]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/nfs/inode.c
Merge branch 'for-3.20' of git://linux-nfs.org/~bfields/linux
[mirror_ubuntu-zesty-kernel.git] / fs / nfs / inode.c
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 */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
62
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
65
66 static struct kmem_cache * nfs_inode_cachep;
67
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
70 {
71 return nfs_fileid_to_ino_t(fattr->fileid);
72 }
73
74 /**
75 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76 * @word: long word containing the bit lock
77 */
78 int nfs_wait_bit_killable(struct wait_bit_key *key)
79 {
80 if (fatal_signal_pending(current))
81 return -ERESTARTSYS;
82 freezable_schedule_unsafe();
83 return 0;
84 }
85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
86
87 /**
88 * nfs_compat_user_ino64 - returns the user-visible inode number
89 * @fileid: 64-bit fileid
90 *
91 * This function returns a 32-bit inode number if the boot parameter
92 * nfs.enable_ino64 is zero.
93 */
94 u64 nfs_compat_user_ino64(u64 fileid)
95 {
96 #ifdef CONFIG_COMPAT
97 compat_ulong_t ino;
98 #else
99 unsigned long ino;
100 #endif
101
102 if (enable_ino64)
103 return fileid;
104 ino = fileid;
105 if (sizeof(ino) < sizeof(fileid))
106 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
107 return ino;
108 }
109
110 int nfs_drop_inode(struct inode *inode)
111 {
112 return NFS_STALE(inode) || generic_drop_inode(inode);
113 }
114 EXPORT_SYMBOL_GPL(nfs_drop_inode);
115
116 void nfs_clear_inode(struct inode *inode)
117 {
118 /*
119 * The following should never happen...
120 */
121 WARN_ON_ONCE(nfs_have_writebacks(inode));
122 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
123 nfs_zap_acl_cache(inode);
124 nfs_access_zap_cache(inode);
125 nfs_fscache_clear_inode(inode);
126 }
127 EXPORT_SYMBOL_GPL(nfs_clear_inode);
128
129 void nfs_evict_inode(struct inode *inode)
130 {
131 truncate_inode_pages_final(&inode->i_data);
132 clear_inode(inode);
133 nfs_clear_inode(inode);
134 }
135
136 /**
137 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
138 */
139 int nfs_sync_mapping(struct address_space *mapping)
140 {
141 int ret = 0;
142
143 if (mapping->nrpages != 0) {
144 unmap_mapping_range(mapping, 0, 0, 0);
145 ret = nfs_wb_all(mapping->host);
146 }
147 return ret;
148 }
149
150 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
151 {
152 struct nfs_inode *nfsi = NFS_I(inode);
153
154 if (inode->i_mapping->nrpages == 0)
155 flags &= ~NFS_INO_INVALID_DATA;
156 nfsi->cache_validity |= flags;
157 if (flags & NFS_INO_INVALID_DATA)
158 nfs_fscache_invalidate(inode);
159 }
160
161 /*
162 * Invalidate the local caches
163 */
164 static void nfs_zap_caches_locked(struct inode *inode)
165 {
166 struct nfs_inode *nfsi = NFS_I(inode);
167 int mode = inode->i_mode;
168
169 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
170
171 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
172 nfsi->attrtimeo_timestamp = jiffies;
173
174 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
175 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
176 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
177 | NFS_INO_INVALID_DATA
178 | NFS_INO_INVALID_ACCESS
179 | NFS_INO_INVALID_ACL
180 | NFS_INO_REVAL_PAGECACHE);
181 } else
182 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
183 | NFS_INO_INVALID_ACCESS
184 | NFS_INO_INVALID_ACL
185 | NFS_INO_REVAL_PAGECACHE);
186 nfs_zap_label_cache_locked(nfsi);
187 }
188
189 void nfs_zap_caches(struct inode *inode)
190 {
191 spin_lock(&inode->i_lock);
192 nfs_zap_caches_locked(inode);
193 spin_unlock(&inode->i_lock);
194 }
195 EXPORT_SYMBOL_GPL(nfs_zap_caches);
196
197 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
198 {
199 if (mapping->nrpages != 0) {
200 spin_lock(&inode->i_lock);
201 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
202 spin_unlock(&inode->i_lock);
203 }
204 }
205
206 void nfs_zap_acl_cache(struct inode *inode)
207 {
208 void (*clear_acl_cache)(struct inode *);
209
210 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
211 if (clear_acl_cache != NULL)
212 clear_acl_cache(inode);
213 spin_lock(&inode->i_lock);
214 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
215 spin_unlock(&inode->i_lock);
216 }
217 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
218
219 void nfs_invalidate_atime(struct inode *inode)
220 {
221 spin_lock(&inode->i_lock);
222 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
223 spin_unlock(&inode->i_lock);
224 }
225 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
226
227 /*
228 * Invalidate, but do not unhash, the inode.
229 * NB: must be called with inode->i_lock held!
230 */
231 static void nfs_invalidate_inode(struct inode *inode)
232 {
233 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
234 nfs_zap_caches_locked(inode);
235 }
236
237 struct nfs_find_desc {
238 struct nfs_fh *fh;
239 struct nfs_fattr *fattr;
240 };
241
242 /*
243 * In NFSv3 we can have 64bit inode numbers. In order to support
244 * this, and re-exported directories (also seen in NFSv2)
245 * we are forced to allow 2 different inodes to have the same
246 * i_ino.
247 */
248 static int
249 nfs_find_actor(struct inode *inode, void *opaque)
250 {
251 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
252 struct nfs_fh *fh = desc->fh;
253 struct nfs_fattr *fattr = desc->fattr;
254
255 if (NFS_FILEID(inode) != fattr->fileid)
256 return 0;
257 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
258 return 0;
259 if (nfs_compare_fh(NFS_FH(inode), fh))
260 return 0;
261 if (is_bad_inode(inode) || NFS_STALE(inode))
262 return 0;
263 return 1;
264 }
265
266 static int
267 nfs_init_locked(struct inode *inode, void *opaque)
268 {
269 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
270 struct nfs_fattr *fattr = desc->fattr;
271
272 set_nfs_fileid(inode, fattr->fileid);
273 nfs_copy_fh(NFS_FH(inode), desc->fh);
274 return 0;
275 }
276
277 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
278 static void nfs_clear_label_invalid(struct inode *inode)
279 {
280 spin_lock(&inode->i_lock);
281 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
282 spin_unlock(&inode->i_lock);
283 }
284
285 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
286 struct nfs4_label *label)
287 {
288 int error;
289
290 if (label == NULL)
291 return;
292
293 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
294 error = security_inode_notifysecctx(inode, label->label,
295 label->len);
296 if (error)
297 printk(KERN_ERR "%s() %s %d "
298 "security_inode_notifysecctx() %d\n",
299 __func__,
300 (char *)label->label,
301 label->len, error);
302 nfs_clear_label_invalid(inode);
303 }
304 }
305
306 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
307 {
308 struct nfs4_label *label = NULL;
309 int minor_version = server->nfs_client->cl_minorversion;
310
311 if (minor_version < 2)
312 return label;
313
314 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
315 return label;
316
317 label = kzalloc(sizeof(struct nfs4_label), flags);
318 if (label == NULL)
319 return ERR_PTR(-ENOMEM);
320
321 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
322 if (label->label == NULL) {
323 kfree(label);
324 return ERR_PTR(-ENOMEM);
325 }
326 label->len = NFS4_MAXLABELLEN;
327
328 return label;
329 }
330 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
331 #else
332 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
333 struct nfs4_label *label)
334 {
335 }
336 #endif
337 EXPORT_SYMBOL_GPL(nfs_setsecurity);
338
339 /*
340 * This is our front-end to iget that looks up inodes by file handle
341 * instead of inode number.
342 */
343 struct inode *
344 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
345 {
346 struct nfs_find_desc desc = {
347 .fh = fh,
348 .fattr = fattr
349 };
350 struct inode *inode = ERR_PTR(-ENOENT);
351 unsigned long hash;
352
353 nfs_attr_check_mountpoint(sb, fattr);
354
355 if (nfs_attr_use_mounted_on_fileid(fattr))
356 fattr->fileid = fattr->mounted_on_fileid;
357 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
358 goto out_no_inode;
359 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
360 goto out_no_inode;
361
362 hash = nfs_fattr_to_ino_t(fattr);
363
364 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
365 if (inode == NULL) {
366 inode = ERR_PTR(-ENOMEM);
367 goto out_no_inode;
368 }
369
370 if (inode->i_state & I_NEW) {
371 struct nfs_inode *nfsi = NFS_I(inode);
372 unsigned long now = jiffies;
373
374 /* We set i_ino for the few things that still rely on it,
375 * such as stat(2) */
376 inode->i_ino = hash;
377
378 /* We can't support update_atime(), since the server will reset it */
379 inode->i_flags |= S_NOATIME|S_NOCMTIME;
380 inode->i_mode = fattr->mode;
381 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
382 && nfs_server_capable(inode, NFS_CAP_MODE))
383 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
384 /* Why so? Because we want revalidate for devices/FIFOs, and
385 * that's precisely what we have in nfs_file_inode_operations.
386 */
387 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
388 if (S_ISREG(inode->i_mode)) {
389 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
390 inode->i_data.a_ops = &nfs_file_aops;
391 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
392 } else if (S_ISDIR(inode->i_mode)) {
393 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
394 inode->i_fop = &nfs_dir_operations;
395 inode->i_data.a_ops = &nfs_dir_aops;
396 /* Deal with crossing mountpoints */
397 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
398 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
399 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
400 inode->i_op = &nfs_referral_inode_operations;
401 else
402 inode->i_op = &nfs_mountpoint_inode_operations;
403 inode->i_fop = NULL;
404 inode->i_flags |= S_AUTOMOUNT;
405 }
406 } else if (S_ISLNK(inode->i_mode))
407 inode->i_op = &nfs_symlink_inode_operations;
408 else
409 init_special_inode(inode, inode->i_mode, fattr->rdev);
410
411 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
412 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
413 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
414 inode->i_version = 0;
415 inode->i_size = 0;
416 clear_nlink(inode);
417 inode->i_uid = make_kuid(&init_user_ns, -2);
418 inode->i_gid = make_kgid(&init_user_ns, -2);
419 inode->i_blocks = 0;
420 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
421 nfsi->write_io = 0;
422 nfsi->read_io = 0;
423
424 nfsi->read_cache_jiffies = fattr->time_start;
425 nfsi->attr_gencount = fattr->gencount;
426 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
427 inode->i_atime = fattr->atime;
428 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
429 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
430 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
431 inode->i_mtime = fattr->mtime;
432 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
433 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
434 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
435 inode->i_ctime = fattr->ctime;
436 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
437 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
438 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
439 inode->i_version = fattr->change_attr;
440 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
441 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
442 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
443 inode->i_size = nfs_size_to_loff_t(fattr->size);
444 else
445 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
446 | NFS_INO_REVAL_PAGECACHE);
447 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
448 set_nlink(inode, fattr->nlink);
449 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
450 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
451 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
452 inode->i_uid = fattr->uid;
453 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
454 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
455 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
456 inode->i_gid = fattr->gid;
457 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
458 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
459 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
460 inode->i_blocks = fattr->du.nfs2.blocks;
461 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
462 /*
463 * report the blocks in 512byte units
464 */
465 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
466 }
467
468 nfs_setsecurity(inode, fattr, label);
469
470 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
471 nfsi->attrtimeo_timestamp = now;
472 nfsi->access_cache = RB_ROOT;
473
474 nfs_fscache_init_inode(inode);
475
476 unlock_new_inode(inode);
477 } else
478 nfs_refresh_inode(inode, fattr);
479 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
480 inode->i_sb->s_id,
481 (unsigned long long)NFS_FILEID(inode),
482 nfs_display_fhandle_hash(fh),
483 atomic_read(&inode->i_count));
484
485 out:
486 return inode;
487
488 out_no_inode:
489 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
490 goto out;
491 }
492 EXPORT_SYMBOL_GPL(nfs_fhget);
493
494 #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)
495
496 int
497 nfs_setattr(struct dentry *dentry, struct iattr *attr)
498 {
499 struct inode *inode = dentry->d_inode;
500 struct nfs_fattr *fattr;
501 int error = -ENOMEM;
502
503 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
504
505 /* skip mode change if it's just for clearing setuid/setgid */
506 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
507 attr->ia_valid &= ~ATTR_MODE;
508
509 if (attr->ia_valid & ATTR_SIZE) {
510 loff_t i_size;
511
512 BUG_ON(!S_ISREG(inode->i_mode));
513
514 i_size = i_size_read(inode);
515 if (attr->ia_size == i_size)
516 attr->ia_valid &= ~ATTR_SIZE;
517 else if (attr->ia_size < i_size && IS_SWAPFILE(inode))
518 return -ETXTBSY;
519 }
520
521 /* Optimization: if the end result is no change, don't RPC */
522 attr->ia_valid &= NFS_VALID_ATTRS;
523 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
524 return 0;
525
526 trace_nfs_setattr_enter(inode);
527
528 /* Write all dirty data */
529 if (S_ISREG(inode->i_mode)) {
530 nfs_inode_dio_wait(inode);
531 nfs_wb_all(inode);
532 }
533
534 fattr = nfs_alloc_fattr();
535 if (fattr == NULL)
536 goto out;
537 /*
538 * Return any delegations if we're going to change ACLs
539 */
540 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
541 NFS_PROTO(inode)->return_delegation(inode);
542 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
543 if (error == 0)
544 error = nfs_refresh_inode(inode, fattr);
545 nfs_free_fattr(fattr);
546 out:
547 trace_nfs_setattr_exit(inode, error);
548 return error;
549 }
550 EXPORT_SYMBOL_GPL(nfs_setattr);
551
552 /**
553 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
554 * @inode: inode of the file used
555 * @offset: file offset to start truncating
556 *
557 * This is a copy of the common vmtruncate, but with the locking
558 * corrected to take into account the fact that NFS requires
559 * inode->i_size to be updated under the inode->i_lock.
560 */
561 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
562 {
563 int err;
564
565 err = inode_newsize_ok(inode, offset);
566 if (err)
567 goto out;
568
569 spin_lock(&inode->i_lock);
570 i_size_write(inode, offset);
571 /* Optimisation */
572 if (offset == 0)
573 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
574 spin_unlock(&inode->i_lock);
575
576 truncate_pagecache(inode, offset);
577 out:
578 return err;
579 }
580
581 /**
582 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
583 * @inode: pointer to struct inode
584 * @attr: pointer to struct iattr
585 *
586 * Note: we do this in the *proc.c in order to ensure that
587 * it works for things like exclusive creates too.
588 */
589 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
590 {
591 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
592 spin_lock(&inode->i_lock);
593 if ((attr->ia_valid & ATTR_MODE) != 0) {
594 int mode = attr->ia_mode & S_IALLUGO;
595 mode |= inode->i_mode & ~S_IALLUGO;
596 inode->i_mode = mode;
597 }
598 if ((attr->ia_valid & ATTR_UID) != 0)
599 inode->i_uid = attr->ia_uid;
600 if ((attr->ia_valid & ATTR_GID) != 0)
601 inode->i_gid = attr->ia_gid;
602 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
603 | NFS_INO_INVALID_ACL);
604 spin_unlock(&inode->i_lock);
605 }
606 if ((attr->ia_valid & ATTR_SIZE) != 0) {
607 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
608 nfs_vmtruncate(inode, attr->ia_size);
609 }
610 }
611 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
612
613 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
614 {
615 struct dentry *parent;
616
617 parent = dget_parent(dentry);
618 nfs_force_use_readdirplus(parent->d_inode);
619 dput(parent);
620 }
621
622 static bool nfs_need_revalidate_inode(struct inode *inode)
623 {
624 if (NFS_I(inode)->cache_validity &
625 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
626 return true;
627 if (nfs_attribute_cache_expired(inode))
628 return true;
629 return false;
630 }
631
632 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
633 {
634 struct inode *inode = dentry->d_inode;
635 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
636 int err = 0;
637
638 trace_nfs_getattr_enter(inode);
639 /* Flush out writes to the server in order to update c/mtime. */
640 if (S_ISREG(inode->i_mode)) {
641 nfs_inode_dio_wait(inode);
642 err = filemap_write_and_wait(inode->i_mapping);
643 if (err)
644 goto out;
645 }
646
647 /*
648 * We may force a getattr if the user cares about atime.
649 *
650 * Note that we only have to check the vfsmount flags here:
651 * - NFS always sets S_NOATIME by so checking it would give a
652 * bogus result
653 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
654 * no point in checking those.
655 */
656 if ((mnt->mnt_flags & MNT_NOATIME) ||
657 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
658 need_atime = 0;
659
660 if (need_atime || nfs_need_revalidate_inode(inode)) {
661 struct nfs_server *server = NFS_SERVER(inode);
662
663 if (server->caps & NFS_CAP_READDIRPLUS)
664 nfs_request_parent_use_readdirplus(dentry);
665 err = __nfs_revalidate_inode(server, inode);
666 }
667 if (!err) {
668 generic_fillattr(inode, stat);
669 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
670 }
671 out:
672 trace_nfs_getattr_exit(inode, err);
673 return err;
674 }
675 EXPORT_SYMBOL_GPL(nfs_getattr);
676
677 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
678 {
679 atomic_set(&l_ctx->count, 1);
680 l_ctx->lockowner.l_owner = current->files;
681 l_ctx->lockowner.l_pid = current->tgid;
682 INIT_LIST_HEAD(&l_ctx->list);
683 nfs_iocounter_init(&l_ctx->io_count);
684 }
685
686 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
687 {
688 struct nfs_lock_context *head = &ctx->lock_context;
689 struct nfs_lock_context *pos = head;
690
691 do {
692 if (pos->lockowner.l_owner != current->files)
693 continue;
694 if (pos->lockowner.l_pid != current->tgid)
695 continue;
696 atomic_inc(&pos->count);
697 return pos;
698 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
699 return NULL;
700 }
701
702 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
703 {
704 struct nfs_lock_context *res, *new = NULL;
705 struct inode *inode = ctx->dentry->d_inode;
706
707 spin_lock(&inode->i_lock);
708 res = __nfs_find_lock_context(ctx);
709 if (res == NULL) {
710 spin_unlock(&inode->i_lock);
711 new = kmalloc(sizeof(*new), GFP_KERNEL);
712 if (new == NULL)
713 return ERR_PTR(-ENOMEM);
714 nfs_init_lock_context(new);
715 spin_lock(&inode->i_lock);
716 res = __nfs_find_lock_context(ctx);
717 if (res == NULL) {
718 list_add_tail(&new->list, &ctx->lock_context.list);
719 new->open_context = ctx;
720 res = new;
721 new = NULL;
722 }
723 }
724 spin_unlock(&inode->i_lock);
725 kfree(new);
726 return res;
727 }
728 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
729
730 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
731 {
732 struct nfs_open_context *ctx = l_ctx->open_context;
733 struct inode *inode = ctx->dentry->d_inode;
734
735 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
736 return;
737 list_del(&l_ctx->list);
738 spin_unlock(&inode->i_lock);
739 kfree(l_ctx);
740 }
741 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
742
743 /**
744 * nfs_close_context - Common close_context() routine NFSv2/v3
745 * @ctx: pointer to context
746 * @is_sync: is this a synchronous close
747 *
748 * always ensure that the attributes are up to date if we're mounted
749 * with close-to-open semantics
750 */
751 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
752 {
753 struct inode *inode;
754 struct nfs_server *server;
755
756 if (!(ctx->mode & FMODE_WRITE))
757 return;
758 if (!is_sync)
759 return;
760 inode = ctx->dentry->d_inode;
761 if (!list_empty(&NFS_I(inode)->open_files))
762 return;
763 server = NFS_SERVER(inode);
764 if (server->flags & NFS_MOUNT_NOCTO)
765 return;
766 nfs_revalidate_inode(server, inode);
767 }
768 EXPORT_SYMBOL_GPL(nfs_close_context);
769
770 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
771 {
772 struct nfs_open_context *ctx;
773 struct rpc_cred *cred = rpc_lookup_cred();
774 if (IS_ERR(cred))
775 return ERR_CAST(cred);
776
777 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
778 if (!ctx) {
779 put_rpccred(cred);
780 return ERR_PTR(-ENOMEM);
781 }
782 nfs_sb_active(dentry->d_sb);
783 ctx->dentry = dget(dentry);
784 ctx->cred = cred;
785 ctx->state = NULL;
786 ctx->mode = f_mode;
787 ctx->flags = 0;
788 ctx->error = 0;
789 nfs_init_lock_context(&ctx->lock_context);
790 ctx->lock_context.open_context = ctx;
791 INIT_LIST_HEAD(&ctx->list);
792 ctx->mdsthreshold = NULL;
793 return ctx;
794 }
795 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
796
797 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
798 {
799 if (ctx != NULL)
800 atomic_inc(&ctx->lock_context.count);
801 return ctx;
802 }
803 EXPORT_SYMBOL_GPL(get_nfs_open_context);
804
805 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
806 {
807 struct inode *inode = ctx->dentry->d_inode;
808 struct super_block *sb = ctx->dentry->d_sb;
809
810 if (!list_empty(&ctx->list)) {
811 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
812 return;
813 list_del(&ctx->list);
814 spin_unlock(&inode->i_lock);
815 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
816 return;
817 if (inode != NULL)
818 NFS_PROTO(inode)->close_context(ctx, is_sync);
819 if (ctx->cred != NULL)
820 put_rpccred(ctx->cred);
821 dput(ctx->dentry);
822 nfs_sb_deactive(sb);
823 kfree(ctx->mdsthreshold);
824 kfree(ctx);
825 }
826
827 void put_nfs_open_context(struct nfs_open_context *ctx)
828 {
829 __put_nfs_open_context(ctx, 0);
830 }
831 EXPORT_SYMBOL_GPL(put_nfs_open_context);
832
833 /*
834 * Ensure that mmap has a recent RPC credential for use when writing out
835 * shared pages
836 */
837 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
838 {
839 struct inode *inode = ctx->dentry->d_inode;
840 struct nfs_inode *nfsi = NFS_I(inode);
841
842 spin_lock(&inode->i_lock);
843 list_add(&ctx->list, &nfsi->open_files);
844 spin_unlock(&inode->i_lock);
845 }
846 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
847
848 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
849 {
850 filp->private_data = get_nfs_open_context(ctx);
851 if (list_empty(&ctx->list))
852 nfs_inode_attach_open_context(ctx);
853 }
854 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
855
856 /*
857 * Given an inode, search for an open context with the desired characteristics
858 */
859 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
860 {
861 struct nfs_inode *nfsi = NFS_I(inode);
862 struct nfs_open_context *pos, *ctx = NULL;
863
864 spin_lock(&inode->i_lock);
865 list_for_each_entry(pos, &nfsi->open_files, list) {
866 if (cred != NULL && pos->cred != cred)
867 continue;
868 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
869 continue;
870 ctx = get_nfs_open_context(pos);
871 break;
872 }
873 spin_unlock(&inode->i_lock);
874 return ctx;
875 }
876
877 static void nfs_file_clear_open_context(struct file *filp)
878 {
879 struct nfs_open_context *ctx = nfs_file_open_context(filp);
880
881 if (ctx) {
882 struct inode *inode = ctx->dentry->d_inode;
883
884 filp->private_data = NULL;
885 spin_lock(&inode->i_lock);
886 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
887 spin_unlock(&inode->i_lock);
888 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
889 }
890 }
891
892 /*
893 * These allocate and release file read/write context information.
894 */
895 int nfs_open(struct inode *inode, struct file *filp)
896 {
897 struct nfs_open_context *ctx;
898
899 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
900 if (IS_ERR(ctx))
901 return PTR_ERR(ctx);
902 nfs_file_set_open_context(filp, ctx);
903 put_nfs_open_context(ctx);
904 nfs_fscache_open_file(inode, filp);
905 return 0;
906 }
907
908 int nfs_release(struct inode *inode, struct file *filp)
909 {
910 nfs_file_clear_open_context(filp);
911 return 0;
912 }
913
914 /*
915 * This function is called whenever some part of NFS notices that
916 * the cached attributes have to be refreshed.
917 */
918 int
919 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
920 {
921 int status = -ESTALE;
922 struct nfs4_label *label = NULL;
923 struct nfs_fattr *fattr = NULL;
924 struct nfs_inode *nfsi = NFS_I(inode);
925
926 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
927 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
928
929 trace_nfs_revalidate_inode_enter(inode);
930
931 if (is_bad_inode(inode))
932 goto out;
933 if (NFS_STALE(inode))
934 goto out;
935
936 status = -ENOMEM;
937 fattr = nfs_alloc_fattr();
938 if (fattr == NULL)
939 goto out;
940
941 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
942
943 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
944 if (IS_ERR(label)) {
945 status = PTR_ERR(label);
946 goto out;
947 }
948
949 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
950 if (status != 0) {
951 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
952 inode->i_sb->s_id,
953 (unsigned long long)NFS_FILEID(inode), status);
954 if (status == -ESTALE) {
955 nfs_zap_caches(inode);
956 if (!S_ISDIR(inode->i_mode))
957 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
958 }
959 goto err_out;
960 }
961
962 status = nfs_refresh_inode(inode, fattr);
963 if (status) {
964 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
965 inode->i_sb->s_id,
966 (unsigned long long)NFS_FILEID(inode), status);
967 goto err_out;
968 }
969
970 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
971 nfs_zap_acl_cache(inode);
972
973 nfs_setsecurity(inode, fattr, label);
974
975 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
976 inode->i_sb->s_id,
977 (unsigned long long)NFS_FILEID(inode));
978
979 err_out:
980 nfs4_label_free(label);
981 out:
982 nfs_free_fattr(fattr);
983 trace_nfs_revalidate_inode_exit(inode, status);
984 return status;
985 }
986
987 int nfs_attribute_timeout(struct inode *inode)
988 {
989 struct nfs_inode *nfsi = NFS_I(inode);
990
991 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
992 }
993
994 int nfs_attribute_cache_expired(struct inode *inode)
995 {
996 if (nfs_have_delegated_attributes(inode))
997 return 0;
998 return nfs_attribute_timeout(inode);
999 }
1000
1001 /**
1002 * nfs_revalidate_inode - Revalidate the inode attributes
1003 * @server - pointer to nfs_server struct
1004 * @inode - pointer to inode struct
1005 *
1006 * Updates inode attribute information by retrieving the data from the server.
1007 */
1008 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1009 {
1010 if (!nfs_need_revalidate_inode(inode))
1011 return NFS_STALE(inode) ? -ESTALE : 0;
1012 return __nfs_revalidate_inode(server, inode);
1013 }
1014 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1015
1016 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1017 {
1018 if (!(NFS_I(inode)->cache_validity &
1019 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1020 && !nfs_attribute_cache_expired(inode))
1021 return NFS_STALE(inode) ? -ESTALE : 0;
1022 return -ECHILD;
1023 }
1024
1025 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1026 {
1027 struct nfs_inode *nfsi = NFS_I(inode);
1028 int ret;
1029
1030 if (mapping->nrpages != 0) {
1031 if (S_ISREG(inode->i_mode)) {
1032 ret = nfs_sync_mapping(mapping);
1033 if (ret < 0)
1034 return ret;
1035 }
1036 ret = invalidate_inode_pages2(mapping);
1037 if (ret < 0)
1038 return ret;
1039 }
1040 if (S_ISDIR(inode->i_mode)) {
1041 spin_lock(&inode->i_lock);
1042 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1043 spin_unlock(&inode->i_lock);
1044 }
1045 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1046 nfs_fscache_wait_on_invalidate(inode);
1047
1048 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1049 inode->i_sb->s_id,
1050 (unsigned long long)NFS_FILEID(inode));
1051 return 0;
1052 }
1053
1054 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1055 {
1056 if (nfs_have_delegated_attributes(inode))
1057 return false;
1058 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1059 || nfs_attribute_timeout(inode)
1060 || NFS_STALE(inode);
1061 }
1062
1063 /**
1064 * nfs_revalidate_mapping - Revalidate the pagecache
1065 * @inode - pointer to host inode
1066 * @mapping - pointer to mapping
1067 */
1068 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1069 {
1070 struct nfs_inode *nfsi = NFS_I(inode);
1071 unsigned long *bitlock = &nfsi->flags;
1072 int ret = 0;
1073
1074 /* swapfiles are not supposed to be shared. */
1075 if (IS_SWAPFILE(inode))
1076 goto out;
1077
1078 if (nfs_mapping_need_revalidate_inode(inode)) {
1079 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1080 if (ret < 0)
1081 goto out;
1082 }
1083
1084 /*
1085 * We must clear NFS_INO_INVALID_DATA first to ensure that
1086 * invalidations that come in while we're shooting down the mappings
1087 * are respected. But, that leaves a race window where one revalidator
1088 * can clear the flag, and then another checks it before the mapping
1089 * gets invalidated. Fix that by serializing access to this part of
1090 * the function.
1091 *
1092 * At the same time, we need to allow other tasks to see whether we
1093 * might be in the middle of invalidating the pages, so we only set
1094 * the bit lock here if it looks like we're going to be doing that.
1095 */
1096 for (;;) {
1097 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1098 nfs_wait_bit_killable, TASK_KILLABLE);
1099 if (ret)
1100 goto out;
1101 spin_lock(&inode->i_lock);
1102 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1103 spin_unlock(&inode->i_lock);
1104 continue;
1105 }
1106 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1107 break;
1108 spin_unlock(&inode->i_lock);
1109 goto out;
1110 }
1111
1112 set_bit(NFS_INO_INVALIDATING, bitlock);
1113 smp_wmb();
1114 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1115 spin_unlock(&inode->i_lock);
1116 trace_nfs_invalidate_mapping_enter(inode);
1117 ret = nfs_invalidate_mapping(inode, mapping);
1118 trace_nfs_invalidate_mapping_exit(inode, ret);
1119
1120 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1121 smp_mb__after_atomic();
1122 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1123 out:
1124 return ret;
1125 }
1126
1127 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1128 {
1129 struct nfs_inode *nfsi = NFS_I(inode);
1130 unsigned long ret = 0;
1131
1132 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1133 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1134 && inode->i_version == fattr->pre_change_attr) {
1135 inode->i_version = fattr->change_attr;
1136 if (S_ISDIR(inode->i_mode))
1137 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1138 ret |= NFS_INO_INVALID_ATTR;
1139 }
1140 /* If we have atomic WCC data, we may update some attributes */
1141 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1142 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1143 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1144 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1145 ret |= NFS_INO_INVALID_ATTR;
1146 }
1147
1148 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1149 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1150 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1151 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1152 if (S_ISDIR(inode->i_mode))
1153 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1154 ret |= NFS_INO_INVALID_ATTR;
1155 }
1156 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1157 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1158 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1159 && nfsi->nrequests == 0) {
1160 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1161 ret |= NFS_INO_INVALID_ATTR;
1162 }
1163
1164 return ret;
1165 }
1166
1167 /**
1168 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1169 * @inode - pointer to inode
1170 * @fattr - updated attributes
1171 *
1172 * Verifies the attribute cache. If we have just changed the attributes,
1173 * so that fattr carries weak cache consistency data, then it may
1174 * also update the ctime/mtime/change_attribute.
1175 */
1176 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1177 {
1178 struct nfs_inode *nfsi = NFS_I(inode);
1179 loff_t cur_size, new_isize;
1180 unsigned long invalid = 0;
1181
1182
1183 if (nfs_have_delegated_attributes(inode))
1184 return 0;
1185 /* Has the inode gone and changed behind our back? */
1186 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1187 return -EIO;
1188 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1189 return -EIO;
1190
1191 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1192 inode->i_version != fattr->change_attr)
1193 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1194
1195 /* Verify a few of the more important attributes */
1196 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1197 invalid |= NFS_INO_INVALID_ATTR;
1198
1199 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1200 cur_size = i_size_read(inode);
1201 new_isize = nfs_size_to_loff_t(fattr->size);
1202 if (cur_size != new_isize && nfsi->nrequests == 0)
1203 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1204 }
1205
1206 /* Have any file permissions changed? */
1207 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1208 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1209 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1210 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1211 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1212 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1213
1214 /* Has the link count changed? */
1215 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1216 invalid |= NFS_INO_INVALID_ATTR;
1217
1218 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1219 invalid |= NFS_INO_INVALID_ATIME;
1220
1221 if (invalid != 0)
1222 nfs_set_cache_invalid(inode, invalid);
1223
1224 nfsi->read_cache_jiffies = fattr->time_start;
1225 return 0;
1226 }
1227
1228 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1229 {
1230 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1231 return 0;
1232 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1233 }
1234
1235 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1236 {
1237 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1238 return 0;
1239 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1240 }
1241
1242 static atomic_long_t nfs_attr_generation_counter;
1243
1244 static unsigned long nfs_read_attr_generation_counter(void)
1245 {
1246 return atomic_long_read(&nfs_attr_generation_counter);
1247 }
1248
1249 unsigned long nfs_inc_attr_generation_counter(void)
1250 {
1251 return atomic_long_inc_return(&nfs_attr_generation_counter);
1252 }
1253
1254 void nfs_fattr_init(struct nfs_fattr *fattr)
1255 {
1256 fattr->valid = 0;
1257 fattr->time_start = jiffies;
1258 fattr->gencount = nfs_inc_attr_generation_counter();
1259 fattr->owner_name = NULL;
1260 fattr->group_name = NULL;
1261 }
1262 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1263
1264 struct nfs_fattr *nfs_alloc_fattr(void)
1265 {
1266 struct nfs_fattr *fattr;
1267
1268 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1269 if (fattr != NULL)
1270 nfs_fattr_init(fattr);
1271 return fattr;
1272 }
1273 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1274
1275 struct nfs_fh *nfs_alloc_fhandle(void)
1276 {
1277 struct nfs_fh *fh;
1278
1279 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1280 if (fh != NULL)
1281 fh->size = 0;
1282 return fh;
1283 }
1284 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1285
1286 #ifdef NFS_DEBUG
1287 /*
1288 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1289 * in the same way that wireshark does
1290 *
1291 * @fh: file handle
1292 *
1293 * For debugging only.
1294 */
1295 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1296 {
1297 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1298 * not on the result */
1299 return nfs_fhandle_hash(fh);
1300 }
1301 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1302
1303 /*
1304 * _nfs_display_fhandle - display an NFS file handle on the console
1305 *
1306 * @fh: file handle to display
1307 * @caption: display caption
1308 *
1309 * For debugging only.
1310 */
1311 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1312 {
1313 unsigned short i;
1314
1315 if (fh == NULL || fh->size == 0) {
1316 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1317 return;
1318 }
1319
1320 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1321 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1322 for (i = 0; i < fh->size; i += 16) {
1323 __be32 *pos = (__be32 *)&fh->data[i];
1324
1325 switch ((fh->size - i - 1) >> 2) {
1326 case 0:
1327 printk(KERN_DEFAULT " %08x\n",
1328 be32_to_cpup(pos));
1329 break;
1330 case 1:
1331 printk(KERN_DEFAULT " %08x %08x\n",
1332 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1333 break;
1334 case 2:
1335 printk(KERN_DEFAULT " %08x %08x %08x\n",
1336 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1337 be32_to_cpup(pos + 2));
1338 break;
1339 default:
1340 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1341 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1342 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1343 }
1344 }
1345 }
1346 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1347 #endif
1348
1349 /**
1350 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1351 * @inode - pointer to inode
1352 * @fattr - attributes
1353 *
1354 * Attempt to divine whether or not an RPC call reply carrying stale
1355 * attributes got scheduled after another call carrying updated ones.
1356 *
1357 * To do so, the function first assumes that a more recent ctime means
1358 * that the attributes in fattr are newer, however it also attempt to
1359 * catch the case where ctime either didn't change, or went backwards
1360 * (if someone reset the clock on the server) by looking at whether
1361 * or not this RPC call was started after the inode was last updated.
1362 * Note also the check for wraparound of 'attr_gencount'
1363 *
1364 * The function returns 'true' if it thinks the attributes in 'fattr' are
1365 * more recent than the ones cached in the inode.
1366 *
1367 */
1368 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1369 {
1370 const struct nfs_inode *nfsi = NFS_I(inode);
1371
1372 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1373 nfs_ctime_need_update(inode, fattr) ||
1374 nfs_size_need_update(inode, fattr) ||
1375 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1376 }
1377
1378 /*
1379 * Don't trust the change_attribute, mtime, ctime or size if
1380 * a pnfs LAYOUTCOMMIT is outstanding
1381 */
1382 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1383 struct nfs_fattr *fattr)
1384 {
1385 if (pnfs_layoutcommit_outstanding(inode))
1386 fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1387 NFS_ATTR_FATTR_MTIME |
1388 NFS_ATTR_FATTR_CTIME |
1389 NFS_ATTR_FATTR_SIZE);
1390 }
1391
1392 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1393 {
1394 int ret;
1395
1396 trace_nfs_refresh_inode_enter(inode);
1397
1398 nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1399
1400 if (nfs_inode_attrs_need_update(inode, fattr))
1401 ret = nfs_update_inode(inode, fattr);
1402 else
1403 ret = nfs_check_inode_attributes(inode, fattr);
1404
1405 trace_nfs_refresh_inode_exit(inode, ret);
1406 return ret;
1407 }
1408
1409 /**
1410 * nfs_refresh_inode - try to update the inode attribute cache
1411 * @inode - pointer to inode
1412 * @fattr - updated attributes
1413 *
1414 * Check that an RPC call that returned attributes has not overlapped with
1415 * other recent updates of the inode metadata, then decide whether it is
1416 * safe to do a full update of the inode attributes, or whether just to
1417 * call nfs_check_inode_attributes.
1418 */
1419 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1420 {
1421 int status;
1422
1423 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1424 return 0;
1425 spin_lock(&inode->i_lock);
1426 status = nfs_refresh_inode_locked(inode, fattr);
1427 spin_unlock(&inode->i_lock);
1428
1429 return status;
1430 }
1431 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1432
1433 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1434 {
1435 unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1436
1437 if (S_ISDIR(inode->i_mode))
1438 invalid |= NFS_INO_INVALID_DATA;
1439 nfs_set_cache_invalid(inode, invalid);
1440 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1441 return 0;
1442 return nfs_refresh_inode_locked(inode, fattr);
1443 }
1444
1445 /**
1446 * nfs_post_op_update_inode - try to update the inode attribute cache
1447 * @inode - pointer to inode
1448 * @fattr - updated attributes
1449 *
1450 * After an operation that has changed the inode metadata, mark the
1451 * attribute cache as being invalid, then try to update it.
1452 *
1453 * NB: if the server didn't return any post op attributes, this
1454 * function will force the retrieval of attributes before the next
1455 * NFS request. Thus it should be used only for operations that
1456 * are expected to change one or more attributes, to avoid
1457 * unnecessary NFS requests and trips through nfs_update_inode().
1458 */
1459 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1460 {
1461 int status;
1462
1463 spin_lock(&inode->i_lock);
1464 status = nfs_post_op_update_inode_locked(inode, fattr);
1465 spin_unlock(&inode->i_lock);
1466
1467 return status;
1468 }
1469 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1470
1471 /**
1472 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1473 * @inode - pointer to inode
1474 * @fattr - updated attributes
1475 *
1476 * After an operation that has changed the inode metadata, mark the
1477 * attribute cache as being invalid, then try to update it. Fake up
1478 * weak cache consistency data, if none exist.
1479 *
1480 * This function is mainly designed to be used by the ->write_done() functions.
1481 */
1482 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1483 {
1484 int status;
1485
1486 spin_lock(&inode->i_lock);
1487 /* Don't do a WCC update if these attributes are already stale */
1488 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1489 !nfs_inode_attrs_need_update(inode, fattr)) {
1490 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1491 | NFS_ATTR_FATTR_PRESIZE
1492 | NFS_ATTR_FATTR_PREMTIME
1493 | NFS_ATTR_FATTR_PRECTIME);
1494 goto out_noforce;
1495 }
1496 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1497 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1498 fattr->pre_change_attr = inode->i_version;
1499 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1500 }
1501 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1502 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1503 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1504 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1505 }
1506 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1507 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1508 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1509 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1510 }
1511 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1512 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1513 fattr->pre_size = i_size_read(inode);
1514 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1515 }
1516 out_noforce:
1517 status = nfs_post_op_update_inode_locked(inode, fattr);
1518 spin_unlock(&inode->i_lock);
1519 return status;
1520 }
1521 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1522
1523 /*
1524 * Many nfs protocol calls return the new file attributes after
1525 * an operation. Here we update the inode to reflect the state
1526 * of the server's inode.
1527 *
1528 * This is a bit tricky because we have to make sure all dirty pages
1529 * have been sent off to the server before calling invalidate_inode_pages.
1530 * To make sure no other process adds more write requests while we try
1531 * our best to flush them, we make them sleep during the attribute refresh.
1532 *
1533 * A very similar scenario holds for the dir cache.
1534 */
1535 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1536 {
1537 struct nfs_server *server;
1538 struct nfs_inode *nfsi = NFS_I(inode);
1539 loff_t cur_isize, new_isize;
1540 unsigned long invalid = 0;
1541 unsigned long now = jiffies;
1542 unsigned long save_cache_validity;
1543
1544 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1545 __func__, inode->i_sb->s_id, inode->i_ino,
1546 nfs_display_fhandle_hash(NFS_FH(inode)),
1547 atomic_read(&inode->i_count), fattr->valid);
1548
1549 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1550 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1551 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1552 NFS_SERVER(inode)->nfs_client->cl_hostname,
1553 inode->i_sb->s_id, (long long)nfsi->fileid,
1554 (long long)fattr->fileid);
1555 goto out_err;
1556 }
1557
1558 /*
1559 * Make sure the inode's type hasn't changed.
1560 */
1561 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1562 /*
1563 * Big trouble! The inode has become a different object.
1564 */
1565 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1566 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1567 goto out_err;
1568 }
1569
1570 server = NFS_SERVER(inode);
1571 /* Update the fsid? */
1572 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1573 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1574 !IS_AUTOMOUNT(inode))
1575 server->fsid = fattr->fsid;
1576
1577 /*
1578 * Update the read time so we don't revalidate too often.
1579 */
1580 nfsi->read_cache_jiffies = fattr->time_start;
1581
1582 save_cache_validity = nfsi->cache_validity;
1583 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1584 | NFS_INO_INVALID_ATIME
1585 | NFS_INO_REVAL_FORCED
1586 | NFS_INO_REVAL_PAGECACHE);
1587
1588 /* Do atomic weak cache consistency updates */
1589 invalid |= nfs_wcc_update_inode(inode, fattr);
1590
1591 /* More cache consistency checks */
1592 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1593 if (inode->i_version != fattr->change_attr) {
1594 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1595 inode->i_sb->s_id, inode->i_ino);
1596 invalid |= NFS_INO_INVALID_ATTR
1597 | NFS_INO_INVALID_DATA
1598 | NFS_INO_INVALID_ACCESS
1599 | NFS_INO_INVALID_ACL
1600 | NFS_INO_REVAL_PAGECACHE;
1601 if (S_ISDIR(inode->i_mode))
1602 nfs_force_lookup_revalidate(inode);
1603 inode->i_version = fattr->change_attr;
1604 }
1605 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1606 nfsi->cache_validity |= save_cache_validity;
1607
1608 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1609 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1610 } else if (server->caps & NFS_CAP_MTIME)
1611 nfsi->cache_validity |= save_cache_validity &
1612 (NFS_INO_INVALID_ATTR
1613 | NFS_INO_REVAL_FORCED);
1614
1615 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1616 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1617 } else if (server->caps & NFS_CAP_CTIME)
1618 nfsi->cache_validity |= save_cache_validity &
1619 (NFS_INO_INVALID_ATTR
1620 | NFS_INO_REVAL_FORCED);
1621
1622 /* Check if our cached file size is stale */
1623 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1624 new_isize = nfs_size_to_loff_t(fattr->size);
1625 cur_isize = i_size_read(inode);
1626 if (new_isize != cur_isize) {
1627 /* Do we perhaps have any outstanding writes, or has
1628 * the file grown beyond our last write? */
1629 if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1630 i_size_write(inode, new_isize);
1631 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1632 invalid &= ~NFS_INO_REVAL_PAGECACHE;
1633 }
1634 dprintk("NFS: isize change on server for file %s/%ld "
1635 "(%Ld to %Ld)\n",
1636 inode->i_sb->s_id,
1637 inode->i_ino,
1638 (long long)cur_isize,
1639 (long long)new_isize);
1640 }
1641 } else
1642 nfsi->cache_validity |= save_cache_validity &
1643 (NFS_INO_INVALID_ATTR
1644 | NFS_INO_REVAL_PAGECACHE
1645 | NFS_INO_REVAL_FORCED);
1646
1647
1648 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1649 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1650 else if (server->caps & NFS_CAP_ATIME)
1651 nfsi->cache_validity |= save_cache_validity &
1652 (NFS_INO_INVALID_ATIME
1653 | NFS_INO_REVAL_FORCED);
1654
1655 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1656 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1657 umode_t newmode = inode->i_mode & S_IFMT;
1658 newmode |= fattr->mode & S_IALLUGO;
1659 inode->i_mode = newmode;
1660 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1661 }
1662 } else if (server->caps & NFS_CAP_MODE)
1663 nfsi->cache_validity |= save_cache_validity &
1664 (NFS_INO_INVALID_ATTR
1665 | NFS_INO_INVALID_ACCESS
1666 | NFS_INO_INVALID_ACL
1667 | NFS_INO_REVAL_FORCED);
1668
1669 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1670 if (!uid_eq(inode->i_uid, fattr->uid)) {
1671 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1672 inode->i_uid = fattr->uid;
1673 }
1674 } else if (server->caps & NFS_CAP_OWNER)
1675 nfsi->cache_validity |= save_cache_validity &
1676 (NFS_INO_INVALID_ATTR
1677 | NFS_INO_INVALID_ACCESS
1678 | NFS_INO_INVALID_ACL
1679 | NFS_INO_REVAL_FORCED);
1680
1681 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1682 if (!gid_eq(inode->i_gid, fattr->gid)) {
1683 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1684 inode->i_gid = fattr->gid;
1685 }
1686 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1687 nfsi->cache_validity |= save_cache_validity &
1688 (NFS_INO_INVALID_ATTR
1689 | NFS_INO_INVALID_ACCESS
1690 | NFS_INO_INVALID_ACL
1691 | NFS_INO_REVAL_FORCED);
1692
1693 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1694 if (inode->i_nlink != fattr->nlink) {
1695 invalid |= NFS_INO_INVALID_ATTR;
1696 if (S_ISDIR(inode->i_mode))
1697 invalid |= NFS_INO_INVALID_DATA;
1698 set_nlink(inode, fattr->nlink);
1699 }
1700 } else if (server->caps & NFS_CAP_NLINK)
1701 nfsi->cache_validity |= save_cache_validity &
1702 (NFS_INO_INVALID_ATTR
1703 | NFS_INO_REVAL_FORCED);
1704
1705 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1706 /*
1707 * report the blocks in 512byte units
1708 */
1709 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1710 }
1711 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1712 inode->i_blocks = fattr->du.nfs2.blocks;
1713
1714 /* Update attrtimeo value if we're out of the unstable period */
1715 if (invalid & NFS_INO_INVALID_ATTR) {
1716 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1717 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1718 nfsi->attrtimeo_timestamp = now;
1719 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1720 } else {
1721 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1722 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1723 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1724 nfsi->attrtimeo_timestamp = now;
1725 }
1726 }
1727 invalid &= ~NFS_INO_INVALID_ATTR;
1728 /* Don't invalidate the data if we were to blame */
1729 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1730 || S_ISLNK(inode->i_mode)))
1731 invalid &= ~NFS_INO_INVALID_DATA;
1732 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1733 (save_cache_validity & NFS_INO_REVAL_FORCED))
1734 nfs_set_cache_invalid(inode, invalid);
1735
1736 return 0;
1737 out_err:
1738 /*
1739 * No need to worry about unhashing the dentry, as the
1740 * lookup validation will know that the inode is bad.
1741 * (But we fall through to invalidate the caches.)
1742 */
1743 nfs_invalidate_inode(inode);
1744 return -ESTALE;
1745 }
1746
1747 struct inode *nfs_alloc_inode(struct super_block *sb)
1748 {
1749 struct nfs_inode *nfsi;
1750 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1751 if (!nfsi)
1752 return NULL;
1753 nfsi->flags = 0UL;
1754 nfsi->cache_validity = 0UL;
1755 #if IS_ENABLED(CONFIG_NFS_V4)
1756 nfsi->nfs4_acl = NULL;
1757 #endif /* CONFIG_NFS_V4 */
1758 return &nfsi->vfs_inode;
1759 }
1760 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1761
1762 static void nfs_i_callback(struct rcu_head *head)
1763 {
1764 struct inode *inode = container_of(head, struct inode, i_rcu);
1765 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1766 }
1767
1768 void nfs_destroy_inode(struct inode *inode)
1769 {
1770 call_rcu(&inode->i_rcu, nfs_i_callback);
1771 }
1772 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1773
1774 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1775 {
1776 #if IS_ENABLED(CONFIG_NFS_V4)
1777 INIT_LIST_HEAD(&nfsi->open_states);
1778 nfsi->delegation = NULL;
1779 nfsi->delegation_state = 0;
1780 init_rwsem(&nfsi->rwsem);
1781 nfsi->layout = NULL;
1782 #endif
1783 }
1784
1785 static void init_once(void *foo)
1786 {
1787 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1788
1789 inode_init_once(&nfsi->vfs_inode);
1790 INIT_LIST_HEAD(&nfsi->open_files);
1791 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1792 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1793 INIT_LIST_HEAD(&nfsi->commit_info.list);
1794 nfsi->nrequests = 0;
1795 nfsi->commit_info.ncommit = 0;
1796 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1797 atomic_set(&nfsi->silly_count, 1);
1798 INIT_HLIST_HEAD(&nfsi->silly_list);
1799 init_waitqueue_head(&nfsi->waitqueue);
1800 nfs4_init_once(nfsi);
1801 }
1802
1803 static int __init nfs_init_inodecache(void)
1804 {
1805 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1806 sizeof(struct nfs_inode),
1807 0, (SLAB_RECLAIM_ACCOUNT|
1808 SLAB_MEM_SPREAD),
1809 init_once);
1810 if (nfs_inode_cachep == NULL)
1811 return -ENOMEM;
1812
1813 return 0;
1814 }
1815
1816 static void nfs_destroy_inodecache(void)
1817 {
1818 /*
1819 * Make sure all delayed rcu free inodes are flushed before we
1820 * destroy cache.
1821 */
1822 rcu_barrier();
1823 kmem_cache_destroy(nfs_inode_cachep);
1824 }
1825
1826 struct workqueue_struct *nfsiod_workqueue;
1827 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1828
1829 /*
1830 * start up the nfsiod workqueue
1831 */
1832 static int nfsiod_start(void)
1833 {
1834 struct workqueue_struct *wq;
1835 dprintk("RPC: creating workqueue nfsiod\n");
1836 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1837 if (wq == NULL)
1838 return -ENOMEM;
1839 nfsiod_workqueue = wq;
1840 return 0;
1841 }
1842
1843 /*
1844 * Destroy the nfsiod workqueue
1845 */
1846 static void nfsiod_stop(void)
1847 {
1848 struct workqueue_struct *wq;
1849
1850 wq = nfsiod_workqueue;
1851 if (wq == NULL)
1852 return;
1853 nfsiod_workqueue = NULL;
1854 destroy_workqueue(wq);
1855 }
1856
1857 int nfs_net_id;
1858 EXPORT_SYMBOL_GPL(nfs_net_id);
1859
1860 static int nfs_net_init(struct net *net)
1861 {
1862 nfs_clients_init(net);
1863 return nfs_fs_proc_net_init(net);
1864 }
1865
1866 static void nfs_net_exit(struct net *net)
1867 {
1868 nfs_fs_proc_net_exit(net);
1869 nfs_cleanup_cb_ident_idr(net);
1870 }
1871
1872 static struct pernet_operations nfs_net_ops = {
1873 .init = nfs_net_init,
1874 .exit = nfs_net_exit,
1875 .id = &nfs_net_id,
1876 .size = sizeof(struct nfs_net),
1877 };
1878
1879 /*
1880 * Initialize NFS
1881 */
1882 static int __init init_nfs_fs(void)
1883 {
1884 int err;
1885
1886 err = register_pernet_subsys(&nfs_net_ops);
1887 if (err < 0)
1888 goto out9;
1889
1890 err = nfs_fscache_register();
1891 if (err < 0)
1892 goto out8;
1893
1894 err = nfsiod_start();
1895 if (err)
1896 goto out7;
1897
1898 err = nfs_fs_proc_init();
1899 if (err)
1900 goto out6;
1901
1902 err = nfs_init_nfspagecache();
1903 if (err)
1904 goto out5;
1905
1906 err = nfs_init_inodecache();
1907 if (err)
1908 goto out4;
1909
1910 err = nfs_init_readpagecache();
1911 if (err)
1912 goto out3;
1913
1914 err = nfs_init_writepagecache();
1915 if (err)
1916 goto out2;
1917
1918 err = nfs_init_directcache();
1919 if (err)
1920 goto out1;
1921
1922 #ifdef CONFIG_PROC_FS
1923 rpc_proc_register(&init_net, &nfs_rpcstat);
1924 #endif
1925 if ((err = register_nfs_fs()) != 0)
1926 goto out0;
1927
1928 return 0;
1929 out0:
1930 #ifdef CONFIG_PROC_FS
1931 rpc_proc_unregister(&init_net, "nfs");
1932 #endif
1933 nfs_destroy_directcache();
1934 out1:
1935 nfs_destroy_writepagecache();
1936 out2:
1937 nfs_destroy_readpagecache();
1938 out3:
1939 nfs_destroy_inodecache();
1940 out4:
1941 nfs_destroy_nfspagecache();
1942 out5:
1943 nfs_fs_proc_exit();
1944 out6:
1945 nfsiod_stop();
1946 out7:
1947 nfs_fscache_unregister();
1948 out8:
1949 unregister_pernet_subsys(&nfs_net_ops);
1950 out9:
1951 return err;
1952 }
1953
1954 static void __exit exit_nfs_fs(void)
1955 {
1956 nfs_destroy_directcache();
1957 nfs_destroy_writepagecache();
1958 nfs_destroy_readpagecache();
1959 nfs_destroy_inodecache();
1960 nfs_destroy_nfspagecache();
1961 nfs_fscache_unregister();
1962 unregister_pernet_subsys(&nfs_net_ops);
1963 #ifdef CONFIG_PROC_FS
1964 rpc_proc_unregister(&init_net, "nfs");
1965 #endif
1966 unregister_nfs_fs();
1967 nfs_fs_proc_exit();
1968 nfsiod_stop();
1969 }
1970
1971 /* Not quite true; I just maintain it */
1972 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1973 MODULE_LICENSE("GPL");
1974 module_param(enable_ino64, bool, 0644);
1975
1976 module_init(init_nfs_fs)
1977 module_exit(exit_nfs_fs)