NFS: Use wait_on_atomic_t() for unlock after readahead
[deliverable/linux.git] / fs / nfs / inode.c
... / ...
CommitLineData
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.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
42#include <asm/uaccess.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
88int nfs_wait_atomic_killable(atomic_t *p)
89{
90 return nfs_wait_killable(TASK_KILLABLE);
91}
92
93/**
94 * nfs_compat_user_ino64 - returns the user-visible inode number
95 * @fileid: 64-bit fileid
96 *
97 * This function returns a 32-bit inode number if the boot parameter
98 * nfs.enable_ino64 is zero.
99 */
100u64 nfs_compat_user_ino64(u64 fileid)
101{
102#ifdef CONFIG_COMPAT
103 compat_ulong_t ino;
104#else
105 unsigned long ino;
106#endif
107
108 if (enable_ino64)
109 return fileid;
110 ino = fileid;
111 if (sizeof(ino) < sizeof(fileid))
112 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
113 return ino;
114}
115
116int nfs_drop_inode(struct inode *inode)
117{
118 return NFS_STALE(inode) || generic_drop_inode(inode);
119}
120EXPORT_SYMBOL_GPL(nfs_drop_inode);
121
122void nfs_clear_inode(struct inode *inode)
123{
124 /*
125 * The following should never happen...
126 */
127 WARN_ON_ONCE(nfs_have_writebacks(inode));
128 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
129 nfs_zap_acl_cache(inode);
130 nfs_access_zap_cache(inode);
131 nfs_fscache_clear_inode(inode);
132}
133EXPORT_SYMBOL_GPL(nfs_clear_inode);
134
135void nfs_evict_inode(struct inode *inode)
136{
137 truncate_inode_pages_final(&inode->i_data);
138 clear_inode(inode);
139 nfs_clear_inode(inode);
140}
141
142int nfs_sync_inode(struct inode *inode)
143{
144 nfs_inode_dio_wait(inode);
145 return nfs_wb_all(inode);
146}
147EXPORT_SYMBOL_GPL(nfs_sync_inode);
148
149/**
150 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
151 */
152int nfs_sync_mapping(struct address_space *mapping)
153{
154 int ret = 0;
155
156 if (mapping->nrpages != 0) {
157 unmap_mapping_range(mapping, 0, 0, 0);
158 ret = nfs_wb_all(mapping->host);
159 }
160 return ret;
161}
162
163static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
164{
165 struct nfs_inode *nfsi = NFS_I(inode);
166
167 if (inode->i_mapping->nrpages == 0)
168 flags &= ~NFS_INO_INVALID_DATA;
169 nfsi->cache_validity |= flags;
170 if (flags & NFS_INO_INVALID_DATA)
171 nfs_fscache_invalidate(inode);
172}
173
174/*
175 * Invalidate the local caches
176 */
177static void nfs_zap_caches_locked(struct inode *inode)
178{
179 struct nfs_inode *nfsi = NFS_I(inode);
180 int mode = inode->i_mode;
181
182 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
183
184 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
185 nfsi->attrtimeo_timestamp = jiffies;
186
187 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
188 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
189 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
190 | NFS_INO_INVALID_DATA
191 | NFS_INO_INVALID_ACCESS
192 | NFS_INO_INVALID_ACL
193 | NFS_INO_REVAL_PAGECACHE);
194 } else
195 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
196 | NFS_INO_INVALID_ACCESS
197 | NFS_INO_INVALID_ACL
198 | NFS_INO_REVAL_PAGECACHE);
199 nfs_zap_label_cache_locked(nfsi);
200}
201
202void nfs_zap_caches(struct inode *inode)
203{
204 spin_lock(&inode->i_lock);
205 nfs_zap_caches_locked(inode);
206 spin_unlock(&inode->i_lock);
207}
208
209void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
210{
211 if (mapping->nrpages != 0) {
212 spin_lock(&inode->i_lock);
213 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
214 spin_unlock(&inode->i_lock);
215 }
216}
217
218void nfs_zap_acl_cache(struct inode *inode)
219{
220 void (*clear_acl_cache)(struct inode *);
221
222 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
223 if (clear_acl_cache != NULL)
224 clear_acl_cache(inode);
225 spin_lock(&inode->i_lock);
226 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
227 spin_unlock(&inode->i_lock);
228}
229EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
230
231void nfs_invalidate_atime(struct inode *inode)
232{
233 spin_lock(&inode->i_lock);
234 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
235 spin_unlock(&inode->i_lock);
236}
237EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
238
239/*
240 * Invalidate, but do not unhash, the inode.
241 * NB: must be called with inode->i_lock held!
242 */
243static void nfs_invalidate_inode(struct inode *inode)
244{
245 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
246 nfs_zap_caches_locked(inode);
247}
248
249struct nfs_find_desc {
250 struct nfs_fh *fh;
251 struct nfs_fattr *fattr;
252};
253
254/*
255 * In NFSv3 we can have 64bit inode numbers. In order to support
256 * this, and re-exported directories (also seen in NFSv2)
257 * we are forced to allow 2 different inodes to have the same
258 * i_ino.
259 */
260static int
261nfs_find_actor(struct inode *inode, void *opaque)
262{
263 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
264 struct nfs_fh *fh = desc->fh;
265 struct nfs_fattr *fattr = desc->fattr;
266
267 if (NFS_FILEID(inode) != fattr->fileid)
268 return 0;
269 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
270 return 0;
271 if (nfs_compare_fh(NFS_FH(inode), fh))
272 return 0;
273 if (is_bad_inode(inode) || NFS_STALE(inode))
274 return 0;
275 return 1;
276}
277
278static int
279nfs_init_locked(struct inode *inode, void *opaque)
280{
281 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
282 struct nfs_fattr *fattr = desc->fattr;
283
284 set_nfs_fileid(inode, fattr->fileid);
285 nfs_copy_fh(NFS_FH(inode), desc->fh);
286 return 0;
287}
288
289#ifdef CONFIG_NFS_V4_SECURITY_LABEL
290static void nfs_clear_label_invalid(struct inode *inode)
291{
292 spin_lock(&inode->i_lock);
293 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
294 spin_unlock(&inode->i_lock);
295}
296
297void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
298 struct nfs4_label *label)
299{
300 int error;
301
302 if (label == NULL)
303 return;
304
305 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
306 error = security_inode_notifysecctx(inode, label->label,
307 label->len);
308 if (error)
309 printk(KERN_ERR "%s() %s %d "
310 "security_inode_notifysecctx() %d\n",
311 __func__,
312 (char *)label->label,
313 label->len, error);
314 nfs_clear_label_invalid(inode);
315 }
316}
317
318struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
319{
320 struct nfs4_label *label = NULL;
321 int minor_version = server->nfs_client->cl_minorversion;
322
323 if (minor_version < 2)
324 return label;
325
326 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
327 return label;
328
329 label = kzalloc(sizeof(struct nfs4_label), flags);
330 if (label == NULL)
331 return ERR_PTR(-ENOMEM);
332
333 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
334 if (label->label == NULL) {
335 kfree(label);
336 return ERR_PTR(-ENOMEM);
337 }
338 label->len = NFS4_MAXLABELLEN;
339
340 return label;
341}
342EXPORT_SYMBOL_GPL(nfs4_label_alloc);
343#else
344void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
345 struct nfs4_label *label)
346{
347}
348#endif
349EXPORT_SYMBOL_GPL(nfs_setsecurity);
350
351/*
352 * This is our front-end to iget that looks up inodes by file handle
353 * instead of inode number.
354 */
355struct inode *
356nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
357{
358 struct nfs_find_desc desc = {
359 .fh = fh,
360 .fattr = fattr
361 };
362 struct inode *inode = ERR_PTR(-ENOENT);
363 unsigned long hash;
364
365 nfs_attr_check_mountpoint(sb, fattr);
366
367 if (nfs_attr_use_mounted_on_fileid(fattr))
368 fattr->fileid = fattr->mounted_on_fileid;
369 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
370 goto out_no_inode;
371 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
372 goto out_no_inode;
373
374 hash = nfs_fattr_to_ino_t(fattr);
375
376 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
377 if (inode == NULL) {
378 inode = ERR_PTR(-ENOMEM);
379 goto out_no_inode;
380 }
381
382 if (inode->i_state & I_NEW) {
383 struct nfs_inode *nfsi = NFS_I(inode);
384 unsigned long now = jiffies;
385
386 /* We set i_ino for the few things that still rely on it,
387 * such as stat(2) */
388 inode->i_ino = hash;
389
390 /* We can't support update_atime(), since the server will reset it */
391 inode->i_flags |= S_NOATIME|S_NOCMTIME;
392 inode->i_mode = fattr->mode;
393 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
394 && nfs_server_capable(inode, NFS_CAP_MODE))
395 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
396 /* Why so? Because we want revalidate for devices/FIFOs, and
397 * that's precisely what we have in nfs_file_inode_operations.
398 */
399 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
400 if (S_ISREG(inode->i_mode)) {
401 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
402 inode->i_data.a_ops = &nfs_file_aops;
403 } else if (S_ISDIR(inode->i_mode)) {
404 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
405 inode->i_fop = &nfs_dir_operations;
406 inode->i_data.a_ops = &nfs_dir_aops;
407 /* Deal with crossing mountpoints */
408 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
409 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
410 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
411 inode->i_op = &nfs_referral_inode_operations;
412 else
413 inode->i_op = &nfs_mountpoint_inode_operations;
414 inode->i_fop = NULL;
415 inode->i_flags |= S_AUTOMOUNT;
416 }
417 } else if (S_ISLNK(inode->i_mode))
418 inode->i_op = &nfs_symlink_inode_operations;
419 else
420 init_special_inode(inode, inode->i_mode, fattr->rdev);
421
422 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
423 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
424 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
425 inode->i_version = 0;
426 inode->i_size = 0;
427 clear_nlink(inode);
428 inode->i_uid = make_kuid(&init_user_ns, -2);
429 inode->i_gid = make_kgid(&init_user_ns, -2);
430 inode->i_blocks = 0;
431 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
432 nfsi->write_io = 0;
433 nfsi->read_io = 0;
434
435 nfsi->read_cache_jiffies = fattr->time_start;
436 nfsi->attr_gencount = fattr->gencount;
437 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
438 inode->i_atime = fattr->atime;
439 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
440 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
441 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
442 inode->i_mtime = fattr->mtime;
443 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
444 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
445 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
446 inode->i_ctime = fattr->ctime;
447 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
448 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
449 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
450 inode->i_version = fattr->change_attr;
451 else
452 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
453 | NFS_INO_REVAL_PAGECACHE);
454 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
455 inode->i_size = nfs_size_to_loff_t(fattr->size);
456 else
457 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
458 | NFS_INO_REVAL_PAGECACHE);
459 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
460 set_nlink(inode, fattr->nlink);
461 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
462 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
463 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
464 inode->i_uid = fattr->uid;
465 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
466 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
467 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
468 inode->i_gid = fattr->gid;
469 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
470 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
471 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
472 inode->i_blocks = fattr->du.nfs2.blocks;
473 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
474 /*
475 * report the blocks in 512byte units
476 */
477 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
478 }
479
480 nfs_setsecurity(inode, fattr, label);
481
482 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
483 nfsi->attrtimeo_timestamp = now;
484 nfsi->access_cache = RB_ROOT;
485
486 nfs_fscache_init_inode(inode);
487
488 unlock_new_inode(inode);
489 } else
490 nfs_refresh_inode(inode, fattr);
491 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
492 inode->i_sb->s_id,
493 (unsigned long long)NFS_FILEID(inode),
494 nfs_display_fhandle_hash(fh),
495 atomic_read(&inode->i_count));
496
497out:
498 return inode;
499
500out_no_inode:
501 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
502 goto out;
503}
504EXPORT_SYMBOL_GPL(nfs_fhget);
505
506#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)
507
508int
509nfs_setattr(struct dentry *dentry, struct iattr *attr)
510{
511 struct inode *inode = d_inode(dentry);
512 struct nfs_fattr *fattr;
513 int error = 0;
514
515 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
516
517 /* skip mode change if it's just for clearing setuid/setgid */
518 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
519 attr->ia_valid &= ~ATTR_MODE;
520
521 if (attr->ia_valid & ATTR_SIZE) {
522 BUG_ON(!S_ISREG(inode->i_mode));
523
524 error = inode_newsize_ok(inode, attr->ia_size);
525 if (error)
526 return error;
527
528 if (attr->ia_size == i_size_read(inode))
529 attr->ia_valid &= ~ATTR_SIZE;
530 }
531
532 /* Optimization: if the end result is no change, don't RPC */
533 attr->ia_valid &= NFS_VALID_ATTRS;
534 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
535 return 0;
536
537 trace_nfs_setattr_enter(inode);
538
539 /* Write all dirty data */
540 if (S_ISREG(inode->i_mode))
541 nfs_sync_inode(inode);
542
543 fattr = nfs_alloc_fattr();
544 if (fattr == NULL) {
545 error = -ENOMEM;
546 goto out;
547 }
548
549 /*
550 * Return any delegations if we're going to change ACLs
551 */
552 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
553 NFS_PROTO(inode)->return_delegation(inode);
554 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
555 if (error == 0)
556 error = nfs_refresh_inode(inode, fattr);
557 nfs_free_fattr(fattr);
558out:
559 trace_nfs_setattr_exit(inode, error);
560 return error;
561}
562EXPORT_SYMBOL_GPL(nfs_setattr);
563
564/**
565 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
566 * @inode: inode of the file used
567 * @offset: file offset to start truncating
568 *
569 * This is a copy of the common vmtruncate, but with the locking
570 * corrected to take into account the fact that NFS requires
571 * inode->i_size to be updated under the inode->i_lock.
572 * Note: must be called with inode->i_lock held!
573 */
574static int nfs_vmtruncate(struct inode * inode, loff_t offset)
575{
576 int err;
577
578 err = inode_newsize_ok(inode, offset);
579 if (err)
580 goto out;
581
582 i_size_write(inode, offset);
583 /* Optimisation */
584 if (offset == 0)
585 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
586
587 spin_unlock(&inode->i_lock);
588 truncate_pagecache(inode, offset);
589 spin_lock(&inode->i_lock);
590out:
591 return err;
592}
593
594/**
595 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
596 * @inode: pointer to struct inode
597 * @attr: pointer to struct iattr
598 *
599 * Note: we do this in the *proc.c in order to ensure that
600 * it works for things like exclusive creates too.
601 */
602void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
603 struct nfs_fattr *fattr)
604{
605 /* Barrier: bump the attribute generation count. */
606 nfs_fattr_set_barrier(fattr);
607
608 spin_lock(&inode->i_lock);
609 NFS_I(inode)->attr_gencount = fattr->gencount;
610 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
611 if ((attr->ia_valid & ATTR_MODE) != 0) {
612 int mode = attr->ia_mode & S_IALLUGO;
613 mode |= inode->i_mode & ~S_IALLUGO;
614 inode->i_mode = mode;
615 }
616 if ((attr->ia_valid & ATTR_UID) != 0)
617 inode->i_uid = attr->ia_uid;
618 if ((attr->ia_valid & ATTR_GID) != 0)
619 inode->i_gid = attr->ia_gid;
620 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
621 | NFS_INO_INVALID_ACL);
622 }
623 if ((attr->ia_valid & ATTR_SIZE) != 0) {
624 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
625 nfs_vmtruncate(inode, attr->ia_size);
626 }
627 if (fattr->valid)
628 nfs_update_inode(inode, fattr);
629 else
630 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
631 spin_unlock(&inode->i_lock);
632}
633EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
634
635static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
636{
637 struct dentry *parent;
638
639 parent = dget_parent(dentry);
640 nfs_force_use_readdirplus(d_inode(parent));
641 dput(parent);
642}
643
644static bool nfs_need_revalidate_inode(struct inode *inode)
645{
646 if (NFS_I(inode)->cache_validity &
647 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
648 return true;
649 if (nfs_attribute_cache_expired(inode))
650 return true;
651 return false;
652}
653
654int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
655{
656 struct inode *inode = d_inode(dentry);
657 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
658 int err = 0;
659
660 trace_nfs_getattr_enter(inode);
661 /* Flush out writes to the server in order to update c/mtime. */
662 if (S_ISREG(inode->i_mode)) {
663 mutex_lock(&inode->i_mutex);
664 err = nfs_sync_inode(inode);
665 mutex_unlock(&inode->i_mutex);
666 if (err)
667 goto out;
668 }
669
670 /*
671 * We may force a getattr if the user cares about atime.
672 *
673 * Note that we only have to check the vfsmount flags here:
674 * - NFS always sets S_NOATIME by so checking it would give a
675 * bogus result
676 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
677 * no point in checking those.
678 */
679 if ((mnt->mnt_flags & MNT_NOATIME) ||
680 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
681 need_atime = 0;
682
683 if (need_atime || nfs_need_revalidate_inode(inode)) {
684 struct nfs_server *server = NFS_SERVER(inode);
685
686 if (server->caps & NFS_CAP_READDIRPLUS)
687 nfs_request_parent_use_readdirplus(dentry);
688 err = __nfs_revalidate_inode(server, inode);
689 }
690 if (!err) {
691 generic_fillattr(inode, stat);
692 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
693 if (S_ISDIR(inode->i_mode))
694 stat->blksize = NFS_SERVER(inode)->dtsize;
695 }
696out:
697 trace_nfs_getattr_exit(inode, err);
698 return err;
699}
700EXPORT_SYMBOL_GPL(nfs_getattr);
701
702static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
703{
704 atomic_set(&l_ctx->count, 1);
705 l_ctx->lockowner.l_owner = current->files;
706 l_ctx->lockowner.l_pid = current->tgid;
707 INIT_LIST_HEAD(&l_ctx->list);
708 atomic_set(&l_ctx->io_count, 0);
709}
710
711static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
712{
713 struct nfs_lock_context *head = &ctx->lock_context;
714 struct nfs_lock_context *pos = head;
715
716 do {
717 if (pos->lockowner.l_owner != current->files)
718 continue;
719 if (pos->lockowner.l_pid != current->tgid)
720 continue;
721 atomic_inc(&pos->count);
722 return pos;
723 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
724 return NULL;
725}
726
727struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
728{
729 struct nfs_lock_context *res, *new = NULL;
730 struct inode *inode = d_inode(ctx->dentry);
731
732 spin_lock(&inode->i_lock);
733 res = __nfs_find_lock_context(ctx);
734 if (res == NULL) {
735 spin_unlock(&inode->i_lock);
736 new = kmalloc(sizeof(*new), GFP_KERNEL);
737 if (new == NULL)
738 return ERR_PTR(-ENOMEM);
739 nfs_init_lock_context(new);
740 spin_lock(&inode->i_lock);
741 res = __nfs_find_lock_context(ctx);
742 if (res == NULL) {
743 list_add_tail(&new->list, &ctx->lock_context.list);
744 new->open_context = ctx;
745 res = new;
746 new = NULL;
747 }
748 }
749 spin_unlock(&inode->i_lock);
750 kfree(new);
751 return res;
752}
753EXPORT_SYMBOL_GPL(nfs_get_lock_context);
754
755void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
756{
757 struct nfs_open_context *ctx = l_ctx->open_context;
758 struct inode *inode = d_inode(ctx->dentry);
759
760 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
761 return;
762 list_del(&l_ctx->list);
763 spin_unlock(&inode->i_lock);
764 kfree(l_ctx);
765}
766EXPORT_SYMBOL_GPL(nfs_put_lock_context);
767
768/**
769 * nfs_close_context - Common close_context() routine NFSv2/v3
770 * @ctx: pointer to context
771 * @is_sync: is this a synchronous close
772 *
773 * Ensure that the attributes are up to date if we're mounted
774 * with close-to-open semantics and we have cached data that will
775 * need to be revalidated on open.
776 */
777void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
778{
779 struct nfs_inode *nfsi;
780 struct inode *inode;
781 struct nfs_server *server;
782
783 if (!(ctx->mode & FMODE_WRITE))
784 return;
785 if (!is_sync)
786 return;
787 inode = d_inode(ctx->dentry);
788 nfsi = NFS_I(inode);
789 if (inode->i_mapping->nrpages == 0)
790 return;
791 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
792 return;
793 if (!list_empty(&nfsi->open_files))
794 return;
795 server = NFS_SERVER(inode);
796 if (server->flags & NFS_MOUNT_NOCTO)
797 return;
798 nfs_revalidate_inode(server, inode);
799}
800EXPORT_SYMBOL_GPL(nfs_close_context);
801
802struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
803{
804 struct nfs_open_context *ctx;
805 struct rpc_cred *cred = rpc_lookup_cred();
806 if (IS_ERR(cred))
807 return ERR_CAST(cred);
808
809 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
810 if (!ctx) {
811 put_rpccred(cred);
812 return ERR_PTR(-ENOMEM);
813 }
814 nfs_sb_active(dentry->d_sb);
815 ctx->dentry = dget(dentry);
816 ctx->cred = cred;
817 ctx->state = NULL;
818 ctx->mode = f_mode;
819 ctx->flags = 0;
820 ctx->error = 0;
821 nfs_init_lock_context(&ctx->lock_context);
822 ctx->lock_context.open_context = ctx;
823 INIT_LIST_HEAD(&ctx->list);
824 ctx->mdsthreshold = NULL;
825 return ctx;
826}
827EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
828
829struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
830{
831 if (ctx != NULL)
832 atomic_inc(&ctx->lock_context.count);
833 return ctx;
834}
835EXPORT_SYMBOL_GPL(get_nfs_open_context);
836
837static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
838{
839 struct inode *inode = d_inode(ctx->dentry);
840 struct super_block *sb = ctx->dentry->d_sb;
841
842 if (!list_empty(&ctx->list)) {
843 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
844 return;
845 list_del(&ctx->list);
846 spin_unlock(&inode->i_lock);
847 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
848 return;
849 if (inode != NULL)
850 NFS_PROTO(inode)->close_context(ctx, is_sync);
851 if (ctx->cred != NULL)
852 put_rpccred(ctx->cred);
853 dput(ctx->dentry);
854 nfs_sb_deactive(sb);
855 kfree(ctx->mdsthreshold);
856 kfree(ctx);
857}
858
859void put_nfs_open_context(struct nfs_open_context *ctx)
860{
861 __put_nfs_open_context(ctx, 0);
862}
863EXPORT_SYMBOL_GPL(put_nfs_open_context);
864
865static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
866{
867 __put_nfs_open_context(ctx, 1);
868}
869
870/*
871 * Ensure that mmap has a recent RPC credential for use when writing out
872 * shared pages
873 */
874void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
875{
876 struct inode *inode = d_inode(ctx->dentry);
877 struct nfs_inode *nfsi = NFS_I(inode);
878
879 spin_lock(&inode->i_lock);
880 list_add(&ctx->list, &nfsi->open_files);
881 spin_unlock(&inode->i_lock);
882}
883EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
884
885void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
886{
887 filp->private_data = get_nfs_open_context(ctx);
888 if (list_empty(&ctx->list))
889 nfs_inode_attach_open_context(ctx);
890}
891EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
892
893/*
894 * Given an inode, search for an open context with the desired characteristics
895 */
896struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
897{
898 struct nfs_inode *nfsi = NFS_I(inode);
899 struct nfs_open_context *pos, *ctx = NULL;
900
901 spin_lock(&inode->i_lock);
902 list_for_each_entry(pos, &nfsi->open_files, list) {
903 if (cred != NULL && pos->cred != cred)
904 continue;
905 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
906 continue;
907 ctx = get_nfs_open_context(pos);
908 break;
909 }
910 spin_unlock(&inode->i_lock);
911 return ctx;
912}
913
914void nfs_file_clear_open_context(struct file *filp)
915{
916 struct nfs_open_context *ctx = nfs_file_open_context(filp);
917
918 if (ctx) {
919 struct inode *inode = d_inode(ctx->dentry);
920
921 /*
922 * We fatal error on write before. Try to writeback
923 * every page again.
924 */
925 if (ctx->error < 0)
926 invalidate_inode_pages2(inode->i_mapping);
927 filp->private_data = NULL;
928 spin_lock(&inode->i_lock);
929 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
930 spin_unlock(&inode->i_lock);
931 put_nfs_open_context_sync(ctx);
932 }
933}
934
935/*
936 * These allocate and release file read/write context information.
937 */
938int nfs_open(struct inode *inode, struct file *filp)
939{
940 struct nfs_open_context *ctx;
941
942 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
943 if (IS_ERR(ctx))
944 return PTR_ERR(ctx);
945 nfs_file_set_open_context(filp, ctx);
946 put_nfs_open_context(ctx);
947 nfs_fscache_open_file(inode, filp);
948 return 0;
949}
950
951/*
952 * This function is called whenever some part of NFS notices that
953 * the cached attributes have to be refreshed.
954 */
955int
956__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
957{
958 int status = -ESTALE;
959 struct nfs4_label *label = NULL;
960 struct nfs_fattr *fattr = NULL;
961 struct nfs_inode *nfsi = NFS_I(inode);
962
963 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
964 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
965
966 trace_nfs_revalidate_inode_enter(inode);
967
968 if (is_bad_inode(inode))
969 goto out;
970 if (NFS_STALE(inode))
971 goto out;
972
973 status = -ENOMEM;
974 fattr = nfs_alloc_fattr();
975 if (fattr == NULL)
976 goto out;
977
978 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
979
980 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
981 if (IS_ERR(label)) {
982 status = PTR_ERR(label);
983 goto out;
984 }
985
986 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
987 if (status != 0) {
988 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
989 inode->i_sb->s_id,
990 (unsigned long long)NFS_FILEID(inode), status);
991 if (status == -ESTALE) {
992 nfs_zap_caches(inode);
993 if (!S_ISDIR(inode->i_mode))
994 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
995 }
996 goto err_out;
997 }
998
999 status = nfs_refresh_inode(inode, fattr);
1000 if (status) {
1001 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1002 inode->i_sb->s_id,
1003 (unsigned long long)NFS_FILEID(inode), status);
1004 goto err_out;
1005 }
1006
1007 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1008 nfs_zap_acl_cache(inode);
1009
1010 nfs_setsecurity(inode, fattr, label);
1011
1012 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1013 inode->i_sb->s_id,
1014 (unsigned long long)NFS_FILEID(inode));
1015
1016err_out:
1017 nfs4_label_free(label);
1018out:
1019 nfs_free_fattr(fattr);
1020 trace_nfs_revalidate_inode_exit(inode, status);
1021 return status;
1022}
1023
1024int nfs_attribute_timeout(struct inode *inode)
1025{
1026 struct nfs_inode *nfsi = NFS_I(inode);
1027
1028 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1029}
1030
1031int nfs_attribute_cache_expired(struct inode *inode)
1032{
1033 if (nfs_have_delegated_attributes(inode))
1034 return 0;
1035 return nfs_attribute_timeout(inode);
1036}
1037
1038/**
1039 * nfs_revalidate_inode - Revalidate the inode attributes
1040 * @server - pointer to nfs_server struct
1041 * @inode - pointer to inode struct
1042 *
1043 * Updates inode attribute information by retrieving the data from the server.
1044 */
1045int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1046{
1047 if (!nfs_need_revalidate_inode(inode))
1048 return NFS_STALE(inode) ? -ESTALE : 0;
1049 return __nfs_revalidate_inode(server, inode);
1050}
1051EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1052
1053int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1054{
1055 if (!(NFS_I(inode)->cache_validity &
1056 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1057 && !nfs_attribute_cache_expired(inode))
1058 return NFS_STALE(inode) ? -ESTALE : 0;
1059 return -ECHILD;
1060}
1061
1062static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1063{
1064 struct nfs_inode *nfsi = NFS_I(inode);
1065 int ret;
1066
1067 if (mapping->nrpages != 0) {
1068 if (S_ISREG(inode->i_mode)) {
1069 unmap_mapping_range(mapping, 0, 0, 0);
1070 ret = nfs_sync_mapping(mapping);
1071 if (ret < 0)
1072 return ret;
1073 }
1074 ret = invalidate_inode_pages2(mapping);
1075 if (ret < 0)
1076 return ret;
1077 }
1078 if (S_ISDIR(inode->i_mode)) {
1079 spin_lock(&inode->i_lock);
1080 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1081 spin_unlock(&inode->i_lock);
1082 }
1083 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1084 nfs_fscache_wait_on_invalidate(inode);
1085
1086 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1087 inode->i_sb->s_id,
1088 (unsigned long long)NFS_FILEID(inode));
1089 return 0;
1090}
1091
1092static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1093{
1094 if (nfs_have_delegated_attributes(inode))
1095 return false;
1096 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1097 || nfs_attribute_timeout(inode)
1098 || NFS_STALE(inode);
1099}
1100
1101/**
1102 * __nfs_revalidate_mapping - Revalidate the pagecache
1103 * @inode - pointer to host inode
1104 * @mapping - pointer to mapping
1105 * @may_lock - take inode->i_mutex?
1106 */
1107static int __nfs_revalidate_mapping(struct inode *inode,
1108 struct address_space *mapping,
1109 bool may_lock)
1110{
1111 struct nfs_inode *nfsi = NFS_I(inode);
1112 unsigned long *bitlock = &nfsi->flags;
1113 int ret = 0;
1114
1115 /* swapfiles are not supposed to be shared. */
1116 if (IS_SWAPFILE(inode))
1117 goto out;
1118
1119 if (nfs_mapping_need_revalidate_inode(inode)) {
1120 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1121 if (ret < 0)
1122 goto out;
1123 }
1124
1125 /*
1126 * We must clear NFS_INO_INVALID_DATA first to ensure that
1127 * invalidations that come in while we're shooting down the mappings
1128 * are respected. But, that leaves a race window where one revalidator
1129 * can clear the flag, and then another checks it before the mapping
1130 * gets invalidated. Fix that by serializing access to this part of
1131 * the function.
1132 *
1133 * At the same time, we need to allow other tasks to see whether we
1134 * might be in the middle of invalidating the pages, so we only set
1135 * the bit lock here if it looks like we're going to be doing that.
1136 */
1137 for (;;) {
1138 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1139 nfs_wait_bit_killable, TASK_KILLABLE);
1140 if (ret)
1141 goto out;
1142 spin_lock(&inode->i_lock);
1143 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1144 spin_unlock(&inode->i_lock);
1145 continue;
1146 }
1147 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1148 break;
1149 spin_unlock(&inode->i_lock);
1150 goto out;
1151 }
1152
1153 set_bit(NFS_INO_INVALIDATING, bitlock);
1154 smp_wmb();
1155 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1156 spin_unlock(&inode->i_lock);
1157 trace_nfs_invalidate_mapping_enter(inode);
1158 if (may_lock) {
1159 mutex_lock(&inode->i_mutex);
1160 ret = nfs_invalidate_mapping(inode, mapping);
1161 mutex_unlock(&inode->i_mutex);
1162 } else
1163 ret = nfs_invalidate_mapping(inode, mapping);
1164 trace_nfs_invalidate_mapping_exit(inode, ret);
1165
1166 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1167 smp_mb__after_atomic();
1168 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1169out:
1170 return ret;
1171}
1172
1173/**
1174 * nfs_revalidate_mapping - Revalidate the pagecache
1175 * @inode - pointer to host inode
1176 * @mapping - pointer to mapping
1177 */
1178int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1179{
1180 return __nfs_revalidate_mapping(inode, mapping, false);
1181}
1182
1183/**
1184 * nfs_revalidate_mapping_protected - Revalidate the pagecache
1185 * @inode - pointer to host inode
1186 * @mapping - pointer to mapping
1187 *
1188 * Differs from nfs_revalidate_mapping() in that it grabs the inode->i_mutex
1189 * while invalidating the mapping.
1190 */
1191int nfs_revalidate_mapping_protected(struct inode *inode, struct address_space *mapping)
1192{
1193 return __nfs_revalidate_mapping(inode, mapping, true);
1194}
1195
1196static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1197{
1198 struct nfs_inode *nfsi = NFS_I(inode);
1199 unsigned long ret = 0;
1200
1201 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1202 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1203 && inode->i_version == fattr->pre_change_attr) {
1204 inode->i_version = fattr->change_attr;
1205 if (S_ISDIR(inode->i_mode))
1206 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1207 ret |= NFS_INO_INVALID_ATTR;
1208 }
1209 /* If we have atomic WCC data, we may update some attributes */
1210 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1211 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1212 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1213 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1214 ret |= NFS_INO_INVALID_ATTR;
1215 }
1216
1217 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1218 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1219 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1220 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1221 if (S_ISDIR(inode->i_mode))
1222 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1223 ret |= NFS_INO_INVALID_ATTR;
1224 }
1225 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1226 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1227 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1228 && nfsi->nrequests == 0) {
1229 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1230 ret |= NFS_INO_INVALID_ATTR;
1231 }
1232
1233 return ret;
1234}
1235
1236/**
1237 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1238 * @inode - pointer to inode
1239 * @fattr - updated attributes
1240 *
1241 * Verifies the attribute cache. If we have just changed the attributes,
1242 * so that fattr carries weak cache consistency data, then it may
1243 * also update the ctime/mtime/change_attribute.
1244 */
1245static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1246{
1247 struct nfs_inode *nfsi = NFS_I(inode);
1248 loff_t cur_size, new_isize;
1249 unsigned long invalid = 0;
1250
1251
1252 if (nfs_have_delegated_attributes(inode))
1253 return 0;
1254 /* Has the inode gone and changed behind our back? */
1255 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1256 return -EIO;
1257 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1258 return -EIO;
1259
1260 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1261 inode->i_version != fattr->change_attr)
1262 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1263
1264 /* Verify a few of the more important attributes */
1265 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1266 invalid |= NFS_INO_INVALID_ATTR;
1267
1268 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1269 cur_size = i_size_read(inode);
1270 new_isize = nfs_size_to_loff_t(fattr->size);
1271 if (cur_size != new_isize)
1272 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1273 }
1274 if (nfsi->nrequests != 0)
1275 invalid &= ~NFS_INO_REVAL_PAGECACHE;
1276
1277 /* Have any file permissions changed? */
1278 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1279 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1280 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1281 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1282 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1283 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1284
1285 /* Has the link count changed? */
1286 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1287 invalid |= NFS_INO_INVALID_ATTR;
1288
1289 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1290 invalid |= NFS_INO_INVALID_ATIME;
1291
1292 if (invalid != 0)
1293 nfs_set_cache_invalid(inode, invalid);
1294
1295 nfsi->read_cache_jiffies = fattr->time_start;
1296 return 0;
1297}
1298
1299static atomic_long_t nfs_attr_generation_counter;
1300
1301static unsigned long nfs_read_attr_generation_counter(void)
1302{
1303 return atomic_long_read(&nfs_attr_generation_counter);
1304}
1305
1306unsigned long nfs_inc_attr_generation_counter(void)
1307{
1308 return atomic_long_inc_return(&nfs_attr_generation_counter);
1309}
1310EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1311
1312void nfs_fattr_init(struct nfs_fattr *fattr)
1313{
1314 fattr->valid = 0;
1315 fattr->time_start = jiffies;
1316 fattr->gencount = nfs_inc_attr_generation_counter();
1317 fattr->owner_name = NULL;
1318 fattr->group_name = NULL;
1319}
1320EXPORT_SYMBOL_GPL(nfs_fattr_init);
1321
1322/**
1323 * nfs_fattr_set_barrier
1324 * @fattr: attributes
1325 *
1326 * Used to set a barrier after an attribute was updated. This
1327 * barrier ensures that older attributes from RPC calls that may
1328 * have raced with our update cannot clobber these new values.
1329 * Note that you are still responsible for ensuring that other
1330 * operations which change the attribute on the server do not
1331 * collide.
1332 */
1333void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1334{
1335 fattr->gencount = nfs_inc_attr_generation_counter();
1336}
1337
1338struct nfs_fattr *nfs_alloc_fattr(void)
1339{
1340 struct nfs_fattr *fattr;
1341
1342 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1343 if (fattr != NULL)
1344 nfs_fattr_init(fattr);
1345 return fattr;
1346}
1347EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1348
1349struct nfs_fh *nfs_alloc_fhandle(void)
1350{
1351 struct nfs_fh *fh;
1352
1353 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1354 if (fh != NULL)
1355 fh->size = 0;
1356 return fh;
1357}
1358EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1359
1360#ifdef NFS_DEBUG
1361/*
1362 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1363 * in the same way that wireshark does
1364 *
1365 * @fh: file handle
1366 *
1367 * For debugging only.
1368 */
1369u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1370{
1371 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1372 * not on the result */
1373 return nfs_fhandle_hash(fh);
1374}
1375EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1376
1377/*
1378 * _nfs_display_fhandle - display an NFS file handle on the console
1379 *
1380 * @fh: file handle to display
1381 * @caption: display caption
1382 *
1383 * For debugging only.
1384 */
1385void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1386{
1387 unsigned short i;
1388
1389 if (fh == NULL || fh->size == 0) {
1390 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1391 return;
1392 }
1393
1394 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1395 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1396 for (i = 0; i < fh->size; i += 16) {
1397 __be32 *pos = (__be32 *)&fh->data[i];
1398
1399 switch ((fh->size - i - 1) >> 2) {
1400 case 0:
1401 printk(KERN_DEFAULT " %08x\n",
1402 be32_to_cpup(pos));
1403 break;
1404 case 1:
1405 printk(KERN_DEFAULT " %08x %08x\n",
1406 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1407 break;
1408 case 2:
1409 printk(KERN_DEFAULT " %08x %08x %08x\n",
1410 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1411 be32_to_cpup(pos + 2));
1412 break;
1413 default:
1414 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1415 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1416 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1417 }
1418 }
1419}
1420EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1421#endif
1422
1423/**
1424 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1425 * @inode - pointer to inode
1426 * @fattr - attributes
1427 *
1428 * Attempt to divine whether or not an RPC call reply carrying stale
1429 * attributes got scheduled after another call carrying updated ones.
1430 *
1431 * To do so, the function first assumes that a more recent ctime means
1432 * that the attributes in fattr are newer, however it also attempt to
1433 * catch the case where ctime either didn't change, or went backwards
1434 * (if someone reset the clock on the server) by looking at whether
1435 * or not this RPC call was started after the inode was last updated.
1436 * Note also the check for wraparound of 'attr_gencount'
1437 *
1438 * The function returns 'true' if it thinks the attributes in 'fattr' are
1439 * more recent than the ones cached in the inode.
1440 *
1441 */
1442static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1443{
1444 const struct nfs_inode *nfsi = NFS_I(inode);
1445
1446 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1447 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1448}
1449
1450/*
1451 * Don't trust the change_attribute, mtime, ctime or size if
1452 * a pnfs LAYOUTCOMMIT is outstanding
1453 */
1454static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1455 struct nfs_fattr *fattr)
1456{
1457 if (pnfs_layoutcommit_outstanding(inode))
1458 fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1459 NFS_ATTR_FATTR_MTIME |
1460 NFS_ATTR_FATTR_CTIME |
1461 NFS_ATTR_FATTR_SIZE);
1462}
1463
1464static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1465{
1466 int ret;
1467
1468 trace_nfs_refresh_inode_enter(inode);
1469
1470 nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1471
1472 if (nfs_inode_attrs_need_update(inode, fattr))
1473 ret = nfs_update_inode(inode, fattr);
1474 else
1475 ret = nfs_check_inode_attributes(inode, fattr);
1476
1477 trace_nfs_refresh_inode_exit(inode, ret);
1478 return ret;
1479}
1480
1481/**
1482 * nfs_refresh_inode - try to update the inode attribute cache
1483 * @inode - pointer to inode
1484 * @fattr - updated attributes
1485 *
1486 * Check that an RPC call that returned attributes has not overlapped with
1487 * other recent updates of the inode metadata, then decide whether it is
1488 * safe to do a full update of the inode attributes, or whether just to
1489 * call nfs_check_inode_attributes.
1490 */
1491int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1492{
1493 int status;
1494
1495 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1496 return 0;
1497 spin_lock(&inode->i_lock);
1498 status = nfs_refresh_inode_locked(inode, fattr);
1499 spin_unlock(&inode->i_lock);
1500
1501 return status;
1502}
1503EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1504
1505static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1506{
1507 unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1508
1509 /*
1510 * Don't revalidate the pagecache if we hold a delegation, but do
1511 * force an attribute update
1512 */
1513 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1514 invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_FORCED;
1515
1516 if (S_ISDIR(inode->i_mode))
1517 invalid |= NFS_INO_INVALID_DATA;
1518 nfs_set_cache_invalid(inode, invalid);
1519 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1520 return 0;
1521 return nfs_refresh_inode_locked(inode, fattr);
1522}
1523
1524/**
1525 * nfs_post_op_update_inode - try to update the inode attribute cache
1526 * @inode - pointer to inode
1527 * @fattr - updated attributes
1528 *
1529 * After an operation that has changed the inode metadata, mark the
1530 * attribute cache as being invalid, then try to update it.
1531 *
1532 * NB: if the server didn't return any post op attributes, this
1533 * function will force the retrieval of attributes before the next
1534 * NFS request. Thus it should be used only for operations that
1535 * are expected to change one or more attributes, to avoid
1536 * unnecessary NFS requests and trips through nfs_update_inode().
1537 */
1538int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1539{
1540 int status;
1541
1542 spin_lock(&inode->i_lock);
1543 nfs_fattr_set_barrier(fattr);
1544 status = nfs_post_op_update_inode_locked(inode, fattr);
1545 spin_unlock(&inode->i_lock);
1546
1547 return status;
1548}
1549EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1550
1551/**
1552 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1553 * @inode - pointer to inode
1554 * @fattr - updated attributes
1555 *
1556 * After an operation that has changed the inode metadata, mark the
1557 * attribute cache as being invalid, then try to update it. Fake up
1558 * weak cache consistency data, if none exist.
1559 *
1560 * This function is mainly designed to be used by the ->write_done() functions.
1561 */
1562int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1563{
1564 int status;
1565
1566 /* Don't do a WCC update if these attributes are already stale */
1567 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1568 !nfs_inode_attrs_need_update(inode, fattr)) {
1569 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1570 | NFS_ATTR_FATTR_PRESIZE
1571 | NFS_ATTR_FATTR_PREMTIME
1572 | NFS_ATTR_FATTR_PRECTIME);
1573 goto out_noforce;
1574 }
1575 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1576 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1577 fattr->pre_change_attr = inode->i_version;
1578 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1579 }
1580 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1581 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1582 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1583 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1584 }
1585 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1586 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1587 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1588 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1589 }
1590 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1591 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1592 fattr->pre_size = i_size_read(inode);
1593 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1594 }
1595out_noforce:
1596 status = nfs_post_op_update_inode_locked(inode, fattr);
1597 return status;
1598}
1599
1600/**
1601 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1602 * @inode - pointer to inode
1603 * @fattr - updated attributes
1604 *
1605 * After an operation that has changed the inode metadata, mark the
1606 * attribute cache as being invalid, then try to update it. Fake up
1607 * weak cache consistency data, if none exist.
1608 *
1609 * This function is mainly designed to be used by the ->write_done() functions.
1610 */
1611int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1612{
1613 int status;
1614
1615 spin_lock(&inode->i_lock);
1616 nfs_fattr_set_barrier(fattr);
1617 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1618 spin_unlock(&inode->i_lock);
1619 return status;
1620}
1621EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1622
1623
1624static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1625 struct nfs_fattr *fattr)
1626{
1627 bool ret1 = true, ret2 = true;
1628
1629 if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1630 ret1 = (nfsi->fileid == fattr->fileid);
1631 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1632 ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1633 return ret1 || ret2;
1634}
1635
1636/*
1637 * Many nfs protocol calls return the new file attributes after
1638 * an operation. Here we update the inode to reflect the state
1639 * of the server's inode.
1640 *
1641 * This is a bit tricky because we have to make sure all dirty pages
1642 * have been sent off to the server before calling invalidate_inode_pages.
1643 * To make sure no other process adds more write requests while we try
1644 * our best to flush them, we make them sleep during the attribute refresh.
1645 *
1646 * A very similar scenario holds for the dir cache.
1647 */
1648static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1649{
1650 struct nfs_server *server;
1651 struct nfs_inode *nfsi = NFS_I(inode);
1652 loff_t cur_isize, new_isize;
1653 unsigned long invalid = 0;
1654 unsigned long now = jiffies;
1655 unsigned long save_cache_validity;
1656
1657 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1658 __func__, inode->i_sb->s_id, inode->i_ino,
1659 nfs_display_fhandle_hash(NFS_FH(inode)),
1660 atomic_read(&inode->i_count), fattr->valid);
1661
1662 if (!nfs_fileid_valid(nfsi, fattr)) {
1663 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1664 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1665 NFS_SERVER(inode)->nfs_client->cl_hostname,
1666 inode->i_sb->s_id, (long long)nfsi->fileid,
1667 (long long)fattr->fileid);
1668 goto out_err;
1669 }
1670
1671 /*
1672 * Make sure the inode's type hasn't changed.
1673 */
1674 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1675 /*
1676 * Big trouble! The inode has become a different object.
1677 */
1678 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1679 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1680 goto out_err;
1681 }
1682
1683 server = NFS_SERVER(inode);
1684 /* Update the fsid? */
1685 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1686 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1687 !IS_AUTOMOUNT(inode))
1688 server->fsid = fattr->fsid;
1689
1690 /*
1691 * Update the read time so we don't revalidate too often.
1692 */
1693 nfsi->read_cache_jiffies = fattr->time_start;
1694
1695 save_cache_validity = nfsi->cache_validity;
1696 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1697 | NFS_INO_INVALID_ATIME
1698 | NFS_INO_REVAL_FORCED
1699 | NFS_INO_REVAL_PAGECACHE);
1700
1701 /* Do atomic weak cache consistency updates */
1702 invalid |= nfs_wcc_update_inode(inode, fattr);
1703
1704 /* More cache consistency checks */
1705 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1706 if (inode->i_version != fattr->change_attr) {
1707 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1708 inode->i_sb->s_id, inode->i_ino);
1709 invalid |= NFS_INO_INVALID_ATTR
1710 | NFS_INO_INVALID_DATA
1711 | NFS_INO_INVALID_ACCESS
1712 | NFS_INO_INVALID_ACL;
1713 if (S_ISDIR(inode->i_mode))
1714 nfs_force_lookup_revalidate(inode);
1715 inode->i_version = fattr->change_attr;
1716 }
1717 } else
1718 nfsi->cache_validity |= save_cache_validity;
1719
1720 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1721 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1722 } else if (server->caps & NFS_CAP_MTIME)
1723 nfsi->cache_validity |= save_cache_validity &
1724 (NFS_INO_INVALID_ATTR
1725 | NFS_INO_REVAL_FORCED);
1726
1727 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1728 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1729 } else if (server->caps & NFS_CAP_CTIME)
1730 nfsi->cache_validity |= save_cache_validity &
1731 (NFS_INO_INVALID_ATTR
1732 | NFS_INO_REVAL_FORCED);
1733
1734 /* Check if our cached file size is stale */
1735 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1736 new_isize = nfs_size_to_loff_t(fattr->size);
1737 cur_isize = i_size_read(inode);
1738 if (new_isize != cur_isize) {
1739 /* Do we perhaps have any outstanding writes, or has
1740 * the file grown beyond our last write? */
1741 if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1742 i_size_write(inode, new_isize);
1743 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1744 }
1745 dprintk("NFS: isize change on server for file %s/%ld "
1746 "(%Ld to %Ld)\n",
1747 inode->i_sb->s_id,
1748 inode->i_ino,
1749 (long long)cur_isize,
1750 (long long)new_isize);
1751 }
1752 } else
1753 nfsi->cache_validity |= save_cache_validity &
1754 (NFS_INO_INVALID_ATTR
1755 | NFS_INO_REVAL_PAGECACHE
1756 | NFS_INO_REVAL_FORCED);
1757
1758
1759 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1760 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1761 else if (server->caps & NFS_CAP_ATIME)
1762 nfsi->cache_validity |= save_cache_validity &
1763 (NFS_INO_INVALID_ATIME
1764 | NFS_INO_REVAL_FORCED);
1765
1766 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1767 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1768 umode_t newmode = inode->i_mode & S_IFMT;
1769 newmode |= fattr->mode & S_IALLUGO;
1770 inode->i_mode = newmode;
1771 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1772 }
1773 } else if (server->caps & NFS_CAP_MODE)
1774 nfsi->cache_validity |= save_cache_validity &
1775 (NFS_INO_INVALID_ATTR
1776 | NFS_INO_INVALID_ACCESS
1777 | NFS_INO_INVALID_ACL
1778 | NFS_INO_REVAL_FORCED);
1779
1780 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1781 if (!uid_eq(inode->i_uid, fattr->uid)) {
1782 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1783 inode->i_uid = fattr->uid;
1784 }
1785 } else if (server->caps & NFS_CAP_OWNER)
1786 nfsi->cache_validity |= save_cache_validity &
1787 (NFS_INO_INVALID_ATTR
1788 | NFS_INO_INVALID_ACCESS
1789 | NFS_INO_INVALID_ACL
1790 | NFS_INO_REVAL_FORCED);
1791
1792 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1793 if (!gid_eq(inode->i_gid, fattr->gid)) {
1794 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1795 inode->i_gid = fattr->gid;
1796 }
1797 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1798 nfsi->cache_validity |= save_cache_validity &
1799 (NFS_INO_INVALID_ATTR
1800 | NFS_INO_INVALID_ACCESS
1801 | NFS_INO_INVALID_ACL
1802 | NFS_INO_REVAL_FORCED);
1803
1804 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1805 if (inode->i_nlink != fattr->nlink) {
1806 invalid |= NFS_INO_INVALID_ATTR;
1807 if (S_ISDIR(inode->i_mode))
1808 invalid |= NFS_INO_INVALID_DATA;
1809 set_nlink(inode, fattr->nlink);
1810 }
1811 } else if (server->caps & NFS_CAP_NLINK)
1812 nfsi->cache_validity |= save_cache_validity &
1813 (NFS_INO_INVALID_ATTR
1814 | NFS_INO_REVAL_FORCED);
1815
1816 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1817 /*
1818 * report the blocks in 512byte units
1819 */
1820 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1821 }
1822 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1823 inode->i_blocks = fattr->du.nfs2.blocks;
1824
1825 /* Update attrtimeo value if we're out of the unstable period */
1826 if (invalid & NFS_INO_INVALID_ATTR) {
1827 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1828 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1829 nfsi->attrtimeo_timestamp = now;
1830 /* Set barrier to be more recent than all outstanding updates */
1831 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1832 } else {
1833 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1834 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1835 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1836 nfsi->attrtimeo_timestamp = now;
1837 }
1838 /* Set the barrier to be more recent than this fattr */
1839 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1840 nfsi->attr_gencount = fattr->gencount;
1841 }
1842
1843 /* Don't declare attrcache up to date if there were no attrs! */
1844 if (fattr->valid != 0)
1845 invalid &= ~NFS_INO_INVALID_ATTR;
1846
1847 /* Don't invalidate the data if we were to blame */
1848 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1849 || S_ISLNK(inode->i_mode)))
1850 invalid &= ~NFS_INO_INVALID_DATA;
1851 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1852 (save_cache_validity & NFS_INO_REVAL_FORCED))
1853 nfs_set_cache_invalid(inode, invalid);
1854
1855 return 0;
1856 out_err:
1857 /*
1858 * No need to worry about unhashing the dentry, as the
1859 * lookup validation will know that the inode is bad.
1860 * (But we fall through to invalidate the caches.)
1861 */
1862 nfs_invalidate_inode(inode);
1863 return -ESTALE;
1864}
1865
1866struct inode *nfs_alloc_inode(struct super_block *sb)
1867{
1868 struct nfs_inode *nfsi;
1869 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1870 if (!nfsi)
1871 return NULL;
1872 nfsi->flags = 0UL;
1873 nfsi->cache_validity = 0UL;
1874#if IS_ENABLED(CONFIG_NFS_V4)
1875 nfsi->nfs4_acl = NULL;
1876#endif /* CONFIG_NFS_V4 */
1877 return &nfsi->vfs_inode;
1878}
1879EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1880
1881static void nfs_i_callback(struct rcu_head *head)
1882{
1883 struct inode *inode = container_of(head, struct inode, i_rcu);
1884 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1885}
1886
1887void nfs_destroy_inode(struct inode *inode)
1888{
1889 call_rcu(&inode->i_rcu, nfs_i_callback);
1890}
1891EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1892
1893static inline void nfs4_init_once(struct nfs_inode *nfsi)
1894{
1895#if IS_ENABLED(CONFIG_NFS_V4)
1896 INIT_LIST_HEAD(&nfsi->open_states);
1897 nfsi->delegation = NULL;
1898 init_rwsem(&nfsi->rwsem);
1899 nfsi->layout = NULL;
1900#endif
1901}
1902
1903static void init_once(void *foo)
1904{
1905 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1906
1907 inode_init_once(&nfsi->vfs_inode);
1908 INIT_LIST_HEAD(&nfsi->open_files);
1909 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1910 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1911 INIT_LIST_HEAD(&nfsi->commit_info.list);
1912 nfsi->nrequests = 0;
1913 nfsi->commit_info.ncommit = 0;
1914 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1915 atomic_set(&nfsi->silly_count, 1);
1916 INIT_HLIST_HEAD(&nfsi->silly_list);
1917 init_waitqueue_head(&nfsi->waitqueue);
1918 nfs4_init_once(nfsi);
1919}
1920
1921static int __init nfs_init_inodecache(void)
1922{
1923 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1924 sizeof(struct nfs_inode),
1925 0, (SLAB_RECLAIM_ACCOUNT|
1926 SLAB_MEM_SPREAD),
1927 init_once);
1928 if (nfs_inode_cachep == NULL)
1929 return -ENOMEM;
1930
1931 return 0;
1932}
1933
1934static void nfs_destroy_inodecache(void)
1935{
1936 /*
1937 * Make sure all delayed rcu free inodes are flushed before we
1938 * destroy cache.
1939 */
1940 rcu_barrier();
1941 kmem_cache_destroy(nfs_inode_cachep);
1942}
1943
1944struct workqueue_struct *nfsiod_workqueue;
1945EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1946
1947/*
1948 * start up the nfsiod workqueue
1949 */
1950static int nfsiod_start(void)
1951{
1952 struct workqueue_struct *wq;
1953 dprintk("RPC: creating workqueue nfsiod\n");
1954 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1955 if (wq == NULL)
1956 return -ENOMEM;
1957 nfsiod_workqueue = wq;
1958 return 0;
1959}
1960
1961/*
1962 * Destroy the nfsiod workqueue
1963 */
1964static void nfsiod_stop(void)
1965{
1966 struct workqueue_struct *wq;
1967
1968 wq = nfsiod_workqueue;
1969 if (wq == NULL)
1970 return;
1971 nfsiod_workqueue = NULL;
1972 destroy_workqueue(wq);
1973}
1974
1975int nfs_net_id;
1976EXPORT_SYMBOL_GPL(nfs_net_id);
1977
1978static int nfs_net_init(struct net *net)
1979{
1980 nfs_clients_init(net);
1981 return nfs_fs_proc_net_init(net);
1982}
1983
1984static void nfs_net_exit(struct net *net)
1985{
1986 nfs_fs_proc_net_exit(net);
1987 nfs_cleanup_cb_ident_idr(net);
1988}
1989
1990static struct pernet_operations nfs_net_ops = {
1991 .init = nfs_net_init,
1992 .exit = nfs_net_exit,
1993 .id = &nfs_net_id,
1994 .size = sizeof(struct nfs_net),
1995};
1996
1997/*
1998 * Initialize NFS
1999 */
2000static int __init init_nfs_fs(void)
2001{
2002 int err;
2003
2004 err = register_pernet_subsys(&nfs_net_ops);
2005 if (err < 0)
2006 goto out9;
2007
2008 err = nfs_fscache_register();
2009 if (err < 0)
2010 goto out8;
2011
2012 err = nfsiod_start();
2013 if (err)
2014 goto out7;
2015
2016 err = nfs_fs_proc_init();
2017 if (err)
2018 goto out6;
2019
2020 err = nfs_init_nfspagecache();
2021 if (err)
2022 goto out5;
2023
2024 err = nfs_init_inodecache();
2025 if (err)
2026 goto out4;
2027
2028 err = nfs_init_readpagecache();
2029 if (err)
2030 goto out3;
2031
2032 err = nfs_init_writepagecache();
2033 if (err)
2034 goto out2;
2035
2036 err = nfs_init_directcache();
2037 if (err)
2038 goto out1;
2039
2040 rpc_proc_register(&init_net, &nfs_rpcstat);
2041
2042 err = register_nfs_fs();
2043 if (err)
2044 goto out0;
2045
2046 return 0;
2047out0:
2048 rpc_proc_unregister(&init_net, "nfs");
2049 nfs_destroy_directcache();
2050out1:
2051 nfs_destroy_writepagecache();
2052out2:
2053 nfs_destroy_readpagecache();
2054out3:
2055 nfs_destroy_inodecache();
2056out4:
2057 nfs_destroy_nfspagecache();
2058out5:
2059 nfs_fs_proc_exit();
2060out6:
2061 nfsiod_stop();
2062out7:
2063 nfs_fscache_unregister();
2064out8:
2065 unregister_pernet_subsys(&nfs_net_ops);
2066out9:
2067 return err;
2068}
2069
2070static void __exit exit_nfs_fs(void)
2071{
2072 nfs_destroy_directcache();
2073 nfs_destroy_writepagecache();
2074 nfs_destroy_readpagecache();
2075 nfs_destroy_inodecache();
2076 nfs_destroy_nfspagecache();
2077 nfs_fscache_unregister();
2078 unregister_pernet_subsys(&nfs_net_ops);
2079 rpc_proc_unregister(&init_net, "nfs");
2080 unregister_nfs_fs();
2081 nfs_fs_proc_exit();
2082 nfsiod_stop();
2083}
2084
2085/* Not quite true; I just maintain it */
2086MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2087MODULE_LICENSE("GPL");
2088module_param(enable_ino64, bool, 0644);
2089
2090module_init(init_nfs_fs)
2091module_exit(exit_nfs_fs)
This page took 0.042945 seconds and 5 git commands to generate.