blob: 87b862a5f598252c353850d4866eadf1db9cf0af [file] [log] [blame]
b.liue9582032025-04-17 19:18:16 +08001// 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 void nfs_invalidate_inode(struct inode *);
66static int nfs_update_inode(struct inode *, struct nfs_fattr *);
67
68static struct kmem_cache * nfs_inode_cachep;
69
70static inline unsigned long
71nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
72{
73 return nfs_fileid_to_ino_t(fattr->fileid);
74}
75
76static int nfs_wait_killable(int mode)
77{
78 freezable_schedule_unsafe();
79 if (signal_pending_state(mode, current))
80 return -ERESTARTSYS;
81 return 0;
82}
83
84int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
85{
86 return nfs_wait_killable(mode);
87}
88EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
89
90/**
91 * nfs_compat_user_ino64 - returns the user-visible inode number
92 * @fileid: 64-bit fileid
93 *
94 * This function returns a 32-bit inode number if the boot parameter
95 * nfs.enable_ino64 is zero.
96 */
97u64 nfs_compat_user_ino64(u64 fileid)
98{
99#ifdef CONFIG_COMPAT
100 compat_ulong_t ino;
101#else
102 unsigned long ino;
103#endif
104
105 if (enable_ino64)
106 return fileid;
107 ino = fileid;
108 if (sizeof(ino) < sizeof(fileid))
109 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
110 return ino;
111}
112
113int nfs_drop_inode(struct inode *inode)
114{
115 return NFS_STALE(inode) || generic_drop_inode(inode);
116}
117EXPORT_SYMBOL_GPL(nfs_drop_inode);
118
119void nfs_clear_inode(struct inode *inode)
120{
121 /*
122 * The following should never happen...
123 */
124 WARN_ON_ONCE(nfs_have_writebacks(inode));
125 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
126 nfs_zap_acl_cache(inode);
127 nfs_access_zap_cache(inode);
128 nfs_fscache_clear_inode(inode);
129}
130EXPORT_SYMBOL_GPL(nfs_clear_inode);
131
132void nfs_evict_inode(struct inode *inode)
133{
134 truncate_inode_pages_final(&inode->i_data);
135 clear_inode(inode);
136 nfs_clear_inode(inode);
137}
138
139int nfs_sync_inode(struct inode *inode)
140{
141 inode_dio_wait(inode);
142 return nfs_wb_all(inode);
143}
144EXPORT_SYMBOL_GPL(nfs_sync_inode);
145
146/**
147 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
148 * @mapping: pointer to struct address_space
149 */
150int nfs_sync_mapping(struct address_space *mapping)
151{
152 int ret = 0;
153
154 if (mapping->nrpages != 0) {
155 unmap_mapping_range(mapping, 0, 0, 0);
156 ret = nfs_wb_all(mapping->host);
157 }
158 return ret;
159}
160
161static int nfs_attribute_timeout(struct inode *inode)
162{
163 struct nfs_inode *nfsi = NFS_I(inode);
164
165 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
166}
167
168static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
169{
170 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
171
172 /* Special case for the pagecache or access cache */
173 if (flags == NFS_INO_REVAL_PAGECACHE &&
174 !(cache_validity & NFS_INO_REVAL_FORCED))
175 return false;
176 return (cache_validity & flags) != 0;
177}
178
179static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
180{
181 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
182
183 if ((cache_validity & flags) != 0)
184 return true;
185 if (nfs_attribute_timeout(inode))
186 return true;
187 return false;
188}
189
190bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
191{
192 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
193 return nfs_check_cache_invalid_delegated(inode, flags);
194
195 return nfs_check_cache_invalid_not_delegated(inode, flags);
196}
197
198static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
199{
200 struct nfs_inode *nfsi = NFS_I(inode);
201 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
202
203 if (have_delegation) {
204 if (!(flags & NFS_INO_REVAL_FORCED))
205 flags &= ~NFS_INO_INVALID_OTHER;
206 flags &= ~(NFS_INO_INVALID_CHANGE
207 | NFS_INO_INVALID_SIZE
208 | NFS_INO_REVAL_PAGECACHE);
209 }
210
211 if (inode->i_mapping->nrpages == 0)
212 flags &= ~(NFS_INO_INVALID_DATA|NFS_INO_DATA_INVAL_DEFER);
213 nfsi->cache_validity |= flags;
214 if (flags & NFS_INO_INVALID_DATA)
215 nfs_fscache_invalidate(inode);
216}
217
218/*
219 * Invalidate the local caches
220 */
221static void nfs_zap_caches_locked(struct inode *inode)
222{
223 struct nfs_inode *nfsi = NFS_I(inode);
224 int mode = inode->i_mode;
225
226 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
227
228 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
229 nfsi->attrtimeo_timestamp = jiffies;
230
231 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
232 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
233 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
234 | NFS_INO_INVALID_DATA
235 | NFS_INO_INVALID_ACCESS
236 | NFS_INO_INVALID_ACL
237 | NFS_INO_REVAL_PAGECACHE);
238 } else
239 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
240 | NFS_INO_INVALID_ACCESS
241 | NFS_INO_INVALID_ACL
242 | NFS_INO_REVAL_PAGECACHE);
243 nfs_zap_label_cache_locked(nfsi);
244}
245
246void nfs_zap_caches(struct inode *inode)
247{
248 spin_lock(&inode->i_lock);
249 nfs_zap_caches_locked(inode);
250 spin_unlock(&inode->i_lock);
251}
252
253void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
254{
255 if (mapping->nrpages != 0) {
256 spin_lock(&inode->i_lock);
257 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
258 spin_unlock(&inode->i_lock);
259 }
260}
261
262void nfs_zap_acl_cache(struct inode *inode)
263{
264 void (*clear_acl_cache)(struct inode *);
265
266 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
267 if (clear_acl_cache != NULL)
268 clear_acl_cache(inode);
269 spin_lock(&inode->i_lock);
270 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
271 spin_unlock(&inode->i_lock);
272}
273EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
274
275void nfs_invalidate_atime(struct inode *inode)
276{
277 spin_lock(&inode->i_lock);
278 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
279 spin_unlock(&inode->i_lock);
280}
281EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
282
283/*
284 * Invalidate, but do not unhash, the inode.
285 * NB: must be called with inode->i_lock held!
286 */
287static void nfs_invalidate_inode(struct inode *inode)
288{
289 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
290 nfs_zap_caches_locked(inode);
291}
292
293struct nfs_find_desc {
294 struct nfs_fh *fh;
295 struct nfs_fattr *fattr;
296};
297
298/*
299 * In NFSv3 we can have 64bit inode numbers. In order to support
300 * this, and re-exported directories (also seen in NFSv2)
301 * we are forced to allow 2 different inodes to have the same
302 * i_ino.
303 */
304static int
305nfs_find_actor(struct inode *inode, void *opaque)
306{
307 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
308 struct nfs_fh *fh = desc->fh;
309 struct nfs_fattr *fattr = desc->fattr;
310
311 if (NFS_FILEID(inode) != fattr->fileid)
312 return 0;
313 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
314 return 0;
315 if (nfs_compare_fh(NFS_FH(inode), fh))
316 return 0;
317 if (is_bad_inode(inode) || NFS_STALE(inode))
318 return 0;
319 return 1;
320}
321
322static int
323nfs_init_locked(struct inode *inode, void *opaque)
324{
325 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
326 struct nfs_fattr *fattr = desc->fattr;
327
328 set_nfs_fileid(inode, fattr->fileid);
329 inode->i_mode = fattr->mode;
330 nfs_copy_fh(NFS_FH(inode), desc->fh);
331 return 0;
332}
333
334#ifdef CONFIG_NFS_V4_SECURITY_LABEL
335static void nfs_clear_label_invalid(struct inode *inode)
336{
337 spin_lock(&inode->i_lock);
338 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
339 spin_unlock(&inode->i_lock);
340}
341
342void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
343 struct nfs4_label *label)
344{
345 int error;
346
347 if (label == NULL)
348 return;
349
350 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
351 error = security_inode_notifysecctx(inode, label->label,
352 label->len);
353 if (error)
354 printk(KERN_ERR "%s() %s %d "
355 "security_inode_notifysecctx() %d\n",
356 __func__,
357 (char *)label->label,
358 label->len, error);
359 nfs_clear_label_invalid(inode);
360 }
361}
362
363struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
364{
365 struct nfs4_label *label = NULL;
366 int minor_version = server->nfs_client->cl_minorversion;
367
368 if (minor_version < 2)
369 return label;
370
371 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
372 return label;
373
374 label = kzalloc(sizeof(struct nfs4_label), flags);
375 if (label == NULL)
376 return ERR_PTR(-ENOMEM);
377
378 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
379 if (label->label == NULL) {
380 kfree(label);
381 return ERR_PTR(-ENOMEM);
382 }
383 label->len = NFS4_MAXLABELLEN;
384
385 return label;
386}
387EXPORT_SYMBOL_GPL(nfs4_label_alloc);
388#else
389void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
390 struct nfs4_label *label)
391{
392}
393#endif
394EXPORT_SYMBOL_GPL(nfs_setsecurity);
395
396/* Search for inode identified by fh, fileid and i_mode in inode cache. */
397struct inode *
398nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
399{
400 struct nfs_find_desc desc = {
401 .fh = fh,
402 .fattr = fattr,
403 };
404 struct inode *inode;
405 unsigned long hash;
406
407 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
408 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
409 return NULL;
410
411 hash = nfs_fattr_to_ino_t(fattr);
412 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
413
414 dprintk("%s: returning %p\n", __func__, inode);
415 return inode;
416}
417
418/*
419 * This is our front-end to iget that looks up inodes by file handle
420 * instead of inode number.
421 */
422struct inode *
423nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
424{
425 struct nfs_find_desc desc = {
426 .fh = fh,
427 .fattr = fattr
428 };
429 struct inode *inode = ERR_PTR(-ENOENT);
430 unsigned long hash;
431
432 nfs_attr_check_mountpoint(sb, fattr);
433
434 if (nfs_attr_use_mounted_on_fileid(fattr))
435 fattr->fileid = fattr->mounted_on_fileid;
436 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
437 goto out_no_inode;
438 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
439 goto out_no_inode;
440
441 hash = nfs_fattr_to_ino_t(fattr);
442
443 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
444 if (inode == NULL) {
445 inode = ERR_PTR(-ENOMEM);
446 goto out_no_inode;
447 }
448
449 if (inode->i_state & I_NEW) {
450 struct nfs_inode *nfsi = NFS_I(inode);
451 unsigned long now = jiffies;
452
453 /* We set i_ino for the few things that still rely on it,
454 * such as stat(2) */
455 inode->i_ino = hash;
456
457 /* We can't support update_atime(), since the server will reset it */
458 inode->i_flags |= S_NOATIME|S_NOCMTIME;
459 inode->i_mode = fattr->mode;
460 nfsi->cache_validity = 0;
461 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
462 && nfs_server_capable(inode, NFS_CAP_MODE))
463 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
464 /* Why so? Because we want revalidate for devices/FIFOs, and
465 * that's precisely what we have in nfs_file_inode_operations.
466 */
467 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
468 if (S_ISREG(inode->i_mode)) {
469 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
470 inode->i_data.a_ops = &nfs_file_aops;
471 } else if (S_ISDIR(inode->i_mode)) {
472 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
473 inode->i_fop = &nfs_dir_operations;
474 inode->i_data.a_ops = &nfs_dir_aops;
475 /* Deal with crossing mountpoints */
476 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
477 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
478 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
479 inode->i_op = &nfs_referral_inode_operations;
480 else
481 inode->i_op = &nfs_mountpoint_inode_operations;
482 inode->i_fop = NULL;
483 inode->i_flags |= S_AUTOMOUNT;
484 }
485 } else if (S_ISLNK(inode->i_mode)) {
486 inode->i_op = &nfs_symlink_inode_operations;
487 inode_nohighmem(inode);
488 } else
489 init_special_inode(inode, inode->i_mode, fattr->rdev);
490
491 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
492 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
493 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
494 inode_set_iversion_raw(inode, 0);
495 inode->i_size = 0;
496 clear_nlink(inode);
497 inode->i_uid = make_kuid(&init_user_ns, -2);
498 inode->i_gid = make_kgid(&init_user_ns, -2);
499 inode->i_blocks = 0;
500 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
501 nfsi->write_io = 0;
502 nfsi->read_io = 0;
503
504 nfsi->read_cache_jiffies = fattr->time_start;
505 nfsi->attr_gencount = fattr->gencount;
506 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
507 inode->i_atime = fattr->atime;
508 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
509 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
510 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
511 inode->i_mtime = fattr->mtime;
512 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
513 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
514 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
515 inode->i_ctime = fattr->ctime;
516 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
517 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
518 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
519 inode_set_iversion_raw(inode, fattr->change_attr);
520 else
521 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
522 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
523 inode->i_size = nfs_size_to_loff_t(fattr->size);
524 else
525 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
526 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
527 set_nlink(inode, fattr->nlink);
528 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
529 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
530 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
531 inode->i_uid = fattr->uid;
532 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
533 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
534 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
535 inode->i_gid = fattr->gid;
536 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
537 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
538 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
539 inode->i_blocks = fattr->du.nfs2.blocks;
540 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
541 /*
542 * report the blocks in 512byte units
543 */
544 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
545 }
546
547 if (nfsi->cache_validity != 0)
548 nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
549
550 nfs_setsecurity(inode, fattr, label);
551
552 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
553 nfsi->attrtimeo_timestamp = now;
554 nfsi->access_cache = RB_ROOT;
555
556 nfs_fscache_init_inode(inode);
557
558 unlock_new_inode(inode);
559 } else {
560 int err = nfs_refresh_inode(inode, fattr);
561 if (err < 0) {
562 iput(inode);
563 inode = ERR_PTR(err);
564 goto out_no_inode;
565 }
566 }
567 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
568 inode->i_sb->s_id,
569 (unsigned long long)NFS_FILEID(inode),
570 nfs_display_fhandle_hash(fh),
571 atomic_read(&inode->i_count));
572
573out:
574 return inode;
575
576out_no_inode:
577 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
578 goto out;
579}
580EXPORT_SYMBOL_GPL(nfs_fhget);
581
582#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)
583
584int
585nfs_setattr(struct dentry *dentry, struct iattr *attr)
586{
587 struct inode *inode = d_inode(dentry);
588 struct nfs_fattr *fattr;
589 int error = 0;
590
591 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
592
593 /* skip mode change if it's just for clearing setuid/setgid */
594 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
595 attr->ia_valid &= ~ATTR_MODE;
596
597 if (attr->ia_valid & ATTR_SIZE) {
598 BUG_ON(!S_ISREG(inode->i_mode));
599
600 error = inode_newsize_ok(inode, attr->ia_size);
601 if (error)
602 return error;
603
604 if (attr->ia_size == i_size_read(inode))
605 attr->ia_valid &= ~ATTR_SIZE;
606 }
607
608 /* Optimization: if the end result is no change, don't RPC */
609 attr->ia_valid &= NFS_VALID_ATTRS;
610 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
611 return 0;
612
613 trace_nfs_setattr_enter(inode);
614
615 /* Write all dirty data */
616 if (S_ISREG(inode->i_mode))
617 nfs_sync_inode(inode);
618
619 fattr = nfs_alloc_fattr();
620 if (fattr == NULL) {
621 error = -ENOMEM;
622 goto out;
623 }
624
625 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
626 if (error == 0)
627 error = nfs_refresh_inode(inode, fattr);
628 nfs_free_fattr(fattr);
629out:
630 trace_nfs_setattr_exit(inode, error);
631 return error;
632}
633EXPORT_SYMBOL_GPL(nfs_setattr);
634
635/**
636 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
637 * @inode: inode of the file used
638 * @offset: file offset to start truncating
639 *
640 * This is a copy of the common vmtruncate, but with the locking
641 * corrected to take into account the fact that NFS requires
642 * inode->i_size to be updated under the inode->i_lock.
643 * Note: must be called with inode->i_lock held!
644 */
645static int nfs_vmtruncate(struct inode * inode, loff_t offset)
646{
647 int err;
648
649 err = inode_newsize_ok(inode, offset);
650 if (err)
651 goto out;
652
653 i_size_write(inode, offset);
654 /* Optimisation */
655 if (offset == 0)
656 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
657 NFS_INO_DATA_INVAL_DEFER);
658 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
659
660 spin_unlock(&inode->i_lock);
661 truncate_pagecache(inode, offset);
662 spin_lock(&inode->i_lock);
663out:
664 return err;
665}
666
667/**
668 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
669 * @inode: pointer to struct inode
670 * @attr: pointer to struct iattr
671 * @fattr: pointer to struct nfs_fattr
672 *
673 * Note: we do this in the *proc.c in order to ensure that
674 * it works for things like exclusive creates too.
675 */
676void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
677 struct nfs_fattr *fattr)
678{
679 /* Barrier: bump the attribute generation count. */
680 nfs_fattr_set_barrier(fattr);
681
682 spin_lock(&inode->i_lock);
683 NFS_I(inode)->attr_gencount = fattr->gencount;
684 if ((attr->ia_valid & ATTR_SIZE) != 0) {
685 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
686 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
687 nfs_vmtruncate(inode, attr->ia_size);
688 }
689 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
690 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
691 if ((attr->ia_valid & ATTR_MODE) != 0) {
692 int mode = attr->ia_mode & S_IALLUGO;
693 mode |= inode->i_mode & ~S_IALLUGO;
694 inode->i_mode = mode;
695 }
696 if ((attr->ia_valid & ATTR_UID) != 0)
697 inode->i_uid = attr->ia_uid;
698 if ((attr->ia_valid & ATTR_GID) != 0)
699 inode->i_gid = attr->ia_gid;
700 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
701 inode->i_ctime = fattr->ctime;
702 else
703 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
704 | NFS_INO_INVALID_CTIME);
705 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
706 | NFS_INO_INVALID_ACL);
707 }
708 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
709 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
710 | NFS_INO_INVALID_CTIME);
711 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
712 inode->i_atime = fattr->atime;
713 else if (attr->ia_valid & ATTR_ATIME_SET)
714 inode->i_atime = attr->ia_atime;
715 else
716 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
717
718 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
719 inode->i_ctime = fattr->ctime;
720 else
721 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
722 | NFS_INO_INVALID_CTIME);
723 }
724 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
725 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
726 | NFS_INO_INVALID_CTIME);
727 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
728 inode->i_mtime = fattr->mtime;
729 else if (attr->ia_valid & ATTR_MTIME_SET)
730 inode->i_mtime = attr->ia_mtime;
731 else
732 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
733
734 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
735 inode->i_ctime = fattr->ctime;
736 else
737 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
738 | NFS_INO_INVALID_CTIME);
739 }
740 if (fattr->valid)
741 nfs_update_inode(inode, fattr);
742 spin_unlock(&inode->i_lock);
743}
744EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
745
746static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
747{
748 struct dentry *parent;
749
750 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
751 return;
752 parent = dget_parent(dentry);
753 nfs_force_use_readdirplus(d_inode(parent));
754 dput(parent);
755}
756
757static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
758{
759 struct dentry *parent;
760
761 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
762 return;
763 parent = dget_parent(dentry);
764 nfs_advise_use_readdirplus(d_inode(parent));
765 dput(parent);
766}
767
768static bool nfs_need_revalidate_inode(struct inode *inode)
769{
770 if (NFS_I(inode)->cache_validity &
771 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
772 return true;
773 if (nfs_attribute_cache_expired(inode))
774 return true;
775 return false;
776}
777
778int nfs_getattr(const struct path *path, struct kstat *stat,
779 u32 request_mask, unsigned int query_flags)
780{
781 struct inode *inode = d_inode(path->dentry);
782 struct nfs_server *server = NFS_SERVER(inode);
783 unsigned long cache_validity;
784 int err = 0;
785 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
786 bool do_update = false;
787
788 trace_nfs_getattr_enter(inode);
789
790 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync)
791 goto out_no_update;
792
793 /* Flush out writes to the server in order to update c/mtime. */
794 if ((request_mask & (STATX_CTIME | STATX_MTIME)) &&
795 S_ISREG(inode->i_mode))
796 filemap_write_and_wait(inode->i_mapping);
797
798 /*
799 * We may force a getattr if the user cares about atime.
800 *
801 * Note that we only have to check the vfsmount flags here:
802 * - NFS always sets S_NOATIME by so checking it would give a
803 * bogus result
804 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
805 * no point in checking those.
806 */
807 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
808 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
809 request_mask &= ~STATX_ATIME;
810
811 /* Is the user requesting attributes that might need revalidation? */
812 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
813 STATX_MTIME|STATX_UID|STATX_GID|
814 STATX_SIZE|STATX_BLOCKS)))
815 goto out_no_revalidate;
816
817 /* Check whether the cached attributes are stale */
818 do_update |= force_sync || nfs_attribute_cache_expired(inode);
819 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
820 do_update |= cache_validity &
821 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
822 if (request_mask & STATX_ATIME)
823 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
824 if (request_mask & (STATX_CTIME|STATX_MTIME))
825 do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
826 if (request_mask & STATX_BLOCKS)
827 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
828 if (do_update) {
829 /* Update the attribute cache */
830 if (!(server->flags & NFS_MOUNT_NOAC))
831 nfs_readdirplus_parent_cache_miss(path->dentry);
832 else
833 nfs_readdirplus_parent_cache_hit(path->dentry);
834 err = __nfs_revalidate_inode(server, inode);
835 if (err)
836 goto out;
837 } else
838 nfs_readdirplus_parent_cache_hit(path->dentry);
839out_no_revalidate:
840 /* Only return attributes that were revalidated. */
841 stat->result_mask &= request_mask;
842out_no_update:
843 generic_fillattr(inode, stat);
844 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
845 if (S_ISDIR(inode->i_mode))
846 stat->blksize = NFS_SERVER(inode)->dtsize;
847out:
848 trace_nfs_getattr_exit(inode, err);
849 return err;
850}
851EXPORT_SYMBOL_GPL(nfs_getattr);
852
853static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
854{
855 refcount_set(&l_ctx->count, 1);
856 l_ctx->lockowner = current->files;
857 INIT_LIST_HEAD(&l_ctx->list);
858 atomic_set(&l_ctx->io_count, 0);
859}
860
861static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
862{
863 struct nfs_lock_context *pos;
864
865 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
866 if (pos->lockowner != current->files)
867 continue;
868 if (refcount_inc_not_zero(&pos->count))
869 return pos;
870 }
871 return NULL;
872}
873
874struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
875{
876 struct nfs_lock_context *res, *new = NULL;
877 struct inode *inode = d_inode(ctx->dentry);
878
879 rcu_read_lock();
880 res = __nfs_find_lock_context(ctx);
881 rcu_read_unlock();
882 if (res == NULL) {
883 new = kmalloc(sizeof(*new), GFP_KERNEL);
884 if (new == NULL)
885 return ERR_PTR(-ENOMEM);
886 nfs_init_lock_context(new);
887 spin_lock(&inode->i_lock);
888 res = __nfs_find_lock_context(ctx);
889 if (res == NULL) {
890 new->open_context = get_nfs_open_context(ctx);
891 if (new->open_context) {
892 list_add_tail_rcu(&new->list,
893 &ctx->lock_context.list);
894 res = new;
895 new = NULL;
896 } else
897 res = ERR_PTR(-EBADF);
898 }
899 spin_unlock(&inode->i_lock);
900 kfree(new);
901 }
902 return res;
903}
904EXPORT_SYMBOL_GPL(nfs_get_lock_context);
905
906void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
907{
908 struct nfs_open_context *ctx = l_ctx->open_context;
909 struct inode *inode = d_inode(ctx->dentry);
910
911 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
912 return;
913 list_del_rcu(&l_ctx->list);
914 spin_unlock(&inode->i_lock);
915 put_nfs_open_context(ctx);
916 kfree_rcu(l_ctx, rcu_head);
917}
918EXPORT_SYMBOL_GPL(nfs_put_lock_context);
919
920/**
921 * nfs_close_context - Common close_context() routine NFSv2/v3
922 * @ctx: pointer to context
923 * @is_sync: is this a synchronous close
924 *
925 * Ensure that the attributes are up to date if we're mounted
926 * with close-to-open semantics and we have cached data that will
927 * need to be revalidated on open.
928 */
929void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
930{
931 struct nfs_inode *nfsi;
932 struct inode *inode;
933 struct nfs_server *server;
934
935 if (!(ctx->mode & FMODE_WRITE))
936 return;
937 if (!is_sync)
938 return;
939 inode = d_inode(ctx->dentry);
940 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
941 return;
942 nfsi = NFS_I(inode);
943 if (inode->i_mapping->nrpages == 0)
944 return;
945 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
946 return;
947 if (!list_empty(&nfsi->open_files))
948 return;
949 server = NFS_SERVER(inode);
950 if (server->flags & NFS_MOUNT_NOCTO)
951 return;
952 nfs_revalidate_inode(server, inode);
953}
954EXPORT_SYMBOL_GPL(nfs_close_context);
955
956struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
957 fmode_t f_mode,
958 struct file *filp)
959{
960 struct nfs_open_context *ctx;
961
962 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
963 if (!ctx)
964 return ERR_PTR(-ENOMEM);
965 nfs_sb_active(dentry->d_sb);
966 ctx->dentry = dget(dentry);
967 if (filp)
968 ctx->cred = get_cred(filp->f_cred);
969 else
970 ctx->cred = get_current_cred();
971 ctx->ll_cred = NULL;
972 ctx->state = NULL;
973 ctx->mode = f_mode;
974 ctx->flags = 0;
975 ctx->error = 0;
976 ctx->flock_owner = (fl_owner_t)filp;
977 nfs_init_lock_context(&ctx->lock_context);
978 ctx->lock_context.open_context = ctx;
979 INIT_LIST_HEAD(&ctx->list);
980 ctx->mdsthreshold = NULL;
981 return ctx;
982}
983EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
984
985struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
986{
987 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
988 return ctx;
989 return NULL;
990}
991EXPORT_SYMBOL_GPL(get_nfs_open_context);
992
993static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
994{
995 struct inode *inode = d_inode(ctx->dentry);
996 struct super_block *sb = ctx->dentry->d_sb;
997
998 if (!refcount_dec_and_test(&ctx->lock_context.count))
999 return;
1000 if (!list_empty(&ctx->list)) {
1001 spin_lock(&inode->i_lock);
1002 list_del_rcu(&ctx->list);
1003 spin_unlock(&inode->i_lock);
1004 }
1005 if (inode != NULL)
1006 NFS_PROTO(inode)->close_context(ctx, is_sync);
1007 put_cred(ctx->cred);
1008 dput(ctx->dentry);
1009 nfs_sb_deactive(sb);
1010 put_rpccred(ctx->ll_cred);
1011 kfree(ctx->mdsthreshold);
1012 kfree_rcu(ctx, rcu_head);
1013}
1014
1015void put_nfs_open_context(struct nfs_open_context *ctx)
1016{
1017 __put_nfs_open_context(ctx, 0);
1018}
1019EXPORT_SYMBOL_GPL(put_nfs_open_context);
1020
1021static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1022{
1023 __put_nfs_open_context(ctx, 1);
1024}
1025
1026/*
1027 * Ensure that mmap has a recent RPC credential for use when writing out
1028 * shared pages
1029 */
1030void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1031{
1032 struct inode *inode = d_inode(ctx->dentry);
1033 struct nfs_inode *nfsi = NFS_I(inode);
1034
1035 spin_lock(&inode->i_lock);
1036 if (list_empty(&nfsi->open_files) &&
1037 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1038 nfsi->cache_validity |= NFS_INO_INVALID_DATA |
1039 NFS_INO_REVAL_FORCED;
1040 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1041 spin_unlock(&inode->i_lock);
1042}
1043EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1044
1045void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1046{
1047 filp->private_data = get_nfs_open_context(ctx);
1048 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1049 if (list_empty(&ctx->list))
1050 nfs_inode_attach_open_context(ctx);
1051}
1052EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1053
1054/*
1055 * Given an inode, search for an open context with the desired characteristics
1056 */
1057struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1058{
1059 struct nfs_inode *nfsi = NFS_I(inode);
1060 struct nfs_open_context *pos, *ctx = NULL;
1061
1062 rcu_read_lock();
1063 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1064 if (cred != NULL && pos->cred != cred)
1065 continue;
1066 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1067 continue;
1068 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1069 continue;
1070 ctx = get_nfs_open_context(pos);
1071 if (ctx)
1072 break;
1073 }
1074 rcu_read_unlock();
1075 return ctx;
1076}
1077
1078void nfs_file_clear_open_context(struct file *filp)
1079{
1080 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1081
1082 if (ctx) {
1083 struct inode *inode = d_inode(ctx->dentry);
1084
1085 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1086 /*
1087 * We fatal error on write before. Try to writeback
1088 * every page again.
1089 */
1090 if (ctx->error < 0)
1091 invalidate_inode_pages2(inode->i_mapping);
1092 filp->private_data = NULL;
1093 put_nfs_open_context_sync(ctx);
1094 }
1095}
1096
1097/*
1098 * These allocate and release file read/write context information.
1099 */
1100int nfs_open(struct inode *inode, struct file *filp)
1101{
1102 struct nfs_open_context *ctx;
1103
1104 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1105 if (IS_ERR(ctx))
1106 return PTR_ERR(ctx);
1107 nfs_file_set_open_context(filp, ctx);
1108 put_nfs_open_context(ctx);
1109 nfs_fscache_open_file(inode, filp);
1110 return 0;
1111}
1112EXPORT_SYMBOL_GPL(nfs_open);
1113
1114/*
1115 * This function is called whenever some part of NFS notices that
1116 * the cached attributes have to be refreshed.
1117 */
1118int
1119__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1120{
1121 int status = -ESTALE;
1122 struct nfs4_label *label = NULL;
1123 struct nfs_fattr *fattr = NULL;
1124 struct nfs_inode *nfsi = NFS_I(inode);
1125
1126 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1127 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1128
1129 trace_nfs_revalidate_inode_enter(inode);
1130
1131 if (is_bad_inode(inode))
1132 goto out;
1133 if (NFS_STALE(inode))
1134 goto out;
1135
1136 /* pNFS: Attributes aren't updated until we layoutcommit */
1137 if (S_ISREG(inode->i_mode)) {
1138 status = pnfs_sync_inode(inode, false);
1139 if (status)
1140 goto out;
1141 }
1142
1143 status = -ENOMEM;
1144 fattr = nfs_alloc_fattr();
1145 if (fattr == NULL)
1146 goto out;
1147
1148 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1149
1150 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1151 if (IS_ERR(label)) {
1152 status = PTR_ERR(label);
1153 goto out;
1154 }
1155
1156 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1157 label, inode);
1158 if (status != 0) {
1159 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1160 inode->i_sb->s_id,
1161 (unsigned long long)NFS_FILEID(inode), status);
1162 if (status == -ESTALE) {
1163 nfs_zap_caches(inode);
1164 if (!S_ISDIR(inode->i_mode))
1165 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1166 }
1167 goto err_out;
1168 }
1169
1170 status = nfs_refresh_inode(inode, fattr);
1171 if (status) {
1172 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1173 inode->i_sb->s_id,
1174 (unsigned long long)NFS_FILEID(inode), status);
1175 goto err_out;
1176 }
1177
1178 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1179 nfs_zap_acl_cache(inode);
1180
1181 nfs_setsecurity(inode, fattr, label);
1182
1183 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1184 inode->i_sb->s_id,
1185 (unsigned long long)NFS_FILEID(inode));
1186
1187err_out:
1188 nfs4_label_free(label);
1189out:
1190 nfs_free_fattr(fattr);
1191 trace_nfs_revalidate_inode_exit(inode, status);
1192 return status;
1193}
1194
1195int nfs_attribute_cache_expired(struct inode *inode)
1196{
1197 if (nfs_have_delegated_attributes(inode))
1198 return 0;
1199 return nfs_attribute_timeout(inode);
1200}
1201
1202/**
1203 * nfs_revalidate_inode - Revalidate the inode attributes
1204 * @server: pointer to nfs_server struct
1205 * @inode: pointer to inode struct
1206 *
1207 * Updates inode attribute information by retrieving the data from the server.
1208 */
1209int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1210{
1211 if (!nfs_need_revalidate_inode(inode))
1212 return NFS_STALE(inode) ? -ESTALE : 0;
1213 return __nfs_revalidate_inode(server, inode);
1214}
1215EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1216
1217static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1218{
1219 struct nfs_inode *nfsi = NFS_I(inode);
1220 int ret;
1221
1222 if (mapping->nrpages != 0) {
1223 if (S_ISREG(inode->i_mode)) {
1224 ret = nfs_sync_mapping(mapping);
1225 if (ret < 0)
1226 return ret;
1227 }
1228 ret = invalidate_inode_pages2(mapping);
1229 if (ret < 0)
1230 return ret;
1231 }
1232 if (S_ISDIR(inode->i_mode)) {
1233 spin_lock(&inode->i_lock);
1234 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1235 spin_unlock(&inode->i_lock);
1236 }
1237 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1238 nfs_fscache_wait_on_invalidate(inode);
1239
1240 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1241 inode->i_sb->s_id,
1242 (unsigned long long)NFS_FILEID(inode));
1243 return 0;
1244}
1245
1246bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1247{
1248 return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1249 NFS_STALE(inode);
1250}
1251
1252int nfs_revalidate_mapping_rcu(struct inode *inode)
1253{
1254 struct nfs_inode *nfsi = NFS_I(inode);
1255 unsigned long *bitlock = &nfsi->flags;
1256 int ret = 0;
1257
1258 if (IS_SWAPFILE(inode))
1259 goto out;
1260 if (nfs_mapping_need_revalidate_inode(inode)) {
1261 ret = -ECHILD;
1262 goto out;
1263 }
1264 spin_lock(&inode->i_lock);
1265 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1266 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1267 ret = -ECHILD;
1268 spin_unlock(&inode->i_lock);
1269out:
1270 return ret;
1271}
1272
1273/**
1274 * nfs_revalidate_mapping - Revalidate the pagecache
1275 * @inode: pointer to host inode
1276 * @mapping: pointer to mapping
1277 */
1278int nfs_revalidate_mapping(struct inode *inode,
1279 struct address_space *mapping)
1280{
1281 struct nfs_inode *nfsi = NFS_I(inode);
1282 unsigned long *bitlock = &nfsi->flags;
1283 int ret = 0;
1284
1285 /* swapfiles are not supposed to be shared. */
1286 if (IS_SWAPFILE(inode))
1287 goto out;
1288
1289 if (nfs_mapping_need_revalidate_inode(inode)) {
1290 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1291 if (ret < 0)
1292 goto out;
1293 }
1294
1295 /*
1296 * We must clear NFS_INO_INVALID_DATA first to ensure that
1297 * invalidations that come in while we're shooting down the mappings
1298 * are respected. But, that leaves a race window where one revalidator
1299 * can clear the flag, and then another checks it before the mapping
1300 * gets invalidated. Fix that by serializing access to this part of
1301 * the function.
1302 *
1303 * At the same time, we need to allow other tasks to see whether we
1304 * might be in the middle of invalidating the pages, so we only set
1305 * the bit lock here if it looks like we're going to be doing that.
1306 */
1307 for (;;) {
1308 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1309 nfs_wait_bit_killable, TASK_KILLABLE);
1310 if (ret)
1311 goto out;
1312 spin_lock(&inode->i_lock);
1313 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1314 spin_unlock(&inode->i_lock);
1315 continue;
1316 }
1317 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1318 break;
1319 spin_unlock(&inode->i_lock);
1320 goto out;
1321 }
1322
1323 set_bit(NFS_INO_INVALIDATING, bitlock);
1324 smp_wmb();
1325 nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA|
1326 NFS_INO_DATA_INVAL_DEFER);
1327 spin_unlock(&inode->i_lock);
1328 trace_nfs_invalidate_mapping_enter(inode);
1329 ret = nfs_invalidate_mapping(inode, mapping);
1330 trace_nfs_invalidate_mapping_exit(inode, ret);
1331
1332 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1333 smp_mb__after_atomic();
1334 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1335out:
1336 return ret;
1337}
1338
1339static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1340{
1341 struct inode *inode = &nfsi->vfs_inode;
1342
1343 if (!S_ISREG(inode->i_mode))
1344 return false;
1345 if (list_empty(&nfsi->open_files))
1346 return false;
1347 return inode_is_open_for_write(inode);
1348}
1349
1350static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1351{
1352 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1353}
1354
1355static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1356{
1357 struct timespec64 ts;
1358
1359 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1360 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1361 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1362 inode_set_iversion_raw(inode, fattr->change_attr);
1363 if (S_ISDIR(inode->i_mode))
1364 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1365 }
1366 /* If we have atomic WCC data, we may update some attributes */
1367 ts = inode->i_ctime;
1368 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1369 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1370 && timespec64_equal(&ts, &fattr->pre_ctime)) {
1371 inode->i_ctime = fattr->ctime;
1372 }
1373
1374 ts = inode->i_mtime;
1375 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1376 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1377 && timespec64_equal(&ts, &fattr->pre_mtime)) {
1378 inode->i_mtime = fattr->mtime;
1379 if (S_ISDIR(inode->i_mode))
1380 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1381 }
1382 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1383 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1384 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1385 && !nfs_have_writebacks(inode)) {
1386 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1387 }
1388}
1389
1390/**
1391 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1392 * @inode: pointer to inode
1393 * @fattr: updated attributes
1394 *
1395 * Verifies the attribute cache. If we have just changed the attributes,
1396 * so that fattr carries weak cache consistency data, then it may
1397 * also update the ctime/mtime/change_attribute.
1398 */
1399static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1400{
1401 struct nfs_inode *nfsi = NFS_I(inode);
1402 loff_t cur_size, new_isize;
1403 unsigned long invalid = 0;
1404 struct timespec64 ts;
1405
1406 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1407 return 0;
1408
1409 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1410 /* Only a mounted-on-fileid? Just exit */
1411 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1412 return 0;
1413 /* Has the inode gone and changed behind our back? */
1414 } else if (nfsi->fileid != fattr->fileid) {
1415 /* Is this perhaps the mounted-on fileid? */
1416 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1417 nfsi->fileid == fattr->mounted_on_fileid)
1418 return 0;
1419 return -ESTALE;
1420 }
1421 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1422 return -ESTALE;
1423
1424
1425 if (!nfs_file_has_buffered_writers(nfsi)) {
1426 /* Verify a few of the more important attributes */
1427 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1428 invalid |= NFS_INO_INVALID_CHANGE
1429 | NFS_INO_REVAL_PAGECACHE;
1430
1431 ts = inode->i_mtime;
1432 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1433 invalid |= NFS_INO_INVALID_MTIME;
1434
1435 ts = inode->i_ctime;
1436 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1437 invalid |= NFS_INO_INVALID_CTIME;
1438
1439 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1440 cur_size = i_size_read(inode);
1441 new_isize = nfs_size_to_loff_t(fattr->size);
1442 if (cur_size != new_isize)
1443 invalid |= NFS_INO_INVALID_SIZE
1444 | NFS_INO_REVAL_PAGECACHE;
1445 }
1446 }
1447
1448 /* Have any file permissions changed? */
1449 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1450 invalid |= NFS_INO_INVALID_ACCESS
1451 | NFS_INO_INVALID_ACL
1452 | NFS_INO_INVALID_OTHER;
1453 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1454 invalid |= NFS_INO_INVALID_ACCESS
1455 | NFS_INO_INVALID_ACL
1456 | NFS_INO_INVALID_OTHER;
1457 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1458 invalid |= NFS_INO_INVALID_ACCESS
1459 | NFS_INO_INVALID_ACL
1460 | NFS_INO_INVALID_OTHER;
1461
1462 /* Has the link count changed? */
1463 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1464 invalid |= NFS_INO_INVALID_OTHER;
1465
1466 ts = inode->i_atime;
1467 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1468 invalid |= NFS_INO_INVALID_ATIME;
1469
1470 if (invalid != 0)
1471 nfs_set_cache_invalid(inode, invalid);
1472
1473 nfsi->read_cache_jiffies = fattr->time_start;
1474 return 0;
1475}
1476
1477static atomic_long_t nfs_attr_generation_counter;
1478
1479static unsigned long nfs_read_attr_generation_counter(void)
1480{
1481 return atomic_long_read(&nfs_attr_generation_counter);
1482}
1483
1484unsigned long nfs_inc_attr_generation_counter(void)
1485{
1486 return atomic_long_inc_return(&nfs_attr_generation_counter);
1487}
1488EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1489
1490void nfs_fattr_init(struct nfs_fattr *fattr)
1491{
1492 fattr->valid = 0;
1493 fattr->time_start = jiffies;
1494 fattr->gencount = nfs_inc_attr_generation_counter();
1495 fattr->owner_name = NULL;
1496 fattr->group_name = NULL;
1497 fattr->mdsthreshold = NULL;
1498}
1499EXPORT_SYMBOL_GPL(nfs_fattr_init);
1500
1501/**
1502 * nfs_fattr_set_barrier
1503 * @fattr: attributes
1504 *
1505 * Used to set a barrier after an attribute was updated. This
1506 * barrier ensures that older attributes from RPC calls that may
1507 * have raced with our update cannot clobber these new values.
1508 * Note that you are still responsible for ensuring that other
1509 * operations which change the attribute on the server do not
1510 * collide.
1511 */
1512void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1513{
1514 fattr->gencount = nfs_inc_attr_generation_counter();
1515}
1516
1517struct nfs_fattr *nfs_alloc_fattr(void)
1518{
1519 struct nfs_fattr *fattr;
1520
1521 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1522 if (fattr != NULL)
1523 nfs_fattr_init(fattr);
1524 return fattr;
1525}
1526EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1527
1528struct nfs_fh *nfs_alloc_fhandle(void)
1529{
1530 struct nfs_fh *fh;
1531
1532 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1533 if (fh != NULL)
1534 fh->size = 0;
1535 return fh;
1536}
1537EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1538
1539#ifdef NFS_DEBUG
1540/*
1541 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1542 * in the same way that wireshark does
1543 *
1544 * @fh: file handle
1545 *
1546 * For debugging only.
1547 */
1548u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1549{
1550 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1551 * not on the result */
1552 return nfs_fhandle_hash(fh);
1553}
1554EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1555
1556/*
1557 * _nfs_display_fhandle - display an NFS file handle on the console
1558 *
1559 * @fh: file handle to display
1560 * @caption: display caption
1561 *
1562 * For debugging only.
1563 */
1564void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1565{
1566 unsigned short i;
1567
1568 if (fh == NULL || fh->size == 0) {
1569 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1570 return;
1571 }
1572
1573 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1574 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1575 for (i = 0; i < fh->size; i += 16) {
1576 __be32 *pos = (__be32 *)&fh->data[i];
1577
1578 switch ((fh->size - i - 1) >> 2) {
1579 case 0:
1580 printk(KERN_DEFAULT " %08x\n",
1581 be32_to_cpup(pos));
1582 break;
1583 case 1:
1584 printk(KERN_DEFAULT " %08x %08x\n",
1585 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1586 break;
1587 case 2:
1588 printk(KERN_DEFAULT " %08x %08x %08x\n",
1589 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1590 be32_to_cpup(pos + 2));
1591 break;
1592 default:
1593 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1594 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1595 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1596 }
1597 }
1598}
1599EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1600#endif
1601
1602/**
1603 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1604 * @inode: pointer to inode
1605 * @fattr: attributes
1606 *
1607 * Attempt to divine whether or not an RPC call reply carrying stale
1608 * attributes got scheduled after another call carrying updated ones.
1609 *
1610 * To do so, the function first assumes that a more recent ctime means
1611 * that the attributes in fattr are newer, however it also attempt to
1612 * catch the case where ctime either didn't change, or went backwards
1613 * (if someone reset the clock on the server) by looking at whether
1614 * or not this RPC call was started after the inode was last updated.
1615 * Note also the check for wraparound of 'attr_gencount'
1616 *
1617 * The function returns 'true' if it thinks the attributes in 'fattr' are
1618 * more recent than the ones cached in the inode.
1619 *
1620 */
1621static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1622{
1623 unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1624
1625 return (long)(fattr->gencount - attr_gencount) > 0 ||
1626 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1627}
1628
1629static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1630{
1631 int ret;
1632
1633 trace_nfs_refresh_inode_enter(inode);
1634
1635 if (nfs_inode_attrs_need_update(inode, fattr))
1636 ret = nfs_update_inode(inode, fattr);
1637 else
1638 ret = nfs_check_inode_attributes(inode, fattr);
1639
1640 trace_nfs_refresh_inode_exit(inode, ret);
1641 return ret;
1642}
1643
1644/**
1645 * nfs_refresh_inode - try to update the inode attribute cache
1646 * @inode: pointer to inode
1647 * @fattr: updated attributes
1648 *
1649 * Check that an RPC call that returned attributes has not overlapped with
1650 * other recent updates of the inode metadata, then decide whether it is
1651 * safe to do a full update of the inode attributes, or whether just to
1652 * call nfs_check_inode_attributes.
1653 */
1654int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1655{
1656 int status;
1657
1658 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1659 return 0;
1660 spin_lock(&inode->i_lock);
1661 status = nfs_refresh_inode_locked(inode, fattr);
1662 spin_unlock(&inode->i_lock);
1663
1664 return status;
1665}
1666EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1667
1668static int nfs_post_op_update_inode_locked(struct inode *inode,
1669 struct nfs_fattr *fattr, unsigned int invalid)
1670{
1671 if (S_ISDIR(inode->i_mode))
1672 invalid |= NFS_INO_INVALID_DATA;
1673 nfs_set_cache_invalid(inode, invalid);
1674 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1675 return 0;
1676 return nfs_refresh_inode_locked(inode, fattr);
1677}
1678
1679/**
1680 * nfs_post_op_update_inode - try to update the inode attribute cache
1681 * @inode: pointer to inode
1682 * @fattr: updated attributes
1683 *
1684 * After an operation that has changed the inode metadata, mark the
1685 * attribute cache as being invalid, then try to update it.
1686 *
1687 * NB: if the server didn't return any post op attributes, this
1688 * function will force the retrieval of attributes before the next
1689 * NFS request. Thus it should be used only for operations that
1690 * are expected to change one or more attributes, to avoid
1691 * unnecessary NFS requests and trips through nfs_update_inode().
1692 */
1693int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1694{
1695 int status;
1696
1697 spin_lock(&inode->i_lock);
1698 nfs_fattr_set_barrier(fattr);
1699 status = nfs_post_op_update_inode_locked(inode, fattr,
1700 NFS_INO_INVALID_CHANGE
1701 | NFS_INO_INVALID_CTIME
1702 | NFS_INO_REVAL_FORCED);
1703 spin_unlock(&inode->i_lock);
1704
1705 return status;
1706}
1707EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1708
1709/**
1710 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1711 * @inode: pointer to inode
1712 * @fattr: updated attributes
1713 *
1714 * After an operation that has changed the inode metadata, mark the
1715 * attribute cache as being invalid, then try to update it. Fake up
1716 * weak cache consistency data, if none exist.
1717 *
1718 * This function is mainly designed to be used by the ->write_done() functions.
1719 */
1720int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1721{
1722 int status;
1723
1724 /* Don't do a WCC update if these attributes are already stale */
1725 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1726 !nfs_inode_attrs_need_update(inode, fattr)) {
1727 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1728 | NFS_ATTR_FATTR_PRESIZE
1729 | NFS_ATTR_FATTR_PREMTIME
1730 | NFS_ATTR_FATTR_PRECTIME);
1731 goto out_noforce;
1732 }
1733 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1734 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1735 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1736 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1737 }
1738 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1739 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1740 fattr->pre_ctime = inode->i_ctime;
1741 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1742 }
1743 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1744 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1745 fattr->pre_mtime = inode->i_mtime;
1746 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1747 }
1748 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1749 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1750 fattr->pre_size = i_size_read(inode);
1751 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1752 }
1753out_noforce:
1754 status = nfs_post_op_update_inode_locked(inode, fattr,
1755 NFS_INO_INVALID_CHANGE
1756 | NFS_INO_INVALID_CTIME
1757 | NFS_INO_INVALID_MTIME
1758 | NFS_INO_INVALID_BLOCKS);
1759 return status;
1760}
1761
1762/**
1763 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1764 * @inode: pointer to inode
1765 * @fattr: updated attributes
1766 *
1767 * After an operation that has changed the inode metadata, mark the
1768 * attribute cache as being invalid, then try to update it. Fake up
1769 * weak cache consistency data, if none exist.
1770 *
1771 * This function is mainly designed to be used by the ->write_done() functions.
1772 */
1773int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1774{
1775 int status;
1776
1777 spin_lock(&inode->i_lock);
1778 nfs_fattr_set_barrier(fattr);
1779 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1780 spin_unlock(&inode->i_lock);
1781 return status;
1782}
1783EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1784
1785
1786/*
1787 * Many nfs protocol calls return the new file attributes after
1788 * an operation. Here we update the inode to reflect the state
1789 * of the server's inode.
1790 *
1791 * This is a bit tricky because we have to make sure all dirty pages
1792 * have been sent off to the server before calling invalidate_inode_pages.
1793 * To make sure no other process adds more write requests while we try
1794 * our best to flush them, we make them sleep during the attribute refresh.
1795 *
1796 * A very similar scenario holds for the dir cache.
1797 */
1798static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1799{
1800 struct nfs_server *server;
1801 struct nfs_inode *nfsi = NFS_I(inode);
1802 loff_t cur_isize, new_isize;
1803 unsigned long invalid = 0;
1804 unsigned long now = jiffies;
1805 unsigned long save_cache_validity;
1806 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1807 bool cache_revalidated = true;
1808 bool attr_changed = false;
1809 bool have_delegation;
1810
1811 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1812 __func__, inode->i_sb->s_id, inode->i_ino,
1813 nfs_display_fhandle_hash(NFS_FH(inode)),
1814 atomic_read(&inode->i_count), fattr->valid);
1815
1816 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1817 /* Only a mounted-on-fileid? Just exit */
1818 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1819 return 0;
1820 /* Has the inode gone and changed behind our back? */
1821 } else if (nfsi->fileid != fattr->fileid) {
1822 /* Is this perhaps the mounted-on fileid? */
1823 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1824 nfsi->fileid == fattr->mounted_on_fileid)
1825 return 0;
1826 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1827 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1828 NFS_SERVER(inode)->nfs_client->cl_hostname,
1829 inode->i_sb->s_id, (long long)nfsi->fileid,
1830 (long long)fattr->fileid);
1831 goto out_err;
1832 }
1833
1834 /*
1835 * Make sure the inode's type hasn't changed.
1836 */
1837 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1838 /*
1839 * Big trouble! The inode has become a different object.
1840 */
1841 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1842 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1843 goto out_err;
1844 }
1845
1846 server = NFS_SERVER(inode);
1847 /* Update the fsid? */
1848 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1849 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1850 !IS_AUTOMOUNT(inode))
1851 server->fsid = fattr->fsid;
1852
1853 /* Save the delegation state before clearing cache_validity */
1854 have_delegation = nfs_have_delegated_attributes(inode);
1855
1856 /*
1857 * Update the read time so we don't revalidate too often.
1858 */
1859 nfsi->read_cache_jiffies = fattr->time_start;
1860
1861 save_cache_validity = nfsi->cache_validity;
1862 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1863 | NFS_INO_INVALID_ATIME
1864 | NFS_INO_REVAL_FORCED
1865 | NFS_INO_REVAL_PAGECACHE
1866 | NFS_INO_INVALID_BLOCKS);
1867
1868 /* Do atomic weak cache consistency updates */
1869 nfs_wcc_update_inode(inode, fattr);
1870
1871 if (pnfs_layoutcommit_outstanding(inode)) {
1872 nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1873 cache_revalidated = false;
1874 }
1875
1876 /* More cache consistency checks */
1877 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1878 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1879 /* Could it be a race with writeback? */
1880 if (!(have_writers || have_delegation)) {
1881 invalid |= NFS_INO_INVALID_DATA
1882 | NFS_INO_INVALID_ACCESS
1883 | NFS_INO_INVALID_ACL;
1884 /* Force revalidate of all attributes */
1885 save_cache_validity |= NFS_INO_INVALID_CTIME
1886 | NFS_INO_INVALID_MTIME
1887 | NFS_INO_INVALID_SIZE
1888 | NFS_INO_INVALID_OTHER;
1889 if (S_ISDIR(inode->i_mode))
1890 nfs_force_lookup_revalidate(inode);
1891 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1892 inode->i_sb->s_id,
1893 inode->i_ino);
1894 } else if (!have_delegation)
1895 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1896 inode_set_iversion_raw(inode, fattr->change_attr);
1897 attr_changed = true;
1898 }
1899 } else {
1900 nfsi->cache_validity |= save_cache_validity &
1901 (NFS_INO_INVALID_CHANGE
1902 | NFS_INO_REVAL_PAGECACHE
1903 | NFS_INO_REVAL_FORCED);
1904 cache_revalidated = false;
1905 }
1906
1907 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1908 inode->i_mtime = fattr->mtime;
1909 } else if (server->caps & NFS_CAP_MTIME) {
1910 nfsi->cache_validity |= save_cache_validity &
1911 (NFS_INO_INVALID_MTIME
1912 | NFS_INO_REVAL_FORCED);
1913 cache_revalidated = false;
1914 }
1915
1916 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1917 inode->i_ctime = fattr->ctime;
1918 } else if (server->caps & NFS_CAP_CTIME) {
1919 nfsi->cache_validity |= save_cache_validity &
1920 (NFS_INO_INVALID_CTIME
1921 | NFS_INO_REVAL_FORCED);
1922 cache_revalidated = false;
1923 }
1924
1925 /* Check if our cached file size is stale */
1926 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1927 new_isize = nfs_size_to_loff_t(fattr->size);
1928 cur_isize = i_size_read(inode);
1929 if (new_isize != cur_isize && !have_delegation) {
1930 /* Do we perhaps have any outstanding writes, or has
1931 * the file grown beyond our last write? */
1932 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1933 i_size_write(inode, new_isize);
1934 if (!have_writers)
1935 invalid |= NFS_INO_INVALID_DATA;
1936 attr_changed = true;
1937 }
1938 dprintk("NFS: isize change on server for file %s/%ld "
1939 "(%Ld to %Ld)\n",
1940 inode->i_sb->s_id,
1941 inode->i_ino,
1942 (long long)cur_isize,
1943 (long long)new_isize);
1944 }
1945 } else {
1946 nfsi->cache_validity |= save_cache_validity &
1947 (NFS_INO_INVALID_SIZE
1948 | NFS_INO_REVAL_PAGECACHE
1949 | NFS_INO_REVAL_FORCED);
1950 cache_revalidated = false;
1951 }
1952
1953
1954 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1955 inode->i_atime = fattr->atime;
1956 else if (server->caps & NFS_CAP_ATIME) {
1957 nfsi->cache_validity |= save_cache_validity &
1958 (NFS_INO_INVALID_ATIME
1959 | NFS_INO_REVAL_FORCED);
1960 cache_revalidated = false;
1961 }
1962
1963 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1964 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1965 umode_t newmode = inode->i_mode & S_IFMT;
1966 newmode |= fattr->mode & S_IALLUGO;
1967 inode->i_mode = newmode;
1968 invalid |= NFS_INO_INVALID_ACCESS
1969 | NFS_INO_INVALID_ACL;
1970 attr_changed = true;
1971 }
1972 } else if (server->caps & NFS_CAP_MODE) {
1973 nfsi->cache_validity |= save_cache_validity &
1974 (NFS_INO_INVALID_OTHER
1975 | NFS_INO_REVAL_FORCED);
1976 cache_revalidated = false;
1977 }
1978
1979 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1980 if (!uid_eq(inode->i_uid, fattr->uid)) {
1981 invalid |= NFS_INO_INVALID_ACCESS
1982 | NFS_INO_INVALID_ACL;
1983 inode->i_uid = fattr->uid;
1984 attr_changed = true;
1985 }
1986 } else if (server->caps & NFS_CAP_OWNER) {
1987 nfsi->cache_validity |= save_cache_validity &
1988 (NFS_INO_INVALID_OTHER
1989 | NFS_INO_REVAL_FORCED);
1990 cache_revalidated = false;
1991 }
1992
1993 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1994 if (!gid_eq(inode->i_gid, fattr->gid)) {
1995 invalid |= NFS_INO_INVALID_ACCESS
1996 | NFS_INO_INVALID_ACL;
1997 inode->i_gid = fattr->gid;
1998 attr_changed = true;
1999 }
2000 } else if (server->caps & NFS_CAP_OWNER_GROUP) {
2001 nfsi->cache_validity |= save_cache_validity &
2002 (NFS_INO_INVALID_OTHER
2003 | NFS_INO_REVAL_FORCED);
2004 cache_revalidated = false;
2005 }
2006
2007 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2008 if (inode->i_nlink != fattr->nlink) {
2009 if (S_ISDIR(inode->i_mode))
2010 invalid |= NFS_INO_INVALID_DATA;
2011 set_nlink(inode, fattr->nlink);
2012 attr_changed = true;
2013 }
2014 } else if (server->caps & NFS_CAP_NLINK) {
2015 nfsi->cache_validity |= save_cache_validity &
2016 (NFS_INO_INVALID_OTHER
2017 | NFS_INO_REVAL_FORCED);
2018 cache_revalidated = false;
2019 }
2020
2021 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2022 /*
2023 * report the blocks in 512byte units
2024 */
2025 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2026 } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2027 inode->i_blocks = fattr->du.nfs2.blocks;
2028 else {
2029 nfsi->cache_validity |= save_cache_validity &
2030 (NFS_INO_INVALID_BLOCKS
2031 | NFS_INO_REVAL_FORCED);
2032 cache_revalidated = false;
2033 }
2034
2035 /* Update attrtimeo value if we're out of the unstable period */
2036 if (attr_changed) {
2037 invalid &= ~NFS_INO_INVALID_ATTR;
2038 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2039 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2040 nfsi->attrtimeo_timestamp = now;
2041 /* Set barrier to be more recent than all outstanding updates */
2042 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2043 } else {
2044 if (cache_revalidated) {
2045 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2046 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2047 nfsi->attrtimeo <<= 1;
2048 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2049 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2050 }
2051 nfsi->attrtimeo_timestamp = now;
2052 }
2053 /* Set the barrier to be more recent than this fattr */
2054 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2055 nfsi->attr_gencount = fattr->gencount;
2056 }
2057
2058 /* Don't invalidate the data if we were to blame */
2059 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2060 || S_ISLNK(inode->i_mode)))
2061 invalid &= ~NFS_INO_INVALID_DATA;
2062 nfs_set_cache_invalid(inode, invalid);
2063
2064 return 0;
2065 out_err:
2066 /*
2067 * No need to worry about unhashing the dentry, as the
2068 * lookup validation will know that the inode is bad.
2069 * (But we fall through to invalidate the caches.)
2070 */
2071 nfs_invalidate_inode(inode);
2072 return -ESTALE;
2073}
2074
2075struct inode *nfs_alloc_inode(struct super_block *sb)
2076{
2077 struct nfs_inode *nfsi;
2078 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2079 if (!nfsi)
2080 return NULL;
2081 nfsi->flags = 0UL;
2082 nfsi->cache_validity = 0UL;
2083#if IS_ENABLED(CONFIG_NFS_V4)
2084 nfsi->nfs4_acl = NULL;
2085#endif /* CONFIG_NFS_V4 */
2086 return &nfsi->vfs_inode;
2087}
2088EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2089
2090void nfs_free_inode(struct inode *inode)
2091{
2092 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2093}
2094EXPORT_SYMBOL_GPL(nfs_free_inode);
2095
2096static inline void nfs4_init_once(struct nfs_inode *nfsi)
2097{
2098#if IS_ENABLED(CONFIG_NFS_V4)
2099 INIT_LIST_HEAD(&nfsi->open_states);
2100 nfsi->delegation = NULL;
2101 init_rwsem(&nfsi->rwsem);
2102 nfsi->layout = NULL;
2103#endif
2104}
2105
2106static void init_once(void *foo)
2107{
2108 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2109
2110 inode_init_once(&nfsi->vfs_inode);
2111 INIT_LIST_HEAD(&nfsi->open_files);
2112 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2113 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2114 INIT_LIST_HEAD(&nfsi->commit_info.list);
2115 atomic_long_set(&nfsi->nrequests, 0);
2116 atomic_long_set(&nfsi->commit_info.ncommit, 0);
2117 atomic_set(&nfsi->commit_info.rpcs_out, 0);
2118 init_rwsem(&nfsi->rmdir_sem);
2119 mutex_init(&nfsi->commit_mutex);
2120 nfs4_init_once(nfsi);
2121}
2122
2123static int __init nfs_init_inodecache(void)
2124{
2125 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2126 sizeof(struct nfs_inode),
2127 0, (SLAB_RECLAIM_ACCOUNT|
2128 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2129 init_once);
2130 if (nfs_inode_cachep == NULL)
2131 return -ENOMEM;
2132
2133 return 0;
2134}
2135
2136static void nfs_destroy_inodecache(void)
2137{
2138 /*
2139 * Make sure all delayed rcu free inodes are flushed before we
2140 * destroy cache.
2141 */
2142 rcu_barrier();
2143 kmem_cache_destroy(nfs_inode_cachep);
2144}
2145
2146struct workqueue_struct *nfsiod_workqueue;
2147EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2148
2149/*
2150 * start up the nfsiod workqueue
2151 */
2152static int nfsiod_start(void)
2153{
2154 struct workqueue_struct *wq;
2155 dprintk("RPC: creating workqueue nfsiod\n");
2156 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2157 if (wq == NULL)
2158 return -ENOMEM;
2159 nfsiod_workqueue = wq;
2160 return 0;
2161}
2162
2163/*
2164 * Destroy the nfsiod workqueue
2165 */
2166static void nfsiod_stop(void)
2167{
2168 struct workqueue_struct *wq;
2169
2170 wq = nfsiod_workqueue;
2171 if (wq == NULL)
2172 return;
2173 nfsiod_workqueue = NULL;
2174 destroy_workqueue(wq);
2175}
2176
2177unsigned int nfs_net_id;
2178EXPORT_SYMBOL_GPL(nfs_net_id);
2179
2180static int nfs_net_init(struct net *net)
2181{
2182 struct nfs_net *nn = net_generic(net, nfs_net_id);
2183
2184 nfs_clients_init(net);
2185
2186 if (!rpc_proc_register(net, &nn->rpcstats)) {
2187 nfs_clients_exit(net);
2188 return -ENOMEM;
2189 }
2190
2191 return nfs_fs_proc_net_init(net);
2192}
2193
2194static void nfs_net_exit(struct net *net)
2195{
2196 rpc_proc_unregister(net, "nfs");
2197 nfs_fs_proc_net_exit(net);
2198 nfs_clients_exit(net);
2199}
2200
2201static struct pernet_operations nfs_net_ops = {
2202 .init = nfs_net_init,
2203 .exit = nfs_net_exit,
2204 .id = &nfs_net_id,
2205 .size = sizeof(struct nfs_net),
2206};
2207
2208/*
2209 * Initialize NFS
2210 */
2211static int __init init_nfs_fs(void)
2212{
2213 int err;
2214
2215 err = nfs_sysfs_init();
2216 if (err < 0)
2217 goto out10;
2218
2219 err = register_pernet_subsys(&nfs_net_ops);
2220 if (err < 0)
2221 goto out9;
2222
2223 err = nfs_fscache_register();
2224 if (err < 0)
2225 goto out8;
2226
2227 err = nfsiod_start();
2228 if (err)
2229 goto out7;
2230
2231 err = nfs_fs_proc_init();
2232 if (err)
2233 goto out6;
2234
2235 err = nfs_init_nfspagecache();
2236 if (err)
2237 goto out5;
2238
2239 err = nfs_init_inodecache();
2240 if (err)
2241 goto out4;
2242
2243 err = nfs_init_readpagecache();
2244 if (err)
2245 goto out3;
2246
2247 err = nfs_init_writepagecache();
2248 if (err)
2249 goto out2;
2250
2251 err = nfs_init_directcache();
2252 if (err)
2253 goto out1;
2254
2255 err = register_nfs_fs();
2256 if (err)
2257 goto out0;
2258
2259 return 0;
2260out0:
2261 nfs_destroy_directcache();
2262out1:
2263 nfs_destroy_writepagecache();
2264out2:
2265 nfs_destroy_readpagecache();
2266out3:
2267 nfs_destroy_inodecache();
2268out4:
2269 nfs_destroy_nfspagecache();
2270out5:
2271 nfs_fs_proc_exit();
2272out6:
2273 nfsiod_stop();
2274out7:
2275 nfs_fscache_unregister();
2276out8:
2277 unregister_pernet_subsys(&nfs_net_ops);
2278out9:
2279 nfs_sysfs_exit();
2280out10:
2281 return err;
2282}
2283
2284static void __exit exit_nfs_fs(void)
2285{
2286 nfs_destroy_directcache();
2287 nfs_destroy_writepagecache();
2288 nfs_destroy_readpagecache();
2289 nfs_destroy_inodecache();
2290 nfs_destroy_nfspagecache();
2291 nfs_fscache_unregister();
2292 unregister_pernet_subsys(&nfs_net_ops);
2293 unregister_nfs_fs();
2294 nfs_fs_proc_exit();
2295 nfsiod_stop();
2296 nfs_sysfs_exit();
2297}
2298
2299/* Not quite true; I just maintain it */
2300MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2301MODULE_LICENSE("GPL");
2302module_param(enable_ino64, bool, 0644);
2303
2304module_init(init_nfs_fs)
2305module_exit(exit_nfs_fs)