xref: /linux/fs/nfs/inode.c (revision dfff0fa65ab15db45acd64b3189787d37ab163cd)
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 
40 #include <asm/system.h>
41 #include <asm/uaccess.h>
42 
43 #include "nfs4_fs.h"
44 #include "callback.h"
45 #include "delegation.h"
46 #include "iostat.h"
47 #include "internal.h"
48 #include "fscache.h"
49 
50 #define NFSDBG_FACILITY		NFSDBG_VFS
51 
52 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
53 
54 /* Default is to see 64-bit inode numbers */
55 static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
56 
57 static void nfs_invalidate_inode(struct inode *);
58 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
59 
60 static struct kmem_cache * nfs_inode_cachep;
61 
62 static inline unsigned long
63 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
64 {
65 	return nfs_fileid_to_ino_t(fattr->fileid);
66 }
67 
68 /**
69  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
70  * @word: long word containing the bit lock
71  */
72 int nfs_wait_bit_killable(void *word)
73 {
74 	if (fatal_signal_pending(current))
75 		return -ERESTARTSYS;
76 	schedule();
77 	return 0;
78 }
79 
80 /**
81  * nfs_compat_user_ino64 - returns the user-visible inode number
82  * @fileid: 64-bit fileid
83  *
84  * This function returns a 32-bit inode number if the boot parameter
85  * nfs.enable_ino64 is zero.
86  */
87 u64 nfs_compat_user_ino64(u64 fileid)
88 {
89 	int ino;
90 
91 	if (enable_ino64)
92 		return fileid;
93 	ino = fileid;
94 	if (sizeof(ino) < sizeof(fileid))
95 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
96 	return ino;
97 }
98 
99 int nfs_write_inode(struct inode *inode, int sync)
100 {
101 	int ret;
102 
103 	if (sync) {
104 		ret = filemap_fdatawait(inode->i_mapping);
105 		if (ret == 0)
106 			ret = nfs_commit_inode(inode, FLUSH_SYNC);
107 	} else
108 		ret = nfs_commit_inode(inode, 0);
109 	if (ret >= 0)
110 		return 0;
111 	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
112 	return ret;
113 }
114 
115 void nfs_clear_inode(struct inode *inode)
116 {
117 	/*
118 	 * The following should never happen...
119 	 */
120 	BUG_ON(nfs_have_writebacks(inode));
121 	BUG_ON(!list_empty(&NFS_I(inode)->open_files));
122 	nfs_zap_acl_cache(inode);
123 	nfs_access_zap_cache(inode);
124 	nfs_fscache_release_inode_cookie(inode);
125 }
126 
127 /**
128  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
129  */
130 int nfs_sync_mapping(struct address_space *mapping)
131 {
132 	int ret;
133 
134 	if (mapping->nrpages == 0)
135 		return 0;
136 	unmap_mapping_range(mapping, 0, 0, 0);
137 	ret = filemap_write_and_wait(mapping);
138 	if (ret != 0)
139 		goto out;
140 	ret = nfs_wb_all(mapping->host);
141 out:
142 	return ret;
143 }
144 
145 /*
146  * Invalidate the local caches
147  */
148 static void nfs_zap_caches_locked(struct inode *inode)
149 {
150 	struct nfs_inode *nfsi = NFS_I(inode);
151 	int mode = inode->i_mode;
152 
153 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
154 
155 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
156 	nfsi->attrtimeo_timestamp = jiffies;
157 
158 	memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
159 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
160 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
161 	else
162 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
163 }
164 
165 void nfs_zap_caches(struct inode *inode)
166 {
167 	spin_lock(&inode->i_lock);
168 	nfs_zap_caches_locked(inode);
169 	spin_unlock(&inode->i_lock);
170 }
171 
172 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
173 {
174 	if (mapping->nrpages != 0) {
175 		spin_lock(&inode->i_lock);
176 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
177 		spin_unlock(&inode->i_lock);
178 	}
179 }
180 
181 void nfs_zap_acl_cache(struct inode *inode)
182 {
183 	void (*clear_acl_cache)(struct inode *);
184 
185 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
186 	if (clear_acl_cache != NULL)
187 		clear_acl_cache(inode);
188 	spin_lock(&inode->i_lock);
189 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
190 	spin_unlock(&inode->i_lock);
191 }
192 
193 void nfs_invalidate_atime(struct inode *inode)
194 {
195 	spin_lock(&inode->i_lock);
196 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
197 	spin_unlock(&inode->i_lock);
198 }
199 
200 /*
201  * Invalidate, but do not unhash, the inode.
202  * NB: must be called with inode->i_lock held!
203  */
204 static void nfs_invalidate_inode(struct inode *inode)
205 {
206 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
207 	nfs_zap_caches_locked(inode);
208 }
209 
210 struct nfs_find_desc {
211 	struct nfs_fh		*fh;
212 	struct nfs_fattr	*fattr;
213 };
214 
215 /*
216  * In NFSv3 we can have 64bit inode numbers. In order to support
217  * this, and re-exported directories (also seen in NFSv2)
218  * we are forced to allow 2 different inodes to have the same
219  * i_ino.
220  */
221 static int
222 nfs_find_actor(struct inode *inode, void *opaque)
223 {
224 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
225 	struct nfs_fh		*fh = desc->fh;
226 	struct nfs_fattr	*fattr = desc->fattr;
227 
228 	if (NFS_FILEID(inode) != fattr->fileid)
229 		return 0;
230 	if (nfs_compare_fh(NFS_FH(inode), fh))
231 		return 0;
232 	if (is_bad_inode(inode) || NFS_STALE(inode))
233 		return 0;
234 	return 1;
235 }
236 
237 static int
238 nfs_init_locked(struct inode *inode, void *opaque)
239 {
240 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
241 	struct nfs_fattr	*fattr = desc->fattr;
242 
243 	set_nfs_fileid(inode, fattr->fileid);
244 	nfs_copy_fh(NFS_FH(inode), desc->fh);
245 	return 0;
246 }
247 
248 /* Don't use READDIRPLUS on directories that we believe are too large */
249 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
250 
251 /*
252  * This is our front-end to iget that looks up inodes by file handle
253  * instead of inode number.
254  */
255 struct inode *
256 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
257 {
258 	struct nfs_find_desc desc = {
259 		.fh	= fh,
260 		.fattr	= fattr
261 	};
262 	struct inode *inode = ERR_PTR(-ENOENT);
263 	unsigned long hash;
264 
265 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
266 		goto out_no_inode;
267 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
268 		goto out_no_inode;
269 
270 	hash = nfs_fattr_to_ino_t(fattr);
271 
272 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
273 	if (inode == NULL) {
274 		inode = ERR_PTR(-ENOMEM);
275 		goto out_no_inode;
276 	}
277 
278 	if (inode->i_state & I_NEW) {
279 		struct nfs_inode *nfsi = NFS_I(inode);
280 		unsigned long now = jiffies;
281 
282 		/* We set i_ino for the few things that still rely on it,
283 		 * such as stat(2) */
284 		inode->i_ino = hash;
285 
286 		/* We can't support update_atime(), since the server will reset it */
287 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
288 		inode->i_mode = fattr->mode;
289 		/* Why so? Because we want revalidate for devices/FIFOs, and
290 		 * that's precisely what we have in nfs_file_inode_operations.
291 		 */
292 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
293 		if (S_ISREG(inode->i_mode)) {
294 			inode->i_fop = &nfs_file_operations;
295 			inode->i_data.a_ops = &nfs_file_aops;
296 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
297 		} else if (S_ISDIR(inode->i_mode)) {
298 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
299 			inode->i_fop = &nfs_dir_operations;
300 			if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
301 			    && fattr->size <= NFS_LIMIT_READDIRPLUS)
302 				set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
303 			/* Deal with crossing mountpoints */
304 			if ((fattr->valid & NFS_ATTR_FATTR_FSID)
305 					&& !nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
306 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
307 					inode->i_op = &nfs_referral_inode_operations;
308 				else
309 					inode->i_op = &nfs_mountpoint_inode_operations;
310 				inode->i_fop = NULL;
311 				set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
312 			}
313 		} else if (S_ISLNK(inode->i_mode))
314 			inode->i_op = &nfs_symlink_inode_operations;
315 		else
316 			init_special_inode(inode, inode->i_mode, fattr->rdev);
317 
318 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
319 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
320 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
321 		nfsi->change_attr = 0;
322 		inode->i_size = 0;
323 		inode->i_nlink = 0;
324 		inode->i_uid = -2;
325 		inode->i_gid = -2;
326 		inode->i_blocks = 0;
327 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
328 
329 		nfsi->read_cache_jiffies = fattr->time_start;
330 		nfsi->attr_gencount = fattr->gencount;
331 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
332 			inode->i_atime = fattr->atime;
333 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
334 			inode->i_mtime = fattr->mtime;
335 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
336 			inode->i_ctime = fattr->ctime;
337 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
338 			nfsi->change_attr = fattr->change_attr;
339 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
340 			inode->i_size = nfs_size_to_loff_t(fattr->size);
341 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
342 			inode->i_nlink = fattr->nlink;
343 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
344 			inode->i_uid = fattr->uid;
345 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
346 			inode->i_gid = fattr->gid;
347 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
348 			inode->i_blocks = fattr->du.nfs2.blocks;
349 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
350 			/*
351 			 * report the blocks in 512byte units
352 			 */
353 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
354 		}
355 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
356 		nfsi->attrtimeo_timestamp = now;
357 		nfsi->access_cache = RB_ROOT;
358 
359 		nfs_fscache_init_inode_cookie(inode);
360 
361 		unlock_new_inode(inode);
362 	} else
363 		nfs_refresh_inode(inode, fattr);
364 	dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
365 		inode->i_sb->s_id,
366 		(long long)NFS_FILEID(inode),
367 		atomic_read(&inode->i_count));
368 
369 out:
370 	return inode;
371 
372 out_no_inode:
373 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
374 	goto out;
375 }
376 
377 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
378 
379 int
380 nfs_setattr(struct dentry *dentry, struct iattr *attr)
381 {
382 	struct inode *inode = dentry->d_inode;
383 	struct nfs_fattr fattr;
384 	int error;
385 
386 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
387 
388 	/* skip mode change if it's just for clearing setuid/setgid */
389 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
390 		attr->ia_valid &= ~ATTR_MODE;
391 
392 	if (attr->ia_valid & ATTR_SIZE) {
393 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
394 			attr->ia_valid &= ~ATTR_SIZE;
395 	}
396 
397 	/* Optimization: if the end result is no change, don't RPC */
398 	attr->ia_valid &= NFS_VALID_ATTRS;
399 	if ((attr->ia_valid & ~ATTR_FILE) == 0)
400 		return 0;
401 
402 	/* Write all dirty data */
403 	if (S_ISREG(inode->i_mode)) {
404 		filemap_write_and_wait(inode->i_mapping);
405 		nfs_wb_all(inode);
406 	}
407 	/*
408 	 * Return any delegations if we're going to change ACLs
409 	 */
410 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
411 		nfs_inode_return_delegation(inode);
412 	error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
413 	if (error == 0)
414 		nfs_refresh_inode(inode, &fattr);
415 	return error;
416 }
417 
418 /**
419  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
420  * @inode: inode of the file used
421  * @offset: file offset to start truncating
422  *
423  * This is a copy of the common vmtruncate, but with the locking
424  * corrected to take into account the fact that NFS requires
425  * inode->i_size to be updated under the inode->i_lock.
426  */
427 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
428 {
429 	if (i_size_read(inode) < offset) {
430 		unsigned long limit;
431 
432 		limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
433 		if (limit != RLIM_INFINITY && offset > limit)
434 			goto out_sig;
435 		if (offset > inode->i_sb->s_maxbytes)
436 			goto out_big;
437 		spin_lock(&inode->i_lock);
438 		i_size_write(inode, offset);
439 		spin_unlock(&inode->i_lock);
440 	} else {
441 		struct address_space *mapping = inode->i_mapping;
442 
443 		/*
444 		 * truncation of in-use swapfiles is disallowed - it would
445 		 * cause subsequent swapout to scribble on the now-freed
446 		 * blocks.
447 		 */
448 		if (IS_SWAPFILE(inode))
449 			return -ETXTBSY;
450 		spin_lock(&inode->i_lock);
451 		i_size_write(inode, offset);
452 		spin_unlock(&inode->i_lock);
453 
454 		/*
455 		 * unmap_mapping_range is called twice, first simply for
456 		 * efficiency so that truncate_inode_pages does fewer
457 		 * single-page unmaps.  However after this first call, and
458 		 * before truncate_inode_pages finishes, it is possible for
459 		 * private pages to be COWed, which remain after
460 		 * truncate_inode_pages finishes, hence the second
461 		 * unmap_mapping_range call must be made for correctness.
462 		 */
463 		unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
464 		truncate_inode_pages(mapping, offset);
465 		unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
466 	}
467 	return 0;
468 out_sig:
469 	send_sig(SIGXFSZ, current, 0);
470 out_big:
471 	return -EFBIG;
472 }
473 
474 /**
475  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
476  * @inode: pointer to struct inode
477  * @attr: pointer to struct iattr
478  *
479  * Note: we do this in the *proc.c in order to ensure that
480  *       it works for things like exclusive creates too.
481  */
482 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
483 {
484 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
485 		spin_lock(&inode->i_lock);
486 		if ((attr->ia_valid & ATTR_MODE) != 0) {
487 			int mode = attr->ia_mode & S_IALLUGO;
488 			mode |= inode->i_mode & ~S_IALLUGO;
489 			inode->i_mode = mode;
490 		}
491 		if ((attr->ia_valid & ATTR_UID) != 0)
492 			inode->i_uid = attr->ia_uid;
493 		if ((attr->ia_valid & ATTR_GID) != 0)
494 			inode->i_gid = attr->ia_gid;
495 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
496 		spin_unlock(&inode->i_lock);
497 	}
498 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
499 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
500 		nfs_vmtruncate(inode, attr->ia_size);
501 	}
502 }
503 
504 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
505 {
506 	struct inode *inode = dentry->d_inode;
507 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
508 	int err;
509 
510 	/*
511 	 * Flush out writes to the server in order to update c/mtime.
512 	 *
513 	 * Hold the i_mutex to suspend application writes temporarily;
514 	 * this prevents long-running writing applications from blocking
515 	 * nfs_wb_nocommit.
516 	 */
517 	if (S_ISREG(inode->i_mode)) {
518 		mutex_lock(&inode->i_mutex);
519 		nfs_wb_nocommit(inode);
520 		mutex_unlock(&inode->i_mutex);
521 	}
522 
523 	/*
524 	 * We may force a getattr if the user cares about atime.
525 	 *
526 	 * Note that we only have to check the vfsmount flags here:
527 	 *  - NFS always sets S_NOATIME by so checking it would give a
528 	 *    bogus result
529 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
530 	 *    no point in checking those.
531 	 */
532  	if ((mnt->mnt_flags & MNT_NOATIME) ||
533  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
534 		need_atime = 0;
535 
536 	if (need_atime)
537 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
538 	else
539 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
540 	if (!err) {
541 		generic_fillattr(inode, stat);
542 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
543 	}
544 	return err;
545 }
546 
547 /**
548  * nfs_close_context - Common close_context() routine NFSv2/v3
549  * @ctx: pointer to context
550  * @is_sync: is this a synchronous close
551  *
552  * always ensure that the attributes are up to date if we're mounted
553  * with close-to-open semantics
554  */
555 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
556 {
557 	struct inode *inode;
558 	struct nfs_server *server;
559 
560 	if (!(ctx->mode & FMODE_WRITE))
561 		return;
562 	if (!is_sync)
563 		return;
564 	inode = ctx->path.dentry->d_inode;
565 	if (!list_empty(&NFS_I(inode)->open_files))
566 		return;
567 	server = NFS_SERVER(inode);
568 	if (server->flags & NFS_MOUNT_NOCTO)
569 		return;
570 	nfs_revalidate_inode(server, inode);
571 }
572 
573 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
574 {
575 	struct nfs_open_context *ctx;
576 
577 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
578 	if (ctx != NULL) {
579 		ctx->path.dentry = dget(dentry);
580 		ctx->path.mnt = mntget(mnt);
581 		ctx->cred = get_rpccred(cred);
582 		ctx->state = NULL;
583 		ctx->lockowner = current->files;
584 		ctx->flags = 0;
585 		ctx->error = 0;
586 		ctx->dir_cookie = 0;
587 		atomic_set(&ctx->count, 1);
588 	}
589 	return ctx;
590 }
591 
592 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
593 {
594 	if (ctx != NULL)
595 		atomic_inc(&ctx->count);
596 	return ctx;
597 }
598 
599 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
600 {
601 	struct inode *inode = ctx->path.dentry->d_inode;
602 
603 	if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
604 		return;
605 	list_del(&ctx->list);
606 	spin_unlock(&inode->i_lock);
607 	NFS_PROTO(inode)->close_context(ctx, is_sync);
608 	if (ctx->cred != NULL)
609 		put_rpccred(ctx->cred);
610 	path_put(&ctx->path);
611 	kfree(ctx);
612 }
613 
614 void put_nfs_open_context(struct nfs_open_context *ctx)
615 {
616 	__put_nfs_open_context(ctx, 0);
617 }
618 
619 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
620 {
621 	__put_nfs_open_context(ctx, 1);
622 }
623 
624 /*
625  * Ensure that mmap has a recent RPC credential for use when writing out
626  * shared pages
627  */
628 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
629 {
630 	struct inode *inode = filp->f_path.dentry->d_inode;
631 	struct nfs_inode *nfsi = NFS_I(inode);
632 
633 	filp->private_data = get_nfs_open_context(ctx);
634 	spin_lock(&inode->i_lock);
635 	list_add(&ctx->list, &nfsi->open_files);
636 	spin_unlock(&inode->i_lock);
637 }
638 
639 /*
640  * Given an inode, search for an open context with the desired characteristics
641  */
642 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
643 {
644 	struct nfs_inode *nfsi = NFS_I(inode);
645 	struct nfs_open_context *pos, *ctx = NULL;
646 
647 	spin_lock(&inode->i_lock);
648 	list_for_each_entry(pos, &nfsi->open_files, list) {
649 		if (cred != NULL && pos->cred != cred)
650 			continue;
651 		if ((pos->mode & mode) == mode) {
652 			ctx = get_nfs_open_context(pos);
653 			break;
654 		}
655 	}
656 	spin_unlock(&inode->i_lock);
657 	return ctx;
658 }
659 
660 static void nfs_file_clear_open_context(struct file *filp)
661 {
662 	struct inode *inode = filp->f_path.dentry->d_inode;
663 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
664 
665 	if (ctx) {
666 		filp->private_data = NULL;
667 		spin_lock(&inode->i_lock);
668 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
669 		spin_unlock(&inode->i_lock);
670 		put_nfs_open_context_sync(ctx);
671 	}
672 }
673 
674 /*
675  * These allocate and release file read/write context information.
676  */
677 int nfs_open(struct inode *inode, struct file *filp)
678 {
679 	struct nfs_open_context *ctx;
680 	struct rpc_cred *cred;
681 
682 	cred = rpc_lookup_cred();
683 	if (IS_ERR(cred))
684 		return PTR_ERR(cred);
685 	ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
686 	put_rpccred(cred);
687 	if (ctx == NULL)
688 		return -ENOMEM;
689 	ctx->mode = filp->f_mode;
690 	nfs_file_set_open_context(filp, ctx);
691 	put_nfs_open_context(ctx);
692 	nfs_fscache_set_inode_cookie(inode, filp);
693 	return 0;
694 }
695 
696 int nfs_release(struct inode *inode, struct file *filp)
697 {
698 	nfs_file_clear_open_context(filp);
699 	return 0;
700 }
701 
702 /*
703  * This function is called whenever some part of NFS notices that
704  * the cached attributes have to be refreshed.
705  */
706 int
707 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
708 {
709 	int		 status = -ESTALE;
710 	struct nfs_fattr fattr;
711 	struct nfs_inode *nfsi = NFS_I(inode);
712 
713 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
714 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
715 
716 	if (is_bad_inode(inode))
717 		goto out;
718 	if (NFS_STALE(inode))
719 		goto out;
720 
721 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
722 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
723 	if (status != 0) {
724 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
725 			 inode->i_sb->s_id,
726 			 (long long)NFS_FILEID(inode), status);
727 		if (status == -ESTALE) {
728 			nfs_zap_caches(inode);
729 			if (!S_ISDIR(inode->i_mode))
730 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
731 		}
732 		goto out;
733 	}
734 
735 	status = nfs_refresh_inode(inode, &fattr);
736 	if (status) {
737 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
738 			 inode->i_sb->s_id,
739 			 (long long)NFS_FILEID(inode), status);
740 		goto out;
741 	}
742 
743 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
744 		nfs_zap_acl_cache(inode);
745 
746 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
747 		inode->i_sb->s_id,
748 		(long long)NFS_FILEID(inode));
749 
750  out:
751 	return status;
752 }
753 
754 int nfs_attribute_timeout(struct inode *inode)
755 {
756 	struct nfs_inode *nfsi = NFS_I(inode);
757 
758 	if (nfs_have_delegation(inode, FMODE_READ))
759 		return 0;
760 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
761 }
762 
763 /**
764  * nfs_revalidate_inode - Revalidate the inode attributes
765  * @server - pointer to nfs_server struct
766  * @inode - pointer to inode struct
767  *
768  * Updates inode attribute information by retrieving the data from the server.
769  */
770 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
771 {
772 	if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
773 			&& !nfs_attribute_timeout(inode))
774 		return NFS_STALE(inode) ? -ESTALE : 0;
775 	return __nfs_revalidate_inode(server, inode);
776 }
777 
778 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
779 {
780 	struct nfs_inode *nfsi = NFS_I(inode);
781 
782 	if (mapping->nrpages != 0) {
783 		int ret = invalidate_inode_pages2(mapping);
784 		if (ret < 0)
785 			return ret;
786 	}
787 	spin_lock(&inode->i_lock);
788 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
789 	if (S_ISDIR(inode->i_mode))
790 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
791 	spin_unlock(&inode->i_lock);
792 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
793 	nfs_fscache_reset_inode_cookie(inode);
794 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
795 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
796 	return 0;
797 }
798 
799 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
800 {
801 	int ret = 0;
802 
803 	mutex_lock(&inode->i_mutex);
804 	if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
805 		ret = nfs_sync_mapping(mapping);
806 		if (ret == 0)
807 			ret = nfs_invalidate_mapping_nolock(inode, mapping);
808 	}
809 	mutex_unlock(&inode->i_mutex);
810 	return ret;
811 }
812 
813 /**
814  * nfs_revalidate_mapping_nolock - Revalidate the pagecache
815  * @inode - pointer to host inode
816  * @mapping - pointer to mapping
817  */
818 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
819 {
820 	struct nfs_inode *nfsi = NFS_I(inode);
821 	int ret = 0;
822 
823 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
824 			|| nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
825 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
826 		if (ret < 0)
827 			goto out;
828 	}
829 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
830 		ret = nfs_invalidate_mapping_nolock(inode, mapping);
831 out:
832 	return ret;
833 }
834 
835 /**
836  * nfs_revalidate_mapping - Revalidate the pagecache
837  * @inode - pointer to host inode
838  * @mapping - pointer to mapping
839  *
840  * This version of the function will take the inode->i_mutex and attempt to
841  * flush out all dirty data if it needs to invalidate the page cache.
842  */
843 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
844 {
845 	struct nfs_inode *nfsi = NFS_I(inode);
846 	int ret = 0;
847 
848 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
849 			|| nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
850 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
851 		if (ret < 0)
852 			goto out;
853 	}
854 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
855 		ret = nfs_invalidate_mapping(inode, mapping);
856 out:
857 	return ret;
858 }
859 
860 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
861 {
862 	struct nfs_inode *nfsi = NFS_I(inode);
863 
864 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
865 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
866 			&& nfsi->change_attr == fattr->pre_change_attr) {
867 		nfsi->change_attr = fattr->change_attr;
868 		if (S_ISDIR(inode->i_mode))
869 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
870 	}
871 	/* If we have atomic WCC data, we may update some attributes */
872 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
873 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
874 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
875 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
876 
877 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
878 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
879 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
880 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
881 			if (S_ISDIR(inode->i_mode))
882 				nfsi->cache_validity |= NFS_INO_INVALID_DATA;
883 	}
884 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
885 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
886 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
887 			&& nfsi->npages == 0)
888 			i_size_write(inode, nfs_size_to_loff_t(fattr->size));
889 }
890 
891 /**
892  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
893  * @inode - pointer to inode
894  * @fattr - updated attributes
895  *
896  * Verifies the attribute cache. If we have just changed the attributes,
897  * so that fattr carries weak cache consistency data, then it may
898  * also update the ctime/mtime/change_attribute.
899  */
900 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
901 {
902 	struct nfs_inode *nfsi = NFS_I(inode);
903 	loff_t cur_size, new_isize;
904 	unsigned long invalid = 0;
905 
906 
907 	/* Has the inode gone and changed behind our back? */
908 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
909 		return -EIO;
910 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
911 		return -EIO;
912 
913 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
914 			nfsi->change_attr != fattr->change_attr)
915 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
916 
917 	/* Verify a few of the more important attributes */
918 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
919 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
920 
921 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
922 		cur_size = i_size_read(inode);
923 		new_isize = nfs_size_to_loff_t(fattr->size);
924 		if (cur_size != new_isize && nfsi->npages == 0)
925 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
926 	}
927 
928 	/* Have any file permissions changed? */
929 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
930 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
931 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
932 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
933 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
934 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
935 
936 	/* Has the link count changed? */
937 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
938 		invalid |= NFS_INO_INVALID_ATTR;
939 
940 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
941 		invalid |= NFS_INO_INVALID_ATIME;
942 
943 	if (invalid != 0)
944 		nfsi->cache_validity |= invalid;
945 
946 	nfsi->read_cache_jiffies = fattr->time_start;
947 	return 0;
948 }
949 
950 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
951 {
952 	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
953 		return 0;
954 	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
955 }
956 
957 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
958 {
959 	if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
960 		return 0;
961 	return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
962 }
963 
964 static atomic_long_t nfs_attr_generation_counter;
965 
966 static unsigned long nfs_read_attr_generation_counter(void)
967 {
968 	return atomic_long_read(&nfs_attr_generation_counter);
969 }
970 
971 unsigned long nfs_inc_attr_generation_counter(void)
972 {
973 	return atomic_long_inc_return(&nfs_attr_generation_counter);
974 }
975 
976 void nfs_fattr_init(struct nfs_fattr *fattr)
977 {
978 	fattr->valid = 0;
979 	fattr->time_start = jiffies;
980 	fattr->gencount = nfs_inc_attr_generation_counter();
981 }
982 
983 /**
984  * nfs_inode_attrs_need_update - check if the inode attributes need updating
985  * @inode - pointer to inode
986  * @fattr - attributes
987  *
988  * Attempt to divine whether or not an RPC call reply carrying stale
989  * attributes got scheduled after another call carrying updated ones.
990  *
991  * To do so, the function first assumes that a more recent ctime means
992  * that the attributes in fattr are newer, however it also attempt to
993  * catch the case where ctime either didn't change, or went backwards
994  * (if someone reset the clock on the server) by looking at whether
995  * or not this RPC call was started after the inode was last updated.
996  * Note also the check for wraparound of 'attr_gencount'
997  *
998  * The function returns 'true' if it thinks the attributes in 'fattr' are
999  * more recent than the ones cached in the inode.
1000  *
1001  */
1002 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1003 {
1004 	const struct nfs_inode *nfsi = NFS_I(inode);
1005 
1006 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1007 		nfs_ctime_need_update(inode, fattr) ||
1008 		nfs_size_need_update(inode, fattr) ||
1009 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1010 }
1011 
1012 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1013 {
1014 	if (nfs_inode_attrs_need_update(inode, fattr))
1015 		return nfs_update_inode(inode, fattr);
1016 	return nfs_check_inode_attributes(inode, fattr);
1017 }
1018 
1019 /**
1020  * nfs_refresh_inode - try to update the inode attribute cache
1021  * @inode - pointer to inode
1022  * @fattr - updated attributes
1023  *
1024  * Check that an RPC call that returned attributes has not overlapped with
1025  * other recent updates of the inode metadata, then decide whether it is
1026  * safe to do a full update of the inode attributes, or whether just to
1027  * call nfs_check_inode_attributes.
1028  */
1029 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1030 {
1031 	int status;
1032 
1033 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1034 		return 0;
1035 	spin_lock(&inode->i_lock);
1036 	status = nfs_refresh_inode_locked(inode, fattr);
1037 	spin_unlock(&inode->i_lock);
1038 
1039 	return status;
1040 }
1041 
1042 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1043 {
1044 	struct nfs_inode *nfsi = NFS_I(inode);
1045 
1046 	nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1047 	if (S_ISDIR(inode->i_mode))
1048 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1049 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1050 		return 0;
1051 	return nfs_refresh_inode_locked(inode, fattr);
1052 }
1053 
1054 /**
1055  * nfs_post_op_update_inode - try to update the inode attribute cache
1056  * @inode - pointer to inode
1057  * @fattr - updated attributes
1058  *
1059  * After an operation that has changed the inode metadata, mark the
1060  * attribute cache as being invalid, then try to update it.
1061  *
1062  * NB: if the server didn't return any post op attributes, this
1063  * function will force the retrieval of attributes before the next
1064  * NFS request.  Thus it should be used only for operations that
1065  * are expected to change one or more attributes, to avoid
1066  * unnecessary NFS requests and trips through nfs_update_inode().
1067  */
1068 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1069 {
1070 	int status;
1071 
1072 	spin_lock(&inode->i_lock);
1073 	status = nfs_post_op_update_inode_locked(inode, fattr);
1074 	spin_unlock(&inode->i_lock);
1075 	return status;
1076 }
1077 
1078 /**
1079  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1080  * @inode - pointer to inode
1081  * @fattr - updated attributes
1082  *
1083  * After an operation that has changed the inode metadata, mark the
1084  * attribute cache as being invalid, then try to update it. Fake up
1085  * weak cache consistency data, if none exist.
1086  *
1087  * This function is mainly designed to be used by the ->write_done() functions.
1088  */
1089 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1090 {
1091 	int status;
1092 
1093 	spin_lock(&inode->i_lock);
1094 	/* Don't do a WCC update if these attributes are already stale */
1095 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1096 			!nfs_inode_attrs_need_update(inode, fattr)) {
1097 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1098 				| NFS_ATTR_FATTR_PRESIZE
1099 				| NFS_ATTR_FATTR_PREMTIME
1100 				| NFS_ATTR_FATTR_PRECTIME);
1101 		goto out_noforce;
1102 	}
1103 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1104 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1105 		fattr->pre_change_attr = NFS_I(inode)->change_attr;
1106 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1107 	}
1108 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1109 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1110 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1111 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1112 	}
1113 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1114 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1115 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1116 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1117 	}
1118 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1119 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1120 		fattr->pre_size = i_size_read(inode);
1121 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1122 	}
1123 out_noforce:
1124 	status = nfs_post_op_update_inode_locked(inode, fattr);
1125 	spin_unlock(&inode->i_lock);
1126 	return status;
1127 }
1128 
1129 /*
1130  * Many nfs protocol calls return the new file attributes after
1131  * an operation.  Here we update the inode to reflect the state
1132  * of the server's inode.
1133  *
1134  * This is a bit tricky because we have to make sure all dirty pages
1135  * have been sent off to the server before calling invalidate_inode_pages.
1136  * To make sure no other process adds more write requests while we try
1137  * our best to flush them, we make them sleep during the attribute refresh.
1138  *
1139  * A very similar scenario holds for the dir cache.
1140  */
1141 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1142 {
1143 	struct nfs_server *server;
1144 	struct nfs_inode *nfsi = NFS_I(inode);
1145 	loff_t cur_isize, new_isize;
1146 	unsigned long invalid = 0;
1147 	unsigned long now = jiffies;
1148 
1149 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1150 			__func__, inode->i_sb->s_id, inode->i_ino,
1151 			atomic_read(&inode->i_count), fattr->valid);
1152 
1153 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1154 		goto out_fileid;
1155 
1156 	/*
1157 	 * Make sure the inode's type hasn't changed.
1158 	 */
1159 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1160 		goto out_changed;
1161 
1162 	server = NFS_SERVER(inode);
1163 	/* Update the fsid? */
1164 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1165 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1166 			!test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
1167 		server->fsid = fattr->fsid;
1168 
1169 	/*
1170 	 * Update the read time so we don't revalidate too often.
1171 	 */
1172 	nfsi->read_cache_jiffies = fattr->time_start;
1173 
1174 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) || (fattr->valid & (NFS_ATTR_FATTR_MTIME|NFS_ATTR_FATTR_CTIME)))
1175 	    nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1176 		    | NFS_INO_INVALID_ATIME
1177 		    | NFS_INO_REVAL_PAGECACHE);
1178 
1179 	/* Do atomic weak cache consistency updates */
1180 	nfs_wcc_update_inode(inode, fattr);
1181 
1182 	/* More cache consistency checks */
1183 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1184 		if (nfsi->change_attr != fattr->change_attr) {
1185 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1186 					inode->i_sb->s_id, inode->i_ino);
1187 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1188 			if (S_ISDIR(inode->i_mode))
1189 				nfs_force_lookup_revalidate(inode);
1190 			nfsi->change_attr = fattr->change_attr;
1191 		}
1192 	}
1193 
1194 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1195 		/* NFSv2/v3: Check if the mtime agrees */
1196 		if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1197 			dprintk("NFS: mtime change on server for file %s/%ld\n",
1198 					inode->i_sb->s_id, inode->i_ino);
1199 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1200 			if (S_ISDIR(inode->i_mode))
1201 				nfs_force_lookup_revalidate(inode);
1202 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1203 		}
1204 	}
1205 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1206 		/* If ctime has changed we should definitely clear access+acl caches */
1207 		if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1208 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1209 			/* and probably clear data for a directory too as utimes can cause
1210 			 * havoc with our cache.
1211 			 */
1212 			if (S_ISDIR(inode->i_mode)) {
1213 				invalid |= NFS_INO_INVALID_DATA;
1214 				nfs_force_lookup_revalidate(inode);
1215 			}
1216 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1217 		}
1218 	}
1219 
1220 	/* Check if our cached file size is stale */
1221 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1222 		new_isize = nfs_size_to_loff_t(fattr->size);
1223 		cur_isize = i_size_read(inode);
1224 		if (new_isize != cur_isize) {
1225 			/* Do we perhaps have any outstanding writes, or has
1226 			 * the file grown beyond our last write? */
1227 			if (nfsi->npages == 0 || new_isize > cur_isize) {
1228 				i_size_write(inode, new_isize);
1229 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1230 			}
1231 			dprintk("NFS: isize change on server for file %s/%ld\n",
1232 					inode->i_sb->s_id, inode->i_ino);
1233 		}
1234 	}
1235 
1236 
1237 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1238 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1239 
1240 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1241 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1242 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1243 			inode->i_mode = fattr->mode;
1244 		}
1245 	}
1246 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1247 		if (inode->i_uid != fattr->uid) {
1248 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1249 			inode->i_uid = fattr->uid;
1250 		}
1251 	}
1252 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1253 		if (inode->i_gid != fattr->gid) {
1254 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1255 			inode->i_gid = fattr->gid;
1256 		}
1257 	}
1258 
1259 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1260 		if (inode->i_nlink != fattr->nlink) {
1261 			invalid |= NFS_INO_INVALID_ATTR;
1262 			if (S_ISDIR(inode->i_mode))
1263 				invalid |= NFS_INO_INVALID_DATA;
1264 			inode->i_nlink = fattr->nlink;
1265 		}
1266 	}
1267 
1268 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1269 		/*
1270 		 * report the blocks in 512byte units
1271 		 */
1272 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1273  	}
1274 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1275 		inode->i_blocks = fattr->du.nfs2.blocks;
1276 
1277 	/* Update attrtimeo value if we're out of the unstable period */
1278 	if (invalid & NFS_INO_INVALID_ATTR) {
1279 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1280 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1281 		nfsi->attrtimeo_timestamp = now;
1282 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1283 	} else {
1284 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1285 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1286 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1287 			nfsi->attrtimeo_timestamp = now;
1288 		}
1289 	}
1290 	invalid &= ~NFS_INO_INVALID_ATTR;
1291 	/* Don't invalidate the data if we were to blame */
1292 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1293 				|| S_ISLNK(inode->i_mode)))
1294 		invalid &= ~NFS_INO_INVALID_DATA;
1295 	if (!nfs_have_delegation(inode, FMODE_READ) ||
1296 			(nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1297 		nfsi->cache_validity |= invalid;
1298 	nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1299 
1300 	return 0;
1301  out_changed:
1302 	/*
1303 	 * Big trouble! The inode has become a different object.
1304 	 */
1305 	printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1306 			__func__, inode->i_ino, inode->i_mode, fattr->mode);
1307  out_err:
1308 	/*
1309 	 * No need to worry about unhashing the dentry, as the
1310 	 * lookup validation will know that the inode is bad.
1311 	 * (But we fall through to invalidate the caches.)
1312 	 */
1313 	nfs_invalidate_inode(inode);
1314 	return -ESTALE;
1315 
1316  out_fileid:
1317 	printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1318 		"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1319 		NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1320 		(long long)nfsi->fileid, (long long)fattr->fileid);
1321 	goto out_err;
1322 }
1323 
1324 
1325 #ifdef CONFIG_NFS_V4
1326 
1327 /*
1328  * Clean out any remaining NFSv4 state that might be left over due
1329  * to open() calls that passed nfs_atomic_lookup, but failed to call
1330  * nfs_open().
1331  */
1332 void nfs4_clear_inode(struct inode *inode)
1333 {
1334 	/* If we are holding a delegation, return it! */
1335 	nfs_inode_return_delegation_noreclaim(inode);
1336 	/* First call standard NFS clear_inode() code */
1337 	nfs_clear_inode(inode);
1338 }
1339 #endif
1340 
1341 struct inode *nfs_alloc_inode(struct super_block *sb)
1342 {
1343 	struct nfs_inode *nfsi;
1344 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1345 	if (!nfsi)
1346 		return NULL;
1347 	nfsi->flags = 0UL;
1348 	nfsi->cache_validity = 0UL;
1349 #ifdef CONFIG_NFS_V3_ACL
1350 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1351 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1352 #endif
1353 #ifdef CONFIG_NFS_V4
1354 	nfsi->nfs4_acl = NULL;
1355 #endif /* CONFIG_NFS_V4 */
1356 	return &nfsi->vfs_inode;
1357 }
1358 
1359 void nfs_destroy_inode(struct inode *inode)
1360 {
1361 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1362 }
1363 
1364 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1365 {
1366 #ifdef CONFIG_NFS_V4
1367 	INIT_LIST_HEAD(&nfsi->open_states);
1368 	nfsi->delegation = NULL;
1369 	nfsi->delegation_state = 0;
1370 	init_rwsem(&nfsi->rwsem);
1371 #endif
1372 }
1373 
1374 static void init_once(void *foo)
1375 {
1376 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1377 
1378 	inode_init_once(&nfsi->vfs_inode);
1379 	INIT_LIST_HEAD(&nfsi->open_files);
1380 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1381 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1382 	INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1383 	nfsi->npages = 0;
1384 	atomic_set(&nfsi->silly_count, 1);
1385 	INIT_HLIST_HEAD(&nfsi->silly_list);
1386 	init_waitqueue_head(&nfsi->waitqueue);
1387 	nfs4_init_once(nfsi);
1388 }
1389 
1390 static int __init nfs_init_inodecache(void)
1391 {
1392 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1393 					     sizeof(struct nfs_inode),
1394 					     0, (SLAB_RECLAIM_ACCOUNT|
1395 						SLAB_MEM_SPREAD),
1396 					     init_once);
1397 	if (nfs_inode_cachep == NULL)
1398 		return -ENOMEM;
1399 
1400 	return 0;
1401 }
1402 
1403 static void nfs_destroy_inodecache(void)
1404 {
1405 	kmem_cache_destroy(nfs_inode_cachep);
1406 }
1407 
1408 struct workqueue_struct *nfsiod_workqueue;
1409 
1410 /*
1411  * start up the nfsiod workqueue
1412  */
1413 static int nfsiod_start(void)
1414 {
1415 	struct workqueue_struct *wq;
1416 	dprintk("RPC:       creating workqueue nfsiod\n");
1417 	wq = create_singlethread_workqueue("nfsiod");
1418 	if (wq == NULL)
1419 		return -ENOMEM;
1420 	nfsiod_workqueue = wq;
1421 	return 0;
1422 }
1423 
1424 /*
1425  * Destroy the nfsiod workqueue
1426  */
1427 static void nfsiod_stop(void)
1428 {
1429 	struct workqueue_struct *wq;
1430 
1431 	wq = nfsiod_workqueue;
1432 	if (wq == NULL)
1433 		return;
1434 	nfsiod_workqueue = NULL;
1435 	destroy_workqueue(wq);
1436 }
1437 
1438 /*
1439  * Initialize NFS
1440  */
1441 static int __init init_nfs_fs(void)
1442 {
1443 	int err;
1444 
1445 	err = nfs_fscache_register();
1446 	if (err < 0)
1447 		goto out7;
1448 
1449 	err = nfsiod_start();
1450 	if (err)
1451 		goto out6;
1452 
1453 	err = nfs_fs_proc_init();
1454 	if (err)
1455 		goto out5;
1456 
1457 	err = nfs_init_nfspagecache();
1458 	if (err)
1459 		goto out4;
1460 
1461 	err = nfs_init_inodecache();
1462 	if (err)
1463 		goto out3;
1464 
1465 	err = nfs_init_readpagecache();
1466 	if (err)
1467 		goto out2;
1468 
1469 	err = nfs_init_writepagecache();
1470 	if (err)
1471 		goto out1;
1472 
1473 	err = nfs_init_directcache();
1474 	if (err)
1475 		goto out0;
1476 
1477 #ifdef CONFIG_PROC_FS
1478 	rpc_proc_register(&nfs_rpcstat);
1479 #endif
1480 	if ((err = register_nfs_fs()) != 0)
1481 		goto out;
1482 	return 0;
1483 out:
1484 #ifdef CONFIG_PROC_FS
1485 	rpc_proc_unregister("nfs");
1486 #endif
1487 	nfs_destroy_directcache();
1488 out0:
1489 	nfs_destroy_writepagecache();
1490 out1:
1491 	nfs_destroy_readpagecache();
1492 out2:
1493 	nfs_destroy_inodecache();
1494 out3:
1495 	nfs_destroy_nfspagecache();
1496 out4:
1497 	nfs_fs_proc_exit();
1498 out5:
1499 	nfsiod_stop();
1500 out6:
1501 	nfs_fscache_unregister();
1502 out7:
1503 	return err;
1504 }
1505 
1506 static void __exit exit_nfs_fs(void)
1507 {
1508 	nfs_destroy_directcache();
1509 	nfs_destroy_writepagecache();
1510 	nfs_destroy_readpagecache();
1511 	nfs_destroy_inodecache();
1512 	nfs_destroy_nfspagecache();
1513 	nfs_fscache_unregister();
1514 #ifdef CONFIG_PROC_FS
1515 	rpc_proc_unregister("nfs");
1516 #endif
1517 	unregister_nfs_fs();
1518 	nfs_fs_proc_exit();
1519 	nfsiod_stop();
1520 }
1521 
1522 /* Not quite true; I just maintain it */
1523 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1524 MODULE_LICENSE("GPL");
1525 module_param(enable_ino64, bool, 0644);
1526 
1527 module_init(init_nfs_fs)
1528 module_exit(exit_nfs_fs)
1529