Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/nfs/inode.c
4 *
5 * Copyright (C) 1992 Rick Sladkey
6 *
7 * nfs inode and superblock handling functions
8 *
9 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
10 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 *
12 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
13 * J.S.Peatfield@damtp.cam.ac.uk
14 *
15 */
16
17#include <linux/module.h>
18#include <linux/init.h>
19#include <linux/sched/signal.h>
20#include <linux/time.h>
21#include <linux/kernel.h>
22#include <linux/mm.h>
23#include <linux/string.h>
24#include <linux/stat.h>
25#include <linux/errno.h>
26#include <linux/unistd.h>
27#include <linux/sunrpc/clnt.h>
28#include <linux/sunrpc/stats.h>
29#include <linux/sunrpc/metrics.h>
30#include <linux/nfs_fs.h>
31#include <linux/nfs_mount.h>
32#include <linux/nfs4_mount.h>
33#include <linux/lockd/bind.h>
34#include <linux/seq_file.h>
35#include <linux/mount.h>
36#include <linux/vfs.h>
37#include <linux/inet.h>
38#include <linux/nfs_xdr.h>
39#include <linux/slab.h>
40#include <linux/compat.h>
41#include <linux/freezer.h>
42#include <linux/uaccess.h>
43#include <linux/iversion.h>
44
45#include "nfs4_fs.h"
46#include "callback.h"
47#include "delegation.h"
48#include "iostat.h"
49#include "internal.h"
50#include "fscache.h"
51#include "pnfs.h"
52#include "nfs.h"
53#include "netns.h"
54#include "sysfs.h"
55
56#include "nfstrace.h"
57
58#define NFSDBG_FACILITY NFSDBG_VFS
59
60#define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
61
62/* Default is to see 64-bit inode numbers */
63static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
64
65static int nfs_update_inode(struct inode *, struct nfs_fattr *);
66
67static struct kmem_cache * nfs_inode_cachep;
68
69static inline unsigned long
70nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71{
72 return nfs_fileid_to_ino_t(fattr->fileid);
73}
74
75static int nfs_wait_killable(int mode)
76{
77 freezable_schedule_unsafe();
78 if (signal_pending_state(mode, current))
79 return -ERESTARTSYS;
80 return 0;
81}
82
83int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
84{
85 return nfs_wait_killable(mode);
86}
87EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
88
89/**
90 * nfs_compat_user_ino64 - returns the user-visible inode number
91 * @fileid: 64-bit fileid
92 *
93 * This function returns a 32-bit inode number if the boot parameter
94 * nfs.enable_ino64 is zero.
95 */
96u64 nfs_compat_user_ino64(u64 fileid)
97{
98#ifdef CONFIG_COMPAT
99 compat_ulong_t ino;
100#else
101 unsigned long ino;
102#endif
103
104 if (enable_ino64)
105 return fileid;
106 ino = fileid;
107 if (sizeof(ino) < sizeof(fileid))
108 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
109 return ino;
110}
111
112int nfs_drop_inode(struct inode *inode)
113{
114 return NFS_STALE(inode) || generic_drop_inode(inode);
115}
116EXPORT_SYMBOL_GPL(nfs_drop_inode);
117
118void nfs_clear_inode(struct inode *inode)
119{
120 /*
121 * The following should never happen...
122 */
123 WARN_ON_ONCE(nfs_have_writebacks(inode));
124 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
125 nfs_zap_acl_cache(inode);
126 nfs_access_zap_cache(inode);
127 nfs_fscache_clear_inode(inode);
128}
129EXPORT_SYMBOL_GPL(nfs_clear_inode);
130
131void nfs_evict_inode(struct inode *inode)
132{
133 truncate_inode_pages_final(&inode->i_data);
134 clear_inode(inode);
135 nfs_clear_inode(inode);
136}
137
138int nfs_sync_inode(struct inode *inode)
139{
140 inode_dio_wait(inode);
141 return nfs_wb_all(inode);
142}
143EXPORT_SYMBOL_GPL(nfs_sync_inode);
144
145/**
146 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
147 * @mapping: pointer to struct address_space
148 */
149int nfs_sync_mapping(struct address_space *mapping)
150{
151 int ret = 0;
152
153 if (mapping->nrpages != 0) {
154 unmap_mapping_range(mapping, 0, 0, 0);
155 ret = nfs_wb_all(mapping->host);
156 }
157 return ret;
158}
159
160static int nfs_attribute_timeout(struct inode *inode)
161{
162 struct nfs_inode *nfsi = NFS_I(inode);
163
164 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
165}
166
167static bool nfs_check_cache_flags_invalid(struct inode *inode,
168 unsigned long flags)
169{
170 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
171
172 return (cache_validity & flags) != 0;
173}
174
175bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
176{
177 if (nfs_check_cache_flags_invalid(inode, flags))
178 return true;
179 return nfs_attribute_cache_expired(inode);
180}
181EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
182
183#ifdef CONFIG_NFS_V4_2
184static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
185{
186 return nfsi->xattr_cache != NULL;
187}
188#else
189static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
190{
191 return false;
192}
193#endif
194
195void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
196{
197 struct nfs_inode *nfsi = NFS_I(inode);
198 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
199
200 if (have_delegation) {
201 if (!(flags & NFS_INO_REVAL_FORCED))
202 flags &= ~(NFS_INO_INVALID_MODE |
203 NFS_INO_INVALID_OTHER |
204 NFS_INO_INVALID_XATTR);
205 flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
206 } else if (flags & NFS_INO_REVAL_PAGECACHE)
207 flags |= NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE;
208
209 if (!nfs_has_xattr_cache(nfsi))
210 flags &= ~NFS_INO_INVALID_XATTR;
211 if (flags & NFS_INO_INVALID_DATA)
212 nfs_fscache_invalidate(inode);
213 if (inode->i_mapping->nrpages == 0)
214 flags &= ~(NFS_INO_INVALID_DATA|NFS_INO_DATA_INVAL_DEFER);
215 flags &= ~(NFS_INO_REVAL_PAGECACHE | NFS_INO_REVAL_FORCED);
216 nfsi->cache_validity |= flags;
217}
218EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
219
220/*
221 * Invalidate the local caches
222 */
223static void nfs_zap_caches_locked(struct inode *inode)
224{
225 struct nfs_inode *nfsi = NFS_I(inode);
226 int mode = inode->i_mode;
227
228 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
229
230 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
231 nfsi->attrtimeo_timestamp = jiffies;
232
233 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
234 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
235 | NFS_INO_INVALID_DATA
236 | NFS_INO_INVALID_ACCESS
237 | NFS_INO_INVALID_ACL
238 | NFS_INO_INVALID_XATTR
239 | NFS_INO_REVAL_PAGECACHE);
240 } else
241 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
242 | NFS_INO_INVALID_ACCESS
243 | NFS_INO_INVALID_ACL
244 | NFS_INO_INVALID_XATTR
245 | NFS_INO_REVAL_PAGECACHE);
246 nfs_zap_label_cache_locked(nfsi);
247}
248
249void nfs_zap_caches(struct inode *inode)
250{
251 spin_lock(&inode->i_lock);
252 nfs_zap_caches_locked(inode);
253 spin_unlock(&inode->i_lock);
254}
255
256void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
257{
258 if (mapping->nrpages != 0) {
259 spin_lock(&inode->i_lock);
260 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
261 spin_unlock(&inode->i_lock);
262 }
263}
264
265void nfs_zap_acl_cache(struct inode *inode)
266{
267 void (*clear_acl_cache)(struct inode *);
268
269 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
270 if (clear_acl_cache != NULL)
271 clear_acl_cache(inode);
272 spin_lock(&inode->i_lock);
273 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
274 spin_unlock(&inode->i_lock);
275}
276EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
277
278void nfs_invalidate_atime(struct inode *inode)
279{
280 spin_lock(&inode->i_lock);
281 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
282 spin_unlock(&inode->i_lock);
283}
284EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
285
286/*
287 * Invalidate, but do not unhash, the inode.
288 * NB: must be called with inode->i_lock held!
289 */
290static void nfs_set_inode_stale_locked(struct inode *inode)
291{
292 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
293 nfs_zap_caches_locked(inode);
294 trace_nfs_set_inode_stale(inode);
295}
296
297void nfs_set_inode_stale(struct inode *inode)
298{
299 spin_lock(&inode->i_lock);
300 nfs_set_inode_stale_locked(inode);
301 spin_unlock(&inode->i_lock);
302}
303
304struct nfs_find_desc {
305 struct nfs_fh *fh;
306 struct nfs_fattr *fattr;
307};
308
309/*
310 * In NFSv3 we can have 64bit inode numbers. In order to support
311 * this, and re-exported directories (also seen in NFSv2)
312 * we are forced to allow 2 different inodes to have the same
313 * i_ino.
314 */
315static int
316nfs_find_actor(struct inode *inode, void *opaque)
317{
318 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
319 struct nfs_fh *fh = desc->fh;
320 struct nfs_fattr *fattr = desc->fattr;
321
322 if (NFS_FILEID(inode) != fattr->fileid)
323 return 0;
324 if (inode_wrong_type(inode, fattr->mode))
325 return 0;
326 if (nfs_compare_fh(NFS_FH(inode), fh))
327 return 0;
328 if (is_bad_inode(inode) || NFS_STALE(inode))
329 return 0;
330 return 1;
331}
332
333static int
334nfs_init_locked(struct inode *inode, void *opaque)
335{
336 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
337 struct nfs_fattr *fattr = desc->fattr;
338
339 set_nfs_fileid(inode, fattr->fileid);
340 inode->i_mode = fattr->mode;
341 nfs_copy_fh(NFS_FH(inode), desc->fh);
342 return 0;
343}
344
345#ifdef CONFIG_NFS_V4_SECURITY_LABEL
346static void nfs_clear_label_invalid(struct inode *inode)
347{
348 spin_lock(&inode->i_lock);
349 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
350 spin_unlock(&inode->i_lock);
351}
352
353void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
354 struct nfs4_label *label)
355{
356 int error;
357
358 if (label == NULL)
359 return;
360
361 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
362 error = security_inode_notifysecctx(inode, label->label,
363 label->len);
364 if (error)
365 printk(KERN_ERR "%s() %s %d "
366 "security_inode_notifysecctx() %d\n",
367 __func__,
368 (char *)label->label,
369 label->len, error);
370 nfs_clear_label_invalid(inode);
371 }
372}
373
374struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
375{
376 struct nfs4_label *label = NULL;
377 int minor_version = server->nfs_client->cl_minorversion;
378
379 if (minor_version < 2)
380 return label;
381
382 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
383 return label;
384
385 label = kzalloc(sizeof(struct nfs4_label), flags);
386 if (label == NULL)
387 return ERR_PTR(-ENOMEM);
388
389 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
390 if (label->label == NULL) {
391 kfree(label);
392 return ERR_PTR(-ENOMEM);
393 }
394 label->len = NFS4_MAXLABELLEN;
395
396 return label;
397}
398EXPORT_SYMBOL_GPL(nfs4_label_alloc);
399#else
400void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
401 struct nfs4_label *label)
402{
403}
404#endif
405EXPORT_SYMBOL_GPL(nfs_setsecurity);
406
407/* Search for inode identified by fh, fileid and i_mode in inode cache. */
408struct inode *
409nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
410{
411 struct nfs_find_desc desc = {
412 .fh = fh,
413 .fattr = fattr,
414 };
415 struct inode *inode;
416 unsigned long hash;
417
418 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
419 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
420 return NULL;
421
422 hash = nfs_fattr_to_ino_t(fattr);
423 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
424
425 dprintk("%s: returning %p\n", __func__, inode);
426 return inode;
427}
428
429/*
430 * This is our front-end to iget that looks up inodes by file handle
431 * instead of inode number.
432 */
433struct inode *
434nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
435{
436 struct nfs_find_desc desc = {
437 .fh = fh,
438 .fattr = fattr
439 };
440 struct inode *inode = ERR_PTR(-ENOENT);
441 u64 fattr_supported = NFS_SB(sb)->fattr_valid;
442 unsigned long hash;
443
444 nfs_attr_check_mountpoint(sb, fattr);
445
446 if (nfs_attr_use_mounted_on_fileid(fattr))
447 fattr->fileid = fattr->mounted_on_fileid;
448 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
449 goto out_no_inode;
450 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
451 goto out_no_inode;
452
453 hash = nfs_fattr_to_ino_t(fattr);
454
455 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
456 if (inode == NULL) {
457 inode = ERR_PTR(-ENOMEM);
458 goto out_no_inode;
459 }
460
461 if (inode->i_state & I_NEW) {
462 struct nfs_inode *nfsi = NFS_I(inode);
463 unsigned long now = jiffies;
464
465 /* We set i_ino for the few things that still rely on it,
466 * such as stat(2) */
467 inode->i_ino = hash;
468
469 /* We can't support update_atime(), since the server will reset it */
470 inode->i_flags |= S_NOATIME|S_NOCMTIME;
471 inode->i_mode = fattr->mode;
472 nfsi->cache_validity = 0;
473 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
474 && (fattr_supported & NFS_ATTR_FATTR_MODE))
475 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
476 /* Why so? Because we want revalidate for devices/FIFOs, and
477 * that's precisely what we have in nfs_file_inode_operations.
478 */
479 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
480 if (S_ISREG(inode->i_mode)) {
481 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
482 inode->i_data.a_ops = &nfs_file_aops;
483 } else if (S_ISDIR(inode->i_mode)) {
484 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
485 inode->i_fop = &nfs_dir_operations;
486 inode->i_data.a_ops = &nfs_dir_aops;
487 /* Deal with crossing mountpoints */
488 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
489 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
490 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
491 inode->i_op = &nfs_referral_inode_operations;
492 else
493 inode->i_op = &nfs_mountpoint_inode_operations;
494 inode->i_fop = NULL;
495 inode->i_flags |= S_AUTOMOUNT;
496 }
497 } else if (S_ISLNK(inode->i_mode)) {
498 inode->i_op = &nfs_symlink_inode_operations;
499 inode_nohighmem(inode);
500 } else
501 init_special_inode(inode, inode->i_mode, fattr->rdev);
502
503 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
504 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
505 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
506 inode_set_iversion_raw(inode, 0);
507 inode->i_size = 0;
508 clear_nlink(inode);
509 inode->i_uid = make_kuid(&init_user_ns, -2);
510 inode->i_gid = make_kgid(&init_user_ns, -2);
511 inode->i_blocks = 0;
512 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
513 nfsi->write_io = 0;
514 nfsi->read_io = 0;
515
516 nfsi->read_cache_jiffies = fattr->time_start;
517 nfsi->attr_gencount = fattr->gencount;
518 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
519 inode->i_atime = fattr->atime;
520 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
521 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
522 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
523 inode->i_mtime = fattr->mtime;
524 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
525 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
526 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
527 inode->i_ctime = fattr->ctime;
528 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
529 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
530 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
531 inode_set_iversion_raw(inode, fattr->change_attr);
532 else
533 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
534 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
535 inode->i_size = nfs_size_to_loff_t(fattr->size);
536 else
537 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
538 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
539 set_nlink(inode, fattr->nlink);
540 else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
541 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
542 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
543 inode->i_uid = fattr->uid;
544 else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
545 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
546 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
547 inode->i_gid = fattr->gid;
548 else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
549 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
550 if (nfs_server_capable(inode, NFS_CAP_XATTR))
551 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
552 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
553 inode->i_blocks = fattr->du.nfs2.blocks;
554 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
555 fattr->size != 0)
556 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
557 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
558 /*
559 * report the blocks in 512byte units
560 */
561 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
562 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
563 fattr->size != 0)
564 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
565
566 nfs_setsecurity(inode, fattr, label);
567
568 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
569 nfsi->attrtimeo_timestamp = now;
570 nfsi->access_cache = RB_ROOT;
571
572 nfs_fscache_init_inode(inode);
573
574 unlock_new_inode(inode);
575 } else {
576 int err = nfs_refresh_inode(inode, fattr);
577 if (err < 0) {
578 iput(inode);
579 inode = ERR_PTR(err);
580 goto out_no_inode;
581 }
582 }
583 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
584 inode->i_sb->s_id,
585 (unsigned long long)NFS_FILEID(inode),
586 nfs_display_fhandle_hash(fh),
587 atomic_read(&inode->i_count));
588
589out:
590 return inode;
591
592out_no_inode:
593 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
594 goto out;
595}
596EXPORT_SYMBOL_GPL(nfs_fhget);
597
598#define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN)
599
600int
601nfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
602 struct iattr *attr)
603{
604 struct inode *inode = d_inode(dentry);
605 struct nfs_fattr *fattr;
606 int error = 0;
607
608 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
609
610 /* skip mode change if it's just for clearing setuid/setgid */
611 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
612 attr->ia_valid &= ~ATTR_MODE;
613
614 if (attr->ia_valid & ATTR_SIZE) {
615 BUG_ON(!S_ISREG(inode->i_mode));
616
617 error = inode_newsize_ok(inode, attr->ia_size);
618 if (error)
619 return error;
620
621 if (attr->ia_size == i_size_read(inode))
622 attr->ia_valid &= ~ATTR_SIZE;
623 }
624
625 /* Optimization: if the end result is no change, don't RPC */
626 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
627 return 0;
628
629 trace_nfs_setattr_enter(inode);
630
631 /* Write all dirty data */
632 if (S_ISREG(inode->i_mode))
633 nfs_sync_inode(inode);
634
635 fattr = nfs_alloc_fattr();
636 if (fattr == NULL) {
637 error = -ENOMEM;
638 goto out;
639 }
640
641 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
642 if (error == 0)
643 error = nfs_refresh_inode(inode, fattr);
644 nfs_free_fattr(fattr);
645out:
646 trace_nfs_setattr_exit(inode, error);
647 return error;
648}
649EXPORT_SYMBOL_GPL(nfs_setattr);
650
651/**
652 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
653 * @inode: inode of the file used
654 * @offset: file offset to start truncating
655 *
656 * This is a copy of the common vmtruncate, but with the locking
657 * corrected to take into account the fact that NFS requires
658 * inode->i_size to be updated under the inode->i_lock.
659 * Note: must be called with inode->i_lock held!
660 */
661static int nfs_vmtruncate(struct inode * inode, loff_t offset)
662{
663 int err;
664
665 err = inode_newsize_ok(inode, offset);
666 if (err)
667 goto out;
668
669 i_size_write(inode, offset);
670 /* Optimisation */
671 if (offset == 0)
672 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
673 NFS_INO_DATA_INVAL_DEFER);
674 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
675
676 spin_unlock(&inode->i_lock);
677 truncate_pagecache(inode, offset);
678 spin_lock(&inode->i_lock);
679out:
680 return err;
681}
682
683/**
684 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
685 * @inode: pointer to struct inode
686 * @attr: pointer to struct iattr
687 * @fattr: pointer to struct nfs_fattr
688 *
689 * Note: we do this in the *proc.c in order to ensure that
690 * it works for things like exclusive creates too.
691 */
692void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
693 struct nfs_fattr *fattr)
694{
695 /* Barrier: bump the attribute generation count. */
696 nfs_fattr_set_barrier(fattr);
697
698 spin_lock(&inode->i_lock);
699 NFS_I(inode)->attr_gencount = fattr->gencount;
700 if ((attr->ia_valid & ATTR_SIZE) != 0) {
701 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME |
702 NFS_INO_INVALID_BLOCKS);
703 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
704 nfs_vmtruncate(inode, attr->ia_size);
705 }
706 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
707 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
708 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
709 inode->i_mode & S_ISUID)
710 inode->i_mode &= ~S_ISUID;
711 if ((attr->ia_valid & ATTR_KILL_SGID) != 0 &&
712 (inode->i_mode & (S_ISGID | S_IXGRP)) ==
713 (S_ISGID | S_IXGRP))
714 inode->i_mode &= ~S_ISGID;
715 if ((attr->ia_valid & ATTR_MODE) != 0) {
716 int mode = attr->ia_mode & S_IALLUGO;
717 mode |= inode->i_mode & ~S_IALLUGO;
718 inode->i_mode = mode;
719 }
720 if ((attr->ia_valid & ATTR_UID) != 0)
721 inode->i_uid = attr->ia_uid;
722 if ((attr->ia_valid & ATTR_GID) != 0)
723 inode->i_gid = attr->ia_gid;
724 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
725 inode->i_ctime = fattr->ctime;
726 else
727 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
728 | NFS_INO_INVALID_CTIME);
729 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
730 | NFS_INO_INVALID_ACL);
731 }
732 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
733 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
734 | NFS_INO_INVALID_CTIME);
735 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
736 inode->i_atime = fattr->atime;
737 else if (attr->ia_valid & ATTR_ATIME_SET)
738 inode->i_atime = attr->ia_atime;
739 else
740 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
741
742 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
743 inode->i_ctime = fattr->ctime;
744 else
745 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
746 | NFS_INO_INVALID_CTIME);
747 }
748 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
749 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
750 | NFS_INO_INVALID_CTIME);
751 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
752 inode->i_mtime = fattr->mtime;
753 else if (attr->ia_valid & ATTR_MTIME_SET)
754 inode->i_mtime = attr->ia_mtime;
755 else
756 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
757
758 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
759 inode->i_ctime = fattr->ctime;
760 else
761 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
762 | NFS_INO_INVALID_CTIME);
763 }
764 if (fattr->valid)
765 nfs_update_inode(inode, fattr);
766 spin_unlock(&inode->i_lock);
767}
768EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
769
770static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
771{
772 struct dentry *parent;
773
774 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
775 return;
776 parent = dget_parent(dentry);
777 nfs_force_use_readdirplus(d_inode(parent));
778 dput(parent);
779}
780
781static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
782{
783 struct dentry *parent;
784
785 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
786 return;
787 parent = dget_parent(dentry);
788 nfs_advise_use_readdirplus(d_inode(parent));
789 dput(parent);
790}
791
792static u32 nfs_get_valid_attrmask(struct inode *inode)
793{
794 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
795 u32 reply_mask = STATX_INO | STATX_TYPE;
796
797 if (!(cache_validity & NFS_INO_INVALID_ATIME))
798 reply_mask |= STATX_ATIME;
799 if (!(cache_validity & NFS_INO_INVALID_CTIME))
800 reply_mask |= STATX_CTIME;
801 if (!(cache_validity & NFS_INO_INVALID_MTIME))
802 reply_mask |= STATX_MTIME;
803 if (!(cache_validity & NFS_INO_INVALID_SIZE))
804 reply_mask |= STATX_SIZE;
805 if (!(cache_validity & NFS_INO_INVALID_NLINK))
806 reply_mask |= STATX_NLINK;
807 if (!(cache_validity & NFS_INO_INVALID_MODE))
808 reply_mask |= STATX_MODE;
809 if (!(cache_validity & NFS_INO_INVALID_OTHER))
810 reply_mask |= STATX_UID | STATX_GID;
811 if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
812 reply_mask |= STATX_BLOCKS;
813 return reply_mask;
814}
815
816int nfs_getattr(struct user_namespace *mnt_userns, const struct path *path,
817 struct kstat *stat, u32 request_mask, unsigned int query_flags)
818{
819 struct inode *inode = d_inode(path->dentry);
820 struct nfs_server *server = NFS_SERVER(inode);
821 unsigned long cache_validity;
822 int err = 0;
823 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
824 bool do_update = false;
825
826 trace_nfs_getattr_enter(inode);
827
828 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
829 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
830 STATX_INO | STATX_SIZE | STATX_BLOCKS;
831
832 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
833 nfs_readdirplus_parent_cache_hit(path->dentry);
834 goto out_no_revalidate;
835 }
836
837 /* Flush out writes to the server in order to update c/mtime. */
838 if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
839 S_ISREG(inode->i_mode)) {
840 err = filemap_write_and_wait(inode->i_mapping);
841 if (err)
842 goto out;
843 }
844
845 /*
846 * We may force a getattr if the user cares about atime.
847 *
848 * Note that we only have to check the vfsmount flags here:
849 * - NFS always sets S_NOATIME by so checking it would give a
850 * bogus result
851 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
852 * no point in checking those.
853 */
854 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
855 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
856 request_mask &= ~STATX_ATIME;
857
858 /* Is the user requesting attributes that might need revalidation? */
859 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
860 STATX_MTIME|STATX_UID|STATX_GID|
861 STATX_SIZE|STATX_BLOCKS)))
862 goto out_no_revalidate;
863
864 /* Check whether the cached attributes are stale */
865 do_update |= force_sync || nfs_attribute_cache_expired(inode);
866 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
867 do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
868 if (request_mask & STATX_ATIME)
869 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
870 if (request_mask & STATX_CTIME)
871 do_update |= cache_validity & NFS_INO_INVALID_CTIME;
872 if (request_mask & STATX_MTIME)
873 do_update |= cache_validity & NFS_INO_INVALID_MTIME;
874 if (request_mask & STATX_SIZE)
875 do_update |= cache_validity & NFS_INO_INVALID_SIZE;
876 if (request_mask & STATX_NLINK)
877 do_update |= cache_validity & NFS_INO_INVALID_NLINK;
878 if (request_mask & STATX_MODE)
879 do_update |= cache_validity & NFS_INO_INVALID_MODE;
880 if (request_mask & (STATX_UID | STATX_GID))
881 do_update |= cache_validity & NFS_INO_INVALID_OTHER;
882 if (request_mask & STATX_BLOCKS)
883 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
884
885 if (do_update) {
886 /* Update the attribute cache */
887 if (!(server->flags & NFS_MOUNT_NOAC))
888 nfs_readdirplus_parent_cache_miss(path->dentry);
889 else
890 nfs_readdirplus_parent_cache_hit(path->dentry);
891 err = __nfs_revalidate_inode(server, inode);
892 if (err)
893 goto out;
894 } else
895 nfs_readdirplus_parent_cache_hit(path->dentry);
896out_no_revalidate:
897 /* Only return attributes that were revalidated. */
898 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
899
900 generic_fillattr(&init_user_ns, inode, stat);
901 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
902 if (S_ISDIR(inode->i_mode))
903 stat->blksize = NFS_SERVER(inode)->dtsize;
904out:
905 trace_nfs_getattr_exit(inode, err);
906 return err;
907}
908EXPORT_SYMBOL_GPL(nfs_getattr);
909
910static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
911{
912 refcount_set(&l_ctx->count, 1);
913 l_ctx->lockowner = current->files;
914 INIT_LIST_HEAD(&l_ctx->list);
915 atomic_set(&l_ctx->io_count, 0);
916}
917
918static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
919{
920 struct nfs_lock_context *pos;
921
922 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
923 if (pos->lockowner != current->files)
924 continue;
925 if (refcount_inc_not_zero(&pos->count))
926 return pos;
927 }
928 return NULL;
929}
930
931struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
932{
933 struct nfs_lock_context *res, *new = NULL;
934 struct inode *inode = d_inode(ctx->dentry);
935
936 rcu_read_lock();
937 res = __nfs_find_lock_context(ctx);
938 rcu_read_unlock();
939 if (res == NULL) {
940 new = kmalloc(sizeof(*new), GFP_KERNEL);
941 if (new == NULL)
942 return ERR_PTR(-ENOMEM);
943 nfs_init_lock_context(new);
944 spin_lock(&inode->i_lock);
945 res = __nfs_find_lock_context(ctx);
946 if (res == NULL) {
947 new->open_context = get_nfs_open_context(ctx);
948 if (new->open_context) {
949 list_add_tail_rcu(&new->list,
950 &ctx->lock_context.list);
951 res = new;
952 new = NULL;
953 } else
954 res = ERR_PTR(-EBADF);
955 }
956 spin_unlock(&inode->i_lock);
957 kfree(new);
958 }
959 return res;
960}
961EXPORT_SYMBOL_GPL(nfs_get_lock_context);
962
963void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
964{
965 struct nfs_open_context *ctx = l_ctx->open_context;
966 struct inode *inode = d_inode(ctx->dentry);
967
968 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
969 return;
970 list_del_rcu(&l_ctx->list);
971 spin_unlock(&inode->i_lock);
972 put_nfs_open_context(ctx);
973 kfree_rcu(l_ctx, rcu_head);
974}
975EXPORT_SYMBOL_GPL(nfs_put_lock_context);
976
977/**
978 * nfs_close_context - Common close_context() routine NFSv2/v3
979 * @ctx: pointer to context
980 * @is_sync: is this a synchronous close
981 *
982 * Ensure that the attributes are up to date if we're mounted
983 * with close-to-open semantics and we have cached data that will
984 * need to be revalidated on open.
985 */
986void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
987{
988 struct nfs_inode *nfsi;
989 struct inode *inode;
990
991 if (!(ctx->mode & FMODE_WRITE))
992 return;
993 if (!is_sync)
994 return;
995 inode = d_inode(ctx->dentry);
996 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
997 return;
998 nfsi = NFS_I(inode);
999 if (inode->i_mapping->nrpages == 0)
1000 return;
1001 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1002 return;
1003 if (!list_empty(&nfsi->open_files))
1004 return;
1005 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1006 return;
1007 nfs_revalidate_inode(inode,
1008 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1009}
1010EXPORT_SYMBOL_GPL(nfs_close_context);
1011
1012struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1013 fmode_t f_mode,
1014 struct file *filp)
1015{
1016 struct nfs_open_context *ctx;
1017
1018 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
1019 if (!ctx)
1020 return ERR_PTR(-ENOMEM);
1021 nfs_sb_active(dentry->d_sb);
1022 ctx->dentry = dget(dentry);
1023 if (filp)
1024 ctx->cred = get_cred(filp->f_cred);
1025 else
1026 ctx->cred = get_current_cred();
1027 ctx->ll_cred = NULL;
1028 ctx->state = NULL;
1029 ctx->mode = f_mode;
1030 ctx->flags = 0;
1031 ctx->error = 0;
1032 ctx->flock_owner = (fl_owner_t)filp;
1033 nfs_init_lock_context(&ctx->lock_context);
1034 ctx->lock_context.open_context = ctx;
1035 INIT_LIST_HEAD(&ctx->list);
1036 ctx->mdsthreshold = NULL;
1037 return ctx;
1038}
1039EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1040
1041struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1042{
1043 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1044 return ctx;
1045 return NULL;
1046}
1047EXPORT_SYMBOL_GPL(get_nfs_open_context);
1048
1049static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1050{
1051 struct inode *inode = d_inode(ctx->dentry);
1052 struct super_block *sb = ctx->dentry->d_sb;
1053
1054 if (!refcount_dec_and_test(&ctx->lock_context.count))
1055 return;
1056 if (!list_empty(&ctx->list)) {
1057 spin_lock(&inode->i_lock);
1058 list_del_rcu(&ctx->list);
1059 spin_unlock(&inode->i_lock);
1060 }
1061 if (inode != NULL)
1062 NFS_PROTO(inode)->close_context(ctx, is_sync);
1063 put_cred(ctx->cred);
1064 dput(ctx->dentry);
1065 nfs_sb_deactive(sb);
1066 put_rpccred(ctx->ll_cred);
1067 kfree(ctx->mdsthreshold);
1068 kfree_rcu(ctx, rcu_head);
1069}
1070
1071void put_nfs_open_context(struct nfs_open_context *ctx)
1072{
1073 __put_nfs_open_context(ctx, 0);
1074}
1075EXPORT_SYMBOL_GPL(put_nfs_open_context);
1076
1077static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1078{
1079 __put_nfs_open_context(ctx, 1);
1080}
1081
1082/*
1083 * Ensure that mmap has a recent RPC credential for use when writing out
1084 * shared pages
1085 */
1086void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1087{
1088 struct inode *inode = d_inode(ctx->dentry);
1089 struct nfs_inode *nfsi = NFS_I(inode);
1090
1091 spin_lock(&inode->i_lock);
1092 if (list_empty(&nfsi->open_files) &&
1093 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1094 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1095 NFS_INO_REVAL_FORCED);
1096 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1097 spin_unlock(&inode->i_lock);
1098}
1099EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1100
1101void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1102{
1103 filp->private_data = get_nfs_open_context(ctx);
1104 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1105 if (list_empty(&ctx->list))
1106 nfs_inode_attach_open_context(ctx);
1107}
1108EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1109
1110/*
1111 * Given an inode, search for an open context with the desired characteristics
1112 */
1113struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1114{
1115 struct nfs_inode *nfsi = NFS_I(inode);
1116 struct nfs_open_context *pos, *ctx = NULL;
1117
1118 rcu_read_lock();
1119 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1120 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1121 continue;
1122 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1123 continue;
1124 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1125 continue;
1126 ctx = get_nfs_open_context(pos);
1127 if (ctx)
1128 break;
1129 }
1130 rcu_read_unlock();
1131 return ctx;
1132}
1133
1134void nfs_file_clear_open_context(struct file *filp)
1135{
1136 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1137
1138 if (ctx) {
1139 struct inode *inode = d_inode(ctx->dentry);
1140
1141 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1142 /*
1143 * We fatal error on write before. Try to writeback
1144 * every page again.
1145 */
1146 if (ctx->error < 0)
1147 invalidate_inode_pages2(inode->i_mapping);
1148 filp->private_data = NULL;
1149 put_nfs_open_context_sync(ctx);
1150 }
1151}
1152
1153/*
1154 * These allocate and release file read/write context information.
1155 */
1156int nfs_open(struct inode *inode, struct file *filp)
1157{
1158 struct nfs_open_context *ctx;
1159
1160 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1161 if (IS_ERR(ctx))
1162 return PTR_ERR(ctx);
1163 nfs_file_set_open_context(filp, ctx);
1164 put_nfs_open_context(ctx);
1165 nfs_fscache_open_file(inode, filp);
1166 return 0;
1167}
1168EXPORT_SYMBOL_GPL(nfs_open);
1169
1170/*
1171 * This function is called whenever some part of NFS notices that
1172 * the cached attributes have to be refreshed.
1173 */
1174int
1175__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1176{
1177 int status = -ESTALE;
1178 struct nfs4_label *label = NULL;
1179 struct nfs_fattr *fattr = NULL;
1180 struct nfs_inode *nfsi = NFS_I(inode);
1181
1182 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1183 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1184
1185 trace_nfs_revalidate_inode_enter(inode);
1186
1187 if (is_bad_inode(inode))
1188 goto out;
1189 if (NFS_STALE(inode))
1190 goto out;
1191
1192 /* pNFS: Attributes aren't updated until we layoutcommit */
1193 if (S_ISREG(inode->i_mode)) {
1194 status = pnfs_sync_inode(inode, false);
1195 if (status)
1196 goto out;
1197 }
1198
1199 status = -ENOMEM;
1200 fattr = nfs_alloc_fattr();
1201 if (fattr == NULL)
1202 goto out;
1203
1204 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1205
1206 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1207 if (IS_ERR(label)) {
1208 status = PTR_ERR(label);
1209 goto out;
1210 }
1211
1212 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1213 label, inode);
1214 if (status != 0) {
1215 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1216 inode->i_sb->s_id,
1217 (unsigned long long)NFS_FILEID(inode), status);
1218 switch (status) {
1219 case -ETIMEDOUT:
1220 /* A soft timeout occurred. Use cached information? */
1221 if (server->flags & NFS_MOUNT_SOFTREVAL)
1222 status = 0;
1223 break;
1224 case -ESTALE:
1225 if (!S_ISDIR(inode->i_mode))
1226 nfs_set_inode_stale(inode);
1227 else
1228 nfs_zap_caches(inode);
1229 }
1230 goto err_out;
1231 }
1232
1233 status = nfs_refresh_inode(inode, fattr);
1234 if (status) {
1235 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1236 inode->i_sb->s_id,
1237 (unsigned long long)NFS_FILEID(inode), status);
1238 goto err_out;
1239 }
1240
1241 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1242 nfs_zap_acl_cache(inode);
1243
1244 nfs_setsecurity(inode, fattr, label);
1245
1246 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1247 inode->i_sb->s_id,
1248 (unsigned long long)NFS_FILEID(inode));
1249
1250err_out:
1251 nfs4_label_free(label);
1252out:
1253 nfs_free_fattr(fattr);
1254 trace_nfs_revalidate_inode_exit(inode, status);
1255 return status;
1256}
1257
1258int nfs_attribute_cache_expired(struct inode *inode)
1259{
1260 if (nfs_have_delegated_attributes(inode))
1261 return 0;
1262 return nfs_attribute_timeout(inode);
1263}
1264
1265/**
1266 * nfs_revalidate_inode - Revalidate the inode attributes
1267 * @inode: pointer to inode struct
1268 * @flags: cache flags to check
1269 *
1270 * Updates inode attribute information by retrieving the data from the server.
1271 */
1272int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1273{
1274 if (!nfs_check_cache_invalid(inode, flags))
1275 return NFS_STALE(inode) ? -ESTALE : 0;
1276 return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1277}
1278EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1279
1280static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1281{
1282 int ret;
1283
1284 if (mapping->nrpages != 0) {
1285 if (S_ISREG(inode->i_mode)) {
1286 ret = nfs_sync_mapping(mapping);
1287 if (ret < 0)
1288 return ret;
1289 }
1290 ret = invalidate_inode_pages2(mapping);
1291 if (ret < 0)
1292 return ret;
1293 }
1294 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1295 nfs_fscache_wait_on_invalidate(inode);
1296
1297 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1298 inode->i_sb->s_id,
1299 (unsigned long long)NFS_FILEID(inode));
1300 return 0;
1301}
1302
1303/**
1304 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1305 * @mapping: pointer to mapping
1306 *
1307 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1308 */
1309int nfs_clear_invalid_mapping(struct address_space *mapping)
1310{
1311 struct inode *inode = mapping->host;
1312 struct nfs_inode *nfsi = NFS_I(inode);
1313 unsigned long *bitlock = &nfsi->flags;
1314 int ret = 0;
1315
1316 /*
1317 * We must clear NFS_INO_INVALID_DATA first to ensure that
1318 * invalidations that come in while we're shooting down the mappings
1319 * are respected. But, that leaves a race window where one revalidator
1320 * can clear the flag, and then another checks it before the mapping
1321 * gets invalidated. Fix that by serializing access to this part of
1322 * the function.
1323 *
1324 * At the same time, we need to allow other tasks to see whether we
1325 * might be in the middle of invalidating the pages, so we only set
1326 * the bit lock here if it looks like we're going to be doing that.
1327 */
1328 for (;;) {
1329 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1330 nfs_wait_bit_killable, TASK_KILLABLE);
1331 if (ret)
1332 goto out;
1333 spin_lock(&inode->i_lock);
1334 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1335 spin_unlock(&inode->i_lock);
1336 continue;
1337 }
1338 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1339 break;
1340 spin_unlock(&inode->i_lock);
1341 goto out;
1342 }
1343
1344 set_bit(NFS_INO_INVALIDATING, bitlock);
1345 smp_wmb();
1346 nfsi->cache_validity &=
1347 ~(NFS_INO_INVALID_DATA | NFS_INO_DATA_INVAL_DEFER);
1348 spin_unlock(&inode->i_lock);
1349 trace_nfs_invalidate_mapping_enter(inode);
1350 ret = nfs_invalidate_mapping(inode, mapping);
1351 trace_nfs_invalidate_mapping_exit(inode, ret);
1352
1353 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1354 smp_mb__after_atomic();
1355 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1356out:
1357 return ret;
1358}
1359
1360bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1361{
1362 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1363 NFS_STALE(inode);
1364}
1365
1366int nfs_revalidate_mapping_rcu(struct inode *inode)
1367{
1368 struct nfs_inode *nfsi = NFS_I(inode);
1369 unsigned long *bitlock = &nfsi->flags;
1370 int ret = 0;
1371
1372 if (IS_SWAPFILE(inode))
1373 goto out;
1374 if (nfs_mapping_need_revalidate_inode(inode)) {
1375 ret = -ECHILD;
1376 goto out;
1377 }
1378 spin_lock(&inode->i_lock);
1379 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1380 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1381 ret = -ECHILD;
1382 spin_unlock(&inode->i_lock);
1383out:
1384 return ret;
1385}
1386
1387/**
1388 * nfs_revalidate_mapping - Revalidate the pagecache
1389 * @inode: pointer to host inode
1390 * @mapping: pointer to mapping
1391 */
1392int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1393{
1394 /* swapfiles are not supposed to be shared. */
1395 if (IS_SWAPFILE(inode))
1396 return 0;
1397
1398 if (nfs_mapping_need_revalidate_inode(inode)) {
1399 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1400 if (ret < 0)
1401 return ret;
1402 }
1403
1404 return nfs_clear_invalid_mapping(mapping);
1405}
1406
1407static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1408{
1409 struct inode *inode = &nfsi->vfs_inode;
1410
1411 if (!S_ISREG(inode->i_mode))
1412 return false;
1413 if (list_empty(&nfsi->open_files))
1414 return false;
1415 return inode_is_open_for_write(inode);
1416}
1417
1418static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1419{
1420 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1421}
1422
1423static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1424{
1425 struct timespec64 ts;
1426
1427 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1428 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1429 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1430 inode_set_iversion_raw(inode, fattr->change_attr);
1431 if (S_ISDIR(inode->i_mode))
1432 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1433 else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1434 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1435 }
1436 /* If we have atomic WCC data, we may update some attributes */
1437 ts = inode->i_ctime;
1438 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1439 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1440 && timespec64_equal(&ts, &fattr->pre_ctime)) {
1441 inode->i_ctime = fattr->ctime;
1442 }
1443
1444 ts = inode->i_mtime;
1445 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1446 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1447 && timespec64_equal(&ts, &fattr->pre_mtime)) {
1448 inode->i_mtime = fattr->mtime;
1449 if (S_ISDIR(inode->i_mode))
1450 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1451 }
1452 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1453 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1454 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1455 && !nfs_have_writebacks(inode)) {
1456 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1457 }
1458}
1459
1460/**
1461 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1462 * @inode: pointer to inode
1463 * @fattr: updated attributes
1464 *
1465 * Verifies the attribute cache. If we have just changed the attributes,
1466 * so that fattr carries weak cache consistency data, then it may
1467 * also update the ctime/mtime/change_attribute.
1468 */
1469static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1470{
1471 struct nfs_inode *nfsi = NFS_I(inode);
1472 loff_t cur_size, new_isize;
1473 unsigned long invalid = 0;
1474 struct timespec64 ts;
1475
1476 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1477 return 0;
1478
1479 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1480 /* Only a mounted-on-fileid? Just exit */
1481 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1482 return 0;
1483 /* Has the inode gone and changed behind our back? */
1484 } else if (nfsi->fileid != fattr->fileid) {
1485 /* Is this perhaps the mounted-on fileid? */
1486 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1487 nfsi->fileid == fattr->mounted_on_fileid)
1488 return 0;
1489 return -ESTALE;
1490 }
1491 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1492 return -ESTALE;
1493
1494
1495 if (!nfs_file_has_buffered_writers(nfsi)) {
1496 /* Verify a few of the more important attributes */
1497 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1498 invalid |= NFS_INO_INVALID_CHANGE;
1499
1500 ts = inode->i_mtime;
1501 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1502 invalid |= NFS_INO_INVALID_MTIME;
1503
1504 ts = inode->i_ctime;
1505 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1506 invalid |= NFS_INO_INVALID_CTIME;
1507
1508 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1509 cur_size = i_size_read(inode);
1510 new_isize = nfs_size_to_loff_t(fattr->size);
1511 if (cur_size != new_isize)
1512 invalid |= NFS_INO_INVALID_SIZE;
1513 }
1514 }
1515
1516 /* Have any file permissions changed? */
1517 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1518 invalid |= NFS_INO_INVALID_MODE;
1519 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1520 invalid |= NFS_INO_INVALID_OTHER;
1521 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1522 invalid |= NFS_INO_INVALID_OTHER;
1523
1524 /* Has the link count changed? */
1525 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1526 invalid |= NFS_INO_INVALID_NLINK;
1527
1528 ts = inode->i_atime;
1529 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1530 invalid |= NFS_INO_INVALID_ATIME;
1531
1532 if (invalid != 0)
1533 nfs_set_cache_invalid(inode, invalid);
1534
1535 nfsi->read_cache_jiffies = fattr->time_start;
1536 return 0;
1537}
1538
1539static atomic_long_t nfs_attr_generation_counter;
1540
1541static unsigned long nfs_read_attr_generation_counter(void)
1542{
1543 return atomic_long_read(&nfs_attr_generation_counter);
1544}
1545
1546unsigned long nfs_inc_attr_generation_counter(void)
1547{
1548 return atomic_long_inc_return(&nfs_attr_generation_counter);
1549}
1550EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1551
1552void nfs_fattr_init(struct nfs_fattr *fattr)
1553{
1554 fattr->valid = 0;
1555 fattr->time_start = jiffies;
1556 fattr->gencount = nfs_inc_attr_generation_counter();
1557 fattr->owner_name = NULL;
1558 fattr->group_name = NULL;
1559}
1560EXPORT_SYMBOL_GPL(nfs_fattr_init);
1561
1562/**
1563 * nfs_fattr_set_barrier
1564 * @fattr: attributes
1565 *
1566 * Used to set a barrier after an attribute was updated. This
1567 * barrier ensures that older attributes from RPC calls that may
1568 * have raced with our update cannot clobber these new values.
1569 * Note that you are still responsible for ensuring that other
1570 * operations which change the attribute on the server do not
1571 * collide.
1572 */
1573void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1574{
1575 fattr->gencount = nfs_inc_attr_generation_counter();
1576}
1577
1578struct nfs_fattr *nfs_alloc_fattr(void)
1579{
1580 struct nfs_fattr *fattr;
1581
1582 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1583 if (fattr != NULL)
1584 nfs_fattr_init(fattr);
1585 return fattr;
1586}
1587EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1588
1589struct nfs_fh *nfs_alloc_fhandle(void)
1590{
1591 struct nfs_fh *fh;
1592
1593 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1594 if (fh != NULL)
1595 fh->size = 0;
1596 return fh;
1597}
1598EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1599
1600#ifdef NFS_DEBUG
1601/*
1602 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1603 * in the same way that wireshark does
1604 *
1605 * @fh: file handle
1606 *
1607 * For debugging only.
1608 */
1609u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1610{
1611 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1612 * not on the result */
1613 return nfs_fhandle_hash(fh);
1614}
1615EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1616
1617/*
1618 * _nfs_display_fhandle - display an NFS file handle on the console
1619 *
1620 * @fh: file handle to display
1621 * @caption: display caption
1622 *
1623 * For debugging only.
1624 */
1625void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1626{
1627 unsigned short i;
1628
1629 if (fh == NULL || fh->size == 0) {
1630 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1631 return;
1632 }
1633
1634 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1635 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1636 for (i = 0; i < fh->size; i += 16) {
1637 __be32 *pos = (__be32 *)&fh->data[i];
1638
1639 switch ((fh->size - i - 1) >> 2) {
1640 case 0:
1641 printk(KERN_DEFAULT " %08x\n",
1642 be32_to_cpup(pos));
1643 break;
1644 case 1:
1645 printk(KERN_DEFAULT " %08x %08x\n",
1646 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1647 break;
1648 case 2:
1649 printk(KERN_DEFAULT " %08x %08x %08x\n",
1650 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1651 be32_to_cpup(pos + 2));
1652 break;
1653 default:
1654 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1655 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1656 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1657 }
1658 }
1659}
1660EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1661#endif
1662
1663/**
1664 * nfs_inode_attrs_cmp_generic - compare attributes
1665 * @fattr: attributes
1666 * @inode: pointer to inode
1667 *
1668 * Attempt to divine whether or not an RPC call reply carrying stale
1669 * attributes got scheduled after another call carrying updated ones.
1670 * Note also the check for wraparound of 'attr_gencount'
1671 *
1672 * The function returns '1' if it thinks the attributes in @fattr are
1673 * more recent than the ones cached in @inode. Otherwise it returns
1674 * the value '0'.
1675 */
1676static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1677 const struct inode *inode)
1678{
1679 unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1680
1681 return (long)(fattr->gencount - attr_gencount) > 0 ||
1682 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1683}
1684
1685/**
1686 * nfs_inode_attrs_cmp_monotonic - compare attributes
1687 * @fattr: attributes
1688 * @inode: pointer to inode
1689 *
1690 * Attempt to divine whether or not an RPC call reply carrying stale
1691 * attributes got scheduled after another call carrying updated ones.
1692 *
1693 * We assume that the server observes monotonic semantics for
1694 * the change attribute, so a larger value means that the attributes in
1695 * @fattr are more recent, in which case the function returns the
1696 * value '1'.
1697 * A return value of '0' indicates no measurable change
1698 * A return value of '-1' means that the attributes in @inode are
1699 * more recent.
1700 */
1701static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1702 const struct inode *inode)
1703{
1704 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1705 if (diff > 0)
1706 return 1;
1707 return diff == 0 ? 0 : -1;
1708}
1709
1710/**
1711 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1712 * @fattr: attributes
1713 * @inode: pointer to inode
1714 *
1715 * Attempt to divine whether or not an RPC call reply carrying stale
1716 * attributes got scheduled after another call carrying updated ones.
1717 *
1718 * We assume that the server observes strictly monotonic semantics for
1719 * the change attribute, so a larger value means that the attributes in
1720 * @fattr are more recent, in which case the function returns the
1721 * value '1'.
1722 * A return value of '-1' means that the attributes in @inode are
1723 * more recent or unchanged.
1724 */
1725static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1726 const struct inode *inode)
1727{
1728 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1729}
1730
1731/**
1732 * nfs_inode_attrs_cmp - compare attributes
1733 * @fattr: attributes
1734 * @inode: pointer to inode
1735 *
1736 * This function returns '1' if it thinks the attributes in @fattr are
1737 * more recent than the ones cached in @inode. It returns '-1' if
1738 * the attributes in @inode are more recent than the ones in @fattr,
1739 * and it returns 0 if not sure.
1740 */
1741static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1742 const struct inode *inode)
1743{
1744 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1745 return 1;
1746 switch (NFS_SERVER(inode)->change_attr_type) {
1747 case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1748 break;
1749 case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1750 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1751 break;
1752 return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1753 default:
1754 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1755 break;
1756 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1757 }
1758 return 0;
1759}
1760
1761/**
1762 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1763 * @fattr: attributes
1764 * @inode: pointer to inode
1765 *
1766 * Returns '1' if the last attribute update left the inode cached
1767 * attributes in a partially unrevalidated state, and @fattr
1768 * matches the change attribute of that partial update.
1769 * Otherwise returns '0'.
1770 */
1771static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1772 const struct inode *inode)
1773{
1774 const unsigned long check_valid =
1775 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1776 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1777 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1778 NFS_INO_INVALID_NLINK;
1779 unsigned long cache_validity = NFS_I(inode)->cache_validity;
1780
1781 if (!(cache_validity & NFS_INO_INVALID_CHANGE) &&
1782 (cache_validity & check_valid) != 0 &&
1783 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1784 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1785 return 1;
1786 return 0;
1787}
1788
1789static int nfs_refresh_inode_locked(struct inode *inode,
1790 struct nfs_fattr *fattr)
1791{
1792 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1793 int ret = 0;
1794
1795 trace_nfs_refresh_inode_enter(inode);
1796
1797 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1798 ret = nfs_update_inode(inode, fattr);
1799 else if (attr_cmp == 0)
1800 ret = nfs_check_inode_attributes(inode, fattr);
1801
1802 trace_nfs_refresh_inode_exit(inode, ret);
1803 return ret;
1804}
1805
1806/**
1807 * nfs_refresh_inode - try to update the inode attribute cache
1808 * @inode: pointer to inode
1809 * @fattr: updated attributes
1810 *
1811 * Check that an RPC call that returned attributes has not overlapped with
1812 * other recent updates of the inode metadata, then decide whether it is
1813 * safe to do a full update of the inode attributes, or whether just to
1814 * call nfs_check_inode_attributes.
1815 */
1816int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1817{
1818 int status;
1819
1820 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1821 return 0;
1822 spin_lock(&inode->i_lock);
1823 status = nfs_refresh_inode_locked(inode, fattr);
1824 spin_unlock(&inode->i_lock);
1825
1826 return status;
1827}
1828EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1829
1830static int nfs_post_op_update_inode_locked(struct inode *inode,
1831 struct nfs_fattr *fattr, unsigned int invalid)
1832{
1833 if (S_ISDIR(inode->i_mode))
1834 invalid |= NFS_INO_INVALID_DATA;
1835 nfs_set_cache_invalid(inode, invalid);
1836 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1837 return 0;
1838 return nfs_refresh_inode_locked(inode, fattr);
1839}
1840
1841/**
1842 * nfs_post_op_update_inode - try to update the inode attribute cache
1843 * @inode: pointer to inode
1844 * @fattr: updated attributes
1845 *
1846 * After an operation that has changed the inode metadata, mark the
1847 * attribute cache as being invalid, then try to update it.
1848 *
1849 * NB: if the server didn't return any post op attributes, this
1850 * function will force the retrieval of attributes before the next
1851 * NFS request. Thus it should be used only for operations that
1852 * are expected to change one or more attributes, to avoid
1853 * unnecessary NFS requests and trips through nfs_update_inode().
1854 */
1855int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1856{
1857 int status;
1858
1859 spin_lock(&inode->i_lock);
1860 nfs_fattr_set_barrier(fattr);
1861 status = nfs_post_op_update_inode_locked(inode, fattr,
1862 NFS_INO_INVALID_CHANGE
1863 | NFS_INO_INVALID_CTIME
1864 | NFS_INO_REVAL_FORCED);
1865 spin_unlock(&inode->i_lock);
1866
1867 return status;
1868}
1869EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1870
1871/**
1872 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1873 * @inode: pointer to inode
1874 * @fattr: updated attributes
1875 *
1876 * After an operation that has changed the inode metadata, mark the
1877 * attribute cache as being invalid, then try to update it. Fake up
1878 * weak cache consistency data, if none exist.
1879 *
1880 * This function is mainly designed to be used by the ->write_done() functions.
1881 */
1882int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1883{
1884 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1885 int status;
1886
1887 /* Don't do a WCC update if these attributes are already stale */
1888 if (attr_cmp < 0)
1889 return 0;
1890 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
1891 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1892 | NFS_ATTR_FATTR_PRESIZE
1893 | NFS_ATTR_FATTR_PREMTIME
1894 | NFS_ATTR_FATTR_PRECTIME);
1895 goto out_noforce;
1896 }
1897 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1898 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1899 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1900 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1901 }
1902 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1903 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1904 fattr->pre_ctime = inode->i_ctime;
1905 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1906 }
1907 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1908 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1909 fattr->pre_mtime = inode->i_mtime;
1910 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1911 }
1912 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1913 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1914 fattr->pre_size = i_size_read(inode);
1915 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1916 }
1917out_noforce:
1918 status = nfs_post_op_update_inode_locked(inode, fattr,
1919 NFS_INO_INVALID_CHANGE
1920 | NFS_INO_INVALID_CTIME
1921 | NFS_INO_INVALID_MTIME
1922 | NFS_INO_INVALID_BLOCKS);
1923 return status;
1924}
1925
1926/**
1927 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1928 * @inode: pointer to inode
1929 * @fattr: updated attributes
1930 *
1931 * After an operation that has changed the inode metadata, mark the
1932 * attribute cache as being invalid, then try to update it. Fake up
1933 * weak cache consistency data, if none exist.
1934 *
1935 * This function is mainly designed to be used by the ->write_done() functions.
1936 */
1937int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1938{
1939 int status;
1940
1941 spin_lock(&inode->i_lock);
1942 nfs_fattr_set_barrier(fattr);
1943 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1944 spin_unlock(&inode->i_lock);
1945 return status;
1946}
1947EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1948
1949
1950/*
1951 * Many nfs protocol calls return the new file attributes after
1952 * an operation. Here we update the inode to reflect the state
1953 * of the server's inode.
1954 *
1955 * This is a bit tricky because we have to make sure all dirty pages
1956 * have been sent off to the server before calling invalidate_inode_pages.
1957 * To make sure no other process adds more write requests while we try
1958 * our best to flush them, we make them sleep during the attribute refresh.
1959 *
1960 * A very similar scenario holds for the dir cache.
1961 */
1962static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1963{
1964 struct nfs_server *server = NFS_SERVER(inode);
1965 struct nfs_inode *nfsi = NFS_I(inode);
1966 loff_t cur_isize, new_isize;
1967 u64 fattr_supported = server->fattr_valid;
1968 unsigned long invalid = 0;
1969 unsigned long now = jiffies;
1970 unsigned long save_cache_validity;
1971 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1972 bool cache_revalidated = true;
1973 bool attr_changed = false;
1974 bool have_delegation;
1975
1976 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1977 __func__, inode->i_sb->s_id, inode->i_ino,
1978 nfs_display_fhandle_hash(NFS_FH(inode)),
1979 atomic_read(&inode->i_count), fattr->valid);
1980
1981 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1982 /* Only a mounted-on-fileid? Just exit */
1983 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1984 return 0;
1985 /* Has the inode gone and changed behind our back? */
1986 } else if (nfsi->fileid != fattr->fileid) {
1987 /* Is this perhaps the mounted-on fileid? */
1988 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1989 nfsi->fileid == fattr->mounted_on_fileid)
1990 return 0;
1991 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1992 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1993 NFS_SERVER(inode)->nfs_client->cl_hostname,
1994 inode->i_sb->s_id, (long long)nfsi->fileid,
1995 (long long)fattr->fileid);
1996 goto out_err;
1997 }
1998
1999 /*
2000 * Make sure the inode's type hasn't changed.
2001 */
2002 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2003 /*
2004 * Big trouble! The inode has become a different object.
2005 */
2006 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2007 __func__, inode->i_ino, inode->i_mode, fattr->mode);
2008 goto out_err;
2009 }
2010
2011 /* Update the fsid? */
2012 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2013 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2014 !IS_AUTOMOUNT(inode))
2015 server->fsid = fattr->fsid;
2016
2017 /* Save the delegation state before clearing cache_validity */
2018 have_delegation = nfs_have_delegated_attributes(inode);
2019
2020 /*
2021 * Update the read time so we don't revalidate too often.
2022 */
2023 nfsi->read_cache_jiffies = fattr->time_start;
2024
2025 save_cache_validity = nfsi->cache_validity;
2026 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2027 | NFS_INO_INVALID_ATIME
2028 | NFS_INO_REVAL_FORCED
2029 | NFS_INO_INVALID_BLOCKS);
2030
2031 /* Do atomic weak cache consistency updates */
2032 nfs_wcc_update_inode(inode, fattr);
2033
2034 if (pnfs_layoutcommit_outstanding(inode)) {
2035 nfsi->cache_validity |=
2036 save_cache_validity &
2037 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2038 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2039 NFS_INO_INVALID_BLOCKS);
2040 cache_revalidated = false;
2041 }
2042
2043 /* More cache consistency checks */
2044 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2045 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2046 /* Could it be a race with writeback? */
2047 if (!(have_writers || have_delegation)) {
2048 invalid |= NFS_INO_INVALID_DATA
2049 | NFS_INO_INVALID_ACCESS
2050 | NFS_INO_INVALID_ACL
2051 | NFS_INO_INVALID_XATTR;
2052 /* Force revalidate of all attributes */
2053 save_cache_validity |= NFS_INO_INVALID_CTIME
2054 | NFS_INO_INVALID_MTIME
2055 | NFS_INO_INVALID_SIZE
2056 | NFS_INO_INVALID_BLOCKS
2057 | NFS_INO_INVALID_NLINK
2058 | NFS_INO_INVALID_MODE
2059 | NFS_INO_INVALID_OTHER;
2060 if (S_ISDIR(inode->i_mode))
2061 nfs_force_lookup_revalidate(inode);
2062 attr_changed = true;
2063 dprintk("NFS: change_attr change on server for file %s/%ld\n",
2064 inode->i_sb->s_id,
2065 inode->i_ino);
2066 } else if (!have_delegation)
2067 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
2068 inode_set_iversion_raw(inode, fattr->change_attr);
2069 }
2070 } else {
2071 nfsi->cache_validity |=
2072 save_cache_validity & NFS_INO_INVALID_CHANGE;
2073 if (!have_delegation ||
2074 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2075 cache_revalidated = false;
2076 }
2077
2078 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2079 inode->i_mtime = fattr->mtime;
2080 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2081 nfsi->cache_validity |=
2082 save_cache_validity & NFS_INO_INVALID_MTIME;
2083
2084 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2085 inode->i_ctime = fattr->ctime;
2086 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2087 nfsi->cache_validity |=
2088 save_cache_validity & NFS_INO_INVALID_CTIME;
2089
2090 /* Check if our cached file size is stale */
2091 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2092 new_isize = nfs_size_to_loff_t(fattr->size);
2093 cur_isize = i_size_read(inode);
2094 if (new_isize != cur_isize && !have_delegation) {
2095 /* Do we perhaps have any outstanding writes, or has
2096 * the file grown beyond our last write? */
2097 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2098 i_size_write(inode, new_isize);
2099 if (!have_writers)
2100 invalid |= NFS_INO_INVALID_DATA;
2101 }
2102 dprintk("NFS: isize change on server for file %s/%ld "
2103 "(%Ld to %Ld)\n",
2104 inode->i_sb->s_id,
2105 inode->i_ino,
2106 (long long)cur_isize,
2107 (long long)new_isize);
2108 }
2109 if (new_isize == 0 &&
2110 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2111 NFS_ATTR_FATTR_BLOCKS_USED))) {
2112 fattr->du.nfs3.used = 0;
2113 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2114 }
2115 } else
2116 nfsi->cache_validity |=
2117 save_cache_validity & NFS_INO_INVALID_SIZE;
2118
2119 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2120 inode->i_atime = fattr->atime;
2121 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2122 nfsi->cache_validity |=
2123 save_cache_validity & NFS_INO_INVALID_ATIME;
2124
2125 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2126 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2127 umode_t newmode = inode->i_mode & S_IFMT;
2128 newmode |= fattr->mode & S_IALLUGO;
2129 inode->i_mode = newmode;
2130 invalid |= NFS_INO_INVALID_ACCESS
2131 | NFS_INO_INVALID_ACL;
2132 }
2133 } else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2134 nfsi->cache_validity |=
2135 save_cache_validity & NFS_INO_INVALID_MODE;
2136
2137 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2138 if (!uid_eq(inode->i_uid, fattr->uid)) {
2139 invalid |= NFS_INO_INVALID_ACCESS
2140 | NFS_INO_INVALID_ACL;
2141 inode->i_uid = fattr->uid;
2142 }
2143 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2144 nfsi->cache_validity |=
2145 save_cache_validity & NFS_INO_INVALID_OTHER;
2146
2147 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2148 if (!gid_eq(inode->i_gid, fattr->gid)) {
2149 invalid |= NFS_INO_INVALID_ACCESS
2150 | NFS_INO_INVALID_ACL;
2151 inode->i_gid = fattr->gid;
2152 }
2153 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2154 nfsi->cache_validity |=
2155 save_cache_validity & NFS_INO_INVALID_OTHER;
2156
2157 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2158 if (inode->i_nlink != fattr->nlink) {
2159 if (S_ISDIR(inode->i_mode))
2160 invalid |= NFS_INO_INVALID_DATA;
2161 set_nlink(inode, fattr->nlink);
2162 }
2163 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2164 nfsi->cache_validity |=
2165 save_cache_validity & NFS_INO_INVALID_NLINK;
2166
2167 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2168 /*
2169 * report the blocks in 512byte units
2170 */
2171 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2172 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2173 nfsi->cache_validity |=
2174 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2175
2176 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2177 inode->i_blocks = fattr->du.nfs2.blocks;
2178 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2179 nfsi->cache_validity |=
2180 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2181
2182 /* Update attrtimeo value if we're out of the unstable period */
2183 if (attr_changed) {
2184 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2185 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2186 nfsi->attrtimeo_timestamp = now;
2187 /* Set barrier to be more recent than all outstanding updates */
2188 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2189 } else {
2190 if (cache_revalidated) {
2191 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2192 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2193 nfsi->attrtimeo <<= 1;
2194 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2195 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2196 }
2197 nfsi->attrtimeo_timestamp = now;
2198 }
2199 /* Set the barrier to be more recent than this fattr */
2200 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2201 nfsi->attr_gencount = fattr->gencount;
2202 }
2203
2204 /* Don't invalidate the data if we were to blame */
2205 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2206 || S_ISLNK(inode->i_mode)))
2207 invalid &= ~NFS_INO_INVALID_DATA;
2208 nfs_set_cache_invalid(inode, invalid);
2209
2210 return 0;
2211 out_err:
2212 /*
2213 * No need to worry about unhashing the dentry, as the
2214 * lookup validation will know that the inode is bad.
2215 * (But we fall through to invalidate the caches.)
2216 */
2217 nfs_set_inode_stale_locked(inode);
2218 return -ESTALE;
2219}
2220
2221struct inode *nfs_alloc_inode(struct super_block *sb)
2222{
2223 struct nfs_inode *nfsi;
2224 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2225 if (!nfsi)
2226 return NULL;
2227 nfsi->flags = 0UL;
2228 nfsi->cache_validity = 0UL;
2229#if IS_ENABLED(CONFIG_NFS_V4)
2230 nfsi->nfs4_acl = NULL;
2231#endif /* CONFIG_NFS_V4 */
2232#ifdef CONFIG_NFS_V4_2
2233 nfsi->xattr_cache = NULL;
2234#endif
2235 return &nfsi->vfs_inode;
2236}
2237EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2238
2239void nfs_free_inode(struct inode *inode)
2240{
2241 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2242}
2243EXPORT_SYMBOL_GPL(nfs_free_inode);
2244
2245static inline void nfs4_init_once(struct nfs_inode *nfsi)
2246{
2247#if IS_ENABLED(CONFIG_NFS_V4)
2248 INIT_LIST_HEAD(&nfsi->open_states);
2249 nfsi->delegation = NULL;
2250 init_rwsem(&nfsi->rwsem);
2251 nfsi->layout = NULL;
2252#endif
2253}
2254
2255static void init_once(void *foo)
2256{
2257 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2258
2259 inode_init_once(&nfsi->vfs_inode);
2260 INIT_LIST_HEAD(&nfsi->open_files);
2261 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2262 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2263 INIT_LIST_HEAD(&nfsi->commit_info.list);
2264 atomic_long_set(&nfsi->nrequests, 0);
2265 atomic_long_set(&nfsi->commit_info.ncommit, 0);
2266 atomic_set(&nfsi->commit_info.rpcs_out, 0);
2267 init_rwsem(&nfsi->rmdir_sem);
2268 mutex_init(&nfsi->commit_mutex);
2269 nfs4_init_once(nfsi);
2270 nfsi->cache_change_attribute = 0;
2271}
2272
2273static int __init nfs_init_inodecache(void)
2274{
2275 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2276 sizeof(struct nfs_inode),
2277 0, (SLAB_RECLAIM_ACCOUNT|
2278 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2279 init_once);
2280 if (nfs_inode_cachep == NULL)
2281 return -ENOMEM;
2282
2283 return 0;
2284}
2285
2286static void nfs_destroy_inodecache(void)
2287{
2288 /*
2289 * Make sure all delayed rcu free inodes are flushed before we
2290 * destroy cache.
2291 */
2292 rcu_barrier();
2293 kmem_cache_destroy(nfs_inode_cachep);
2294}
2295
2296struct workqueue_struct *nfsiod_workqueue;
2297EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2298
2299/*
2300 * start up the nfsiod workqueue
2301 */
2302static int nfsiod_start(void)
2303{
2304 struct workqueue_struct *wq;
2305 dprintk("RPC: creating workqueue nfsiod\n");
2306 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2307 if (wq == NULL)
2308 return -ENOMEM;
2309 nfsiod_workqueue = wq;
2310 return 0;
2311}
2312
2313/*
2314 * Destroy the nfsiod workqueue
2315 */
2316static void nfsiod_stop(void)
2317{
2318 struct workqueue_struct *wq;
2319
2320 wq = nfsiod_workqueue;
2321 if (wq == NULL)
2322 return;
2323 nfsiod_workqueue = NULL;
2324 destroy_workqueue(wq);
2325}
2326
2327unsigned int nfs_net_id;
2328EXPORT_SYMBOL_GPL(nfs_net_id);
2329
2330static int nfs_net_init(struct net *net)
2331{
2332 nfs_clients_init(net);
2333 return nfs_fs_proc_net_init(net);
2334}
2335
2336static void nfs_net_exit(struct net *net)
2337{
2338 nfs_fs_proc_net_exit(net);
2339 nfs_clients_exit(net);
2340}
2341
2342static struct pernet_operations nfs_net_ops = {
2343 .init = nfs_net_init,
2344 .exit = nfs_net_exit,
2345 .id = &nfs_net_id,
2346 .size = sizeof(struct nfs_net),
2347};
2348
2349/*
2350 * Initialize NFS
2351 */
2352static int __init init_nfs_fs(void)
2353{
2354 int err;
2355
2356 err = nfs_sysfs_init();
2357 if (err < 0)
2358 goto out10;
2359
2360 err = register_pernet_subsys(&nfs_net_ops);
2361 if (err < 0)
2362 goto out9;
2363
2364 err = nfs_fscache_register();
2365 if (err < 0)
2366 goto out8;
2367
2368 err = nfsiod_start();
2369 if (err)
2370 goto out7;
2371
2372 err = nfs_fs_proc_init();
2373 if (err)
2374 goto out6;
2375
2376 err = nfs_init_nfspagecache();
2377 if (err)
2378 goto out5;
2379
2380 err = nfs_init_inodecache();
2381 if (err)
2382 goto out4;
2383
2384 err = nfs_init_readpagecache();
2385 if (err)
2386 goto out3;
2387
2388 err = nfs_init_writepagecache();
2389 if (err)
2390 goto out2;
2391
2392 err = nfs_init_directcache();
2393 if (err)
2394 goto out1;
2395
2396 rpc_proc_register(&init_net, &nfs_rpcstat);
2397
2398 err = register_nfs_fs();
2399 if (err)
2400 goto out0;
2401
2402 return 0;
2403out0:
2404 rpc_proc_unregister(&init_net, "nfs");
2405 nfs_destroy_directcache();
2406out1:
2407 nfs_destroy_writepagecache();
2408out2:
2409 nfs_destroy_readpagecache();
2410out3:
2411 nfs_destroy_inodecache();
2412out4:
2413 nfs_destroy_nfspagecache();
2414out5:
2415 nfs_fs_proc_exit();
2416out6:
2417 nfsiod_stop();
2418out7:
2419 nfs_fscache_unregister();
2420out8:
2421 unregister_pernet_subsys(&nfs_net_ops);
2422out9:
2423 nfs_sysfs_exit();
2424out10:
2425 return err;
2426}
2427
2428static void __exit exit_nfs_fs(void)
2429{
2430 nfs_destroy_directcache();
2431 nfs_destroy_writepagecache();
2432 nfs_destroy_readpagecache();
2433 nfs_destroy_inodecache();
2434 nfs_destroy_nfspagecache();
2435 nfs_fscache_unregister();
2436 unregister_pernet_subsys(&nfs_net_ops);
2437 rpc_proc_unregister(&init_net, "nfs");
2438 unregister_nfs_fs();
2439 nfs_fs_proc_exit();
2440 nfsiod_stop();
2441 nfs_sysfs_exit();
2442}
2443
2444/* Not quite true; I just maintain it */
2445MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2446MODULE_LICENSE("GPL");
2447module_param(enable_ino64, bool, 0644);
2448
2449module_init(init_nfs_fs)
2450module_exit(exit_nfs_fs)
1// 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
75int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
76{
77 schedule();
78 if (signal_pending_state(mode, current))
79 return -ERESTARTSYS;
80 return 0;
81}
82EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
83
84/**
85 * nfs_compat_user_ino64 - returns the user-visible inode number
86 * @fileid: 64-bit fileid
87 *
88 * This function returns a 32-bit inode number if the boot parameter
89 * nfs.enable_ino64 is zero.
90 */
91u64 nfs_compat_user_ino64(u64 fileid)
92{
93#ifdef CONFIG_COMPAT
94 compat_ulong_t ino;
95#else
96 unsigned long ino;
97#endif
98
99 if (enable_ino64)
100 return fileid;
101 ino = fileid;
102 if (sizeof(ino) < sizeof(fileid))
103 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
104 return ino;
105}
106
107int nfs_drop_inode(struct inode *inode)
108{
109 return NFS_STALE(inode) || generic_drop_inode(inode);
110}
111EXPORT_SYMBOL_GPL(nfs_drop_inode);
112
113void nfs_clear_inode(struct inode *inode)
114{
115 /*
116 * The following should never happen...
117 */
118 WARN_ON_ONCE(nfs_have_writebacks(inode));
119 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
120 nfs_zap_acl_cache(inode);
121 nfs_access_zap_cache(inode);
122 nfs_fscache_clear_inode(inode);
123}
124EXPORT_SYMBOL_GPL(nfs_clear_inode);
125
126void nfs_evict_inode(struct inode *inode)
127{
128 truncate_inode_pages_final(&inode->i_data);
129 clear_inode(inode);
130 nfs_clear_inode(inode);
131}
132
133int nfs_sync_inode(struct inode *inode)
134{
135 inode_dio_wait(inode);
136 return nfs_wb_all(inode);
137}
138EXPORT_SYMBOL_GPL(nfs_sync_inode);
139
140/**
141 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
142 * @mapping: pointer to struct address_space
143 */
144int nfs_sync_mapping(struct address_space *mapping)
145{
146 int ret = 0;
147
148 if (mapping->nrpages != 0) {
149 unmap_mapping_range(mapping, 0, 0, 0);
150 ret = nfs_wb_all(mapping->host);
151 }
152 return ret;
153}
154
155static int nfs_attribute_timeout(struct inode *inode)
156{
157 struct nfs_inode *nfsi = NFS_I(inode);
158
159 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
160}
161
162static bool nfs_check_cache_flags_invalid(struct inode *inode,
163 unsigned long flags)
164{
165 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
166
167 return (cache_validity & flags) != 0;
168}
169
170bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
171{
172 if (nfs_check_cache_flags_invalid(inode, flags))
173 return true;
174 return nfs_attribute_cache_expired(inode);
175}
176EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
177
178#ifdef CONFIG_NFS_V4_2
179static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
180{
181 return nfsi->xattr_cache != NULL;
182}
183#else
184static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
185{
186 return false;
187}
188#endif
189
190void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
191{
192 struct nfs_inode *nfsi = NFS_I(inode);
193 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
194
195 if (have_delegation) {
196 if (!(flags & NFS_INO_REVAL_FORCED))
197 flags &= ~(NFS_INO_INVALID_MODE |
198 NFS_INO_INVALID_OTHER |
199 NFS_INO_INVALID_XATTR);
200 flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
201 }
202
203 if (!nfs_has_xattr_cache(nfsi))
204 flags &= ~NFS_INO_INVALID_XATTR;
205 if (flags & NFS_INO_INVALID_DATA)
206 nfs_fscache_invalidate(inode, 0);
207 flags &= ~NFS_INO_REVAL_FORCED;
208
209 nfsi->cache_validity |= flags;
210
211 if (inode->i_mapping->nrpages == 0) {
212 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
213 nfs_ooo_clear(nfsi);
214 } else if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
215 nfs_ooo_clear(nfsi);
216 }
217 trace_nfs_set_cache_invalid(inode, 0);
218}
219EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
220
221/*
222 * Invalidate the local caches
223 */
224static void nfs_zap_caches_locked(struct inode *inode)
225{
226 struct nfs_inode *nfsi = NFS_I(inode);
227 int mode = inode->i_mode;
228
229 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
230
231 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
232 nfsi->attrtimeo_timestamp = jiffies;
233
234 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
235 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
236 NFS_INO_INVALID_DATA |
237 NFS_INO_INVALID_ACCESS |
238 NFS_INO_INVALID_ACL |
239 NFS_INO_INVALID_XATTR);
240 else
241 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
242 NFS_INO_INVALID_ACCESS |
243 NFS_INO_INVALID_ACL |
244 NFS_INO_INVALID_XATTR);
245 nfs_zap_label_cache_locked(nfsi);
246}
247
248void nfs_zap_caches(struct inode *inode)
249{
250 spin_lock(&inode->i_lock);
251 nfs_zap_caches_locked(inode);
252 spin_unlock(&inode->i_lock);
253}
254
255void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
256{
257 if (mapping->nrpages != 0) {
258 spin_lock(&inode->i_lock);
259 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
260 spin_unlock(&inode->i_lock);
261 }
262}
263
264void nfs_zap_acl_cache(struct inode *inode)
265{
266 void (*clear_acl_cache)(struct inode *);
267
268 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
269 if (clear_acl_cache != NULL)
270 clear_acl_cache(inode);
271 spin_lock(&inode->i_lock);
272 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
273 spin_unlock(&inode->i_lock);
274}
275EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
276
277void nfs_invalidate_atime(struct inode *inode)
278{
279 spin_lock(&inode->i_lock);
280 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
281 spin_unlock(&inode->i_lock);
282}
283EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
284
285/*
286 * Invalidate, but do not unhash, the inode.
287 * NB: must be called with inode->i_lock held!
288 */
289static void nfs_set_inode_stale_locked(struct inode *inode)
290{
291 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
292 nfs_zap_caches_locked(inode);
293 trace_nfs_set_inode_stale(inode);
294}
295
296void nfs_set_inode_stale(struct inode *inode)
297{
298 spin_lock(&inode->i_lock);
299 nfs_set_inode_stale_locked(inode);
300 spin_unlock(&inode->i_lock);
301}
302
303struct nfs_find_desc {
304 struct nfs_fh *fh;
305 struct nfs_fattr *fattr;
306};
307
308/*
309 * In NFSv3 we can have 64bit inode numbers. In order to support
310 * this, and re-exported directories (also seen in NFSv2)
311 * we are forced to allow 2 different inodes to have the same
312 * i_ino.
313 */
314static int
315nfs_find_actor(struct inode *inode, void *opaque)
316{
317 struct nfs_find_desc *desc = opaque;
318 struct nfs_fh *fh = desc->fh;
319 struct nfs_fattr *fattr = desc->fattr;
320
321 if (NFS_FILEID(inode) != fattr->fileid)
322 return 0;
323 if (inode_wrong_type(inode, fattr->mode))
324 return 0;
325 if (nfs_compare_fh(NFS_FH(inode), fh))
326 return 0;
327 if (is_bad_inode(inode) || NFS_STALE(inode))
328 return 0;
329 return 1;
330}
331
332static int
333nfs_init_locked(struct inode *inode, void *opaque)
334{
335 struct nfs_find_desc *desc = opaque;
336 struct nfs_fattr *fattr = desc->fattr;
337
338 set_nfs_fileid(inode, fattr->fileid);
339 inode->i_mode = fattr->mode;
340 nfs_copy_fh(NFS_FH(inode), desc->fh);
341 return 0;
342}
343
344#ifdef CONFIG_NFS_V4_SECURITY_LABEL
345static void nfs_clear_label_invalid(struct inode *inode)
346{
347 spin_lock(&inode->i_lock);
348 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
349 spin_unlock(&inode->i_lock);
350}
351
352void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
353{
354 int error;
355
356 if (fattr->label == NULL)
357 return;
358
359 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
360 error = security_inode_notifysecctx(inode, fattr->label->label,
361 fattr->label->len);
362 if (error)
363 printk(KERN_ERR "%s() %s %d "
364 "security_inode_notifysecctx() %d\n",
365 __func__,
366 (char *)fattr->label->label,
367 fattr->label->len, error);
368 nfs_clear_label_invalid(inode);
369 }
370}
371
372struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
373{
374 struct nfs4_label *label;
375
376 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
377 return NULL;
378
379 label = kzalloc(sizeof(struct nfs4_label), flags);
380 if (label == NULL)
381 return ERR_PTR(-ENOMEM);
382
383 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
384 if (label->label == NULL) {
385 kfree(label);
386 return ERR_PTR(-ENOMEM);
387 }
388 label->len = NFS4_MAXLABELLEN;
389
390 return label;
391}
392EXPORT_SYMBOL_GPL(nfs4_label_alloc);
393#else
394void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
395{
396}
397#endif
398EXPORT_SYMBOL_GPL(nfs_setsecurity);
399
400/* Search for inode identified by fh, fileid and i_mode in inode cache. */
401struct inode *
402nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
403{
404 struct nfs_find_desc desc = {
405 .fh = fh,
406 .fattr = fattr,
407 };
408 struct inode *inode;
409 unsigned long hash;
410
411 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
412 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
413 return NULL;
414
415 hash = nfs_fattr_to_ino_t(fattr);
416 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
417
418 dprintk("%s: returning %p\n", __func__, inode);
419 return inode;
420}
421
422static void nfs_inode_init_regular(struct nfs_inode *nfsi)
423{
424 atomic_long_set(&nfsi->nrequests, 0);
425 atomic_long_set(&nfsi->redirtied_pages, 0);
426 INIT_LIST_HEAD(&nfsi->commit_info.list);
427 atomic_long_set(&nfsi->commit_info.ncommit, 0);
428 atomic_set(&nfsi->commit_info.rpcs_out, 0);
429 mutex_init(&nfsi->commit_mutex);
430}
431
432static void nfs_inode_init_dir(struct nfs_inode *nfsi)
433{
434 nfsi->cache_change_attribute = 0;
435 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
436 init_rwsem(&nfsi->rmdir_sem);
437}
438
439/*
440 * This is our front-end to iget that looks up inodes by file handle
441 * instead of inode number.
442 */
443struct inode *
444nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
445{
446 struct nfs_find_desc desc = {
447 .fh = fh,
448 .fattr = fattr
449 };
450 struct inode *inode = ERR_PTR(-ENOENT);
451 u64 fattr_supported = NFS_SB(sb)->fattr_valid;
452 unsigned long hash;
453
454 nfs_attr_check_mountpoint(sb, fattr);
455
456 if (nfs_attr_use_mounted_on_fileid(fattr))
457 fattr->fileid = fattr->mounted_on_fileid;
458 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
459 goto out_no_inode;
460 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
461 goto out_no_inode;
462
463 hash = nfs_fattr_to_ino_t(fattr);
464
465 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
466 if (inode == NULL) {
467 inode = ERR_PTR(-ENOMEM);
468 goto out_no_inode;
469 }
470
471 if (inode->i_state & I_NEW) {
472 struct nfs_inode *nfsi = NFS_I(inode);
473 unsigned long now = jiffies;
474
475 /* We set i_ino for the few things that still rely on it,
476 * such as stat(2) */
477 inode->i_ino = hash;
478
479 /* We can't support update_atime(), since the server will reset it */
480 inode->i_flags |= S_NOATIME|S_NOCMTIME;
481 inode->i_mode = fattr->mode;
482 nfsi->cache_validity = 0;
483 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
484 && (fattr_supported & NFS_ATTR_FATTR_MODE))
485 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
486 /* Why so? Because we want revalidate for devices/FIFOs, and
487 * that's precisely what we have in nfs_file_inode_operations.
488 */
489 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
490 if (S_ISREG(inode->i_mode)) {
491 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
492 inode->i_data.a_ops = &nfs_file_aops;
493 nfs_inode_init_regular(nfsi);
494 } else if (S_ISDIR(inode->i_mode)) {
495 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
496 inode->i_fop = &nfs_dir_operations;
497 inode->i_data.a_ops = &nfs_dir_aops;
498 nfs_inode_init_dir(nfsi);
499 /* Deal with crossing mountpoints */
500 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
501 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
502 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
503 inode->i_op = &nfs_referral_inode_operations;
504 else
505 inode->i_op = &nfs_mountpoint_inode_operations;
506 inode->i_fop = NULL;
507 inode->i_flags |= S_AUTOMOUNT;
508 }
509 } else if (S_ISLNK(inode->i_mode)) {
510 inode->i_op = &nfs_symlink_inode_operations;
511 inode_nohighmem(inode);
512 } else
513 init_special_inode(inode, inode->i_mode, fattr->rdev);
514
515 inode_set_atime(inode, 0, 0);
516 inode_set_mtime(inode, 0, 0);
517 inode_set_ctime(inode, 0, 0);
518 inode_set_iversion_raw(inode, 0);
519 inode->i_size = 0;
520 clear_nlink(inode);
521 inode->i_uid = make_kuid(&init_user_ns, -2);
522 inode->i_gid = make_kgid(&init_user_ns, -2);
523 inode->i_blocks = 0;
524 nfsi->write_io = 0;
525 nfsi->read_io = 0;
526
527 nfsi->read_cache_jiffies = fattr->time_start;
528 nfsi->attr_gencount = fattr->gencount;
529 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
530 inode_set_atime_to_ts(inode, fattr->atime);
531 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
532 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
533 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
534 inode_set_mtime_to_ts(inode, fattr->mtime);
535 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
536 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
537 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
538 inode_set_ctime_to_ts(inode, fattr->ctime);
539 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
540 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
541 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
542 inode_set_iversion_raw(inode, fattr->change_attr);
543 else
544 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
545 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
546 inode->i_size = nfs_size_to_loff_t(fattr->size);
547 else
548 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
549 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
550 set_nlink(inode, fattr->nlink);
551 else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
552 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
553 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
554 inode->i_uid = fattr->uid;
555 else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
556 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
557 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
558 inode->i_gid = fattr->gid;
559 else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
560 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
561 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
562 inode->i_blocks = fattr->du.nfs2.blocks;
563 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
564 fattr->size != 0)
565 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
566 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
567 /*
568 * report the blocks in 512byte units
569 */
570 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
571 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
572 fattr->size != 0)
573 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
574
575 nfs_setsecurity(inode, fattr);
576
577 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
578 nfsi->attrtimeo_timestamp = now;
579 nfsi->access_cache = RB_ROOT;
580
581 nfs_fscache_init_inode(inode);
582
583 unlock_new_inode(inode);
584 } else {
585 int err = nfs_refresh_inode(inode, fattr);
586 if (err < 0) {
587 iput(inode);
588 inode = ERR_PTR(err);
589 goto out_no_inode;
590 }
591 }
592 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
593 inode->i_sb->s_id,
594 (unsigned long long)NFS_FILEID(inode),
595 nfs_display_fhandle_hash(fh),
596 atomic_read(&inode->i_count));
597
598out:
599 return inode;
600
601out_no_inode:
602 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
603 goto out;
604}
605EXPORT_SYMBOL_GPL(nfs_fhget);
606
607#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)
608
609int
610nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
611 struct iattr *attr)
612{
613 struct inode *inode = d_inode(dentry);
614 struct nfs_fattr *fattr;
615 int error = 0;
616
617 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
618
619 /* skip mode change if it's just for clearing setuid/setgid */
620 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
621 attr->ia_valid &= ~ATTR_MODE;
622
623 if (attr->ia_valid & ATTR_SIZE) {
624 BUG_ON(!S_ISREG(inode->i_mode));
625
626 error = inode_newsize_ok(inode, attr->ia_size);
627 if (error)
628 return error;
629
630 if (attr->ia_size == i_size_read(inode))
631 attr->ia_valid &= ~ATTR_SIZE;
632 }
633
634 /* Optimization: if the end result is no change, don't RPC */
635 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
636 return 0;
637
638 trace_nfs_setattr_enter(inode);
639
640 /* Write all dirty data */
641 if (S_ISREG(inode->i_mode))
642 nfs_sync_inode(inode);
643
644 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
645 if (fattr == NULL) {
646 error = -ENOMEM;
647 goto out;
648 }
649
650 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
651 if (error == 0)
652 error = nfs_refresh_inode(inode, fattr);
653 nfs_free_fattr(fattr);
654out:
655 trace_nfs_setattr_exit(inode, error);
656 return error;
657}
658EXPORT_SYMBOL_GPL(nfs_setattr);
659
660/**
661 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
662 * @inode: inode of the file used
663 * @offset: file offset to start truncating
664 *
665 * This is a copy of the common vmtruncate, but with the locking
666 * corrected to take into account the fact that NFS requires
667 * inode->i_size to be updated under the inode->i_lock.
668 * Note: must be called with inode->i_lock held!
669 */
670static int nfs_vmtruncate(struct inode * inode, loff_t offset)
671{
672 int err;
673
674 err = inode_newsize_ok(inode, offset);
675 if (err)
676 goto out;
677
678 trace_nfs_size_truncate(inode, offset);
679 i_size_write(inode, offset);
680 /* Optimisation */
681 if (offset == 0) {
682 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
683 nfs_ooo_clear(NFS_I(inode));
684 }
685 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
686
687 spin_unlock(&inode->i_lock);
688 truncate_pagecache(inode, offset);
689 spin_lock(&inode->i_lock);
690out:
691 return err;
692}
693
694/**
695 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
696 * @inode: pointer to struct inode
697 * @attr: pointer to struct iattr
698 * @fattr: pointer to struct nfs_fattr
699 *
700 * Note: we do this in the *proc.c in order to ensure that
701 * it works for things like exclusive creates too.
702 */
703void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
704 struct nfs_fattr *fattr)
705{
706 /* Barrier: bump the attribute generation count. */
707 nfs_fattr_set_barrier(fattr);
708
709 spin_lock(&inode->i_lock);
710 NFS_I(inode)->attr_gencount = fattr->gencount;
711 if ((attr->ia_valid & ATTR_SIZE) != 0) {
712 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME |
713 NFS_INO_INVALID_BLOCKS);
714 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
715 nfs_vmtruncate(inode, attr->ia_size);
716 }
717 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
718 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
719 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
720 inode->i_mode & S_ISUID)
721 inode->i_mode &= ~S_ISUID;
722 if (setattr_should_drop_sgid(&nop_mnt_idmap, inode))
723 inode->i_mode &= ~S_ISGID;
724 if ((attr->ia_valid & ATTR_MODE) != 0) {
725 int mode = attr->ia_mode & S_IALLUGO;
726 mode |= inode->i_mode & ~S_IALLUGO;
727 inode->i_mode = mode;
728 }
729 if ((attr->ia_valid & ATTR_UID) != 0)
730 inode->i_uid = attr->ia_uid;
731 if ((attr->ia_valid & ATTR_GID) != 0)
732 inode->i_gid = attr->ia_gid;
733 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
734 inode_set_ctime_to_ts(inode, fattr->ctime);
735 else
736 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
737 | NFS_INO_INVALID_CTIME);
738 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
739 | NFS_INO_INVALID_ACL);
740 }
741 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
742 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
743 | NFS_INO_INVALID_CTIME);
744 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
745 inode_set_atime_to_ts(inode, fattr->atime);
746 else if (attr->ia_valid & ATTR_ATIME_SET)
747 inode_set_atime_to_ts(inode, attr->ia_atime);
748 else
749 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
750
751 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
752 inode_set_ctime_to_ts(inode, fattr->ctime);
753 else
754 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
755 | NFS_INO_INVALID_CTIME);
756 }
757 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
758 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
759 | NFS_INO_INVALID_CTIME);
760 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
761 inode_set_mtime_to_ts(inode, fattr->mtime);
762 else if (attr->ia_valid & ATTR_MTIME_SET)
763 inode_set_mtime_to_ts(inode, attr->ia_mtime);
764 else
765 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
766
767 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
768 inode_set_ctime_to_ts(inode, fattr->ctime);
769 else
770 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
771 | NFS_INO_INVALID_CTIME);
772 }
773 if (fattr->valid)
774 nfs_update_inode(inode, fattr);
775 spin_unlock(&inode->i_lock);
776}
777EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
778
779/*
780 * Don't request help from readdirplus if the file is being written to,
781 * or if attribute caching is turned off
782 */
783static bool nfs_getattr_readdirplus_enable(const struct inode *inode)
784{
785 return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) &&
786 !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ;
787}
788
789static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
790{
791 if (!IS_ROOT(dentry)) {
792 struct dentry *parent = dget_parent(dentry);
793 nfs_readdir_record_entry_cache_miss(d_inode(parent));
794 dput(parent);
795 }
796}
797
798static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
799{
800 if (!IS_ROOT(dentry)) {
801 struct dentry *parent = dget_parent(dentry);
802 nfs_readdir_record_entry_cache_hit(d_inode(parent));
803 dput(parent);
804 }
805}
806
807static u32 nfs_get_valid_attrmask(struct inode *inode)
808{
809 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
810 u32 reply_mask = STATX_INO | STATX_TYPE;
811
812 if (!(cache_validity & NFS_INO_INVALID_ATIME))
813 reply_mask |= STATX_ATIME;
814 if (!(cache_validity & NFS_INO_INVALID_CTIME))
815 reply_mask |= STATX_CTIME;
816 if (!(cache_validity & NFS_INO_INVALID_MTIME))
817 reply_mask |= STATX_MTIME;
818 if (!(cache_validity & NFS_INO_INVALID_SIZE))
819 reply_mask |= STATX_SIZE;
820 if (!(cache_validity & NFS_INO_INVALID_NLINK))
821 reply_mask |= STATX_NLINK;
822 if (!(cache_validity & NFS_INO_INVALID_MODE))
823 reply_mask |= STATX_MODE;
824 if (!(cache_validity & NFS_INO_INVALID_OTHER))
825 reply_mask |= STATX_UID | STATX_GID;
826 if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
827 reply_mask |= STATX_BLOCKS;
828 if (!(cache_validity & NFS_INO_INVALID_CHANGE))
829 reply_mask |= STATX_CHANGE_COOKIE;
830 return reply_mask;
831}
832
833int nfs_getattr(struct mnt_idmap *idmap, const struct path *path,
834 struct kstat *stat, u32 request_mask, unsigned int query_flags)
835{
836 struct inode *inode = d_inode(path->dentry);
837 struct nfs_server *server = NFS_SERVER(inode);
838 unsigned long cache_validity;
839 int err = 0;
840 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
841 bool do_update = false;
842 bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
843
844 trace_nfs_getattr_enter(inode);
845
846 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
847 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
848 STATX_INO | STATX_SIZE | STATX_BLOCKS |
849 STATX_CHANGE_COOKIE;
850
851 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
852 if (readdirplus_enabled)
853 nfs_readdirplus_parent_cache_hit(path->dentry);
854 goto out_no_revalidate;
855 }
856
857 /* Flush out writes to the server in order to update c/mtime/version. */
858 if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) &&
859 S_ISREG(inode->i_mode))
860 filemap_write_and_wait(inode->i_mapping);
861
862 /*
863 * We may force a getattr if the user cares about atime.
864 *
865 * Note that we only have to check the vfsmount flags here:
866 * - NFS always sets S_NOATIME by so checking it would give a
867 * bogus result
868 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
869 * no point in checking those.
870 */
871 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
872 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
873 request_mask &= ~STATX_ATIME;
874
875 /* Is the user requesting attributes that might need revalidation? */
876 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
877 STATX_MTIME|STATX_UID|STATX_GID|
878 STATX_SIZE|STATX_BLOCKS|
879 STATX_CHANGE_COOKIE)))
880 goto out_no_revalidate;
881
882 /* Check whether the cached attributes are stale */
883 do_update |= force_sync || nfs_attribute_cache_expired(inode);
884 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
885 do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
886 if (request_mask & STATX_ATIME)
887 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
888 if (request_mask & STATX_CTIME)
889 do_update |= cache_validity & NFS_INO_INVALID_CTIME;
890 if (request_mask & STATX_MTIME)
891 do_update |= cache_validity & NFS_INO_INVALID_MTIME;
892 if (request_mask & STATX_SIZE)
893 do_update |= cache_validity & NFS_INO_INVALID_SIZE;
894 if (request_mask & STATX_NLINK)
895 do_update |= cache_validity & NFS_INO_INVALID_NLINK;
896 if (request_mask & STATX_MODE)
897 do_update |= cache_validity & NFS_INO_INVALID_MODE;
898 if (request_mask & (STATX_UID | STATX_GID))
899 do_update |= cache_validity & NFS_INO_INVALID_OTHER;
900 if (request_mask & STATX_BLOCKS)
901 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
902
903 if (do_update) {
904 if (readdirplus_enabled)
905 nfs_readdirplus_parent_cache_miss(path->dentry);
906 err = __nfs_revalidate_inode(server, inode);
907 if (err)
908 goto out;
909 } else if (readdirplus_enabled)
910 nfs_readdirplus_parent_cache_hit(path->dentry);
911out_no_revalidate:
912 /* Only return attributes that were revalidated. */
913 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
914
915 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
916 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
917 stat->change_cookie = inode_peek_iversion_raw(inode);
918 stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC;
919 if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED)
920 stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
921 if (S_ISDIR(inode->i_mode))
922 stat->blksize = NFS_SERVER(inode)->dtsize;
923out:
924 trace_nfs_getattr_exit(inode, err);
925 return err;
926}
927EXPORT_SYMBOL_GPL(nfs_getattr);
928
929static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
930{
931 refcount_set(&l_ctx->count, 1);
932 l_ctx->lockowner = current->files;
933 INIT_LIST_HEAD(&l_ctx->list);
934 atomic_set(&l_ctx->io_count, 0);
935}
936
937static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
938{
939 struct nfs_lock_context *pos;
940
941 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
942 if (pos->lockowner != current->files)
943 continue;
944 if (refcount_inc_not_zero(&pos->count))
945 return pos;
946 }
947 return NULL;
948}
949
950struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
951{
952 struct nfs_lock_context *res, *new = NULL;
953 struct inode *inode = d_inode(ctx->dentry);
954
955 rcu_read_lock();
956 res = __nfs_find_lock_context(ctx);
957 rcu_read_unlock();
958 if (res == NULL) {
959 new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
960 if (new == NULL)
961 return ERR_PTR(-ENOMEM);
962 nfs_init_lock_context(new);
963 spin_lock(&inode->i_lock);
964 res = __nfs_find_lock_context(ctx);
965 if (res == NULL) {
966 new->open_context = get_nfs_open_context(ctx);
967 if (new->open_context) {
968 list_add_tail_rcu(&new->list,
969 &ctx->lock_context.list);
970 res = new;
971 new = NULL;
972 } else
973 res = ERR_PTR(-EBADF);
974 }
975 spin_unlock(&inode->i_lock);
976 kfree(new);
977 }
978 return res;
979}
980EXPORT_SYMBOL_GPL(nfs_get_lock_context);
981
982void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
983{
984 struct nfs_open_context *ctx = l_ctx->open_context;
985 struct inode *inode = d_inode(ctx->dentry);
986
987 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
988 return;
989 list_del_rcu(&l_ctx->list);
990 spin_unlock(&inode->i_lock);
991 put_nfs_open_context(ctx);
992 kfree_rcu(l_ctx, rcu_head);
993}
994EXPORT_SYMBOL_GPL(nfs_put_lock_context);
995
996/**
997 * nfs_close_context - Common close_context() routine NFSv2/v3
998 * @ctx: pointer to context
999 * @is_sync: is this a synchronous close
1000 *
1001 * Ensure that the attributes are up to date if we're mounted
1002 * with close-to-open semantics and we have cached data that will
1003 * need to be revalidated on open.
1004 */
1005void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1006{
1007 struct nfs_inode *nfsi;
1008 struct inode *inode;
1009
1010 if (!(ctx->mode & FMODE_WRITE))
1011 return;
1012 if (!is_sync)
1013 return;
1014 inode = d_inode(ctx->dentry);
1015 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1016 return;
1017 nfsi = NFS_I(inode);
1018 if (inode->i_mapping->nrpages == 0)
1019 return;
1020 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1021 return;
1022 if (!list_empty(&nfsi->open_files))
1023 return;
1024 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1025 return;
1026 nfs_revalidate_inode(inode,
1027 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1028}
1029EXPORT_SYMBOL_GPL(nfs_close_context);
1030
1031struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1032 fmode_t f_mode,
1033 struct file *filp)
1034{
1035 struct nfs_open_context *ctx;
1036
1037 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1038 if (!ctx)
1039 return ERR_PTR(-ENOMEM);
1040 nfs_sb_active(dentry->d_sb);
1041 ctx->dentry = dget(dentry);
1042 if (filp)
1043 ctx->cred = get_cred(filp->f_cred);
1044 else
1045 ctx->cred = get_current_cred();
1046 rcu_assign_pointer(ctx->ll_cred, NULL);
1047 ctx->state = NULL;
1048 ctx->mode = f_mode;
1049 ctx->flags = 0;
1050 ctx->error = 0;
1051 ctx->flock_owner = (fl_owner_t)filp;
1052 nfs_init_lock_context(&ctx->lock_context);
1053 ctx->lock_context.open_context = ctx;
1054 INIT_LIST_HEAD(&ctx->list);
1055 ctx->mdsthreshold = NULL;
1056 return ctx;
1057}
1058EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1059
1060struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1061{
1062 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1063 return ctx;
1064 return NULL;
1065}
1066EXPORT_SYMBOL_GPL(get_nfs_open_context);
1067
1068static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1069{
1070 struct inode *inode = d_inode(ctx->dentry);
1071 struct super_block *sb = ctx->dentry->d_sb;
1072
1073 if (!refcount_dec_and_test(&ctx->lock_context.count))
1074 return;
1075 if (!list_empty(&ctx->list)) {
1076 spin_lock(&inode->i_lock);
1077 list_del_rcu(&ctx->list);
1078 spin_unlock(&inode->i_lock);
1079 }
1080 if (inode != NULL)
1081 NFS_PROTO(inode)->close_context(ctx, is_sync);
1082 put_cred(ctx->cred);
1083 dput(ctx->dentry);
1084 nfs_sb_deactive(sb);
1085 put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1086 kfree(ctx->mdsthreshold);
1087 kfree_rcu(ctx, rcu_head);
1088}
1089
1090void put_nfs_open_context(struct nfs_open_context *ctx)
1091{
1092 __put_nfs_open_context(ctx, 0);
1093}
1094EXPORT_SYMBOL_GPL(put_nfs_open_context);
1095
1096static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1097{
1098 __put_nfs_open_context(ctx, 1);
1099}
1100
1101/*
1102 * Ensure that mmap has a recent RPC credential for use when writing out
1103 * shared pages
1104 */
1105void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1106{
1107 struct inode *inode = d_inode(ctx->dentry);
1108 struct nfs_inode *nfsi = NFS_I(inode);
1109
1110 spin_lock(&inode->i_lock);
1111 if (list_empty(&nfsi->open_files) &&
1112 nfs_ooo_test(nfsi))
1113 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1114 NFS_INO_REVAL_FORCED);
1115 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1116 spin_unlock(&inode->i_lock);
1117}
1118EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1119
1120void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1121{
1122 filp->private_data = get_nfs_open_context(ctx);
1123 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1124 if (list_empty(&ctx->list))
1125 nfs_inode_attach_open_context(ctx);
1126}
1127EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1128
1129/*
1130 * Given an inode, search for an open context with the desired characteristics
1131 */
1132struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1133{
1134 struct nfs_inode *nfsi = NFS_I(inode);
1135 struct nfs_open_context *pos, *ctx = NULL;
1136
1137 rcu_read_lock();
1138 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1139 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1140 continue;
1141 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1142 continue;
1143 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1144 continue;
1145 ctx = get_nfs_open_context(pos);
1146 if (ctx)
1147 break;
1148 }
1149 rcu_read_unlock();
1150 return ctx;
1151}
1152
1153void nfs_file_clear_open_context(struct file *filp)
1154{
1155 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1156
1157 if (ctx) {
1158 struct inode *inode = d_inode(ctx->dentry);
1159
1160 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1161 /*
1162 * We fatal error on write before. Try to writeback
1163 * every page again.
1164 */
1165 if (ctx->error < 0)
1166 invalidate_inode_pages2(inode->i_mapping);
1167 filp->private_data = NULL;
1168 put_nfs_open_context_sync(ctx);
1169 }
1170}
1171
1172/*
1173 * These allocate and release file read/write context information.
1174 */
1175int nfs_open(struct inode *inode, struct file *filp)
1176{
1177 struct nfs_open_context *ctx;
1178
1179 ctx = alloc_nfs_open_context(file_dentry(filp),
1180 flags_to_mode(filp->f_flags), filp);
1181 if (IS_ERR(ctx))
1182 return PTR_ERR(ctx);
1183 nfs_file_set_open_context(filp, ctx);
1184 put_nfs_open_context(ctx);
1185 nfs_fscache_open_file(inode, filp);
1186 return 0;
1187}
1188
1189/*
1190 * This function is called whenever some part of NFS notices that
1191 * the cached attributes have to be refreshed.
1192 */
1193int
1194__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1195{
1196 int status = -ESTALE;
1197 struct nfs_fattr *fattr = NULL;
1198 struct nfs_inode *nfsi = NFS_I(inode);
1199
1200 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1201 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1202
1203 trace_nfs_revalidate_inode_enter(inode);
1204
1205 if (is_bad_inode(inode))
1206 goto out;
1207 if (NFS_STALE(inode))
1208 goto out;
1209
1210 /* pNFS: Attributes aren't updated until we layoutcommit */
1211 if (S_ISREG(inode->i_mode)) {
1212 status = pnfs_sync_inode(inode, false);
1213 if (status)
1214 goto out;
1215 }
1216
1217 status = -ENOMEM;
1218 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1219 if (fattr == NULL)
1220 goto out;
1221
1222 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1223
1224 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1225 if (status != 0) {
1226 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1227 inode->i_sb->s_id,
1228 (unsigned long long)NFS_FILEID(inode), status);
1229 switch (status) {
1230 case -ETIMEDOUT:
1231 /* A soft timeout occurred. Use cached information? */
1232 if (server->flags & NFS_MOUNT_SOFTREVAL)
1233 status = 0;
1234 break;
1235 case -ESTALE:
1236 if (!S_ISDIR(inode->i_mode))
1237 nfs_set_inode_stale(inode);
1238 else
1239 nfs_zap_caches(inode);
1240 }
1241 goto out;
1242 }
1243
1244 status = nfs_refresh_inode(inode, fattr);
1245 if (status) {
1246 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1247 inode->i_sb->s_id,
1248 (unsigned long long)NFS_FILEID(inode), status);
1249 goto out;
1250 }
1251
1252 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1253 nfs_zap_acl_cache(inode);
1254
1255 nfs_setsecurity(inode, fattr);
1256
1257 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1258 inode->i_sb->s_id,
1259 (unsigned long long)NFS_FILEID(inode));
1260
1261out:
1262 nfs_free_fattr(fattr);
1263 trace_nfs_revalidate_inode_exit(inode, status);
1264 return status;
1265}
1266
1267int nfs_attribute_cache_expired(struct inode *inode)
1268{
1269 if (nfs_have_delegated_attributes(inode))
1270 return 0;
1271 return nfs_attribute_timeout(inode);
1272}
1273
1274/**
1275 * nfs_revalidate_inode - Revalidate the inode attributes
1276 * @inode: pointer to inode struct
1277 * @flags: cache flags to check
1278 *
1279 * Updates inode attribute information by retrieving the data from the server.
1280 */
1281int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1282{
1283 if (!nfs_check_cache_invalid(inode, flags))
1284 return NFS_STALE(inode) ? -ESTALE : 0;
1285 return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1286}
1287EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1288
1289static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1290{
1291 int ret;
1292
1293 nfs_fscache_invalidate(inode, 0);
1294 if (mapping->nrpages != 0) {
1295 if (S_ISREG(inode->i_mode)) {
1296 ret = nfs_sync_mapping(mapping);
1297 if (ret < 0)
1298 return ret;
1299 }
1300 ret = invalidate_inode_pages2(mapping);
1301 if (ret < 0)
1302 return ret;
1303 }
1304 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1305
1306 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1307 inode->i_sb->s_id,
1308 (unsigned long long)NFS_FILEID(inode));
1309 return 0;
1310}
1311
1312/**
1313 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1314 * @mapping: pointer to mapping
1315 *
1316 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1317 */
1318int nfs_clear_invalid_mapping(struct address_space *mapping)
1319{
1320 struct inode *inode = mapping->host;
1321 struct nfs_inode *nfsi = NFS_I(inode);
1322 unsigned long *bitlock = &nfsi->flags;
1323 int ret = 0;
1324
1325 /*
1326 * We must clear NFS_INO_INVALID_DATA first to ensure that
1327 * invalidations that come in while we're shooting down the mappings
1328 * are respected. But, that leaves a race window where one revalidator
1329 * can clear the flag, and then another checks it before the mapping
1330 * gets invalidated. Fix that by serializing access to this part of
1331 * the function.
1332 *
1333 * At the same time, we need to allow other tasks to see whether we
1334 * might be in the middle of invalidating the pages, so we only set
1335 * the bit lock here if it looks like we're going to be doing that.
1336 */
1337 for (;;) {
1338 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1339 nfs_wait_bit_killable,
1340 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1341 if (ret)
1342 goto out;
1343 spin_lock(&inode->i_lock);
1344 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1345 spin_unlock(&inode->i_lock);
1346 continue;
1347 }
1348 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1349 break;
1350 spin_unlock(&inode->i_lock);
1351 goto out;
1352 }
1353
1354 set_bit(NFS_INO_INVALIDATING, bitlock);
1355 smp_wmb();
1356 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1357 nfs_ooo_clear(nfsi);
1358 spin_unlock(&inode->i_lock);
1359 trace_nfs_invalidate_mapping_enter(inode);
1360 ret = nfs_invalidate_mapping(inode, mapping);
1361 trace_nfs_invalidate_mapping_exit(inode, ret);
1362
1363 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1364 smp_mb__after_atomic();
1365 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1366out:
1367 return ret;
1368}
1369
1370bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1371{
1372 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1373 NFS_STALE(inode);
1374}
1375
1376int nfs_revalidate_mapping_rcu(struct inode *inode)
1377{
1378 struct nfs_inode *nfsi = NFS_I(inode);
1379 unsigned long *bitlock = &nfsi->flags;
1380 int ret = 0;
1381
1382 if (IS_SWAPFILE(inode))
1383 goto out;
1384 if (nfs_mapping_need_revalidate_inode(inode)) {
1385 ret = -ECHILD;
1386 goto out;
1387 }
1388 spin_lock(&inode->i_lock);
1389 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1390 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1391 ret = -ECHILD;
1392 spin_unlock(&inode->i_lock);
1393out:
1394 return ret;
1395}
1396
1397/**
1398 * nfs_revalidate_mapping - Revalidate the pagecache
1399 * @inode: pointer to host inode
1400 * @mapping: pointer to mapping
1401 */
1402int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1403{
1404 /* swapfiles are not supposed to be shared. */
1405 if (IS_SWAPFILE(inode))
1406 return 0;
1407
1408 if (nfs_mapping_need_revalidate_inode(inode)) {
1409 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1410 if (ret < 0)
1411 return ret;
1412 }
1413
1414 return nfs_clear_invalid_mapping(mapping);
1415}
1416
1417static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1418{
1419 struct inode *inode = &nfsi->vfs_inode;
1420
1421 if (!S_ISREG(inode->i_mode))
1422 return false;
1423 if (list_empty(&nfsi->open_files))
1424 return false;
1425 return inode_is_open_for_write(inode);
1426}
1427
1428static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1429{
1430 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1431}
1432
1433static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1434{
1435 struct timespec64 ts;
1436
1437 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1438 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1439 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1440 inode_set_iversion_raw(inode, fattr->change_attr);
1441 if (S_ISDIR(inode->i_mode))
1442 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1443 else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1444 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1445 }
1446 /* If we have atomic WCC data, we may update some attributes */
1447 ts = inode_get_ctime(inode);
1448 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1449 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1450 && timespec64_equal(&ts, &fattr->pre_ctime)) {
1451 inode_set_ctime_to_ts(inode, fattr->ctime);
1452 }
1453
1454 ts = inode_get_mtime(inode);
1455 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1456 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1457 && timespec64_equal(&ts, &fattr->pre_mtime)) {
1458 inode_set_mtime_to_ts(inode, fattr->mtime);
1459 }
1460 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1461 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1462 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1463 && !nfs_have_writebacks(inode)) {
1464 trace_nfs_size_wcc(inode, fattr->size);
1465 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1466 }
1467}
1468
1469/**
1470 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1471 * @inode: pointer to inode
1472 * @fattr: updated attributes
1473 *
1474 * Verifies the attribute cache. If we have just changed the attributes,
1475 * so that fattr carries weak cache consistency data, then it may
1476 * also update the ctime/mtime/change_attribute.
1477 */
1478static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1479{
1480 struct nfs_inode *nfsi = NFS_I(inode);
1481 loff_t cur_size, new_isize;
1482 unsigned long invalid = 0;
1483 struct timespec64 ts;
1484
1485 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1486 return 0;
1487
1488 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1489 /* Only a mounted-on-fileid? Just exit */
1490 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1491 return 0;
1492 /* Has the inode gone and changed behind our back? */
1493 } else if (nfsi->fileid != fattr->fileid) {
1494 /* Is this perhaps the mounted-on fileid? */
1495 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1496 nfsi->fileid == fattr->mounted_on_fileid)
1497 return 0;
1498 return -ESTALE;
1499 }
1500 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1501 return -ESTALE;
1502
1503
1504 if (!nfs_file_has_buffered_writers(nfsi)) {
1505 /* Verify a few of the more important attributes */
1506 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1507 invalid |= NFS_INO_INVALID_CHANGE;
1508
1509 ts = inode_get_mtime(inode);
1510 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1511 invalid |= NFS_INO_INVALID_MTIME;
1512
1513 ts = inode_get_ctime(inode);
1514 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1515 invalid |= NFS_INO_INVALID_CTIME;
1516
1517 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1518 cur_size = i_size_read(inode);
1519 new_isize = nfs_size_to_loff_t(fattr->size);
1520 if (cur_size != new_isize)
1521 invalid |= NFS_INO_INVALID_SIZE;
1522 }
1523 }
1524
1525 /* Have any file permissions changed? */
1526 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1527 invalid |= NFS_INO_INVALID_MODE;
1528 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1529 invalid |= NFS_INO_INVALID_OTHER;
1530 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1531 invalid |= NFS_INO_INVALID_OTHER;
1532
1533 /* Has the link count changed? */
1534 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1535 invalid |= NFS_INO_INVALID_NLINK;
1536
1537 ts = inode_get_atime(inode);
1538 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1539 invalid |= NFS_INO_INVALID_ATIME;
1540
1541 if (invalid != 0)
1542 nfs_set_cache_invalid(inode, invalid);
1543
1544 nfsi->read_cache_jiffies = fattr->time_start;
1545 return 0;
1546}
1547
1548static atomic_long_t nfs_attr_generation_counter;
1549
1550static unsigned long nfs_read_attr_generation_counter(void)
1551{
1552 return atomic_long_read(&nfs_attr_generation_counter);
1553}
1554
1555unsigned long nfs_inc_attr_generation_counter(void)
1556{
1557 return atomic_long_inc_return(&nfs_attr_generation_counter);
1558}
1559EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1560
1561void nfs_fattr_init(struct nfs_fattr *fattr)
1562{
1563 fattr->valid = 0;
1564 fattr->time_start = jiffies;
1565 fattr->gencount = nfs_inc_attr_generation_counter();
1566 fattr->owner_name = NULL;
1567 fattr->group_name = NULL;
1568}
1569EXPORT_SYMBOL_GPL(nfs_fattr_init);
1570
1571/**
1572 * nfs_fattr_set_barrier
1573 * @fattr: attributes
1574 *
1575 * Used to set a barrier after an attribute was updated. This
1576 * barrier ensures that older attributes from RPC calls that may
1577 * have raced with our update cannot clobber these new values.
1578 * Note that you are still responsible for ensuring that other
1579 * operations which change the attribute on the server do not
1580 * collide.
1581 */
1582void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1583{
1584 fattr->gencount = nfs_inc_attr_generation_counter();
1585}
1586
1587struct nfs_fattr *nfs_alloc_fattr(void)
1588{
1589 struct nfs_fattr *fattr;
1590
1591 fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1592 if (fattr != NULL) {
1593 nfs_fattr_init(fattr);
1594 fattr->label = NULL;
1595 }
1596 return fattr;
1597}
1598EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1599
1600struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1601{
1602 struct nfs_fattr *fattr = nfs_alloc_fattr();
1603
1604 if (!fattr)
1605 return NULL;
1606
1607 fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1608 if (IS_ERR(fattr->label)) {
1609 kfree(fattr);
1610 return NULL;
1611 }
1612
1613 return fattr;
1614}
1615EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1616
1617struct nfs_fh *nfs_alloc_fhandle(void)
1618{
1619 struct nfs_fh *fh;
1620
1621 fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1622 if (fh != NULL)
1623 fh->size = 0;
1624 return fh;
1625}
1626EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1627
1628#ifdef NFS_DEBUG
1629/*
1630 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1631 * in the same way that wireshark does
1632 *
1633 * @fh: file handle
1634 *
1635 * For debugging only.
1636 */
1637u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1638{
1639 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1640 * not on the result */
1641 return nfs_fhandle_hash(fh);
1642}
1643EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1644
1645/*
1646 * _nfs_display_fhandle - display an NFS file handle on the console
1647 *
1648 * @fh: file handle to display
1649 * @caption: display caption
1650 *
1651 * For debugging only.
1652 */
1653void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1654{
1655 unsigned short i;
1656
1657 if (fh == NULL || fh->size == 0) {
1658 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1659 return;
1660 }
1661
1662 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1663 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1664 for (i = 0; i < fh->size; i += 16) {
1665 __be32 *pos = (__be32 *)&fh->data[i];
1666
1667 switch ((fh->size - i - 1) >> 2) {
1668 case 0:
1669 printk(KERN_DEFAULT " %08x\n",
1670 be32_to_cpup(pos));
1671 break;
1672 case 1:
1673 printk(KERN_DEFAULT " %08x %08x\n",
1674 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1675 break;
1676 case 2:
1677 printk(KERN_DEFAULT " %08x %08x %08x\n",
1678 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1679 be32_to_cpup(pos + 2));
1680 break;
1681 default:
1682 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1683 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1684 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1685 }
1686 }
1687}
1688EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1689#endif
1690
1691/**
1692 * nfs_inode_attrs_cmp_generic - compare attributes
1693 * @fattr: attributes
1694 * @inode: pointer to inode
1695 *
1696 * Attempt to divine whether or not an RPC call reply carrying stale
1697 * attributes got scheduled after another call carrying updated ones.
1698 * Note also the check for wraparound of 'attr_gencount'
1699 *
1700 * The function returns '1' if it thinks the attributes in @fattr are
1701 * more recent than the ones cached in @inode. Otherwise it returns
1702 * the value '0'.
1703 */
1704static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1705 const struct inode *inode)
1706{
1707 unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1708
1709 return (long)(fattr->gencount - attr_gencount) > 0 ||
1710 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1711}
1712
1713/**
1714 * nfs_inode_attrs_cmp_monotonic - compare attributes
1715 * @fattr: attributes
1716 * @inode: pointer to inode
1717 *
1718 * Attempt to divine whether or not an RPC call reply carrying stale
1719 * attributes got scheduled after another call carrying updated ones.
1720 *
1721 * We assume that the server observes monotonic semantics for
1722 * the change attribute, so a larger value means that the attributes in
1723 * @fattr are more recent, in which case the function returns the
1724 * value '1'.
1725 * A return value of '0' indicates no measurable change
1726 * A return value of '-1' means that the attributes in @inode are
1727 * more recent.
1728 */
1729static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1730 const struct inode *inode)
1731{
1732 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1733 if (diff > 0)
1734 return 1;
1735 return diff == 0 ? 0 : -1;
1736}
1737
1738/**
1739 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1740 * @fattr: attributes
1741 * @inode: pointer to inode
1742 *
1743 * Attempt to divine whether or not an RPC call reply carrying stale
1744 * attributes got scheduled after another call carrying updated ones.
1745 *
1746 * We assume that the server observes strictly monotonic semantics for
1747 * the change attribute, so a larger value means that the attributes in
1748 * @fattr are more recent, in which case the function returns the
1749 * value '1'.
1750 * A return value of '-1' means that the attributes in @inode are
1751 * more recent or unchanged.
1752 */
1753static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1754 const struct inode *inode)
1755{
1756 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1757}
1758
1759/**
1760 * nfs_inode_attrs_cmp - compare attributes
1761 * @fattr: attributes
1762 * @inode: pointer to inode
1763 *
1764 * This function returns '1' if it thinks the attributes in @fattr are
1765 * more recent than the ones cached in @inode. It returns '-1' if
1766 * the attributes in @inode are more recent than the ones in @fattr,
1767 * and it returns 0 if not sure.
1768 */
1769static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1770 const struct inode *inode)
1771{
1772 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1773 return 1;
1774 switch (NFS_SERVER(inode)->change_attr_type) {
1775 case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1776 break;
1777 case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1778 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1779 break;
1780 return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1781 default:
1782 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1783 break;
1784 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1785 }
1786 return 0;
1787}
1788
1789/**
1790 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1791 * @fattr: attributes
1792 * @inode: pointer to inode
1793 *
1794 * Returns '1' if the last attribute update left the inode cached
1795 * attributes in a partially unrevalidated state, and @fattr
1796 * matches the change attribute of that partial update.
1797 * Otherwise returns '0'.
1798 */
1799static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1800 const struct inode *inode)
1801{
1802 const unsigned long check_valid =
1803 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1804 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1805 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1806 NFS_INO_INVALID_NLINK;
1807 unsigned long cache_validity = NFS_I(inode)->cache_validity;
1808 enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1809
1810 if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1811 !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1812 (cache_validity & check_valid) != 0 &&
1813 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1814 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1815 return 1;
1816 return 0;
1817}
1818
1819static void nfs_ooo_merge(struct nfs_inode *nfsi,
1820 u64 start, u64 end)
1821{
1822 int i, cnt;
1823
1824 if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)
1825 /* No point merging anything */
1826 return;
1827
1828 if (!nfsi->ooo) {
1829 nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC);
1830 if (!nfsi->ooo) {
1831 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1832 return;
1833 }
1834 nfsi->ooo->cnt = 0;
1835 }
1836
1837 /* add this range, merging if possible */
1838 cnt = nfsi->ooo->cnt;
1839 for (i = 0; i < cnt; i++) {
1840 if (end == nfsi->ooo->gap[i].start)
1841 end = nfsi->ooo->gap[i].end;
1842 else if (start == nfsi->ooo->gap[i].end)
1843 start = nfsi->ooo->gap[i].start;
1844 else
1845 continue;
1846 /* Remove 'i' from table and loop to insert the new range */
1847 cnt -= 1;
1848 nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt];
1849 i = -1;
1850 }
1851 if (start != end) {
1852 if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) {
1853 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1854 kfree(nfsi->ooo);
1855 nfsi->ooo = NULL;
1856 return;
1857 }
1858 nfsi->ooo->gap[cnt].start = start;
1859 nfsi->ooo->gap[cnt].end = end;
1860 cnt += 1;
1861 }
1862 nfsi->ooo->cnt = cnt;
1863}
1864
1865static void nfs_ooo_record(struct nfs_inode *nfsi,
1866 struct nfs_fattr *fattr)
1867{
1868 /* This reply was out-of-order, so record in the
1869 * pre/post change id, possibly cancelling
1870 * gaps created when iversion was jumpped forward.
1871 */
1872 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) &&
1873 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE))
1874 nfs_ooo_merge(nfsi,
1875 fattr->change_attr,
1876 fattr->pre_change_attr);
1877}
1878
1879static int nfs_refresh_inode_locked(struct inode *inode,
1880 struct nfs_fattr *fattr)
1881{
1882 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1883 int ret = 0;
1884
1885 trace_nfs_refresh_inode_enter(inode);
1886
1887 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1888 ret = nfs_update_inode(inode, fattr);
1889 else {
1890 nfs_ooo_record(NFS_I(inode), fattr);
1891
1892 if (attr_cmp == 0)
1893 ret = nfs_check_inode_attributes(inode, fattr);
1894 }
1895
1896 trace_nfs_refresh_inode_exit(inode, ret);
1897 return ret;
1898}
1899
1900/**
1901 * nfs_refresh_inode - try to update the inode attribute cache
1902 * @inode: pointer to inode
1903 * @fattr: updated attributes
1904 *
1905 * Check that an RPC call that returned attributes has not overlapped with
1906 * other recent updates of the inode metadata, then decide whether it is
1907 * safe to do a full update of the inode attributes, or whether just to
1908 * call nfs_check_inode_attributes.
1909 */
1910int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1911{
1912 int status;
1913
1914 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1915 return 0;
1916 spin_lock(&inode->i_lock);
1917 status = nfs_refresh_inode_locked(inode, fattr);
1918 spin_unlock(&inode->i_lock);
1919
1920 return status;
1921}
1922EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1923
1924static int nfs_post_op_update_inode_locked(struct inode *inode,
1925 struct nfs_fattr *fattr, unsigned int invalid)
1926{
1927 if (S_ISDIR(inode->i_mode))
1928 invalid |= NFS_INO_INVALID_DATA;
1929 nfs_set_cache_invalid(inode, invalid);
1930 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1931 return 0;
1932 return nfs_refresh_inode_locked(inode, fattr);
1933}
1934
1935/**
1936 * nfs_post_op_update_inode - try to update the inode attribute cache
1937 * @inode: pointer to inode
1938 * @fattr: updated attributes
1939 *
1940 * After an operation that has changed the inode metadata, mark the
1941 * attribute cache as being invalid, then try to update it.
1942 *
1943 * NB: if the server didn't return any post op attributes, this
1944 * function will force the retrieval of attributes before the next
1945 * NFS request. Thus it should be used only for operations that
1946 * are expected to change one or more attributes, to avoid
1947 * unnecessary NFS requests and trips through nfs_update_inode().
1948 */
1949int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1950{
1951 int status;
1952
1953 spin_lock(&inode->i_lock);
1954 nfs_fattr_set_barrier(fattr);
1955 status = nfs_post_op_update_inode_locked(inode, fattr,
1956 NFS_INO_INVALID_CHANGE
1957 | NFS_INO_INVALID_CTIME
1958 | NFS_INO_REVAL_FORCED);
1959 spin_unlock(&inode->i_lock);
1960
1961 return status;
1962}
1963EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1964
1965/**
1966 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1967 * @inode: pointer to inode
1968 * @fattr: updated attributes
1969 *
1970 * After an operation that has changed the inode metadata, mark the
1971 * attribute cache as being invalid, then try to update it. Fake up
1972 * weak cache consistency data, if none exist.
1973 *
1974 * This function is mainly designed to be used by the ->write_done() functions.
1975 */
1976int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1977{
1978 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1979 int status;
1980
1981 /* Don't do a WCC update if these attributes are already stale */
1982 if (attr_cmp < 0)
1983 return 0;
1984 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
1985 /* Record the pre/post change info before clearing PRECHANGE */
1986 nfs_ooo_record(NFS_I(inode), fattr);
1987 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1988 | NFS_ATTR_FATTR_PRESIZE
1989 | NFS_ATTR_FATTR_PREMTIME
1990 | NFS_ATTR_FATTR_PRECTIME);
1991 goto out_noforce;
1992 }
1993 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1994 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1995 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1996 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1997 }
1998 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1999 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
2000 fattr->pre_ctime = inode_get_ctime(inode);
2001 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
2002 }
2003 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
2004 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
2005 fattr->pre_mtime = inode_get_mtime(inode);
2006 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
2007 }
2008 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
2009 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
2010 fattr->pre_size = i_size_read(inode);
2011 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
2012 }
2013out_noforce:
2014 status = nfs_post_op_update_inode_locked(inode, fattr,
2015 NFS_INO_INVALID_CHANGE
2016 | NFS_INO_INVALID_CTIME
2017 | NFS_INO_INVALID_MTIME
2018 | NFS_INO_INVALID_BLOCKS);
2019 return status;
2020}
2021
2022/**
2023 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
2024 * @inode: pointer to inode
2025 * @fattr: updated attributes
2026 *
2027 * After an operation that has changed the inode metadata, mark the
2028 * attribute cache as being invalid, then try to update it. Fake up
2029 * weak cache consistency data, if none exist.
2030 *
2031 * This function is mainly designed to be used by the ->write_done() functions.
2032 */
2033int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
2034{
2035 int status;
2036
2037 spin_lock(&inode->i_lock);
2038 nfs_fattr_set_barrier(fattr);
2039 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
2040 spin_unlock(&inode->i_lock);
2041 return status;
2042}
2043EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
2044
2045
2046/*
2047 * Many nfs protocol calls return the new file attributes after
2048 * an operation. Here we update the inode to reflect the state
2049 * of the server's inode.
2050 *
2051 * This is a bit tricky because we have to make sure all dirty pages
2052 * have been sent off to the server before calling invalidate_inode_pages.
2053 * To make sure no other process adds more write requests while we try
2054 * our best to flush them, we make them sleep during the attribute refresh.
2055 *
2056 * A very similar scenario holds for the dir cache.
2057 */
2058static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2059{
2060 struct nfs_server *server = NFS_SERVER(inode);
2061 struct nfs_inode *nfsi = NFS_I(inode);
2062 loff_t cur_isize, new_isize;
2063 u64 fattr_supported = server->fattr_valid;
2064 unsigned long invalid = 0;
2065 unsigned long now = jiffies;
2066 unsigned long save_cache_validity;
2067 bool have_writers = nfs_file_has_buffered_writers(nfsi);
2068 bool cache_revalidated = true;
2069 bool attr_changed = false;
2070 bool have_delegation;
2071
2072 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
2073 __func__, inode->i_sb->s_id, inode->i_ino,
2074 nfs_display_fhandle_hash(NFS_FH(inode)),
2075 atomic_read(&inode->i_count), fattr->valid);
2076
2077 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2078 /* Only a mounted-on-fileid? Just exit */
2079 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2080 return 0;
2081 /* Has the inode gone and changed behind our back? */
2082 } else if (nfsi->fileid != fattr->fileid) {
2083 /* Is this perhaps the mounted-on fileid? */
2084 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2085 nfsi->fileid == fattr->mounted_on_fileid)
2086 return 0;
2087 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2088 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2089 NFS_SERVER(inode)->nfs_client->cl_hostname,
2090 inode->i_sb->s_id, (long long)nfsi->fileid,
2091 (long long)fattr->fileid);
2092 goto out_err;
2093 }
2094
2095 /*
2096 * Make sure the inode's type hasn't changed.
2097 */
2098 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2099 /*
2100 * Big trouble! The inode has become a different object.
2101 */
2102 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2103 __func__, inode->i_ino, inode->i_mode, fattr->mode);
2104 goto out_err;
2105 }
2106
2107 /* Update the fsid? */
2108 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2109 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2110 !IS_AUTOMOUNT(inode))
2111 server->fsid = fattr->fsid;
2112
2113 /* Save the delegation state before clearing cache_validity */
2114 have_delegation = nfs_have_delegated_attributes(inode);
2115
2116 /*
2117 * Update the read time so we don't revalidate too often.
2118 */
2119 nfsi->read_cache_jiffies = fattr->time_start;
2120
2121 save_cache_validity = nfsi->cache_validity;
2122 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2123 | NFS_INO_INVALID_ATIME
2124 | NFS_INO_REVAL_FORCED
2125 | NFS_INO_INVALID_BLOCKS);
2126
2127 /* Do atomic weak cache consistency updates */
2128 nfs_wcc_update_inode(inode, fattr);
2129
2130 if (pnfs_layoutcommit_outstanding(inode)) {
2131 nfsi->cache_validity |=
2132 save_cache_validity &
2133 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2134 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2135 NFS_INO_INVALID_BLOCKS);
2136 cache_revalidated = false;
2137 }
2138
2139 /* More cache consistency checks */
2140 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2141 if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 &&
2142 nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) {
2143 /* There is one remaining gap that hasn't been
2144 * merged into iversion - do that now.
2145 */
2146 inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start);
2147 kfree(nfsi->ooo);
2148 nfsi->ooo = NULL;
2149 }
2150 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2151 /* Could it be a race with writeback? */
2152 if (!(have_writers || have_delegation)) {
2153 invalid |= NFS_INO_INVALID_DATA
2154 | NFS_INO_INVALID_ACCESS
2155 | NFS_INO_INVALID_ACL
2156 | NFS_INO_INVALID_XATTR;
2157 /* Force revalidate of all attributes */
2158 save_cache_validity |= NFS_INO_INVALID_CTIME
2159 | NFS_INO_INVALID_MTIME
2160 | NFS_INO_INVALID_SIZE
2161 | NFS_INO_INVALID_BLOCKS
2162 | NFS_INO_INVALID_NLINK
2163 | NFS_INO_INVALID_MODE
2164 | NFS_INO_INVALID_OTHER;
2165 if (S_ISDIR(inode->i_mode))
2166 nfs_force_lookup_revalidate(inode);
2167 attr_changed = true;
2168 dprintk("NFS: change_attr change on server for file %s/%ld\n",
2169 inode->i_sb->s_id,
2170 inode->i_ino);
2171 } else if (!have_delegation) {
2172 nfs_ooo_record(nfsi, fattr);
2173 nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode),
2174 fattr->change_attr);
2175 }
2176 inode_set_iversion_raw(inode, fattr->change_attr);
2177 }
2178 } else {
2179 nfsi->cache_validity |=
2180 save_cache_validity & NFS_INO_INVALID_CHANGE;
2181 if (!have_delegation ||
2182 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2183 cache_revalidated = false;
2184 }
2185
2186 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2187 inode_set_mtime_to_ts(inode, fattr->mtime);
2188 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2189 nfsi->cache_validity |=
2190 save_cache_validity & NFS_INO_INVALID_MTIME;
2191
2192 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2193 inode_set_ctime_to_ts(inode, fattr->ctime);
2194 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2195 nfsi->cache_validity |=
2196 save_cache_validity & NFS_INO_INVALID_CTIME;
2197
2198 /* Check if our cached file size is stale */
2199 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2200 new_isize = nfs_size_to_loff_t(fattr->size);
2201 cur_isize = i_size_read(inode);
2202 if (new_isize != cur_isize && !have_delegation) {
2203 /* Do we perhaps have any outstanding writes, or has
2204 * the file grown beyond our last write? */
2205 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2206 trace_nfs_size_update(inode, new_isize);
2207 i_size_write(inode, new_isize);
2208 if (!have_writers)
2209 invalid |= NFS_INO_INVALID_DATA;
2210 }
2211 }
2212 if (new_isize == 0 &&
2213 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2214 NFS_ATTR_FATTR_BLOCKS_USED))) {
2215 fattr->du.nfs3.used = 0;
2216 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2217 }
2218 } else
2219 nfsi->cache_validity |=
2220 save_cache_validity & NFS_INO_INVALID_SIZE;
2221
2222 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2223 inode_set_atime_to_ts(inode, fattr->atime);
2224 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2225 nfsi->cache_validity |=
2226 save_cache_validity & NFS_INO_INVALID_ATIME;
2227
2228 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2229 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2230 umode_t newmode = inode->i_mode & S_IFMT;
2231 newmode |= fattr->mode & S_IALLUGO;
2232 inode->i_mode = newmode;
2233 invalid |= NFS_INO_INVALID_ACCESS
2234 | NFS_INO_INVALID_ACL;
2235 }
2236 } else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2237 nfsi->cache_validity |=
2238 save_cache_validity & NFS_INO_INVALID_MODE;
2239
2240 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2241 if (!uid_eq(inode->i_uid, fattr->uid)) {
2242 invalid |= NFS_INO_INVALID_ACCESS
2243 | NFS_INO_INVALID_ACL;
2244 inode->i_uid = fattr->uid;
2245 }
2246 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2247 nfsi->cache_validity |=
2248 save_cache_validity & NFS_INO_INVALID_OTHER;
2249
2250 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2251 if (!gid_eq(inode->i_gid, fattr->gid)) {
2252 invalid |= NFS_INO_INVALID_ACCESS
2253 | NFS_INO_INVALID_ACL;
2254 inode->i_gid = fattr->gid;
2255 }
2256 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2257 nfsi->cache_validity |=
2258 save_cache_validity & NFS_INO_INVALID_OTHER;
2259
2260 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2261 if (inode->i_nlink != fattr->nlink)
2262 set_nlink(inode, fattr->nlink);
2263 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2264 nfsi->cache_validity |=
2265 save_cache_validity & NFS_INO_INVALID_NLINK;
2266
2267 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2268 /*
2269 * report the blocks in 512byte units
2270 */
2271 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2272 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2273 nfsi->cache_validity |=
2274 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2275
2276 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2277 inode->i_blocks = fattr->du.nfs2.blocks;
2278 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2279 nfsi->cache_validity |=
2280 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2281
2282 /* Update attrtimeo value if we're out of the unstable period */
2283 if (attr_changed) {
2284 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2285 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2286 nfsi->attrtimeo_timestamp = now;
2287 /* Set barrier to be more recent than all outstanding updates */
2288 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2289 } else {
2290 if (cache_revalidated) {
2291 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2292 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2293 nfsi->attrtimeo <<= 1;
2294 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2295 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2296 }
2297 nfsi->attrtimeo_timestamp = now;
2298 }
2299 /* Set the barrier to be more recent than this fattr */
2300 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2301 nfsi->attr_gencount = fattr->gencount;
2302 }
2303
2304 /* Don't invalidate the data if we were to blame */
2305 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2306 || S_ISLNK(inode->i_mode)))
2307 invalid &= ~NFS_INO_INVALID_DATA;
2308 nfs_set_cache_invalid(inode, invalid);
2309
2310 return 0;
2311 out_err:
2312 /*
2313 * No need to worry about unhashing the dentry, as the
2314 * lookup validation will know that the inode is bad.
2315 * (But we fall through to invalidate the caches.)
2316 */
2317 nfs_set_inode_stale_locked(inode);
2318 return -ESTALE;
2319}
2320
2321struct inode *nfs_alloc_inode(struct super_block *sb)
2322{
2323 struct nfs_inode *nfsi;
2324 nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2325 if (!nfsi)
2326 return NULL;
2327 nfsi->flags = 0UL;
2328 nfsi->cache_validity = 0UL;
2329 nfsi->ooo = NULL;
2330#if IS_ENABLED(CONFIG_NFS_V4)
2331 nfsi->nfs4_acl = NULL;
2332#endif /* CONFIG_NFS_V4 */
2333#ifdef CONFIG_NFS_V4_2
2334 nfsi->xattr_cache = NULL;
2335#endif
2336 nfs_netfs_inode_init(nfsi);
2337
2338 return &nfsi->vfs_inode;
2339}
2340EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2341
2342void nfs_free_inode(struct inode *inode)
2343{
2344 kfree(NFS_I(inode)->ooo);
2345 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2346}
2347EXPORT_SYMBOL_GPL(nfs_free_inode);
2348
2349static inline void nfs4_init_once(struct nfs_inode *nfsi)
2350{
2351#if IS_ENABLED(CONFIG_NFS_V4)
2352 INIT_LIST_HEAD(&nfsi->open_states);
2353 nfsi->delegation = NULL;
2354 init_rwsem(&nfsi->rwsem);
2355 nfsi->layout = NULL;
2356#endif
2357}
2358
2359static void init_once(void *foo)
2360{
2361 struct nfs_inode *nfsi = foo;
2362
2363 inode_init_once(&nfsi->vfs_inode);
2364 INIT_LIST_HEAD(&nfsi->open_files);
2365 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2366 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2367 nfs4_init_once(nfsi);
2368}
2369
2370static int __init nfs_init_inodecache(void)
2371{
2372 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2373 sizeof(struct nfs_inode),
2374 0, (SLAB_RECLAIM_ACCOUNT|
2375 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2376 init_once);
2377 if (nfs_inode_cachep == NULL)
2378 return -ENOMEM;
2379
2380 return 0;
2381}
2382
2383static void nfs_destroy_inodecache(void)
2384{
2385 /*
2386 * Make sure all delayed rcu free inodes are flushed before we
2387 * destroy cache.
2388 */
2389 rcu_barrier();
2390 kmem_cache_destroy(nfs_inode_cachep);
2391}
2392
2393struct workqueue_struct *nfsiod_workqueue;
2394EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2395
2396/*
2397 * start up the nfsiod workqueue
2398 */
2399static int nfsiod_start(void)
2400{
2401 struct workqueue_struct *wq;
2402 dprintk("RPC: creating workqueue nfsiod\n");
2403 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2404 if (wq == NULL)
2405 return -ENOMEM;
2406 nfsiod_workqueue = wq;
2407 return 0;
2408}
2409
2410/*
2411 * Destroy the nfsiod workqueue
2412 */
2413static void nfsiod_stop(void)
2414{
2415 struct workqueue_struct *wq;
2416
2417 wq = nfsiod_workqueue;
2418 if (wq == NULL)
2419 return;
2420 nfsiod_workqueue = NULL;
2421 destroy_workqueue(wq);
2422}
2423
2424unsigned int nfs_net_id;
2425EXPORT_SYMBOL_GPL(nfs_net_id);
2426
2427static int nfs_net_init(struct net *net)
2428{
2429 nfs_clients_init(net);
2430 return nfs_fs_proc_net_init(net);
2431}
2432
2433static void nfs_net_exit(struct net *net)
2434{
2435 nfs_fs_proc_net_exit(net);
2436 nfs_clients_exit(net);
2437}
2438
2439static struct pernet_operations nfs_net_ops = {
2440 .init = nfs_net_init,
2441 .exit = nfs_net_exit,
2442 .id = &nfs_net_id,
2443 .size = sizeof(struct nfs_net),
2444};
2445
2446/*
2447 * Initialize NFS
2448 */
2449static int __init init_nfs_fs(void)
2450{
2451 int err;
2452
2453 err = nfs_sysfs_init();
2454 if (err < 0)
2455 goto out10;
2456
2457 err = register_pernet_subsys(&nfs_net_ops);
2458 if (err < 0)
2459 goto out9;
2460
2461 err = nfsiod_start();
2462 if (err)
2463 goto out7;
2464
2465 err = nfs_fs_proc_init();
2466 if (err)
2467 goto out6;
2468
2469 err = nfs_init_nfspagecache();
2470 if (err)
2471 goto out5;
2472
2473 err = nfs_init_inodecache();
2474 if (err)
2475 goto out4;
2476
2477 err = nfs_init_readpagecache();
2478 if (err)
2479 goto out3;
2480
2481 err = nfs_init_writepagecache();
2482 if (err)
2483 goto out2;
2484
2485 err = nfs_init_directcache();
2486 if (err)
2487 goto out1;
2488
2489 rpc_proc_register(&init_net, &nfs_rpcstat);
2490
2491 err = register_nfs_fs();
2492 if (err)
2493 goto out0;
2494
2495 return 0;
2496out0:
2497 rpc_proc_unregister(&init_net, "nfs");
2498 nfs_destroy_directcache();
2499out1:
2500 nfs_destroy_writepagecache();
2501out2:
2502 nfs_destroy_readpagecache();
2503out3:
2504 nfs_destroy_inodecache();
2505out4:
2506 nfs_destroy_nfspagecache();
2507out5:
2508 nfs_fs_proc_exit();
2509out6:
2510 nfsiod_stop();
2511out7:
2512 unregister_pernet_subsys(&nfs_net_ops);
2513out9:
2514 nfs_sysfs_exit();
2515out10:
2516 return err;
2517}
2518
2519static void __exit exit_nfs_fs(void)
2520{
2521 nfs_destroy_directcache();
2522 nfs_destroy_writepagecache();
2523 nfs_destroy_readpagecache();
2524 nfs_destroy_inodecache();
2525 nfs_destroy_nfspagecache();
2526 unregister_pernet_subsys(&nfs_net_ops);
2527 rpc_proc_unregister(&init_net, "nfs");
2528 unregister_nfs_fs();
2529 nfs_fs_proc_exit();
2530 nfsiod_stop();
2531 nfs_sysfs_exit();
2532}
2533
2534/* Not quite true; I just maintain it */
2535MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2536MODULE_LICENSE("GPL");
2537module_param(enable_ino64, bool, 0644);
2538
2539module_init(init_nfs_fs)
2540module_exit(exit_nfs_fs)