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/signal.h>
19#include <linux/time.h>
20#include <linux/kernel.h>
21#include <linux/mm.h>
22#include <linux/string.h>
23#include <linux/stat.h>
24#include <linux/errno.h>
25#include <linux/unistd.h>
26#include <linux/sunrpc/clnt.h>
27#include <linux/sunrpc/stats.h>
28#include <linux/sunrpc/metrics.h>
29#include <linux/nfs_fs.h>
30#include <linux/nfs_mount.h>
31#include <linux/nfs4_mount.h>
32#include <linux/lockd/bind.h>
33#include <linux/seq_file.h>
34#include <linux/mount.h>
35#include <linux/vfs.h>
36#include <linux/inet.h>
37#include <linux/nfs_xdr.h>
38#include <linux/slab.h>
39#include <linux/compat.h>
40#include <linux/freezer.h>
41#include <linux/uaccess.h>
42#include <linux/iversion.h>
43
44#include "nfs4_fs.h"
45#include "callback.h"
46#include "delegation.h"
47#include "iostat.h"
48#include "internal.h"
49#include "fscache.h"
50#include "pnfs.h"
51#include "nfs.h"
52#include "netns.h"
53
54#include "nfstrace.h"
55
56#define NFSDBG_FACILITY NFSDBG_VFS
57
58#define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
59
60/* Default is to see 64-bit inode numbers */
61static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
62
63static void nfs_invalidate_inode(struct inode *);
64static int nfs_update_inode(struct inode *, struct nfs_fattr *);
65
66static struct kmem_cache * nfs_inode_cachep;
67
68static inline unsigned long
69nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
70{
71 return nfs_fileid_to_ino_t(fattr->fileid);
72}
73
74static int nfs_wait_killable(int mode)
75{
76 freezable_schedule_unsafe();
77 if (signal_pending_state(mode, current))
78 return -ERESTARTSYS;
79 return 0;
80}
81
82int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
83{
84 return nfs_wait_killable(mode);
85}
86EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
87
88/**
89 * nfs_compat_user_ino64 - returns the user-visible inode number
90 * @fileid: 64-bit fileid
91 *
92 * This function returns a 32-bit inode number if the boot parameter
93 * nfs.enable_ino64 is zero.
94 */
95u64 nfs_compat_user_ino64(u64 fileid)
96{
97#ifdef CONFIG_COMPAT
98 compat_ulong_t ino;
99#else
100 unsigned long ino;
101#endif
102
103 if (enable_ino64)
104 return fileid;
105 ino = fileid;
106 if (sizeof(ino) < sizeof(fileid))
107 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
108 return ino;
109}
110
111int nfs_drop_inode(struct inode *inode)
112{
113 return NFS_STALE(inode) || generic_drop_inode(inode);
114}
115EXPORT_SYMBOL_GPL(nfs_drop_inode);
116
117void nfs_clear_inode(struct inode *inode)
118{
119 /*
120 * The following should never happen...
121 */
122 WARN_ON_ONCE(nfs_have_writebacks(inode));
123 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
124 nfs_zap_acl_cache(inode);
125 nfs_access_zap_cache(inode);
126 nfs_fscache_clear_inode(inode);
127}
128EXPORT_SYMBOL_GPL(nfs_clear_inode);
129
130void nfs_evict_inode(struct inode *inode)
131{
132 truncate_inode_pages_final(&inode->i_data);
133 clear_inode(inode);
134 nfs_clear_inode(inode);
135}
136
137int nfs_sync_inode(struct inode *inode)
138{
139 inode_dio_wait(inode);
140 return nfs_wb_all(inode);
141}
142EXPORT_SYMBOL_GPL(nfs_sync_inode);
143
144/**
145 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
146 */
147int nfs_sync_mapping(struct address_space *mapping)
148{
149 int ret = 0;
150
151 if (mapping->nrpages != 0) {
152 unmap_mapping_range(mapping, 0, 0, 0);
153 ret = nfs_wb_all(mapping->host);
154 }
155 return ret;
156}
157
158static int nfs_attribute_timeout(struct inode *inode)
159{
160 struct nfs_inode *nfsi = NFS_I(inode);
161
162 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
163}
164
165static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
166{
167 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
168
169 /* Special case for the pagecache or access cache */
170 if (flags == NFS_INO_REVAL_PAGECACHE &&
171 !(cache_validity & NFS_INO_REVAL_FORCED))
172 return false;
173 return (cache_validity & flags) != 0;
174}
175
176static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
177{
178 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
179
180 if ((cache_validity & flags) != 0)
181 return true;
182 if (nfs_attribute_timeout(inode))
183 return true;
184 return false;
185}
186
187bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
188{
189 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
190 return nfs_check_cache_invalid_delegated(inode, flags);
191
192 return nfs_check_cache_invalid_not_delegated(inode, flags);
193}
194
195static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
196{
197 struct nfs_inode *nfsi = NFS_I(inode);
198 bool have_delegation = nfs_have_delegated_attributes(inode);
199
200 if (have_delegation)
201 flags &= ~(NFS_INO_INVALID_CHANGE|NFS_INO_REVAL_PAGECACHE);
202 if (inode->i_mapping->nrpages == 0)
203 flags &= ~NFS_INO_INVALID_DATA;
204 nfsi->cache_validity |= flags;
205 if (flags & NFS_INO_INVALID_DATA)
206 nfs_fscache_invalidate(inode);
207}
208
209/*
210 * Invalidate the local caches
211 */
212static void nfs_zap_caches_locked(struct inode *inode)
213{
214 struct nfs_inode *nfsi = NFS_I(inode);
215 int mode = inode->i_mode;
216
217 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
218
219 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
220 nfsi->attrtimeo_timestamp = jiffies;
221
222 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
223 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
224 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
225 | NFS_INO_INVALID_DATA
226 | NFS_INO_INVALID_ACCESS
227 | NFS_INO_INVALID_ACL
228 | NFS_INO_REVAL_PAGECACHE);
229 } else
230 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
231 | NFS_INO_INVALID_ACCESS
232 | NFS_INO_INVALID_ACL
233 | NFS_INO_REVAL_PAGECACHE);
234 nfs_zap_label_cache_locked(nfsi);
235}
236
237void nfs_zap_caches(struct inode *inode)
238{
239 spin_lock(&inode->i_lock);
240 nfs_zap_caches_locked(inode);
241 spin_unlock(&inode->i_lock);
242}
243
244void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
245{
246 if (mapping->nrpages != 0) {
247 spin_lock(&inode->i_lock);
248 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
249 spin_unlock(&inode->i_lock);
250 }
251}
252
253void nfs_zap_acl_cache(struct inode *inode)
254{
255 void (*clear_acl_cache)(struct inode *);
256
257 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
258 if (clear_acl_cache != NULL)
259 clear_acl_cache(inode);
260 spin_lock(&inode->i_lock);
261 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
262 spin_unlock(&inode->i_lock);
263}
264EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
265
266void nfs_invalidate_atime(struct inode *inode)
267{
268 spin_lock(&inode->i_lock);
269 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
270 spin_unlock(&inode->i_lock);
271}
272EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
273
274/*
275 * Invalidate, but do not unhash, the inode.
276 * NB: must be called with inode->i_lock held!
277 */
278static void nfs_invalidate_inode(struct inode *inode)
279{
280 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
281 nfs_zap_caches_locked(inode);
282}
283
284struct nfs_find_desc {
285 struct nfs_fh *fh;
286 struct nfs_fattr *fattr;
287};
288
289/*
290 * In NFSv3 we can have 64bit inode numbers. In order to support
291 * this, and re-exported directories (also seen in NFSv2)
292 * we are forced to allow 2 different inodes to have the same
293 * i_ino.
294 */
295static int
296nfs_find_actor(struct inode *inode, void *opaque)
297{
298 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
299 struct nfs_fh *fh = desc->fh;
300 struct nfs_fattr *fattr = desc->fattr;
301
302 if (NFS_FILEID(inode) != fattr->fileid)
303 return 0;
304 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
305 return 0;
306 if (nfs_compare_fh(NFS_FH(inode), fh))
307 return 0;
308 if (is_bad_inode(inode) || NFS_STALE(inode))
309 return 0;
310 return 1;
311}
312
313static int
314nfs_init_locked(struct inode *inode, void *opaque)
315{
316 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
317 struct nfs_fattr *fattr = desc->fattr;
318
319 set_nfs_fileid(inode, fattr->fileid);
320 inode->i_mode = fattr->mode;
321 nfs_copy_fh(NFS_FH(inode), desc->fh);
322 return 0;
323}
324
325#ifdef CONFIG_NFS_V4_SECURITY_LABEL
326static void nfs_clear_label_invalid(struct inode *inode)
327{
328 spin_lock(&inode->i_lock);
329 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
330 spin_unlock(&inode->i_lock);
331}
332
333void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
334 struct nfs4_label *label)
335{
336 int error;
337
338 if (label == NULL)
339 return;
340
341 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
342 error = security_inode_notifysecctx(inode, label->label,
343 label->len);
344 if (error)
345 printk(KERN_ERR "%s() %s %d "
346 "security_inode_notifysecctx() %d\n",
347 __func__,
348 (char *)label->label,
349 label->len, error);
350 nfs_clear_label_invalid(inode);
351 }
352}
353
354struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
355{
356 struct nfs4_label *label = NULL;
357 int minor_version = server->nfs_client->cl_minorversion;
358
359 if (minor_version < 2)
360 return label;
361
362 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
363 return label;
364
365 label = kzalloc(sizeof(struct nfs4_label), flags);
366 if (label == NULL)
367 return ERR_PTR(-ENOMEM);
368
369 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
370 if (label->label == NULL) {
371 kfree(label);
372 return ERR_PTR(-ENOMEM);
373 }
374 label->len = NFS4_MAXLABELLEN;
375
376 return label;
377}
378EXPORT_SYMBOL_GPL(nfs4_label_alloc);
379#else
380void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
381 struct nfs4_label *label)
382{
383}
384#endif
385EXPORT_SYMBOL_GPL(nfs_setsecurity);
386
387/* Search for inode identified by fh, fileid and i_mode in inode cache. */
388struct inode *
389nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
390{
391 struct nfs_find_desc desc = {
392 .fh = fh,
393 .fattr = fattr,
394 };
395 struct inode *inode;
396 unsigned long hash;
397
398 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
399 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
400 return NULL;
401
402 hash = nfs_fattr_to_ino_t(fattr);
403 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
404
405 dprintk("%s: returning %p\n", __func__, inode);
406 return inode;
407}
408
409/*
410 * This is our front-end to iget that looks up inodes by file handle
411 * instead of inode number.
412 */
413struct inode *
414nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
415{
416 struct nfs_find_desc desc = {
417 .fh = fh,
418 .fattr = fattr
419 };
420 struct inode *inode = ERR_PTR(-ENOENT);
421 unsigned long hash;
422
423 nfs_attr_check_mountpoint(sb, fattr);
424
425 if (nfs_attr_use_mounted_on_fileid(fattr))
426 fattr->fileid = fattr->mounted_on_fileid;
427 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
428 goto out_no_inode;
429 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
430 goto out_no_inode;
431
432 hash = nfs_fattr_to_ino_t(fattr);
433
434 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
435 if (inode == NULL) {
436 inode = ERR_PTR(-ENOMEM);
437 goto out_no_inode;
438 }
439
440 if (inode->i_state & I_NEW) {
441 struct nfs_inode *nfsi = NFS_I(inode);
442 unsigned long now = jiffies;
443
444 /* We set i_ino for the few things that still rely on it,
445 * such as stat(2) */
446 inode->i_ino = hash;
447
448 /* We can't support update_atime(), since the server will reset it */
449 inode->i_flags |= S_NOATIME|S_NOCMTIME;
450 inode->i_mode = fattr->mode;
451 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
452 && nfs_server_capable(inode, NFS_CAP_MODE))
453 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
454 /* Why so? Because we want revalidate for devices/FIFOs, and
455 * that's precisely what we have in nfs_file_inode_operations.
456 */
457 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
458 if (S_ISREG(inode->i_mode)) {
459 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
460 inode->i_data.a_ops = &nfs_file_aops;
461 } else if (S_ISDIR(inode->i_mode)) {
462 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
463 inode->i_fop = &nfs_dir_operations;
464 inode->i_data.a_ops = &nfs_dir_aops;
465 /* Deal with crossing mountpoints */
466 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
467 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
468 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
469 inode->i_op = &nfs_referral_inode_operations;
470 else
471 inode->i_op = &nfs_mountpoint_inode_operations;
472 inode->i_fop = NULL;
473 inode->i_flags |= S_AUTOMOUNT;
474 }
475 } else if (S_ISLNK(inode->i_mode)) {
476 inode->i_op = &nfs_symlink_inode_operations;
477 inode_nohighmem(inode);
478 } else
479 init_special_inode(inode, inode->i_mode, fattr->rdev);
480
481 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
482 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
483 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
484 inode_set_iversion_raw(inode, 0);
485 inode->i_size = 0;
486 clear_nlink(inode);
487 inode->i_uid = make_kuid(&init_user_ns, -2);
488 inode->i_gid = make_kgid(&init_user_ns, -2);
489 inode->i_blocks = 0;
490 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
491 nfsi->write_io = 0;
492 nfsi->read_io = 0;
493
494 nfsi->read_cache_jiffies = fattr->time_start;
495 nfsi->attr_gencount = fattr->gencount;
496 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
497 inode->i_atime = fattr->atime;
498 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
499 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
500 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
501 inode->i_mtime = fattr->mtime;
502 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
503 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
504 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
505 inode->i_ctime = fattr->ctime;
506 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
507 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
508 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
509 inode_set_iversion_raw(inode, fattr->change_attr);
510 else
511 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
512 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
513 inode->i_size = nfs_size_to_loff_t(fattr->size);
514 else
515 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
516 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
517 set_nlink(inode, fattr->nlink);
518 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
519 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
520 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
521 inode->i_uid = fattr->uid;
522 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
523 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
524 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
525 inode->i_gid = fattr->gid;
526 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
527 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
528 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
529 inode->i_blocks = fattr->du.nfs2.blocks;
530 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
531 /*
532 * report the blocks in 512byte units
533 */
534 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
535 }
536
537 nfs_setsecurity(inode, fattr, label);
538
539 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
540 nfsi->attrtimeo_timestamp = now;
541 nfsi->access_cache = RB_ROOT;
542
543 nfs_fscache_init_inode(inode);
544
545 unlock_new_inode(inode);
546 } else {
547 int err = nfs_refresh_inode(inode, fattr);
548 if (err < 0) {
549 iput(inode);
550 inode = ERR_PTR(err);
551 goto out_no_inode;
552 }
553 }
554 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
555 inode->i_sb->s_id,
556 (unsigned long long)NFS_FILEID(inode),
557 nfs_display_fhandle_hash(fh),
558 atomic_read(&inode->i_count));
559
560out:
561 return inode;
562
563out_no_inode:
564 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
565 goto out;
566}
567EXPORT_SYMBOL_GPL(nfs_fhget);
568
569#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)
570
571int
572nfs_setattr(struct dentry *dentry, struct iattr *attr)
573{
574 struct inode *inode = d_inode(dentry);
575 struct nfs_fattr *fattr;
576 int error = 0;
577
578 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
579
580 /* skip mode change if it's just for clearing setuid/setgid */
581 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
582 attr->ia_valid &= ~ATTR_MODE;
583
584 if (attr->ia_valid & ATTR_SIZE) {
585 BUG_ON(!S_ISREG(inode->i_mode));
586
587 error = inode_newsize_ok(inode, attr->ia_size);
588 if (error)
589 return error;
590
591 if (attr->ia_size == i_size_read(inode))
592 attr->ia_valid &= ~ATTR_SIZE;
593 }
594
595 /* Optimization: if the end result is no change, don't RPC */
596 attr->ia_valid &= NFS_VALID_ATTRS;
597 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
598 return 0;
599
600 trace_nfs_setattr_enter(inode);
601
602 /* Write all dirty data */
603 if (S_ISREG(inode->i_mode))
604 nfs_sync_inode(inode);
605
606 fattr = nfs_alloc_fattr();
607 if (fattr == NULL) {
608 error = -ENOMEM;
609 goto out;
610 }
611
612 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
613 if (error == 0)
614 error = nfs_refresh_inode(inode, fattr);
615 nfs_free_fattr(fattr);
616out:
617 trace_nfs_setattr_exit(inode, error);
618 return error;
619}
620EXPORT_SYMBOL_GPL(nfs_setattr);
621
622/**
623 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
624 * @inode: inode of the file used
625 * @offset: file offset to start truncating
626 *
627 * This is a copy of the common vmtruncate, but with the locking
628 * corrected to take into account the fact that NFS requires
629 * inode->i_size to be updated under the inode->i_lock.
630 * Note: must be called with inode->i_lock held!
631 */
632static int nfs_vmtruncate(struct inode * inode, loff_t offset)
633{
634 int err;
635
636 err = inode_newsize_ok(inode, offset);
637 if (err)
638 goto out;
639
640 i_size_write(inode, offset);
641 /* Optimisation */
642 if (offset == 0)
643 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
644 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
645
646 spin_unlock(&inode->i_lock);
647 truncate_pagecache(inode, offset);
648 spin_lock(&inode->i_lock);
649out:
650 return err;
651}
652
653/**
654 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
655 * @inode: pointer to struct inode
656 * @attr: pointer to struct iattr
657 * @fattr: pointer to struct nfs_fattr
658 *
659 * Note: we do this in the *proc.c in order to ensure that
660 * it works for things like exclusive creates too.
661 */
662void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
663 struct nfs_fattr *fattr)
664{
665 /* Barrier: bump the attribute generation count. */
666 nfs_fattr_set_barrier(fattr);
667
668 spin_lock(&inode->i_lock);
669 NFS_I(inode)->attr_gencount = fattr->gencount;
670 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
671 | NFS_INO_INVALID_CTIME);
672 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
673 if ((attr->ia_valid & ATTR_MODE) != 0) {
674 int mode = attr->ia_mode & S_IALLUGO;
675 mode |= inode->i_mode & ~S_IALLUGO;
676 inode->i_mode = mode;
677 }
678 if ((attr->ia_valid & ATTR_UID) != 0)
679 inode->i_uid = attr->ia_uid;
680 if ((attr->ia_valid & ATTR_GID) != 0)
681 inode->i_gid = attr->ia_gid;
682 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
683 | NFS_INO_INVALID_ACL);
684 }
685 if ((attr->ia_valid & ATTR_SIZE) != 0) {
686 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
687 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
688 nfs_vmtruncate(inode, attr->ia_size);
689 }
690 if (fattr->valid)
691 nfs_update_inode(inode, fattr);
692 spin_unlock(&inode->i_lock);
693}
694EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
695
696static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
697{
698 struct dentry *parent;
699
700 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
701 return;
702 parent = dget_parent(dentry);
703 nfs_force_use_readdirplus(d_inode(parent));
704 dput(parent);
705}
706
707static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
708{
709 struct dentry *parent;
710
711 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
712 return;
713 parent = dget_parent(dentry);
714 nfs_advise_use_readdirplus(d_inode(parent));
715 dput(parent);
716}
717
718static bool nfs_need_revalidate_inode(struct inode *inode)
719{
720 if (NFS_I(inode)->cache_validity &
721 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
722 return true;
723 if (nfs_attribute_cache_expired(inode))
724 return true;
725 return false;
726}
727
728int nfs_getattr(const struct path *path, struct kstat *stat,
729 u32 request_mask, unsigned int query_flags)
730{
731 struct inode *inode = d_inode(path->dentry);
732 struct nfs_server *server = NFS_SERVER(inode);
733 unsigned long cache_validity;
734 int err = 0;
735 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
736 bool do_update = false;
737
738 trace_nfs_getattr_enter(inode);
739
740 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync)
741 goto out_no_update;
742
743 /* Flush out writes to the server in order to update c/mtime. */
744 if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
745 S_ISREG(inode->i_mode)) {
746 err = filemap_write_and_wait(inode->i_mapping);
747 if (err)
748 goto out;
749 }
750
751 /*
752 * We may force a getattr if the user cares about atime.
753 *
754 * Note that we only have to check the vfsmount flags here:
755 * - NFS always sets S_NOATIME by so checking it would give a
756 * bogus result
757 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
758 * no point in checking those.
759 */
760 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
761 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
762 request_mask &= ~STATX_ATIME;
763
764 /* Is the user requesting attributes that might need revalidation? */
765 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
766 STATX_MTIME|STATX_UID|STATX_GID|
767 STATX_SIZE|STATX_BLOCKS)))
768 goto out_no_revalidate;
769
770 /* Check whether the cached attributes are stale */
771 do_update |= force_sync || nfs_attribute_cache_expired(inode);
772 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
773 do_update |= cache_validity &
774 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
775 if (request_mask & STATX_ATIME)
776 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
777 if (request_mask & (STATX_CTIME|STATX_MTIME))
778 do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
779 if (do_update) {
780 /* Update the attribute cache */
781 if (!(server->flags & NFS_MOUNT_NOAC))
782 nfs_readdirplus_parent_cache_miss(path->dentry);
783 else
784 nfs_readdirplus_parent_cache_hit(path->dentry);
785 err = __nfs_revalidate_inode(server, inode);
786 if (err)
787 goto out;
788 } else
789 nfs_readdirplus_parent_cache_hit(path->dentry);
790out_no_revalidate:
791 /* Only return attributes that were revalidated. */
792 stat->result_mask &= request_mask;
793out_no_update:
794 generic_fillattr(inode, stat);
795 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
796 if (S_ISDIR(inode->i_mode))
797 stat->blksize = NFS_SERVER(inode)->dtsize;
798out:
799 trace_nfs_getattr_exit(inode, err);
800 return err;
801}
802EXPORT_SYMBOL_GPL(nfs_getattr);
803
804static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
805{
806 refcount_set(&l_ctx->count, 1);
807 l_ctx->lockowner = current->files;
808 INIT_LIST_HEAD(&l_ctx->list);
809 atomic_set(&l_ctx->io_count, 0);
810}
811
812static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
813{
814 struct nfs_lock_context *head = &ctx->lock_context;
815 struct nfs_lock_context *pos = head;
816
817 do {
818 if (pos->lockowner != current->files)
819 continue;
820 refcount_inc(&pos->count);
821 return pos;
822 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
823 return NULL;
824}
825
826struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
827{
828 struct nfs_lock_context *res, *new = NULL;
829 struct inode *inode = d_inode(ctx->dentry);
830
831 spin_lock(&inode->i_lock);
832 res = __nfs_find_lock_context(ctx);
833 if (res == NULL) {
834 spin_unlock(&inode->i_lock);
835 new = kmalloc(sizeof(*new), GFP_KERNEL);
836 if (new == NULL)
837 return ERR_PTR(-ENOMEM);
838 nfs_init_lock_context(new);
839 spin_lock(&inode->i_lock);
840 res = __nfs_find_lock_context(ctx);
841 if (res == NULL) {
842 list_add_tail(&new->list, &ctx->lock_context.list);
843 new->open_context = ctx;
844 res = new;
845 new = NULL;
846 }
847 }
848 spin_unlock(&inode->i_lock);
849 kfree(new);
850 return res;
851}
852EXPORT_SYMBOL_GPL(nfs_get_lock_context);
853
854void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
855{
856 struct nfs_open_context *ctx = l_ctx->open_context;
857 struct inode *inode = d_inode(ctx->dentry);
858
859 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
860 return;
861 list_del(&l_ctx->list);
862 spin_unlock(&inode->i_lock);
863 kfree(l_ctx);
864}
865EXPORT_SYMBOL_GPL(nfs_put_lock_context);
866
867/**
868 * nfs_close_context - Common close_context() routine NFSv2/v3
869 * @ctx: pointer to context
870 * @is_sync: is this a synchronous close
871 *
872 * Ensure that the attributes are up to date if we're mounted
873 * with close-to-open semantics and we have cached data that will
874 * need to be revalidated on open.
875 */
876void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
877{
878 struct nfs_inode *nfsi;
879 struct inode *inode;
880 struct nfs_server *server;
881
882 if (!(ctx->mode & FMODE_WRITE))
883 return;
884 if (!is_sync)
885 return;
886 inode = d_inode(ctx->dentry);
887 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
888 return;
889 nfsi = NFS_I(inode);
890 if (inode->i_mapping->nrpages == 0)
891 return;
892 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
893 return;
894 if (!list_empty(&nfsi->open_files))
895 return;
896 server = NFS_SERVER(inode);
897 if (server->flags & NFS_MOUNT_NOCTO)
898 return;
899 nfs_revalidate_inode(server, inode);
900}
901EXPORT_SYMBOL_GPL(nfs_close_context);
902
903struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
904 fmode_t f_mode,
905 struct file *filp)
906{
907 struct nfs_open_context *ctx;
908 struct rpc_cred *cred = rpc_lookup_cred();
909 if (IS_ERR(cred))
910 return ERR_CAST(cred);
911
912 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
913 if (!ctx) {
914 put_rpccred(cred);
915 return ERR_PTR(-ENOMEM);
916 }
917 nfs_sb_active(dentry->d_sb);
918 ctx->dentry = dget(dentry);
919 ctx->cred = cred;
920 ctx->state = NULL;
921 ctx->mode = f_mode;
922 ctx->flags = 0;
923 ctx->error = 0;
924 ctx->flock_owner = (fl_owner_t)filp;
925 nfs_init_lock_context(&ctx->lock_context);
926 ctx->lock_context.open_context = ctx;
927 INIT_LIST_HEAD(&ctx->list);
928 ctx->mdsthreshold = NULL;
929 return ctx;
930}
931EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
932
933struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
934{
935 if (ctx != NULL)
936 refcount_inc(&ctx->lock_context.count);
937 return ctx;
938}
939EXPORT_SYMBOL_GPL(get_nfs_open_context);
940
941static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
942{
943 struct inode *inode = d_inode(ctx->dentry);
944 struct super_block *sb = ctx->dentry->d_sb;
945
946 if (!list_empty(&ctx->list)) {
947 if (!refcount_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
948 return;
949 list_del(&ctx->list);
950 spin_unlock(&inode->i_lock);
951 } else if (!refcount_dec_and_test(&ctx->lock_context.count))
952 return;
953 if (inode != NULL)
954 NFS_PROTO(inode)->close_context(ctx, is_sync);
955 if (ctx->cred != NULL)
956 put_rpccred(ctx->cred);
957 dput(ctx->dentry);
958 nfs_sb_deactive(sb);
959 kfree(ctx->mdsthreshold);
960 kfree(ctx);
961}
962
963void put_nfs_open_context(struct nfs_open_context *ctx)
964{
965 __put_nfs_open_context(ctx, 0);
966}
967EXPORT_SYMBOL_GPL(put_nfs_open_context);
968
969static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
970{
971 __put_nfs_open_context(ctx, 1);
972}
973
974/*
975 * Ensure that mmap has a recent RPC credential for use when writing out
976 * shared pages
977 */
978void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
979{
980 struct inode *inode = d_inode(ctx->dentry);
981 struct nfs_inode *nfsi = NFS_I(inode);
982
983 spin_lock(&inode->i_lock);
984 if (ctx->mode & FMODE_WRITE)
985 list_add(&ctx->list, &nfsi->open_files);
986 else
987 list_add_tail(&ctx->list, &nfsi->open_files);
988 spin_unlock(&inode->i_lock);
989}
990EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
991
992void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
993{
994 filp->private_data = get_nfs_open_context(ctx);
995 if (list_empty(&ctx->list))
996 nfs_inode_attach_open_context(ctx);
997}
998EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
999
1000/*
1001 * Given an inode, search for an open context with the desired characteristics
1002 */
1003struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
1004{
1005 struct nfs_inode *nfsi = NFS_I(inode);
1006 struct nfs_open_context *pos, *ctx = NULL;
1007
1008 spin_lock(&inode->i_lock);
1009 list_for_each_entry(pos, &nfsi->open_files, list) {
1010 if (cred != NULL && pos->cred != cred)
1011 continue;
1012 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1013 continue;
1014 ctx = get_nfs_open_context(pos);
1015 break;
1016 }
1017 spin_unlock(&inode->i_lock);
1018 return ctx;
1019}
1020
1021void nfs_file_clear_open_context(struct file *filp)
1022{
1023 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1024
1025 if (ctx) {
1026 struct inode *inode = d_inode(ctx->dentry);
1027
1028 /*
1029 * We fatal error on write before. Try to writeback
1030 * every page again.
1031 */
1032 if (ctx->error < 0)
1033 invalidate_inode_pages2(inode->i_mapping);
1034 filp->private_data = NULL;
1035 spin_lock(&inode->i_lock);
1036 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1037 spin_unlock(&inode->i_lock);
1038 put_nfs_open_context_sync(ctx);
1039 }
1040}
1041
1042/*
1043 * These allocate and release file read/write context information.
1044 */
1045int nfs_open(struct inode *inode, struct file *filp)
1046{
1047 struct nfs_open_context *ctx;
1048
1049 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1050 if (IS_ERR(ctx))
1051 return PTR_ERR(ctx);
1052 nfs_file_set_open_context(filp, ctx);
1053 put_nfs_open_context(ctx);
1054 nfs_fscache_open_file(inode, filp);
1055 return 0;
1056}
1057
1058/*
1059 * This function is called whenever some part of NFS notices that
1060 * the cached attributes have to be refreshed.
1061 */
1062int
1063__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1064{
1065 int status = -ESTALE;
1066 struct nfs4_label *label = NULL;
1067 struct nfs_fattr *fattr = NULL;
1068 struct nfs_inode *nfsi = NFS_I(inode);
1069
1070 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1071 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1072
1073 trace_nfs_revalidate_inode_enter(inode);
1074
1075 if (is_bad_inode(inode))
1076 goto out;
1077 if (NFS_STALE(inode))
1078 goto out;
1079
1080 /* pNFS: Attributes aren't updated until we layoutcommit */
1081 if (S_ISREG(inode->i_mode)) {
1082 status = pnfs_sync_inode(inode, false);
1083 if (status)
1084 goto out;
1085 }
1086
1087 status = -ENOMEM;
1088 fattr = nfs_alloc_fattr();
1089 if (fattr == NULL)
1090 goto out;
1091
1092 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1093
1094 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1095 if (IS_ERR(label)) {
1096 status = PTR_ERR(label);
1097 goto out;
1098 }
1099
1100 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
1101 if (status != 0) {
1102 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1103 inode->i_sb->s_id,
1104 (unsigned long long)NFS_FILEID(inode), status);
1105 if (status == -ESTALE) {
1106 nfs_zap_caches(inode);
1107 if (!S_ISDIR(inode->i_mode))
1108 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1109 }
1110 goto err_out;
1111 }
1112
1113 status = nfs_refresh_inode(inode, fattr);
1114 if (status) {
1115 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1116 inode->i_sb->s_id,
1117 (unsigned long long)NFS_FILEID(inode), status);
1118 goto err_out;
1119 }
1120
1121 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1122 nfs_zap_acl_cache(inode);
1123
1124 nfs_setsecurity(inode, fattr, label);
1125
1126 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1127 inode->i_sb->s_id,
1128 (unsigned long long)NFS_FILEID(inode));
1129
1130err_out:
1131 nfs4_label_free(label);
1132out:
1133 nfs_free_fattr(fattr);
1134 trace_nfs_revalidate_inode_exit(inode, status);
1135 return status;
1136}
1137
1138int nfs_attribute_cache_expired(struct inode *inode)
1139{
1140 if (nfs_have_delegated_attributes(inode))
1141 return 0;
1142 return nfs_attribute_timeout(inode);
1143}
1144
1145/**
1146 * nfs_revalidate_inode - Revalidate the inode attributes
1147 * @server - pointer to nfs_server struct
1148 * @inode - pointer to inode struct
1149 *
1150 * Updates inode attribute information by retrieving the data from the server.
1151 */
1152int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1153{
1154 if (!nfs_need_revalidate_inode(inode))
1155 return NFS_STALE(inode) ? -ESTALE : 0;
1156 return __nfs_revalidate_inode(server, inode);
1157}
1158EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1159
1160static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1161{
1162 struct nfs_inode *nfsi = NFS_I(inode);
1163 int ret;
1164
1165 if (mapping->nrpages != 0) {
1166 if (S_ISREG(inode->i_mode)) {
1167 ret = nfs_sync_mapping(mapping);
1168 if (ret < 0)
1169 return ret;
1170 }
1171 ret = invalidate_inode_pages2(mapping);
1172 if (ret < 0)
1173 return ret;
1174 }
1175 if (S_ISDIR(inode->i_mode)) {
1176 spin_lock(&inode->i_lock);
1177 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1178 spin_unlock(&inode->i_lock);
1179 }
1180 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1181 nfs_fscache_wait_on_invalidate(inode);
1182
1183 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1184 inode->i_sb->s_id,
1185 (unsigned long long)NFS_FILEID(inode));
1186 return 0;
1187}
1188
1189bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1190{
1191 return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1192 NFS_STALE(inode);
1193}
1194
1195int nfs_revalidate_mapping_rcu(struct inode *inode)
1196{
1197 struct nfs_inode *nfsi = NFS_I(inode);
1198 unsigned long *bitlock = &nfsi->flags;
1199 int ret = 0;
1200
1201 if (IS_SWAPFILE(inode))
1202 goto out;
1203 if (nfs_mapping_need_revalidate_inode(inode)) {
1204 ret = -ECHILD;
1205 goto out;
1206 }
1207 spin_lock(&inode->i_lock);
1208 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1209 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1210 ret = -ECHILD;
1211 spin_unlock(&inode->i_lock);
1212out:
1213 return ret;
1214}
1215
1216/**
1217 * nfs_revalidate_mapping - Revalidate the pagecache
1218 * @inode - pointer to host inode
1219 * @mapping - pointer to mapping
1220 */
1221int nfs_revalidate_mapping(struct inode *inode,
1222 struct address_space *mapping)
1223{
1224 struct nfs_inode *nfsi = NFS_I(inode);
1225 unsigned long *bitlock = &nfsi->flags;
1226 int ret = 0;
1227
1228 /* swapfiles are not supposed to be shared. */
1229 if (IS_SWAPFILE(inode))
1230 goto out;
1231
1232 if (nfs_mapping_need_revalidate_inode(inode)) {
1233 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1234 if (ret < 0)
1235 goto out;
1236 }
1237
1238 /*
1239 * We must clear NFS_INO_INVALID_DATA first to ensure that
1240 * invalidations that come in while we're shooting down the mappings
1241 * are respected. But, that leaves a race window where one revalidator
1242 * can clear the flag, and then another checks it before the mapping
1243 * gets invalidated. Fix that by serializing access to this part of
1244 * the function.
1245 *
1246 * At the same time, we need to allow other tasks to see whether we
1247 * might be in the middle of invalidating the pages, so we only set
1248 * the bit lock here if it looks like we're going to be doing that.
1249 */
1250 for (;;) {
1251 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1252 nfs_wait_bit_killable, TASK_KILLABLE);
1253 if (ret)
1254 goto out;
1255 spin_lock(&inode->i_lock);
1256 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1257 spin_unlock(&inode->i_lock);
1258 continue;
1259 }
1260 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1261 break;
1262 spin_unlock(&inode->i_lock);
1263 goto out;
1264 }
1265
1266 set_bit(NFS_INO_INVALIDATING, bitlock);
1267 smp_wmb();
1268 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1269 spin_unlock(&inode->i_lock);
1270 trace_nfs_invalidate_mapping_enter(inode);
1271 ret = nfs_invalidate_mapping(inode, mapping);
1272 trace_nfs_invalidate_mapping_exit(inode, ret);
1273
1274 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1275 smp_mb__after_atomic();
1276 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1277out:
1278 return ret;
1279}
1280
1281static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1282{
1283 struct inode *inode = &nfsi->vfs_inode;
1284
1285 assert_spin_locked(&inode->i_lock);
1286
1287 if (!S_ISREG(inode->i_mode))
1288 return false;
1289 if (list_empty(&nfsi->open_files))
1290 return false;
1291 /* Note: This relies on nfsi->open_files being ordered with writers
1292 * being placed at the head of the list.
1293 * See nfs_inode_attach_open_context()
1294 */
1295 return (list_first_entry(&nfsi->open_files,
1296 struct nfs_open_context,
1297 list)->mode & FMODE_WRITE) == FMODE_WRITE;
1298}
1299
1300static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1301{
1302 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1303}
1304
1305static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1306{
1307 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1308 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1309 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1310 inode_set_iversion_raw(inode, fattr->change_attr);
1311 if (S_ISDIR(inode->i_mode))
1312 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1313 }
1314 /* If we have atomic WCC data, we may update some attributes */
1315 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1316 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1317 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1318 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1319 }
1320
1321 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1322 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1323 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1324 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1325 if (S_ISDIR(inode->i_mode))
1326 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1327 }
1328 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1329 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1330 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1331 && !nfs_have_writebacks(inode)) {
1332 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1333 }
1334}
1335
1336/**
1337 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1338 * @inode - pointer to inode
1339 * @fattr - updated attributes
1340 *
1341 * Verifies the attribute cache. If we have just changed the attributes,
1342 * so that fattr carries weak cache consistency data, then it may
1343 * also update the ctime/mtime/change_attribute.
1344 */
1345static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1346{
1347 struct nfs_inode *nfsi = NFS_I(inode);
1348 loff_t cur_size, new_isize;
1349 unsigned long invalid = 0;
1350
1351
1352 if (nfs_have_delegated_attributes(inode))
1353 return 0;
1354 /* Has the inode gone and changed behind our back? */
1355 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1356 return -ESTALE;
1357 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1358 return -ESTALE;
1359
1360 if (!nfs_file_has_buffered_writers(nfsi)) {
1361 /* Verify a few of the more important attributes */
1362 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1363 invalid |= NFS_INO_INVALID_CHANGE
1364 | NFS_INO_REVAL_PAGECACHE;
1365
1366 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1367 invalid |= NFS_INO_INVALID_MTIME;
1368
1369 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&inode->i_ctime, &fattr->ctime))
1370 invalid |= NFS_INO_INVALID_CTIME;
1371
1372 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1373 cur_size = i_size_read(inode);
1374 new_isize = nfs_size_to_loff_t(fattr->size);
1375 if (cur_size != new_isize)
1376 invalid |= NFS_INO_INVALID_SIZE
1377 | NFS_INO_REVAL_PAGECACHE;
1378 }
1379 }
1380
1381 /* Have any file permissions changed? */
1382 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1383 invalid |= NFS_INO_INVALID_ACCESS
1384 | NFS_INO_INVALID_ACL
1385 | NFS_INO_INVALID_OTHER;
1386 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1387 invalid |= NFS_INO_INVALID_ACCESS
1388 | NFS_INO_INVALID_ACL
1389 | NFS_INO_INVALID_OTHER;
1390 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1391 invalid |= NFS_INO_INVALID_ACCESS
1392 | NFS_INO_INVALID_ACL
1393 | NFS_INO_INVALID_OTHER;
1394
1395 /* Has the link count changed? */
1396 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1397 invalid |= NFS_INO_INVALID_OTHER;
1398
1399 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1400 invalid |= NFS_INO_INVALID_ATIME;
1401
1402 if (invalid != 0)
1403 nfs_set_cache_invalid(inode, invalid | NFS_INO_REVAL_FORCED);
1404
1405 nfsi->read_cache_jiffies = fattr->time_start;
1406 return 0;
1407}
1408
1409static atomic_long_t nfs_attr_generation_counter;
1410
1411static unsigned long nfs_read_attr_generation_counter(void)
1412{
1413 return atomic_long_read(&nfs_attr_generation_counter);
1414}
1415
1416unsigned long nfs_inc_attr_generation_counter(void)
1417{
1418 return atomic_long_inc_return(&nfs_attr_generation_counter);
1419}
1420EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1421
1422void nfs_fattr_init(struct nfs_fattr *fattr)
1423{
1424 fattr->valid = 0;
1425 fattr->time_start = jiffies;
1426 fattr->gencount = nfs_inc_attr_generation_counter();
1427 fattr->owner_name = NULL;
1428 fattr->group_name = NULL;
1429}
1430EXPORT_SYMBOL_GPL(nfs_fattr_init);
1431
1432/**
1433 * nfs_fattr_set_barrier
1434 * @fattr: attributes
1435 *
1436 * Used to set a barrier after an attribute was updated. This
1437 * barrier ensures that older attributes from RPC calls that may
1438 * have raced with our update cannot clobber these new values.
1439 * Note that you are still responsible for ensuring that other
1440 * operations which change the attribute on the server do not
1441 * collide.
1442 */
1443void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1444{
1445 fattr->gencount = nfs_inc_attr_generation_counter();
1446}
1447
1448struct nfs_fattr *nfs_alloc_fattr(void)
1449{
1450 struct nfs_fattr *fattr;
1451
1452 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1453 if (fattr != NULL)
1454 nfs_fattr_init(fattr);
1455 return fattr;
1456}
1457EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1458
1459struct nfs_fh *nfs_alloc_fhandle(void)
1460{
1461 struct nfs_fh *fh;
1462
1463 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1464 if (fh != NULL)
1465 fh->size = 0;
1466 return fh;
1467}
1468EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1469
1470#ifdef NFS_DEBUG
1471/*
1472 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1473 * in the same way that wireshark does
1474 *
1475 * @fh: file handle
1476 *
1477 * For debugging only.
1478 */
1479u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1480{
1481 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1482 * not on the result */
1483 return nfs_fhandle_hash(fh);
1484}
1485EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1486
1487/*
1488 * _nfs_display_fhandle - display an NFS file handle on the console
1489 *
1490 * @fh: file handle to display
1491 * @caption: display caption
1492 *
1493 * For debugging only.
1494 */
1495void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1496{
1497 unsigned short i;
1498
1499 if (fh == NULL || fh->size == 0) {
1500 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1501 return;
1502 }
1503
1504 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1505 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1506 for (i = 0; i < fh->size; i += 16) {
1507 __be32 *pos = (__be32 *)&fh->data[i];
1508
1509 switch ((fh->size - i - 1) >> 2) {
1510 case 0:
1511 printk(KERN_DEFAULT " %08x\n",
1512 be32_to_cpup(pos));
1513 break;
1514 case 1:
1515 printk(KERN_DEFAULT " %08x %08x\n",
1516 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1517 break;
1518 case 2:
1519 printk(KERN_DEFAULT " %08x %08x %08x\n",
1520 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1521 be32_to_cpup(pos + 2));
1522 break;
1523 default:
1524 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1525 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1526 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1527 }
1528 }
1529}
1530EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1531#endif
1532
1533/**
1534 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1535 * @inode - pointer to inode
1536 * @fattr - attributes
1537 *
1538 * Attempt to divine whether or not an RPC call reply carrying stale
1539 * attributes got scheduled after another call carrying updated ones.
1540 *
1541 * To do so, the function first assumes that a more recent ctime means
1542 * that the attributes in fattr are newer, however it also attempt to
1543 * catch the case where ctime either didn't change, or went backwards
1544 * (if someone reset the clock on the server) by looking at whether
1545 * or not this RPC call was started after the inode was last updated.
1546 * Note also the check for wraparound of 'attr_gencount'
1547 *
1548 * The function returns 'true' if it thinks the attributes in 'fattr' are
1549 * more recent than the ones cached in the inode.
1550 *
1551 */
1552static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1553{
1554 const struct nfs_inode *nfsi = NFS_I(inode);
1555
1556 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1557 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1558}
1559
1560static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1561{
1562 int ret;
1563
1564 trace_nfs_refresh_inode_enter(inode);
1565
1566 if (nfs_inode_attrs_need_update(inode, fattr))
1567 ret = nfs_update_inode(inode, fattr);
1568 else
1569 ret = nfs_check_inode_attributes(inode, fattr);
1570
1571 trace_nfs_refresh_inode_exit(inode, ret);
1572 return ret;
1573}
1574
1575/**
1576 * nfs_refresh_inode - try to update the inode attribute cache
1577 * @inode - pointer to inode
1578 * @fattr - updated attributes
1579 *
1580 * Check that an RPC call that returned attributes has not overlapped with
1581 * other recent updates of the inode metadata, then decide whether it is
1582 * safe to do a full update of the inode attributes, or whether just to
1583 * call nfs_check_inode_attributes.
1584 */
1585int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1586{
1587 int status;
1588
1589 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1590 return 0;
1591 spin_lock(&inode->i_lock);
1592 status = nfs_refresh_inode_locked(inode, fattr);
1593 spin_unlock(&inode->i_lock);
1594
1595 return status;
1596}
1597EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1598
1599static int nfs_post_op_update_inode_locked(struct inode *inode,
1600 struct nfs_fattr *fattr, unsigned int invalid)
1601{
1602 if (S_ISDIR(inode->i_mode))
1603 invalid |= NFS_INO_INVALID_DATA;
1604 nfs_set_cache_invalid(inode, invalid);
1605 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1606 return 0;
1607 return nfs_refresh_inode_locked(inode, fattr);
1608}
1609
1610/**
1611 * nfs_post_op_update_inode - try to update the inode attribute cache
1612 * @inode - pointer to inode
1613 * @fattr - updated attributes
1614 *
1615 * After an operation that has changed the inode metadata, mark the
1616 * attribute cache as being invalid, then try to update it.
1617 *
1618 * NB: if the server didn't return any post op attributes, this
1619 * function will force the retrieval of attributes before the next
1620 * NFS request. Thus it should be used only for operations that
1621 * are expected to change one or more attributes, to avoid
1622 * unnecessary NFS requests and trips through nfs_update_inode().
1623 */
1624int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1625{
1626 int status;
1627
1628 spin_lock(&inode->i_lock);
1629 nfs_fattr_set_barrier(fattr);
1630 status = nfs_post_op_update_inode_locked(inode, fattr,
1631 NFS_INO_INVALID_CHANGE
1632 | NFS_INO_INVALID_CTIME);
1633 spin_unlock(&inode->i_lock);
1634
1635 return status;
1636}
1637EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1638
1639/**
1640 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1641 * @inode - pointer to inode
1642 * @fattr - updated attributes
1643 *
1644 * After an operation that has changed the inode metadata, mark the
1645 * attribute cache as being invalid, then try to update it. Fake up
1646 * weak cache consistency data, if none exist.
1647 *
1648 * This function is mainly designed to be used by the ->write_done() functions.
1649 */
1650int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1651{
1652 int status;
1653
1654 /* Don't do a WCC update if these attributes are already stale */
1655 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1656 !nfs_inode_attrs_need_update(inode, fattr)) {
1657 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1658 | NFS_ATTR_FATTR_PRESIZE
1659 | NFS_ATTR_FATTR_PREMTIME
1660 | NFS_ATTR_FATTR_PRECTIME);
1661 goto out_noforce;
1662 }
1663 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1664 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1665 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1666 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1667 }
1668 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1669 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1670 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1671 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1672 }
1673 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1674 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1675 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1676 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1677 }
1678 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1679 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1680 fattr->pre_size = i_size_read(inode);
1681 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1682 }
1683out_noforce:
1684 status = nfs_post_op_update_inode_locked(inode, fattr,
1685 NFS_INO_INVALID_CHANGE
1686 | NFS_INO_INVALID_CTIME
1687 | NFS_INO_INVALID_MTIME);
1688 return status;
1689}
1690
1691/**
1692 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1693 * @inode - pointer to inode
1694 * @fattr - updated attributes
1695 *
1696 * After an operation that has changed the inode metadata, mark the
1697 * attribute cache as being invalid, then try to update it. Fake up
1698 * weak cache consistency data, if none exist.
1699 *
1700 * This function is mainly designed to be used by the ->write_done() functions.
1701 */
1702int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1703{
1704 int status;
1705
1706 spin_lock(&inode->i_lock);
1707 nfs_fattr_set_barrier(fattr);
1708 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1709 spin_unlock(&inode->i_lock);
1710 return status;
1711}
1712EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1713
1714
1715static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1716 struct nfs_fattr *fattr)
1717{
1718 bool ret1 = true, ret2 = true;
1719
1720 if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1721 ret1 = (nfsi->fileid == fattr->fileid);
1722 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1723 ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1724 return ret1 || ret2;
1725}
1726
1727/*
1728 * Many nfs protocol calls return the new file attributes after
1729 * an operation. Here we update the inode to reflect the state
1730 * of the server's inode.
1731 *
1732 * This is a bit tricky because we have to make sure all dirty pages
1733 * have been sent off to the server before calling invalidate_inode_pages.
1734 * To make sure no other process adds more write requests while we try
1735 * our best to flush them, we make them sleep during the attribute refresh.
1736 *
1737 * A very similar scenario holds for the dir cache.
1738 */
1739static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1740{
1741 struct nfs_server *server;
1742 struct nfs_inode *nfsi = NFS_I(inode);
1743 loff_t cur_isize, new_isize;
1744 unsigned long invalid = 0;
1745 unsigned long now = jiffies;
1746 unsigned long save_cache_validity;
1747 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1748 bool cache_revalidated = true;
1749
1750 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1751 __func__, inode->i_sb->s_id, inode->i_ino,
1752 nfs_display_fhandle_hash(NFS_FH(inode)),
1753 atomic_read(&inode->i_count), fattr->valid);
1754
1755 if (!nfs_fileid_valid(nfsi, fattr)) {
1756 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1757 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1758 NFS_SERVER(inode)->nfs_client->cl_hostname,
1759 inode->i_sb->s_id, (long long)nfsi->fileid,
1760 (long long)fattr->fileid);
1761 goto out_err;
1762 }
1763
1764 /*
1765 * Make sure the inode's type hasn't changed.
1766 */
1767 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1768 /*
1769 * Big trouble! The inode has become a different object.
1770 */
1771 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1772 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1773 goto out_err;
1774 }
1775
1776 server = NFS_SERVER(inode);
1777 /* Update the fsid? */
1778 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1779 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1780 !IS_AUTOMOUNT(inode))
1781 server->fsid = fattr->fsid;
1782
1783 /*
1784 * Update the read time so we don't revalidate too often.
1785 */
1786 nfsi->read_cache_jiffies = fattr->time_start;
1787
1788 save_cache_validity = nfsi->cache_validity;
1789 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1790 | NFS_INO_INVALID_ATIME
1791 | NFS_INO_REVAL_FORCED
1792 | NFS_INO_REVAL_PAGECACHE);
1793
1794 /* Do atomic weak cache consistency updates */
1795 nfs_wcc_update_inode(inode, fattr);
1796
1797 if (pnfs_layoutcommit_outstanding(inode)) {
1798 nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1799 cache_revalidated = false;
1800 }
1801
1802 /* More cache consistency checks */
1803 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1804 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1805 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1806 inode->i_sb->s_id, inode->i_ino);
1807 /* Could it be a race with writeback? */
1808 if (!have_writers) {
1809 invalid |= NFS_INO_INVALID_CHANGE
1810 | NFS_INO_INVALID_DATA
1811 | NFS_INO_INVALID_ACCESS
1812 | NFS_INO_INVALID_ACL;
1813 /* Force revalidate of all attributes */
1814 save_cache_validity |= NFS_INO_INVALID_CTIME
1815 | NFS_INO_INVALID_MTIME
1816 | NFS_INO_INVALID_SIZE
1817 | NFS_INO_INVALID_OTHER;
1818 if (S_ISDIR(inode->i_mode))
1819 nfs_force_lookup_revalidate(inode);
1820 }
1821 inode_set_iversion_raw(inode, fattr->change_attr);
1822 }
1823 } else {
1824 nfsi->cache_validity |= save_cache_validity &
1825 (NFS_INO_INVALID_CHANGE
1826 | NFS_INO_REVAL_PAGECACHE
1827 | NFS_INO_REVAL_FORCED);
1828 cache_revalidated = false;
1829 }
1830
1831 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1832 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1833 } else if (server->caps & NFS_CAP_MTIME) {
1834 nfsi->cache_validity |= save_cache_validity &
1835 (NFS_INO_INVALID_MTIME
1836 | NFS_INO_REVAL_FORCED);
1837 cache_revalidated = false;
1838 }
1839
1840 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1841 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1842 } else if (server->caps & NFS_CAP_CTIME) {
1843 nfsi->cache_validity |= save_cache_validity &
1844 (NFS_INO_INVALID_CTIME
1845 | NFS_INO_REVAL_FORCED);
1846 cache_revalidated = false;
1847 }
1848
1849 /* Check if our cached file size is stale */
1850 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1851 new_isize = nfs_size_to_loff_t(fattr->size);
1852 cur_isize = i_size_read(inode);
1853 if (new_isize != cur_isize) {
1854 /* Do we perhaps have any outstanding writes, or has
1855 * the file grown beyond our last write? */
1856 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1857 i_size_write(inode, new_isize);
1858 if (!have_writers)
1859 invalid |= NFS_INO_INVALID_DATA;
1860 }
1861 dprintk("NFS: isize change on server for file %s/%ld "
1862 "(%Ld to %Ld)\n",
1863 inode->i_sb->s_id,
1864 inode->i_ino,
1865 (long long)cur_isize,
1866 (long long)new_isize);
1867 }
1868 } else {
1869 nfsi->cache_validity |= save_cache_validity &
1870 (NFS_INO_INVALID_SIZE
1871 | NFS_INO_REVAL_PAGECACHE
1872 | NFS_INO_REVAL_FORCED);
1873 cache_revalidated = false;
1874 }
1875
1876
1877 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1878 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1879 else if (server->caps & NFS_CAP_ATIME) {
1880 nfsi->cache_validity |= save_cache_validity &
1881 (NFS_INO_INVALID_ATIME
1882 | NFS_INO_REVAL_FORCED);
1883 cache_revalidated = false;
1884 }
1885
1886 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1887 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1888 umode_t newmode = inode->i_mode & S_IFMT;
1889 newmode |= fattr->mode & S_IALLUGO;
1890 inode->i_mode = newmode;
1891 invalid |= NFS_INO_INVALID_ACCESS
1892 | NFS_INO_INVALID_ACL
1893 | NFS_INO_INVALID_OTHER;
1894 }
1895 } else if (server->caps & NFS_CAP_MODE) {
1896 nfsi->cache_validity |= save_cache_validity &
1897 (NFS_INO_INVALID_ACCESS
1898 | NFS_INO_INVALID_ACL
1899 | NFS_INO_INVALID_OTHER
1900 | NFS_INO_REVAL_FORCED);
1901 cache_revalidated = false;
1902 }
1903
1904 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1905 if (!uid_eq(inode->i_uid, fattr->uid)) {
1906 invalid |= NFS_INO_INVALID_ACCESS
1907 | NFS_INO_INVALID_ACL
1908 | NFS_INO_INVALID_OTHER;
1909 inode->i_uid = fattr->uid;
1910 }
1911 } else if (server->caps & NFS_CAP_OWNER) {
1912 nfsi->cache_validity |= save_cache_validity &
1913 (NFS_INO_INVALID_ACCESS
1914 | NFS_INO_INVALID_ACL
1915 | NFS_INO_INVALID_OTHER
1916 | NFS_INO_REVAL_FORCED);
1917 cache_revalidated = false;
1918 }
1919
1920 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1921 if (!gid_eq(inode->i_gid, fattr->gid)) {
1922 invalid |= NFS_INO_INVALID_ACCESS
1923 | NFS_INO_INVALID_ACL
1924 | NFS_INO_INVALID_OTHER;
1925 inode->i_gid = fattr->gid;
1926 }
1927 } else if (server->caps & NFS_CAP_OWNER_GROUP) {
1928 nfsi->cache_validity |= save_cache_validity &
1929 (NFS_INO_INVALID_ACCESS
1930 | NFS_INO_INVALID_ACL
1931 | NFS_INO_INVALID_OTHER
1932 | NFS_INO_REVAL_FORCED);
1933 cache_revalidated = false;
1934 }
1935
1936 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1937 if (inode->i_nlink != fattr->nlink) {
1938 invalid |= NFS_INO_INVALID_OTHER;
1939 if (S_ISDIR(inode->i_mode))
1940 invalid |= NFS_INO_INVALID_DATA;
1941 set_nlink(inode, fattr->nlink);
1942 }
1943 } else if (server->caps & NFS_CAP_NLINK) {
1944 nfsi->cache_validity |= save_cache_validity &
1945 (NFS_INO_INVALID_OTHER
1946 | NFS_INO_REVAL_FORCED);
1947 cache_revalidated = false;
1948 }
1949
1950 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1951 /*
1952 * report the blocks in 512byte units
1953 */
1954 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1955 } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1956 inode->i_blocks = fattr->du.nfs2.blocks;
1957 else
1958 cache_revalidated = false;
1959
1960 /* Update attrtimeo value if we're out of the unstable period */
1961 if (invalid & NFS_INO_INVALID_ATTR) {
1962 invalid &= ~NFS_INO_INVALID_ATTR;
1963 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1964 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1965 nfsi->attrtimeo_timestamp = now;
1966 /* Set barrier to be more recent than all outstanding updates */
1967 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1968 } else {
1969 if (cache_revalidated) {
1970 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1971 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1972 nfsi->attrtimeo <<= 1;
1973 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1974 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1975 }
1976 nfsi->attrtimeo_timestamp = now;
1977 }
1978 /* Set the barrier to be more recent than this fattr */
1979 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1980 nfsi->attr_gencount = fattr->gencount;
1981 }
1982
1983 /* Don't invalidate the data if we were to blame */
1984 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1985 || S_ISLNK(inode->i_mode)))
1986 invalid &= ~NFS_INO_INVALID_DATA;
1987 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1988 (save_cache_validity & NFS_INO_REVAL_FORCED))
1989 nfs_set_cache_invalid(inode, invalid);
1990
1991 return 0;
1992 out_err:
1993 /*
1994 * No need to worry about unhashing the dentry, as the
1995 * lookup validation will know that the inode is bad.
1996 * (But we fall through to invalidate the caches.)
1997 */
1998 nfs_invalidate_inode(inode);
1999 return -ESTALE;
2000}
2001
2002struct inode *nfs_alloc_inode(struct super_block *sb)
2003{
2004 struct nfs_inode *nfsi;
2005 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2006 if (!nfsi)
2007 return NULL;
2008 nfsi->flags = 0UL;
2009 nfsi->cache_validity = 0UL;
2010#if IS_ENABLED(CONFIG_NFS_V4)
2011 nfsi->nfs4_acl = NULL;
2012#endif /* CONFIG_NFS_V4 */
2013 return &nfsi->vfs_inode;
2014}
2015EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2016
2017static void nfs_i_callback(struct rcu_head *head)
2018{
2019 struct inode *inode = container_of(head, struct inode, i_rcu);
2020 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2021}
2022
2023void nfs_destroy_inode(struct inode *inode)
2024{
2025 call_rcu(&inode->i_rcu, nfs_i_callback);
2026}
2027EXPORT_SYMBOL_GPL(nfs_destroy_inode);
2028
2029static inline void nfs4_init_once(struct nfs_inode *nfsi)
2030{
2031#if IS_ENABLED(CONFIG_NFS_V4)
2032 INIT_LIST_HEAD(&nfsi->open_states);
2033 nfsi->delegation = NULL;
2034 init_rwsem(&nfsi->rwsem);
2035 nfsi->layout = NULL;
2036#endif
2037}
2038
2039static void init_once(void *foo)
2040{
2041 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2042
2043 inode_init_once(&nfsi->vfs_inode);
2044 INIT_LIST_HEAD(&nfsi->open_files);
2045 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2046 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2047 INIT_LIST_HEAD(&nfsi->commit_info.list);
2048 atomic_long_set(&nfsi->nrequests, 0);
2049 atomic_long_set(&nfsi->commit_info.ncommit, 0);
2050 atomic_set(&nfsi->commit_info.rpcs_out, 0);
2051 init_rwsem(&nfsi->rmdir_sem);
2052 mutex_init(&nfsi->commit_mutex);
2053 nfs4_init_once(nfsi);
2054}
2055
2056static int __init nfs_init_inodecache(void)
2057{
2058 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2059 sizeof(struct nfs_inode),
2060 0, (SLAB_RECLAIM_ACCOUNT|
2061 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2062 init_once);
2063 if (nfs_inode_cachep == NULL)
2064 return -ENOMEM;
2065
2066 return 0;
2067}
2068
2069static void nfs_destroy_inodecache(void)
2070{
2071 /*
2072 * Make sure all delayed rcu free inodes are flushed before we
2073 * destroy cache.
2074 */
2075 rcu_barrier();
2076 kmem_cache_destroy(nfs_inode_cachep);
2077}
2078
2079struct workqueue_struct *nfsiod_workqueue;
2080EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2081
2082/*
2083 * start up the nfsiod workqueue
2084 */
2085static int nfsiod_start(void)
2086{
2087 struct workqueue_struct *wq;
2088 dprintk("RPC: creating workqueue nfsiod\n");
2089 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2090 if (wq == NULL)
2091 return -ENOMEM;
2092 nfsiod_workqueue = wq;
2093 return 0;
2094}
2095
2096/*
2097 * Destroy the nfsiod workqueue
2098 */
2099static void nfsiod_stop(void)
2100{
2101 struct workqueue_struct *wq;
2102
2103 wq = nfsiod_workqueue;
2104 if (wq == NULL)
2105 return;
2106 nfsiod_workqueue = NULL;
2107 destroy_workqueue(wq);
2108}
2109
2110unsigned int nfs_net_id;
2111EXPORT_SYMBOL_GPL(nfs_net_id);
2112
2113static int nfs_net_init(struct net *net)
2114{
2115 nfs_clients_init(net);
2116 return nfs_fs_proc_net_init(net);
2117}
2118
2119static void nfs_net_exit(struct net *net)
2120{
2121 struct nfs_net *nn = net_generic(net, nfs_net_id);
2122
2123 nfs_fs_proc_net_exit(net);
2124 nfs_cleanup_cb_ident_idr(net);
2125 WARN_ON_ONCE(!list_empty(&nn->nfs_client_list));
2126 WARN_ON_ONCE(!list_empty(&nn->nfs_volume_list));
2127}
2128
2129static struct pernet_operations nfs_net_ops = {
2130 .init = nfs_net_init,
2131 .exit = nfs_net_exit,
2132 .id = &nfs_net_id,
2133 .size = sizeof(struct nfs_net),
2134};
2135
2136/*
2137 * Initialize NFS
2138 */
2139static int __init init_nfs_fs(void)
2140{
2141 int err;
2142
2143 err = register_pernet_subsys(&nfs_net_ops);
2144 if (err < 0)
2145 goto out9;
2146
2147 err = nfs_fscache_register();
2148 if (err < 0)
2149 goto out8;
2150
2151 err = nfsiod_start();
2152 if (err)
2153 goto out7;
2154
2155 err = nfs_fs_proc_init();
2156 if (err)
2157 goto out6;
2158
2159 err = nfs_init_nfspagecache();
2160 if (err)
2161 goto out5;
2162
2163 err = nfs_init_inodecache();
2164 if (err)
2165 goto out4;
2166
2167 err = nfs_init_readpagecache();
2168 if (err)
2169 goto out3;
2170
2171 err = nfs_init_writepagecache();
2172 if (err)
2173 goto out2;
2174
2175 err = nfs_init_directcache();
2176 if (err)
2177 goto out1;
2178
2179 rpc_proc_register(&init_net, &nfs_rpcstat);
2180
2181 err = register_nfs_fs();
2182 if (err)
2183 goto out0;
2184
2185 return 0;
2186out0:
2187 rpc_proc_unregister(&init_net, "nfs");
2188 nfs_destroy_directcache();
2189out1:
2190 nfs_destroy_writepagecache();
2191out2:
2192 nfs_destroy_readpagecache();
2193out3:
2194 nfs_destroy_inodecache();
2195out4:
2196 nfs_destroy_nfspagecache();
2197out5:
2198 nfs_fs_proc_exit();
2199out6:
2200 nfsiod_stop();
2201out7:
2202 nfs_fscache_unregister();
2203out8:
2204 unregister_pernet_subsys(&nfs_net_ops);
2205out9:
2206 return err;
2207}
2208
2209static void __exit exit_nfs_fs(void)
2210{
2211 nfs_destroy_directcache();
2212 nfs_destroy_writepagecache();
2213 nfs_destroy_readpagecache();
2214 nfs_destroy_inodecache();
2215 nfs_destroy_nfspagecache();
2216 nfs_fscache_unregister();
2217 unregister_pernet_subsys(&nfs_net_ops);
2218 rpc_proc_unregister(&init_net, "nfs");
2219 unregister_nfs_fs();
2220 nfs_fs_proc_exit();
2221 nfsiod_stop();
2222}
2223
2224/* Not quite true; I just maintain it */
2225MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2226MODULE_LICENSE("GPL");
2227module_param(enable_ino64, bool, 0644);
2228
2229module_init(init_nfs_fs)
2230module_exit(exit_nfs_fs)