Loading...
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 */
63static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
64
65static int nfs_update_inode(struct inode *, struct nfs_fattr *);
66
67static struct kmem_cache * nfs_inode_cachep;
68
69static inline unsigned long
70nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71{
72 return nfs_fileid_to_ino_t(fattr->fileid);
73}
74
75static int nfs_wait_killable(int mode)
76{
77 freezable_schedule_unsafe();
78 if (signal_pending_state(mode, current))
79 return -ERESTARTSYS;
80 return 0;
81}
82
83int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
84{
85 return nfs_wait_killable(mode);
86}
87EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
88
89/**
90 * nfs_compat_user_ino64 - returns the user-visible inode number
91 * @fileid: 64-bit fileid
92 *
93 * This function returns a 32-bit inode number if the boot parameter
94 * nfs.enable_ino64 is zero.
95 */
96u64 nfs_compat_user_ino64(u64 fileid)
97{
98#ifdef CONFIG_COMPAT
99 compat_ulong_t ino;
100#else
101 unsigned long ino;
102#endif
103
104 if (enable_ino64)
105 return fileid;
106 ino = fileid;
107 if (sizeof(ino) < sizeof(fileid))
108 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
109 return ino;
110}
111
112int nfs_drop_inode(struct inode *inode)
113{
114 return NFS_STALE(inode) || generic_drop_inode(inode);
115}
116EXPORT_SYMBOL_GPL(nfs_drop_inode);
117
118void nfs_clear_inode(struct inode *inode)
119{
120 /*
121 * The following should never happen...
122 */
123 WARN_ON_ONCE(nfs_have_writebacks(inode));
124 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
125 nfs_zap_acl_cache(inode);
126 nfs_access_zap_cache(inode);
127 nfs_fscache_clear_inode(inode);
128}
129EXPORT_SYMBOL_GPL(nfs_clear_inode);
130
131void nfs_evict_inode(struct inode *inode)
132{
133 truncate_inode_pages_final(&inode->i_data);
134 clear_inode(inode);
135 nfs_clear_inode(inode);
136}
137
138int nfs_sync_inode(struct inode *inode)
139{
140 inode_dio_wait(inode);
141 return nfs_wb_all(inode);
142}
143EXPORT_SYMBOL_GPL(nfs_sync_inode);
144
145/**
146 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
147 * @mapping: pointer to struct address_space
148 */
149int nfs_sync_mapping(struct address_space *mapping)
150{
151 int ret = 0;
152
153 if (mapping->nrpages != 0) {
154 unmap_mapping_range(mapping, 0, 0, 0);
155 ret = nfs_wb_all(mapping->host);
156 }
157 return ret;
158}
159
160static int nfs_attribute_timeout(struct inode *inode)
161{
162 struct nfs_inode *nfsi = NFS_I(inode);
163
164 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
165}
166
167static bool nfs_check_cache_flags_invalid(struct inode *inode,
168 unsigned long flags)
169{
170 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
171
172 return (cache_validity & flags) != 0;
173}
174
175bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
176{
177 if (nfs_check_cache_flags_invalid(inode, flags))
178 return true;
179 return nfs_attribute_cache_expired(inode);
180}
181EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
182
183#ifdef CONFIG_NFS_V4_2
184static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
185{
186 return nfsi->xattr_cache != NULL;
187}
188#else
189static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
190{
191 return false;
192}
193#endif
194
195void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
196{
197 struct nfs_inode *nfsi = NFS_I(inode);
198 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
199
200 if (have_delegation) {
201 if (!(flags & NFS_INO_REVAL_FORCED))
202 flags &= ~(NFS_INO_INVALID_MODE |
203 NFS_INO_INVALID_OTHER |
204 NFS_INO_INVALID_XATTR);
205 flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
206 } else if (flags & NFS_INO_REVAL_PAGECACHE)
207 flags |= NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE;
208
209 if (!nfs_has_xattr_cache(nfsi))
210 flags &= ~NFS_INO_INVALID_XATTR;
211 if (flags & NFS_INO_INVALID_DATA)
212 nfs_fscache_invalidate(inode);
213 if (inode->i_mapping->nrpages == 0)
214 flags &= ~(NFS_INO_INVALID_DATA|NFS_INO_DATA_INVAL_DEFER);
215 flags &= ~(NFS_INO_REVAL_PAGECACHE | NFS_INO_REVAL_FORCED);
216 nfsi->cache_validity |= flags;
217}
218EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
219
220/*
221 * Invalidate the local caches
222 */
223static void nfs_zap_caches_locked(struct inode *inode)
224{
225 struct nfs_inode *nfsi = NFS_I(inode);
226 int mode = inode->i_mode;
227
228 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
229
230 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
231 nfsi->attrtimeo_timestamp = jiffies;
232
233 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
234 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
235 | NFS_INO_INVALID_DATA
236 | NFS_INO_INVALID_ACCESS
237 | NFS_INO_INVALID_ACL
238 | NFS_INO_INVALID_XATTR
239 | NFS_INO_REVAL_PAGECACHE);
240 } else
241 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
242 | NFS_INO_INVALID_ACCESS
243 | NFS_INO_INVALID_ACL
244 | NFS_INO_INVALID_XATTR
245 | NFS_INO_REVAL_PAGECACHE);
246 nfs_zap_label_cache_locked(nfsi);
247}
248
249void nfs_zap_caches(struct inode *inode)
250{
251 spin_lock(&inode->i_lock);
252 nfs_zap_caches_locked(inode);
253 spin_unlock(&inode->i_lock);
254}
255
256void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
257{
258 if (mapping->nrpages != 0) {
259 spin_lock(&inode->i_lock);
260 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
261 spin_unlock(&inode->i_lock);
262 }
263}
264
265void nfs_zap_acl_cache(struct inode *inode)
266{
267 void (*clear_acl_cache)(struct inode *);
268
269 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
270 if (clear_acl_cache != NULL)
271 clear_acl_cache(inode);
272 spin_lock(&inode->i_lock);
273 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
274 spin_unlock(&inode->i_lock);
275}
276EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
277
278void nfs_invalidate_atime(struct inode *inode)
279{
280 spin_lock(&inode->i_lock);
281 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
282 spin_unlock(&inode->i_lock);
283}
284EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
285
286/*
287 * Invalidate, but do not unhash, the inode.
288 * NB: must be called with inode->i_lock held!
289 */
290static void nfs_set_inode_stale_locked(struct inode *inode)
291{
292 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
293 nfs_zap_caches_locked(inode);
294 trace_nfs_set_inode_stale(inode);
295}
296
297void nfs_set_inode_stale(struct inode *inode)
298{
299 spin_lock(&inode->i_lock);
300 nfs_set_inode_stale_locked(inode);
301 spin_unlock(&inode->i_lock);
302}
303
304struct nfs_find_desc {
305 struct nfs_fh *fh;
306 struct nfs_fattr *fattr;
307};
308
309/*
310 * In NFSv3 we can have 64bit inode numbers. In order to support
311 * this, and re-exported directories (also seen in NFSv2)
312 * we are forced to allow 2 different inodes to have the same
313 * i_ino.
314 */
315static int
316nfs_find_actor(struct inode *inode, void *opaque)
317{
318 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
319 struct nfs_fh *fh = desc->fh;
320 struct nfs_fattr *fattr = desc->fattr;
321
322 if (NFS_FILEID(inode) != fattr->fileid)
323 return 0;
324 if (inode_wrong_type(inode, fattr->mode))
325 return 0;
326 if (nfs_compare_fh(NFS_FH(inode), fh))
327 return 0;
328 if (is_bad_inode(inode) || NFS_STALE(inode))
329 return 0;
330 return 1;
331}
332
333static int
334nfs_init_locked(struct inode *inode, void *opaque)
335{
336 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
337 struct nfs_fattr *fattr = desc->fattr;
338
339 set_nfs_fileid(inode, fattr->fileid);
340 inode->i_mode = fattr->mode;
341 nfs_copy_fh(NFS_FH(inode), desc->fh);
342 return 0;
343}
344
345#ifdef CONFIG_NFS_V4_SECURITY_LABEL
346static void nfs_clear_label_invalid(struct inode *inode)
347{
348 spin_lock(&inode->i_lock);
349 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
350 spin_unlock(&inode->i_lock);
351}
352
353void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
354 struct nfs4_label *label)
355{
356 int error;
357
358 if (label == NULL)
359 return;
360
361 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
362 error = security_inode_notifysecctx(inode, label->label,
363 label->len);
364 if (error)
365 printk(KERN_ERR "%s() %s %d "
366 "security_inode_notifysecctx() %d\n",
367 __func__,
368 (char *)label->label,
369 label->len, error);
370 nfs_clear_label_invalid(inode);
371 }
372}
373
374struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
375{
376 struct nfs4_label *label = NULL;
377 int minor_version = server->nfs_client->cl_minorversion;
378
379 if (minor_version < 2)
380 return label;
381
382 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
383 return label;
384
385 label = kzalloc(sizeof(struct nfs4_label), flags);
386 if (label == NULL)
387 return ERR_PTR(-ENOMEM);
388
389 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
390 if (label->label == NULL) {
391 kfree(label);
392 return ERR_PTR(-ENOMEM);
393 }
394 label->len = NFS4_MAXLABELLEN;
395
396 return label;
397}
398EXPORT_SYMBOL_GPL(nfs4_label_alloc);
399#else
400void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
401 struct nfs4_label *label)
402{
403}
404#endif
405EXPORT_SYMBOL_GPL(nfs_setsecurity);
406
407/* Search for inode identified by fh, fileid and i_mode in inode cache. */
408struct inode *
409nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
410{
411 struct nfs_find_desc desc = {
412 .fh = fh,
413 .fattr = fattr,
414 };
415 struct inode *inode;
416 unsigned long hash;
417
418 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
419 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
420 return NULL;
421
422 hash = nfs_fattr_to_ino_t(fattr);
423 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
424
425 dprintk("%s: returning %p\n", __func__, inode);
426 return inode;
427}
428
429/*
430 * This is our front-end to iget that looks up inodes by file handle
431 * instead of inode number.
432 */
433struct inode *
434nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
435{
436 struct nfs_find_desc desc = {
437 .fh = fh,
438 .fattr = fattr
439 };
440 struct inode *inode = ERR_PTR(-ENOENT);
441 u64 fattr_supported = NFS_SB(sb)->fattr_valid;
442 unsigned long hash;
443
444 nfs_attr_check_mountpoint(sb, fattr);
445
446 if (nfs_attr_use_mounted_on_fileid(fattr))
447 fattr->fileid = fattr->mounted_on_fileid;
448 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
449 goto out_no_inode;
450 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
451 goto out_no_inode;
452
453 hash = nfs_fattr_to_ino_t(fattr);
454
455 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
456 if (inode == NULL) {
457 inode = ERR_PTR(-ENOMEM);
458 goto out_no_inode;
459 }
460
461 if (inode->i_state & I_NEW) {
462 struct nfs_inode *nfsi = NFS_I(inode);
463 unsigned long now = jiffies;
464
465 /* We set i_ino for the few things that still rely on it,
466 * such as stat(2) */
467 inode->i_ino = hash;
468
469 /* We can't support update_atime(), since the server will reset it */
470 inode->i_flags |= S_NOATIME|S_NOCMTIME;
471 inode->i_mode = fattr->mode;
472 nfsi->cache_validity = 0;
473 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
474 && (fattr_supported & NFS_ATTR_FATTR_MODE))
475 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
476 /* Why so? Because we want revalidate for devices/FIFOs, and
477 * that's precisely what we have in nfs_file_inode_operations.
478 */
479 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
480 if (S_ISREG(inode->i_mode)) {
481 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
482 inode->i_data.a_ops = &nfs_file_aops;
483 } else if (S_ISDIR(inode->i_mode)) {
484 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
485 inode->i_fop = &nfs_dir_operations;
486 inode->i_data.a_ops = &nfs_dir_aops;
487 /* Deal with crossing mountpoints */
488 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
489 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
490 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
491 inode->i_op = &nfs_referral_inode_operations;
492 else
493 inode->i_op = &nfs_mountpoint_inode_operations;
494 inode->i_fop = NULL;
495 inode->i_flags |= S_AUTOMOUNT;
496 }
497 } else if (S_ISLNK(inode->i_mode)) {
498 inode->i_op = &nfs_symlink_inode_operations;
499 inode_nohighmem(inode);
500 } else
501 init_special_inode(inode, inode->i_mode, fattr->rdev);
502
503 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
504 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
505 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
506 inode_set_iversion_raw(inode, 0);
507 inode->i_size = 0;
508 clear_nlink(inode);
509 inode->i_uid = make_kuid(&init_user_ns, -2);
510 inode->i_gid = make_kgid(&init_user_ns, -2);
511 inode->i_blocks = 0;
512 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
513 nfsi->write_io = 0;
514 nfsi->read_io = 0;
515
516 nfsi->read_cache_jiffies = fattr->time_start;
517 nfsi->attr_gencount = fattr->gencount;
518 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
519 inode->i_atime = fattr->atime;
520 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
521 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
522 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
523 inode->i_mtime = fattr->mtime;
524 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
525 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
526 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
527 inode->i_ctime = fattr->ctime;
528 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
529 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
530 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
531 inode_set_iversion_raw(inode, fattr->change_attr);
532 else
533 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
534 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
535 inode->i_size = nfs_size_to_loff_t(fattr->size);
536 else
537 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
538 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
539 set_nlink(inode, fattr->nlink);
540 else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
541 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
542 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
543 inode->i_uid = fattr->uid;
544 else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
545 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
546 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
547 inode->i_gid = fattr->gid;
548 else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
549 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
550 if (nfs_server_capable(inode, NFS_CAP_XATTR))
551 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
552 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
553 inode->i_blocks = fattr->du.nfs2.blocks;
554 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
555 fattr->size != 0)
556 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
557 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
558 /*
559 * report the blocks in 512byte units
560 */
561 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
562 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
563 fattr->size != 0)
564 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
565
566 nfs_setsecurity(inode, fattr, label);
567
568 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
569 nfsi->attrtimeo_timestamp = now;
570 nfsi->access_cache = RB_ROOT;
571
572 nfs_fscache_init_inode(inode);
573
574 unlock_new_inode(inode);
575 } else {
576 int err = nfs_refresh_inode(inode, fattr);
577 if (err < 0) {
578 iput(inode);
579 inode = ERR_PTR(err);
580 goto out_no_inode;
581 }
582 }
583 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
584 inode->i_sb->s_id,
585 (unsigned long long)NFS_FILEID(inode),
586 nfs_display_fhandle_hash(fh),
587 atomic_read(&inode->i_count));
588
589out:
590 return inode;
591
592out_no_inode:
593 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
594 goto out;
595}
596EXPORT_SYMBOL_GPL(nfs_fhget);
597
598#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)
599
600int
601nfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
602 struct iattr *attr)
603{
604 struct inode *inode = d_inode(dentry);
605 struct nfs_fattr *fattr;
606 int error = 0;
607
608 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
609
610 /* skip mode change if it's just for clearing setuid/setgid */
611 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
612 attr->ia_valid &= ~ATTR_MODE;
613
614 if (attr->ia_valid & ATTR_SIZE) {
615 BUG_ON(!S_ISREG(inode->i_mode));
616
617 error = inode_newsize_ok(inode, attr->ia_size);
618 if (error)
619 return error;
620
621 if (attr->ia_size == i_size_read(inode))
622 attr->ia_valid &= ~ATTR_SIZE;
623 }
624
625 /* Optimization: if the end result is no change, don't RPC */
626 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
627 return 0;
628
629 trace_nfs_setattr_enter(inode);
630
631 /* Write all dirty data */
632 if (S_ISREG(inode->i_mode))
633 nfs_sync_inode(inode);
634
635 fattr = nfs_alloc_fattr();
636 if (fattr == NULL) {
637 error = -ENOMEM;
638 goto out;
639 }
640
641 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
642 if (error == 0)
643 error = nfs_refresh_inode(inode, fattr);
644 nfs_free_fattr(fattr);
645out:
646 trace_nfs_setattr_exit(inode, error);
647 return error;
648}
649EXPORT_SYMBOL_GPL(nfs_setattr);
650
651/**
652 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
653 * @inode: inode of the file used
654 * @offset: file offset to start truncating
655 *
656 * This is a copy of the common vmtruncate, but with the locking
657 * corrected to take into account the fact that NFS requires
658 * inode->i_size to be updated under the inode->i_lock.
659 * Note: must be called with inode->i_lock held!
660 */
661static int nfs_vmtruncate(struct inode * inode, loff_t offset)
662{
663 int err;
664
665 err = inode_newsize_ok(inode, offset);
666 if (err)
667 goto out;
668
669 i_size_write(inode, offset);
670 /* Optimisation */
671 if (offset == 0)
672 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
673 NFS_INO_DATA_INVAL_DEFER);
674 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
675
676 spin_unlock(&inode->i_lock);
677 truncate_pagecache(inode, offset);
678 spin_lock(&inode->i_lock);
679out:
680 return err;
681}
682
683/**
684 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
685 * @inode: pointer to struct inode
686 * @attr: pointer to struct iattr
687 * @fattr: pointer to struct nfs_fattr
688 *
689 * Note: we do this in the *proc.c in order to ensure that
690 * it works for things like exclusive creates too.
691 */
692void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
693 struct nfs_fattr *fattr)
694{
695 /* Barrier: bump the attribute generation count. */
696 nfs_fattr_set_barrier(fattr);
697
698 spin_lock(&inode->i_lock);
699 NFS_I(inode)->attr_gencount = fattr->gencount;
700 if ((attr->ia_valid & ATTR_SIZE) != 0) {
701 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME |
702 NFS_INO_INVALID_BLOCKS);
703 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
704 nfs_vmtruncate(inode, attr->ia_size);
705 }
706 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
707 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
708 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
709 inode->i_mode & S_ISUID)
710 inode->i_mode &= ~S_ISUID;
711 if ((attr->ia_valid & ATTR_KILL_SGID) != 0 &&
712 (inode->i_mode & (S_ISGID | S_IXGRP)) ==
713 (S_ISGID | S_IXGRP))
714 inode->i_mode &= ~S_ISGID;
715 if ((attr->ia_valid & ATTR_MODE) != 0) {
716 int mode = attr->ia_mode & S_IALLUGO;
717 mode |= inode->i_mode & ~S_IALLUGO;
718 inode->i_mode = mode;
719 }
720 if ((attr->ia_valid & ATTR_UID) != 0)
721 inode->i_uid = attr->ia_uid;
722 if ((attr->ia_valid & ATTR_GID) != 0)
723 inode->i_gid = attr->ia_gid;
724 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
725 inode->i_ctime = fattr->ctime;
726 else
727 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
728 | NFS_INO_INVALID_CTIME);
729 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
730 | NFS_INO_INVALID_ACL);
731 }
732 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
733 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
734 | NFS_INO_INVALID_CTIME);
735 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
736 inode->i_atime = fattr->atime;
737 else if (attr->ia_valid & ATTR_ATIME_SET)
738 inode->i_atime = attr->ia_atime;
739 else
740 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
741
742 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
743 inode->i_ctime = fattr->ctime;
744 else
745 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
746 | NFS_INO_INVALID_CTIME);
747 }
748 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
749 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
750 | NFS_INO_INVALID_CTIME);
751 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
752 inode->i_mtime = fattr->mtime;
753 else if (attr->ia_valid & ATTR_MTIME_SET)
754 inode->i_mtime = attr->ia_mtime;
755 else
756 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
757
758 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
759 inode->i_ctime = fattr->ctime;
760 else
761 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
762 | NFS_INO_INVALID_CTIME);
763 }
764 if (fattr->valid)
765 nfs_update_inode(inode, fattr);
766 spin_unlock(&inode->i_lock);
767}
768EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
769
770static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
771{
772 struct dentry *parent;
773
774 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
775 return;
776 parent = dget_parent(dentry);
777 nfs_force_use_readdirplus(d_inode(parent));
778 dput(parent);
779}
780
781static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
782{
783 struct dentry *parent;
784
785 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
786 return;
787 parent = dget_parent(dentry);
788 nfs_advise_use_readdirplus(d_inode(parent));
789 dput(parent);
790}
791
792static u32 nfs_get_valid_attrmask(struct inode *inode)
793{
794 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
795 u32 reply_mask = STATX_INO | STATX_TYPE;
796
797 if (!(cache_validity & NFS_INO_INVALID_ATIME))
798 reply_mask |= STATX_ATIME;
799 if (!(cache_validity & NFS_INO_INVALID_CTIME))
800 reply_mask |= STATX_CTIME;
801 if (!(cache_validity & NFS_INO_INVALID_MTIME))
802 reply_mask |= STATX_MTIME;
803 if (!(cache_validity & NFS_INO_INVALID_SIZE))
804 reply_mask |= STATX_SIZE;
805 if (!(cache_validity & NFS_INO_INVALID_NLINK))
806 reply_mask |= STATX_NLINK;
807 if (!(cache_validity & NFS_INO_INVALID_MODE))
808 reply_mask |= STATX_MODE;
809 if (!(cache_validity & NFS_INO_INVALID_OTHER))
810 reply_mask |= STATX_UID | STATX_GID;
811 if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
812 reply_mask |= STATX_BLOCKS;
813 return reply_mask;
814}
815
816int nfs_getattr(struct user_namespace *mnt_userns, const struct path *path,
817 struct kstat *stat, u32 request_mask, unsigned int query_flags)
818{
819 struct inode *inode = d_inode(path->dentry);
820 struct nfs_server *server = NFS_SERVER(inode);
821 unsigned long cache_validity;
822 int err = 0;
823 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
824 bool do_update = false;
825
826 trace_nfs_getattr_enter(inode);
827
828 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
829 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
830 STATX_INO | STATX_SIZE | STATX_BLOCKS;
831
832 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
833 nfs_readdirplus_parent_cache_hit(path->dentry);
834 goto out_no_revalidate;
835 }
836
837 /* Flush out writes to the server in order to update c/mtime. */
838 if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
839 S_ISREG(inode->i_mode)) {
840 err = filemap_write_and_wait(inode->i_mapping);
841 if (err)
842 goto out;
843 }
844
845 /*
846 * We may force a getattr if the user cares about atime.
847 *
848 * Note that we only have to check the vfsmount flags here:
849 * - NFS always sets S_NOATIME by so checking it would give a
850 * bogus result
851 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
852 * no point in checking those.
853 */
854 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
855 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
856 request_mask &= ~STATX_ATIME;
857
858 /* Is the user requesting attributes that might need revalidation? */
859 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
860 STATX_MTIME|STATX_UID|STATX_GID|
861 STATX_SIZE|STATX_BLOCKS)))
862 goto out_no_revalidate;
863
864 /* Check whether the cached attributes are stale */
865 do_update |= force_sync || nfs_attribute_cache_expired(inode);
866 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
867 do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
868 if (request_mask & STATX_ATIME)
869 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
870 if (request_mask & STATX_CTIME)
871 do_update |= cache_validity & NFS_INO_INVALID_CTIME;
872 if (request_mask & STATX_MTIME)
873 do_update |= cache_validity & NFS_INO_INVALID_MTIME;
874 if (request_mask & STATX_SIZE)
875 do_update |= cache_validity & NFS_INO_INVALID_SIZE;
876 if (request_mask & STATX_NLINK)
877 do_update |= cache_validity & NFS_INO_INVALID_NLINK;
878 if (request_mask & STATX_MODE)
879 do_update |= cache_validity & NFS_INO_INVALID_MODE;
880 if (request_mask & (STATX_UID | STATX_GID))
881 do_update |= cache_validity & NFS_INO_INVALID_OTHER;
882 if (request_mask & STATX_BLOCKS)
883 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
884
885 if (do_update) {
886 /* Update the attribute cache */
887 if (!(server->flags & NFS_MOUNT_NOAC))
888 nfs_readdirplus_parent_cache_miss(path->dentry);
889 else
890 nfs_readdirplus_parent_cache_hit(path->dentry);
891 err = __nfs_revalidate_inode(server, inode);
892 if (err)
893 goto out;
894 } else
895 nfs_readdirplus_parent_cache_hit(path->dentry);
896out_no_revalidate:
897 /* Only return attributes that were revalidated. */
898 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
899
900 generic_fillattr(&init_user_ns, inode, stat);
901 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
902 if (S_ISDIR(inode->i_mode))
903 stat->blksize = NFS_SERVER(inode)->dtsize;
904out:
905 trace_nfs_getattr_exit(inode, err);
906 return err;
907}
908EXPORT_SYMBOL_GPL(nfs_getattr);
909
910static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
911{
912 refcount_set(&l_ctx->count, 1);
913 l_ctx->lockowner = current->files;
914 INIT_LIST_HEAD(&l_ctx->list);
915 atomic_set(&l_ctx->io_count, 0);
916}
917
918static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
919{
920 struct nfs_lock_context *pos;
921
922 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
923 if (pos->lockowner != current->files)
924 continue;
925 if (refcount_inc_not_zero(&pos->count))
926 return pos;
927 }
928 return NULL;
929}
930
931struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
932{
933 struct nfs_lock_context *res, *new = NULL;
934 struct inode *inode = d_inode(ctx->dentry);
935
936 rcu_read_lock();
937 res = __nfs_find_lock_context(ctx);
938 rcu_read_unlock();
939 if (res == NULL) {
940 new = kmalloc(sizeof(*new), GFP_KERNEL);
941 if (new == NULL)
942 return ERR_PTR(-ENOMEM);
943 nfs_init_lock_context(new);
944 spin_lock(&inode->i_lock);
945 res = __nfs_find_lock_context(ctx);
946 if (res == NULL) {
947 new->open_context = get_nfs_open_context(ctx);
948 if (new->open_context) {
949 list_add_tail_rcu(&new->list,
950 &ctx->lock_context.list);
951 res = new;
952 new = NULL;
953 } else
954 res = ERR_PTR(-EBADF);
955 }
956 spin_unlock(&inode->i_lock);
957 kfree(new);
958 }
959 return res;
960}
961EXPORT_SYMBOL_GPL(nfs_get_lock_context);
962
963void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
964{
965 struct nfs_open_context *ctx = l_ctx->open_context;
966 struct inode *inode = d_inode(ctx->dentry);
967
968 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
969 return;
970 list_del_rcu(&l_ctx->list);
971 spin_unlock(&inode->i_lock);
972 put_nfs_open_context(ctx);
973 kfree_rcu(l_ctx, rcu_head);
974}
975EXPORT_SYMBOL_GPL(nfs_put_lock_context);
976
977/**
978 * nfs_close_context - Common close_context() routine NFSv2/v3
979 * @ctx: pointer to context
980 * @is_sync: is this a synchronous close
981 *
982 * Ensure that the attributes are up to date if we're mounted
983 * with close-to-open semantics and we have cached data that will
984 * need to be revalidated on open.
985 */
986void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
987{
988 struct nfs_inode *nfsi;
989 struct inode *inode;
990
991 if (!(ctx->mode & FMODE_WRITE))
992 return;
993 if (!is_sync)
994 return;
995 inode = d_inode(ctx->dentry);
996 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
997 return;
998 nfsi = NFS_I(inode);
999 if (inode->i_mapping->nrpages == 0)
1000 return;
1001 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1002 return;
1003 if (!list_empty(&nfsi->open_files))
1004 return;
1005 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1006 return;
1007 nfs_revalidate_inode(inode,
1008 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1009}
1010EXPORT_SYMBOL_GPL(nfs_close_context);
1011
1012struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1013 fmode_t f_mode,
1014 struct file *filp)
1015{
1016 struct nfs_open_context *ctx;
1017
1018 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
1019 if (!ctx)
1020 return ERR_PTR(-ENOMEM);
1021 nfs_sb_active(dentry->d_sb);
1022 ctx->dentry = dget(dentry);
1023 if (filp)
1024 ctx->cred = get_cred(filp->f_cred);
1025 else
1026 ctx->cred = get_current_cred();
1027 ctx->ll_cred = NULL;
1028 ctx->state = NULL;
1029 ctx->mode = f_mode;
1030 ctx->flags = 0;
1031 ctx->error = 0;
1032 ctx->flock_owner = (fl_owner_t)filp;
1033 nfs_init_lock_context(&ctx->lock_context);
1034 ctx->lock_context.open_context = ctx;
1035 INIT_LIST_HEAD(&ctx->list);
1036 ctx->mdsthreshold = NULL;
1037 return ctx;
1038}
1039EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1040
1041struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1042{
1043 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1044 return ctx;
1045 return NULL;
1046}
1047EXPORT_SYMBOL_GPL(get_nfs_open_context);
1048
1049static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1050{
1051 struct inode *inode = d_inode(ctx->dentry);
1052 struct super_block *sb = ctx->dentry->d_sb;
1053
1054 if (!refcount_dec_and_test(&ctx->lock_context.count))
1055 return;
1056 if (!list_empty(&ctx->list)) {
1057 spin_lock(&inode->i_lock);
1058 list_del_rcu(&ctx->list);
1059 spin_unlock(&inode->i_lock);
1060 }
1061 if (inode != NULL)
1062 NFS_PROTO(inode)->close_context(ctx, is_sync);
1063 put_cred(ctx->cred);
1064 dput(ctx->dentry);
1065 nfs_sb_deactive(sb);
1066 put_rpccred(ctx->ll_cred);
1067 kfree(ctx->mdsthreshold);
1068 kfree_rcu(ctx, rcu_head);
1069}
1070
1071void put_nfs_open_context(struct nfs_open_context *ctx)
1072{
1073 __put_nfs_open_context(ctx, 0);
1074}
1075EXPORT_SYMBOL_GPL(put_nfs_open_context);
1076
1077static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1078{
1079 __put_nfs_open_context(ctx, 1);
1080}
1081
1082/*
1083 * Ensure that mmap has a recent RPC credential for use when writing out
1084 * shared pages
1085 */
1086void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1087{
1088 struct inode *inode = d_inode(ctx->dentry);
1089 struct nfs_inode *nfsi = NFS_I(inode);
1090
1091 spin_lock(&inode->i_lock);
1092 if (list_empty(&nfsi->open_files) &&
1093 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1094 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1095 NFS_INO_REVAL_FORCED);
1096 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1097 spin_unlock(&inode->i_lock);
1098}
1099EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1100
1101void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1102{
1103 filp->private_data = get_nfs_open_context(ctx);
1104 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1105 if (list_empty(&ctx->list))
1106 nfs_inode_attach_open_context(ctx);
1107}
1108EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1109
1110/*
1111 * Given an inode, search for an open context with the desired characteristics
1112 */
1113struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1114{
1115 struct nfs_inode *nfsi = NFS_I(inode);
1116 struct nfs_open_context *pos, *ctx = NULL;
1117
1118 rcu_read_lock();
1119 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1120 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1121 continue;
1122 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1123 continue;
1124 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1125 continue;
1126 ctx = get_nfs_open_context(pos);
1127 if (ctx)
1128 break;
1129 }
1130 rcu_read_unlock();
1131 return ctx;
1132}
1133
1134void nfs_file_clear_open_context(struct file *filp)
1135{
1136 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1137
1138 if (ctx) {
1139 struct inode *inode = d_inode(ctx->dentry);
1140
1141 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1142 /*
1143 * We fatal error on write before. Try to writeback
1144 * every page again.
1145 */
1146 if (ctx->error < 0)
1147 invalidate_inode_pages2(inode->i_mapping);
1148 filp->private_data = NULL;
1149 put_nfs_open_context_sync(ctx);
1150 }
1151}
1152
1153/*
1154 * These allocate and release file read/write context information.
1155 */
1156int nfs_open(struct inode *inode, struct file *filp)
1157{
1158 struct nfs_open_context *ctx;
1159
1160 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1161 if (IS_ERR(ctx))
1162 return PTR_ERR(ctx);
1163 nfs_file_set_open_context(filp, ctx);
1164 put_nfs_open_context(ctx);
1165 nfs_fscache_open_file(inode, filp);
1166 return 0;
1167}
1168EXPORT_SYMBOL_GPL(nfs_open);
1169
1170/*
1171 * This function is called whenever some part of NFS notices that
1172 * the cached attributes have to be refreshed.
1173 */
1174int
1175__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1176{
1177 int status = -ESTALE;
1178 struct nfs4_label *label = NULL;
1179 struct nfs_fattr *fattr = NULL;
1180 struct nfs_inode *nfsi = NFS_I(inode);
1181
1182 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1183 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1184
1185 trace_nfs_revalidate_inode_enter(inode);
1186
1187 if (is_bad_inode(inode))
1188 goto out;
1189 if (NFS_STALE(inode))
1190 goto out;
1191
1192 /* pNFS: Attributes aren't updated until we layoutcommit */
1193 if (S_ISREG(inode->i_mode)) {
1194 status = pnfs_sync_inode(inode, false);
1195 if (status)
1196 goto out;
1197 }
1198
1199 status = -ENOMEM;
1200 fattr = nfs_alloc_fattr();
1201 if (fattr == NULL)
1202 goto out;
1203
1204 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1205
1206 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1207 if (IS_ERR(label)) {
1208 status = PTR_ERR(label);
1209 goto out;
1210 }
1211
1212 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1213 label, inode);
1214 if (status != 0) {
1215 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1216 inode->i_sb->s_id,
1217 (unsigned long long)NFS_FILEID(inode), status);
1218 switch (status) {
1219 case -ETIMEDOUT:
1220 /* A soft timeout occurred. Use cached information? */
1221 if (server->flags & NFS_MOUNT_SOFTREVAL)
1222 status = 0;
1223 break;
1224 case -ESTALE:
1225 if (!S_ISDIR(inode->i_mode))
1226 nfs_set_inode_stale(inode);
1227 else
1228 nfs_zap_caches(inode);
1229 }
1230 goto err_out;
1231 }
1232
1233 status = nfs_refresh_inode(inode, fattr);
1234 if (status) {
1235 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1236 inode->i_sb->s_id,
1237 (unsigned long long)NFS_FILEID(inode), status);
1238 goto err_out;
1239 }
1240
1241 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1242 nfs_zap_acl_cache(inode);
1243
1244 nfs_setsecurity(inode, fattr, label);
1245
1246 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1247 inode->i_sb->s_id,
1248 (unsigned long long)NFS_FILEID(inode));
1249
1250err_out:
1251 nfs4_label_free(label);
1252out:
1253 nfs_free_fattr(fattr);
1254 trace_nfs_revalidate_inode_exit(inode, status);
1255 return status;
1256}
1257
1258int nfs_attribute_cache_expired(struct inode *inode)
1259{
1260 if (nfs_have_delegated_attributes(inode))
1261 return 0;
1262 return nfs_attribute_timeout(inode);
1263}
1264
1265/**
1266 * nfs_revalidate_inode - Revalidate the inode attributes
1267 * @inode: pointer to inode struct
1268 * @flags: cache flags to check
1269 *
1270 * Updates inode attribute information by retrieving the data from the server.
1271 */
1272int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1273{
1274 if (!nfs_check_cache_invalid(inode, flags))
1275 return NFS_STALE(inode) ? -ESTALE : 0;
1276 return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1277}
1278EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1279
1280static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1281{
1282 int ret;
1283
1284 if (mapping->nrpages != 0) {
1285 if (S_ISREG(inode->i_mode)) {
1286 ret = nfs_sync_mapping(mapping);
1287 if (ret < 0)
1288 return ret;
1289 }
1290 ret = invalidate_inode_pages2(mapping);
1291 if (ret < 0)
1292 return ret;
1293 }
1294 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1295 nfs_fscache_wait_on_invalidate(inode);
1296
1297 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1298 inode->i_sb->s_id,
1299 (unsigned long long)NFS_FILEID(inode));
1300 return 0;
1301}
1302
1303/**
1304 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1305 * @mapping: pointer to mapping
1306 *
1307 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1308 */
1309int nfs_clear_invalid_mapping(struct address_space *mapping)
1310{
1311 struct inode *inode = mapping->host;
1312 struct nfs_inode *nfsi = NFS_I(inode);
1313 unsigned long *bitlock = &nfsi->flags;
1314 int ret = 0;
1315
1316 /*
1317 * We must clear NFS_INO_INVALID_DATA first to ensure that
1318 * invalidations that come in while we're shooting down the mappings
1319 * are respected. But, that leaves a race window where one revalidator
1320 * can clear the flag, and then another checks it before the mapping
1321 * gets invalidated. Fix that by serializing access to this part of
1322 * the function.
1323 *
1324 * At the same time, we need to allow other tasks to see whether we
1325 * might be in the middle of invalidating the pages, so we only set
1326 * the bit lock here if it looks like we're going to be doing that.
1327 */
1328 for (;;) {
1329 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1330 nfs_wait_bit_killable, TASK_KILLABLE);
1331 if (ret)
1332 goto out;
1333 spin_lock(&inode->i_lock);
1334 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1335 spin_unlock(&inode->i_lock);
1336 continue;
1337 }
1338 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1339 break;
1340 spin_unlock(&inode->i_lock);
1341 goto out;
1342 }
1343
1344 set_bit(NFS_INO_INVALIDATING, bitlock);
1345 smp_wmb();
1346 nfsi->cache_validity &=
1347 ~(NFS_INO_INVALID_DATA | NFS_INO_DATA_INVAL_DEFER);
1348 spin_unlock(&inode->i_lock);
1349 trace_nfs_invalidate_mapping_enter(inode);
1350 ret = nfs_invalidate_mapping(inode, mapping);
1351 trace_nfs_invalidate_mapping_exit(inode, ret);
1352
1353 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1354 smp_mb__after_atomic();
1355 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1356out:
1357 return ret;
1358}
1359
1360bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1361{
1362 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1363 NFS_STALE(inode);
1364}
1365
1366int nfs_revalidate_mapping_rcu(struct inode *inode)
1367{
1368 struct nfs_inode *nfsi = NFS_I(inode);
1369 unsigned long *bitlock = &nfsi->flags;
1370 int ret = 0;
1371
1372 if (IS_SWAPFILE(inode))
1373 goto out;
1374 if (nfs_mapping_need_revalidate_inode(inode)) {
1375 ret = -ECHILD;
1376 goto out;
1377 }
1378 spin_lock(&inode->i_lock);
1379 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1380 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1381 ret = -ECHILD;
1382 spin_unlock(&inode->i_lock);
1383out:
1384 return ret;
1385}
1386
1387/**
1388 * nfs_revalidate_mapping - Revalidate the pagecache
1389 * @inode: pointer to host inode
1390 * @mapping: pointer to mapping
1391 */
1392int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1393{
1394 /* swapfiles are not supposed to be shared. */
1395 if (IS_SWAPFILE(inode))
1396 return 0;
1397
1398 if (nfs_mapping_need_revalidate_inode(inode)) {
1399 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1400 if (ret < 0)
1401 return ret;
1402 }
1403
1404 return nfs_clear_invalid_mapping(mapping);
1405}
1406
1407static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1408{
1409 struct inode *inode = &nfsi->vfs_inode;
1410
1411 if (!S_ISREG(inode->i_mode))
1412 return false;
1413 if (list_empty(&nfsi->open_files))
1414 return false;
1415 return inode_is_open_for_write(inode);
1416}
1417
1418static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1419{
1420 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1421}
1422
1423static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1424{
1425 struct timespec64 ts;
1426
1427 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1428 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1429 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1430 inode_set_iversion_raw(inode, fattr->change_attr);
1431 if (S_ISDIR(inode->i_mode))
1432 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1433 else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1434 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1435 }
1436 /* If we have atomic WCC data, we may update some attributes */
1437 ts = inode->i_ctime;
1438 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1439 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1440 && timespec64_equal(&ts, &fattr->pre_ctime)) {
1441 inode->i_ctime = fattr->ctime;
1442 }
1443
1444 ts = inode->i_mtime;
1445 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1446 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1447 && timespec64_equal(&ts, &fattr->pre_mtime)) {
1448 inode->i_mtime = fattr->mtime;
1449 if (S_ISDIR(inode->i_mode))
1450 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1451 }
1452 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1453 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1454 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1455 && !nfs_have_writebacks(inode)) {
1456 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1457 }
1458}
1459
1460/**
1461 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1462 * @inode: pointer to inode
1463 * @fattr: updated attributes
1464 *
1465 * Verifies the attribute cache. If we have just changed the attributes,
1466 * so that fattr carries weak cache consistency data, then it may
1467 * also update the ctime/mtime/change_attribute.
1468 */
1469static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1470{
1471 struct nfs_inode *nfsi = NFS_I(inode);
1472 loff_t cur_size, new_isize;
1473 unsigned long invalid = 0;
1474 struct timespec64 ts;
1475
1476 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1477 return 0;
1478
1479 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1480 /* Only a mounted-on-fileid? Just exit */
1481 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1482 return 0;
1483 /* Has the inode gone and changed behind our back? */
1484 } else if (nfsi->fileid != fattr->fileid) {
1485 /* Is this perhaps the mounted-on fileid? */
1486 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1487 nfsi->fileid == fattr->mounted_on_fileid)
1488 return 0;
1489 return -ESTALE;
1490 }
1491 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1492 return -ESTALE;
1493
1494
1495 if (!nfs_file_has_buffered_writers(nfsi)) {
1496 /* Verify a few of the more important attributes */
1497 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1498 invalid |= NFS_INO_INVALID_CHANGE;
1499
1500 ts = inode->i_mtime;
1501 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1502 invalid |= NFS_INO_INVALID_MTIME;
1503
1504 ts = inode->i_ctime;
1505 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1506 invalid |= NFS_INO_INVALID_CTIME;
1507
1508 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1509 cur_size = i_size_read(inode);
1510 new_isize = nfs_size_to_loff_t(fattr->size);
1511 if (cur_size != new_isize)
1512 invalid |= NFS_INO_INVALID_SIZE;
1513 }
1514 }
1515
1516 /* Have any file permissions changed? */
1517 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1518 invalid |= NFS_INO_INVALID_MODE;
1519 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1520 invalid |= NFS_INO_INVALID_OTHER;
1521 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1522 invalid |= NFS_INO_INVALID_OTHER;
1523
1524 /* Has the link count changed? */
1525 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1526 invalid |= NFS_INO_INVALID_NLINK;
1527
1528 ts = inode->i_atime;
1529 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1530 invalid |= NFS_INO_INVALID_ATIME;
1531
1532 if (invalid != 0)
1533 nfs_set_cache_invalid(inode, invalid);
1534
1535 nfsi->read_cache_jiffies = fattr->time_start;
1536 return 0;
1537}
1538
1539static atomic_long_t nfs_attr_generation_counter;
1540
1541static unsigned long nfs_read_attr_generation_counter(void)
1542{
1543 return atomic_long_read(&nfs_attr_generation_counter);
1544}
1545
1546unsigned long nfs_inc_attr_generation_counter(void)
1547{
1548 return atomic_long_inc_return(&nfs_attr_generation_counter);
1549}
1550EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1551
1552void nfs_fattr_init(struct nfs_fattr *fattr)
1553{
1554 fattr->valid = 0;
1555 fattr->time_start = jiffies;
1556 fattr->gencount = nfs_inc_attr_generation_counter();
1557 fattr->owner_name = NULL;
1558 fattr->group_name = NULL;
1559}
1560EXPORT_SYMBOL_GPL(nfs_fattr_init);
1561
1562/**
1563 * nfs_fattr_set_barrier
1564 * @fattr: attributes
1565 *
1566 * Used to set a barrier after an attribute was updated. This
1567 * barrier ensures that older attributes from RPC calls that may
1568 * have raced with our update cannot clobber these new values.
1569 * Note that you are still responsible for ensuring that other
1570 * operations which change the attribute on the server do not
1571 * collide.
1572 */
1573void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1574{
1575 fattr->gencount = nfs_inc_attr_generation_counter();
1576}
1577
1578struct nfs_fattr *nfs_alloc_fattr(void)
1579{
1580 struct nfs_fattr *fattr;
1581
1582 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1583 if (fattr != NULL)
1584 nfs_fattr_init(fattr);
1585 return fattr;
1586}
1587EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1588
1589struct nfs_fh *nfs_alloc_fhandle(void)
1590{
1591 struct nfs_fh *fh;
1592
1593 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1594 if (fh != NULL)
1595 fh->size = 0;
1596 return fh;
1597}
1598EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1599
1600#ifdef NFS_DEBUG
1601/*
1602 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1603 * in the same way that wireshark does
1604 *
1605 * @fh: file handle
1606 *
1607 * For debugging only.
1608 */
1609u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1610{
1611 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1612 * not on the result */
1613 return nfs_fhandle_hash(fh);
1614}
1615EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1616
1617/*
1618 * _nfs_display_fhandle - display an NFS file handle on the console
1619 *
1620 * @fh: file handle to display
1621 * @caption: display caption
1622 *
1623 * For debugging only.
1624 */
1625void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1626{
1627 unsigned short i;
1628
1629 if (fh == NULL || fh->size == 0) {
1630 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1631 return;
1632 }
1633
1634 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1635 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1636 for (i = 0; i < fh->size; i += 16) {
1637 __be32 *pos = (__be32 *)&fh->data[i];
1638
1639 switch ((fh->size - i - 1) >> 2) {
1640 case 0:
1641 printk(KERN_DEFAULT " %08x\n",
1642 be32_to_cpup(pos));
1643 break;
1644 case 1:
1645 printk(KERN_DEFAULT " %08x %08x\n",
1646 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1647 break;
1648 case 2:
1649 printk(KERN_DEFAULT " %08x %08x %08x\n",
1650 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1651 be32_to_cpup(pos + 2));
1652 break;
1653 default:
1654 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1655 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1656 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1657 }
1658 }
1659}
1660EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1661#endif
1662
1663/**
1664 * nfs_inode_attrs_cmp_generic - compare attributes
1665 * @fattr: attributes
1666 * @inode: pointer to inode
1667 *
1668 * Attempt to divine whether or not an RPC call reply carrying stale
1669 * attributes got scheduled after another call carrying updated ones.
1670 * Note also the check for wraparound of 'attr_gencount'
1671 *
1672 * The function returns '1' if it thinks the attributes in @fattr are
1673 * more recent than the ones cached in @inode. Otherwise it returns
1674 * the value '0'.
1675 */
1676static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1677 const struct inode *inode)
1678{
1679 unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1680
1681 return (long)(fattr->gencount - attr_gencount) > 0 ||
1682 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1683}
1684
1685/**
1686 * nfs_inode_attrs_cmp_monotonic - compare attributes
1687 * @fattr: attributes
1688 * @inode: pointer to inode
1689 *
1690 * Attempt to divine whether or not an RPC call reply carrying stale
1691 * attributes got scheduled after another call carrying updated ones.
1692 *
1693 * We assume that the server observes monotonic semantics for
1694 * the change attribute, so a larger value means that the attributes in
1695 * @fattr are more recent, in which case the function returns the
1696 * value '1'.
1697 * A return value of '0' indicates no measurable change
1698 * A return value of '-1' means that the attributes in @inode are
1699 * more recent.
1700 */
1701static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1702 const struct inode *inode)
1703{
1704 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1705 if (diff > 0)
1706 return 1;
1707 return diff == 0 ? 0 : -1;
1708}
1709
1710/**
1711 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1712 * @fattr: attributes
1713 * @inode: pointer to inode
1714 *
1715 * Attempt to divine whether or not an RPC call reply carrying stale
1716 * attributes got scheduled after another call carrying updated ones.
1717 *
1718 * We assume that the server observes strictly monotonic semantics for
1719 * the change attribute, so a larger value means that the attributes in
1720 * @fattr are more recent, in which case the function returns the
1721 * value '1'.
1722 * A return value of '-1' means that the attributes in @inode are
1723 * more recent or unchanged.
1724 */
1725static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1726 const struct inode *inode)
1727{
1728 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1729}
1730
1731/**
1732 * nfs_inode_attrs_cmp - compare attributes
1733 * @fattr: attributes
1734 * @inode: pointer to inode
1735 *
1736 * This function returns '1' if it thinks the attributes in @fattr are
1737 * more recent than the ones cached in @inode. It returns '-1' if
1738 * the attributes in @inode are more recent than the ones in @fattr,
1739 * and it returns 0 if not sure.
1740 */
1741static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1742 const struct inode *inode)
1743{
1744 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1745 return 1;
1746 switch (NFS_SERVER(inode)->change_attr_type) {
1747 case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1748 break;
1749 case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1750 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1751 break;
1752 return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1753 default:
1754 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1755 break;
1756 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1757 }
1758 return 0;
1759}
1760
1761/**
1762 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1763 * @fattr: attributes
1764 * @inode: pointer to inode
1765 *
1766 * Returns '1' if the last attribute update left the inode cached
1767 * attributes in a partially unrevalidated state, and @fattr
1768 * matches the change attribute of that partial update.
1769 * Otherwise returns '0'.
1770 */
1771static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1772 const struct inode *inode)
1773{
1774 const unsigned long check_valid =
1775 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1776 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1777 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1778 NFS_INO_INVALID_NLINK;
1779 unsigned long cache_validity = NFS_I(inode)->cache_validity;
1780
1781 if (!(cache_validity & NFS_INO_INVALID_CHANGE) &&
1782 (cache_validity & check_valid) != 0 &&
1783 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1784 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1785 return 1;
1786 return 0;
1787}
1788
1789static int nfs_refresh_inode_locked(struct inode *inode,
1790 struct nfs_fattr *fattr)
1791{
1792 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1793 int ret = 0;
1794
1795 trace_nfs_refresh_inode_enter(inode);
1796
1797 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1798 ret = nfs_update_inode(inode, fattr);
1799 else if (attr_cmp == 0)
1800 ret = nfs_check_inode_attributes(inode, fattr);
1801
1802 trace_nfs_refresh_inode_exit(inode, ret);
1803 return ret;
1804}
1805
1806/**
1807 * nfs_refresh_inode - try to update the inode attribute cache
1808 * @inode: pointer to inode
1809 * @fattr: updated attributes
1810 *
1811 * Check that an RPC call that returned attributes has not overlapped with
1812 * other recent updates of the inode metadata, then decide whether it is
1813 * safe to do a full update of the inode attributes, or whether just to
1814 * call nfs_check_inode_attributes.
1815 */
1816int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1817{
1818 int status;
1819
1820 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1821 return 0;
1822 spin_lock(&inode->i_lock);
1823 status = nfs_refresh_inode_locked(inode, fattr);
1824 spin_unlock(&inode->i_lock);
1825
1826 return status;
1827}
1828EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1829
1830static int nfs_post_op_update_inode_locked(struct inode *inode,
1831 struct nfs_fattr *fattr, unsigned int invalid)
1832{
1833 if (S_ISDIR(inode->i_mode))
1834 invalid |= NFS_INO_INVALID_DATA;
1835 nfs_set_cache_invalid(inode, invalid);
1836 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1837 return 0;
1838 return nfs_refresh_inode_locked(inode, fattr);
1839}
1840
1841/**
1842 * nfs_post_op_update_inode - try to update the inode attribute cache
1843 * @inode: pointer to inode
1844 * @fattr: updated attributes
1845 *
1846 * After an operation that has changed the inode metadata, mark the
1847 * attribute cache as being invalid, then try to update it.
1848 *
1849 * NB: if the server didn't return any post op attributes, this
1850 * function will force the retrieval of attributes before the next
1851 * NFS request. Thus it should be used only for operations that
1852 * are expected to change one or more attributes, to avoid
1853 * unnecessary NFS requests and trips through nfs_update_inode().
1854 */
1855int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1856{
1857 int status;
1858
1859 spin_lock(&inode->i_lock);
1860 nfs_fattr_set_barrier(fattr);
1861 status = nfs_post_op_update_inode_locked(inode, fattr,
1862 NFS_INO_INVALID_CHANGE
1863 | NFS_INO_INVALID_CTIME
1864 | NFS_INO_REVAL_FORCED);
1865 spin_unlock(&inode->i_lock);
1866
1867 return status;
1868}
1869EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1870
1871/**
1872 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1873 * @inode: pointer to inode
1874 * @fattr: updated attributes
1875 *
1876 * After an operation that has changed the inode metadata, mark the
1877 * attribute cache as being invalid, then try to update it. Fake up
1878 * weak cache consistency data, if none exist.
1879 *
1880 * This function is mainly designed to be used by the ->write_done() functions.
1881 */
1882int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1883{
1884 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1885 int status;
1886
1887 /* Don't do a WCC update if these attributes are already stale */
1888 if (attr_cmp < 0)
1889 return 0;
1890 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
1891 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1892 | NFS_ATTR_FATTR_PRESIZE
1893 | NFS_ATTR_FATTR_PREMTIME
1894 | NFS_ATTR_FATTR_PRECTIME);
1895 goto out_noforce;
1896 }
1897 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1898 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1899 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1900 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1901 }
1902 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1903 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1904 fattr->pre_ctime = inode->i_ctime;
1905 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1906 }
1907 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1908 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1909 fattr->pre_mtime = inode->i_mtime;
1910 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1911 }
1912 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1913 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1914 fattr->pre_size = i_size_read(inode);
1915 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1916 }
1917out_noforce:
1918 status = nfs_post_op_update_inode_locked(inode, fattr,
1919 NFS_INO_INVALID_CHANGE
1920 | NFS_INO_INVALID_CTIME
1921 | NFS_INO_INVALID_MTIME
1922 | NFS_INO_INVALID_BLOCKS);
1923 return status;
1924}
1925
1926/**
1927 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1928 * @inode: pointer to inode
1929 * @fattr: updated attributes
1930 *
1931 * After an operation that has changed the inode metadata, mark the
1932 * attribute cache as being invalid, then try to update it. Fake up
1933 * weak cache consistency data, if none exist.
1934 *
1935 * This function is mainly designed to be used by the ->write_done() functions.
1936 */
1937int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1938{
1939 int status;
1940
1941 spin_lock(&inode->i_lock);
1942 nfs_fattr_set_barrier(fattr);
1943 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1944 spin_unlock(&inode->i_lock);
1945 return status;
1946}
1947EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1948
1949
1950/*
1951 * Many nfs protocol calls return the new file attributes after
1952 * an operation. Here we update the inode to reflect the state
1953 * of the server's inode.
1954 *
1955 * This is a bit tricky because we have to make sure all dirty pages
1956 * have been sent off to the server before calling invalidate_inode_pages.
1957 * To make sure no other process adds more write requests while we try
1958 * our best to flush them, we make them sleep during the attribute refresh.
1959 *
1960 * A very similar scenario holds for the dir cache.
1961 */
1962static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1963{
1964 struct nfs_server *server = NFS_SERVER(inode);
1965 struct nfs_inode *nfsi = NFS_I(inode);
1966 loff_t cur_isize, new_isize;
1967 u64 fattr_supported = server->fattr_valid;
1968 unsigned long invalid = 0;
1969 unsigned long now = jiffies;
1970 unsigned long save_cache_validity;
1971 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1972 bool cache_revalidated = true;
1973 bool attr_changed = false;
1974 bool have_delegation;
1975
1976 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1977 __func__, inode->i_sb->s_id, inode->i_ino,
1978 nfs_display_fhandle_hash(NFS_FH(inode)),
1979 atomic_read(&inode->i_count), fattr->valid);
1980
1981 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1982 /* Only a mounted-on-fileid? Just exit */
1983 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1984 return 0;
1985 /* Has the inode gone and changed behind our back? */
1986 } else if (nfsi->fileid != fattr->fileid) {
1987 /* Is this perhaps the mounted-on fileid? */
1988 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1989 nfsi->fileid == fattr->mounted_on_fileid)
1990 return 0;
1991 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1992 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1993 NFS_SERVER(inode)->nfs_client->cl_hostname,
1994 inode->i_sb->s_id, (long long)nfsi->fileid,
1995 (long long)fattr->fileid);
1996 goto out_err;
1997 }
1998
1999 /*
2000 * Make sure the inode's type hasn't changed.
2001 */
2002 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2003 /*
2004 * Big trouble! The inode has become a different object.
2005 */
2006 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2007 __func__, inode->i_ino, inode->i_mode, fattr->mode);
2008 goto out_err;
2009 }
2010
2011 /* Update the fsid? */
2012 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2013 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2014 !IS_AUTOMOUNT(inode))
2015 server->fsid = fattr->fsid;
2016
2017 /* Save the delegation state before clearing cache_validity */
2018 have_delegation = nfs_have_delegated_attributes(inode);
2019
2020 /*
2021 * Update the read time so we don't revalidate too often.
2022 */
2023 nfsi->read_cache_jiffies = fattr->time_start;
2024
2025 save_cache_validity = nfsi->cache_validity;
2026 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2027 | NFS_INO_INVALID_ATIME
2028 | NFS_INO_REVAL_FORCED
2029 | NFS_INO_INVALID_BLOCKS);
2030
2031 /* Do atomic weak cache consistency updates */
2032 nfs_wcc_update_inode(inode, fattr);
2033
2034 if (pnfs_layoutcommit_outstanding(inode)) {
2035 nfsi->cache_validity |=
2036 save_cache_validity &
2037 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2038 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2039 NFS_INO_INVALID_BLOCKS);
2040 cache_revalidated = false;
2041 }
2042
2043 /* More cache consistency checks */
2044 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2045 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2046 /* Could it be a race with writeback? */
2047 if (!(have_writers || have_delegation)) {
2048 invalid |= NFS_INO_INVALID_DATA
2049 | NFS_INO_INVALID_ACCESS
2050 | NFS_INO_INVALID_ACL
2051 | NFS_INO_INVALID_XATTR;
2052 /* Force revalidate of all attributes */
2053 save_cache_validity |= NFS_INO_INVALID_CTIME
2054 | NFS_INO_INVALID_MTIME
2055 | NFS_INO_INVALID_SIZE
2056 | NFS_INO_INVALID_BLOCKS
2057 | NFS_INO_INVALID_NLINK
2058 | NFS_INO_INVALID_MODE
2059 | NFS_INO_INVALID_OTHER;
2060 if (S_ISDIR(inode->i_mode))
2061 nfs_force_lookup_revalidate(inode);
2062 attr_changed = true;
2063 dprintk("NFS: change_attr change on server for file %s/%ld\n",
2064 inode->i_sb->s_id,
2065 inode->i_ino);
2066 } else if (!have_delegation)
2067 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
2068 inode_set_iversion_raw(inode, fattr->change_attr);
2069 }
2070 } else {
2071 nfsi->cache_validity |=
2072 save_cache_validity & NFS_INO_INVALID_CHANGE;
2073 if (!have_delegation ||
2074 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2075 cache_revalidated = false;
2076 }
2077
2078 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2079 inode->i_mtime = fattr->mtime;
2080 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2081 nfsi->cache_validity |=
2082 save_cache_validity & NFS_INO_INVALID_MTIME;
2083
2084 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2085 inode->i_ctime = fattr->ctime;
2086 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2087 nfsi->cache_validity |=
2088 save_cache_validity & NFS_INO_INVALID_CTIME;
2089
2090 /* Check if our cached file size is stale */
2091 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2092 new_isize = nfs_size_to_loff_t(fattr->size);
2093 cur_isize = i_size_read(inode);
2094 if (new_isize != cur_isize && !have_delegation) {
2095 /* Do we perhaps have any outstanding writes, or has
2096 * the file grown beyond our last write? */
2097 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2098 i_size_write(inode, new_isize);
2099 if (!have_writers)
2100 invalid |= NFS_INO_INVALID_DATA;
2101 }
2102 dprintk("NFS: isize change on server for file %s/%ld "
2103 "(%Ld to %Ld)\n",
2104 inode->i_sb->s_id,
2105 inode->i_ino,
2106 (long long)cur_isize,
2107 (long long)new_isize);
2108 }
2109 if (new_isize == 0 &&
2110 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2111 NFS_ATTR_FATTR_BLOCKS_USED))) {
2112 fattr->du.nfs3.used = 0;
2113 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2114 }
2115 } else
2116 nfsi->cache_validity |=
2117 save_cache_validity & NFS_INO_INVALID_SIZE;
2118
2119 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2120 inode->i_atime = fattr->atime;
2121 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2122 nfsi->cache_validity |=
2123 save_cache_validity & NFS_INO_INVALID_ATIME;
2124
2125 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2126 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2127 umode_t newmode = inode->i_mode & S_IFMT;
2128 newmode |= fattr->mode & S_IALLUGO;
2129 inode->i_mode = newmode;
2130 invalid |= NFS_INO_INVALID_ACCESS
2131 | NFS_INO_INVALID_ACL;
2132 }
2133 } else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2134 nfsi->cache_validity |=
2135 save_cache_validity & NFS_INO_INVALID_MODE;
2136
2137 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2138 if (!uid_eq(inode->i_uid, fattr->uid)) {
2139 invalid |= NFS_INO_INVALID_ACCESS
2140 | NFS_INO_INVALID_ACL;
2141 inode->i_uid = fattr->uid;
2142 }
2143 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2144 nfsi->cache_validity |=
2145 save_cache_validity & NFS_INO_INVALID_OTHER;
2146
2147 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2148 if (!gid_eq(inode->i_gid, fattr->gid)) {
2149 invalid |= NFS_INO_INVALID_ACCESS
2150 | NFS_INO_INVALID_ACL;
2151 inode->i_gid = fattr->gid;
2152 }
2153 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2154 nfsi->cache_validity |=
2155 save_cache_validity & NFS_INO_INVALID_OTHER;
2156
2157 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2158 if (inode->i_nlink != fattr->nlink) {
2159 if (S_ISDIR(inode->i_mode))
2160 invalid |= NFS_INO_INVALID_DATA;
2161 set_nlink(inode, fattr->nlink);
2162 }
2163 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2164 nfsi->cache_validity |=
2165 save_cache_validity & NFS_INO_INVALID_NLINK;
2166
2167 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2168 /*
2169 * report the blocks in 512byte units
2170 */
2171 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2172 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2173 nfsi->cache_validity |=
2174 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2175
2176 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2177 inode->i_blocks = fattr->du.nfs2.blocks;
2178 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2179 nfsi->cache_validity |=
2180 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2181
2182 /* Update attrtimeo value if we're out of the unstable period */
2183 if (attr_changed) {
2184 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2185 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2186 nfsi->attrtimeo_timestamp = now;
2187 /* Set barrier to be more recent than all outstanding updates */
2188 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2189 } else {
2190 if (cache_revalidated) {
2191 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2192 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2193 nfsi->attrtimeo <<= 1;
2194 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2195 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2196 }
2197 nfsi->attrtimeo_timestamp = now;
2198 }
2199 /* Set the barrier to be more recent than this fattr */
2200 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2201 nfsi->attr_gencount = fattr->gencount;
2202 }
2203
2204 /* Don't invalidate the data if we were to blame */
2205 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2206 || S_ISLNK(inode->i_mode)))
2207 invalid &= ~NFS_INO_INVALID_DATA;
2208 nfs_set_cache_invalid(inode, invalid);
2209
2210 return 0;
2211 out_err:
2212 /*
2213 * No need to worry about unhashing the dentry, as the
2214 * lookup validation will know that the inode is bad.
2215 * (But we fall through to invalidate the caches.)
2216 */
2217 nfs_set_inode_stale_locked(inode);
2218 return -ESTALE;
2219}
2220
2221struct inode *nfs_alloc_inode(struct super_block *sb)
2222{
2223 struct nfs_inode *nfsi;
2224 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2225 if (!nfsi)
2226 return NULL;
2227 nfsi->flags = 0UL;
2228 nfsi->cache_validity = 0UL;
2229#if IS_ENABLED(CONFIG_NFS_V4)
2230 nfsi->nfs4_acl = NULL;
2231#endif /* CONFIG_NFS_V4 */
2232#ifdef CONFIG_NFS_V4_2
2233 nfsi->xattr_cache = NULL;
2234#endif
2235 return &nfsi->vfs_inode;
2236}
2237EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2238
2239void nfs_free_inode(struct inode *inode)
2240{
2241 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2242}
2243EXPORT_SYMBOL_GPL(nfs_free_inode);
2244
2245static inline void nfs4_init_once(struct nfs_inode *nfsi)
2246{
2247#if IS_ENABLED(CONFIG_NFS_V4)
2248 INIT_LIST_HEAD(&nfsi->open_states);
2249 nfsi->delegation = NULL;
2250 init_rwsem(&nfsi->rwsem);
2251 nfsi->layout = NULL;
2252#endif
2253}
2254
2255static void init_once(void *foo)
2256{
2257 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2258
2259 inode_init_once(&nfsi->vfs_inode);
2260 INIT_LIST_HEAD(&nfsi->open_files);
2261 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2262 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2263 INIT_LIST_HEAD(&nfsi->commit_info.list);
2264 atomic_long_set(&nfsi->nrequests, 0);
2265 atomic_long_set(&nfsi->commit_info.ncommit, 0);
2266 atomic_set(&nfsi->commit_info.rpcs_out, 0);
2267 init_rwsem(&nfsi->rmdir_sem);
2268 mutex_init(&nfsi->commit_mutex);
2269 nfs4_init_once(nfsi);
2270 nfsi->cache_change_attribute = 0;
2271}
2272
2273static int __init nfs_init_inodecache(void)
2274{
2275 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2276 sizeof(struct nfs_inode),
2277 0, (SLAB_RECLAIM_ACCOUNT|
2278 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2279 init_once);
2280 if (nfs_inode_cachep == NULL)
2281 return -ENOMEM;
2282
2283 return 0;
2284}
2285
2286static void nfs_destroy_inodecache(void)
2287{
2288 /*
2289 * Make sure all delayed rcu free inodes are flushed before we
2290 * destroy cache.
2291 */
2292 rcu_barrier();
2293 kmem_cache_destroy(nfs_inode_cachep);
2294}
2295
2296struct workqueue_struct *nfsiod_workqueue;
2297EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2298
2299/*
2300 * start up the nfsiod workqueue
2301 */
2302static int nfsiod_start(void)
2303{
2304 struct workqueue_struct *wq;
2305 dprintk("RPC: creating workqueue nfsiod\n");
2306 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2307 if (wq == NULL)
2308 return -ENOMEM;
2309 nfsiod_workqueue = wq;
2310 return 0;
2311}
2312
2313/*
2314 * Destroy the nfsiod workqueue
2315 */
2316static void nfsiod_stop(void)
2317{
2318 struct workqueue_struct *wq;
2319
2320 wq = nfsiod_workqueue;
2321 if (wq == NULL)
2322 return;
2323 nfsiod_workqueue = NULL;
2324 destroy_workqueue(wq);
2325}
2326
2327unsigned int nfs_net_id;
2328EXPORT_SYMBOL_GPL(nfs_net_id);
2329
2330static int nfs_net_init(struct net *net)
2331{
2332 nfs_clients_init(net);
2333 return nfs_fs_proc_net_init(net);
2334}
2335
2336static void nfs_net_exit(struct net *net)
2337{
2338 nfs_fs_proc_net_exit(net);
2339 nfs_clients_exit(net);
2340}
2341
2342static struct pernet_operations nfs_net_ops = {
2343 .init = nfs_net_init,
2344 .exit = nfs_net_exit,
2345 .id = &nfs_net_id,
2346 .size = sizeof(struct nfs_net),
2347};
2348
2349/*
2350 * Initialize NFS
2351 */
2352static int __init init_nfs_fs(void)
2353{
2354 int err;
2355
2356 err = nfs_sysfs_init();
2357 if (err < 0)
2358 goto out10;
2359
2360 err = register_pernet_subsys(&nfs_net_ops);
2361 if (err < 0)
2362 goto out9;
2363
2364 err = nfs_fscache_register();
2365 if (err < 0)
2366 goto out8;
2367
2368 err = nfsiod_start();
2369 if (err)
2370 goto out7;
2371
2372 err = nfs_fs_proc_init();
2373 if (err)
2374 goto out6;
2375
2376 err = nfs_init_nfspagecache();
2377 if (err)
2378 goto out5;
2379
2380 err = nfs_init_inodecache();
2381 if (err)
2382 goto out4;
2383
2384 err = nfs_init_readpagecache();
2385 if (err)
2386 goto out3;
2387
2388 err = nfs_init_writepagecache();
2389 if (err)
2390 goto out2;
2391
2392 err = nfs_init_directcache();
2393 if (err)
2394 goto out1;
2395
2396 rpc_proc_register(&init_net, &nfs_rpcstat);
2397
2398 err = register_nfs_fs();
2399 if (err)
2400 goto out0;
2401
2402 return 0;
2403out0:
2404 rpc_proc_unregister(&init_net, "nfs");
2405 nfs_destroy_directcache();
2406out1:
2407 nfs_destroy_writepagecache();
2408out2:
2409 nfs_destroy_readpagecache();
2410out3:
2411 nfs_destroy_inodecache();
2412out4:
2413 nfs_destroy_nfspagecache();
2414out5:
2415 nfs_fs_proc_exit();
2416out6:
2417 nfsiod_stop();
2418out7:
2419 nfs_fscache_unregister();
2420out8:
2421 unregister_pernet_subsys(&nfs_net_ops);
2422out9:
2423 nfs_sysfs_exit();
2424out10:
2425 return err;
2426}
2427
2428static void __exit exit_nfs_fs(void)
2429{
2430 nfs_destroy_directcache();
2431 nfs_destroy_writepagecache();
2432 nfs_destroy_readpagecache();
2433 nfs_destroy_inodecache();
2434 nfs_destroy_nfspagecache();
2435 nfs_fscache_unregister();
2436 unregister_pernet_subsys(&nfs_net_ops);
2437 rpc_proc_unregister(&init_net, "nfs");
2438 unregister_nfs_fs();
2439 nfs_fs_proc_exit();
2440 nfsiod_stop();
2441 nfs_sysfs_exit();
2442}
2443
2444/* Not quite true; I just maintain it */
2445MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2446MODULE_LICENSE("GPL");
2447module_param(enable_ino64, bool, 0644);
2448
2449module_init(init_nfs_fs)
2450module_exit(exit_nfs_fs)
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.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/nfs_idmap.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
42#include <asm/system.h>
43#include <asm/uaccess.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 "dns_resolve.h"
52#include "pnfs.h"
53
54#define NFSDBG_FACILITY NFSDBG_VFS
55
56#define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
57
58/* Default is to see 64-bit inode numbers */
59static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
60
61static void nfs_invalidate_inode(struct inode *);
62static int nfs_update_inode(struct inode *, struct nfs_fattr *);
63
64static struct kmem_cache * nfs_inode_cachep;
65
66static inline unsigned long
67nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
68{
69 return nfs_fileid_to_ino_t(fattr->fileid);
70}
71
72/**
73 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
74 * @word: long word containing the bit lock
75 */
76int nfs_wait_bit_killable(void *word)
77{
78 if (fatal_signal_pending(current))
79 return -ERESTARTSYS;
80 schedule();
81 return 0;
82}
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 */
91u64 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
107static void nfs_clear_inode(struct inode *inode)
108{
109 /*
110 * The following should never happen...
111 */
112 BUG_ON(nfs_have_writebacks(inode));
113 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
114 nfs_zap_acl_cache(inode);
115 nfs_access_zap_cache(inode);
116 nfs_fscache_release_inode_cookie(inode);
117}
118
119void nfs_evict_inode(struct inode *inode)
120{
121 truncate_inode_pages(&inode->i_data, 0);
122 end_writeback(inode);
123 nfs_clear_inode(inode);
124}
125
126/**
127 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
128 */
129int nfs_sync_mapping(struct address_space *mapping)
130{
131 int ret = 0;
132
133 if (mapping->nrpages != 0) {
134 unmap_mapping_range(mapping, 0, 0, 0);
135 ret = nfs_wb_all(mapping->host);
136 }
137 return ret;
138}
139
140/*
141 * Invalidate the local caches
142 */
143static void nfs_zap_caches_locked(struct inode *inode)
144{
145 struct nfs_inode *nfsi = NFS_I(inode);
146 int mode = inode->i_mode;
147
148 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
149
150 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
151 nfsi->attrtimeo_timestamp = jiffies;
152
153 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
154 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
155 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
156 else
157 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
158}
159
160void nfs_zap_caches(struct inode *inode)
161{
162 spin_lock(&inode->i_lock);
163 nfs_zap_caches_locked(inode);
164 spin_unlock(&inode->i_lock);
165}
166
167void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
168{
169 if (mapping->nrpages != 0) {
170 spin_lock(&inode->i_lock);
171 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
172 spin_unlock(&inode->i_lock);
173 }
174}
175
176void nfs_zap_acl_cache(struct inode *inode)
177{
178 void (*clear_acl_cache)(struct inode *);
179
180 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
181 if (clear_acl_cache != NULL)
182 clear_acl_cache(inode);
183 spin_lock(&inode->i_lock);
184 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
185 spin_unlock(&inode->i_lock);
186}
187
188void nfs_invalidate_atime(struct inode *inode)
189{
190 spin_lock(&inode->i_lock);
191 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
192 spin_unlock(&inode->i_lock);
193}
194
195/*
196 * Invalidate, but do not unhash, the inode.
197 * NB: must be called with inode->i_lock held!
198 */
199static void nfs_invalidate_inode(struct inode *inode)
200{
201 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
202 nfs_zap_caches_locked(inode);
203}
204
205struct nfs_find_desc {
206 struct nfs_fh *fh;
207 struct nfs_fattr *fattr;
208};
209
210/*
211 * In NFSv3 we can have 64bit inode numbers. In order to support
212 * this, and re-exported directories (also seen in NFSv2)
213 * we are forced to allow 2 different inodes to have the same
214 * i_ino.
215 */
216static int
217nfs_find_actor(struct inode *inode, void *opaque)
218{
219 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
220 struct nfs_fh *fh = desc->fh;
221 struct nfs_fattr *fattr = desc->fattr;
222
223 if (NFS_FILEID(inode) != fattr->fileid)
224 return 0;
225 if (nfs_compare_fh(NFS_FH(inode), fh))
226 return 0;
227 if (is_bad_inode(inode) || NFS_STALE(inode))
228 return 0;
229 return 1;
230}
231
232static int
233nfs_init_locked(struct inode *inode, void *opaque)
234{
235 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
236 struct nfs_fattr *fattr = desc->fattr;
237
238 set_nfs_fileid(inode, fattr->fileid);
239 nfs_copy_fh(NFS_FH(inode), desc->fh);
240 return 0;
241}
242
243/*
244 * This is our front-end to iget that looks up inodes by file handle
245 * instead of inode number.
246 */
247struct inode *
248nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
249{
250 struct nfs_find_desc desc = {
251 .fh = fh,
252 .fattr = fattr
253 };
254 struct inode *inode = ERR_PTR(-ENOENT);
255 unsigned long hash;
256
257 nfs_attr_check_mountpoint(sb, fattr);
258
259 if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
260 !nfs_attr_use_mounted_on_fileid(fattr))
261 goto out_no_inode;
262 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
263 goto out_no_inode;
264
265 hash = nfs_fattr_to_ino_t(fattr);
266
267 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
268 if (inode == NULL) {
269 inode = ERR_PTR(-ENOMEM);
270 goto out_no_inode;
271 }
272
273 if (inode->i_state & I_NEW) {
274 struct nfs_inode *nfsi = NFS_I(inode);
275 unsigned long now = jiffies;
276
277 /* We set i_ino for the few things that still rely on it,
278 * such as stat(2) */
279 inode->i_ino = hash;
280
281 /* We can't support update_atime(), since the server will reset it */
282 inode->i_flags |= S_NOATIME|S_NOCMTIME;
283 inode->i_mode = fattr->mode;
284 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
285 && nfs_server_capable(inode, NFS_CAP_MODE))
286 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
287 | NFS_INO_INVALID_ACCESS
288 | NFS_INO_INVALID_ACL;
289 /* Why so? Because we want revalidate for devices/FIFOs, and
290 * that's precisely what we have in nfs_file_inode_operations.
291 */
292 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
293 if (S_ISREG(inode->i_mode)) {
294 inode->i_fop = &nfs_file_operations;
295 inode->i_data.a_ops = &nfs_file_aops;
296 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
297 } else if (S_ISDIR(inode->i_mode)) {
298 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
299 inode->i_fop = &nfs_dir_operations;
300 inode->i_data.a_ops = &nfs_dir_aops;
301 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS))
302 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
303 /* Deal with crossing mountpoints */
304 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
305 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
306 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
307 inode->i_op = &nfs_referral_inode_operations;
308 else
309 inode->i_op = &nfs_mountpoint_inode_operations;
310 inode->i_fop = NULL;
311 inode->i_flags |= S_AUTOMOUNT;
312 }
313 } else if (S_ISLNK(inode->i_mode))
314 inode->i_op = &nfs_symlink_inode_operations;
315 else
316 init_special_inode(inode, inode->i_mode, fattr->rdev);
317
318 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
319 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
320 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
321 nfsi->change_attr = 0;
322 inode->i_size = 0;
323 inode->i_nlink = 0;
324 inode->i_uid = -2;
325 inode->i_gid = -2;
326 inode->i_blocks = 0;
327 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
328
329 nfsi->read_cache_jiffies = fattr->time_start;
330 nfsi->attr_gencount = fattr->gencount;
331 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
332 inode->i_atime = fattr->atime;
333 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
334 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
335 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
336 inode->i_mtime = fattr->mtime;
337 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
338 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
339 | NFS_INO_INVALID_DATA;
340 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
341 inode->i_ctime = fattr->ctime;
342 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
343 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
344 | NFS_INO_INVALID_ACCESS
345 | NFS_INO_INVALID_ACL;
346 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
347 nfsi->change_attr = fattr->change_attr;
348 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
349 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
350 | NFS_INO_INVALID_DATA;
351 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
352 inode->i_size = nfs_size_to_loff_t(fattr->size);
353 else
354 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
355 | NFS_INO_INVALID_DATA
356 | NFS_INO_REVAL_PAGECACHE;
357 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
358 inode->i_nlink = fattr->nlink;
359 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
360 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
361 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
362 inode->i_uid = fattr->uid;
363 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
364 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
365 | NFS_INO_INVALID_ACCESS
366 | NFS_INO_INVALID_ACL;
367 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
368 inode->i_gid = fattr->gid;
369 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
370 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
371 | NFS_INO_INVALID_ACCESS
372 | NFS_INO_INVALID_ACL;
373 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
374 inode->i_blocks = fattr->du.nfs2.blocks;
375 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
376 /*
377 * report the blocks in 512byte units
378 */
379 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
380 }
381 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
382 nfsi->attrtimeo_timestamp = now;
383 nfsi->access_cache = RB_ROOT;
384
385 nfs_fscache_init_inode_cookie(inode);
386
387 unlock_new_inode(inode);
388 } else
389 nfs_refresh_inode(inode, fattr);
390 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
391 inode->i_sb->s_id,
392 (long long)NFS_FILEID(inode),
393 atomic_read(&inode->i_count));
394
395out:
396 return inode;
397
398out_no_inode:
399 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
400 goto out;
401}
402
403#define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
404
405int
406nfs_setattr(struct dentry *dentry, struct iattr *attr)
407{
408 struct inode *inode = dentry->d_inode;
409 struct nfs_fattr *fattr;
410 int error = -ENOMEM;
411
412 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
413
414 /* skip mode change if it's just for clearing setuid/setgid */
415 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
416 attr->ia_valid &= ~ATTR_MODE;
417
418 if (attr->ia_valid & ATTR_SIZE) {
419 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
420 attr->ia_valid &= ~ATTR_SIZE;
421 }
422
423 /* Optimization: if the end result is no change, don't RPC */
424 attr->ia_valid &= NFS_VALID_ATTRS;
425 if ((attr->ia_valid & ~ATTR_FILE) == 0)
426 return 0;
427
428 /* Write all dirty data */
429 if (S_ISREG(inode->i_mode))
430 nfs_wb_all(inode);
431
432 fattr = nfs_alloc_fattr();
433 if (fattr == NULL)
434 goto out;
435 /*
436 * Return any delegations if we're going to change ACLs
437 */
438 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
439 nfs_inode_return_delegation(inode);
440 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
441 if (error == 0)
442 nfs_refresh_inode(inode, fattr);
443 nfs_free_fattr(fattr);
444out:
445 return error;
446}
447
448/**
449 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
450 * @inode: inode of the file used
451 * @offset: file offset to start truncating
452 *
453 * This is a copy of the common vmtruncate, but with the locking
454 * corrected to take into account the fact that NFS requires
455 * inode->i_size to be updated under the inode->i_lock.
456 */
457static int nfs_vmtruncate(struct inode * inode, loff_t offset)
458{
459 loff_t oldsize;
460 int err;
461
462 err = inode_newsize_ok(inode, offset);
463 if (err)
464 goto out;
465
466 spin_lock(&inode->i_lock);
467 oldsize = inode->i_size;
468 i_size_write(inode, offset);
469 spin_unlock(&inode->i_lock);
470
471 truncate_pagecache(inode, oldsize, offset);
472out:
473 return err;
474}
475
476/**
477 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
478 * @inode: pointer to struct inode
479 * @attr: pointer to struct iattr
480 *
481 * Note: we do this in the *proc.c in order to ensure that
482 * it works for things like exclusive creates too.
483 */
484void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
485{
486 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
487 spin_lock(&inode->i_lock);
488 if ((attr->ia_valid & ATTR_MODE) != 0) {
489 int mode = attr->ia_mode & S_IALLUGO;
490 mode |= inode->i_mode & ~S_IALLUGO;
491 inode->i_mode = mode;
492 }
493 if ((attr->ia_valid & ATTR_UID) != 0)
494 inode->i_uid = attr->ia_uid;
495 if ((attr->ia_valid & ATTR_GID) != 0)
496 inode->i_gid = attr->ia_gid;
497 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
498 spin_unlock(&inode->i_lock);
499 }
500 if ((attr->ia_valid & ATTR_SIZE) != 0) {
501 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
502 nfs_vmtruncate(inode, attr->ia_size);
503 }
504}
505
506int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
507{
508 struct inode *inode = dentry->d_inode;
509 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
510 int err;
511
512 /* Flush out writes to the server in order to update c/mtime. */
513 if (S_ISREG(inode->i_mode)) {
514 err = filemap_write_and_wait(inode->i_mapping);
515 if (err)
516 goto out;
517 }
518
519 /*
520 * We may force a getattr if the user cares about atime.
521 *
522 * Note that we only have to check the vfsmount flags here:
523 * - NFS always sets S_NOATIME by so checking it would give a
524 * bogus result
525 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
526 * no point in checking those.
527 */
528 if ((mnt->mnt_flags & MNT_NOATIME) ||
529 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
530 need_atime = 0;
531
532 if (need_atime)
533 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
534 else
535 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
536 if (!err) {
537 generic_fillattr(inode, stat);
538 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
539 }
540out:
541 return err;
542}
543
544static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
545{
546 atomic_set(&l_ctx->count, 1);
547 l_ctx->lockowner = current->files;
548 l_ctx->pid = current->tgid;
549 INIT_LIST_HEAD(&l_ctx->list);
550}
551
552static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
553{
554 struct nfs_lock_context *pos;
555
556 list_for_each_entry(pos, &ctx->lock_context.list, list) {
557 if (pos->lockowner != current->files)
558 continue;
559 if (pos->pid != current->tgid)
560 continue;
561 atomic_inc(&pos->count);
562 return pos;
563 }
564 return NULL;
565}
566
567struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
568{
569 struct nfs_lock_context *res, *new = NULL;
570 struct inode *inode = ctx->dentry->d_inode;
571
572 spin_lock(&inode->i_lock);
573 res = __nfs_find_lock_context(ctx);
574 if (res == NULL) {
575 spin_unlock(&inode->i_lock);
576 new = kmalloc(sizeof(*new), GFP_KERNEL);
577 if (new == NULL)
578 return NULL;
579 nfs_init_lock_context(new);
580 spin_lock(&inode->i_lock);
581 res = __nfs_find_lock_context(ctx);
582 if (res == NULL) {
583 list_add_tail(&new->list, &ctx->lock_context.list);
584 new->open_context = ctx;
585 res = new;
586 new = NULL;
587 }
588 }
589 spin_unlock(&inode->i_lock);
590 kfree(new);
591 return res;
592}
593
594void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
595{
596 struct nfs_open_context *ctx = l_ctx->open_context;
597 struct inode *inode = ctx->dentry->d_inode;
598
599 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
600 return;
601 list_del(&l_ctx->list);
602 spin_unlock(&inode->i_lock);
603 kfree(l_ctx);
604}
605
606/**
607 * nfs_close_context - Common close_context() routine NFSv2/v3
608 * @ctx: pointer to context
609 * @is_sync: is this a synchronous close
610 *
611 * always ensure that the attributes are up to date if we're mounted
612 * with close-to-open semantics
613 */
614void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
615{
616 struct inode *inode;
617 struct nfs_server *server;
618
619 if (!(ctx->mode & FMODE_WRITE))
620 return;
621 if (!is_sync)
622 return;
623 inode = ctx->dentry->d_inode;
624 if (!list_empty(&NFS_I(inode)->open_files))
625 return;
626 server = NFS_SERVER(inode);
627 if (server->flags & NFS_MOUNT_NOCTO)
628 return;
629 nfs_revalidate_inode(server, inode);
630}
631
632struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred, fmode_t f_mode)
633{
634 struct nfs_open_context *ctx;
635
636 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
637 if (ctx != NULL) {
638 nfs_sb_active(dentry->d_sb);
639 ctx->dentry = dget(dentry);
640 ctx->cred = get_rpccred(cred);
641 ctx->state = NULL;
642 ctx->mode = f_mode;
643 ctx->flags = 0;
644 ctx->error = 0;
645 nfs_init_lock_context(&ctx->lock_context);
646 ctx->lock_context.open_context = ctx;
647 INIT_LIST_HEAD(&ctx->list);
648 }
649 return ctx;
650}
651
652struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
653{
654 if (ctx != NULL)
655 atomic_inc(&ctx->lock_context.count);
656 return ctx;
657}
658
659static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
660{
661 struct inode *inode = ctx->dentry->d_inode;
662 struct super_block *sb = ctx->dentry->d_sb;
663
664 if (!list_empty(&ctx->list)) {
665 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
666 return;
667 list_del(&ctx->list);
668 spin_unlock(&inode->i_lock);
669 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
670 return;
671 if (inode != NULL)
672 NFS_PROTO(inode)->close_context(ctx, is_sync);
673 if (ctx->cred != NULL)
674 put_rpccred(ctx->cred);
675 dput(ctx->dentry);
676 nfs_sb_deactive(sb);
677 kfree(ctx);
678}
679
680void put_nfs_open_context(struct nfs_open_context *ctx)
681{
682 __put_nfs_open_context(ctx, 0);
683}
684
685/*
686 * Ensure that mmap has a recent RPC credential for use when writing out
687 * shared pages
688 */
689void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
690{
691 struct inode *inode = filp->f_path.dentry->d_inode;
692 struct nfs_inode *nfsi = NFS_I(inode);
693
694 filp->private_data = get_nfs_open_context(ctx);
695 spin_lock(&inode->i_lock);
696 list_add(&ctx->list, &nfsi->open_files);
697 spin_unlock(&inode->i_lock);
698}
699
700/*
701 * Given an inode, search for an open context with the desired characteristics
702 */
703struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
704{
705 struct nfs_inode *nfsi = NFS_I(inode);
706 struct nfs_open_context *pos, *ctx = NULL;
707
708 spin_lock(&inode->i_lock);
709 list_for_each_entry(pos, &nfsi->open_files, list) {
710 if (cred != NULL && pos->cred != cred)
711 continue;
712 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
713 continue;
714 ctx = get_nfs_open_context(pos);
715 break;
716 }
717 spin_unlock(&inode->i_lock);
718 return ctx;
719}
720
721static void nfs_file_clear_open_context(struct file *filp)
722{
723 struct inode *inode = filp->f_path.dentry->d_inode;
724 struct nfs_open_context *ctx = nfs_file_open_context(filp);
725
726 if (ctx) {
727 filp->private_data = NULL;
728 spin_lock(&inode->i_lock);
729 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
730 spin_unlock(&inode->i_lock);
731 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
732 }
733}
734
735/*
736 * These allocate and release file read/write context information.
737 */
738int nfs_open(struct inode *inode, struct file *filp)
739{
740 struct nfs_open_context *ctx;
741 struct rpc_cred *cred;
742
743 cred = rpc_lookup_cred();
744 if (IS_ERR(cred))
745 return PTR_ERR(cred);
746 ctx = alloc_nfs_open_context(filp->f_path.dentry, cred, filp->f_mode);
747 put_rpccred(cred);
748 if (ctx == NULL)
749 return -ENOMEM;
750 nfs_file_set_open_context(filp, ctx);
751 put_nfs_open_context(ctx);
752 nfs_fscache_set_inode_cookie(inode, filp);
753 return 0;
754}
755
756int nfs_release(struct inode *inode, struct file *filp)
757{
758 nfs_file_clear_open_context(filp);
759 return 0;
760}
761
762/*
763 * This function is called whenever some part of NFS notices that
764 * the cached attributes have to be refreshed.
765 */
766int
767__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
768{
769 int status = -ESTALE;
770 struct nfs_fattr *fattr = NULL;
771 struct nfs_inode *nfsi = NFS_I(inode);
772
773 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
774 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
775
776 if (is_bad_inode(inode))
777 goto out;
778 if (NFS_STALE(inode))
779 goto out;
780
781 status = -ENOMEM;
782 fattr = nfs_alloc_fattr();
783 if (fattr == NULL)
784 goto out;
785
786 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
787 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr);
788 if (status != 0) {
789 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
790 inode->i_sb->s_id,
791 (long long)NFS_FILEID(inode), status);
792 if (status == -ESTALE) {
793 nfs_zap_caches(inode);
794 if (!S_ISDIR(inode->i_mode))
795 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
796 }
797 goto out;
798 }
799
800 status = nfs_refresh_inode(inode, fattr);
801 if (status) {
802 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
803 inode->i_sb->s_id,
804 (long long)NFS_FILEID(inode), status);
805 goto out;
806 }
807
808 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
809 nfs_zap_acl_cache(inode);
810
811 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
812 inode->i_sb->s_id,
813 (long long)NFS_FILEID(inode));
814
815 out:
816 nfs_free_fattr(fattr);
817 return status;
818}
819
820int nfs_attribute_timeout(struct inode *inode)
821{
822 struct nfs_inode *nfsi = NFS_I(inode);
823
824 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
825}
826
827static int nfs_attribute_cache_expired(struct inode *inode)
828{
829 if (nfs_have_delegated_attributes(inode))
830 return 0;
831 return nfs_attribute_timeout(inode);
832}
833
834/**
835 * nfs_revalidate_inode - Revalidate the inode attributes
836 * @server - pointer to nfs_server struct
837 * @inode - pointer to inode struct
838 *
839 * Updates inode attribute information by retrieving the data from the server.
840 */
841int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
842{
843 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
844 && !nfs_attribute_cache_expired(inode))
845 return NFS_STALE(inode) ? -ESTALE : 0;
846 return __nfs_revalidate_inode(server, inode);
847}
848
849static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
850{
851 struct nfs_inode *nfsi = NFS_I(inode);
852
853 if (mapping->nrpages != 0) {
854 int ret = invalidate_inode_pages2(mapping);
855 if (ret < 0)
856 return ret;
857 }
858 spin_lock(&inode->i_lock);
859 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
860 if (S_ISDIR(inode->i_mode))
861 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
862 spin_unlock(&inode->i_lock);
863 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
864 nfs_fscache_reset_inode_cookie(inode);
865 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
866 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
867 return 0;
868}
869
870/**
871 * nfs_revalidate_mapping - Revalidate the pagecache
872 * @inode - pointer to host inode
873 * @mapping - pointer to mapping
874 */
875int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
876{
877 struct nfs_inode *nfsi = NFS_I(inode);
878 int ret = 0;
879
880 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
881 || nfs_attribute_cache_expired(inode)
882 || NFS_STALE(inode)) {
883 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
884 if (ret < 0)
885 goto out;
886 }
887 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
888 ret = nfs_invalidate_mapping(inode, mapping);
889out:
890 return ret;
891}
892
893static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
894{
895 struct nfs_inode *nfsi = NFS_I(inode);
896 unsigned long ret = 0;
897
898 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
899 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
900 && nfsi->change_attr == fattr->pre_change_attr) {
901 nfsi->change_attr = fattr->change_attr;
902 if (S_ISDIR(inode->i_mode))
903 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
904 ret |= NFS_INO_INVALID_ATTR;
905 }
906 /* If we have atomic WCC data, we may update some attributes */
907 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
908 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
909 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
910 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
911 ret |= NFS_INO_INVALID_ATTR;
912 }
913
914 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
915 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
916 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
917 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
918 if (S_ISDIR(inode->i_mode))
919 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
920 ret |= NFS_INO_INVALID_ATTR;
921 }
922 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
923 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
924 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
925 && nfsi->npages == 0) {
926 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
927 ret |= NFS_INO_INVALID_ATTR;
928 }
929 return ret;
930}
931
932/**
933 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
934 * @inode - pointer to inode
935 * @fattr - updated attributes
936 *
937 * Verifies the attribute cache. If we have just changed the attributes,
938 * so that fattr carries weak cache consistency data, then it may
939 * also update the ctime/mtime/change_attribute.
940 */
941static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
942{
943 struct nfs_inode *nfsi = NFS_I(inode);
944 loff_t cur_size, new_isize;
945 unsigned long invalid = 0;
946
947
948 /* Has the inode gone and changed behind our back? */
949 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
950 return -EIO;
951 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
952 return -EIO;
953
954 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
955 nfsi->change_attr != fattr->change_attr)
956 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
957
958 /* Verify a few of the more important attributes */
959 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
960 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
961
962 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
963 cur_size = i_size_read(inode);
964 new_isize = nfs_size_to_loff_t(fattr->size);
965 if (cur_size != new_isize && nfsi->npages == 0)
966 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
967 }
968
969 /* Have any file permissions changed? */
970 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
971 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
972 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
973 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
974 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
975 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
976
977 /* Has the link count changed? */
978 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
979 invalid |= NFS_INO_INVALID_ATTR;
980
981 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
982 invalid |= NFS_INO_INVALID_ATIME;
983
984 if (invalid != 0)
985 nfsi->cache_validity |= invalid;
986
987 nfsi->read_cache_jiffies = fattr->time_start;
988 return 0;
989}
990
991static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
992{
993 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
994 return 0;
995 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
996}
997
998static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
999{
1000 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1001 return 0;
1002 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1003}
1004
1005static atomic_long_t nfs_attr_generation_counter;
1006
1007static unsigned long nfs_read_attr_generation_counter(void)
1008{
1009 return atomic_long_read(&nfs_attr_generation_counter);
1010}
1011
1012unsigned long nfs_inc_attr_generation_counter(void)
1013{
1014 return atomic_long_inc_return(&nfs_attr_generation_counter);
1015}
1016
1017void nfs_fattr_init(struct nfs_fattr *fattr)
1018{
1019 fattr->valid = 0;
1020 fattr->time_start = jiffies;
1021 fattr->gencount = nfs_inc_attr_generation_counter();
1022}
1023
1024struct nfs_fattr *nfs_alloc_fattr(void)
1025{
1026 struct nfs_fattr *fattr;
1027
1028 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1029 if (fattr != NULL)
1030 nfs_fattr_init(fattr);
1031 return fattr;
1032}
1033
1034struct nfs_fh *nfs_alloc_fhandle(void)
1035{
1036 struct nfs_fh *fh;
1037
1038 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1039 if (fh != NULL)
1040 fh->size = 0;
1041 return fh;
1042}
1043
1044/**
1045 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1046 * @inode - pointer to inode
1047 * @fattr - attributes
1048 *
1049 * Attempt to divine whether or not an RPC call reply carrying stale
1050 * attributes got scheduled after another call carrying updated ones.
1051 *
1052 * To do so, the function first assumes that a more recent ctime means
1053 * that the attributes in fattr are newer, however it also attempt to
1054 * catch the case where ctime either didn't change, or went backwards
1055 * (if someone reset the clock on the server) by looking at whether
1056 * or not this RPC call was started after the inode was last updated.
1057 * Note also the check for wraparound of 'attr_gencount'
1058 *
1059 * The function returns 'true' if it thinks the attributes in 'fattr' are
1060 * more recent than the ones cached in the inode.
1061 *
1062 */
1063static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1064{
1065 const struct nfs_inode *nfsi = NFS_I(inode);
1066
1067 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1068 nfs_ctime_need_update(inode, fattr) ||
1069 nfs_size_need_update(inode, fattr) ||
1070 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1071}
1072
1073static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1074{
1075 if (nfs_inode_attrs_need_update(inode, fattr))
1076 return nfs_update_inode(inode, fattr);
1077 return nfs_check_inode_attributes(inode, fattr);
1078}
1079
1080/**
1081 * nfs_refresh_inode - try to update the inode attribute cache
1082 * @inode - pointer to inode
1083 * @fattr - updated attributes
1084 *
1085 * Check that an RPC call that returned attributes has not overlapped with
1086 * other recent updates of the inode metadata, then decide whether it is
1087 * safe to do a full update of the inode attributes, or whether just to
1088 * call nfs_check_inode_attributes.
1089 */
1090int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1091{
1092 int status;
1093
1094 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1095 return 0;
1096 spin_lock(&inode->i_lock);
1097 status = nfs_refresh_inode_locked(inode, fattr);
1098 spin_unlock(&inode->i_lock);
1099
1100 return status;
1101}
1102
1103static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1104{
1105 struct nfs_inode *nfsi = NFS_I(inode);
1106
1107 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1108 if (S_ISDIR(inode->i_mode))
1109 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1110 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1111 return 0;
1112 return nfs_refresh_inode_locked(inode, fattr);
1113}
1114
1115/**
1116 * nfs_post_op_update_inode - try to update the inode attribute cache
1117 * @inode - pointer to inode
1118 * @fattr - updated attributes
1119 *
1120 * After an operation that has changed the inode metadata, mark the
1121 * attribute cache as being invalid, then try to update it.
1122 *
1123 * NB: if the server didn't return any post op attributes, this
1124 * function will force the retrieval of attributes before the next
1125 * NFS request. Thus it should be used only for operations that
1126 * are expected to change one or more attributes, to avoid
1127 * unnecessary NFS requests and trips through nfs_update_inode().
1128 */
1129int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1130{
1131 int status;
1132
1133 spin_lock(&inode->i_lock);
1134 status = nfs_post_op_update_inode_locked(inode, fattr);
1135 spin_unlock(&inode->i_lock);
1136 return status;
1137}
1138
1139/**
1140 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1141 * @inode - pointer to inode
1142 * @fattr - updated attributes
1143 *
1144 * After an operation that has changed the inode metadata, mark the
1145 * attribute cache as being invalid, then try to update it. Fake up
1146 * weak cache consistency data, if none exist.
1147 *
1148 * This function is mainly designed to be used by the ->write_done() functions.
1149 */
1150int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1151{
1152 int status;
1153
1154 spin_lock(&inode->i_lock);
1155 /* Don't do a WCC update if these attributes are already stale */
1156 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1157 !nfs_inode_attrs_need_update(inode, fattr)) {
1158 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1159 | NFS_ATTR_FATTR_PRESIZE
1160 | NFS_ATTR_FATTR_PREMTIME
1161 | NFS_ATTR_FATTR_PRECTIME);
1162 goto out_noforce;
1163 }
1164 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1165 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1166 fattr->pre_change_attr = NFS_I(inode)->change_attr;
1167 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1168 }
1169 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1170 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1171 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1172 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1173 }
1174 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1175 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1176 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1177 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1178 }
1179 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1180 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1181 fattr->pre_size = i_size_read(inode);
1182 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1183 }
1184out_noforce:
1185 status = nfs_post_op_update_inode_locked(inode, fattr);
1186 spin_unlock(&inode->i_lock);
1187 return status;
1188}
1189
1190/*
1191 * Many nfs protocol calls return the new file attributes after
1192 * an operation. Here we update the inode to reflect the state
1193 * of the server's inode.
1194 *
1195 * This is a bit tricky because we have to make sure all dirty pages
1196 * have been sent off to the server before calling invalidate_inode_pages.
1197 * To make sure no other process adds more write requests while we try
1198 * our best to flush them, we make them sleep during the attribute refresh.
1199 *
1200 * A very similar scenario holds for the dir cache.
1201 */
1202static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1203{
1204 struct nfs_server *server;
1205 struct nfs_inode *nfsi = NFS_I(inode);
1206 loff_t cur_isize, new_isize;
1207 unsigned long invalid = 0;
1208 unsigned long now = jiffies;
1209 unsigned long save_cache_validity;
1210
1211 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1212 __func__, inode->i_sb->s_id, inode->i_ino,
1213 atomic_read(&inode->i_count), fattr->valid);
1214
1215 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1216 goto out_fileid;
1217
1218 /*
1219 * Make sure the inode's type hasn't changed.
1220 */
1221 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1222 goto out_changed;
1223
1224 server = NFS_SERVER(inode);
1225 /* Update the fsid? */
1226 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1227 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1228 !IS_AUTOMOUNT(inode))
1229 server->fsid = fattr->fsid;
1230
1231 /*
1232 * Update the read time so we don't revalidate too often.
1233 */
1234 nfsi->read_cache_jiffies = fattr->time_start;
1235
1236 save_cache_validity = nfsi->cache_validity;
1237 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1238 | NFS_INO_INVALID_ATIME
1239 | NFS_INO_REVAL_FORCED
1240 | NFS_INO_REVAL_PAGECACHE);
1241
1242 /* Do atomic weak cache consistency updates */
1243 invalid |= nfs_wcc_update_inode(inode, fattr);
1244
1245 /* More cache consistency checks */
1246 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1247 if (nfsi->change_attr != fattr->change_attr) {
1248 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1249 inode->i_sb->s_id, inode->i_ino);
1250 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1251 if (S_ISDIR(inode->i_mode))
1252 nfs_force_lookup_revalidate(inode);
1253 nfsi->change_attr = fattr->change_attr;
1254 }
1255 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1256 invalid |= save_cache_validity;
1257
1258 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1259 /* NFSv2/v3: Check if the mtime agrees */
1260 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1261 dprintk("NFS: mtime change on server for file %s/%ld\n",
1262 inode->i_sb->s_id, inode->i_ino);
1263 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1264 if (S_ISDIR(inode->i_mode))
1265 nfs_force_lookup_revalidate(inode);
1266 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1267 }
1268 } else if (server->caps & NFS_CAP_MTIME)
1269 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1270 | NFS_INO_INVALID_DATA
1271 | NFS_INO_REVAL_PAGECACHE
1272 | NFS_INO_REVAL_FORCED);
1273
1274 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1275 /* If ctime has changed we should definitely clear access+acl caches */
1276 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1277 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1278 /* and probably clear data for a directory too as utimes can cause
1279 * havoc with our cache.
1280 */
1281 if (S_ISDIR(inode->i_mode)) {
1282 invalid |= NFS_INO_INVALID_DATA;
1283 nfs_force_lookup_revalidate(inode);
1284 }
1285 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1286 }
1287 } else if (server->caps & NFS_CAP_CTIME)
1288 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1289 | NFS_INO_INVALID_ACCESS
1290 | NFS_INO_INVALID_ACL
1291 | NFS_INO_REVAL_FORCED);
1292
1293 /* Check if our cached file size is stale */
1294 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1295 new_isize = nfs_size_to_loff_t(fattr->size);
1296 cur_isize = i_size_read(inode);
1297 if (new_isize != cur_isize) {
1298 /* Do we perhaps have any outstanding writes, or has
1299 * the file grown beyond our last write? */
1300 if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) ||
1301 new_isize > cur_isize) {
1302 i_size_write(inode, new_isize);
1303 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1304 }
1305 dprintk("NFS: isize change on server for file %s/%ld "
1306 "(%Ld to %Ld)\n",
1307 inode->i_sb->s_id,
1308 inode->i_ino,
1309 (long long)cur_isize,
1310 (long long)new_isize);
1311 }
1312 } else
1313 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1314 | NFS_INO_REVAL_PAGECACHE
1315 | NFS_INO_REVAL_FORCED);
1316
1317
1318 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1319 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1320 else if (server->caps & NFS_CAP_ATIME)
1321 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1322 | NFS_INO_REVAL_FORCED);
1323
1324 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1325 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1326 umode_t newmode = inode->i_mode & S_IFMT;
1327 newmode |= fattr->mode & S_IALLUGO;
1328 inode->i_mode = newmode;
1329 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1330 }
1331 } else if (server->caps & NFS_CAP_MODE)
1332 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1333 | NFS_INO_INVALID_ACCESS
1334 | NFS_INO_INVALID_ACL
1335 | NFS_INO_REVAL_FORCED);
1336
1337 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1338 if (inode->i_uid != fattr->uid) {
1339 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1340 inode->i_uid = fattr->uid;
1341 }
1342 } else if (server->caps & NFS_CAP_OWNER)
1343 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1344 | NFS_INO_INVALID_ACCESS
1345 | NFS_INO_INVALID_ACL
1346 | NFS_INO_REVAL_FORCED);
1347
1348 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1349 if (inode->i_gid != fattr->gid) {
1350 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1351 inode->i_gid = fattr->gid;
1352 }
1353 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1354 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1355 | NFS_INO_INVALID_ACCESS
1356 | NFS_INO_INVALID_ACL
1357 | NFS_INO_REVAL_FORCED);
1358
1359 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1360 if (inode->i_nlink != fattr->nlink) {
1361 invalid |= NFS_INO_INVALID_ATTR;
1362 if (S_ISDIR(inode->i_mode))
1363 invalid |= NFS_INO_INVALID_DATA;
1364 inode->i_nlink = fattr->nlink;
1365 }
1366 } else if (server->caps & NFS_CAP_NLINK)
1367 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1368 | NFS_INO_REVAL_FORCED);
1369
1370 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1371 /*
1372 * report the blocks in 512byte units
1373 */
1374 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1375 }
1376 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1377 inode->i_blocks = fattr->du.nfs2.blocks;
1378
1379 /* Update attrtimeo value if we're out of the unstable period */
1380 if (invalid & NFS_INO_INVALID_ATTR) {
1381 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1382 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1383 nfsi->attrtimeo_timestamp = now;
1384 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1385 } else {
1386 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1387 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1388 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1389 nfsi->attrtimeo_timestamp = now;
1390 }
1391 }
1392 invalid &= ~NFS_INO_INVALID_ATTR;
1393 /* Don't invalidate the data if we were to blame */
1394 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1395 || S_ISLNK(inode->i_mode)))
1396 invalid &= ~NFS_INO_INVALID_DATA;
1397 if (!nfs_have_delegation(inode, FMODE_READ) ||
1398 (save_cache_validity & NFS_INO_REVAL_FORCED))
1399 nfsi->cache_validity |= invalid;
1400
1401 return 0;
1402 out_changed:
1403 /*
1404 * Big trouble! The inode has become a different object.
1405 */
1406 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1407 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1408 out_err:
1409 /*
1410 * No need to worry about unhashing the dentry, as the
1411 * lookup validation will know that the inode is bad.
1412 * (But we fall through to invalidate the caches.)
1413 */
1414 nfs_invalidate_inode(inode);
1415 return -ESTALE;
1416
1417 out_fileid:
1418 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1419 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1420 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1421 (long long)nfsi->fileid, (long long)fattr->fileid);
1422 goto out_err;
1423}
1424
1425
1426#ifdef CONFIG_NFS_V4
1427
1428/*
1429 * Clean out any remaining NFSv4 state that might be left over due
1430 * to open() calls that passed nfs_atomic_lookup, but failed to call
1431 * nfs_open().
1432 */
1433void nfs4_evict_inode(struct inode *inode)
1434{
1435 truncate_inode_pages(&inode->i_data, 0);
1436 end_writeback(inode);
1437 pnfs_return_layout(inode);
1438 pnfs_destroy_layout(NFS_I(inode));
1439 /* If we are holding a delegation, return it! */
1440 nfs_inode_return_delegation_noreclaim(inode);
1441 /* First call standard NFS clear_inode() code */
1442 nfs_clear_inode(inode);
1443}
1444#endif
1445
1446struct inode *nfs_alloc_inode(struct super_block *sb)
1447{
1448 struct nfs_inode *nfsi;
1449 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1450 if (!nfsi)
1451 return NULL;
1452 nfsi->flags = 0UL;
1453 nfsi->cache_validity = 0UL;
1454#ifdef CONFIG_NFS_V3_ACL
1455 nfsi->acl_access = ERR_PTR(-EAGAIN);
1456 nfsi->acl_default = ERR_PTR(-EAGAIN);
1457#endif
1458#ifdef CONFIG_NFS_V4
1459 nfsi->nfs4_acl = NULL;
1460#endif /* CONFIG_NFS_V4 */
1461 return &nfsi->vfs_inode;
1462}
1463
1464static void nfs_i_callback(struct rcu_head *head)
1465{
1466 struct inode *inode = container_of(head, struct inode, i_rcu);
1467 INIT_LIST_HEAD(&inode->i_dentry);
1468 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1469}
1470
1471void nfs_destroy_inode(struct inode *inode)
1472{
1473 call_rcu(&inode->i_rcu, nfs_i_callback);
1474}
1475
1476static inline void nfs4_init_once(struct nfs_inode *nfsi)
1477{
1478#ifdef CONFIG_NFS_V4
1479 INIT_LIST_HEAD(&nfsi->open_states);
1480 nfsi->delegation = NULL;
1481 nfsi->delegation_state = 0;
1482 init_rwsem(&nfsi->rwsem);
1483 nfsi->layout = NULL;
1484 atomic_set(&nfsi->commits_outstanding, 0);
1485#endif
1486}
1487
1488static void init_once(void *foo)
1489{
1490 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1491
1492 inode_init_once(&nfsi->vfs_inode);
1493 INIT_LIST_HEAD(&nfsi->open_files);
1494 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1495 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1496 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1497 nfsi->npages = 0;
1498 nfsi->ncommit = 0;
1499 atomic_set(&nfsi->silly_count, 1);
1500 INIT_HLIST_HEAD(&nfsi->silly_list);
1501 init_waitqueue_head(&nfsi->waitqueue);
1502 nfs4_init_once(nfsi);
1503}
1504
1505static int __init nfs_init_inodecache(void)
1506{
1507 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1508 sizeof(struct nfs_inode),
1509 0, (SLAB_RECLAIM_ACCOUNT|
1510 SLAB_MEM_SPREAD),
1511 init_once);
1512 if (nfs_inode_cachep == NULL)
1513 return -ENOMEM;
1514
1515 return 0;
1516}
1517
1518static void nfs_destroy_inodecache(void)
1519{
1520 kmem_cache_destroy(nfs_inode_cachep);
1521}
1522
1523struct workqueue_struct *nfsiod_workqueue;
1524
1525/*
1526 * start up the nfsiod workqueue
1527 */
1528static int nfsiod_start(void)
1529{
1530 struct workqueue_struct *wq;
1531 dprintk("RPC: creating workqueue nfsiod\n");
1532 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1533 if (wq == NULL)
1534 return -ENOMEM;
1535 nfsiod_workqueue = wq;
1536 return 0;
1537}
1538
1539/*
1540 * Destroy the nfsiod workqueue
1541 */
1542static void nfsiod_stop(void)
1543{
1544 struct workqueue_struct *wq;
1545
1546 wq = nfsiod_workqueue;
1547 if (wq == NULL)
1548 return;
1549 nfsiod_workqueue = NULL;
1550 destroy_workqueue(wq);
1551}
1552
1553/*
1554 * Initialize NFS
1555 */
1556static int __init init_nfs_fs(void)
1557{
1558 int err;
1559
1560 err = nfs_idmap_init();
1561 if (err < 0)
1562 goto out9;
1563
1564 err = nfs_dns_resolver_init();
1565 if (err < 0)
1566 goto out8;
1567
1568 err = nfs_fscache_register();
1569 if (err < 0)
1570 goto out7;
1571
1572 err = nfsiod_start();
1573 if (err)
1574 goto out6;
1575
1576 err = nfs_fs_proc_init();
1577 if (err)
1578 goto out5;
1579
1580 err = nfs_init_nfspagecache();
1581 if (err)
1582 goto out4;
1583
1584 err = nfs_init_inodecache();
1585 if (err)
1586 goto out3;
1587
1588 err = nfs_init_readpagecache();
1589 if (err)
1590 goto out2;
1591
1592 err = nfs_init_writepagecache();
1593 if (err)
1594 goto out1;
1595
1596 err = nfs_init_directcache();
1597 if (err)
1598 goto out0;
1599
1600#ifdef CONFIG_PROC_FS
1601 rpc_proc_register(&nfs_rpcstat);
1602#endif
1603 if ((err = register_nfs_fs()) != 0)
1604 goto out;
1605 return 0;
1606out:
1607#ifdef CONFIG_PROC_FS
1608 rpc_proc_unregister("nfs");
1609#endif
1610 nfs_destroy_directcache();
1611out0:
1612 nfs_destroy_writepagecache();
1613out1:
1614 nfs_destroy_readpagecache();
1615out2:
1616 nfs_destroy_inodecache();
1617out3:
1618 nfs_destroy_nfspagecache();
1619out4:
1620 nfs_fs_proc_exit();
1621out5:
1622 nfsiod_stop();
1623out6:
1624 nfs_fscache_unregister();
1625out7:
1626 nfs_dns_resolver_destroy();
1627out8:
1628 nfs_idmap_quit();
1629out9:
1630 return err;
1631}
1632
1633static void __exit exit_nfs_fs(void)
1634{
1635 nfs_destroy_directcache();
1636 nfs_destroy_writepagecache();
1637 nfs_destroy_readpagecache();
1638 nfs_destroy_inodecache();
1639 nfs_destroy_nfspagecache();
1640 nfs_fscache_unregister();
1641 nfs_dns_resolver_destroy();
1642 nfs_idmap_quit();
1643#ifdef CONFIG_PROC_FS
1644 rpc_proc_unregister("nfs");
1645#endif
1646 nfs_cleanup_cb_ident_idr();
1647 unregister_nfs_fs();
1648 nfs_fs_proc_exit();
1649 nfsiod_stop();
1650}
1651
1652/* Not quite true; I just maintain it */
1653MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1654MODULE_LICENSE("GPL");
1655module_param(enable_ino64, bool, 0644);
1656
1657module_init(init_nfs_fs)
1658module_exit(exit_nfs_fs)