728 *
1729 * Attempt to divine whether or not an RPC call reply carrying stale
1730 * attributes got scheduled after another call carrying updated ones.
1731 *
1732 * We assume that the server observes monotonic semantics for
1733 * the change attribute, so a larger value means that the attributes in
1734 * @fattr are more recent, in which case the function returns the
1735 * value '1'.
1736 * A return value of '0' indicates no measurable change
1737 * A return value of '-1' means that the attributes in @inode are
1738 * more recent.
1739 */
nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1740 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1741 const struct inode *inode)
1742 {
1743 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1744 if (diff > 0)
1745 return 1;
1746 return diff == 0 ? 0 : -1;
1747 }
1748
1749 /**
1750 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1751 * @fattr: attributes
1752 * @inode: pointer to inode
1753 *
1754 * Attempt to divine whether or not an RPC call reply carrying stale
1755 * attributes got scheduled after another call carrying updated ones.
1756 *
1757 * We assume that the server observes strictly monotonic semantics for
1758 * the change attribute, so a larger value means that the attributes in
1759 * @fattr are more recent, in which case the function returns the
1760 * value '1'.
1761 * A return value of '-1' means that the attributes in @inode are
1762 * more recent or unchanged.
1763 */
nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr * fattr,const struct inode * inode)1764 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1765 const struct inode *inode)
1766 {
1767 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1768 }
1769
1770 /**
1771 * nfs_inode_attrs_cmp - compare attributes
1772 * @fattr: attributes
1773 * @inode: pointer to inode
1774 *
1775 * This function returns '1' if it thinks the attributes in @fattr are
1776 * more recent than the ones cached in @inode. It returns '-1' if
1777 * the attributes in @inode are more recent than the ones in @fattr,
1778 * and it returns 0 if not sure.
1779 */
nfs_inode_attrs_cmp(const struct nfs_fattr * fattr,const struct inode * inode)1780 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1781 const struct inode *inode)
1782 {
1783 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1784 return 1;
1785 switch (NFS_SERVER(inode)->change_attr_type) {
1786 case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1787 break;
1788 case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1789 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1790 break;
1791 return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1792 default:
1793 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1794 break;
1795 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1796 }
1797 return 0;
1798 }
1799
1800 /**
1801 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1802 * @fattr: attributes
1803 * @inode: pointer to inode
1804 *
1805 * Returns '1' if the last attribute update left the inode cached
1806 * attributes in a partially unrevalidated state, and @fattr
1807 * matches the change attribute of that partial update.
1808 * Otherwise returns '0'.
1809 */
nfs_inode_finish_partial_attr_update(const struct nfs_fattr * fattr,const struct inode * inode)1810 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1811 const struct inode *inode)
1812 {
1813 const unsigned long check_valid =
1814 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1815 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1816 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1817 NFS_INO_INVALID_NLINK;
1818 unsigned long cache_validity = NFS_I(inode)->cache_validity;
1819 enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1820
1821 if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1822 !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1823 (cache_validity & check_valid) != 0 &&
1824 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1825 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1826 return 1;
1827 return 0;
1828 }
1829
nfs_ooo_merge(struct nfs_inode * nfsi,u64 start,u64 end)1830 static void nfs_ooo_merge(struct nfs_inode *nfsi,
1831 u64 start, u64 end)
1832 {
1833 int i, cnt;
1834
1835 if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)
1836 /* No point merging anything */
1837 return;
1838
1839 if (!nfsi->ooo) {
1840 nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC);
1841 if (!nfsi->ooo) {
1842 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1843 return;
1844 }
1845 nfsi->ooo->cnt = 0;
1846 }
1847
1848 /* add this range, merging if possible */
1849 cnt = nfsi->ooo->cnt;
1850 for (i = 0; i < cnt; i++) {
1851 if (end == nfsi->ooo->gap[i].start)
1852 end = nfsi->ooo->gap[i].end;
1853 else if (start == nfsi->ooo->gap[i].end)
1854 start = nfsi->ooo->gap[i].start;
1855 else
1856 continue;
1857 /* Remove 'i' from table and loop to insert the new range */
1858 cnt -= 1;
1859 nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt];
1860 i = -1;
1861 }
1862 if (start != end) {
1863 if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) {
1864 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1865 kfree(nfsi->ooo);
1866 nfsi->ooo = NULL;
1867 return;
1868 }
1869 nfsi->ooo->gap[cnt].start = start;
1870 nfsi->ooo->gap[cnt].end = end;
1871 cnt += 1;
1872 }
1873 nfsi->ooo->cnt = cnt;
1874 }
1875
nfs_ooo_record(struct nfs_inode * nfsi,struct nfs_fattr * fattr)1876 static void nfs_ooo_record(struct nfs_inode *nfsi,
1877 struct nfs_fattr *fattr)
1878 {
1879 /* This reply was out-of-order, so record in the
1880 * pre/post change id, possibly cancelling
1881 * gaps created when iversion was jumpped forward.
1882 */
1883 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) &&
1884 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE))
1885 nfs_ooo_merge(nfsi,
1886 fattr->change_attr,
1887 fattr->pre_change_attr);
1888 }
1889
nfs_refresh_inode_locked(struct inode * inode,struct nfs_fattr * fattr)1890 static int nfs_refresh_inode_locked(struct inode *inode,
1891 struct nfs_fattr *fattr)
1892 {
1893 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1894 int ret = 0;
1895
1896 trace_nfs_refresh_inode_enter(inode);
1897
1898 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1899 ret = nfs_update_inode(inode, fattr);
1900 else {
1901 nfs_ooo_record(NFS_I(inode), fattr);
1902
1903 if (attr_cmp == 0)
1904 ret = nfs_check_inode_attributes(inode, fattr);
1905 }
1906
1907 trace_nfs_refresh_inode_exit(inode, ret);
1908 return ret;
1909 }
1910
1911 /**
1912 * nfs_refresh_inode - try to update the inode attribute cache
1913 * @inode: pointer to inode
1914 * @fattr: updated attributes
1915 *
1916 * Check that an RPC call that returned attributes has not overlapped with
1917 * other recent updates of the inode metadata, then decide whether it is
1918 * safe to do a full update of the inode attributes, or whether just to
1919 * call nfs_check_inode_attributes.
1920 */
nfs_refresh_inode(struct inode * inode,struct nfs_fattr * fattr)1921 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1922 {
1923 int status;
1924
1925 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1926 return 0;
1927 spin_lock(&inode->i_lock);
1928 status = nfs_refresh_inode_locked(inode, fattr);
1929 spin_unlock(&inode->i_lock);
1930
1931 return status;
1932 }
1933 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1934
nfs_post_op_update_inode_locked(struct inode * inode,struct nfs_fattr * fattr,unsigned int invalid)1935 static int nfs_post_op_update_inode_locked(struct inode *inode,
1936 struct nfs_fattr *fattr, unsigned int invalid)
1937 {
1938 if (S_ISDIR(inode->i_mode))
1939 invalid |= NFS_INO_INVALID_DATA;
1940 nfs_set_cache_invalid(inode, invalid);
1941 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1942 return 0;
1943 return nfs_refresh_inode_locked(inode, fattr);
1944 }
1945
1946 /**
1947 * nfs_post_op_update_inode - try to update the inode attribute cache
1948 * @inode: pointer to inode
1949 * @fattr: updated attributes
1950 *
1951 * After an operation that has changed the inode metadata, mark the
1952 * attribute cache as being invalid, then try to update it.
1953 *
1954 * NB: if the server didn't return any post op attributes, this
1955 * function will force the retrieval of attributes before the next
1956 * NFS request. Thus it should be used only for operations that
1957 * are expected to change one or more attributes, to avoid
1958 * unnecessary NFS requests and trips through nfs_update_inode().
1959 */
nfs_post_op_update_inode(struct inode * inode,struct nfs_fattr * fattr)1960 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1961 {
1962 int status;
1963
1964 spin_lock(&inode->i_lock);
1965 nfs_fattr_set_barrier(fattr);
1966 status = nfs_post_op_update_inode_locked(inode, fattr,
1967 NFS_INO_INVALID_CHANGE
1968 | NFS_INO_INVALID_CTIME
1969 | NFS_INO_REVAL_FORCED);
1970 spin_unlock(&inode->i_lock);
1971
1972 return status;
1973 }
1974 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1975
1976 /**
1977 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1978 * @inode: pointer to inode
1979 * @fattr: updated attributes
1980 *
1981 * After an operation that has changed the inode metadata, mark the
1982 * attribute cache as being invalid, then try to update it. Fake up
1983 * weak cache consistency data, if none exist.
1984 *
1985 * This function is mainly designed to be used by the ->write_done() functions.
1986 */
nfs_post_op_update_inode_force_wcc_locked(struct inode * inode,struct nfs_fattr * fattr)1987 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1988 {
1989 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1990 int status;
1991
1992 /* Don't do a WCC update if these attributes are already stale */
1993 if (attr_cmp < 0)
1994 return 0;
1995 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
1996 /* Record the pre/post change info before clearing PRECHANGE */
1997 nfs_ooo_record(NFS_I(inode), fattr);
1998 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1999 | NFS_ATTR_FATTR_PRESIZE
2000 | NFS_ATTR_FATTR_PREMTIME
2001 | NFS_ATTR_FATTR_PRECTIME);
2002 goto out_noforce;
2003 }
2004 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
2005 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
2006 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
2007 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
2008 }
2009 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
2010 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
2011 fattr->pre_ctime = inode_get_ctime(inode);
2012 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
2013 }
2014 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
2015 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
2016 fattr->pre_mtime = inode->i_mtime;
2017 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
2018 }
2019 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
2020 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
2021 fattr->pre_size = i_size_read(inode);
2022 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
2023 }
2024 out_noforce:
2025 status = nfs_post_op_update_inode_locked(inode, fattr,
2026 NFS_INO_INVALID_CHANGE
2027 | NFS_INO_INVALID_CTIME
2028 | NFS_INO_INVALID_MTIME
2029 | NFS_INO_INVALID_BLOCKS);
2030 return status;
2031 }
2032
2033 /**
2034 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
2035 * @inode: pointer to inode
2036 * @fattr: updated attributes
2037 *
2038 * After an operation that has changed the inode metadata, mark the
2039 * attribute cache as being invalid, then try to update it. Fake up
2040 * weak cache consistency data, if none exist.
2041 *
2042 * This function is mainly designed to be used by the ->write_done() functions.
2043 */
nfs_post_op_update_inode_force_wcc(struct inode * inode,struct nfs_fattr * fattr)2044 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
2045 {
2046 int status;
2047
2048 spin_lock(&inode->i_lock);
2049 nfs_fattr_set_barrier(fattr);
2050 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
2051 spin_unlock(&inode->i_lock);
2052 return status;
2053 }
2054 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
2055
2056
2057 /*
2058 * Many nfs protocol calls return the new file attributes after
2059 * an operation. Here we update the inode to reflect the state
2060 * of the server's inode.
2061 *
2062 * This is a bit tricky because we have to make sure all dirty pages
2063 * have been sent off to the server before calling invalidate_inode_pages.
2064 * To make sure no other process adds more write requests while we try
2065 * our best to flush them, we make them sleep during the attribute refresh.
2066 *
2067 * A very similar scenario holds for the dir cache.
2068 */
nfs_update_inode(struct inode * inode,struct nfs_fattr * fattr)2069 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2070 {
2071 struct nfs_server *server = NFS_SERVER(inode);
2072 struct nfs_inode *nfsi = NFS_I(inode);
2073 loff_t cur_isize, new_isize;
2074 u64 fattr_supported = server->fattr_valid;
2075 unsigned long invalid = 0;
2076 unsigned long now = jiffies;
2077 unsigned long save_cache_validity;
2078 bool have_writers = nfs_file_has_buffered_writers(nfsi);
2079 bool cache_revalidated = true;
2080 bool attr_changed = false;
2081 bool have_delegation;
2082
2083 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
2084 __func__, inode->i_sb->s_id, inode->i_ino,
2085 nfs_display_fhandle_hash(NFS_FH(inode)),
2086 atomic_read(&inode->i_count), fattr->valid);
2087
2088 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2089 /* Only a mounted-on-fileid? Just exit */
2090 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2091 return 0;
2092 /* Has the inode gone and changed behind our back? */
2093 } else if (nfsi->fileid != fattr->fileid) {
2094 /* Is this perhaps the mounted-on fileid? */
2095 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2096 nfsi->fileid == fattr->mounted_on_fileid)
2097 return 0;
2098 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2099 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2100 NFS_SERVER(inode)->nfs_client->cl_hostname,
2101 inode->i_sb->s_id, (long long)nfsi->fileid,
2102 (long long)fattr->fileid);
2103 goto out_err;
2104 }
2105
2106 /*
2107 * Make sure the inode's type hasn't changed.
2108 */
2109 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2110 /*
2111 * Big trouble! The inode has become a different object.
2112 */
2113 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2114 __func__, inode->i_ino, inode->i_mode, fattr->mode);
2115 goto out_err;
2116 }
2117
2118 /* Update the fsid? */
2119 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2120 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2121 !IS_AUTOMOUNT(inode))
2122 server->fsid = fattr->fsid;
2123
2124 /* Save the delegation state before clearing cache_validity */
2125 have_delegation = nfs_have_delegated_attributes(inode);
2126
2127 /*
2128 * Update the read time so we don't revalidate too often.
2129 */
2130 nfsi->read_cache_jiffies = fattr->time_start;
2131
2132 save_cache_validity = nfsi->cache_validity;
2133 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2134 | NFS_INO_INVALID_ATIME
2135 | NFS_INO_REVAL_FORCED
2136 | NFS_INO_INVALID_BLOCKS);
2137
2138 /* Do atomic weak cache consistency updates */
2139 nfs_wcc_update_inode(inode, fattr);
2140
2141 if (pnfs_layoutcommit_outstanding(inode)) {
2142 nfsi->cache_validity |=
2143 save_cache_validity &
2144 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2145 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2146 NFS_INO_INVALID_BLOCKS);
2147 cache_revalidated = false;
2148 }
2149
2150 /* More cache consistency checks */
2151 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2152 if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 &&
2153 nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) {
2154 /* There is one remaining gap that hasn't been
2155 * merged into iversion - do that now.
2156 */
2157 inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start);
2158 kfree(nfsi->ooo);
2159 nfsi->ooo = NULL;
2160 }
2161 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2162 /* Could it be a race with writeback? */
2163 if (!(have_writers || have_delegation)) {
2164 invalid |= NFS_INO_INVALID_DATA
2165 | NFS_INO_INVALID_ACCESS
2166 | NFS_INO_INVALID_ACL
2167 | NFS_INO_INVALID_XATTR;
2168 /* Force revalidate of all attributes */
2169 save_cache_validity |= NFS_INO_INVALID_CTIME
2170 | NFS_INO_INVALID_MTIME
2171 | NFS_INO_INVALID_SIZE
2172 | NFS_INO_INVALID_BLOCKS
2173 | NFS_INO_INVALID_NLINK
2174 | NFS_INO_INVALID_MODE
2175 | NFS_INO_INVALID_OTHER;
2176 if (S_ISDIR(inode->i_mode))
2177 nfs_force_lookup_revalidate(inode);
2178 attr_changed = true;
2179 dprintk("NFS: change_attr change on server for file %s/%ld\n",
2180 inode->i_sb->s_id,
2181 inode->i_ino);
2182 } else if (!have_delegation) {
2183 nfs_ooo_record(nfsi, fattr);
2184 nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode),
2185 fattr->change_attr);
2186 }
2187 inode_set_iversion_raw(inode, fattr->change_attr);
2188 }
2189 } else {
2190 nfsi->cache_validity |=
2191 save_cache_validity & NFS_INO_INVALID_CHANGE;
2192 if (!have_delegation ||
2193 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2194 cache_revalidated = false;
2195 }
2196
2197 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2198 inode->i_mtime = fattr->mtime;
2199 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2200 nfsi->cache_validity |=
2201 save_cache_validity & NFS_INO_INVALID_MTIME;
2202
2203 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2204 inode_set_ctime_to_ts(inode, fattr->ctime);
2205 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2206 nfsi->cache_validity |=
2207 save_cache_validity & NFS_INO_INVALID_CTIME;
2208
2209 /* Check if our cached file size is stale */
2210 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2211 new_isize = nfs_size_to_loff_t(fattr->size);
2212 cur_isize = i_size_read(inode);
2213 if (new_isize != cur_isize && !have_delegation) {
2214 /* Do we perhaps have any outstanding writes, or has
2215 * the file grown beyond our last write? */
2216 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2217 trace_nfs_size_update(inode, new_isize);
2218 i_size_write(inode, new_isize);
2219 if (!have_writers)
2220 invalid |= NFS_INO_INVALID_DATA;
2221 }
2222 }
2223 if (new_isize == 0 &&
2224 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2225 NFS_ATTR_FATTR_BLOCKS_USED))) {
2226 fattr->du.nfs3.used = 0;
2227 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2228 }
2229 } else
2230 nfsi->cache_validity |=
2231 save_cache_validity & NFS_INO_INVALID_SIZE;
2232
2233 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2234 inode->i_atime = fattr->atime;
2235 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2236 nfsi->cache_validity |=
2237 save_cache_validity & NFS_INO_INVALID_ATIME;
2238
2239 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2240 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2241 umode_t newmode = inode->i_mode & S_IFMT;
2242 newmode |= fattr->mode & S_IALLUGO;
2243 inode->i_mode = newmode;
2244 invalid |= NFS_INO_INVALID_ACCESS
2245 | NFS_INO_INVALID_ACL;
2246 }
2247 } else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2248 nfsi->cache_validity |=
2249 save_cache_validity & NFS_INO_INVALID_MODE;
2250
2251 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2252 if (!uid_eq(inode->i_uid, fattr->uid)) {
2253 invalid |= NFS_INO_INVALID_ACCESS
2254 | NFS_INO_INVALID_ACL;
2255 inode->i_uid = fattr->uid;
2256 }
2257 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2258 nfsi->cache_validity |=
2259 save_cache_validity & NFS_INO_INVALID_OTHER;
2260
2261 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2262 if (!gid_eq(inode->i_gid, fattr->gid)) {
2263 invalid |= NFS_INO_INVALID_ACCESS
2264 | NFS_INO_INVALID_ACL;
2265 inode->i_gid = fattr->gid;
2266 }
2267 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2268 nfsi->cache_validity |=
2269 save_cache_validity & NFS_INO_INVALID_OTHER;
2270
2271 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2272 if (inode->i_nlink != fattr->nlink)
2273 set_nlink(inode, fattr->nlink);
2274 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2275 nfsi->cache_validity |=
2276 save_cache_validity & NFS_INO_INVALID_NLINK;
2277
2278 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2279 /*
2280 * report the blocks in 512byte units
2281 */
2282 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2283 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2284 nfsi->cache_validity |=
2285 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2286
2287 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2288 inode->i_blocks = fattr->du.nfs2.blocks;
2289 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2290 nfsi->cache_validity |=
2291 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2292
2293 /* Update attrtimeo value if we're out of the unstable period */
2294 if (attr_changed) {
2295 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2296 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2297 nfsi->attrtimeo_timestamp = now;
2298 /* Set barrier to be more recent than all outstanding updates */
2299 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2300 } else {
2301 if (cache_revalidated) {
2302 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2303 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2304 nfsi->attrtimeo <<= 1;
2305 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2306 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2307 }
2308 nfsi->attrtimeo_timestamp = now;
2309 }
2310 /* Set the barrier to be more recent than this fattr */
2311 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2312 nfsi->attr_gencount = fattr->gencount;
2313 }
2314
2315 /* Don't invalidate the data if we were to blame */
2316 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2317 || S_ISLNK(inode->i_mode)))
2318 invalid &= ~NFS_INO_INVALID_DATA;
2319 nfs_set_cache_invalid(inode, invalid);
2320
2321 return 0;
2322 out_err:
2323 /*
2324 * No need to worry about unhashing the dentry, as the
2325 * lookup validation will know that the inode is bad.
2326 * (But we fall through to invalidate the caches.)
2327 */
2328 nfs_set_inode_stale_locked(inode);
2329 return -ESTALE;
2330 }
2331
nfs_alloc_inode(struct super_block * sb)2332 struct inode *nfs_alloc_inode(struct super_block *sb)
2333 {
2334 struct nfs_inode *nfsi;
2335 nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2336 if (!nfsi)
2337 return NULL;
2338 nfsi->flags = 0UL;
2339 nfsi->cache_validity = 0UL;
2340 nfsi->ooo = NULL;
2341 #if IS_ENABLED(CONFIG_NFS_V4)
2342 nfsi->nfs4_acl = NULL;
2343 #endif /* CONFIG_NFS_V4 */
2344 #ifdef CONFIG_NFS_V4_2
2345 nfsi->xattr_cache = NULL;
2346 #endif
2347 nfs_netfs_inode_init(nfsi);
2348
2349 return &nfsi->vfs_inode;
2350 }
2351 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2352
nfs_free_inode(struct inode * inode)2353 void nfs_free_inode(struct inode *inode)
2354 {
2355 kfree(NFS_I(inode)->ooo);
2356 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2357 }
2358 EXPORT_SYMBOL_GPL(nfs_free_inode);
2359
nfs4_init_once(struct nfs_inode * nfsi)2360 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2361 {
2362 #if IS_ENABLED(CONFIG_NFS_V4)
2363 INIT_LIST_HEAD(&nfsi->open_states);
2364 nfsi->delegation = NULL;
2365 init_rwsem(&nfsi->rwsem);
2366 nfsi->layout = NULL;
2367 #endif
2368 }
2369
init_once(void * foo)2370 static void init_once(void *foo)
2371 {
2372 struct nfs_inode *nfsi = foo;
2373
2374 inode_init_once(&nfsi->vfs_inode);
2375 INIT_LIST_HEAD(&nfsi->open_files);
2376 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2377 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2378 nfs4_init_once(nfsi);
2379 }
2380
nfs_init_inodecache(void)2381 static int __init nfs_init_inodecache(void)
2382 {
2383 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2384 sizeof(struct nfs_inode),
2385 0, (SLAB_RECLAIM_ACCOUNT|
2386 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2387 init_once);
2388 if (nfs_inode_cachep == NULL)
2389 return -ENOMEM;
2390
2391 return 0;
2392 }
2393
nfs_destroy_inodecache(void)2394 static void nfs_destroy_inodecache(void)
2395 {
2396 /*
2397 * Make sure all delayed rcu free inodes are flushed before we
2398 * destroy cache.
2399 */
2400 rcu_barrier();
2401 kmem_cache_destroy(nfs_inode_cachep);
2402 }
2403
2404 struct workqueue_struct *nfsiod_workqueue;
2405 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2406
2407 /*
2408 * start up the nfsiod workqueue
2409 */
nfsiod_start(void)2410 static int nfsiod_start(void)
2411 {
2412 struct workqueue_struct *wq;
2413 dprintk("RPC: creating workqueue nfsiod\n");
2414 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2415 if (wq == NULL)
2416 return -ENOMEM;
2417 nfsiod_workqueue = wq;
2418 return 0;
2419 }
2420
2421 /*
2422 * Destroy the nfsiod workqueue
2423 */
nfsiod_stop(void)2424 static void nfsiod_stop(void)
2425 {
2426 struct workqueue_struct *wq;
2427
2428 wq = nfsiod_workqueue;
2429 if (wq == NULL)
2430 return;
2431 nfsiod_workqueue = NULL;
2432 destroy_workqueue(wq);
2433 }
2434
2435 unsigned int nfs_net_id;
2436 EXPORT_SYMBOL_GPL(nfs_net_id);
2437
nfs_net_init(struct net * net)2438 static int nfs_net_init(struct net *net)
2439 {
2440 struct nfs_net *nn = net_generic(net, nfs_net_id);
2441
2442 nfs_clients_init(net);
2443
2444 if (!rpc_proc_register(net, &nn->rpcstats)) {
2445 nfs_clients_exit(net);
2446 return -ENOMEM;
2447 }
2448
2449 return nfs_fs_proc_net_init(net);
2450 }
2451
nfs_net_exit(struct net * net)2452 static void nfs_net_exit(struct net *net)
2453 {
2454 rpc_proc_unregister(net, "nfs");
2455 nfs_fs_proc_net_exit(net);
2456 nfs_clients_exit(net);
2457 }
2458
2459 static struct pernet_operations nfs_net_ops = {
2460 .init = nfs_net_init,
2461 .exit = nfs_net_exit,
2462 .id = &nfs_net_id,
2463 .size = sizeof(struct nfs_net),
2464 };
2465
2466 /*
2467 * Initialize NFS
2468 */
init_nfs_fs(void)2469 static int __init init_nfs_fs(void)
2470 {
2471 int err;
2472
2473 err = nfs_sysfs_init();
2474 if (err < 0)
2475 goto out10;
2476
2477 err = register_pernet_subsys(&nfs_net_ops);
2478 if (err < 0)
2479 goto out9;
2480
2481 err = nfsiod_start();
2482 if (err)
2483 goto out7;
2484
2485 err = nfs_fs_proc_init();
2486 if (err)
2487 goto out6;
2488
2489 err = nfs_init_nfspagecache();
2490 if (err)
2491 goto out5;
2492
2493 err = nfs_init_inodecache();
2494 if (err)
2495 goto out4;
2496
2497 err = nfs_init_readpagecache();
2498 if (err)
2499 goto out3;
2500
2501 err = nfs_init_writepagecache();
2502 if (err)
2503 goto out2;
2504
2505 err = nfs_init_directcache();
2506 if (err)
2507 goto out1;
2508
2509 err = register_nfs_fs();
2510 if (err)
2511 goto out0;
2512
2513 return 0;
2514 out0:
2515 nfs_destroy_directcache();
2516 out1:
2517 nfs_destroy_writepagecache();
2518 out2:
2519 nfs_destroy_readpagecache();
2520 out3:
2521 nfs_destroy_inodecache();
2522 out4:
2523 nfs_destroy_nfspagecache();
2524 out5:
2525 nfs_fs_proc_exit();
2526 out6:
2527 nfsiod_stop();
2528 out7:
2529 unregister_pernet_subsys(&nfs_net_ops);
2530 out9:
2531 nfs_sysfs_exit();
2532 out10:
2533 return err;
2534 }
2535
exit_nfs_fs(void)2536 static void __exit exit_nfs_fs(void)
2537 {
2538 nfs_destroy_directcache();
2539 nfs_destroy_writepagecache();
2540 nfs_destroy_readpagecache();
2541 nfs_destroy_inodecache();
2542 nfs_destroy_nfspagecache();
2543 unregister_pernet_subsys(&nfs_net_ops);
2544 unregister_nfs_fs();
2545 nfs_fs_proc_exit();
2546 nfsiod_stop();
2547 nfs_sysfs_exit();
2548 }
2549
2550 /* Not quite true; I just maintain it */
2551 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2552 MODULE_LICENSE("GPL");
2553 module_param(enable_ino64, bool, 0644);
2554
2555 module_init(init_nfs_fs)
2556 module_exit(exit_nfs_fs)
2557