]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/nfs/inode.c
scsi: qedf: Fix a potential NULL pointer dereference
[mirror_ubuntu-artful-kernel.git] / fs / nfs / inode.c
1 /*
2 * linux/fs/nfs/inode.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * nfs inode and superblock handling functions
7 *
8 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
10 *
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
13 *
14 */
15
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched/signal.h>
19 #include <linux/time.h>
20 #include <linux/kernel.h>
21 #include <linux/mm.h>
22 #include <linux/string.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/unistd.h>
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/metrics.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/vfs.h>
36 #include <linux/inet.h>
37 #include <linux/nfs_xdr.h>
38 #include <linux/slab.h>
39 #include <linux/compat.h>
40 #include <linux/freezer.h>
41
42 #include <linux/uaccess.h>
43
44 #include "nfs4_fs.h"
45 #include "callback.h"
46 #include "delegation.h"
47 #include "iostat.h"
48 #include "internal.h"
49 #include "fscache.h"
50 #include "pnfs.h"
51 #include "nfs.h"
52 #include "netns.h"
53
54 #include "nfstrace.h"
55
56 #define NFSDBG_FACILITY NFSDBG_VFS
57
58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
59
60 /* Default is to see 64-bit inode numbers */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
62
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
65
66 static struct kmem_cache * nfs_inode_cachep;
67
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
70 {
71 return nfs_fileid_to_ino_t(fattr->fileid);
72 }
73
74 static int nfs_wait_killable(int mode)
75 {
76 freezable_schedule_unsafe();
77 if (signal_pending_state(mode, current))
78 return -ERESTARTSYS;
79 return 0;
80 }
81
82 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
83 {
84 return nfs_wait_killable(mode);
85 }
86 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
87
88 int nfs_wait_atomic_killable(atomic_t *p)
89 {
90 return nfs_wait_killable(TASK_KILLABLE);
91 }
92
93 /**
94 * nfs_compat_user_ino64 - returns the user-visible inode number
95 * @fileid: 64-bit fileid
96 *
97 * This function returns a 32-bit inode number if the boot parameter
98 * nfs.enable_ino64 is zero.
99 */
100 u64 nfs_compat_user_ino64(u64 fileid)
101 {
102 #ifdef CONFIG_COMPAT
103 compat_ulong_t ino;
104 #else
105 unsigned long ino;
106 #endif
107
108 if (enable_ino64)
109 return fileid;
110 ino = fileid;
111 if (sizeof(ino) < sizeof(fileid))
112 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
113 return ino;
114 }
115
116 int nfs_drop_inode(struct inode *inode)
117 {
118 return NFS_STALE(inode) || generic_drop_inode(inode);
119 }
120 EXPORT_SYMBOL_GPL(nfs_drop_inode);
121
122 void nfs_clear_inode(struct inode *inode)
123 {
124 /*
125 * The following should never happen...
126 */
127 WARN_ON_ONCE(nfs_have_writebacks(inode));
128 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
129 nfs_zap_acl_cache(inode);
130 nfs_access_zap_cache(inode);
131 nfs_fscache_clear_inode(inode);
132 }
133 EXPORT_SYMBOL_GPL(nfs_clear_inode);
134
135 void nfs_evict_inode(struct inode *inode)
136 {
137 truncate_inode_pages_final(&inode->i_data);
138 clear_inode(inode);
139 nfs_clear_inode(inode);
140 }
141
142 int nfs_sync_inode(struct inode *inode)
143 {
144 inode_dio_wait(inode);
145 return nfs_wb_all(inode);
146 }
147 EXPORT_SYMBOL_GPL(nfs_sync_inode);
148
149 /**
150 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
151 */
152 int nfs_sync_mapping(struct address_space *mapping)
153 {
154 int ret = 0;
155
156 if (mapping->nrpages != 0) {
157 unmap_mapping_range(mapping, 0, 0, 0);
158 ret = nfs_wb_all(mapping->host);
159 }
160 return ret;
161 }
162
163 static int nfs_attribute_timeout(struct inode *inode)
164 {
165 struct nfs_inode *nfsi = NFS_I(inode);
166
167 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
168 }
169
170 static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
171 {
172 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
173
174 /* Special case for the pagecache or access cache */
175 if (flags == NFS_INO_REVAL_PAGECACHE &&
176 !(cache_validity & NFS_INO_REVAL_FORCED))
177 return false;
178 return (cache_validity & flags) != 0;
179 }
180
181 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
182 {
183 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
184
185 if ((cache_validity & flags) != 0)
186 return true;
187 if (nfs_attribute_timeout(inode))
188 return true;
189 return false;
190 }
191
192 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
193 {
194 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
195 return nfs_check_cache_invalid_delegated(inode, flags);
196
197 return nfs_check_cache_invalid_not_delegated(inode, flags);
198 }
199
200 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
201 {
202 struct nfs_inode *nfsi = NFS_I(inode);
203
204 if (inode->i_mapping->nrpages == 0)
205 flags &= ~NFS_INO_INVALID_DATA;
206 nfsi->cache_validity |= flags;
207 if (flags & NFS_INO_INVALID_DATA)
208 nfs_fscache_invalidate(inode);
209 }
210
211 /*
212 * Invalidate the local caches
213 */
214 static void nfs_zap_caches_locked(struct inode *inode)
215 {
216 struct nfs_inode *nfsi = NFS_I(inode);
217 int mode = inode->i_mode;
218
219 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
220
221 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
222 nfsi->attrtimeo_timestamp = jiffies;
223
224 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
225 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
226 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
227 | NFS_INO_INVALID_DATA
228 | NFS_INO_INVALID_ACCESS
229 | NFS_INO_INVALID_ACL
230 | NFS_INO_REVAL_PAGECACHE);
231 } else
232 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
233 | NFS_INO_INVALID_ACCESS
234 | NFS_INO_INVALID_ACL
235 | NFS_INO_REVAL_PAGECACHE);
236 nfs_zap_label_cache_locked(nfsi);
237 }
238
239 void nfs_zap_caches(struct inode *inode)
240 {
241 spin_lock(&inode->i_lock);
242 nfs_zap_caches_locked(inode);
243 spin_unlock(&inode->i_lock);
244 }
245
246 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
247 {
248 if (mapping->nrpages != 0) {
249 spin_lock(&inode->i_lock);
250 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
251 spin_unlock(&inode->i_lock);
252 }
253 }
254
255 void nfs_zap_acl_cache(struct inode *inode)
256 {
257 void (*clear_acl_cache)(struct inode *);
258
259 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
260 if (clear_acl_cache != NULL)
261 clear_acl_cache(inode);
262 spin_lock(&inode->i_lock);
263 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
264 spin_unlock(&inode->i_lock);
265 }
266 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
267
268 void nfs_invalidate_atime(struct inode *inode)
269 {
270 spin_lock(&inode->i_lock);
271 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
272 spin_unlock(&inode->i_lock);
273 }
274 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
275
276 /*
277 * Invalidate, but do not unhash, the inode.
278 * NB: must be called with inode->i_lock held!
279 */
280 static void nfs_invalidate_inode(struct inode *inode)
281 {
282 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
283 nfs_zap_caches_locked(inode);
284 }
285
286 struct nfs_find_desc {
287 struct nfs_fh *fh;
288 struct nfs_fattr *fattr;
289 };
290
291 /*
292 * In NFSv3 we can have 64bit inode numbers. In order to support
293 * this, and re-exported directories (also seen in NFSv2)
294 * we are forced to allow 2 different inodes to have the same
295 * i_ino.
296 */
297 static int
298 nfs_find_actor(struct inode *inode, void *opaque)
299 {
300 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
301 struct nfs_fh *fh = desc->fh;
302 struct nfs_fattr *fattr = desc->fattr;
303
304 if (NFS_FILEID(inode) != fattr->fileid)
305 return 0;
306 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
307 return 0;
308 if (nfs_compare_fh(NFS_FH(inode), fh))
309 return 0;
310 if (is_bad_inode(inode) || NFS_STALE(inode))
311 return 0;
312 return 1;
313 }
314
315 static int
316 nfs_init_locked(struct inode *inode, void *opaque)
317 {
318 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
319 struct nfs_fattr *fattr = desc->fattr;
320
321 set_nfs_fileid(inode, fattr->fileid);
322 inode->i_mode = fattr->mode;
323 nfs_copy_fh(NFS_FH(inode), desc->fh);
324 return 0;
325 }
326
327 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
328 static void nfs_clear_label_invalid(struct inode *inode)
329 {
330 spin_lock(&inode->i_lock);
331 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
332 spin_unlock(&inode->i_lock);
333 }
334
335 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
336 struct nfs4_label *label)
337 {
338 int error;
339
340 if (label == NULL)
341 return;
342
343 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
344 error = security_inode_notifysecctx(inode, label->label,
345 label->len);
346 if (error)
347 printk(KERN_ERR "%s() %s %d "
348 "security_inode_notifysecctx() %d\n",
349 __func__,
350 (char *)label->label,
351 label->len, error);
352 nfs_clear_label_invalid(inode);
353 }
354 }
355
356 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
357 {
358 struct nfs4_label *label = NULL;
359 int minor_version = server->nfs_client->cl_minorversion;
360
361 if (minor_version < 2)
362 return label;
363
364 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
365 return label;
366
367 label = kzalloc(sizeof(struct nfs4_label), flags);
368 if (label == NULL)
369 return ERR_PTR(-ENOMEM);
370
371 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
372 if (label->label == NULL) {
373 kfree(label);
374 return ERR_PTR(-ENOMEM);
375 }
376 label->len = NFS4_MAXLABELLEN;
377
378 return label;
379 }
380 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
381 #else
382 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
383 struct nfs4_label *label)
384 {
385 }
386 #endif
387 EXPORT_SYMBOL_GPL(nfs_setsecurity);
388
389 /*
390 * This is our front-end to iget that looks up inodes by file handle
391 * instead of inode number.
392 */
393 struct inode *
394 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
395 {
396 struct nfs_find_desc desc = {
397 .fh = fh,
398 .fattr = fattr
399 };
400 struct inode *inode = ERR_PTR(-ENOENT);
401 unsigned long hash;
402
403 nfs_attr_check_mountpoint(sb, fattr);
404
405 if (nfs_attr_use_mounted_on_fileid(fattr))
406 fattr->fileid = fattr->mounted_on_fileid;
407 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
408 goto out_no_inode;
409 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
410 goto out_no_inode;
411
412 hash = nfs_fattr_to_ino_t(fattr);
413
414 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
415 if (inode == NULL) {
416 inode = ERR_PTR(-ENOMEM);
417 goto out_no_inode;
418 }
419
420 if (inode->i_state & I_NEW) {
421 struct nfs_inode *nfsi = NFS_I(inode);
422 unsigned long now = jiffies;
423
424 /* We set i_ino for the few things that still rely on it,
425 * such as stat(2) */
426 inode->i_ino = hash;
427
428 /* We can't support update_atime(), since the server will reset it */
429 inode->i_flags |= S_NOATIME|S_NOCMTIME;
430 inode->i_mode = fattr->mode;
431 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
432 && nfs_server_capable(inode, NFS_CAP_MODE))
433 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
434 /* Why so? Because we want revalidate for devices/FIFOs, and
435 * that's precisely what we have in nfs_file_inode_operations.
436 */
437 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
438 if (S_ISREG(inode->i_mode)) {
439 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
440 inode->i_data.a_ops = &nfs_file_aops;
441 } else if (S_ISDIR(inode->i_mode)) {
442 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
443 inode->i_fop = &nfs_dir_operations;
444 inode->i_data.a_ops = &nfs_dir_aops;
445 /* Deal with crossing mountpoints */
446 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
447 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
448 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
449 inode->i_op = &nfs_referral_inode_operations;
450 else
451 inode->i_op = &nfs_mountpoint_inode_operations;
452 inode->i_fop = NULL;
453 inode->i_flags |= S_AUTOMOUNT;
454 }
455 } else if (S_ISLNK(inode->i_mode)) {
456 inode->i_op = &nfs_symlink_inode_operations;
457 inode_nohighmem(inode);
458 } else
459 init_special_inode(inode, inode->i_mode, fattr->rdev);
460
461 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
462 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
463 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
464 inode->i_version = 0;
465 inode->i_size = 0;
466 clear_nlink(inode);
467 inode->i_uid = make_kuid(&init_user_ns, -2);
468 inode->i_gid = make_kgid(&init_user_ns, -2);
469 inode->i_blocks = 0;
470 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
471 nfsi->write_io = 0;
472 nfsi->read_io = 0;
473
474 nfsi->read_cache_jiffies = fattr->time_start;
475 nfsi->attr_gencount = fattr->gencount;
476 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
477 inode->i_atime = fattr->atime;
478 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
479 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
480 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
481 inode->i_mtime = fattr->mtime;
482 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
483 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
484 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
485 inode->i_ctime = fattr->ctime;
486 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
487 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
488 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
489 inode->i_version = fattr->change_attr;
490 else
491 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
492 | NFS_INO_REVAL_PAGECACHE);
493 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
494 inode->i_size = nfs_size_to_loff_t(fattr->size);
495 else
496 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
497 | NFS_INO_REVAL_PAGECACHE);
498 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
499 set_nlink(inode, fattr->nlink);
500 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
501 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
502 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
503 inode->i_uid = fattr->uid;
504 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
505 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
506 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
507 inode->i_gid = fattr->gid;
508 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
509 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
510 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
511 inode->i_blocks = fattr->du.nfs2.blocks;
512 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
513 /*
514 * report the blocks in 512byte units
515 */
516 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
517 }
518
519 nfs_setsecurity(inode, fattr, label);
520
521 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
522 nfsi->attrtimeo_timestamp = now;
523 nfsi->access_cache = RB_ROOT;
524
525 nfs_fscache_init_inode(inode);
526
527 unlock_new_inode(inode);
528 } else
529 nfs_refresh_inode(inode, fattr);
530 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
531 inode->i_sb->s_id,
532 (unsigned long long)NFS_FILEID(inode),
533 nfs_display_fhandle_hash(fh),
534 atomic_read(&inode->i_count));
535
536 out:
537 return inode;
538
539 out_no_inode:
540 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
541 goto out;
542 }
543 EXPORT_SYMBOL_GPL(nfs_fhget);
544
545 #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)
546
547 int
548 nfs_setattr(struct dentry *dentry, struct iattr *attr)
549 {
550 struct inode *inode = d_inode(dentry);
551 struct nfs_fattr *fattr;
552 int error = 0;
553
554 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
555
556 /* skip mode change if it's just for clearing setuid/setgid */
557 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
558 attr->ia_valid &= ~ATTR_MODE;
559
560 if (attr->ia_valid & ATTR_SIZE) {
561 BUG_ON(!S_ISREG(inode->i_mode));
562
563 error = inode_newsize_ok(inode, attr->ia_size);
564 if (error)
565 return error;
566
567 if (attr->ia_size == i_size_read(inode))
568 attr->ia_valid &= ~ATTR_SIZE;
569 }
570
571 /* Optimization: if the end result is no change, don't RPC */
572 attr->ia_valid &= NFS_VALID_ATTRS;
573 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
574 return 0;
575
576 trace_nfs_setattr_enter(inode);
577
578 /* Write all dirty data */
579 if (S_ISREG(inode->i_mode))
580 nfs_sync_inode(inode);
581
582 fattr = nfs_alloc_fattr();
583 if (fattr == NULL) {
584 error = -ENOMEM;
585 goto out;
586 }
587
588 /*
589 * Return any delegations if we're going to change ACLs
590 */
591 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
592 NFS_PROTO(inode)->return_delegation(inode);
593 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
594 if (error == 0)
595 error = nfs_refresh_inode(inode, fattr);
596 nfs_free_fattr(fattr);
597 out:
598 trace_nfs_setattr_exit(inode, error);
599 return error;
600 }
601 EXPORT_SYMBOL_GPL(nfs_setattr);
602
603 /**
604 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
605 * @inode: inode of the file used
606 * @offset: file offset to start truncating
607 *
608 * This is a copy of the common vmtruncate, but with the locking
609 * corrected to take into account the fact that NFS requires
610 * inode->i_size to be updated under the inode->i_lock.
611 * Note: must be called with inode->i_lock held!
612 */
613 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
614 {
615 int err;
616
617 err = inode_newsize_ok(inode, offset);
618 if (err)
619 goto out;
620
621 i_size_write(inode, offset);
622 /* Optimisation */
623 if (offset == 0)
624 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
625
626 spin_unlock(&inode->i_lock);
627 truncate_pagecache(inode, offset);
628 spin_lock(&inode->i_lock);
629 out:
630 return err;
631 }
632
633 /**
634 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
635 * @inode: pointer to struct inode
636 * @attr: pointer to struct iattr
637 *
638 * Note: we do this in the *proc.c in order to ensure that
639 * it works for things like exclusive creates too.
640 */
641 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
642 struct nfs_fattr *fattr)
643 {
644 /* Barrier: bump the attribute generation count. */
645 nfs_fattr_set_barrier(fattr);
646
647 spin_lock(&inode->i_lock);
648 NFS_I(inode)->attr_gencount = fattr->gencount;
649 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
650 if ((attr->ia_valid & ATTR_MODE) != 0) {
651 int mode = attr->ia_mode & S_IALLUGO;
652 mode |= inode->i_mode & ~S_IALLUGO;
653 inode->i_mode = mode;
654 }
655 if ((attr->ia_valid & ATTR_UID) != 0)
656 inode->i_uid = attr->ia_uid;
657 if ((attr->ia_valid & ATTR_GID) != 0)
658 inode->i_gid = attr->ia_gid;
659 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
660 | NFS_INO_INVALID_ACL);
661 }
662 if ((attr->ia_valid & ATTR_SIZE) != 0) {
663 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
664 nfs_vmtruncate(inode, attr->ia_size);
665 }
666 if (fattr->valid)
667 nfs_update_inode(inode, fattr);
668 else
669 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
670 spin_unlock(&inode->i_lock);
671 }
672 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
673
674 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
675 {
676 struct dentry *parent;
677
678 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
679 return;
680 parent = dget_parent(dentry);
681 nfs_force_use_readdirplus(d_inode(parent));
682 dput(parent);
683 }
684
685 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
686 {
687 struct dentry *parent;
688
689 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
690 return;
691 parent = dget_parent(dentry);
692 nfs_advise_use_readdirplus(d_inode(parent));
693 dput(parent);
694 }
695
696 static bool nfs_need_revalidate_inode(struct inode *inode)
697 {
698 if (NFS_I(inode)->cache_validity &
699 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
700 return true;
701 if (nfs_attribute_cache_expired(inode))
702 return true;
703 return false;
704 }
705
706 int nfs_getattr(const struct path *path, struct kstat *stat,
707 u32 request_mask, unsigned int query_flags)
708 {
709 struct inode *inode = d_inode(path->dentry);
710 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
711 int err = 0;
712
713 trace_nfs_getattr_enter(inode);
714 /* Flush out writes to the server in order to update c/mtime. */
715 if (S_ISREG(inode->i_mode)) {
716 err = filemap_write_and_wait(inode->i_mapping);
717 if (err)
718 goto out;
719 }
720
721 /*
722 * We may force a getattr if the user cares about atime.
723 *
724 * Note that we only have to check the vfsmount flags here:
725 * - NFS always sets S_NOATIME by so checking it would give a
726 * bogus result
727 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
728 * no point in checking those.
729 */
730 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
731 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
732 need_atime = 0;
733
734 if (need_atime || nfs_need_revalidate_inode(inode)) {
735 struct nfs_server *server = NFS_SERVER(inode);
736
737 if (!(server->flags & NFS_MOUNT_NOAC))
738 nfs_readdirplus_parent_cache_miss(path->dentry);
739 else
740 nfs_readdirplus_parent_cache_hit(path->dentry);
741 err = __nfs_revalidate_inode(server, inode);
742 } else
743 nfs_readdirplus_parent_cache_hit(path->dentry);
744 if (!err) {
745 generic_fillattr(inode, stat);
746 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
747 if (S_ISDIR(inode->i_mode))
748 stat->blksize = NFS_SERVER(inode)->dtsize;
749 }
750 out:
751 trace_nfs_getattr_exit(inode, err);
752 return err;
753 }
754 EXPORT_SYMBOL_GPL(nfs_getattr);
755
756 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
757 {
758 atomic_set(&l_ctx->count, 1);
759 l_ctx->lockowner = current->files;
760 INIT_LIST_HEAD(&l_ctx->list);
761 atomic_set(&l_ctx->io_count, 0);
762 }
763
764 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
765 {
766 struct nfs_lock_context *head = &ctx->lock_context;
767 struct nfs_lock_context *pos = head;
768
769 do {
770 if (pos->lockowner != current->files)
771 continue;
772 atomic_inc(&pos->count);
773 return pos;
774 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
775 return NULL;
776 }
777
778 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
779 {
780 struct nfs_lock_context *res, *new = NULL;
781 struct inode *inode = d_inode(ctx->dentry);
782
783 spin_lock(&inode->i_lock);
784 res = __nfs_find_lock_context(ctx);
785 if (res == NULL) {
786 spin_unlock(&inode->i_lock);
787 new = kmalloc(sizeof(*new), GFP_KERNEL);
788 if (new == NULL)
789 return ERR_PTR(-ENOMEM);
790 nfs_init_lock_context(new);
791 spin_lock(&inode->i_lock);
792 res = __nfs_find_lock_context(ctx);
793 if (res == NULL) {
794 list_add_tail(&new->list, &ctx->lock_context.list);
795 new->open_context = ctx;
796 res = new;
797 new = NULL;
798 }
799 }
800 spin_unlock(&inode->i_lock);
801 kfree(new);
802 return res;
803 }
804 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
805
806 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
807 {
808 struct nfs_open_context *ctx = l_ctx->open_context;
809 struct inode *inode = d_inode(ctx->dentry);
810
811 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
812 return;
813 list_del(&l_ctx->list);
814 spin_unlock(&inode->i_lock);
815 kfree(l_ctx);
816 }
817 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
818
819 /**
820 * nfs_close_context - Common close_context() routine NFSv2/v3
821 * @ctx: pointer to context
822 * @is_sync: is this a synchronous close
823 *
824 * Ensure that the attributes are up to date if we're mounted
825 * with close-to-open semantics and we have cached data that will
826 * need to be revalidated on open.
827 */
828 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
829 {
830 struct nfs_inode *nfsi;
831 struct inode *inode;
832 struct nfs_server *server;
833
834 if (!(ctx->mode & FMODE_WRITE))
835 return;
836 if (!is_sync)
837 return;
838 inode = d_inode(ctx->dentry);
839 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
840 return;
841 nfsi = NFS_I(inode);
842 if (inode->i_mapping->nrpages == 0)
843 return;
844 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
845 return;
846 if (!list_empty(&nfsi->open_files))
847 return;
848 server = NFS_SERVER(inode);
849 if (server->flags & NFS_MOUNT_NOCTO)
850 return;
851 nfs_revalidate_inode(server, inode);
852 }
853 EXPORT_SYMBOL_GPL(nfs_close_context);
854
855 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
856 fmode_t f_mode,
857 struct file *filp)
858 {
859 struct nfs_open_context *ctx;
860 struct rpc_cred *cred = rpc_lookup_cred();
861 if (IS_ERR(cred))
862 return ERR_CAST(cred);
863
864 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
865 if (!ctx) {
866 put_rpccred(cred);
867 return ERR_PTR(-ENOMEM);
868 }
869 nfs_sb_active(dentry->d_sb);
870 ctx->dentry = dget(dentry);
871 ctx->cred = cred;
872 ctx->state = NULL;
873 ctx->mode = f_mode;
874 ctx->flags = 0;
875 ctx->error = 0;
876 ctx->flock_owner = (fl_owner_t)filp;
877 nfs_init_lock_context(&ctx->lock_context);
878 ctx->lock_context.open_context = ctx;
879 INIT_LIST_HEAD(&ctx->list);
880 ctx->mdsthreshold = NULL;
881 return ctx;
882 }
883 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
884
885 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
886 {
887 if (ctx != NULL)
888 atomic_inc(&ctx->lock_context.count);
889 return ctx;
890 }
891 EXPORT_SYMBOL_GPL(get_nfs_open_context);
892
893 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
894 {
895 struct inode *inode = d_inode(ctx->dentry);
896 struct super_block *sb = ctx->dentry->d_sb;
897
898 if (!list_empty(&ctx->list)) {
899 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
900 return;
901 list_del(&ctx->list);
902 spin_unlock(&inode->i_lock);
903 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
904 return;
905 if (inode != NULL)
906 NFS_PROTO(inode)->close_context(ctx, is_sync);
907 if (ctx->cred != NULL)
908 put_rpccred(ctx->cred);
909 dput(ctx->dentry);
910 nfs_sb_deactive(sb);
911 kfree(ctx->mdsthreshold);
912 kfree(ctx);
913 }
914
915 void put_nfs_open_context(struct nfs_open_context *ctx)
916 {
917 __put_nfs_open_context(ctx, 0);
918 }
919 EXPORT_SYMBOL_GPL(put_nfs_open_context);
920
921 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
922 {
923 __put_nfs_open_context(ctx, 1);
924 }
925
926 /*
927 * Ensure that mmap has a recent RPC credential for use when writing out
928 * shared pages
929 */
930 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
931 {
932 struct inode *inode = d_inode(ctx->dentry);
933 struct nfs_inode *nfsi = NFS_I(inode);
934
935 spin_lock(&inode->i_lock);
936 if (ctx->mode & FMODE_WRITE)
937 list_add(&ctx->list, &nfsi->open_files);
938 else
939 list_add_tail(&ctx->list, &nfsi->open_files);
940 spin_unlock(&inode->i_lock);
941 }
942 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
943
944 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
945 {
946 filp->private_data = get_nfs_open_context(ctx);
947 if (list_empty(&ctx->list))
948 nfs_inode_attach_open_context(ctx);
949 }
950 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
951
952 /*
953 * Given an inode, search for an open context with the desired characteristics
954 */
955 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
956 {
957 struct nfs_inode *nfsi = NFS_I(inode);
958 struct nfs_open_context *pos, *ctx = NULL;
959
960 spin_lock(&inode->i_lock);
961 list_for_each_entry(pos, &nfsi->open_files, list) {
962 if (cred != NULL && pos->cred != cred)
963 continue;
964 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
965 continue;
966 ctx = get_nfs_open_context(pos);
967 break;
968 }
969 spin_unlock(&inode->i_lock);
970 return ctx;
971 }
972
973 void nfs_file_clear_open_context(struct file *filp)
974 {
975 struct nfs_open_context *ctx = nfs_file_open_context(filp);
976
977 if (ctx) {
978 struct inode *inode = d_inode(ctx->dentry);
979
980 /*
981 * We fatal error on write before. Try to writeback
982 * every page again.
983 */
984 if (ctx->error < 0)
985 invalidate_inode_pages2(inode->i_mapping);
986 filp->private_data = NULL;
987 spin_lock(&inode->i_lock);
988 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
989 spin_unlock(&inode->i_lock);
990 put_nfs_open_context_sync(ctx);
991 }
992 }
993
994 /*
995 * These allocate and release file read/write context information.
996 */
997 int nfs_open(struct inode *inode, struct file *filp)
998 {
999 struct nfs_open_context *ctx;
1000
1001 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1002 if (IS_ERR(ctx))
1003 return PTR_ERR(ctx);
1004 nfs_file_set_open_context(filp, ctx);
1005 put_nfs_open_context(ctx);
1006 nfs_fscache_open_file(inode, filp);
1007 return 0;
1008 }
1009
1010 /*
1011 * This function is called whenever some part of NFS notices that
1012 * the cached attributes have to be refreshed.
1013 */
1014 int
1015 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1016 {
1017 int status = -ESTALE;
1018 struct nfs4_label *label = NULL;
1019 struct nfs_fattr *fattr = NULL;
1020 struct nfs_inode *nfsi = NFS_I(inode);
1021
1022 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1023 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1024
1025 trace_nfs_revalidate_inode_enter(inode);
1026
1027 if (is_bad_inode(inode))
1028 goto out;
1029 if (NFS_STALE(inode))
1030 goto out;
1031
1032 /* pNFS: Attributes aren't updated until we layoutcommit */
1033 if (S_ISREG(inode->i_mode)) {
1034 status = pnfs_sync_inode(inode, false);
1035 if (status)
1036 goto out;
1037 }
1038
1039 status = -ENOMEM;
1040 fattr = nfs_alloc_fattr();
1041 if (fattr == NULL)
1042 goto out;
1043
1044 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1045
1046 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1047 if (IS_ERR(label)) {
1048 status = PTR_ERR(label);
1049 goto out;
1050 }
1051
1052 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
1053 if (status != 0) {
1054 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1055 inode->i_sb->s_id,
1056 (unsigned long long)NFS_FILEID(inode), status);
1057 if (status == -ESTALE) {
1058 nfs_zap_caches(inode);
1059 if (!S_ISDIR(inode->i_mode))
1060 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1061 }
1062 goto err_out;
1063 }
1064
1065 status = nfs_refresh_inode(inode, fattr);
1066 if (status) {
1067 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1068 inode->i_sb->s_id,
1069 (unsigned long long)NFS_FILEID(inode), status);
1070 goto err_out;
1071 }
1072
1073 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1074 nfs_zap_acl_cache(inode);
1075
1076 nfs_setsecurity(inode, fattr, label);
1077
1078 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1079 inode->i_sb->s_id,
1080 (unsigned long long)NFS_FILEID(inode));
1081
1082 err_out:
1083 nfs4_label_free(label);
1084 out:
1085 nfs_free_fattr(fattr);
1086 trace_nfs_revalidate_inode_exit(inode, status);
1087 return status;
1088 }
1089
1090 int nfs_attribute_cache_expired(struct inode *inode)
1091 {
1092 if (nfs_have_delegated_attributes(inode))
1093 return 0;
1094 return nfs_attribute_timeout(inode);
1095 }
1096
1097 /**
1098 * nfs_revalidate_inode - Revalidate the inode attributes
1099 * @server - pointer to nfs_server struct
1100 * @inode - pointer to inode struct
1101 *
1102 * Updates inode attribute information by retrieving the data from the server.
1103 */
1104 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1105 {
1106 if (!nfs_need_revalidate_inode(inode))
1107 return NFS_STALE(inode) ? -ESTALE : 0;
1108 return __nfs_revalidate_inode(server, inode);
1109 }
1110 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1111
1112 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1113 {
1114 struct nfs_inode *nfsi = NFS_I(inode);
1115 int ret;
1116
1117 if (mapping->nrpages != 0) {
1118 if (S_ISREG(inode->i_mode)) {
1119 unmap_mapping_range(mapping, 0, 0, 0);
1120 ret = nfs_sync_mapping(mapping);
1121 if (ret < 0)
1122 return ret;
1123 }
1124 ret = invalidate_inode_pages2(mapping);
1125 if (ret < 0)
1126 return ret;
1127 }
1128 if (S_ISDIR(inode->i_mode)) {
1129 spin_lock(&inode->i_lock);
1130 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1131 spin_unlock(&inode->i_lock);
1132 }
1133 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1134 nfs_fscache_wait_on_invalidate(inode);
1135
1136 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1137 inode->i_sb->s_id,
1138 (unsigned long long)NFS_FILEID(inode));
1139 return 0;
1140 }
1141
1142 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1143 {
1144 return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1145 NFS_STALE(inode);
1146 }
1147
1148 int nfs_revalidate_mapping_rcu(struct inode *inode)
1149 {
1150 struct nfs_inode *nfsi = NFS_I(inode);
1151 unsigned long *bitlock = &nfsi->flags;
1152 int ret = 0;
1153
1154 if (IS_SWAPFILE(inode))
1155 goto out;
1156 if (nfs_mapping_need_revalidate_inode(inode)) {
1157 ret = -ECHILD;
1158 goto out;
1159 }
1160 spin_lock(&inode->i_lock);
1161 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1162 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1163 ret = -ECHILD;
1164 spin_unlock(&inode->i_lock);
1165 out:
1166 return ret;
1167 }
1168
1169 /**
1170 * nfs_revalidate_mapping - Revalidate the pagecache
1171 * @inode - pointer to host inode
1172 * @mapping - pointer to mapping
1173 */
1174 int nfs_revalidate_mapping(struct inode *inode,
1175 struct address_space *mapping)
1176 {
1177 struct nfs_inode *nfsi = NFS_I(inode);
1178 unsigned long *bitlock = &nfsi->flags;
1179 int ret = 0;
1180
1181 /* swapfiles are not supposed to be shared. */
1182 if (IS_SWAPFILE(inode))
1183 goto out;
1184
1185 if (nfs_mapping_need_revalidate_inode(inode)) {
1186 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1187 if (ret < 0)
1188 goto out;
1189 }
1190
1191 /*
1192 * We must clear NFS_INO_INVALID_DATA first to ensure that
1193 * invalidations that come in while we're shooting down the mappings
1194 * are respected. But, that leaves a race window where one revalidator
1195 * can clear the flag, and then another checks it before the mapping
1196 * gets invalidated. Fix that by serializing access to this part of
1197 * the function.
1198 *
1199 * At the same time, we need to allow other tasks to see whether we
1200 * might be in the middle of invalidating the pages, so we only set
1201 * the bit lock here if it looks like we're going to be doing that.
1202 */
1203 for (;;) {
1204 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1205 nfs_wait_bit_killable, TASK_KILLABLE);
1206 if (ret)
1207 goto out;
1208 spin_lock(&inode->i_lock);
1209 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1210 spin_unlock(&inode->i_lock);
1211 continue;
1212 }
1213 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1214 break;
1215 spin_unlock(&inode->i_lock);
1216 goto out;
1217 }
1218
1219 set_bit(NFS_INO_INVALIDATING, bitlock);
1220 smp_wmb();
1221 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1222 spin_unlock(&inode->i_lock);
1223 trace_nfs_invalidate_mapping_enter(inode);
1224 ret = nfs_invalidate_mapping(inode, mapping);
1225 trace_nfs_invalidate_mapping_exit(inode, ret);
1226
1227 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1228 smp_mb__after_atomic();
1229 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1230 out:
1231 return ret;
1232 }
1233
1234 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1235 {
1236 struct inode *inode = &nfsi->vfs_inode;
1237
1238 assert_spin_locked(&inode->i_lock);
1239
1240 if (!S_ISREG(inode->i_mode))
1241 return false;
1242 if (list_empty(&nfsi->open_files))
1243 return false;
1244 /* Note: This relies on nfsi->open_files being ordered with writers
1245 * being placed at the head of the list.
1246 * See nfs_inode_attach_open_context()
1247 */
1248 return (list_first_entry(&nfsi->open_files,
1249 struct nfs_open_context,
1250 list)->mode & FMODE_WRITE) == FMODE_WRITE;
1251 }
1252
1253 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1254 {
1255 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1256 }
1257
1258 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1259 {
1260 struct nfs_inode *nfsi = NFS_I(inode);
1261 unsigned long ret = 0;
1262
1263 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1264 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1265 && inode->i_version == fattr->pre_change_attr) {
1266 inode->i_version = fattr->change_attr;
1267 if (S_ISDIR(inode->i_mode))
1268 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1269 ret |= NFS_INO_INVALID_ATTR;
1270 }
1271 /* If we have atomic WCC data, we may update some attributes */
1272 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1273 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1274 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1275 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1276 ret |= NFS_INO_INVALID_ATTR;
1277 }
1278
1279 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1280 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1281 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1282 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1283 if (S_ISDIR(inode->i_mode))
1284 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1285 ret |= NFS_INO_INVALID_ATTR;
1286 }
1287 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1288 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1289 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1290 && nfsi->nrequests == 0) {
1291 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1292 ret |= NFS_INO_INVALID_ATTR;
1293 }
1294
1295 return ret;
1296 }
1297
1298 /**
1299 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1300 * @inode - pointer to inode
1301 * @fattr - updated attributes
1302 *
1303 * Verifies the attribute cache. If we have just changed the attributes,
1304 * so that fattr carries weak cache consistency data, then it may
1305 * also update the ctime/mtime/change_attribute.
1306 */
1307 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1308 {
1309 struct nfs_inode *nfsi = NFS_I(inode);
1310 loff_t cur_size, new_isize;
1311 unsigned long invalid = 0;
1312
1313
1314 if (nfs_have_delegated_attributes(inode))
1315 return 0;
1316 /* Has the inode gone and changed behind our back? */
1317 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1318 return -EIO;
1319 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1320 return -EIO;
1321
1322 if (!nfs_file_has_buffered_writers(nfsi)) {
1323 /* Verify a few of the more important attributes */
1324 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && inode->i_version != fattr->change_attr)
1325 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_REVAL_PAGECACHE;
1326
1327 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1328 invalid |= NFS_INO_INVALID_ATTR;
1329
1330 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&inode->i_ctime, &fattr->ctime))
1331 invalid |= NFS_INO_INVALID_ATTR;
1332
1333 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1334 cur_size = i_size_read(inode);
1335 new_isize = nfs_size_to_loff_t(fattr->size);
1336 if (cur_size != new_isize)
1337 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1338 }
1339 }
1340
1341 /* Have any file permissions changed? */
1342 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1343 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1344 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1345 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1346 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1347 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1348
1349 /* Has the link count changed? */
1350 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1351 invalid |= NFS_INO_INVALID_ATTR;
1352
1353 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1354 invalid |= NFS_INO_INVALID_ATIME;
1355
1356 if (invalid != 0)
1357 nfs_set_cache_invalid(inode, invalid | NFS_INO_REVAL_FORCED);
1358
1359 nfsi->read_cache_jiffies = fattr->time_start;
1360 return 0;
1361 }
1362
1363 static atomic_long_t nfs_attr_generation_counter;
1364
1365 static unsigned long nfs_read_attr_generation_counter(void)
1366 {
1367 return atomic_long_read(&nfs_attr_generation_counter);
1368 }
1369
1370 unsigned long nfs_inc_attr_generation_counter(void)
1371 {
1372 return atomic_long_inc_return(&nfs_attr_generation_counter);
1373 }
1374 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1375
1376 void nfs_fattr_init(struct nfs_fattr *fattr)
1377 {
1378 fattr->valid = 0;
1379 fattr->time_start = jiffies;
1380 fattr->gencount = nfs_inc_attr_generation_counter();
1381 fattr->owner_name = NULL;
1382 fattr->group_name = NULL;
1383 }
1384 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1385
1386 /**
1387 * nfs_fattr_set_barrier
1388 * @fattr: attributes
1389 *
1390 * Used to set a barrier after an attribute was updated. This
1391 * barrier ensures that older attributes from RPC calls that may
1392 * have raced with our update cannot clobber these new values.
1393 * Note that you are still responsible for ensuring that other
1394 * operations which change the attribute on the server do not
1395 * collide.
1396 */
1397 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1398 {
1399 fattr->gencount = nfs_inc_attr_generation_counter();
1400 }
1401
1402 struct nfs_fattr *nfs_alloc_fattr(void)
1403 {
1404 struct nfs_fattr *fattr;
1405
1406 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1407 if (fattr != NULL)
1408 nfs_fattr_init(fattr);
1409 return fattr;
1410 }
1411 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1412
1413 struct nfs_fh *nfs_alloc_fhandle(void)
1414 {
1415 struct nfs_fh *fh;
1416
1417 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1418 if (fh != NULL)
1419 fh->size = 0;
1420 return fh;
1421 }
1422 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1423
1424 #ifdef NFS_DEBUG
1425 /*
1426 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1427 * in the same way that wireshark does
1428 *
1429 * @fh: file handle
1430 *
1431 * For debugging only.
1432 */
1433 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1434 {
1435 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1436 * not on the result */
1437 return nfs_fhandle_hash(fh);
1438 }
1439 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1440
1441 /*
1442 * _nfs_display_fhandle - display an NFS file handle on the console
1443 *
1444 * @fh: file handle to display
1445 * @caption: display caption
1446 *
1447 * For debugging only.
1448 */
1449 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1450 {
1451 unsigned short i;
1452
1453 if (fh == NULL || fh->size == 0) {
1454 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1455 return;
1456 }
1457
1458 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1459 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1460 for (i = 0; i < fh->size; i += 16) {
1461 __be32 *pos = (__be32 *)&fh->data[i];
1462
1463 switch ((fh->size - i - 1) >> 2) {
1464 case 0:
1465 printk(KERN_DEFAULT " %08x\n",
1466 be32_to_cpup(pos));
1467 break;
1468 case 1:
1469 printk(KERN_DEFAULT " %08x %08x\n",
1470 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1471 break;
1472 case 2:
1473 printk(KERN_DEFAULT " %08x %08x %08x\n",
1474 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1475 be32_to_cpup(pos + 2));
1476 break;
1477 default:
1478 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1479 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1480 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1481 }
1482 }
1483 }
1484 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1485 #endif
1486
1487 /**
1488 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1489 * @inode - pointer to inode
1490 * @fattr - attributes
1491 *
1492 * Attempt to divine whether or not an RPC call reply carrying stale
1493 * attributes got scheduled after another call carrying updated ones.
1494 *
1495 * To do so, the function first assumes that a more recent ctime means
1496 * that the attributes in fattr are newer, however it also attempt to
1497 * catch the case where ctime either didn't change, or went backwards
1498 * (if someone reset the clock on the server) by looking at whether
1499 * or not this RPC call was started after the inode was last updated.
1500 * Note also the check for wraparound of 'attr_gencount'
1501 *
1502 * The function returns 'true' if it thinks the attributes in 'fattr' are
1503 * more recent than the ones cached in the inode.
1504 *
1505 */
1506 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1507 {
1508 const struct nfs_inode *nfsi = NFS_I(inode);
1509
1510 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1511 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1512 }
1513
1514 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1515 {
1516 int ret;
1517
1518 trace_nfs_refresh_inode_enter(inode);
1519
1520 if (nfs_inode_attrs_need_update(inode, fattr))
1521 ret = nfs_update_inode(inode, fattr);
1522 else
1523 ret = nfs_check_inode_attributes(inode, fattr);
1524
1525 trace_nfs_refresh_inode_exit(inode, ret);
1526 return ret;
1527 }
1528
1529 /**
1530 * nfs_refresh_inode - try to update the inode attribute cache
1531 * @inode - pointer to inode
1532 * @fattr - updated attributes
1533 *
1534 * Check that an RPC call that returned attributes has not overlapped with
1535 * other recent updates of the inode metadata, then decide whether it is
1536 * safe to do a full update of the inode attributes, or whether just to
1537 * call nfs_check_inode_attributes.
1538 */
1539 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1540 {
1541 int status;
1542
1543 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1544 return 0;
1545 spin_lock(&inode->i_lock);
1546 status = nfs_refresh_inode_locked(inode, fattr);
1547 spin_unlock(&inode->i_lock);
1548
1549 return status;
1550 }
1551 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1552
1553 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1554 {
1555 unsigned long invalid = NFS_INO_INVALID_ATTR;
1556
1557 if (S_ISDIR(inode->i_mode))
1558 invalid |= NFS_INO_INVALID_DATA;
1559 nfs_set_cache_invalid(inode, invalid);
1560 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1561 return 0;
1562 return nfs_refresh_inode_locked(inode, fattr);
1563 }
1564
1565 /**
1566 * nfs_post_op_update_inode - try to update the inode attribute cache
1567 * @inode - pointer to inode
1568 * @fattr - updated attributes
1569 *
1570 * After an operation that has changed the inode metadata, mark the
1571 * attribute cache as being invalid, then try to update it.
1572 *
1573 * NB: if the server didn't return any post op attributes, this
1574 * function will force the retrieval of attributes before the next
1575 * NFS request. Thus it should be used only for operations that
1576 * are expected to change one or more attributes, to avoid
1577 * unnecessary NFS requests and trips through nfs_update_inode().
1578 */
1579 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1580 {
1581 int status;
1582
1583 spin_lock(&inode->i_lock);
1584 nfs_fattr_set_barrier(fattr);
1585 status = nfs_post_op_update_inode_locked(inode, fattr);
1586 spin_unlock(&inode->i_lock);
1587
1588 return status;
1589 }
1590 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1591
1592 /**
1593 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1594 * @inode - pointer to inode
1595 * @fattr - updated attributes
1596 *
1597 * After an operation that has changed the inode metadata, mark the
1598 * attribute cache as being invalid, then try to update it. Fake up
1599 * weak cache consistency data, if none exist.
1600 *
1601 * This function is mainly designed to be used by the ->write_done() functions.
1602 */
1603 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1604 {
1605 int status;
1606
1607 /* Don't do a WCC update if these attributes are already stale */
1608 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1609 !nfs_inode_attrs_need_update(inode, fattr)) {
1610 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1611 | NFS_ATTR_FATTR_PRESIZE
1612 | NFS_ATTR_FATTR_PREMTIME
1613 | NFS_ATTR_FATTR_PRECTIME);
1614 goto out_noforce;
1615 }
1616 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1617 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1618 fattr->pre_change_attr = inode->i_version;
1619 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1620 }
1621 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1622 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1623 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1624 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1625 }
1626 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1627 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1628 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1629 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1630 }
1631 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1632 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1633 fattr->pre_size = i_size_read(inode);
1634 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1635 }
1636 out_noforce:
1637 status = nfs_post_op_update_inode_locked(inode, fattr);
1638 return status;
1639 }
1640
1641 /**
1642 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1643 * @inode - pointer to inode
1644 * @fattr - updated attributes
1645 *
1646 * After an operation that has changed the inode metadata, mark the
1647 * attribute cache as being invalid, then try to update it. Fake up
1648 * weak cache consistency data, if none exist.
1649 *
1650 * This function is mainly designed to be used by the ->write_done() functions.
1651 */
1652 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1653 {
1654 int status;
1655
1656 spin_lock(&inode->i_lock);
1657 nfs_fattr_set_barrier(fattr);
1658 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1659 spin_unlock(&inode->i_lock);
1660 return status;
1661 }
1662 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1663
1664
1665 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1666 struct nfs_fattr *fattr)
1667 {
1668 bool ret1 = true, ret2 = true;
1669
1670 if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1671 ret1 = (nfsi->fileid == fattr->fileid);
1672 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1673 ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1674 return ret1 || ret2;
1675 }
1676
1677 /*
1678 * Many nfs protocol calls return the new file attributes after
1679 * an operation. Here we update the inode to reflect the state
1680 * of the server's inode.
1681 *
1682 * This is a bit tricky because we have to make sure all dirty pages
1683 * have been sent off to the server before calling invalidate_inode_pages.
1684 * To make sure no other process adds more write requests while we try
1685 * our best to flush them, we make them sleep during the attribute refresh.
1686 *
1687 * A very similar scenario holds for the dir cache.
1688 */
1689 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1690 {
1691 struct nfs_server *server;
1692 struct nfs_inode *nfsi = NFS_I(inode);
1693 loff_t cur_isize, new_isize;
1694 unsigned long invalid = 0;
1695 unsigned long now = jiffies;
1696 unsigned long save_cache_validity;
1697 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1698 bool cache_revalidated = true;
1699
1700 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1701 __func__, inode->i_sb->s_id, inode->i_ino,
1702 nfs_display_fhandle_hash(NFS_FH(inode)),
1703 atomic_read(&inode->i_count), fattr->valid);
1704
1705 if (!nfs_fileid_valid(nfsi, fattr)) {
1706 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1707 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1708 NFS_SERVER(inode)->nfs_client->cl_hostname,
1709 inode->i_sb->s_id, (long long)nfsi->fileid,
1710 (long long)fattr->fileid);
1711 goto out_err;
1712 }
1713
1714 /*
1715 * Make sure the inode's type hasn't changed.
1716 */
1717 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1718 /*
1719 * Big trouble! The inode has become a different object.
1720 */
1721 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1722 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1723 goto out_err;
1724 }
1725
1726 server = NFS_SERVER(inode);
1727 /* Update the fsid? */
1728 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1729 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1730 !IS_AUTOMOUNT(inode))
1731 server->fsid = fattr->fsid;
1732
1733 /*
1734 * Update the read time so we don't revalidate too often.
1735 */
1736 nfsi->read_cache_jiffies = fattr->time_start;
1737
1738 save_cache_validity = nfsi->cache_validity;
1739 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1740 | NFS_INO_INVALID_ATIME
1741 | NFS_INO_REVAL_FORCED
1742 | NFS_INO_REVAL_PAGECACHE);
1743
1744 /* Do atomic weak cache consistency updates */
1745 invalid |= nfs_wcc_update_inode(inode, fattr);
1746
1747 if (pnfs_layoutcommit_outstanding(inode)) {
1748 nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1749 cache_revalidated = false;
1750 }
1751
1752 /* More cache consistency checks */
1753 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1754 if (inode->i_version != fattr->change_attr) {
1755 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1756 inode->i_sb->s_id, inode->i_ino);
1757 /* Could it be a race with writeback? */
1758 if (!have_writers) {
1759 invalid |= NFS_INO_INVALID_ATTR
1760 | NFS_INO_INVALID_DATA
1761 | NFS_INO_INVALID_ACCESS
1762 | NFS_INO_INVALID_ACL;
1763 if (S_ISDIR(inode->i_mode))
1764 nfs_force_lookup_revalidate(inode);
1765 }
1766 inode->i_version = fattr->change_attr;
1767 }
1768 } else {
1769 nfsi->cache_validity |= save_cache_validity;
1770 cache_revalidated = false;
1771 }
1772
1773 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1774 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1775 } else if (server->caps & NFS_CAP_MTIME) {
1776 nfsi->cache_validity |= save_cache_validity &
1777 (NFS_INO_INVALID_ATTR
1778 | NFS_INO_REVAL_FORCED);
1779 cache_revalidated = false;
1780 }
1781
1782 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1783 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1784 } else if (server->caps & NFS_CAP_CTIME) {
1785 nfsi->cache_validity |= save_cache_validity &
1786 (NFS_INO_INVALID_ATTR
1787 | NFS_INO_REVAL_FORCED);
1788 cache_revalidated = false;
1789 }
1790
1791 /* Check if our cached file size is stale */
1792 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1793 new_isize = nfs_size_to_loff_t(fattr->size);
1794 cur_isize = i_size_read(inode);
1795 if (new_isize != cur_isize) {
1796 /* Do we perhaps have any outstanding writes, or has
1797 * the file grown beyond our last write? */
1798 if (nfsi->nrequests == 0 || new_isize > cur_isize) {
1799 i_size_write(inode, new_isize);
1800 if (!have_writers)
1801 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1802 }
1803 dprintk("NFS: isize change on server for file %s/%ld "
1804 "(%Ld to %Ld)\n",
1805 inode->i_sb->s_id,
1806 inode->i_ino,
1807 (long long)cur_isize,
1808 (long long)new_isize);
1809 }
1810 } else {
1811 nfsi->cache_validity |= save_cache_validity &
1812 (NFS_INO_INVALID_ATTR
1813 | NFS_INO_REVAL_PAGECACHE
1814 | NFS_INO_REVAL_FORCED);
1815 cache_revalidated = false;
1816 }
1817
1818
1819 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1820 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1821 else if (server->caps & NFS_CAP_ATIME) {
1822 nfsi->cache_validity |= save_cache_validity &
1823 (NFS_INO_INVALID_ATIME
1824 | NFS_INO_REVAL_FORCED);
1825 cache_revalidated = false;
1826 }
1827
1828 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1829 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1830 umode_t newmode = inode->i_mode & S_IFMT;
1831 newmode |= fattr->mode & S_IALLUGO;
1832 inode->i_mode = newmode;
1833 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1834 }
1835 } else if (server->caps & NFS_CAP_MODE) {
1836 nfsi->cache_validity |= save_cache_validity &
1837 (NFS_INO_INVALID_ATTR
1838 | NFS_INO_INVALID_ACCESS
1839 | NFS_INO_INVALID_ACL
1840 | NFS_INO_REVAL_FORCED);
1841 cache_revalidated = false;
1842 }
1843
1844 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1845 if (!uid_eq(inode->i_uid, fattr->uid)) {
1846 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1847 inode->i_uid = fattr->uid;
1848 }
1849 } else if (server->caps & NFS_CAP_OWNER) {
1850 nfsi->cache_validity |= save_cache_validity &
1851 (NFS_INO_INVALID_ATTR
1852 | NFS_INO_INVALID_ACCESS
1853 | NFS_INO_INVALID_ACL
1854 | NFS_INO_REVAL_FORCED);
1855 cache_revalidated = false;
1856 }
1857
1858 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1859 if (!gid_eq(inode->i_gid, fattr->gid)) {
1860 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1861 inode->i_gid = fattr->gid;
1862 }
1863 } else if (server->caps & NFS_CAP_OWNER_GROUP) {
1864 nfsi->cache_validity |= save_cache_validity &
1865 (NFS_INO_INVALID_ATTR
1866 | NFS_INO_INVALID_ACCESS
1867 | NFS_INO_INVALID_ACL
1868 | NFS_INO_REVAL_FORCED);
1869 cache_revalidated = false;
1870 }
1871
1872 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1873 if (inode->i_nlink != fattr->nlink) {
1874 invalid |= NFS_INO_INVALID_ATTR;
1875 if (S_ISDIR(inode->i_mode))
1876 invalid |= NFS_INO_INVALID_DATA;
1877 set_nlink(inode, fattr->nlink);
1878 }
1879 } else if (server->caps & NFS_CAP_NLINK) {
1880 nfsi->cache_validity |= save_cache_validity &
1881 (NFS_INO_INVALID_ATTR
1882 | NFS_INO_REVAL_FORCED);
1883 cache_revalidated = false;
1884 }
1885
1886 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1887 /*
1888 * report the blocks in 512byte units
1889 */
1890 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1891 } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1892 inode->i_blocks = fattr->du.nfs2.blocks;
1893 else
1894 cache_revalidated = false;
1895
1896 /* Update attrtimeo value if we're out of the unstable period */
1897 if (invalid & NFS_INO_INVALID_ATTR) {
1898 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1899 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1900 nfsi->attrtimeo_timestamp = now;
1901 /* Set barrier to be more recent than all outstanding updates */
1902 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1903 } else {
1904 if (cache_revalidated) {
1905 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1906 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1907 nfsi->attrtimeo <<= 1;
1908 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1909 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1910 }
1911 nfsi->attrtimeo_timestamp = now;
1912 }
1913 /* Set the barrier to be more recent than this fattr */
1914 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1915 nfsi->attr_gencount = fattr->gencount;
1916 }
1917
1918 /* Don't declare attrcache up to date if there were no attrs! */
1919 if (cache_revalidated)
1920 invalid &= ~NFS_INO_INVALID_ATTR;
1921
1922 /* Don't invalidate the data if we were to blame */
1923 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1924 || S_ISLNK(inode->i_mode)))
1925 invalid &= ~NFS_INO_INVALID_DATA;
1926 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1927 (save_cache_validity & NFS_INO_REVAL_FORCED))
1928 nfs_set_cache_invalid(inode, invalid);
1929
1930 return 0;
1931 out_err:
1932 /*
1933 * No need to worry about unhashing the dentry, as the
1934 * lookup validation will know that the inode is bad.
1935 * (But we fall through to invalidate the caches.)
1936 */
1937 nfs_invalidate_inode(inode);
1938 return -ESTALE;
1939 }
1940
1941 struct inode *nfs_alloc_inode(struct super_block *sb)
1942 {
1943 struct nfs_inode *nfsi;
1944 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1945 if (!nfsi)
1946 return NULL;
1947 nfsi->flags = 0UL;
1948 nfsi->cache_validity = 0UL;
1949 #if IS_ENABLED(CONFIG_NFS_V4)
1950 nfsi->nfs4_acl = NULL;
1951 #endif /* CONFIG_NFS_V4 */
1952 return &nfsi->vfs_inode;
1953 }
1954 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1955
1956 static void nfs_i_callback(struct rcu_head *head)
1957 {
1958 struct inode *inode = container_of(head, struct inode, i_rcu);
1959 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1960 }
1961
1962 void nfs_destroy_inode(struct inode *inode)
1963 {
1964 call_rcu(&inode->i_rcu, nfs_i_callback);
1965 }
1966 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1967
1968 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1969 {
1970 #if IS_ENABLED(CONFIG_NFS_V4)
1971 INIT_LIST_HEAD(&nfsi->open_states);
1972 nfsi->delegation = NULL;
1973 init_rwsem(&nfsi->rwsem);
1974 nfsi->layout = NULL;
1975 #endif
1976 }
1977
1978 static void init_once(void *foo)
1979 {
1980 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1981
1982 inode_init_once(&nfsi->vfs_inode);
1983 INIT_LIST_HEAD(&nfsi->open_files);
1984 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1985 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1986 INIT_LIST_HEAD(&nfsi->commit_info.list);
1987 nfsi->nrequests = 0;
1988 nfsi->commit_info.ncommit = 0;
1989 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1990 init_rwsem(&nfsi->rmdir_sem);
1991 nfs4_init_once(nfsi);
1992 }
1993
1994 static int __init nfs_init_inodecache(void)
1995 {
1996 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1997 sizeof(struct nfs_inode),
1998 0, (SLAB_RECLAIM_ACCOUNT|
1999 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2000 init_once);
2001 if (nfs_inode_cachep == NULL)
2002 return -ENOMEM;
2003
2004 return 0;
2005 }
2006
2007 static void nfs_destroy_inodecache(void)
2008 {
2009 /*
2010 * Make sure all delayed rcu free inodes are flushed before we
2011 * destroy cache.
2012 */
2013 rcu_barrier();
2014 kmem_cache_destroy(nfs_inode_cachep);
2015 }
2016
2017 struct workqueue_struct *nfsiod_workqueue;
2018 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2019
2020 /*
2021 * start up the nfsiod workqueue
2022 */
2023 static int nfsiod_start(void)
2024 {
2025 struct workqueue_struct *wq;
2026 dprintk("RPC: creating workqueue nfsiod\n");
2027 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2028 if (wq == NULL)
2029 return -ENOMEM;
2030 nfsiod_workqueue = wq;
2031 return 0;
2032 }
2033
2034 /*
2035 * Destroy the nfsiod workqueue
2036 */
2037 static void nfsiod_stop(void)
2038 {
2039 struct workqueue_struct *wq;
2040
2041 wq = nfsiod_workqueue;
2042 if (wq == NULL)
2043 return;
2044 nfsiod_workqueue = NULL;
2045 destroy_workqueue(wq);
2046 }
2047
2048 unsigned int nfs_net_id;
2049 EXPORT_SYMBOL_GPL(nfs_net_id);
2050
2051 static int nfs_net_init(struct net *net)
2052 {
2053 nfs_clients_init(net);
2054 return nfs_fs_proc_net_init(net);
2055 }
2056
2057 static void nfs_net_exit(struct net *net)
2058 {
2059 nfs_fs_proc_net_exit(net);
2060 nfs_cleanup_cb_ident_idr(net);
2061 }
2062
2063 static struct pernet_operations nfs_net_ops = {
2064 .init = nfs_net_init,
2065 .exit = nfs_net_exit,
2066 .id = &nfs_net_id,
2067 .size = sizeof(struct nfs_net),
2068 };
2069
2070 /*
2071 * Initialize NFS
2072 */
2073 static int __init init_nfs_fs(void)
2074 {
2075 int err;
2076
2077 err = register_pernet_subsys(&nfs_net_ops);
2078 if (err < 0)
2079 goto out9;
2080
2081 err = nfs_fscache_register();
2082 if (err < 0)
2083 goto out8;
2084
2085 err = nfsiod_start();
2086 if (err)
2087 goto out7;
2088
2089 err = nfs_fs_proc_init();
2090 if (err)
2091 goto out6;
2092
2093 err = nfs_init_nfspagecache();
2094 if (err)
2095 goto out5;
2096
2097 err = nfs_init_inodecache();
2098 if (err)
2099 goto out4;
2100
2101 err = nfs_init_readpagecache();
2102 if (err)
2103 goto out3;
2104
2105 err = nfs_init_writepagecache();
2106 if (err)
2107 goto out2;
2108
2109 err = nfs_init_directcache();
2110 if (err)
2111 goto out1;
2112
2113 rpc_proc_register(&init_net, &nfs_rpcstat);
2114
2115 err = register_nfs_fs();
2116 if (err)
2117 goto out0;
2118
2119 return 0;
2120 out0:
2121 rpc_proc_unregister(&init_net, "nfs");
2122 nfs_destroy_directcache();
2123 out1:
2124 nfs_destroy_writepagecache();
2125 out2:
2126 nfs_destroy_readpagecache();
2127 out3:
2128 nfs_destroy_inodecache();
2129 out4:
2130 nfs_destroy_nfspagecache();
2131 out5:
2132 nfs_fs_proc_exit();
2133 out6:
2134 nfsiod_stop();
2135 out7:
2136 nfs_fscache_unregister();
2137 out8:
2138 unregister_pernet_subsys(&nfs_net_ops);
2139 out9:
2140 return err;
2141 }
2142
2143 static void __exit exit_nfs_fs(void)
2144 {
2145 nfs_destroy_directcache();
2146 nfs_destroy_writepagecache();
2147 nfs_destroy_readpagecache();
2148 nfs_destroy_inodecache();
2149 nfs_destroy_nfspagecache();
2150 nfs_fscache_unregister();
2151 unregister_pernet_subsys(&nfs_net_ops);
2152 rpc_proc_unregister(&init_net, "nfs");
2153 unregister_nfs_fs();
2154 nfs_fs_proc_exit();
2155 nfsiod_stop();
2156 }
2157
2158 /* Not quite true; I just maintain it */
2159 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2160 MODULE_LICENSE("GPL");
2161 module_param(enable_ino64, bool, 0644);
2162
2163 module_init(init_nfs_fs)
2164 module_exit(exit_nfs_fs)