xref: /linux/fs/nfs/inode.c (revision 5a0e3ad6af8660be21ca98a971cd00f331318c05)
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/nfs_idmap.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 
41 #include <asm/system.h>
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 "dns_resolve.h"
51 
52 #define NFSDBG_FACILITY		NFSDBG_VFS
53 
54 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
55 
56 /* Default is to see 64-bit inode numbers */
57 static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
58 
59 static void nfs_invalidate_inode(struct inode *);
60 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
61 
62 static struct kmem_cache * nfs_inode_cachep;
63 
64 static inline unsigned long
65 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
66 {
67 	return nfs_fileid_to_ino_t(fattr->fileid);
68 }
69 
70 /**
71  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
72  * @word: long word containing the bit lock
73  */
74 int nfs_wait_bit_killable(void *word)
75 {
76 	if (fatal_signal_pending(current))
77 		return -ERESTARTSYS;
78 	schedule();
79 	return 0;
80 }
81 
82 /**
83  * nfs_compat_user_ino64 - returns the user-visible inode number
84  * @fileid: 64-bit fileid
85  *
86  * This function returns a 32-bit inode number if the boot parameter
87  * nfs.enable_ino64 is zero.
88  */
89 u64 nfs_compat_user_ino64(u64 fileid)
90 {
91 	int ino;
92 
93 	if (enable_ino64)
94 		return fileid;
95 	ino = fileid;
96 	if (sizeof(ino) < sizeof(fileid))
97 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
98 	return ino;
99 }
100 
101 void nfs_clear_inode(struct inode *inode)
102 {
103 	/*
104 	 * The following should never happen...
105 	 */
106 	BUG_ON(nfs_have_writebacks(inode));
107 	BUG_ON(!list_empty(&NFS_I(inode)->open_files));
108 	nfs_zap_acl_cache(inode);
109 	nfs_access_zap_cache(inode);
110 	nfs_fscache_release_inode_cookie(inode);
111 }
112 
113 /**
114  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
115  */
116 int nfs_sync_mapping(struct address_space *mapping)
117 {
118 	int ret = 0;
119 
120 	if (mapping->nrpages != 0) {
121 		unmap_mapping_range(mapping, 0, 0, 0);
122 		ret = nfs_wb_all(mapping->host);
123 	}
124 	return ret;
125 }
126 
127 /*
128  * Invalidate the local caches
129  */
130 static void nfs_zap_caches_locked(struct inode *inode)
131 {
132 	struct nfs_inode *nfsi = NFS_I(inode);
133 	int mode = inode->i_mode;
134 
135 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
136 
137 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
138 	nfsi->attrtimeo_timestamp = jiffies;
139 
140 	memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
141 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
142 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
143 	else
144 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
145 }
146 
147 void nfs_zap_caches(struct inode *inode)
148 {
149 	spin_lock(&inode->i_lock);
150 	nfs_zap_caches_locked(inode);
151 	spin_unlock(&inode->i_lock);
152 }
153 
154 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
155 {
156 	if (mapping->nrpages != 0) {
157 		spin_lock(&inode->i_lock);
158 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
159 		spin_unlock(&inode->i_lock);
160 	}
161 }
162 
163 void nfs_zap_acl_cache(struct inode *inode)
164 {
165 	void (*clear_acl_cache)(struct inode *);
166 
167 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
168 	if (clear_acl_cache != NULL)
169 		clear_acl_cache(inode);
170 	spin_lock(&inode->i_lock);
171 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
172 	spin_unlock(&inode->i_lock);
173 }
174 
175 void nfs_invalidate_atime(struct inode *inode)
176 {
177 	spin_lock(&inode->i_lock);
178 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
179 	spin_unlock(&inode->i_lock);
180 }
181 
182 /*
183  * Invalidate, but do not unhash, the inode.
184  * NB: must be called with inode->i_lock held!
185  */
186 static void nfs_invalidate_inode(struct inode *inode)
187 {
188 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
189 	nfs_zap_caches_locked(inode);
190 }
191 
192 struct nfs_find_desc {
193 	struct nfs_fh		*fh;
194 	struct nfs_fattr	*fattr;
195 };
196 
197 /*
198  * In NFSv3 we can have 64bit inode numbers. In order to support
199  * this, and re-exported directories (also seen in NFSv2)
200  * we are forced to allow 2 different inodes to have the same
201  * i_ino.
202  */
203 static int
204 nfs_find_actor(struct inode *inode, void *opaque)
205 {
206 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
207 	struct nfs_fh		*fh = desc->fh;
208 	struct nfs_fattr	*fattr = desc->fattr;
209 
210 	if (NFS_FILEID(inode) != fattr->fileid)
211 		return 0;
212 	if (nfs_compare_fh(NFS_FH(inode), fh))
213 		return 0;
214 	if (is_bad_inode(inode) || NFS_STALE(inode))
215 		return 0;
216 	return 1;
217 }
218 
219 static int
220 nfs_init_locked(struct inode *inode, void *opaque)
221 {
222 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
223 	struct nfs_fattr	*fattr = desc->fattr;
224 
225 	set_nfs_fileid(inode, fattr->fileid);
226 	nfs_copy_fh(NFS_FH(inode), desc->fh);
227 	return 0;
228 }
229 
230 /* Don't use READDIRPLUS on directories that we believe are too large */
231 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
232 
233 /*
234  * This is our front-end to iget that looks up inodes by file handle
235  * instead of inode number.
236  */
237 struct inode *
238 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
239 {
240 	struct nfs_find_desc desc = {
241 		.fh	= fh,
242 		.fattr	= fattr
243 	};
244 	struct inode *inode = ERR_PTR(-ENOENT);
245 	unsigned long hash;
246 
247 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
248 		goto out_no_inode;
249 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
250 		goto out_no_inode;
251 
252 	hash = nfs_fattr_to_ino_t(fattr);
253 
254 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
255 	if (inode == NULL) {
256 		inode = ERR_PTR(-ENOMEM);
257 		goto out_no_inode;
258 	}
259 
260 	if (inode->i_state & I_NEW) {
261 		struct nfs_inode *nfsi = NFS_I(inode);
262 		unsigned long now = jiffies;
263 
264 		/* We set i_ino for the few things that still rely on it,
265 		 * such as stat(2) */
266 		inode->i_ino = hash;
267 
268 		/* We can't support update_atime(), since the server will reset it */
269 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
270 		inode->i_mode = fattr->mode;
271 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
272 				&& nfs_server_capable(inode, NFS_CAP_MODE))
273 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
274 				| NFS_INO_INVALID_ACCESS
275 				| NFS_INO_INVALID_ACL;
276 		/* Why so? Because we want revalidate for devices/FIFOs, and
277 		 * that's precisely what we have in nfs_file_inode_operations.
278 		 */
279 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
280 		if (S_ISREG(inode->i_mode)) {
281 			inode->i_fop = &nfs_file_operations;
282 			inode->i_data.a_ops = &nfs_file_aops;
283 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
284 		} else if (S_ISDIR(inode->i_mode)) {
285 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
286 			inode->i_fop = &nfs_dir_operations;
287 			if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
288 			    && fattr->size <= NFS_LIMIT_READDIRPLUS)
289 				set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
290 			/* Deal with crossing mountpoints */
291 			if ((fattr->valid & NFS_ATTR_FATTR_FSID)
292 					&& !nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
293 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
294 					inode->i_op = &nfs_referral_inode_operations;
295 				else
296 					inode->i_op = &nfs_mountpoint_inode_operations;
297 				inode->i_fop = NULL;
298 				set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
299 			}
300 		} else if (S_ISLNK(inode->i_mode))
301 			inode->i_op = &nfs_symlink_inode_operations;
302 		else
303 			init_special_inode(inode, inode->i_mode, fattr->rdev);
304 
305 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
306 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
307 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
308 		nfsi->change_attr = 0;
309 		inode->i_size = 0;
310 		inode->i_nlink = 0;
311 		inode->i_uid = -2;
312 		inode->i_gid = -2;
313 		inode->i_blocks = 0;
314 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
315 
316 		nfsi->read_cache_jiffies = fattr->time_start;
317 		nfsi->attr_gencount = fattr->gencount;
318 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
319 			inode->i_atime = fattr->atime;
320 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
321 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
322 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
323 			inode->i_mtime = fattr->mtime;
324 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
325 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
326 				| NFS_INO_INVALID_DATA;
327 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
328 			inode->i_ctime = fattr->ctime;
329 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
330 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
331 				| NFS_INO_INVALID_ACCESS
332 				| NFS_INO_INVALID_ACL;
333 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
334 			nfsi->change_attr = fattr->change_attr;
335 		else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
336 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
337 				| NFS_INO_INVALID_DATA;
338 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
339 			inode->i_size = nfs_size_to_loff_t(fattr->size);
340 		else
341 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
342 				| NFS_INO_INVALID_DATA
343 				| NFS_INO_REVAL_PAGECACHE;
344 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
345 			inode->i_nlink = fattr->nlink;
346 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
347 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
348 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
349 			inode->i_uid = fattr->uid;
350 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
351 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
352 				| NFS_INO_INVALID_ACCESS
353 				| NFS_INO_INVALID_ACL;
354 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
355 			inode->i_gid = fattr->gid;
356 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
357 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
358 				| NFS_INO_INVALID_ACCESS
359 				| NFS_INO_INVALID_ACL;
360 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
361 			inode->i_blocks = fattr->du.nfs2.blocks;
362 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
363 			/*
364 			 * report the blocks in 512byte units
365 			 */
366 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
367 		}
368 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
369 		nfsi->attrtimeo_timestamp = now;
370 		nfsi->access_cache = RB_ROOT;
371 
372 		nfs_fscache_init_inode_cookie(inode);
373 
374 		unlock_new_inode(inode);
375 	} else
376 		nfs_refresh_inode(inode, fattr);
377 	dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
378 		inode->i_sb->s_id,
379 		(long long)NFS_FILEID(inode),
380 		atomic_read(&inode->i_count));
381 
382 out:
383 	return inode;
384 
385 out_no_inode:
386 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
387 	goto out;
388 }
389 
390 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
391 
392 int
393 nfs_setattr(struct dentry *dentry, struct iattr *attr)
394 {
395 	struct inode *inode = dentry->d_inode;
396 	struct nfs_fattr fattr;
397 	int error;
398 
399 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
400 
401 	/* skip mode change if it's just for clearing setuid/setgid */
402 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
403 		attr->ia_valid &= ~ATTR_MODE;
404 
405 	if (attr->ia_valid & ATTR_SIZE) {
406 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
407 			attr->ia_valid &= ~ATTR_SIZE;
408 	}
409 
410 	/* Optimization: if the end result is no change, don't RPC */
411 	attr->ia_valid &= NFS_VALID_ATTRS;
412 	if ((attr->ia_valid & ~ATTR_FILE) == 0)
413 		return 0;
414 
415 	/* Write all dirty data */
416 	if (S_ISREG(inode->i_mode)) {
417 		filemap_write_and_wait(inode->i_mapping);
418 		nfs_wb_all(inode);
419 	}
420 	/*
421 	 * Return any delegations if we're going to change ACLs
422 	 */
423 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
424 		nfs_inode_return_delegation(inode);
425 	error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
426 	if (error == 0)
427 		nfs_refresh_inode(inode, &fattr);
428 	return error;
429 }
430 
431 /**
432  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
433  * @inode: inode of the file used
434  * @offset: file offset to start truncating
435  *
436  * This is a copy of the common vmtruncate, but with the locking
437  * corrected to take into account the fact that NFS requires
438  * inode->i_size to be updated under the inode->i_lock.
439  */
440 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
441 {
442 	loff_t oldsize;
443 	int err;
444 
445 	err = inode_newsize_ok(inode, offset);
446 	if (err)
447 		goto out;
448 
449 	spin_lock(&inode->i_lock);
450 	oldsize = inode->i_size;
451 	i_size_write(inode, offset);
452 	spin_unlock(&inode->i_lock);
453 
454 	truncate_pagecache(inode, oldsize, offset);
455 out:
456 	return err;
457 }
458 
459 /**
460  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
461  * @inode: pointer to struct inode
462  * @attr: pointer to struct iattr
463  *
464  * Note: we do this in the *proc.c in order to ensure that
465  *       it works for things like exclusive creates too.
466  */
467 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
468 {
469 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
470 		spin_lock(&inode->i_lock);
471 		if ((attr->ia_valid & ATTR_MODE) != 0) {
472 			int mode = attr->ia_mode & S_IALLUGO;
473 			mode |= inode->i_mode & ~S_IALLUGO;
474 			inode->i_mode = mode;
475 		}
476 		if ((attr->ia_valid & ATTR_UID) != 0)
477 			inode->i_uid = attr->ia_uid;
478 		if ((attr->ia_valid & ATTR_GID) != 0)
479 			inode->i_gid = attr->ia_gid;
480 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
481 		spin_unlock(&inode->i_lock);
482 	}
483 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
484 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
485 		nfs_vmtruncate(inode, attr->ia_size);
486 	}
487 }
488 
489 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
490 {
491 	struct inode *inode = dentry->d_inode;
492 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
493 	int err;
494 
495 	/* Flush out writes to the server in order to update c/mtime.  */
496 	if (S_ISREG(inode->i_mode)) {
497 		err = filemap_write_and_wait(inode->i_mapping);
498 		if (err)
499 			goto out;
500 	}
501 
502 	/*
503 	 * We may force a getattr if the user cares about atime.
504 	 *
505 	 * Note that we only have to check the vfsmount flags here:
506 	 *  - NFS always sets S_NOATIME by so checking it would give a
507 	 *    bogus result
508 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
509 	 *    no point in checking those.
510 	 */
511  	if ((mnt->mnt_flags & MNT_NOATIME) ||
512  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
513 		need_atime = 0;
514 
515 	if (need_atime)
516 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
517 	else
518 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
519 	if (!err) {
520 		generic_fillattr(inode, stat);
521 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
522 	}
523 out:
524 	return err;
525 }
526 
527 /**
528  * nfs_close_context - Common close_context() routine NFSv2/v3
529  * @ctx: pointer to context
530  * @is_sync: is this a synchronous close
531  *
532  * always ensure that the attributes are up to date if we're mounted
533  * with close-to-open semantics
534  */
535 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
536 {
537 	struct inode *inode;
538 	struct nfs_server *server;
539 
540 	if (!(ctx->mode & FMODE_WRITE))
541 		return;
542 	if (!is_sync)
543 		return;
544 	inode = ctx->path.dentry->d_inode;
545 	if (!list_empty(&NFS_I(inode)->open_files))
546 		return;
547 	server = NFS_SERVER(inode);
548 	if (server->flags & NFS_MOUNT_NOCTO)
549 		return;
550 	nfs_revalidate_inode(server, inode);
551 }
552 
553 static struct nfs_open_context *alloc_nfs_open_context(struct path *path, struct rpc_cred *cred)
554 {
555 	struct nfs_open_context *ctx;
556 
557 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
558 	if (ctx != NULL) {
559 		ctx->path = *path;
560 		path_get(&ctx->path);
561 		ctx->cred = get_rpccred(cred);
562 		ctx->state = NULL;
563 		ctx->lockowner = current->files;
564 		ctx->flags = 0;
565 		ctx->error = 0;
566 		ctx->dir_cookie = 0;
567 		atomic_set(&ctx->count, 1);
568 	}
569 	return ctx;
570 }
571 
572 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
573 {
574 	if (ctx != NULL)
575 		atomic_inc(&ctx->count);
576 	return ctx;
577 }
578 
579 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
580 {
581 	struct inode *inode = ctx->path.dentry->d_inode;
582 
583 	if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
584 		return;
585 	list_del(&ctx->list);
586 	spin_unlock(&inode->i_lock);
587 	NFS_PROTO(inode)->close_context(ctx, is_sync);
588 	if (ctx->cred != NULL)
589 		put_rpccred(ctx->cred);
590 	path_put(&ctx->path);
591 	kfree(ctx);
592 }
593 
594 void put_nfs_open_context(struct nfs_open_context *ctx)
595 {
596 	__put_nfs_open_context(ctx, 0);
597 }
598 
599 /*
600  * Ensure that mmap has a recent RPC credential for use when writing out
601  * shared pages
602  */
603 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
604 {
605 	struct inode *inode = filp->f_path.dentry->d_inode;
606 	struct nfs_inode *nfsi = NFS_I(inode);
607 
608 	filp->private_data = get_nfs_open_context(ctx);
609 	spin_lock(&inode->i_lock);
610 	list_add(&ctx->list, &nfsi->open_files);
611 	spin_unlock(&inode->i_lock);
612 }
613 
614 /*
615  * Given an inode, search for an open context with the desired characteristics
616  */
617 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
618 {
619 	struct nfs_inode *nfsi = NFS_I(inode);
620 	struct nfs_open_context *pos, *ctx = NULL;
621 
622 	spin_lock(&inode->i_lock);
623 	list_for_each_entry(pos, &nfsi->open_files, list) {
624 		if (cred != NULL && pos->cred != cred)
625 			continue;
626 		if ((pos->mode & mode) == mode) {
627 			ctx = get_nfs_open_context(pos);
628 			break;
629 		}
630 	}
631 	spin_unlock(&inode->i_lock);
632 	return ctx;
633 }
634 
635 static void nfs_file_clear_open_context(struct file *filp)
636 {
637 	struct inode *inode = filp->f_path.dentry->d_inode;
638 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
639 
640 	if (ctx) {
641 		filp->private_data = NULL;
642 		spin_lock(&inode->i_lock);
643 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
644 		spin_unlock(&inode->i_lock);
645 		__put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
646 	}
647 }
648 
649 /*
650  * These allocate and release file read/write context information.
651  */
652 int nfs_open(struct inode *inode, struct file *filp)
653 {
654 	struct nfs_open_context *ctx;
655 	struct rpc_cred *cred;
656 
657 	cred = rpc_lookup_cred();
658 	if (IS_ERR(cred))
659 		return PTR_ERR(cred);
660 	ctx = alloc_nfs_open_context(&filp->f_path, cred);
661 	put_rpccred(cred);
662 	if (ctx == NULL)
663 		return -ENOMEM;
664 	ctx->mode = filp->f_mode;
665 	nfs_file_set_open_context(filp, ctx);
666 	put_nfs_open_context(ctx);
667 	nfs_fscache_set_inode_cookie(inode, filp);
668 	return 0;
669 }
670 
671 int nfs_release(struct inode *inode, struct file *filp)
672 {
673 	nfs_file_clear_open_context(filp);
674 	return 0;
675 }
676 
677 /*
678  * This function is called whenever some part of NFS notices that
679  * the cached attributes have to be refreshed.
680  */
681 int
682 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
683 {
684 	int		 status = -ESTALE;
685 	struct nfs_fattr fattr;
686 	struct nfs_inode *nfsi = NFS_I(inode);
687 
688 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
689 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
690 
691 	if (is_bad_inode(inode))
692 		goto out;
693 	if (NFS_STALE(inode))
694 		goto out;
695 
696 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
697 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
698 	if (status != 0) {
699 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
700 			 inode->i_sb->s_id,
701 			 (long long)NFS_FILEID(inode), status);
702 		if (status == -ESTALE) {
703 			nfs_zap_caches(inode);
704 			if (!S_ISDIR(inode->i_mode))
705 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
706 		}
707 		goto out;
708 	}
709 
710 	status = nfs_refresh_inode(inode, &fattr);
711 	if (status) {
712 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
713 			 inode->i_sb->s_id,
714 			 (long long)NFS_FILEID(inode), status);
715 		goto out;
716 	}
717 
718 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
719 		nfs_zap_acl_cache(inode);
720 
721 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
722 		inode->i_sb->s_id,
723 		(long long)NFS_FILEID(inode));
724 
725  out:
726 	return status;
727 }
728 
729 int nfs_attribute_timeout(struct inode *inode)
730 {
731 	struct nfs_inode *nfsi = NFS_I(inode);
732 
733 	if (nfs_have_delegated_attributes(inode))
734 		return 0;
735 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
736 }
737 
738 /**
739  * nfs_revalidate_inode - Revalidate the inode attributes
740  * @server - pointer to nfs_server struct
741  * @inode - pointer to inode struct
742  *
743  * Updates inode attribute information by retrieving the data from the server.
744  */
745 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
746 {
747 	if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
748 			&& !nfs_attribute_timeout(inode))
749 		return NFS_STALE(inode) ? -ESTALE : 0;
750 	return __nfs_revalidate_inode(server, inode);
751 }
752 
753 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
754 {
755 	struct nfs_inode *nfsi = NFS_I(inode);
756 
757 	if (mapping->nrpages != 0) {
758 		int ret = invalidate_inode_pages2(mapping);
759 		if (ret < 0)
760 			return ret;
761 	}
762 	spin_lock(&inode->i_lock);
763 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
764 	if (S_ISDIR(inode->i_mode))
765 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
766 	spin_unlock(&inode->i_lock);
767 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
768 	nfs_fscache_reset_inode_cookie(inode);
769 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
770 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
771 	return 0;
772 }
773 
774 /**
775  * nfs_revalidate_mapping - Revalidate the pagecache
776  * @inode - pointer to host inode
777  * @mapping - pointer to mapping
778  */
779 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
780 {
781 	struct nfs_inode *nfsi = NFS_I(inode);
782 	int ret = 0;
783 
784 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
785 			|| nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
786 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
787 		if (ret < 0)
788 			goto out;
789 	}
790 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
791 		ret = nfs_invalidate_mapping(inode, mapping);
792 out:
793 	return ret;
794 }
795 
796 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
797 {
798 	struct nfs_inode *nfsi = NFS_I(inode);
799 
800 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
801 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
802 			&& nfsi->change_attr == fattr->pre_change_attr) {
803 		nfsi->change_attr = fattr->change_attr;
804 		if (S_ISDIR(inode->i_mode))
805 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
806 	}
807 	/* If we have atomic WCC data, we may update some attributes */
808 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
809 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
810 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
811 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
812 
813 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
814 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
815 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
816 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
817 			if (S_ISDIR(inode->i_mode))
818 				nfsi->cache_validity |= NFS_INO_INVALID_DATA;
819 	}
820 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
821 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
822 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
823 			&& nfsi->npages == 0)
824 			i_size_write(inode, nfs_size_to_loff_t(fattr->size));
825 }
826 
827 /**
828  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
829  * @inode - pointer to inode
830  * @fattr - updated attributes
831  *
832  * Verifies the attribute cache. If we have just changed the attributes,
833  * so that fattr carries weak cache consistency data, then it may
834  * also update the ctime/mtime/change_attribute.
835  */
836 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
837 {
838 	struct nfs_inode *nfsi = NFS_I(inode);
839 	loff_t cur_size, new_isize;
840 	unsigned long invalid = 0;
841 
842 
843 	/* Has the inode gone and changed behind our back? */
844 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
845 		return -EIO;
846 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
847 		return -EIO;
848 
849 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
850 			nfsi->change_attr != fattr->change_attr)
851 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
852 
853 	/* Verify a few of the more important attributes */
854 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
855 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
856 
857 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
858 		cur_size = i_size_read(inode);
859 		new_isize = nfs_size_to_loff_t(fattr->size);
860 		if (cur_size != new_isize && nfsi->npages == 0)
861 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
862 	}
863 
864 	/* Have any file permissions changed? */
865 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
866 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
867 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
868 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
869 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
870 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
871 
872 	/* Has the link count changed? */
873 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
874 		invalid |= NFS_INO_INVALID_ATTR;
875 
876 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
877 		invalid |= NFS_INO_INVALID_ATIME;
878 
879 	if (invalid != 0)
880 		nfsi->cache_validity |= invalid;
881 
882 	nfsi->read_cache_jiffies = fattr->time_start;
883 	return 0;
884 }
885 
886 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
887 {
888 	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
889 		return 0;
890 	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
891 }
892 
893 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
894 {
895 	if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
896 		return 0;
897 	return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
898 }
899 
900 static atomic_long_t nfs_attr_generation_counter;
901 
902 static unsigned long nfs_read_attr_generation_counter(void)
903 {
904 	return atomic_long_read(&nfs_attr_generation_counter);
905 }
906 
907 unsigned long nfs_inc_attr_generation_counter(void)
908 {
909 	return atomic_long_inc_return(&nfs_attr_generation_counter);
910 }
911 
912 void nfs_fattr_init(struct nfs_fattr *fattr)
913 {
914 	fattr->valid = 0;
915 	fattr->time_start = jiffies;
916 	fattr->gencount = nfs_inc_attr_generation_counter();
917 }
918 
919 /**
920  * nfs_inode_attrs_need_update - check if the inode attributes need updating
921  * @inode - pointer to inode
922  * @fattr - attributes
923  *
924  * Attempt to divine whether or not an RPC call reply carrying stale
925  * attributes got scheduled after another call carrying updated ones.
926  *
927  * To do so, the function first assumes that a more recent ctime means
928  * that the attributes in fattr are newer, however it also attempt to
929  * catch the case where ctime either didn't change, or went backwards
930  * (if someone reset the clock on the server) by looking at whether
931  * or not this RPC call was started after the inode was last updated.
932  * Note also the check for wraparound of 'attr_gencount'
933  *
934  * The function returns 'true' if it thinks the attributes in 'fattr' are
935  * more recent than the ones cached in the inode.
936  *
937  */
938 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
939 {
940 	const struct nfs_inode *nfsi = NFS_I(inode);
941 
942 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
943 		nfs_ctime_need_update(inode, fattr) ||
944 		nfs_size_need_update(inode, fattr) ||
945 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
946 }
947 
948 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
949 {
950 	if (nfs_inode_attrs_need_update(inode, fattr))
951 		return nfs_update_inode(inode, fattr);
952 	return nfs_check_inode_attributes(inode, fattr);
953 }
954 
955 /**
956  * nfs_refresh_inode - try to update the inode attribute cache
957  * @inode - pointer to inode
958  * @fattr - updated attributes
959  *
960  * Check that an RPC call that returned attributes has not overlapped with
961  * other recent updates of the inode metadata, then decide whether it is
962  * safe to do a full update of the inode attributes, or whether just to
963  * call nfs_check_inode_attributes.
964  */
965 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
966 {
967 	int status;
968 
969 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
970 		return 0;
971 	spin_lock(&inode->i_lock);
972 	status = nfs_refresh_inode_locked(inode, fattr);
973 	spin_unlock(&inode->i_lock);
974 
975 	return status;
976 }
977 
978 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
979 {
980 	struct nfs_inode *nfsi = NFS_I(inode);
981 
982 	nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
983 	if (S_ISDIR(inode->i_mode))
984 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
985 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
986 		return 0;
987 	return nfs_refresh_inode_locked(inode, fattr);
988 }
989 
990 /**
991  * nfs_post_op_update_inode - try to update the inode attribute cache
992  * @inode - pointer to inode
993  * @fattr - updated attributes
994  *
995  * After an operation that has changed the inode metadata, mark the
996  * attribute cache as being invalid, then try to update it.
997  *
998  * NB: if the server didn't return any post op attributes, this
999  * function will force the retrieval of attributes before the next
1000  * NFS request.  Thus it should be used only for operations that
1001  * are expected to change one or more attributes, to avoid
1002  * unnecessary NFS requests and trips through nfs_update_inode().
1003  */
1004 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1005 {
1006 	int status;
1007 
1008 	spin_lock(&inode->i_lock);
1009 	status = nfs_post_op_update_inode_locked(inode, fattr);
1010 	spin_unlock(&inode->i_lock);
1011 	return status;
1012 }
1013 
1014 /**
1015  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1016  * @inode - pointer to inode
1017  * @fattr - updated attributes
1018  *
1019  * After an operation that has changed the inode metadata, mark the
1020  * attribute cache as being invalid, then try to update it. Fake up
1021  * weak cache consistency data, if none exist.
1022  *
1023  * This function is mainly designed to be used by the ->write_done() functions.
1024  */
1025 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1026 {
1027 	int status;
1028 
1029 	spin_lock(&inode->i_lock);
1030 	/* Don't do a WCC update if these attributes are already stale */
1031 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1032 			!nfs_inode_attrs_need_update(inode, fattr)) {
1033 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1034 				| NFS_ATTR_FATTR_PRESIZE
1035 				| NFS_ATTR_FATTR_PREMTIME
1036 				| NFS_ATTR_FATTR_PRECTIME);
1037 		goto out_noforce;
1038 	}
1039 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1040 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1041 		fattr->pre_change_attr = NFS_I(inode)->change_attr;
1042 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1043 	}
1044 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1045 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1046 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1047 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1048 	}
1049 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1050 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1051 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1052 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1053 	}
1054 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1055 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1056 		fattr->pre_size = i_size_read(inode);
1057 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1058 	}
1059 out_noforce:
1060 	status = nfs_post_op_update_inode_locked(inode, fattr);
1061 	spin_unlock(&inode->i_lock);
1062 	return status;
1063 }
1064 
1065 /*
1066  * Many nfs protocol calls return the new file attributes after
1067  * an operation.  Here we update the inode to reflect the state
1068  * of the server's inode.
1069  *
1070  * This is a bit tricky because we have to make sure all dirty pages
1071  * have been sent off to the server before calling invalidate_inode_pages.
1072  * To make sure no other process adds more write requests while we try
1073  * our best to flush them, we make them sleep during the attribute refresh.
1074  *
1075  * A very similar scenario holds for the dir cache.
1076  */
1077 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1078 {
1079 	struct nfs_server *server;
1080 	struct nfs_inode *nfsi = NFS_I(inode);
1081 	loff_t cur_isize, new_isize;
1082 	unsigned long invalid = 0;
1083 	unsigned long now = jiffies;
1084 	unsigned long save_cache_validity;
1085 
1086 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1087 			__func__, inode->i_sb->s_id, inode->i_ino,
1088 			atomic_read(&inode->i_count), fattr->valid);
1089 
1090 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1091 		goto out_fileid;
1092 
1093 	/*
1094 	 * Make sure the inode's type hasn't changed.
1095 	 */
1096 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1097 		goto out_changed;
1098 
1099 	server = NFS_SERVER(inode);
1100 	/* Update the fsid? */
1101 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1102 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1103 			!test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
1104 		server->fsid = fattr->fsid;
1105 
1106 	/*
1107 	 * Update the read time so we don't revalidate too often.
1108 	 */
1109 	nfsi->read_cache_jiffies = fattr->time_start;
1110 
1111 	save_cache_validity = nfsi->cache_validity;
1112 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1113 			| NFS_INO_INVALID_ATIME
1114 			| NFS_INO_REVAL_FORCED
1115 			| NFS_INO_REVAL_PAGECACHE);
1116 
1117 	/* Do atomic weak cache consistency updates */
1118 	nfs_wcc_update_inode(inode, fattr);
1119 
1120 	/* More cache consistency checks */
1121 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1122 		if (nfsi->change_attr != fattr->change_attr) {
1123 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1124 					inode->i_sb->s_id, inode->i_ino);
1125 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1126 			if (S_ISDIR(inode->i_mode))
1127 				nfs_force_lookup_revalidate(inode);
1128 			nfsi->change_attr = fattr->change_attr;
1129 		}
1130 	} else if (server->caps & NFS_CAP_CHANGE_ATTR)
1131 		invalid |= save_cache_validity;
1132 
1133 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1134 		/* NFSv2/v3: Check if the mtime agrees */
1135 		if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1136 			dprintk("NFS: mtime change on server for file %s/%ld\n",
1137 					inode->i_sb->s_id, inode->i_ino);
1138 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1139 			if (S_ISDIR(inode->i_mode))
1140 				nfs_force_lookup_revalidate(inode);
1141 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1142 		}
1143 	} else if (server->caps & NFS_CAP_MTIME)
1144 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1145 				| NFS_INO_INVALID_DATA
1146 				| NFS_INO_REVAL_PAGECACHE
1147 				| NFS_INO_REVAL_FORCED);
1148 
1149 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1150 		/* If ctime has changed we should definitely clear access+acl caches */
1151 		if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1152 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1153 			/* and probably clear data for a directory too as utimes can cause
1154 			 * havoc with our cache.
1155 			 */
1156 			if (S_ISDIR(inode->i_mode)) {
1157 				invalid |= NFS_INO_INVALID_DATA;
1158 				nfs_force_lookup_revalidate(inode);
1159 			}
1160 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1161 		}
1162 	} else if (server->caps & NFS_CAP_CTIME)
1163 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1164 				| NFS_INO_INVALID_ACCESS
1165 				| NFS_INO_INVALID_ACL
1166 				| NFS_INO_REVAL_FORCED);
1167 
1168 	/* Check if our cached file size is stale */
1169 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1170 		new_isize = nfs_size_to_loff_t(fattr->size);
1171 		cur_isize = i_size_read(inode);
1172 		if (new_isize != cur_isize) {
1173 			/* Do we perhaps have any outstanding writes, or has
1174 			 * the file grown beyond our last write? */
1175 			if (nfsi->npages == 0 || new_isize > cur_isize) {
1176 				i_size_write(inode, new_isize);
1177 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1178 			}
1179 			dprintk("NFS: isize change on server for file %s/%ld\n",
1180 					inode->i_sb->s_id, inode->i_ino);
1181 		}
1182 	} else
1183 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1184 				| NFS_INO_REVAL_PAGECACHE
1185 				| NFS_INO_REVAL_FORCED);
1186 
1187 
1188 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1189 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1190 	else if (server->caps & NFS_CAP_ATIME)
1191 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1192 				| NFS_INO_REVAL_FORCED);
1193 
1194 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1195 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1196 			umode_t newmode = inode->i_mode & S_IFMT;
1197 			newmode |= fattr->mode & S_IALLUGO;
1198 			inode->i_mode = newmode;
1199 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1200 		}
1201 	} else if (server->caps & NFS_CAP_MODE)
1202 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1203 				| NFS_INO_INVALID_ACCESS
1204 				| NFS_INO_INVALID_ACL
1205 				| NFS_INO_REVAL_FORCED);
1206 
1207 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1208 		if (inode->i_uid != fattr->uid) {
1209 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1210 			inode->i_uid = fattr->uid;
1211 		}
1212 	} else if (server->caps & NFS_CAP_OWNER)
1213 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1214 				| NFS_INO_INVALID_ACCESS
1215 				| NFS_INO_INVALID_ACL
1216 				| NFS_INO_REVAL_FORCED);
1217 
1218 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1219 		if (inode->i_gid != fattr->gid) {
1220 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1221 			inode->i_gid = fattr->gid;
1222 		}
1223 	} else if (server->caps & NFS_CAP_OWNER_GROUP)
1224 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1225 				| NFS_INO_INVALID_ACCESS
1226 				| NFS_INO_INVALID_ACL
1227 				| NFS_INO_REVAL_FORCED);
1228 
1229 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1230 		if (inode->i_nlink != fattr->nlink) {
1231 			invalid |= NFS_INO_INVALID_ATTR;
1232 			if (S_ISDIR(inode->i_mode))
1233 				invalid |= NFS_INO_INVALID_DATA;
1234 			inode->i_nlink = fattr->nlink;
1235 		}
1236 	} else if (server->caps & NFS_CAP_NLINK)
1237 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1238 				| NFS_INO_REVAL_FORCED);
1239 
1240 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1241 		/*
1242 		 * report the blocks in 512byte units
1243 		 */
1244 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1245  	}
1246 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1247 		inode->i_blocks = fattr->du.nfs2.blocks;
1248 
1249 	/* Update attrtimeo value if we're out of the unstable period */
1250 	if (invalid & NFS_INO_INVALID_ATTR) {
1251 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1252 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1253 		nfsi->attrtimeo_timestamp = now;
1254 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1255 	} else {
1256 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1257 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1258 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1259 			nfsi->attrtimeo_timestamp = now;
1260 		}
1261 	}
1262 	invalid &= ~NFS_INO_INVALID_ATTR;
1263 	/* Don't invalidate the data if we were to blame */
1264 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1265 				|| S_ISLNK(inode->i_mode)))
1266 		invalid &= ~NFS_INO_INVALID_DATA;
1267 	if (!nfs_have_delegation(inode, FMODE_READ) ||
1268 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1269 		nfsi->cache_validity |= invalid;
1270 
1271 	return 0;
1272  out_changed:
1273 	/*
1274 	 * Big trouble! The inode has become a different object.
1275 	 */
1276 	printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1277 			__func__, inode->i_ino, inode->i_mode, fattr->mode);
1278  out_err:
1279 	/*
1280 	 * No need to worry about unhashing the dentry, as the
1281 	 * lookup validation will know that the inode is bad.
1282 	 * (But we fall through to invalidate the caches.)
1283 	 */
1284 	nfs_invalidate_inode(inode);
1285 	return -ESTALE;
1286 
1287  out_fileid:
1288 	printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1289 		"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1290 		NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1291 		(long long)nfsi->fileid, (long long)fattr->fileid);
1292 	goto out_err;
1293 }
1294 
1295 
1296 #ifdef CONFIG_NFS_V4
1297 
1298 /*
1299  * Clean out any remaining NFSv4 state that might be left over due
1300  * to open() calls that passed nfs_atomic_lookup, but failed to call
1301  * nfs_open().
1302  */
1303 void nfs4_clear_inode(struct inode *inode)
1304 {
1305 	/* If we are holding a delegation, return it! */
1306 	nfs_inode_return_delegation_noreclaim(inode);
1307 	/* First call standard NFS clear_inode() code */
1308 	nfs_clear_inode(inode);
1309 }
1310 #endif
1311 
1312 struct inode *nfs_alloc_inode(struct super_block *sb)
1313 {
1314 	struct nfs_inode *nfsi;
1315 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1316 	if (!nfsi)
1317 		return NULL;
1318 	nfsi->flags = 0UL;
1319 	nfsi->cache_validity = 0UL;
1320 #ifdef CONFIG_NFS_V3_ACL
1321 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1322 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1323 #endif
1324 #ifdef CONFIG_NFS_V4
1325 	nfsi->nfs4_acl = NULL;
1326 #endif /* CONFIG_NFS_V4 */
1327 	return &nfsi->vfs_inode;
1328 }
1329 
1330 void nfs_destroy_inode(struct inode *inode)
1331 {
1332 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1333 }
1334 
1335 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1336 {
1337 #ifdef CONFIG_NFS_V4
1338 	INIT_LIST_HEAD(&nfsi->open_states);
1339 	nfsi->delegation = NULL;
1340 	nfsi->delegation_state = 0;
1341 	init_rwsem(&nfsi->rwsem);
1342 #endif
1343 }
1344 
1345 static void init_once(void *foo)
1346 {
1347 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1348 
1349 	inode_init_once(&nfsi->vfs_inode);
1350 	INIT_LIST_HEAD(&nfsi->open_files);
1351 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1352 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1353 	INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1354 	nfsi->npages = 0;
1355 	nfsi->ncommit = 0;
1356 	atomic_set(&nfsi->silly_count, 1);
1357 	INIT_HLIST_HEAD(&nfsi->silly_list);
1358 	init_waitqueue_head(&nfsi->waitqueue);
1359 	nfs4_init_once(nfsi);
1360 }
1361 
1362 static int __init nfs_init_inodecache(void)
1363 {
1364 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1365 					     sizeof(struct nfs_inode),
1366 					     0, (SLAB_RECLAIM_ACCOUNT|
1367 						SLAB_MEM_SPREAD),
1368 					     init_once);
1369 	if (nfs_inode_cachep == NULL)
1370 		return -ENOMEM;
1371 
1372 	return 0;
1373 }
1374 
1375 static void nfs_destroy_inodecache(void)
1376 {
1377 	kmem_cache_destroy(nfs_inode_cachep);
1378 }
1379 
1380 struct workqueue_struct *nfsiod_workqueue;
1381 
1382 /*
1383  * start up the nfsiod workqueue
1384  */
1385 static int nfsiod_start(void)
1386 {
1387 	struct workqueue_struct *wq;
1388 	dprintk("RPC:       creating workqueue nfsiod\n");
1389 	wq = create_singlethread_workqueue("nfsiod");
1390 	if (wq == NULL)
1391 		return -ENOMEM;
1392 	nfsiod_workqueue = wq;
1393 	return 0;
1394 }
1395 
1396 /*
1397  * Destroy the nfsiod workqueue
1398  */
1399 static void nfsiod_stop(void)
1400 {
1401 	struct workqueue_struct *wq;
1402 
1403 	wq = nfsiod_workqueue;
1404 	if (wq == NULL)
1405 		return;
1406 	nfsiod_workqueue = NULL;
1407 	destroy_workqueue(wq);
1408 }
1409 
1410 /*
1411  * Initialize NFS
1412  */
1413 static int __init init_nfs_fs(void)
1414 {
1415 	int err;
1416 
1417 	err = nfs_dns_resolver_init();
1418 	if (err < 0)
1419 		goto out8;
1420 
1421 	err = nfs_fscache_register();
1422 	if (err < 0)
1423 		goto out7;
1424 
1425 	err = nfsiod_start();
1426 	if (err)
1427 		goto out6;
1428 
1429 	err = nfs_fs_proc_init();
1430 	if (err)
1431 		goto out5;
1432 
1433 	err = nfs_init_nfspagecache();
1434 	if (err)
1435 		goto out4;
1436 
1437 	err = nfs_init_inodecache();
1438 	if (err)
1439 		goto out3;
1440 
1441 	err = nfs_init_readpagecache();
1442 	if (err)
1443 		goto out2;
1444 
1445 	err = nfs_init_writepagecache();
1446 	if (err)
1447 		goto out1;
1448 
1449 	err = nfs_init_directcache();
1450 	if (err)
1451 		goto out0;
1452 
1453 #ifdef CONFIG_PROC_FS
1454 	rpc_proc_register(&nfs_rpcstat);
1455 #endif
1456 	if ((err = register_nfs_fs()) != 0)
1457 		goto out;
1458 	return 0;
1459 out:
1460 #ifdef CONFIG_PROC_FS
1461 	rpc_proc_unregister("nfs");
1462 #endif
1463 	nfs_destroy_directcache();
1464 out0:
1465 	nfs_destroy_writepagecache();
1466 out1:
1467 	nfs_destroy_readpagecache();
1468 out2:
1469 	nfs_destroy_inodecache();
1470 out3:
1471 	nfs_destroy_nfspagecache();
1472 out4:
1473 	nfs_fs_proc_exit();
1474 out5:
1475 	nfsiod_stop();
1476 out6:
1477 	nfs_fscache_unregister();
1478 out7:
1479 	nfs_dns_resolver_destroy();
1480 out8:
1481 	return err;
1482 }
1483 
1484 static void __exit exit_nfs_fs(void)
1485 {
1486 	nfs_destroy_directcache();
1487 	nfs_destroy_writepagecache();
1488 	nfs_destroy_readpagecache();
1489 	nfs_destroy_inodecache();
1490 	nfs_destroy_nfspagecache();
1491 	nfs_fscache_unregister();
1492 	nfs_dns_resolver_destroy();
1493 #ifdef CONFIG_PROC_FS
1494 	rpc_proc_unregister("nfs");
1495 #endif
1496 	unregister_nfs_fs();
1497 	nfs_fs_proc_exit();
1498 	nfsiod_stop();
1499 }
1500 
1501 /* Not quite true; I just maintain it */
1502 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1503 MODULE_LICENSE("GPL");
1504 module_param(enable_ino64, bool, 0644);
1505 
1506 module_init(init_nfs_fs)
1507 module_exit(exit_nfs_fs)
1508