Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/*
2 * linux/fs/nfs/inode.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * nfs inode and superblock handling functions
7 *
8 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
10 *
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
13 *
14 */
15
16#include <linux/module.h>
17#include <linux/init.h>
18#include <linux/sched/signal.h>
19#include <linux/time.h>
20#include <linux/kernel.h>
21#include <linux/mm.h>
22#include <linux/string.h>
23#include <linux/stat.h>
24#include <linux/errno.h>
25#include <linux/unistd.h>
26#include <linux/sunrpc/clnt.h>
27#include <linux/sunrpc/stats.h>
28#include <linux/sunrpc/metrics.h>
29#include <linux/nfs_fs.h>
30#include <linux/nfs_mount.h>
31#include <linux/nfs4_mount.h>
32#include <linux/lockd/bind.h>
33#include <linux/seq_file.h>
34#include <linux/mount.h>
35#include <linux/vfs.h>
36#include <linux/inet.h>
37#include <linux/nfs_xdr.h>
38#include <linux/slab.h>
39#include <linux/compat.h>
40#include <linux/freezer.h>
41#include <linux/uaccess.h>
42#include <linux/iversion.h>
43
44#include "nfs4_fs.h"
45#include "callback.h"
46#include "delegation.h"
47#include "iostat.h"
48#include "internal.h"
49#include "fscache.h"
50#include "pnfs.h"
51#include "nfs.h"
52#include "netns.h"
53
54#include "nfstrace.h"
55
56#define NFSDBG_FACILITY NFSDBG_VFS
57
58#define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
59
60/* Default is to see 64-bit inode numbers */
61static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
62
63static void nfs_invalidate_inode(struct inode *);
64static int nfs_update_inode(struct inode *, struct nfs_fattr *);
65
66static struct kmem_cache * nfs_inode_cachep;
67
68static inline unsigned long
69nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
70{
71 return nfs_fileid_to_ino_t(fattr->fileid);
72}
73
74static int nfs_wait_killable(int mode)
75{
76 freezable_schedule_unsafe();
77 if (signal_pending_state(mode, current))
78 return -ERESTARTSYS;
79 return 0;
80}
81
82int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
83{
84 return nfs_wait_killable(mode);
85}
86EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
87
88/**
89 * nfs_compat_user_ino64 - returns the user-visible inode number
90 * @fileid: 64-bit fileid
91 *
92 * This function returns a 32-bit inode number if the boot parameter
93 * nfs.enable_ino64 is zero.
94 */
95u64 nfs_compat_user_ino64(u64 fileid)
96{
97#ifdef CONFIG_COMPAT
98 compat_ulong_t ino;
99#else
100 unsigned long ino;
101#endif
102
103 if (enable_ino64)
104 return fileid;
105 ino = fileid;
106 if (sizeof(ino) < sizeof(fileid))
107 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
108 return ino;
109}
110
111int nfs_drop_inode(struct inode *inode)
112{
113 return NFS_STALE(inode) || generic_drop_inode(inode);
114}
115EXPORT_SYMBOL_GPL(nfs_drop_inode);
116
117void nfs_clear_inode(struct inode *inode)
118{
119 /*
120 * The following should never happen...
121 */
122 WARN_ON_ONCE(nfs_have_writebacks(inode));
123 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
124 nfs_zap_acl_cache(inode);
125 nfs_access_zap_cache(inode);
126 nfs_fscache_clear_inode(inode);
127}
128EXPORT_SYMBOL_GPL(nfs_clear_inode);
129
130void nfs_evict_inode(struct inode *inode)
131{
132 truncate_inode_pages_final(&inode->i_data);
133 clear_inode(inode);
134 nfs_clear_inode(inode);
135}
136
137int nfs_sync_inode(struct inode *inode)
138{
139 inode_dio_wait(inode);
140 return nfs_wb_all(inode);
141}
142EXPORT_SYMBOL_GPL(nfs_sync_inode);
143
144/**
145 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
146 * @mapping: pointer to struct address_space
147 */
148int nfs_sync_mapping(struct address_space *mapping)
149{
150 int ret = 0;
151
152 if (mapping->nrpages != 0) {
153 unmap_mapping_range(mapping, 0, 0, 0);
154 ret = nfs_wb_all(mapping->host);
155 }
156 return ret;
157}
158
159static int nfs_attribute_timeout(struct inode *inode)
160{
161 struct nfs_inode *nfsi = NFS_I(inode);
162
163 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
164}
165
166static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
167{
168 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
169
170 /* Special case for the pagecache or access cache */
171 if (flags == NFS_INO_REVAL_PAGECACHE &&
172 !(cache_validity & NFS_INO_REVAL_FORCED))
173 return false;
174 return (cache_validity & flags) != 0;
175}
176
177static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
178{
179 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
180
181 if ((cache_validity & flags) != 0)
182 return true;
183 if (nfs_attribute_timeout(inode))
184 return true;
185 return false;
186}
187
188bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
189{
190 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
191 return nfs_check_cache_invalid_delegated(inode, flags);
192
193 return nfs_check_cache_invalid_not_delegated(inode, flags);
194}
195
196static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
197{
198 struct nfs_inode *nfsi = NFS_I(inode);
199 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
200
201 if (have_delegation) {
202 if (!(flags & NFS_INO_REVAL_FORCED))
203 flags &= ~NFS_INO_INVALID_OTHER;
204 flags &= ~(NFS_INO_INVALID_CHANGE
205 | NFS_INO_INVALID_SIZE
206 | NFS_INO_REVAL_PAGECACHE);
207 }
208
209 if (inode->i_mapping->nrpages == 0)
210 flags &= ~NFS_INO_INVALID_DATA;
211 nfsi->cache_validity |= flags;
212 if (flags & NFS_INO_INVALID_DATA)
213 nfs_fscache_invalidate(inode);
214}
215
216/*
217 * Invalidate the local caches
218 */
219static void nfs_zap_caches_locked(struct inode *inode)
220{
221 struct nfs_inode *nfsi = NFS_I(inode);
222 int mode = inode->i_mode;
223
224 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
225
226 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
227 nfsi->attrtimeo_timestamp = jiffies;
228
229 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
230 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
231 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
232 | NFS_INO_INVALID_DATA
233 | NFS_INO_INVALID_ACCESS
234 | NFS_INO_INVALID_ACL
235 | NFS_INO_REVAL_PAGECACHE);
236 } else
237 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
238 | NFS_INO_INVALID_ACCESS
239 | NFS_INO_INVALID_ACL
240 | NFS_INO_REVAL_PAGECACHE);
241 nfs_zap_label_cache_locked(nfsi);
242}
243
244void nfs_zap_caches(struct inode *inode)
245{
246 spin_lock(&inode->i_lock);
247 nfs_zap_caches_locked(inode);
248 spin_unlock(&inode->i_lock);
249}
250
251void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
252{
253 if (mapping->nrpages != 0) {
254 spin_lock(&inode->i_lock);
255 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
256 spin_unlock(&inode->i_lock);
257 }
258}
259
260void nfs_zap_acl_cache(struct inode *inode)
261{
262 void (*clear_acl_cache)(struct inode *);
263
264 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
265 if (clear_acl_cache != NULL)
266 clear_acl_cache(inode);
267 spin_lock(&inode->i_lock);
268 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
269 spin_unlock(&inode->i_lock);
270}
271EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
272
273void nfs_invalidate_atime(struct inode *inode)
274{
275 spin_lock(&inode->i_lock);
276 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
277 spin_unlock(&inode->i_lock);
278}
279EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
280
281/*
282 * Invalidate, but do not unhash, the inode.
283 * NB: must be called with inode->i_lock held!
284 */
285static void nfs_invalidate_inode(struct inode *inode)
286{
287 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
288 nfs_zap_caches_locked(inode);
289}
290
291struct nfs_find_desc {
292 struct nfs_fh *fh;
293 struct nfs_fattr *fattr;
294};
295
296/*
297 * In NFSv3 we can have 64bit inode numbers. In order to support
298 * this, and re-exported directories (also seen in NFSv2)
299 * we are forced to allow 2 different inodes to have the same
300 * i_ino.
301 */
302static int
303nfs_find_actor(struct inode *inode, void *opaque)
304{
305 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
306 struct nfs_fh *fh = desc->fh;
307 struct nfs_fattr *fattr = desc->fattr;
308
309 if (NFS_FILEID(inode) != fattr->fileid)
310 return 0;
311 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
312 return 0;
313 if (nfs_compare_fh(NFS_FH(inode), fh))
314 return 0;
315 if (is_bad_inode(inode) || NFS_STALE(inode))
316 return 0;
317 return 1;
318}
319
320static int
321nfs_init_locked(struct inode *inode, void *opaque)
322{
323 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
324 struct nfs_fattr *fattr = desc->fattr;
325
326 set_nfs_fileid(inode, fattr->fileid);
327 inode->i_mode = fattr->mode;
328 nfs_copy_fh(NFS_FH(inode), desc->fh);
329 return 0;
330}
331
332#ifdef CONFIG_NFS_V4_SECURITY_LABEL
333static void nfs_clear_label_invalid(struct inode *inode)
334{
335 spin_lock(&inode->i_lock);
336 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
337 spin_unlock(&inode->i_lock);
338}
339
340void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
341 struct nfs4_label *label)
342{
343 int error;
344
345 if (label == NULL)
346 return;
347
348 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
349 error = security_inode_notifysecctx(inode, label->label,
350 label->len);
351 if (error)
352 printk(KERN_ERR "%s() %s %d "
353 "security_inode_notifysecctx() %d\n",
354 __func__,
355 (char *)label->label,
356 label->len, error);
357 nfs_clear_label_invalid(inode);
358 }
359}
360
361struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
362{
363 struct nfs4_label *label = NULL;
364 int minor_version = server->nfs_client->cl_minorversion;
365
366 if (minor_version < 2)
367 return label;
368
369 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
370 return label;
371
372 label = kzalloc(sizeof(struct nfs4_label), flags);
373 if (label == NULL)
374 return ERR_PTR(-ENOMEM);
375
376 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
377 if (label->label == NULL) {
378 kfree(label);
379 return ERR_PTR(-ENOMEM);
380 }
381 label->len = NFS4_MAXLABELLEN;
382
383 return label;
384}
385EXPORT_SYMBOL_GPL(nfs4_label_alloc);
386#else
387void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
388 struct nfs4_label *label)
389{
390}
391#endif
392EXPORT_SYMBOL_GPL(nfs_setsecurity);
393
394/* Search for inode identified by fh, fileid and i_mode in inode cache. */
395struct inode *
396nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
397{
398 struct nfs_find_desc desc = {
399 .fh = fh,
400 .fattr = fattr,
401 };
402 struct inode *inode;
403 unsigned long hash;
404
405 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
406 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
407 return NULL;
408
409 hash = nfs_fattr_to_ino_t(fattr);
410 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
411
412 dprintk("%s: returning %p\n", __func__, inode);
413 return inode;
414}
415
416/*
417 * This is our front-end to iget that looks up inodes by file handle
418 * instead of inode number.
419 */
420struct inode *
421nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
422{
423 struct nfs_find_desc desc = {
424 .fh = fh,
425 .fattr = fattr
426 };
427 struct inode *inode = ERR_PTR(-ENOENT);
428 unsigned long hash;
429
430 nfs_attr_check_mountpoint(sb, fattr);
431
432 if (nfs_attr_use_mounted_on_fileid(fattr))
433 fattr->fileid = fattr->mounted_on_fileid;
434 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
435 goto out_no_inode;
436 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
437 goto out_no_inode;
438
439 hash = nfs_fattr_to_ino_t(fattr);
440
441 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
442 if (inode == NULL) {
443 inode = ERR_PTR(-ENOMEM);
444 goto out_no_inode;
445 }
446
447 if (inode->i_state & I_NEW) {
448 struct nfs_inode *nfsi = NFS_I(inode);
449 unsigned long now = jiffies;
450
451 /* We set i_ino for the few things that still rely on it,
452 * such as stat(2) */
453 inode->i_ino = hash;
454
455 /* We can't support update_atime(), since the server will reset it */
456 inode->i_flags |= S_NOATIME|S_NOCMTIME;
457 inode->i_mode = fattr->mode;
458 nfsi->cache_validity = 0;
459 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
460 && nfs_server_capable(inode, NFS_CAP_MODE))
461 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
462 /* Why so? Because we want revalidate for devices/FIFOs, and
463 * that's precisely what we have in nfs_file_inode_operations.
464 */
465 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
466 if (S_ISREG(inode->i_mode)) {
467 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
468 inode->i_data.a_ops = &nfs_file_aops;
469 } else if (S_ISDIR(inode->i_mode)) {
470 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
471 inode->i_fop = &nfs_dir_operations;
472 inode->i_data.a_ops = &nfs_dir_aops;
473 /* Deal with crossing mountpoints */
474 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
475 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
476 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
477 inode->i_op = &nfs_referral_inode_operations;
478 else
479 inode->i_op = &nfs_mountpoint_inode_operations;
480 inode->i_fop = NULL;
481 inode->i_flags |= S_AUTOMOUNT;
482 }
483 } else if (S_ISLNK(inode->i_mode)) {
484 inode->i_op = &nfs_symlink_inode_operations;
485 inode_nohighmem(inode);
486 } else
487 init_special_inode(inode, inode->i_mode, fattr->rdev);
488
489 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
490 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
491 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
492 inode_set_iversion_raw(inode, 0);
493 inode->i_size = 0;
494 clear_nlink(inode);
495 inode->i_uid = make_kuid(&init_user_ns, -2);
496 inode->i_gid = make_kgid(&init_user_ns, -2);
497 inode->i_blocks = 0;
498 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
499 nfsi->write_io = 0;
500 nfsi->read_io = 0;
501
502 nfsi->read_cache_jiffies = fattr->time_start;
503 nfsi->attr_gencount = fattr->gencount;
504 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
505 inode->i_atime = timespec_to_timespec64(fattr->atime);
506 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
507 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
508 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
509 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
510 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
511 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
512 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
513 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
514 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
515 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
516 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
517 inode_set_iversion_raw(inode, fattr->change_attr);
518 else
519 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
520 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
521 inode->i_size = nfs_size_to_loff_t(fattr->size);
522 else
523 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
524 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
525 set_nlink(inode, fattr->nlink);
526 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
527 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
528 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
529 inode->i_uid = fattr->uid;
530 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
531 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
532 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
533 inode->i_gid = fattr->gid;
534 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
535 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
536 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
537 inode->i_blocks = fattr->du.nfs2.blocks;
538 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
539 /*
540 * report the blocks in 512byte units
541 */
542 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
543 }
544
545 if (nfsi->cache_validity != 0)
546 nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
547
548 nfs_setsecurity(inode, fattr, label);
549
550 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
551 nfsi->attrtimeo_timestamp = now;
552 nfsi->access_cache = RB_ROOT;
553
554 nfs_fscache_init_inode(inode);
555
556 unlock_new_inode(inode);
557 } else {
558 int err = nfs_refresh_inode(inode, fattr);
559 if (err < 0) {
560 iput(inode);
561 inode = ERR_PTR(err);
562 goto out_no_inode;
563 }
564 }
565 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
566 inode->i_sb->s_id,
567 (unsigned long long)NFS_FILEID(inode),
568 nfs_display_fhandle_hash(fh),
569 atomic_read(&inode->i_count));
570
571out:
572 return inode;
573
574out_no_inode:
575 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
576 goto out;
577}
578EXPORT_SYMBOL_GPL(nfs_fhget);
579
580#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)
581
582int
583nfs_setattr(struct dentry *dentry, struct iattr *attr)
584{
585 struct inode *inode = d_inode(dentry);
586 struct nfs_fattr *fattr;
587 int error = 0;
588
589 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
590
591 /* skip mode change if it's just for clearing setuid/setgid */
592 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
593 attr->ia_valid &= ~ATTR_MODE;
594
595 if (attr->ia_valid & ATTR_SIZE) {
596 BUG_ON(!S_ISREG(inode->i_mode));
597
598 error = inode_newsize_ok(inode, attr->ia_size);
599 if (error)
600 return error;
601
602 if (attr->ia_size == i_size_read(inode))
603 attr->ia_valid &= ~ATTR_SIZE;
604 }
605
606 /* Optimization: if the end result is no change, don't RPC */
607 attr->ia_valid &= NFS_VALID_ATTRS;
608 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
609 return 0;
610
611 trace_nfs_setattr_enter(inode);
612
613 /* Write all dirty data */
614 if (S_ISREG(inode->i_mode))
615 nfs_sync_inode(inode);
616
617 fattr = nfs_alloc_fattr();
618 if (fattr == NULL) {
619 error = -ENOMEM;
620 goto out;
621 }
622
623 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
624 if (error == 0)
625 error = nfs_refresh_inode(inode, fattr);
626 nfs_free_fattr(fattr);
627out:
628 trace_nfs_setattr_exit(inode, error);
629 return error;
630}
631EXPORT_SYMBOL_GPL(nfs_setattr);
632
633/**
634 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
635 * @inode: inode of the file used
636 * @offset: file offset to start truncating
637 *
638 * This is a copy of the common vmtruncate, but with the locking
639 * corrected to take into account the fact that NFS requires
640 * inode->i_size to be updated under the inode->i_lock.
641 * Note: must be called with inode->i_lock held!
642 */
643static int nfs_vmtruncate(struct inode * inode, loff_t offset)
644{
645 int err;
646
647 err = inode_newsize_ok(inode, offset);
648 if (err)
649 goto out;
650
651 i_size_write(inode, offset);
652 /* Optimisation */
653 if (offset == 0)
654 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
655 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
656
657 spin_unlock(&inode->i_lock);
658 truncate_pagecache(inode, offset);
659 spin_lock(&inode->i_lock);
660out:
661 return err;
662}
663
664/**
665 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
666 * @inode: pointer to struct inode
667 * @attr: pointer to struct iattr
668 * @fattr: pointer to struct nfs_fattr
669 *
670 * Note: we do this in the *proc.c in order to ensure that
671 * it works for things like exclusive creates too.
672 */
673void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
674 struct nfs_fattr *fattr)
675{
676 /* Barrier: bump the attribute generation count. */
677 nfs_fattr_set_barrier(fattr);
678
679 spin_lock(&inode->i_lock);
680 NFS_I(inode)->attr_gencount = fattr->gencount;
681 if ((attr->ia_valid & ATTR_SIZE) != 0) {
682 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
683 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
684 nfs_vmtruncate(inode, attr->ia_size);
685 }
686 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
687 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
688 if ((attr->ia_valid & ATTR_MODE) != 0) {
689 int mode = attr->ia_mode & S_IALLUGO;
690 mode |= inode->i_mode & ~S_IALLUGO;
691 inode->i_mode = mode;
692 }
693 if ((attr->ia_valid & ATTR_UID) != 0)
694 inode->i_uid = attr->ia_uid;
695 if ((attr->ia_valid & ATTR_GID) != 0)
696 inode->i_gid = attr->ia_gid;
697 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
698 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
699 else
700 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
701 | NFS_INO_INVALID_CTIME);
702 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
703 | NFS_INO_INVALID_ACL);
704 }
705 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
706 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
707 | NFS_INO_INVALID_CTIME);
708 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
709 inode->i_atime = timespec_to_timespec64(fattr->atime);
710 else if (attr->ia_valid & ATTR_ATIME_SET)
711 inode->i_atime = attr->ia_atime;
712 else
713 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
714
715 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
716 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
717 else
718 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
719 | NFS_INO_INVALID_CTIME);
720 }
721 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
722 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
723 | NFS_INO_INVALID_CTIME);
724 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
725 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
726 else if (attr->ia_valid & ATTR_MTIME_SET)
727 inode->i_mtime = attr->ia_mtime;
728 else
729 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
730
731 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
732 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
733 else
734 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
735 | NFS_INO_INVALID_CTIME);
736 }
737 if (fattr->valid)
738 nfs_update_inode(inode, fattr);
739 spin_unlock(&inode->i_lock);
740}
741EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
742
743static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
744{
745 struct dentry *parent;
746
747 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
748 return;
749 parent = dget_parent(dentry);
750 nfs_force_use_readdirplus(d_inode(parent));
751 dput(parent);
752}
753
754static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
755{
756 struct dentry *parent;
757
758 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
759 return;
760 parent = dget_parent(dentry);
761 nfs_advise_use_readdirplus(d_inode(parent));
762 dput(parent);
763}
764
765static bool nfs_need_revalidate_inode(struct inode *inode)
766{
767 if (NFS_I(inode)->cache_validity &
768 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
769 return true;
770 if (nfs_attribute_cache_expired(inode))
771 return true;
772 return false;
773}
774
775int nfs_getattr(const struct path *path, struct kstat *stat,
776 u32 request_mask, unsigned int query_flags)
777{
778 struct inode *inode = d_inode(path->dentry);
779 struct nfs_server *server = NFS_SERVER(inode);
780 unsigned long cache_validity;
781 int err = 0;
782 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
783 bool do_update = false;
784
785 trace_nfs_getattr_enter(inode);
786
787 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync)
788 goto out_no_update;
789
790 /* Flush out writes to the server in order to update c/mtime. */
791 if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
792 S_ISREG(inode->i_mode)) {
793 err = filemap_write_and_wait(inode->i_mapping);
794 if (err)
795 goto out;
796 }
797
798 /*
799 * We may force a getattr if the user cares about atime.
800 *
801 * Note that we only have to check the vfsmount flags here:
802 * - NFS always sets S_NOATIME by so checking it would give a
803 * bogus result
804 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
805 * no point in checking those.
806 */
807 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
808 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
809 request_mask &= ~STATX_ATIME;
810
811 /* Is the user requesting attributes that might need revalidation? */
812 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
813 STATX_MTIME|STATX_UID|STATX_GID|
814 STATX_SIZE|STATX_BLOCKS)))
815 goto out_no_revalidate;
816
817 /* Check whether the cached attributes are stale */
818 do_update |= force_sync || nfs_attribute_cache_expired(inode);
819 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
820 do_update |= cache_validity &
821 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
822 if (request_mask & STATX_ATIME)
823 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
824 if (request_mask & (STATX_CTIME|STATX_MTIME))
825 do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
826 if (do_update) {
827 /* Update the attribute cache */
828 if (!(server->flags & NFS_MOUNT_NOAC))
829 nfs_readdirplus_parent_cache_miss(path->dentry);
830 else
831 nfs_readdirplus_parent_cache_hit(path->dentry);
832 err = __nfs_revalidate_inode(server, inode);
833 if (err)
834 goto out;
835 } else
836 nfs_readdirplus_parent_cache_hit(path->dentry);
837out_no_revalidate:
838 /* Only return attributes that were revalidated. */
839 stat->result_mask &= request_mask;
840out_no_update:
841 generic_fillattr(inode, stat);
842 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
843 if (S_ISDIR(inode->i_mode))
844 stat->blksize = NFS_SERVER(inode)->dtsize;
845out:
846 trace_nfs_getattr_exit(inode, err);
847 return err;
848}
849EXPORT_SYMBOL_GPL(nfs_getattr);
850
851static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
852{
853 refcount_set(&l_ctx->count, 1);
854 l_ctx->lockowner = current->files;
855 INIT_LIST_HEAD(&l_ctx->list);
856 atomic_set(&l_ctx->io_count, 0);
857}
858
859static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
860{
861 struct nfs_lock_context *pos;
862
863 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
864 if (pos->lockowner != current->files)
865 continue;
866 if (refcount_inc_not_zero(&pos->count))
867 return pos;
868 }
869 return NULL;
870}
871
872struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
873{
874 struct nfs_lock_context *res, *new = NULL;
875 struct inode *inode = d_inode(ctx->dentry);
876
877 rcu_read_lock();
878 res = __nfs_find_lock_context(ctx);
879 rcu_read_unlock();
880 if (res == NULL) {
881 new = kmalloc(sizeof(*new), GFP_KERNEL);
882 if (new == NULL)
883 return ERR_PTR(-ENOMEM);
884 nfs_init_lock_context(new);
885 spin_lock(&inode->i_lock);
886 res = __nfs_find_lock_context(ctx);
887 if (res == NULL) {
888 list_add_tail_rcu(&new->list, &ctx->lock_context.list);
889 new->open_context = ctx;
890 res = new;
891 new = NULL;
892 }
893 spin_unlock(&inode->i_lock);
894 kfree(new);
895 }
896 return res;
897}
898EXPORT_SYMBOL_GPL(nfs_get_lock_context);
899
900void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
901{
902 struct nfs_open_context *ctx = l_ctx->open_context;
903 struct inode *inode = d_inode(ctx->dentry);
904
905 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
906 return;
907 list_del_rcu(&l_ctx->list);
908 spin_unlock(&inode->i_lock);
909 kfree_rcu(l_ctx, rcu_head);
910}
911EXPORT_SYMBOL_GPL(nfs_put_lock_context);
912
913/**
914 * nfs_close_context - Common close_context() routine NFSv2/v3
915 * @ctx: pointer to context
916 * @is_sync: is this a synchronous close
917 *
918 * Ensure that the attributes are up to date if we're mounted
919 * with close-to-open semantics and we have cached data that will
920 * need to be revalidated on open.
921 */
922void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
923{
924 struct nfs_inode *nfsi;
925 struct inode *inode;
926 struct nfs_server *server;
927
928 if (!(ctx->mode & FMODE_WRITE))
929 return;
930 if (!is_sync)
931 return;
932 inode = d_inode(ctx->dentry);
933 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
934 return;
935 nfsi = NFS_I(inode);
936 if (inode->i_mapping->nrpages == 0)
937 return;
938 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
939 return;
940 if (!list_empty(&nfsi->open_files))
941 return;
942 server = NFS_SERVER(inode);
943 if (server->flags & NFS_MOUNT_NOCTO)
944 return;
945 nfs_revalidate_inode(server, inode);
946}
947EXPORT_SYMBOL_GPL(nfs_close_context);
948
949struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
950 fmode_t f_mode,
951 struct file *filp)
952{
953 struct nfs_open_context *ctx;
954 const struct cred *cred = get_current_cred();
955
956 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
957 if (!ctx) {
958 put_cred(cred);
959 return ERR_PTR(-ENOMEM);
960 }
961 nfs_sb_active(dentry->d_sb);
962 ctx->dentry = dget(dentry);
963 ctx->cred = cred;
964 ctx->ll_cred = NULL;
965 ctx->state = NULL;
966 ctx->mode = f_mode;
967 ctx->flags = 0;
968 ctx->error = 0;
969 ctx->flock_owner = (fl_owner_t)filp;
970 nfs_init_lock_context(&ctx->lock_context);
971 ctx->lock_context.open_context = ctx;
972 INIT_LIST_HEAD(&ctx->list);
973 ctx->mdsthreshold = NULL;
974 return ctx;
975}
976EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
977
978struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
979{
980 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
981 return ctx;
982 return NULL;
983}
984EXPORT_SYMBOL_GPL(get_nfs_open_context);
985
986static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
987{
988 struct inode *inode = d_inode(ctx->dentry);
989 struct super_block *sb = ctx->dentry->d_sb;
990
991 if (!refcount_dec_and_test(&ctx->lock_context.count))
992 return;
993 if (!list_empty(&ctx->list)) {
994 spin_lock(&inode->i_lock);
995 list_del_rcu(&ctx->list);
996 spin_unlock(&inode->i_lock);
997 }
998 if (inode != NULL)
999 NFS_PROTO(inode)->close_context(ctx, is_sync);
1000 put_cred(ctx->cred);
1001 dput(ctx->dentry);
1002 nfs_sb_deactive(sb);
1003 put_rpccred(ctx->ll_cred);
1004 kfree(ctx->mdsthreshold);
1005 kfree_rcu(ctx, rcu_head);
1006}
1007
1008void put_nfs_open_context(struct nfs_open_context *ctx)
1009{
1010 __put_nfs_open_context(ctx, 0);
1011}
1012EXPORT_SYMBOL_GPL(put_nfs_open_context);
1013
1014static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1015{
1016 __put_nfs_open_context(ctx, 1);
1017}
1018
1019/*
1020 * Ensure that mmap has a recent RPC credential for use when writing out
1021 * shared pages
1022 */
1023void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1024{
1025 struct inode *inode = d_inode(ctx->dentry);
1026 struct nfs_inode *nfsi = NFS_I(inode);
1027
1028 spin_lock(&inode->i_lock);
1029 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1030 spin_unlock(&inode->i_lock);
1031}
1032EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1033
1034void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1035{
1036 filp->private_data = get_nfs_open_context(ctx);
1037 if (list_empty(&ctx->list))
1038 nfs_inode_attach_open_context(ctx);
1039}
1040EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1041
1042/*
1043 * Given an inode, search for an open context with the desired characteristics
1044 */
1045struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1046{
1047 struct nfs_inode *nfsi = NFS_I(inode);
1048 struct nfs_open_context *pos, *ctx = NULL;
1049
1050 rcu_read_lock();
1051 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1052 if (cred != NULL && pos->cred != cred)
1053 continue;
1054 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1055 continue;
1056 ctx = get_nfs_open_context(pos);
1057 if (ctx)
1058 break;
1059 }
1060 rcu_read_unlock();
1061 return ctx;
1062}
1063
1064void nfs_file_clear_open_context(struct file *filp)
1065{
1066 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1067
1068 if (ctx) {
1069 struct inode *inode = d_inode(ctx->dentry);
1070
1071 /*
1072 * We fatal error on write before. Try to writeback
1073 * every page again.
1074 */
1075 if (ctx->error < 0)
1076 invalidate_inode_pages2(inode->i_mapping);
1077 filp->private_data = NULL;
1078 put_nfs_open_context_sync(ctx);
1079 }
1080}
1081
1082/*
1083 * These allocate and release file read/write context information.
1084 */
1085int nfs_open(struct inode *inode, struct file *filp)
1086{
1087 struct nfs_open_context *ctx;
1088
1089 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1090 if (IS_ERR(ctx))
1091 return PTR_ERR(ctx);
1092 nfs_file_set_open_context(filp, ctx);
1093 put_nfs_open_context(ctx);
1094 nfs_fscache_open_file(inode, filp);
1095 return 0;
1096}
1097
1098/*
1099 * This function is called whenever some part of NFS notices that
1100 * the cached attributes have to be refreshed.
1101 */
1102int
1103__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1104{
1105 int status = -ESTALE;
1106 struct nfs4_label *label = NULL;
1107 struct nfs_fattr *fattr = NULL;
1108 struct nfs_inode *nfsi = NFS_I(inode);
1109
1110 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1111 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1112
1113 trace_nfs_revalidate_inode_enter(inode);
1114
1115 if (is_bad_inode(inode))
1116 goto out;
1117 if (NFS_STALE(inode))
1118 goto out;
1119
1120 /* pNFS: Attributes aren't updated until we layoutcommit */
1121 if (S_ISREG(inode->i_mode)) {
1122 status = pnfs_sync_inode(inode, false);
1123 if (status)
1124 goto out;
1125 }
1126
1127 status = -ENOMEM;
1128 fattr = nfs_alloc_fattr();
1129 if (fattr == NULL)
1130 goto out;
1131
1132 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1133
1134 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1135 if (IS_ERR(label)) {
1136 status = PTR_ERR(label);
1137 goto out;
1138 }
1139
1140 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1141 label, inode);
1142 if (status != 0) {
1143 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1144 inode->i_sb->s_id,
1145 (unsigned long long)NFS_FILEID(inode), status);
1146 if (status == -ESTALE) {
1147 nfs_zap_caches(inode);
1148 if (!S_ISDIR(inode->i_mode))
1149 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1150 }
1151 goto err_out;
1152 }
1153
1154 status = nfs_refresh_inode(inode, fattr);
1155 if (status) {
1156 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1157 inode->i_sb->s_id,
1158 (unsigned long long)NFS_FILEID(inode), status);
1159 goto err_out;
1160 }
1161
1162 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1163 nfs_zap_acl_cache(inode);
1164
1165 nfs_setsecurity(inode, fattr, label);
1166
1167 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1168 inode->i_sb->s_id,
1169 (unsigned long long)NFS_FILEID(inode));
1170
1171err_out:
1172 nfs4_label_free(label);
1173out:
1174 nfs_free_fattr(fattr);
1175 trace_nfs_revalidate_inode_exit(inode, status);
1176 return status;
1177}
1178
1179int nfs_attribute_cache_expired(struct inode *inode)
1180{
1181 if (nfs_have_delegated_attributes(inode))
1182 return 0;
1183 return nfs_attribute_timeout(inode);
1184}
1185
1186/**
1187 * nfs_revalidate_inode - Revalidate the inode attributes
1188 * @server: pointer to nfs_server struct
1189 * @inode: pointer to inode struct
1190 *
1191 * Updates inode attribute information by retrieving the data from the server.
1192 */
1193int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1194{
1195 if (!nfs_need_revalidate_inode(inode))
1196 return NFS_STALE(inode) ? -ESTALE : 0;
1197 return __nfs_revalidate_inode(server, inode);
1198}
1199EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1200
1201static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1202{
1203 struct nfs_inode *nfsi = NFS_I(inode);
1204 int ret;
1205
1206 if (mapping->nrpages != 0) {
1207 if (S_ISREG(inode->i_mode)) {
1208 ret = nfs_sync_mapping(mapping);
1209 if (ret < 0)
1210 return ret;
1211 }
1212 ret = invalidate_inode_pages2(mapping);
1213 if (ret < 0)
1214 return ret;
1215 }
1216 if (S_ISDIR(inode->i_mode)) {
1217 spin_lock(&inode->i_lock);
1218 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1219 spin_unlock(&inode->i_lock);
1220 }
1221 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1222 nfs_fscache_wait_on_invalidate(inode);
1223
1224 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1225 inode->i_sb->s_id,
1226 (unsigned long long)NFS_FILEID(inode));
1227 return 0;
1228}
1229
1230bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1231{
1232 return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1233 NFS_STALE(inode);
1234}
1235
1236int nfs_revalidate_mapping_rcu(struct inode *inode)
1237{
1238 struct nfs_inode *nfsi = NFS_I(inode);
1239 unsigned long *bitlock = &nfsi->flags;
1240 int ret = 0;
1241
1242 if (IS_SWAPFILE(inode))
1243 goto out;
1244 if (nfs_mapping_need_revalidate_inode(inode)) {
1245 ret = -ECHILD;
1246 goto out;
1247 }
1248 spin_lock(&inode->i_lock);
1249 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1250 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1251 ret = -ECHILD;
1252 spin_unlock(&inode->i_lock);
1253out:
1254 return ret;
1255}
1256
1257/**
1258 * nfs_revalidate_mapping - Revalidate the pagecache
1259 * @inode: pointer to host inode
1260 * @mapping: pointer to mapping
1261 */
1262int nfs_revalidate_mapping(struct inode *inode,
1263 struct address_space *mapping)
1264{
1265 struct nfs_inode *nfsi = NFS_I(inode);
1266 unsigned long *bitlock = &nfsi->flags;
1267 int ret = 0;
1268
1269 /* swapfiles are not supposed to be shared. */
1270 if (IS_SWAPFILE(inode))
1271 goto out;
1272
1273 if (nfs_mapping_need_revalidate_inode(inode)) {
1274 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1275 if (ret < 0)
1276 goto out;
1277 }
1278
1279 /*
1280 * We must clear NFS_INO_INVALID_DATA first to ensure that
1281 * invalidations that come in while we're shooting down the mappings
1282 * are respected. But, that leaves a race window where one revalidator
1283 * can clear the flag, and then another checks it before the mapping
1284 * gets invalidated. Fix that by serializing access to this part of
1285 * the function.
1286 *
1287 * At the same time, we need to allow other tasks to see whether we
1288 * might be in the middle of invalidating the pages, so we only set
1289 * the bit lock here if it looks like we're going to be doing that.
1290 */
1291 for (;;) {
1292 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1293 nfs_wait_bit_killable, TASK_KILLABLE);
1294 if (ret)
1295 goto out;
1296 spin_lock(&inode->i_lock);
1297 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1298 spin_unlock(&inode->i_lock);
1299 continue;
1300 }
1301 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1302 break;
1303 spin_unlock(&inode->i_lock);
1304 goto out;
1305 }
1306
1307 set_bit(NFS_INO_INVALIDATING, bitlock);
1308 smp_wmb();
1309 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1310 spin_unlock(&inode->i_lock);
1311 trace_nfs_invalidate_mapping_enter(inode);
1312 ret = nfs_invalidate_mapping(inode, mapping);
1313 trace_nfs_invalidate_mapping_exit(inode, ret);
1314
1315 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1316 smp_mb__after_atomic();
1317 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1318out:
1319 return ret;
1320}
1321
1322static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1323{
1324 struct inode *inode = &nfsi->vfs_inode;
1325
1326 if (!S_ISREG(inode->i_mode))
1327 return false;
1328 if (list_empty(&nfsi->open_files))
1329 return false;
1330 return inode_is_open_for_write(inode);
1331}
1332
1333static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1334{
1335 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1336}
1337
1338static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1339{
1340 struct timespec ts;
1341
1342 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1343 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1344 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1345 inode_set_iversion_raw(inode, fattr->change_attr);
1346 if (S_ISDIR(inode->i_mode))
1347 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1348 }
1349 /* If we have atomic WCC data, we may update some attributes */
1350 ts = timespec64_to_timespec(inode->i_ctime);
1351 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1352 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1353 && timespec_equal(&ts, &fattr->pre_ctime)) {
1354 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
1355 }
1356
1357 ts = timespec64_to_timespec(inode->i_mtime);
1358 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1359 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1360 && timespec_equal(&ts, &fattr->pre_mtime)) {
1361 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
1362 if (S_ISDIR(inode->i_mode))
1363 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1364 }
1365 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1366 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1367 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1368 && !nfs_have_writebacks(inode)) {
1369 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1370 }
1371}
1372
1373/**
1374 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1375 * @inode: pointer to inode
1376 * @fattr: updated attributes
1377 *
1378 * Verifies the attribute cache. If we have just changed the attributes,
1379 * so that fattr carries weak cache consistency data, then it may
1380 * also update the ctime/mtime/change_attribute.
1381 */
1382static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1383{
1384 struct nfs_inode *nfsi = NFS_I(inode);
1385 loff_t cur_size, new_isize;
1386 unsigned long invalid = 0;
1387 struct timespec ts;
1388
1389 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1390 return 0;
1391
1392 /* Has the inode gone and changed behind our back? */
1393 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1394 return -ESTALE;
1395 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1396 return -ESTALE;
1397
1398 if (!nfs_file_has_buffered_writers(nfsi)) {
1399 /* Verify a few of the more important attributes */
1400 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1401 invalid |= NFS_INO_INVALID_CHANGE
1402 | NFS_INO_REVAL_PAGECACHE;
1403
1404 ts = timespec64_to_timespec(inode->i_mtime);
1405 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&ts, &fattr->mtime))
1406 invalid |= NFS_INO_INVALID_MTIME;
1407
1408 ts = timespec64_to_timespec(inode->i_ctime);
1409 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&ts, &fattr->ctime))
1410 invalid |= NFS_INO_INVALID_CTIME;
1411
1412 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1413 cur_size = i_size_read(inode);
1414 new_isize = nfs_size_to_loff_t(fattr->size);
1415 if (cur_size != new_isize)
1416 invalid |= NFS_INO_INVALID_SIZE
1417 | NFS_INO_REVAL_PAGECACHE;
1418 }
1419 }
1420
1421 /* Have any file permissions changed? */
1422 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1423 invalid |= NFS_INO_INVALID_ACCESS
1424 | NFS_INO_INVALID_ACL
1425 | NFS_INO_INVALID_OTHER;
1426 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1427 invalid |= NFS_INO_INVALID_ACCESS
1428 | NFS_INO_INVALID_ACL
1429 | NFS_INO_INVALID_OTHER;
1430 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1431 invalid |= NFS_INO_INVALID_ACCESS
1432 | NFS_INO_INVALID_ACL
1433 | NFS_INO_INVALID_OTHER;
1434
1435 /* Has the link count changed? */
1436 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1437 invalid |= NFS_INO_INVALID_OTHER;
1438
1439 ts = timespec64_to_timespec(inode->i_atime);
1440 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&ts, &fattr->atime))
1441 invalid |= NFS_INO_INVALID_ATIME;
1442
1443 if (invalid != 0)
1444 nfs_set_cache_invalid(inode, invalid);
1445
1446 nfsi->read_cache_jiffies = fattr->time_start;
1447 return 0;
1448}
1449
1450static atomic_long_t nfs_attr_generation_counter;
1451
1452static unsigned long nfs_read_attr_generation_counter(void)
1453{
1454 return atomic_long_read(&nfs_attr_generation_counter);
1455}
1456
1457unsigned long nfs_inc_attr_generation_counter(void)
1458{
1459 return atomic_long_inc_return(&nfs_attr_generation_counter);
1460}
1461EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1462
1463void nfs_fattr_init(struct nfs_fattr *fattr)
1464{
1465 fattr->valid = 0;
1466 fattr->time_start = jiffies;
1467 fattr->gencount = nfs_inc_attr_generation_counter();
1468 fattr->owner_name = NULL;
1469 fattr->group_name = NULL;
1470}
1471EXPORT_SYMBOL_GPL(nfs_fattr_init);
1472
1473/**
1474 * nfs_fattr_set_barrier
1475 * @fattr: attributes
1476 *
1477 * Used to set a barrier after an attribute was updated. This
1478 * barrier ensures that older attributes from RPC calls that may
1479 * have raced with our update cannot clobber these new values.
1480 * Note that you are still responsible for ensuring that other
1481 * operations which change the attribute on the server do not
1482 * collide.
1483 */
1484void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1485{
1486 fattr->gencount = nfs_inc_attr_generation_counter();
1487}
1488
1489struct nfs_fattr *nfs_alloc_fattr(void)
1490{
1491 struct nfs_fattr *fattr;
1492
1493 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1494 if (fattr != NULL)
1495 nfs_fattr_init(fattr);
1496 return fattr;
1497}
1498EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1499
1500struct nfs_fh *nfs_alloc_fhandle(void)
1501{
1502 struct nfs_fh *fh;
1503
1504 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1505 if (fh != NULL)
1506 fh->size = 0;
1507 return fh;
1508}
1509EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1510
1511#ifdef NFS_DEBUG
1512/*
1513 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1514 * in the same way that wireshark does
1515 *
1516 * @fh: file handle
1517 *
1518 * For debugging only.
1519 */
1520u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1521{
1522 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1523 * not on the result */
1524 return nfs_fhandle_hash(fh);
1525}
1526EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1527
1528/*
1529 * _nfs_display_fhandle - display an NFS file handle on the console
1530 *
1531 * @fh: file handle to display
1532 * @caption: display caption
1533 *
1534 * For debugging only.
1535 */
1536void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1537{
1538 unsigned short i;
1539
1540 if (fh == NULL || fh->size == 0) {
1541 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1542 return;
1543 }
1544
1545 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1546 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1547 for (i = 0; i < fh->size; i += 16) {
1548 __be32 *pos = (__be32 *)&fh->data[i];
1549
1550 switch ((fh->size - i - 1) >> 2) {
1551 case 0:
1552 printk(KERN_DEFAULT " %08x\n",
1553 be32_to_cpup(pos));
1554 break;
1555 case 1:
1556 printk(KERN_DEFAULT " %08x %08x\n",
1557 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1558 break;
1559 case 2:
1560 printk(KERN_DEFAULT " %08x %08x %08x\n",
1561 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1562 be32_to_cpup(pos + 2));
1563 break;
1564 default:
1565 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1566 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1567 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1568 }
1569 }
1570}
1571EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1572#endif
1573
1574/**
1575 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1576 * @inode: pointer to inode
1577 * @fattr: attributes
1578 *
1579 * Attempt to divine whether or not an RPC call reply carrying stale
1580 * attributes got scheduled after another call carrying updated ones.
1581 *
1582 * To do so, the function first assumes that a more recent ctime means
1583 * that the attributes in fattr are newer, however it also attempt to
1584 * catch the case where ctime either didn't change, or went backwards
1585 * (if someone reset the clock on the server) by looking at whether
1586 * or not this RPC call was started after the inode was last updated.
1587 * Note also the check for wraparound of 'attr_gencount'
1588 *
1589 * The function returns 'true' if it thinks the attributes in 'fattr' are
1590 * more recent than the ones cached in the inode.
1591 *
1592 */
1593static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1594{
1595 const struct nfs_inode *nfsi = NFS_I(inode);
1596
1597 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1598 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1599}
1600
1601static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1602{
1603 int ret;
1604
1605 trace_nfs_refresh_inode_enter(inode);
1606
1607 if (nfs_inode_attrs_need_update(inode, fattr))
1608 ret = nfs_update_inode(inode, fattr);
1609 else
1610 ret = nfs_check_inode_attributes(inode, fattr);
1611
1612 trace_nfs_refresh_inode_exit(inode, ret);
1613 return ret;
1614}
1615
1616/**
1617 * nfs_refresh_inode - try to update the inode attribute cache
1618 * @inode: pointer to inode
1619 * @fattr: updated attributes
1620 *
1621 * Check that an RPC call that returned attributes has not overlapped with
1622 * other recent updates of the inode metadata, then decide whether it is
1623 * safe to do a full update of the inode attributes, or whether just to
1624 * call nfs_check_inode_attributes.
1625 */
1626int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1627{
1628 int status;
1629
1630 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1631 return 0;
1632 spin_lock(&inode->i_lock);
1633 status = nfs_refresh_inode_locked(inode, fattr);
1634 spin_unlock(&inode->i_lock);
1635
1636 return status;
1637}
1638EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1639
1640static int nfs_post_op_update_inode_locked(struct inode *inode,
1641 struct nfs_fattr *fattr, unsigned int invalid)
1642{
1643 if (S_ISDIR(inode->i_mode))
1644 invalid |= NFS_INO_INVALID_DATA;
1645 nfs_set_cache_invalid(inode, invalid);
1646 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1647 return 0;
1648 return nfs_refresh_inode_locked(inode, fattr);
1649}
1650
1651/**
1652 * nfs_post_op_update_inode - try to update the inode attribute cache
1653 * @inode: pointer to inode
1654 * @fattr: updated attributes
1655 *
1656 * After an operation that has changed the inode metadata, mark the
1657 * attribute cache as being invalid, then try to update it.
1658 *
1659 * NB: if the server didn't return any post op attributes, this
1660 * function will force the retrieval of attributes before the next
1661 * NFS request. Thus it should be used only for operations that
1662 * are expected to change one or more attributes, to avoid
1663 * unnecessary NFS requests and trips through nfs_update_inode().
1664 */
1665int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1666{
1667 int status;
1668
1669 spin_lock(&inode->i_lock);
1670 nfs_fattr_set_barrier(fattr);
1671 status = nfs_post_op_update_inode_locked(inode, fattr,
1672 NFS_INO_INVALID_CHANGE
1673 | NFS_INO_INVALID_CTIME
1674 | NFS_INO_REVAL_FORCED);
1675 spin_unlock(&inode->i_lock);
1676
1677 return status;
1678}
1679EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1680
1681/**
1682 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1683 * @inode: pointer to inode
1684 * @fattr: updated attributes
1685 *
1686 * After an operation that has changed the inode metadata, mark the
1687 * attribute cache as being invalid, then try to update it. Fake up
1688 * weak cache consistency data, if none exist.
1689 *
1690 * This function is mainly designed to be used by the ->write_done() functions.
1691 */
1692int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1693{
1694 int status;
1695
1696 /* Don't do a WCC update if these attributes are already stale */
1697 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1698 !nfs_inode_attrs_need_update(inode, fattr)) {
1699 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1700 | NFS_ATTR_FATTR_PRESIZE
1701 | NFS_ATTR_FATTR_PREMTIME
1702 | NFS_ATTR_FATTR_PRECTIME);
1703 goto out_noforce;
1704 }
1705 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1706 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1707 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1708 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1709 }
1710 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1711 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1712 fattr->pre_ctime = timespec64_to_timespec(inode->i_ctime);
1713 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1714 }
1715 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1716 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1717 fattr->pre_mtime = timespec64_to_timespec(inode->i_mtime);
1718 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1719 }
1720 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1721 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1722 fattr->pre_size = i_size_read(inode);
1723 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1724 }
1725out_noforce:
1726 status = nfs_post_op_update_inode_locked(inode, fattr,
1727 NFS_INO_INVALID_CHANGE
1728 | NFS_INO_INVALID_CTIME
1729 | NFS_INO_INVALID_MTIME);
1730 return status;
1731}
1732
1733/**
1734 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1735 * @inode: pointer to inode
1736 * @fattr: updated attributes
1737 *
1738 * After an operation that has changed the inode metadata, mark the
1739 * attribute cache as being invalid, then try to update it. Fake up
1740 * weak cache consistency data, if none exist.
1741 *
1742 * This function is mainly designed to be used by the ->write_done() functions.
1743 */
1744int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1745{
1746 int status;
1747
1748 spin_lock(&inode->i_lock);
1749 nfs_fattr_set_barrier(fattr);
1750 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1751 spin_unlock(&inode->i_lock);
1752 return status;
1753}
1754EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1755
1756
1757static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1758 struct nfs_fattr *fattr)
1759{
1760 bool ret1 = true, ret2 = true;
1761
1762 if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1763 ret1 = (nfsi->fileid == fattr->fileid);
1764 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1765 ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1766 return ret1 || ret2;
1767}
1768
1769/*
1770 * Many nfs protocol calls return the new file attributes after
1771 * an operation. Here we update the inode to reflect the state
1772 * of the server's inode.
1773 *
1774 * This is a bit tricky because we have to make sure all dirty pages
1775 * have been sent off to the server before calling invalidate_inode_pages.
1776 * To make sure no other process adds more write requests while we try
1777 * our best to flush them, we make them sleep during the attribute refresh.
1778 *
1779 * A very similar scenario holds for the dir cache.
1780 */
1781static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1782{
1783 struct nfs_server *server;
1784 struct nfs_inode *nfsi = NFS_I(inode);
1785 loff_t cur_isize, new_isize;
1786 unsigned long invalid = 0;
1787 unsigned long now = jiffies;
1788 unsigned long save_cache_validity;
1789 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1790 bool cache_revalidated = true;
1791 bool attr_changed = false;
1792 bool have_delegation;
1793
1794 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1795 __func__, inode->i_sb->s_id, inode->i_ino,
1796 nfs_display_fhandle_hash(NFS_FH(inode)),
1797 atomic_read(&inode->i_count), fattr->valid);
1798
1799 if (!nfs_fileid_valid(nfsi, fattr)) {
1800 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1801 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1802 NFS_SERVER(inode)->nfs_client->cl_hostname,
1803 inode->i_sb->s_id, (long long)nfsi->fileid,
1804 (long long)fattr->fileid);
1805 goto out_err;
1806 }
1807
1808 /*
1809 * Make sure the inode's type hasn't changed.
1810 */
1811 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1812 /*
1813 * Big trouble! The inode has become a different object.
1814 */
1815 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1816 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1817 goto out_err;
1818 }
1819
1820 server = NFS_SERVER(inode);
1821 /* Update the fsid? */
1822 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1823 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1824 !IS_AUTOMOUNT(inode))
1825 server->fsid = fattr->fsid;
1826
1827 /* Save the delegation state before clearing cache_validity */
1828 have_delegation = nfs_have_delegated_attributes(inode);
1829
1830 /*
1831 * Update the read time so we don't revalidate too often.
1832 */
1833 nfsi->read_cache_jiffies = fattr->time_start;
1834
1835 save_cache_validity = nfsi->cache_validity;
1836 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1837 | NFS_INO_INVALID_ATIME
1838 | NFS_INO_REVAL_FORCED
1839 | NFS_INO_REVAL_PAGECACHE);
1840
1841 /* Do atomic weak cache consistency updates */
1842 nfs_wcc_update_inode(inode, fattr);
1843
1844 if (pnfs_layoutcommit_outstanding(inode)) {
1845 nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1846 cache_revalidated = false;
1847 }
1848
1849 /* More cache consistency checks */
1850 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1851 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1852 /* Could it be a race with writeback? */
1853 if (!(have_writers || have_delegation)) {
1854 invalid |= NFS_INO_INVALID_DATA
1855 | NFS_INO_INVALID_ACCESS
1856 | NFS_INO_INVALID_ACL;
1857 /* Force revalidate of all attributes */
1858 save_cache_validity |= NFS_INO_INVALID_CTIME
1859 | NFS_INO_INVALID_MTIME
1860 | NFS_INO_INVALID_SIZE
1861 | NFS_INO_INVALID_OTHER;
1862 if (S_ISDIR(inode->i_mode))
1863 nfs_force_lookup_revalidate(inode);
1864 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1865 inode->i_sb->s_id,
1866 inode->i_ino);
1867 }
1868 inode_set_iversion_raw(inode, fattr->change_attr);
1869 attr_changed = true;
1870 }
1871 } else {
1872 nfsi->cache_validity |= save_cache_validity &
1873 (NFS_INO_INVALID_CHANGE
1874 | NFS_INO_REVAL_PAGECACHE
1875 | NFS_INO_REVAL_FORCED);
1876 cache_revalidated = false;
1877 }
1878
1879 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1880 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
1881 } else if (server->caps & NFS_CAP_MTIME) {
1882 nfsi->cache_validity |= save_cache_validity &
1883 (NFS_INO_INVALID_MTIME
1884 | NFS_INO_REVAL_FORCED);
1885 cache_revalidated = false;
1886 }
1887
1888 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1889 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
1890 } else if (server->caps & NFS_CAP_CTIME) {
1891 nfsi->cache_validity |= save_cache_validity &
1892 (NFS_INO_INVALID_CTIME
1893 | NFS_INO_REVAL_FORCED);
1894 cache_revalidated = false;
1895 }
1896
1897 /* Check if our cached file size is stale */
1898 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1899 new_isize = nfs_size_to_loff_t(fattr->size);
1900 cur_isize = i_size_read(inode);
1901 if (new_isize != cur_isize && !have_delegation) {
1902 /* Do we perhaps have any outstanding writes, or has
1903 * the file grown beyond our last write? */
1904 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1905 i_size_write(inode, new_isize);
1906 if (!have_writers)
1907 invalid |= NFS_INO_INVALID_DATA;
1908 attr_changed = true;
1909 }
1910 dprintk("NFS: isize change on server for file %s/%ld "
1911 "(%Ld to %Ld)\n",
1912 inode->i_sb->s_id,
1913 inode->i_ino,
1914 (long long)cur_isize,
1915 (long long)new_isize);
1916 }
1917 } else {
1918 nfsi->cache_validity |= save_cache_validity &
1919 (NFS_INO_INVALID_SIZE
1920 | NFS_INO_REVAL_PAGECACHE
1921 | NFS_INO_REVAL_FORCED);
1922 cache_revalidated = false;
1923 }
1924
1925
1926 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1927 inode->i_atime = timespec_to_timespec64(fattr->atime);
1928 else if (server->caps & NFS_CAP_ATIME) {
1929 nfsi->cache_validity |= save_cache_validity &
1930 (NFS_INO_INVALID_ATIME
1931 | NFS_INO_REVAL_FORCED);
1932 cache_revalidated = false;
1933 }
1934
1935 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1936 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1937 umode_t newmode = inode->i_mode & S_IFMT;
1938 newmode |= fattr->mode & S_IALLUGO;
1939 inode->i_mode = newmode;
1940 invalid |= NFS_INO_INVALID_ACCESS
1941 | NFS_INO_INVALID_ACL;
1942 attr_changed = true;
1943 }
1944 } else if (server->caps & NFS_CAP_MODE) {
1945 nfsi->cache_validity |= save_cache_validity &
1946 (NFS_INO_INVALID_OTHER
1947 | NFS_INO_REVAL_FORCED);
1948 cache_revalidated = false;
1949 }
1950
1951 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1952 if (!uid_eq(inode->i_uid, fattr->uid)) {
1953 invalid |= NFS_INO_INVALID_ACCESS
1954 | NFS_INO_INVALID_ACL;
1955 inode->i_uid = fattr->uid;
1956 attr_changed = true;
1957 }
1958 } else if (server->caps & NFS_CAP_OWNER) {
1959 nfsi->cache_validity |= save_cache_validity &
1960 (NFS_INO_INVALID_OTHER
1961 | NFS_INO_REVAL_FORCED);
1962 cache_revalidated = false;
1963 }
1964
1965 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1966 if (!gid_eq(inode->i_gid, fattr->gid)) {
1967 invalid |= NFS_INO_INVALID_ACCESS
1968 | NFS_INO_INVALID_ACL;
1969 inode->i_gid = fattr->gid;
1970 attr_changed = true;
1971 }
1972 } else if (server->caps & NFS_CAP_OWNER_GROUP) {
1973 nfsi->cache_validity |= save_cache_validity &
1974 (NFS_INO_INVALID_OTHER
1975 | NFS_INO_REVAL_FORCED);
1976 cache_revalidated = false;
1977 }
1978
1979 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1980 if (inode->i_nlink != fattr->nlink) {
1981 if (S_ISDIR(inode->i_mode))
1982 invalid |= NFS_INO_INVALID_DATA;
1983 set_nlink(inode, fattr->nlink);
1984 attr_changed = true;
1985 }
1986 } else if (server->caps & NFS_CAP_NLINK) {
1987 nfsi->cache_validity |= save_cache_validity &
1988 (NFS_INO_INVALID_OTHER
1989 | NFS_INO_REVAL_FORCED);
1990 cache_revalidated = false;
1991 }
1992
1993 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1994 /*
1995 * report the blocks in 512byte units
1996 */
1997 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1998 } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1999 inode->i_blocks = fattr->du.nfs2.blocks;
2000 else
2001 cache_revalidated = false;
2002
2003 /* Update attrtimeo value if we're out of the unstable period */
2004 if (attr_changed) {
2005 invalid &= ~NFS_INO_INVALID_ATTR;
2006 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2007 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2008 nfsi->attrtimeo_timestamp = now;
2009 /* Set barrier to be more recent than all outstanding updates */
2010 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2011 } else {
2012 if (cache_revalidated) {
2013 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2014 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2015 nfsi->attrtimeo <<= 1;
2016 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2017 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2018 }
2019 nfsi->attrtimeo_timestamp = now;
2020 }
2021 /* Set the barrier to be more recent than this fattr */
2022 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
2023 nfsi->attr_gencount = fattr->gencount;
2024 }
2025
2026 /* Don't invalidate the data if we were to blame */
2027 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2028 || S_ISLNK(inode->i_mode)))
2029 invalid &= ~NFS_INO_INVALID_DATA;
2030 nfs_set_cache_invalid(inode, invalid);
2031
2032 return 0;
2033 out_err:
2034 /*
2035 * No need to worry about unhashing the dentry, as the
2036 * lookup validation will know that the inode is bad.
2037 * (But we fall through to invalidate the caches.)
2038 */
2039 nfs_invalidate_inode(inode);
2040 return -ESTALE;
2041}
2042
2043struct inode *nfs_alloc_inode(struct super_block *sb)
2044{
2045 struct nfs_inode *nfsi;
2046 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2047 if (!nfsi)
2048 return NULL;
2049 nfsi->flags = 0UL;
2050 nfsi->cache_validity = 0UL;
2051#if IS_ENABLED(CONFIG_NFS_V4)
2052 nfsi->nfs4_acl = NULL;
2053#endif /* CONFIG_NFS_V4 */
2054 return &nfsi->vfs_inode;
2055}
2056EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2057
2058static void nfs_i_callback(struct rcu_head *head)
2059{
2060 struct inode *inode = container_of(head, struct inode, i_rcu);
2061 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2062}
2063
2064void nfs_destroy_inode(struct inode *inode)
2065{
2066 call_rcu(&inode->i_rcu, nfs_i_callback);
2067}
2068EXPORT_SYMBOL_GPL(nfs_destroy_inode);
2069
2070static inline void nfs4_init_once(struct nfs_inode *nfsi)
2071{
2072#if IS_ENABLED(CONFIG_NFS_V4)
2073 INIT_LIST_HEAD(&nfsi->open_states);
2074 nfsi->delegation = NULL;
2075 init_rwsem(&nfsi->rwsem);
2076 nfsi->layout = NULL;
2077#endif
2078}
2079
2080static void init_once(void *foo)
2081{
2082 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2083
2084 inode_init_once(&nfsi->vfs_inode);
2085 INIT_LIST_HEAD(&nfsi->open_files);
2086 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2087 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2088 INIT_LIST_HEAD(&nfsi->commit_info.list);
2089 atomic_long_set(&nfsi->nrequests, 0);
2090 atomic_long_set(&nfsi->commit_info.ncommit, 0);
2091 atomic_set(&nfsi->commit_info.rpcs_out, 0);
2092 init_rwsem(&nfsi->rmdir_sem);
2093 mutex_init(&nfsi->commit_mutex);
2094 nfs4_init_once(nfsi);
2095}
2096
2097static int __init nfs_init_inodecache(void)
2098{
2099 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2100 sizeof(struct nfs_inode),
2101 0, (SLAB_RECLAIM_ACCOUNT|
2102 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2103 init_once);
2104 if (nfs_inode_cachep == NULL)
2105 return -ENOMEM;
2106
2107 return 0;
2108}
2109
2110static void nfs_destroy_inodecache(void)
2111{
2112 /*
2113 * Make sure all delayed rcu free inodes are flushed before we
2114 * destroy cache.
2115 */
2116 rcu_barrier();
2117 kmem_cache_destroy(nfs_inode_cachep);
2118}
2119
2120struct workqueue_struct *nfsiod_workqueue;
2121EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2122
2123/*
2124 * start up the nfsiod workqueue
2125 */
2126static int nfsiod_start(void)
2127{
2128 struct workqueue_struct *wq;
2129 dprintk("RPC: creating workqueue nfsiod\n");
2130 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2131 if (wq == NULL)
2132 return -ENOMEM;
2133 nfsiod_workqueue = wq;
2134 return 0;
2135}
2136
2137/*
2138 * Destroy the nfsiod workqueue
2139 */
2140static void nfsiod_stop(void)
2141{
2142 struct workqueue_struct *wq;
2143
2144 wq = nfsiod_workqueue;
2145 if (wq == NULL)
2146 return;
2147 nfsiod_workqueue = NULL;
2148 destroy_workqueue(wq);
2149}
2150
2151unsigned int nfs_net_id;
2152EXPORT_SYMBOL_GPL(nfs_net_id);
2153
2154static int nfs_net_init(struct net *net)
2155{
2156 nfs_clients_init(net);
2157 return nfs_fs_proc_net_init(net);
2158}
2159
2160static void nfs_net_exit(struct net *net)
2161{
2162 struct nfs_net *nn = net_generic(net, nfs_net_id);
2163
2164 nfs_fs_proc_net_exit(net);
2165 nfs_cleanup_cb_ident_idr(net);
2166 WARN_ON_ONCE(!list_empty(&nn->nfs_client_list));
2167 WARN_ON_ONCE(!list_empty(&nn->nfs_volume_list));
2168}
2169
2170static struct pernet_operations nfs_net_ops = {
2171 .init = nfs_net_init,
2172 .exit = nfs_net_exit,
2173 .id = &nfs_net_id,
2174 .size = sizeof(struct nfs_net),
2175};
2176
2177/*
2178 * Initialize NFS
2179 */
2180static int __init init_nfs_fs(void)
2181{
2182 int err;
2183
2184 err = register_pernet_subsys(&nfs_net_ops);
2185 if (err < 0)
2186 goto out9;
2187
2188 err = nfs_fscache_register();
2189 if (err < 0)
2190 goto out8;
2191
2192 err = nfsiod_start();
2193 if (err)
2194 goto out7;
2195
2196 err = nfs_fs_proc_init();
2197 if (err)
2198 goto out6;
2199
2200 err = nfs_init_nfspagecache();
2201 if (err)
2202 goto out5;
2203
2204 err = nfs_init_inodecache();
2205 if (err)
2206 goto out4;
2207
2208 err = nfs_init_readpagecache();
2209 if (err)
2210 goto out3;
2211
2212 err = nfs_init_writepagecache();
2213 if (err)
2214 goto out2;
2215
2216 err = nfs_init_directcache();
2217 if (err)
2218 goto out1;
2219
2220 rpc_proc_register(&init_net, &nfs_rpcstat);
2221
2222 err = register_nfs_fs();
2223 if (err)
2224 goto out0;
2225
2226 return 0;
2227out0:
2228 rpc_proc_unregister(&init_net, "nfs");
2229 nfs_destroy_directcache();
2230out1:
2231 nfs_destroy_writepagecache();
2232out2:
2233 nfs_destroy_readpagecache();
2234out3:
2235 nfs_destroy_inodecache();
2236out4:
2237 nfs_destroy_nfspagecache();
2238out5:
2239 nfs_fs_proc_exit();
2240out6:
2241 nfsiod_stop();
2242out7:
2243 nfs_fscache_unregister();
2244out8:
2245 unregister_pernet_subsys(&nfs_net_ops);
2246out9:
2247 return err;
2248}
2249
2250static void __exit exit_nfs_fs(void)
2251{
2252 nfs_destroy_directcache();
2253 nfs_destroy_writepagecache();
2254 nfs_destroy_readpagecache();
2255 nfs_destroy_inodecache();
2256 nfs_destroy_nfspagecache();
2257 nfs_fscache_unregister();
2258 unregister_pernet_subsys(&nfs_net_ops);
2259 rpc_proc_unregister(&init_net, "nfs");
2260 unregister_nfs_fs();
2261 nfs_fs_proc_exit();
2262 nfsiod_stop();
2263}
2264
2265/* Not quite true; I just maintain it */
2266MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2267MODULE_LICENSE("GPL");
2268module_param(enable_ino64, bool, 0644);
2269
2270module_init(init_nfs_fs)
2271module_exit(exit_nfs_fs)