projects / linux-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge tag 'docs-5.7-2' of git://git.lwn.net/linux
[linux-block.git] / fs / nfs / inode.c
... / ...
CommitLineData
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_invalid_delegated(struct inode *inode, unsigned long flags)
168{
169 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
170
171 /* Special case for the pagecache or access cache */
172 if (flags == NFS_INO_REVAL_PAGECACHE &&
173 !(cache_validity & NFS_INO_REVAL_FORCED))
174 return false;
175 return (cache_validity & flags) != 0;
176}
177
178static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
179{
180 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
181
182 if ((cache_validity & flags) != 0)
183 return true;
184 if (nfs_attribute_timeout(inode))
185 return true;
186 return false;
187}
188
189bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
190{
191 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
192 return nfs_check_cache_invalid_delegated(inode, flags);
193
194 return nfs_check_cache_invalid_not_delegated(inode, flags);
195}
196
197static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
198{
199 struct nfs_inode *nfsi = NFS_I(inode);
200 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
201
202 if (have_delegation) {
203 if (!(flags & NFS_INO_REVAL_FORCED))
204 flags &= ~NFS_INO_INVALID_OTHER;
205 flags &= ~(NFS_INO_INVALID_CHANGE
206 | NFS_INO_INVALID_SIZE
207 | NFS_INO_REVAL_PAGECACHE);
208 }
209
210 if (inode->i_mapping->nrpages == 0)
211 flags &= ~(NFS_INO_INVALID_DATA|NFS_INO_DATA_INVAL_DEFER);
212 nfsi->cache_validity |= flags;
213 if (flags & NFS_INO_INVALID_DATA)
214 nfs_fscache_invalidate(inode);
215}
216
217/*
218 * Invalidate the local caches
219 */
220static void nfs_zap_caches_locked(struct inode *inode)
221{
222 struct nfs_inode *nfsi = NFS_I(inode);
223 int mode = inode->i_mode;
224
225 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
226
227 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
228 nfsi->attrtimeo_timestamp = jiffies;
229
230 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
231 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
232 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
233 | NFS_INO_INVALID_DATA
234 | NFS_INO_INVALID_ACCESS
235 | NFS_INO_INVALID_ACL
236 | NFS_INO_REVAL_PAGECACHE);
237 } else
238 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
239 | NFS_INO_INVALID_ACCESS
240 | NFS_INO_INVALID_ACL
241 | NFS_INO_REVAL_PAGECACHE);
242 nfs_zap_label_cache_locked(nfsi);
243}
244
245void nfs_zap_caches(struct inode *inode)
246{
247 spin_lock(&inode->i_lock);
248 nfs_zap_caches_locked(inode);
249 spin_unlock(&inode->i_lock);
250}
251
252void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
253{
254 if (mapping->nrpages != 0) {
255 spin_lock(&inode->i_lock);
256 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
257 spin_unlock(&inode->i_lock);
258 }
259}
260
261void nfs_zap_acl_cache(struct inode *inode)
262{
263 void (*clear_acl_cache)(struct inode *);
264
265 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
266 if (clear_acl_cache != NULL)
267 clear_acl_cache(inode);
268 spin_lock(&inode->i_lock);
269 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
270 spin_unlock(&inode->i_lock);
271}
272EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
273
274void nfs_invalidate_atime(struct inode *inode)
275{
276 spin_lock(&inode->i_lock);
277 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
278 spin_unlock(&inode->i_lock);
279}
280EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
281
282/*
283 * Invalidate, but do not unhash, the inode.
284 * NB: must be called with inode->i_lock held!
285 */
286static void nfs_set_inode_stale_locked(struct inode *inode)
287{
288 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
289 nfs_zap_caches_locked(inode);
290 trace_nfs_set_inode_stale(inode);
291}
292
293void nfs_set_inode_stale(struct inode *inode)
294{
295 spin_lock(&inode->i_lock);
296 nfs_set_inode_stale_locked(inode);
297 spin_unlock(&inode->i_lock);
298}
299
300struct nfs_find_desc {
301 struct nfs_fh *fh;
302 struct nfs_fattr *fattr;
303};
304
305/*
306 * In NFSv3 we can have 64bit inode numbers. In order to support
307 * this, and re-exported directories (also seen in NFSv2)
308 * we are forced to allow 2 different inodes to have the same
309 * i_ino.
310 */
311static int
312nfs_find_actor(struct inode *inode, void *opaque)
313{
314 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
315 struct nfs_fh *fh = desc->fh;
316 struct nfs_fattr *fattr = desc->fattr;
317
318 if (NFS_FILEID(inode) != fattr->fileid)
319 return 0;
320 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
321 return 0;
322 if (nfs_compare_fh(NFS_FH(inode), fh))
323 return 0;
324 if (is_bad_inode(inode) || NFS_STALE(inode))
325 return 0;
326 return 1;
327}
328
329static int
330nfs_init_locked(struct inode *inode, void *opaque)
331{
332 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
333 struct nfs_fattr *fattr = desc->fattr;
334
335 set_nfs_fileid(inode, fattr->fileid);
336 inode->i_mode = fattr->mode;
337 nfs_copy_fh(NFS_FH(inode), desc->fh);
338 return 0;
339}
340
341#ifdef CONFIG_NFS_V4_SECURITY_LABEL
342static void nfs_clear_label_invalid(struct inode *inode)
343{
344 spin_lock(&inode->i_lock);
345 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
346 spin_unlock(&inode->i_lock);
347}
348
349void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
350 struct nfs4_label *label)
351{
352 int error;
353
354 if (label == NULL)
355 return;
356
357 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
358 error = security_inode_notifysecctx(inode, label->label,
359 label->len);
360 if (error)
361 printk(KERN_ERR "%s() %s %d "
362 "security_inode_notifysecctx() %d\n",
363 __func__,
364 (char *)label->label,
365 label->len, error);
366 nfs_clear_label_invalid(inode);
367 }
368}
369
370struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
371{
372 struct nfs4_label *label = NULL;
373 int minor_version = server->nfs_client->cl_minorversion;
374
375 if (minor_version < 2)
376 return label;
377
378 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
379 return label;
380
381 label = kzalloc(sizeof(struct nfs4_label), flags);
382 if (label == NULL)
383 return ERR_PTR(-ENOMEM);
384
385 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
386 if (label->label == NULL) {
387 kfree(label);
388 return ERR_PTR(-ENOMEM);
389 }
390 label->len = NFS4_MAXLABELLEN;
391
392 return label;
393}
394EXPORT_SYMBOL_GPL(nfs4_label_alloc);
395#else
396void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
397 struct nfs4_label *label)
398{
399}
400#endif
401EXPORT_SYMBOL_GPL(nfs_setsecurity);
402
403/* Search for inode identified by fh, fileid and i_mode in inode cache. */
404struct inode *
405nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
406{
407 struct nfs_find_desc desc = {
408 .fh = fh,
409 .fattr = fattr,
410 };
411 struct inode *inode;
412 unsigned long hash;
413
414 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
415 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
416 return NULL;
417
418 hash = nfs_fattr_to_ino_t(fattr);
419 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
420
421 dprintk("%s: returning %p\n", __func__, inode);
422 return inode;
423}
424
425/*
426 * This is our front-end to iget that looks up inodes by file handle
427 * instead of inode number.
428 */
429struct inode *
430nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
431{
432 struct nfs_find_desc desc = {
433 .fh = fh,
434 .fattr = fattr
435 };
436 struct inode *inode = ERR_PTR(-ENOENT);
437 unsigned long hash;
438
439 nfs_attr_check_mountpoint(sb, fattr);
440
441 if (nfs_attr_use_mounted_on_fileid(fattr))
442 fattr->fileid = fattr->mounted_on_fileid;
443 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
444 goto out_no_inode;
445 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
446 goto out_no_inode;
447
448 hash = nfs_fattr_to_ino_t(fattr);
449
450 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
451 if (inode == NULL) {
452 inode = ERR_PTR(-ENOMEM);
453 goto out_no_inode;
454 }
455
456 if (inode->i_state & I_NEW) {
457 struct nfs_inode *nfsi = NFS_I(inode);
458 unsigned long now = jiffies;
459
460 /* We set i_ino for the few things that still rely on it,
461 * such as stat(2) */
462 inode->i_ino = hash;
463
464 /* We can't support update_atime(), since the server will reset it */
465 inode->i_flags |= S_NOATIME|S_NOCMTIME;
466 inode->i_mode = fattr->mode;
467 nfsi->cache_validity = 0;
468 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
469 && nfs_server_capable(inode, NFS_CAP_MODE))
470 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
471 /* Why so? Because we want revalidate for devices/FIFOs, and
472 * that's precisely what we have in nfs_file_inode_operations.
473 */
474 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
475 if (S_ISREG(inode->i_mode)) {
476 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
477 inode->i_data.a_ops = &nfs_file_aops;
478 } else if (S_ISDIR(inode->i_mode)) {
479 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
480 inode->i_fop = &nfs_dir_operations;
481 inode->i_data.a_ops = &nfs_dir_aops;
482 /* Deal with crossing mountpoints */
483 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
484 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
485 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
486 inode->i_op = &nfs_referral_inode_operations;
487 else
488 inode->i_op = &nfs_mountpoint_inode_operations;
489 inode->i_fop = NULL;
490 inode->i_flags |= S_AUTOMOUNT;
491 }
492 } else if (S_ISLNK(inode->i_mode)) {
493 inode->i_op = &nfs_symlink_inode_operations;
494 inode_nohighmem(inode);
495 } else
496 init_special_inode(inode, inode->i_mode, fattr->rdev);
497
498 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
499 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
500 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
501 inode_set_iversion_raw(inode, 0);
502 inode->i_size = 0;
503 clear_nlink(inode);
504 inode->i_uid = make_kuid(&init_user_ns, -2);
505 inode->i_gid = make_kgid(&init_user_ns, -2);
506 inode->i_blocks = 0;
507 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
508 nfsi->write_io = 0;
509 nfsi->read_io = 0;
510
511 nfsi->read_cache_jiffies = fattr->time_start;
512 nfsi->attr_gencount = fattr->gencount;
513 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
514 inode->i_atime = fattr->atime;
515 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
516 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
517 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
518 inode->i_mtime = fattr->mtime;
519 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
520 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
521 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
522 inode->i_ctime = fattr->ctime;
523 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
524 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
525 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
526 inode_set_iversion_raw(inode, fattr->change_attr);
527 else
528 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
529 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
530 inode->i_size = nfs_size_to_loff_t(fattr->size);
531 else
532 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
533 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
534 set_nlink(inode, fattr->nlink);
535 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
536 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
537 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
538 inode->i_uid = fattr->uid;
539 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
540 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
541 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
542 inode->i_gid = fattr->gid;
543 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
544 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
545 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
546 inode->i_blocks = fattr->du.nfs2.blocks;
547 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
548 /*
549 * report the blocks in 512byte units
550 */
551 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
552 }
553
554 if (nfsi->cache_validity != 0)
555 nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
556
557 nfs_setsecurity(inode, fattr, label);
558
559 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
560 nfsi->attrtimeo_timestamp = now;
561 nfsi->access_cache = RB_ROOT;
562
563 nfs_fscache_init_inode(inode);
564
565 unlock_new_inode(inode);
566 } else {
567 int err = nfs_refresh_inode(inode, fattr);
568 if (err < 0) {
569 iput(inode);
570 inode = ERR_PTR(err);
571 goto out_no_inode;
572 }
573 }
574 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
575 inode->i_sb->s_id,
576 (unsigned long long)NFS_FILEID(inode),
577 nfs_display_fhandle_hash(fh),
578 atomic_read(&inode->i_count));
579
580out:
581 return inode;
582
583out_no_inode:
584 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
585 goto out;
586}
587EXPORT_SYMBOL_GPL(nfs_fhget);
588
589#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)
590
591int
592nfs_setattr(struct dentry *dentry, struct iattr *attr)
593{
594 struct inode *inode = d_inode(dentry);
595 struct nfs_fattr *fattr;
596 int error = 0;
597
598 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
599
600 /* skip mode change if it's just for clearing setuid/setgid */
601 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
602 attr->ia_valid &= ~ATTR_MODE;
603
604 if (attr->ia_valid & ATTR_SIZE) {
605 BUG_ON(!S_ISREG(inode->i_mode));
606
607 error = inode_newsize_ok(inode, attr->ia_size);
608 if (error)
609 return error;
610
611 if (attr->ia_size == i_size_read(inode))
612 attr->ia_valid &= ~ATTR_SIZE;
613 }
614
615 /* Optimization: if the end result is no change, don't RPC */
616 attr->ia_valid &= NFS_VALID_ATTRS;
617 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
618 return 0;
619
620 trace_nfs_setattr_enter(inode);
621
622 /* Write all dirty data */
623 if (S_ISREG(inode->i_mode))
624 nfs_sync_inode(inode);
625
626 fattr = nfs_alloc_fattr();
627 if (fattr == NULL) {
628 error = -ENOMEM;
629 goto out;
630 }
631
632 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
633 if (error == 0)
634 error = nfs_refresh_inode(inode, fattr);
635 nfs_free_fattr(fattr);
636out:
637 trace_nfs_setattr_exit(inode, error);
638 return error;
639}
640EXPORT_SYMBOL_GPL(nfs_setattr);
641
642/**
643 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
644 * @inode: inode of the file used
645 * @offset: file offset to start truncating
646 *
647 * This is a copy of the common vmtruncate, but with the locking
648 * corrected to take into account the fact that NFS requires
649 * inode->i_size to be updated under the inode->i_lock.
650 * Note: must be called with inode->i_lock held!
651 */
652static int nfs_vmtruncate(struct inode * inode, loff_t offset)
653{
654 int err;
655
656 err = inode_newsize_ok(inode, offset);
657 if (err)
658 goto out;
659
660 i_size_write(inode, offset);
661 /* Optimisation */
662 if (offset == 0)
663 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
664 NFS_INO_DATA_INVAL_DEFER);
665 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
666
667 spin_unlock(&inode->i_lock);
668 truncate_pagecache(inode, offset);
669 spin_lock(&inode->i_lock);
670out:
671 return err;
672}
673
674/**
675 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
676 * @inode: pointer to struct inode
677 * @attr: pointer to struct iattr
678 * @fattr: pointer to struct nfs_fattr
679 *
680 * Note: we do this in the *proc.c in order to ensure that
681 * it works for things like exclusive creates too.
682 */
683void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
684 struct nfs_fattr *fattr)
685{
686 /* Barrier: bump the attribute generation count. */
687 nfs_fattr_set_barrier(fattr);
688
689 spin_lock(&inode->i_lock);
690 NFS_I(inode)->attr_gencount = fattr->gencount;
691 if ((attr->ia_valid & ATTR_SIZE) != 0) {
692 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
693 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
694 nfs_vmtruncate(inode, attr->ia_size);
695 }
696 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
697 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
698 if ((attr->ia_valid & ATTR_MODE) != 0) {
699 int mode = attr->ia_mode & S_IALLUGO;
700 mode |= inode->i_mode & ~S_IALLUGO;
701 inode->i_mode = mode;
702 }
703 if ((attr->ia_valid & ATTR_UID) != 0)
704 inode->i_uid = attr->ia_uid;
705 if ((attr->ia_valid & ATTR_GID) != 0)
706 inode->i_gid = attr->ia_gid;
707 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
708 inode->i_ctime = fattr->ctime;
709 else
710 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
711 | NFS_INO_INVALID_CTIME);
712 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
713 | NFS_INO_INVALID_ACL);
714 }
715 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
716 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
717 | NFS_INO_INVALID_CTIME);
718 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
719 inode->i_atime = fattr->atime;
720 else if (attr->ia_valid & ATTR_ATIME_SET)
721 inode->i_atime = attr->ia_atime;
722 else
723 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
724
725 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
726 inode->i_ctime = fattr->ctime;
727 else
728 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
729 | NFS_INO_INVALID_CTIME);
730 }
731 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
732 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
733 | NFS_INO_INVALID_CTIME);
734 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
735 inode->i_mtime = fattr->mtime;
736 else if (attr->ia_valid & ATTR_MTIME_SET)
737 inode->i_mtime = attr->ia_mtime;
738 else
739 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
740
741 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
742 inode->i_ctime = fattr->ctime;
743 else
744 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
745 | NFS_INO_INVALID_CTIME);
746 }
747 if (fattr->valid)
748 nfs_update_inode(inode, fattr);
749 spin_unlock(&inode->i_lock);
750}
751EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
752
753static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
754{
755 struct dentry *parent;
756
757 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
758 return;
759 parent = dget_parent(dentry);
760 nfs_force_use_readdirplus(d_inode(parent));
761 dput(parent);
762}
763
764static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
765{
766 struct dentry *parent;
767
768 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
769 return;
770 parent = dget_parent(dentry);
771 nfs_advise_use_readdirplus(d_inode(parent));
772 dput(parent);
773}
774
775static bool nfs_need_revalidate_inode(struct inode *inode)
776{
777 if (NFS_I(inode)->cache_validity &
778 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
779 return true;
780 if (nfs_attribute_cache_expired(inode))
781 return true;
782 return false;
783}
784
785int nfs_getattr(const struct path *path, struct kstat *stat,
786 u32 request_mask, unsigned int query_flags)
787{
788 struct inode *inode = d_inode(path->dentry);
789 struct nfs_server *server = NFS_SERVER(inode);
790 unsigned long cache_validity;
791 int err = 0;
792 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
793 bool do_update = false;
794
795 trace_nfs_getattr_enter(inode);
796
797 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync)
798 goto out_no_update;
799
800 /* Flush out writes to the server in order to update c/mtime. */
801 if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
802 S_ISREG(inode->i_mode)) {
803 err = filemap_write_and_wait(inode->i_mapping);
804 if (err)
805 goto out;
806 }
807
808 /*
809 * We may force a getattr if the user cares about atime.
810 *
811 * Note that we only have to check the vfsmount flags here:
812 * - NFS always sets S_NOATIME by so checking it would give a
813 * bogus result
814 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
815 * no point in checking those.
816 */
817 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
818 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
819 request_mask &= ~STATX_ATIME;
820
821 /* Is the user requesting attributes that might need revalidation? */
822 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
823 STATX_MTIME|STATX_UID|STATX_GID|
824 STATX_SIZE|STATX_BLOCKS)))
825 goto out_no_revalidate;
826
827 /* Check whether the cached attributes are stale */
828 do_update |= force_sync || nfs_attribute_cache_expired(inode);
829 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
830 do_update |= cache_validity &
831 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
832 if (request_mask & STATX_ATIME)
833 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
834 if (request_mask & (STATX_CTIME|STATX_MTIME))
835 do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
836 if (do_update) {
837 /* Update the attribute cache */
838 if (!(server->flags & NFS_MOUNT_NOAC))
839 nfs_readdirplus_parent_cache_miss(path->dentry);
840 else
841 nfs_readdirplus_parent_cache_hit(path->dentry);
842 err = __nfs_revalidate_inode(server, inode);
843 if (err)
844 goto out;
845 } else
846 nfs_readdirplus_parent_cache_hit(path->dentry);
847out_no_revalidate:
848 /* Only return attributes that were revalidated. */
849 stat->result_mask &= request_mask;
850out_no_update:
851 generic_fillattr(inode, stat);
852 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
853 if (S_ISDIR(inode->i_mode))
854 stat->blksize = NFS_SERVER(inode)->dtsize;
855out:
856 trace_nfs_getattr_exit(inode, err);
857 return err;
858}
859EXPORT_SYMBOL_GPL(nfs_getattr);
860
861static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
862{
863 refcount_set(&l_ctx->count, 1);
864 l_ctx->lockowner = current->files;
865 INIT_LIST_HEAD(&l_ctx->list);
866 atomic_set(&l_ctx->io_count, 0);
867}
868
869static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
870{
871 struct nfs_lock_context *pos;
872
873 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
874 if (pos->lockowner != current->files)
875 continue;
876 if (refcount_inc_not_zero(&pos->count))
877 return pos;
878 }
879 return NULL;
880}
881
882struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
883{
884 struct nfs_lock_context *res, *new = NULL;
885 struct inode *inode = d_inode(ctx->dentry);
886
887 rcu_read_lock();
888 res = __nfs_find_lock_context(ctx);
889 rcu_read_unlock();
890 if (res == NULL) {
891 new = kmalloc(sizeof(*new), GFP_KERNEL);
892 if (new == NULL)
893 return ERR_PTR(-ENOMEM);
894 nfs_init_lock_context(new);
895 spin_lock(&inode->i_lock);
896 res = __nfs_find_lock_context(ctx);
897 if (res == NULL) {
898 new->open_context = get_nfs_open_context(ctx);
899 if (new->open_context) {
900 list_add_tail_rcu(&new->list,
901 &ctx->lock_context.list);
902 res = new;
903 new = NULL;
904 } else
905 res = ERR_PTR(-EBADF);
906 }
907 spin_unlock(&inode->i_lock);
908 kfree(new);
909 }
910 return res;
911}
912EXPORT_SYMBOL_GPL(nfs_get_lock_context);
913
914void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
915{
916 struct nfs_open_context *ctx = l_ctx->open_context;
917 struct inode *inode = d_inode(ctx->dentry);
918
919 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
920 return;
921 list_del_rcu(&l_ctx->list);
922 spin_unlock(&inode->i_lock);
923 put_nfs_open_context(ctx);
924 kfree_rcu(l_ctx, rcu_head);
925}
926EXPORT_SYMBOL_GPL(nfs_put_lock_context);
927
928/**
929 * nfs_close_context - Common close_context() routine NFSv2/v3
930 * @ctx: pointer to context
931 * @is_sync: is this a synchronous close
932 *
933 * Ensure that the attributes are up to date if we're mounted
934 * with close-to-open semantics and we have cached data that will
935 * need to be revalidated on open.
936 */
937void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
938{
939 struct nfs_inode *nfsi;
940 struct inode *inode;
941 struct nfs_server *server;
942
943 if (!(ctx->mode & FMODE_WRITE))
944 return;
945 if (!is_sync)
946 return;
947 inode = d_inode(ctx->dentry);
948 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
949 return;
950 nfsi = NFS_I(inode);
951 if (inode->i_mapping->nrpages == 0)
952 return;
953 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
954 return;
955 if (!list_empty(&nfsi->open_files))
956 return;
957 server = NFS_SERVER(inode);
958 if (server->flags & NFS_MOUNT_NOCTO)
959 return;
960 nfs_revalidate_inode(server, inode);
961}
962EXPORT_SYMBOL_GPL(nfs_close_context);
963
964struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
965 fmode_t f_mode,
966 struct file *filp)
967{
968 struct nfs_open_context *ctx;
969
970 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
971 if (!ctx)
972 return ERR_PTR(-ENOMEM);
973 nfs_sb_active(dentry->d_sb);
974 ctx->dentry = dget(dentry);
975 if (filp)
976 ctx->cred = get_cred(filp->f_cred);
977 else
978 ctx->cred = get_current_cred();
979 ctx->ll_cred = NULL;
980 ctx->state = NULL;
981 ctx->mode = f_mode;
982 ctx->flags = 0;
983 ctx->error = 0;
984 ctx->flock_owner = (fl_owner_t)filp;
985 nfs_init_lock_context(&ctx->lock_context);
986 ctx->lock_context.open_context = ctx;
987 INIT_LIST_HEAD(&ctx->list);
988 ctx->mdsthreshold = NULL;
989 return ctx;
990}
991EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
992
993struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
994{
995 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
996 return ctx;
997 return NULL;
998}
999EXPORT_SYMBOL_GPL(get_nfs_open_context);
1000
1001static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1002{
1003 struct inode *inode = d_inode(ctx->dentry);
1004 struct super_block *sb = ctx->dentry->d_sb;
1005
1006 if (!refcount_dec_and_test(&ctx->lock_context.count))
1007 return;
1008 if (!list_empty(&ctx->list)) {
1009 spin_lock(&inode->i_lock);
1010 list_del_rcu(&ctx->list);
1011 spin_unlock(&inode->i_lock);
1012 }
1013 if (inode != NULL)
1014 NFS_PROTO(inode)->close_context(ctx, is_sync);
1015 put_cred(ctx->cred);
1016 dput(ctx->dentry);
1017 nfs_sb_deactive(sb);
1018 put_rpccred(ctx->ll_cred);
1019 kfree(ctx->mdsthreshold);
1020 kfree_rcu(ctx, rcu_head);
1021}
1022
1023void put_nfs_open_context(struct nfs_open_context *ctx)
1024{
1025 __put_nfs_open_context(ctx, 0);
1026}
1027EXPORT_SYMBOL_GPL(put_nfs_open_context);
1028
1029static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1030{
1031 __put_nfs_open_context(ctx, 1);
1032}
1033
1034/*
1035 * Ensure that mmap has a recent RPC credential for use when writing out
1036 * shared pages
1037 */
1038void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1039{
1040 struct inode *inode = d_inode(ctx->dentry);
1041 struct nfs_inode *nfsi = NFS_I(inode);
1042
1043 spin_lock(&inode->i_lock);
1044 if (list_empty(&nfsi->open_files) &&
1045 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1046 nfsi->cache_validity |= NFS_INO_INVALID_DATA |
1047 NFS_INO_REVAL_FORCED;
1048 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1049 spin_unlock(&inode->i_lock);
1050}
1051EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1052
1053void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1054{
1055 filp->private_data = get_nfs_open_context(ctx);
1056 if (list_empty(&ctx->list))
1057 nfs_inode_attach_open_context(ctx);
1058}
1059EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1060
1061/*
1062 * Given an inode, search for an open context with the desired characteristics
1063 */
1064struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1065{
1066 struct nfs_inode *nfsi = NFS_I(inode);
1067 struct nfs_open_context *pos, *ctx = NULL;
1068
1069 rcu_read_lock();
1070 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1071 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1072 continue;
1073 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1074 continue;
1075 ctx = get_nfs_open_context(pos);
1076 if (ctx)
1077 break;
1078 }
1079 rcu_read_unlock();
1080 return ctx;
1081}
1082
1083void nfs_file_clear_open_context(struct file *filp)
1084{
1085 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1086
1087 if (ctx) {
1088 struct inode *inode = d_inode(ctx->dentry);
1089
1090 /*
1091 * We fatal error on write before. Try to writeback
1092 * every page again.
1093 */
1094 if (ctx->error < 0)
1095 invalidate_inode_pages2(inode->i_mapping);
1096 filp->private_data = NULL;
1097 put_nfs_open_context_sync(ctx);
1098 }
1099}
1100
1101/*
1102 * These allocate and release file read/write context information.
1103 */
1104int nfs_open(struct inode *inode, struct file *filp)
1105{
1106 struct nfs_open_context *ctx;
1107
1108 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1109 if (IS_ERR(ctx))
1110 return PTR_ERR(ctx);
1111 nfs_file_set_open_context(filp, ctx);
1112 put_nfs_open_context(ctx);
1113 nfs_fscache_open_file(inode, filp);
1114 return 0;
1115}
1116EXPORT_SYMBOL_GPL(nfs_open);
1117
1118/*
1119 * This function is called whenever some part of NFS notices that
1120 * the cached attributes have to be refreshed.
1121 */
1122int
1123__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1124{
1125 int status = -ESTALE;
1126 struct nfs4_label *label = NULL;
1127 struct nfs_fattr *fattr = NULL;
1128 struct nfs_inode *nfsi = NFS_I(inode);
1129
1130 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1131 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1132
1133 trace_nfs_revalidate_inode_enter(inode);
1134
1135 if (is_bad_inode(inode))
1136 goto out;
1137 if (NFS_STALE(inode))
1138 goto out;
1139
1140 /* pNFS: Attributes aren't updated until we layoutcommit */
1141 if (S_ISREG(inode->i_mode)) {
1142 status = pnfs_sync_inode(inode, false);
1143 if (status)
1144 goto out;
1145 }
1146
1147 status = -ENOMEM;
1148 fattr = nfs_alloc_fattr();
1149 if (fattr == NULL)
1150 goto out;
1151
1152 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1153
1154 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1155 if (IS_ERR(label)) {
1156 status = PTR_ERR(label);
1157 goto out;
1158 }
1159
1160 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1161 label, inode);
1162 if (status != 0) {
1163 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1164 inode->i_sb->s_id,
1165 (unsigned long long)NFS_FILEID(inode), status);
1166 switch (status) {
1167 case -ETIMEDOUT:
1168 /* A soft timeout occurred. Use cached information? */
1169 if (server->flags & NFS_MOUNT_SOFTREVAL)
1170 status = 0;
1171 break;
1172 case -ESTALE:
1173 if (!S_ISDIR(inode->i_mode))
1174 nfs_set_inode_stale(inode);
1175 else
1176 nfs_zap_caches(inode);
1177 }
1178 goto err_out;
1179 }
1180
1181 status = nfs_refresh_inode(inode, fattr);
1182 if (status) {
1183 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1184 inode->i_sb->s_id,
1185 (unsigned long long)NFS_FILEID(inode), status);
1186 goto err_out;
1187 }
1188
1189 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1190 nfs_zap_acl_cache(inode);
1191
1192 nfs_setsecurity(inode, fattr, label);
1193
1194 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1195 inode->i_sb->s_id,
1196 (unsigned long long)NFS_FILEID(inode));
1197
1198err_out:
1199 nfs4_label_free(label);
1200out:
1201 nfs_free_fattr(fattr);
1202 trace_nfs_revalidate_inode_exit(inode, status);
1203 return status;
1204}
1205
1206int nfs_attribute_cache_expired(struct inode *inode)
1207{
1208 if (nfs_have_delegated_attributes(inode))
1209 return 0;
1210 return nfs_attribute_timeout(inode);
1211}
1212
1213/**
1214 * nfs_revalidate_inode - Revalidate the inode attributes
1215 * @server: pointer to nfs_server struct
1216 * @inode: pointer to inode struct
1217 *
1218 * Updates inode attribute information by retrieving the data from the server.
1219 */
1220int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1221{
1222 if (!nfs_need_revalidate_inode(inode))
1223 return NFS_STALE(inode) ? -ESTALE : 0;
1224 return __nfs_revalidate_inode(server, inode);
1225}
1226EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1227
1228static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1229{
1230 struct nfs_inode *nfsi = NFS_I(inode);
1231 int ret;
1232
1233 if (mapping->nrpages != 0) {
1234 if (S_ISREG(inode->i_mode)) {
1235 ret = nfs_sync_mapping(mapping);
1236 if (ret < 0)
1237 return ret;
1238 }
1239 ret = invalidate_inode_pages2(mapping);
1240 if (ret < 0)
1241 return ret;
1242 }
1243 if (S_ISDIR(inode->i_mode)) {
1244 spin_lock(&inode->i_lock);
1245 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1246 spin_unlock(&inode->i_lock);
1247 }
1248 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1249 nfs_fscache_wait_on_invalidate(inode);
1250
1251 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1252 inode->i_sb->s_id,
1253 (unsigned long long)NFS_FILEID(inode));
1254 return 0;
1255}
1256
1257bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1258{
1259 return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1260 NFS_STALE(inode);
1261}
1262
1263int nfs_revalidate_mapping_rcu(struct inode *inode)
1264{
1265 struct nfs_inode *nfsi = NFS_I(inode);
1266 unsigned long *bitlock = &nfsi->flags;
1267 int ret = 0;
1268
1269 if (IS_SWAPFILE(inode))
1270 goto out;
1271 if (nfs_mapping_need_revalidate_inode(inode)) {
1272 ret = -ECHILD;
1273 goto out;
1274 }
1275 spin_lock(&inode->i_lock);
1276 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1277 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1278 ret = -ECHILD;
1279 spin_unlock(&inode->i_lock);
1280out:
1281 return ret;
1282}
1283
1284/**
1285 * nfs_revalidate_mapping - Revalidate the pagecache
1286 * @inode: pointer to host inode
1287 * @mapping: pointer to mapping
1288 */
1289int nfs_revalidate_mapping(struct inode *inode,
1290 struct address_space *mapping)
1291{
1292 struct nfs_inode *nfsi = NFS_I(inode);
1293 unsigned long *bitlock = &nfsi->flags;
1294 int ret = 0;
1295
1296 /* swapfiles are not supposed to be shared. */
1297 if (IS_SWAPFILE(inode))
1298 goto out;
1299
1300 if (nfs_mapping_need_revalidate_inode(inode)) {
1301 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1302 if (ret < 0)
1303 goto out;
1304 }
1305
1306 /*
1307 * We must clear NFS_INO_INVALID_DATA first to ensure that
1308 * invalidations that come in while we're shooting down the mappings
1309 * are respected. But, that leaves a race window where one revalidator
1310 * can clear the flag, and then another checks it before the mapping
1311 * gets invalidated. Fix that by serializing access to this part of
1312 * the function.
1313 *
1314 * At the same time, we need to allow other tasks to see whether we
1315 * might be in the middle of invalidating the pages, so we only set
1316 * the bit lock here if it looks like we're going to be doing that.
1317 */
1318 for (;;) {
1319 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1320 nfs_wait_bit_killable, TASK_KILLABLE);
1321 if (ret)
1322 goto out;
1323 spin_lock(&inode->i_lock);
1324 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1325 spin_unlock(&inode->i_lock);
1326 continue;
1327 }
1328 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1329 break;
1330 spin_unlock(&inode->i_lock);
1331 goto out;
1332 }
1333
1334 set_bit(NFS_INO_INVALIDATING, bitlock);
1335 smp_wmb();
1336 nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA|
1337 NFS_INO_DATA_INVAL_DEFER);
1338 spin_unlock(&inode->i_lock);
1339 trace_nfs_invalidate_mapping_enter(inode);
1340 ret = nfs_invalidate_mapping(inode, mapping);
1341 trace_nfs_invalidate_mapping_exit(inode, ret);
1342
1343 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1344 smp_mb__after_atomic();
1345 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1346out:
1347 return ret;
1348}
1349
1350static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1351{
1352 struct inode *inode = &nfsi->vfs_inode;
1353
1354 if (!S_ISREG(inode->i_mode))
1355 return false;
1356 if (list_empty(&nfsi->open_files))
1357 return false;
1358 return inode_is_open_for_write(inode);
1359}
1360
1361static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1362{
1363 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1364}
1365
1366static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1367{
1368 struct timespec64 ts;
1369
1370 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1371 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1372 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1373 inode_set_iversion_raw(inode, fattr->change_attr);
1374 if (S_ISDIR(inode->i_mode))
1375 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1376 }
1377 /* If we have atomic WCC data, we may update some attributes */
1378 ts = inode->i_ctime;
1379 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1380 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1381 && timespec64_equal(&ts, &fattr->pre_ctime)) {
1382 inode->i_ctime = fattr->ctime;
1383 }
1384
1385 ts = inode->i_mtime;
1386 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1387 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1388 && timespec64_equal(&ts, &fattr->pre_mtime)) {
1389 inode->i_mtime = fattr->mtime;
1390 if (S_ISDIR(inode->i_mode))
1391 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1392 }
1393 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1394 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1395 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1396 && !nfs_have_writebacks(inode)) {
1397 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1398 }
1399}
1400
1401/**
1402 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1403 * @inode: pointer to inode
1404 * @fattr: updated attributes
1405 *
1406 * Verifies the attribute cache. If we have just changed the attributes,
1407 * so that fattr carries weak cache consistency data, then it may
1408 * also update the ctime/mtime/change_attribute.
1409 */
1410static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1411{
1412 struct nfs_inode *nfsi = NFS_I(inode);
1413 loff_t cur_size, new_isize;
1414 unsigned long invalid = 0;
1415 struct timespec64 ts;
1416
1417 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1418 return 0;
1419
1420 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1421 /* Only a mounted-on-fileid? Just exit */
1422 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1423 return 0;
1424 /* Has the inode gone and changed behind our back? */
1425 } else if (nfsi->fileid != fattr->fileid) {
1426 /* Is this perhaps the mounted-on fileid? */
1427 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1428 nfsi->fileid == fattr->mounted_on_fileid)
1429 return 0;
1430 return -ESTALE;
1431 }
1432 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1433 return -ESTALE;
1434
1435
1436 if (!nfs_file_has_buffered_writers(nfsi)) {
1437 /* Verify a few of the more important attributes */
1438 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1439 invalid |= NFS_INO_INVALID_CHANGE
1440 | NFS_INO_REVAL_PAGECACHE;
1441
1442 ts = inode->i_mtime;
1443 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1444 invalid |= NFS_INO_INVALID_MTIME;
1445
1446 ts = inode->i_ctime;
1447 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1448 invalid |= NFS_INO_INVALID_CTIME;
1449
1450 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1451 cur_size = i_size_read(inode);
1452 new_isize = nfs_size_to_loff_t(fattr->size);
1453 if (cur_size != new_isize)
1454 invalid |= NFS_INO_INVALID_SIZE
1455 | NFS_INO_REVAL_PAGECACHE;
1456 }
1457 }
1458
1459 /* Have any file permissions changed? */
1460 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1461 invalid |= NFS_INO_INVALID_ACCESS
1462 | NFS_INO_INVALID_ACL
1463 | NFS_INO_INVALID_OTHER;
1464 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1465 invalid |= NFS_INO_INVALID_ACCESS
1466 | NFS_INO_INVALID_ACL
1467 | NFS_INO_INVALID_OTHER;
1468 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1469 invalid |= NFS_INO_INVALID_ACCESS
1470 | NFS_INO_INVALID_ACL
1471 | NFS_INO_INVALID_OTHER;
1472
1473 /* Has the link count changed? */
1474 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1475 invalid |= NFS_INO_INVALID_OTHER;
1476
1477 ts = inode->i_atime;
1478 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1479 invalid |= NFS_INO_INVALID_ATIME;
1480
1481 if (invalid != 0)
1482 nfs_set_cache_invalid(inode, invalid);
1483
1484 nfsi->read_cache_jiffies = fattr->time_start;
1485 return 0;
1486}
1487
1488static atomic_long_t nfs_attr_generation_counter;
1489
1490static unsigned long nfs_read_attr_generation_counter(void)
1491{
1492 return atomic_long_read(&nfs_attr_generation_counter);
1493}
1494
1495unsigned long nfs_inc_attr_generation_counter(void)
1496{
1497 return atomic_long_inc_return(&nfs_attr_generation_counter);
1498}
1499EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1500
1501void nfs_fattr_init(struct nfs_fattr *fattr)
1502{
1503 fattr->valid = 0;
1504 fattr->time_start = jiffies;
1505 fattr->gencount = nfs_inc_attr_generation_counter();
1506 fattr->owner_name = NULL;
1507 fattr->group_name = NULL;
1508}
1509EXPORT_SYMBOL_GPL(nfs_fattr_init);
1510
1511/**
1512 * nfs_fattr_set_barrier
1513 * @fattr: attributes
1514 *
1515 * Used to set a barrier after an attribute was updated. This
1516 * barrier ensures that older attributes from RPC calls that may
1517 * have raced with our update cannot clobber these new values.
1518 * Note that you are still responsible for ensuring that other
1519 * operations which change the attribute on the server do not
1520 * collide.
1521 */
1522void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1523{
1524 fattr->gencount = nfs_inc_attr_generation_counter();
1525}
1526
1527struct nfs_fattr *nfs_alloc_fattr(void)
1528{
1529 struct nfs_fattr *fattr;
1530
1531 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1532 if (fattr != NULL)
1533 nfs_fattr_init(fattr);
1534 return fattr;
1535}
1536EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1537
1538struct nfs_fh *nfs_alloc_fhandle(void)
1539{
1540 struct nfs_fh *fh;
1541
1542 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1543 if (fh != NULL)
1544 fh->size = 0;
1545 return fh;
1546}
1547EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1548
1549#ifdef NFS_DEBUG
1550/*
1551 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1552 * in the same way that wireshark does
1553 *
1554 * @fh: file handle
1555 *
1556 * For debugging only.
1557 */
1558u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1559{
1560 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1561 * not on the result */
1562 return nfs_fhandle_hash(fh);
1563}
1564EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1565
1566/*
1567 * _nfs_display_fhandle - display an NFS file handle on the console
1568 *
1569 * @fh: file handle to display
1570 * @caption: display caption
1571 *
1572 * For debugging only.
1573 */
1574void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1575{
1576 unsigned short i;
1577
1578 if (fh == NULL || fh->size == 0) {
1579 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1580 return;
1581 }
1582
1583 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1584 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1585 for (i = 0; i < fh->size; i += 16) {
1586 __be32 *pos = (__be32 *)&fh->data[i];
1587
1588 switch ((fh->size - i - 1) >> 2) {
1589 case 0:
1590 printk(KERN_DEFAULT " %08x\n",
1591 be32_to_cpup(pos));
1592 break;
1593 case 1:
1594 printk(KERN_DEFAULT " %08x %08x\n",
1595 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1596 break;
1597 case 2:
1598 printk(KERN_DEFAULT " %08x %08x %08x\n",
1599 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1600 be32_to_cpup(pos + 2));
1601 break;
1602 default:
1603 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1604 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1605 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1606 }
1607 }
1608}
1609EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1610#endif
1611
1612/**
1613 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1614 * @inode: pointer to inode
1615 * @fattr: attributes
1616 *
1617 * Attempt to divine whether or not an RPC call reply carrying stale
1618 * attributes got scheduled after another call carrying updated ones.
1619 *
1620 * To do so, the function first assumes that a more recent ctime means
1621 * that the attributes in fattr are newer, however it also attempt to
1622 * catch the case where ctime either didn't change, or went backwards
1623 * (if someone reset the clock on the server) by looking at whether
1624 * or not this RPC call was started after the inode was last updated.
1625 * Note also the check for wraparound of 'attr_gencount'
1626 *
1627 * The function returns 'true' if it thinks the attributes in 'fattr' are
1628 * more recent than the ones cached in the inode.
1629 *
1630 */
1631static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1632{
1633 const struct nfs_inode *nfsi = NFS_I(inode);
1634
1635 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1636 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1637}
1638
1639static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1640{
1641 int ret;
1642
1643 trace_nfs_refresh_inode_enter(inode);
1644
1645 if (nfs_inode_attrs_need_update(inode, fattr))
1646 ret = nfs_update_inode(inode, fattr);
1647 else
1648 ret = nfs_check_inode_attributes(inode, fattr);
1649
1650 trace_nfs_refresh_inode_exit(inode, ret);
1651 return ret;
1652}
1653
1654/**
1655 * nfs_refresh_inode - try to update the inode attribute cache
1656 * @inode: pointer to inode
1657 * @fattr: updated attributes
1658 *
1659 * Check that an RPC call that returned attributes has not overlapped with
1660 * other recent updates of the inode metadata, then decide whether it is
1661 * safe to do a full update of the inode attributes, or whether just to
1662 * call nfs_check_inode_attributes.
1663 */
1664int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1665{
1666 int status;
1667
1668 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1669 return 0;
1670 spin_lock(&inode->i_lock);
1671 status = nfs_refresh_inode_locked(inode, fattr);
1672 spin_unlock(&inode->i_lock);
1673
1674 return status;
1675}
1676EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1677
1678static int nfs_post_op_update_inode_locked(struct inode *inode,
1679 struct nfs_fattr *fattr, unsigned int invalid)
1680{
1681 if (S_ISDIR(inode->i_mode))
1682 invalid |= NFS_INO_INVALID_DATA;
1683 nfs_set_cache_invalid(inode, invalid);
1684 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1685 return 0;
1686 return nfs_refresh_inode_locked(inode, fattr);
1687}
1688
1689/**
1690 * nfs_post_op_update_inode - try to update the inode attribute cache
1691 * @inode: pointer to inode
1692 * @fattr: updated attributes
1693 *
1694 * After an operation that has changed the inode metadata, mark the
1695 * attribute cache as being invalid, then try to update it.
1696 *
1697 * NB: if the server didn't return any post op attributes, this
1698 * function will force the retrieval of attributes before the next
1699 * NFS request. Thus it should be used only for operations that
1700 * are expected to change one or more attributes, to avoid
1701 * unnecessary NFS requests and trips through nfs_update_inode().
1702 */
1703int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1704{
1705 int status;
1706
1707 spin_lock(&inode->i_lock);
1708 nfs_fattr_set_barrier(fattr);
1709 status = nfs_post_op_update_inode_locked(inode, fattr,
1710 NFS_INO_INVALID_CHANGE
1711 | NFS_INO_INVALID_CTIME
1712 | NFS_INO_REVAL_FORCED);
1713 spin_unlock(&inode->i_lock);
1714
1715 return status;
1716}
1717EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1718
1719/**
1720 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1721 * @inode: pointer to inode
1722 * @fattr: updated attributes
1723 *
1724 * After an operation that has changed the inode metadata, mark the
1725 * attribute cache as being invalid, then try to update it. Fake up
1726 * weak cache consistency data, if none exist.
1727 *
1728 * This function is mainly designed to be used by the ->write_done() functions.
1729 */
1730int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1731{
1732 int status;
1733
1734 /* Don't do a WCC update if these attributes are already stale */
1735 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1736 !nfs_inode_attrs_need_update(inode, fattr)) {
1737 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1738 | NFS_ATTR_FATTR_PRESIZE
1739 | NFS_ATTR_FATTR_PREMTIME
1740 | NFS_ATTR_FATTR_PRECTIME);
1741 goto out_noforce;
1742 }
1743 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1744 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1745 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1746 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1747 }
1748 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1749 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1750 fattr->pre_ctime = inode->i_ctime;
1751 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1752 }
1753 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1754 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1755 fattr->pre_mtime = inode->i_mtime;
1756 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1757 }
1758 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1759 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1760 fattr->pre_size = i_size_read(inode);
1761 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1762 }
1763out_noforce:
1764 status = nfs_post_op_update_inode_locked(inode, fattr,
1765 NFS_INO_INVALID_CHANGE
1766 | NFS_INO_INVALID_CTIME
1767 | NFS_INO_INVALID_MTIME);
1768 return status;
1769}
1770
1771/**
1772 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1773 * @inode: pointer to inode
1774 * @fattr: updated attributes
1775 *
1776 * After an operation that has changed the inode metadata, mark the
1777 * attribute cache as being invalid, then try to update it. Fake up
1778 * weak cache consistency data, if none exist.
1779 *
1780 * This function is mainly designed to be used by the ->write_done() functions.
1781 */
1782int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1783{
1784 int status;
1785
1786 spin_lock(&inode->i_lock);
1787 nfs_fattr_set_barrier(fattr);
1788 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1789 spin_unlock(&inode->i_lock);
1790 return status;
1791}
1792EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1793
1794
1795/*
1796 * Many nfs protocol calls return the new file attributes after
1797 * an operation. Here we update the inode to reflect the state
1798 * of the server's inode.
1799 *
1800 * This is a bit tricky because we have to make sure all dirty pages
1801 * have been sent off to the server before calling invalidate_inode_pages.
1802 * To make sure no other process adds more write requests while we try
1803 * our best to flush them, we make them sleep during the attribute refresh.
1804 *
1805 * A very similar scenario holds for the dir cache.
1806 */
1807static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1808{
1809 struct nfs_server *server;
1810 struct nfs_inode *nfsi = NFS_I(inode);
1811 loff_t cur_isize, new_isize;
1812 unsigned long invalid = 0;
1813 unsigned long now = jiffies;
1814 unsigned long save_cache_validity;
1815 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1816 bool cache_revalidated = true;
1817 bool attr_changed = false;
1818 bool have_delegation;
1819
1820 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1821 __func__, inode->i_sb->s_id, inode->i_ino,
1822 nfs_display_fhandle_hash(NFS_FH(inode)),
1823 atomic_read(&inode->i_count), fattr->valid);
1824
1825 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1826 /* Only a mounted-on-fileid? Just exit */
1827 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1828 return 0;
1829 /* Has the inode gone and changed behind our back? */
1830 } else if (nfsi->fileid != fattr->fileid) {
1831 /* Is this perhaps the mounted-on fileid? */
1832 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1833 nfsi->fileid == fattr->mounted_on_fileid)
1834 return 0;
1835 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1836 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1837 NFS_SERVER(inode)->nfs_client->cl_hostname,
1838 inode->i_sb->s_id, (long long)nfsi->fileid,
1839 (long long)fattr->fileid);
1840 goto out_err;
1841 }
1842
1843 /*
1844 * Make sure the inode's type hasn't changed.
1845 */
1846 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1847 /*
1848 * Big trouble! The inode has become a different object.
1849 */
1850 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1851 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1852 goto out_err;
1853 }
1854
1855 server = NFS_SERVER(inode);
1856 /* Update the fsid? */
1857 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1858 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1859 !IS_AUTOMOUNT(inode))
1860 server->fsid = fattr->fsid;
1861
1862 /* Save the delegation state before clearing cache_validity */
1863 have_delegation = nfs_have_delegated_attributes(inode);
1864
1865 /*
1866 * Update the read time so we don't revalidate too often.
1867 */
1868 nfsi->read_cache_jiffies = fattr->time_start;
1869
1870 save_cache_validity = nfsi->cache_validity;
1871 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1872 | NFS_INO_INVALID_ATIME
1873 | NFS_INO_REVAL_FORCED
1874 | NFS_INO_REVAL_PAGECACHE);
1875
1876 /* Do atomic weak cache consistency updates */
1877 nfs_wcc_update_inode(inode, fattr);
1878
1879 if (pnfs_layoutcommit_outstanding(inode)) {
1880 nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1881 cache_revalidated = false;
1882 }
1883
1884 /* More cache consistency checks */
1885 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1886 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1887 /* Could it be a race with writeback? */
1888 if (!(have_writers || have_delegation)) {
1889 invalid |= NFS_INO_INVALID_DATA
1890 | NFS_INO_INVALID_ACCESS
1891 | NFS_INO_INVALID_ACL;
1892 /* Force revalidate of all attributes */
1893 save_cache_validity |= NFS_INO_INVALID_CTIME
1894 | NFS_INO_INVALID_MTIME
1895 | NFS_INO_INVALID_SIZE
1896 | NFS_INO_INVALID_OTHER;
1897 if (S_ISDIR(inode->i_mode))
1898 nfs_force_lookup_revalidate(inode);
1899 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1900 inode->i_sb->s_id,
1901 inode->i_ino);
1902 } else if (!have_delegation)
1903 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1904 inode_set_iversion_raw(inode, fattr->change_attr);
1905 attr_changed = true;
1906 }
1907 } else {
1908 nfsi->cache_validity |= save_cache_validity &
1909 (NFS_INO_INVALID_CHANGE
1910 | NFS_INO_REVAL_PAGECACHE
1911 | NFS_INO_REVAL_FORCED);
1912 cache_revalidated = false;
1913 }
1914
1915 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1916 inode->i_mtime = fattr->mtime;
1917 } else if (server->caps & NFS_CAP_MTIME) {
1918 nfsi->cache_validity |= save_cache_validity &
1919 (NFS_INO_INVALID_MTIME
1920 | NFS_INO_REVAL_FORCED);
1921 cache_revalidated = false;
1922 }
1923
1924 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1925 inode->i_ctime = fattr->ctime;
1926 } else if (server->caps & NFS_CAP_CTIME) {
1927 nfsi->cache_validity |= save_cache_validity &
1928 (NFS_INO_INVALID_CTIME
1929 | NFS_INO_REVAL_FORCED);
1930 cache_revalidated = false;
1931 }
1932
1933 /* Check if our cached file size is stale */
1934 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1935 new_isize = nfs_size_to_loff_t(fattr->size);
1936 cur_isize = i_size_read(inode);
1937 if (new_isize != cur_isize && !have_delegation) {
1938 /* Do we perhaps have any outstanding writes, or has
1939 * the file grown beyond our last write? */
1940 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1941 i_size_write(inode, new_isize);
1942 if (!have_writers)
1943 invalid |= NFS_INO_INVALID_DATA;
1944 attr_changed = true;
1945 }
1946 dprintk("NFS: isize change on server for file %s/%ld "
1947 "(%Ld to %Ld)\n",
1948 inode->i_sb->s_id,
1949 inode->i_ino,
1950 (long long)cur_isize,
1951 (long long)new_isize);
1952 }
1953 } else {
1954 nfsi->cache_validity |= save_cache_validity &
1955 (NFS_INO_INVALID_SIZE
1956 | NFS_INO_REVAL_PAGECACHE
1957 | NFS_INO_REVAL_FORCED);
1958 cache_revalidated = false;
1959 }
1960
1961
1962 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1963 inode->i_atime = fattr->atime;
1964 else if (server->caps & NFS_CAP_ATIME) {
1965 nfsi->cache_validity |= save_cache_validity &
1966 (NFS_INO_INVALID_ATIME
1967 | NFS_INO_REVAL_FORCED);
1968 cache_revalidated = false;
1969 }
1970
1971 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1972 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1973 umode_t newmode = inode->i_mode & S_IFMT;
1974 newmode |= fattr->mode & S_IALLUGO;
1975 inode->i_mode = newmode;
1976 invalid |= NFS_INO_INVALID_ACCESS
1977 | NFS_INO_INVALID_ACL;
1978 attr_changed = true;
1979 }
1980 } else if (server->caps & NFS_CAP_MODE) {
1981 nfsi->cache_validity |= save_cache_validity &
1982 (NFS_INO_INVALID_OTHER
1983 | NFS_INO_REVAL_FORCED);
1984 cache_revalidated = false;
1985 }
1986
1987 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1988 if (!uid_eq(inode->i_uid, fattr->uid)) {
1989 invalid |= NFS_INO_INVALID_ACCESS
1990 | NFS_INO_INVALID_ACL;
1991 inode->i_uid = fattr->uid;
1992 attr_changed = true;
1993 }
1994 } else if (server->caps & NFS_CAP_OWNER) {
1995 nfsi->cache_validity |= save_cache_validity &
1996 (NFS_INO_INVALID_OTHER
1997 | NFS_INO_REVAL_FORCED);
1998 cache_revalidated = false;
1999 }
2000
2001 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2002 if (!gid_eq(inode->i_gid, fattr->gid)) {
2003 invalid |= NFS_INO_INVALID_ACCESS
2004 | NFS_INO_INVALID_ACL;
2005 inode->i_gid = fattr->gid;
2006 attr_changed = true;
2007 }
2008 } else if (server->caps & NFS_CAP_OWNER_GROUP) {
2009 nfsi->cache_validity |= save_cache_validity &
2010 (NFS_INO_INVALID_OTHER
2011 | NFS_INO_REVAL_FORCED);
2012 cache_revalidated = false;
2013 }
2014
2015 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2016 if (inode->i_nlink != fattr->nlink) {
2017 if (S_ISDIR(inode->i_mode))
2018 invalid |= NFS_INO_INVALID_DATA;
2019 set_nlink(inode, fattr->nlink);
2020 attr_changed = true;
2021 }
2022 } else if (server->caps & NFS_CAP_NLINK) {
2023 nfsi->cache_validity |= save_cache_validity &
2024 (NFS_INO_INVALID_OTHER
2025 | NFS_INO_REVAL_FORCED);
2026 cache_revalidated = false;
2027 }
2028
2029 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2030 /*
2031 * report the blocks in 512byte units
2032 */
2033 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2034 } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2035 inode->i_blocks = fattr->du.nfs2.blocks;
2036 else
2037 cache_revalidated = false;
2038
2039 /* Update attrtimeo value if we're out of the unstable period */
2040 if (attr_changed) {
2041 invalid &= ~NFS_INO_INVALID_ATTR;
2042 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2043 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2044 nfsi->attrtimeo_timestamp = now;
2045 /* Set barrier to be more recent than all outstanding updates */
2046 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2047 } else {
2048 if (cache_revalidated) {
2049 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2050 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2051 nfsi->attrtimeo <<= 1;
2052 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2053 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2054 }
2055 nfsi->attrtimeo_timestamp = now;
2056 }
2057 /* Set the barrier to be more recent than this fattr */
2058 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
2059 nfsi->attr_gencount = fattr->gencount;
2060 }
2061
2062 /* Don't invalidate the data if we were to blame */
2063 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2064 || S_ISLNK(inode->i_mode)))
2065 invalid &= ~NFS_INO_INVALID_DATA;
2066 nfs_set_cache_invalid(inode, invalid);
2067
2068 return 0;
2069 out_err:
2070 /*
2071 * No need to worry about unhashing the dentry, as the
2072 * lookup validation will know that the inode is bad.
2073 * (But we fall through to invalidate the caches.)
2074 */
2075 nfs_set_inode_stale_locked(inode);
2076 return -ESTALE;
2077}
2078
2079struct inode *nfs_alloc_inode(struct super_block *sb)
2080{
2081 struct nfs_inode *nfsi;
2082 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2083 if (!nfsi)
2084 return NULL;
2085 nfsi->flags = 0UL;
2086 nfsi->cache_validity = 0UL;
2087#if IS_ENABLED(CONFIG_NFS_V4)
2088 nfsi->nfs4_acl = NULL;
2089#endif /* CONFIG_NFS_V4 */
2090 return &nfsi->vfs_inode;
2091}
2092EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2093
2094void nfs_free_inode(struct inode *inode)
2095{
2096 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2097}
2098EXPORT_SYMBOL_GPL(nfs_free_inode);
2099
2100static inline void nfs4_init_once(struct nfs_inode *nfsi)
2101{
2102#if IS_ENABLED(CONFIG_NFS_V4)
2103 INIT_LIST_HEAD(&nfsi->open_states);
2104 nfsi->delegation = NULL;
2105 init_rwsem(&nfsi->rwsem);
2106 nfsi->layout = NULL;
2107#endif
2108}
2109
2110static void init_once(void *foo)
2111{
2112 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2113
2114 inode_init_once(&nfsi->vfs_inode);
2115 INIT_LIST_HEAD(&nfsi->open_files);
2116 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2117 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2118 INIT_LIST_HEAD(&nfsi->commit_info.list);
2119 atomic_long_set(&nfsi->nrequests, 0);
2120 atomic_long_set(&nfsi->commit_info.ncommit, 0);
2121 atomic_set(&nfsi->commit_info.rpcs_out, 0);
2122 init_rwsem(&nfsi->rmdir_sem);
2123 mutex_init(&nfsi->commit_mutex);
2124 nfs4_init_once(nfsi);
2125 nfsi->cache_change_attribute = 0;
2126}
2127
2128static int __init nfs_init_inodecache(void)
2129{
2130 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2131 sizeof(struct nfs_inode),
2132 0, (SLAB_RECLAIM_ACCOUNT|
2133 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2134 init_once);
2135 if (nfs_inode_cachep == NULL)
2136 return -ENOMEM;
2137
2138 return 0;
2139}
2140
2141static void nfs_destroy_inodecache(void)
2142{
2143 /*
2144 * Make sure all delayed rcu free inodes are flushed before we
2145 * destroy cache.
2146 */
2147 rcu_barrier();
2148 kmem_cache_destroy(nfs_inode_cachep);
2149}
2150
2151struct workqueue_struct *nfsiod_workqueue;
2152EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2153
2154/*
2155 * start up the nfsiod workqueue
2156 */
2157static int nfsiod_start(void)
2158{
2159 struct workqueue_struct *wq;
2160 dprintk("RPC: creating workqueue nfsiod\n");
2161 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2162 if (wq == NULL)
2163 return -ENOMEM;
2164 nfsiod_workqueue = wq;
2165 return 0;
2166}
2167
2168/*
2169 * Destroy the nfsiod workqueue
2170 */
2171static void nfsiod_stop(void)
2172{
2173 struct workqueue_struct *wq;
2174
2175 wq = nfsiod_workqueue;
2176 if (wq == NULL)
2177 return;
2178 nfsiod_workqueue = NULL;
2179 destroy_workqueue(wq);
2180}
2181
2182unsigned int nfs_net_id;
2183EXPORT_SYMBOL_GPL(nfs_net_id);
2184
2185static int nfs_net_init(struct net *net)
2186{
2187 nfs_clients_init(net);
2188 return nfs_fs_proc_net_init(net);
2189}
2190
2191static void nfs_net_exit(struct net *net)
2192{
2193 nfs_fs_proc_net_exit(net);
2194 nfs_clients_exit(net);
2195}
2196
2197static struct pernet_operations nfs_net_ops = {
2198 .init = nfs_net_init,
2199 .exit = nfs_net_exit,
2200 .id = &nfs_net_id,
2201 .size = sizeof(struct nfs_net),
2202};
2203
2204/*
2205 * Initialize NFS
2206 */
2207static int __init init_nfs_fs(void)
2208{
2209 int err;
2210
2211 err = nfs_sysfs_init();
2212 if (err < 0)
2213 goto out10;
2214
2215 err = register_pernet_subsys(&nfs_net_ops);
2216 if (err < 0)
2217 goto out9;
2218
2219 err = nfs_fscache_register();
2220 if (err < 0)
2221 goto out8;
2222
2223 err = nfsiod_start();
2224 if (err)
2225 goto out7;
2226
2227 err = nfs_fs_proc_init();
2228 if (err)
2229 goto out6;
2230
2231 err = nfs_init_nfspagecache();
2232 if (err)
2233 goto out5;
2234
2235 err = nfs_init_inodecache();
2236 if (err)
2237 goto out4;
2238
2239 err = nfs_init_readpagecache();
2240 if (err)
2241 goto out3;
2242
2243 err = nfs_init_writepagecache();
2244 if (err)
2245 goto out2;
2246
2247 err = nfs_init_directcache();
2248 if (err)
2249 goto out1;
2250
2251 rpc_proc_register(&init_net, &nfs_rpcstat);
2252
2253 err = register_nfs_fs();
2254 if (err)
2255 goto out0;
2256
2257 return 0;
2258out0:
2259 rpc_proc_unregister(&init_net, "nfs");
2260 nfs_destroy_directcache();
2261out1:
2262 nfs_destroy_writepagecache();
2263out2:
2264 nfs_destroy_readpagecache();
2265out3:
2266 nfs_destroy_inodecache();
2267out4:
2268 nfs_destroy_nfspagecache();
2269out5:
2270 nfs_fs_proc_exit();
2271out6:
2272 nfsiod_stop();
2273out7:
2274 nfs_fscache_unregister();
2275out8:
2276 unregister_pernet_subsys(&nfs_net_ops);
2277out9:
2278 nfs_sysfs_exit();
2279out10:
2280 return err;
2281}
2282
2283static void __exit exit_nfs_fs(void)
2284{
2285 nfs_destroy_directcache();
2286 nfs_destroy_writepagecache();
2287 nfs_destroy_readpagecache();
2288 nfs_destroy_inodecache();
2289 nfs_destroy_nfspagecache();
2290 nfs_fscache_unregister();
2291 unregister_pernet_subsys(&nfs_net_ops);
2292 rpc_proc_unregister(&init_net, "nfs");
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)
Linux 6.x block layer and io_uring tree(s)