xref: /openbmc/linux/fs/nfs/inode.c (revision 9d56dd3b083a3bec56e9da35ce07baca81030b03)
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 #include "dns_resolve.h"
50 
51 #define NFSDBG_FACILITY		NFSDBG_VFS
52 
53 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
54 
55 /* Default is to see 64-bit inode numbers */
56 static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
57 
58 static void nfs_invalidate_inode(struct inode *);
59 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
60 
61 static struct kmem_cache * nfs_inode_cachep;
62 
63 static inline unsigned long
64 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
65 {
66 	return nfs_fileid_to_ino_t(fattr->fileid);
67 }
68 
69 /**
70  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
71  * @word: long word containing the bit lock
72  */
73 int nfs_wait_bit_killable(void *word)
74 {
75 	if (fatal_signal_pending(current))
76 		return -ERESTARTSYS;
77 	schedule();
78 	return 0;
79 }
80 
81 /**
82  * nfs_compat_user_ino64 - returns the user-visible inode number
83  * @fileid: 64-bit fileid
84  *
85  * This function returns a 32-bit inode number if the boot parameter
86  * nfs.enable_ino64 is zero.
87  */
88 u64 nfs_compat_user_ino64(u64 fileid)
89 {
90 	int ino;
91 
92 	if (enable_ino64)
93 		return fileid;
94 	ino = fileid;
95 	if (sizeof(ino) < sizeof(fileid))
96 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
97 	return ino;
98 }
99 
100 int nfs_write_inode(struct inode *inode, int sync)
101 {
102 	int ret;
103 
104 	if (sync) {
105 		ret = filemap_fdatawait(inode->i_mapping);
106 		if (ret == 0)
107 			ret = nfs_commit_inode(inode, FLUSH_SYNC);
108 	} else
109 		ret = nfs_commit_inode(inode, 0);
110 	if (ret >= 0)
111 		return 0;
112 	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
113 	return ret;
114 }
115 
116 void nfs_clear_inode(struct inode *inode)
117 {
118 	/*
119 	 * The following should never happen...
120 	 */
121 	BUG_ON(nfs_have_writebacks(inode));
122 	BUG_ON(!list_empty(&NFS_I(inode)->open_files));
123 	nfs_zap_acl_cache(inode);
124 	nfs_access_zap_cache(inode);
125 	nfs_fscache_release_inode_cookie(inode);
126 }
127 
128 /**
129  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
130  */
131 int nfs_sync_mapping(struct address_space *mapping)
132 {
133 	int ret;
134 
135 	if (mapping->nrpages == 0)
136 		return 0;
137 	unmap_mapping_range(mapping, 0, 0, 0);
138 	ret = filemap_write_and_wait(mapping);
139 	if (ret != 0)
140 		goto out;
141 	ret = nfs_wb_all(mapping->host);
142 out:
143 	return ret;
144 }
145 
146 /*
147  * Invalidate the local caches
148  */
149 static void nfs_zap_caches_locked(struct inode *inode)
150 {
151 	struct nfs_inode *nfsi = NFS_I(inode);
152 	int mode = inode->i_mode;
153 
154 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
155 
156 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
157 	nfsi->attrtimeo_timestamp = jiffies;
158 
159 	memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
160 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
161 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
162 	else
163 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
164 }
165 
166 void nfs_zap_caches(struct inode *inode)
167 {
168 	spin_lock(&inode->i_lock);
169 	nfs_zap_caches_locked(inode);
170 	spin_unlock(&inode->i_lock);
171 }
172 
173 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
174 {
175 	if (mapping->nrpages != 0) {
176 		spin_lock(&inode->i_lock);
177 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
178 		spin_unlock(&inode->i_lock);
179 	}
180 }
181 
182 void nfs_zap_acl_cache(struct inode *inode)
183 {
184 	void (*clear_acl_cache)(struct inode *);
185 
186 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
187 	if (clear_acl_cache != NULL)
188 		clear_acl_cache(inode);
189 	spin_lock(&inode->i_lock);
190 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
191 	spin_unlock(&inode->i_lock);
192 }
193 
194 void nfs_invalidate_atime(struct inode *inode)
195 {
196 	spin_lock(&inode->i_lock);
197 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
198 	spin_unlock(&inode->i_lock);
199 }
200 
201 /*
202  * Invalidate, but do not unhash, the inode.
203  * NB: must be called with inode->i_lock held!
204  */
205 static void nfs_invalidate_inode(struct inode *inode)
206 {
207 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
208 	nfs_zap_caches_locked(inode);
209 }
210 
211 struct nfs_find_desc {
212 	struct nfs_fh		*fh;
213 	struct nfs_fattr	*fattr;
214 };
215 
216 /*
217  * In NFSv3 we can have 64bit inode numbers. In order to support
218  * this, and re-exported directories (also seen in NFSv2)
219  * we are forced to allow 2 different inodes to have the same
220  * i_ino.
221  */
222 static int
223 nfs_find_actor(struct inode *inode, void *opaque)
224 {
225 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
226 	struct nfs_fh		*fh = desc->fh;
227 	struct nfs_fattr	*fattr = desc->fattr;
228 
229 	if (NFS_FILEID(inode) != fattr->fileid)
230 		return 0;
231 	if (nfs_compare_fh(NFS_FH(inode), fh))
232 		return 0;
233 	if (is_bad_inode(inode) || NFS_STALE(inode))
234 		return 0;
235 	return 1;
236 }
237 
238 static int
239 nfs_init_locked(struct inode *inode, void *opaque)
240 {
241 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
242 	struct nfs_fattr	*fattr = desc->fattr;
243 
244 	set_nfs_fileid(inode, fattr->fileid);
245 	nfs_copy_fh(NFS_FH(inode), desc->fh);
246 	return 0;
247 }
248 
249 /* Don't use READDIRPLUS on directories that we believe are too large */
250 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
251 
252 /*
253  * This is our front-end to iget that looks up inodes by file handle
254  * instead of inode number.
255  */
256 struct inode *
257 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
258 {
259 	struct nfs_find_desc desc = {
260 		.fh	= fh,
261 		.fattr	= fattr
262 	};
263 	struct inode *inode = ERR_PTR(-ENOENT);
264 	unsigned long hash;
265 
266 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
267 		goto out_no_inode;
268 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
269 		goto out_no_inode;
270 
271 	hash = nfs_fattr_to_ino_t(fattr);
272 
273 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
274 	if (inode == NULL) {
275 		inode = ERR_PTR(-ENOMEM);
276 		goto out_no_inode;
277 	}
278 
279 	if (inode->i_state & I_NEW) {
280 		struct nfs_inode *nfsi = NFS_I(inode);
281 		unsigned long now = jiffies;
282 
283 		/* We set i_ino for the few things that still rely on it,
284 		 * such as stat(2) */
285 		inode->i_ino = hash;
286 
287 		/* We can't support update_atime(), since the server will reset it */
288 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
289 		inode->i_mode = fattr->mode;
290 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
291 				&& nfs_server_capable(inode, NFS_CAP_MODE))
292 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
293 				| NFS_INO_INVALID_ACCESS
294 				| NFS_INO_INVALID_ACL;
295 		/* Why so? Because we want revalidate for devices/FIFOs, and
296 		 * that's precisely what we have in nfs_file_inode_operations.
297 		 */
298 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
299 		if (S_ISREG(inode->i_mode)) {
300 			inode->i_fop = &nfs_file_operations;
301 			inode->i_data.a_ops = &nfs_file_aops;
302 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
303 		} else if (S_ISDIR(inode->i_mode)) {
304 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
305 			inode->i_fop = &nfs_dir_operations;
306 			if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
307 			    && fattr->size <= NFS_LIMIT_READDIRPLUS)
308 				set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
309 			/* Deal with crossing mountpoints */
310 			if ((fattr->valid & NFS_ATTR_FATTR_FSID)
311 					&& !nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
312 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
313 					inode->i_op = &nfs_referral_inode_operations;
314 				else
315 					inode->i_op = &nfs_mountpoint_inode_operations;
316 				inode->i_fop = NULL;
317 				set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
318 			}
319 		} else if (S_ISLNK(inode->i_mode))
320 			inode->i_op = &nfs_symlink_inode_operations;
321 		else
322 			init_special_inode(inode, inode->i_mode, fattr->rdev);
323 
324 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
325 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
326 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
327 		nfsi->change_attr = 0;
328 		inode->i_size = 0;
329 		inode->i_nlink = 0;
330 		inode->i_uid = -2;
331 		inode->i_gid = -2;
332 		inode->i_blocks = 0;
333 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
334 
335 		nfsi->read_cache_jiffies = fattr->time_start;
336 		nfsi->attr_gencount = fattr->gencount;
337 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
338 			inode->i_atime = fattr->atime;
339 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
340 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
341 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
342 			inode->i_mtime = fattr->mtime;
343 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
344 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
345 				| NFS_INO_INVALID_DATA;
346 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
347 			inode->i_ctime = fattr->ctime;
348 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
349 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
350 				| NFS_INO_INVALID_ACCESS
351 				| NFS_INO_INVALID_ACL;
352 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
353 			nfsi->change_attr = fattr->change_attr;
354 		else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
355 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
356 				| NFS_INO_INVALID_DATA;
357 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
358 			inode->i_size = nfs_size_to_loff_t(fattr->size);
359 		else
360 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
361 				| NFS_INO_INVALID_DATA
362 				| NFS_INO_REVAL_PAGECACHE;
363 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
364 			inode->i_nlink = fattr->nlink;
365 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
366 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
367 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
368 			inode->i_uid = fattr->uid;
369 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
370 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
371 				| NFS_INO_INVALID_ACCESS
372 				| NFS_INO_INVALID_ACL;
373 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
374 			inode->i_gid = fattr->gid;
375 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
376 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
377 				| NFS_INO_INVALID_ACCESS
378 				| NFS_INO_INVALID_ACL;
379 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
380 			inode->i_blocks = fattr->du.nfs2.blocks;
381 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
382 			/*
383 			 * report the blocks in 512byte units
384 			 */
385 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
386 		}
387 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
388 		nfsi->attrtimeo_timestamp = now;
389 		nfsi->access_cache = RB_ROOT;
390 
391 		nfs_fscache_init_inode_cookie(inode);
392 
393 		unlock_new_inode(inode);
394 	} else
395 		nfs_refresh_inode(inode, fattr);
396 	dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
397 		inode->i_sb->s_id,
398 		(long long)NFS_FILEID(inode),
399 		atomic_read(&inode->i_count));
400 
401 out:
402 	return inode;
403 
404 out_no_inode:
405 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
406 	goto out;
407 }
408 
409 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
410 
411 int
412 nfs_setattr(struct dentry *dentry, struct iattr *attr)
413 {
414 	struct inode *inode = dentry->d_inode;
415 	struct nfs_fattr fattr;
416 	int error;
417 
418 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
419 
420 	/* skip mode change if it's just for clearing setuid/setgid */
421 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
422 		attr->ia_valid &= ~ATTR_MODE;
423 
424 	if (attr->ia_valid & ATTR_SIZE) {
425 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
426 			attr->ia_valid &= ~ATTR_SIZE;
427 	}
428 
429 	/* Optimization: if the end result is no change, don't RPC */
430 	attr->ia_valid &= NFS_VALID_ATTRS;
431 	if ((attr->ia_valid & ~ATTR_FILE) == 0)
432 		return 0;
433 
434 	/* Write all dirty data */
435 	if (S_ISREG(inode->i_mode)) {
436 		filemap_write_and_wait(inode->i_mapping);
437 		nfs_wb_all(inode);
438 	}
439 	/*
440 	 * Return any delegations if we're going to change ACLs
441 	 */
442 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
443 		nfs_inode_return_delegation(inode);
444 	error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
445 	if (error == 0)
446 		nfs_refresh_inode(inode, &fattr);
447 	return error;
448 }
449 
450 /**
451  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
452  * @inode: inode of the file used
453  * @offset: file offset to start truncating
454  *
455  * This is a copy of the common vmtruncate, but with the locking
456  * corrected to take into account the fact that NFS requires
457  * inode->i_size to be updated under the inode->i_lock.
458  */
459 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
460 {
461 	loff_t oldsize;
462 	int err;
463 
464 	err = inode_newsize_ok(inode, offset);
465 	if (err)
466 		goto out;
467 
468 	spin_lock(&inode->i_lock);
469 	oldsize = inode->i_size;
470 	i_size_write(inode, offset);
471 	spin_unlock(&inode->i_lock);
472 
473 	truncate_pagecache(inode, oldsize, offset);
474 out:
475 	return err;
476 }
477 
478 /**
479  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
480  * @inode: pointer to struct inode
481  * @attr: pointer to struct iattr
482  *
483  * Note: we do this in the *proc.c in order to ensure that
484  *       it works for things like exclusive creates too.
485  */
486 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
487 {
488 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
489 		spin_lock(&inode->i_lock);
490 		if ((attr->ia_valid & ATTR_MODE) != 0) {
491 			int mode = attr->ia_mode & S_IALLUGO;
492 			mode |= inode->i_mode & ~S_IALLUGO;
493 			inode->i_mode = mode;
494 		}
495 		if ((attr->ia_valid & ATTR_UID) != 0)
496 			inode->i_uid = attr->ia_uid;
497 		if ((attr->ia_valid & ATTR_GID) != 0)
498 			inode->i_gid = attr->ia_gid;
499 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
500 		spin_unlock(&inode->i_lock);
501 	}
502 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
503 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
504 		nfs_vmtruncate(inode, attr->ia_size);
505 	}
506 }
507 
508 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
509 {
510 	struct inode *inode = dentry->d_inode;
511 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
512 	int err;
513 
514 	/*
515 	 * Flush out writes to the server in order to update c/mtime.
516 	 *
517 	 * Hold the i_mutex to suspend application writes temporarily;
518 	 * this prevents long-running writing applications from blocking
519 	 * nfs_wb_nocommit.
520 	 */
521 	if (S_ISREG(inode->i_mode)) {
522 		mutex_lock(&inode->i_mutex);
523 		nfs_wb_nocommit(inode);
524 		mutex_unlock(&inode->i_mutex);
525 	}
526 
527 	/*
528 	 * We may force a getattr if the user cares about atime.
529 	 *
530 	 * Note that we only have to check the vfsmount flags here:
531 	 *  - NFS always sets S_NOATIME by so checking it would give a
532 	 *    bogus result
533 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
534 	 *    no point in checking those.
535 	 */
536  	if ((mnt->mnt_flags & MNT_NOATIME) ||
537  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
538 		need_atime = 0;
539 
540 	if (need_atime)
541 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
542 	else
543 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
544 	if (!err) {
545 		generic_fillattr(inode, stat);
546 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
547 	}
548 	return err;
549 }
550 
551 /**
552  * nfs_close_context - Common close_context() routine NFSv2/v3
553  * @ctx: pointer to context
554  * @is_sync: is this a synchronous close
555  *
556  * always ensure that the attributes are up to date if we're mounted
557  * with close-to-open semantics
558  */
559 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
560 {
561 	struct inode *inode;
562 	struct nfs_server *server;
563 
564 	if (!(ctx->mode & FMODE_WRITE))
565 		return;
566 	if (!is_sync)
567 		return;
568 	inode = ctx->path.dentry->d_inode;
569 	if (!list_empty(&NFS_I(inode)->open_files))
570 		return;
571 	server = NFS_SERVER(inode);
572 	if (server->flags & NFS_MOUNT_NOCTO)
573 		return;
574 	nfs_revalidate_inode(server, inode);
575 }
576 
577 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
578 {
579 	struct nfs_open_context *ctx;
580 
581 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
582 	if (ctx != NULL) {
583 		ctx->path.dentry = dget(dentry);
584 		ctx->path.mnt = mntget(mnt);
585 		ctx->cred = get_rpccred(cred);
586 		ctx->state = NULL;
587 		ctx->lockowner = current->files;
588 		ctx->flags = 0;
589 		ctx->error = 0;
590 		ctx->dir_cookie = 0;
591 		atomic_set(&ctx->count, 1);
592 	}
593 	return ctx;
594 }
595 
596 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
597 {
598 	if (ctx != NULL)
599 		atomic_inc(&ctx->count);
600 	return ctx;
601 }
602 
603 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
604 {
605 	struct inode *inode = ctx->path.dentry->d_inode;
606 
607 	if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
608 		return;
609 	list_del(&ctx->list);
610 	spin_unlock(&inode->i_lock);
611 	NFS_PROTO(inode)->close_context(ctx, is_sync);
612 	if (ctx->cred != NULL)
613 		put_rpccred(ctx->cred);
614 	path_put(&ctx->path);
615 	kfree(ctx);
616 }
617 
618 void put_nfs_open_context(struct nfs_open_context *ctx)
619 {
620 	__put_nfs_open_context(ctx, 0);
621 }
622 
623 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
624 {
625 	__put_nfs_open_context(ctx, 1);
626 }
627 
628 /*
629  * Ensure that mmap has a recent RPC credential for use when writing out
630  * shared pages
631  */
632 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
633 {
634 	struct inode *inode = filp->f_path.dentry->d_inode;
635 	struct nfs_inode *nfsi = NFS_I(inode);
636 
637 	filp->private_data = get_nfs_open_context(ctx);
638 	spin_lock(&inode->i_lock);
639 	list_add(&ctx->list, &nfsi->open_files);
640 	spin_unlock(&inode->i_lock);
641 }
642 
643 /*
644  * Given an inode, search for an open context with the desired characteristics
645  */
646 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
647 {
648 	struct nfs_inode *nfsi = NFS_I(inode);
649 	struct nfs_open_context *pos, *ctx = NULL;
650 
651 	spin_lock(&inode->i_lock);
652 	list_for_each_entry(pos, &nfsi->open_files, list) {
653 		if (cred != NULL && pos->cred != cred)
654 			continue;
655 		if ((pos->mode & mode) == mode) {
656 			ctx = get_nfs_open_context(pos);
657 			break;
658 		}
659 	}
660 	spin_unlock(&inode->i_lock);
661 	return ctx;
662 }
663 
664 static void nfs_file_clear_open_context(struct file *filp)
665 {
666 	struct inode *inode = filp->f_path.dentry->d_inode;
667 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
668 
669 	if (ctx) {
670 		filp->private_data = NULL;
671 		spin_lock(&inode->i_lock);
672 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
673 		spin_unlock(&inode->i_lock);
674 		put_nfs_open_context_sync(ctx);
675 	}
676 }
677 
678 /*
679  * These allocate and release file read/write context information.
680  */
681 int nfs_open(struct inode *inode, struct file *filp)
682 {
683 	struct nfs_open_context *ctx;
684 	struct rpc_cred *cred;
685 
686 	cred = rpc_lookup_cred();
687 	if (IS_ERR(cred))
688 		return PTR_ERR(cred);
689 	ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
690 	put_rpccred(cred);
691 	if (ctx == NULL)
692 		return -ENOMEM;
693 	ctx->mode = filp->f_mode;
694 	nfs_file_set_open_context(filp, ctx);
695 	put_nfs_open_context(ctx);
696 	nfs_fscache_set_inode_cookie(inode, filp);
697 	return 0;
698 }
699 
700 int nfs_release(struct inode *inode, struct file *filp)
701 {
702 	nfs_file_clear_open_context(filp);
703 	return 0;
704 }
705 
706 /*
707  * This function is called whenever some part of NFS notices that
708  * the cached attributes have to be refreshed.
709  */
710 int
711 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
712 {
713 	int		 status = -ESTALE;
714 	struct nfs_fattr fattr;
715 	struct nfs_inode *nfsi = NFS_I(inode);
716 
717 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
718 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
719 
720 	if (is_bad_inode(inode))
721 		goto out;
722 	if (NFS_STALE(inode))
723 		goto out;
724 
725 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
726 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
727 	if (status != 0) {
728 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
729 			 inode->i_sb->s_id,
730 			 (long long)NFS_FILEID(inode), status);
731 		if (status == -ESTALE) {
732 			nfs_zap_caches(inode);
733 			if (!S_ISDIR(inode->i_mode))
734 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
735 		}
736 		goto out;
737 	}
738 
739 	status = nfs_refresh_inode(inode, &fattr);
740 	if (status) {
741 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
742 			 inode->i_sb->s_id,
743 			 (long long)NFS_FILEID(inode), status);
744 		goto out;
745 	}
746 
747 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
748 		nfs_zap_acl_cache(inode);
749 
750 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
751 		inode->i_sb->s_id,
752 		(long long)NFS_FILEID(inode));
753 
754  out:
755 	return status;
756 }
757 
758 int nfs_attribute_timeout(struct inode *inode)
759 {
760 	struct nfs_inode *nfsi = NFS_I(inode);
761 
762 	if (nfs_have_delegation(inode, FMODE_READ))
763 		return 0;
764 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
765 }
766 
767 /**
768  * nfs_revalidate_inode - Revalidate the inode attributes
769  * @server - pointer to nfs_server struct
770  * @inode - pointer to inode struct
771  *
772  * Updates inode attribute information by retrieving the data from the server.
773  */
774 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
775 {
776 	if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
777 			&& !nfs_attribute_timeout(inode))
778 		return NFS_STALE(inode) ? -ESTALE : 0;
779 	return __nfs_revalidate_inode(server, inode);
780 }
781 
782 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
783 {
784 	struct nfs_inode *nfsi = NFS_I(inode);
785 
786 	if (mapping->nrpages != 0) {
787 		int ret = invalidate_inode_pages2(mapping);
788 		if (ret < 0)
789 			return ret;
790 	}
791 	spin_lock(&inode->i_lock);
792 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
793 	if (S_ISDIR(inode->i_mode))
794 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
795 	spin_unlock(&inode->i_lock);
796 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
797 	nfs_fscache_reset_inode_cookie(inode);
798 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
799 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
800 	return 0;
801 }
802 
803 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
804 {
805 	int ret = 0;
806 
807 	mutex_lock(&inode->i_mutex);
808 	if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
809 		ret = nfs_sync_mapping(mapping);
810 		if (ret == 0)
811 			ret = nfs_invalidate_mapping_nolock(inode, mapping);
812 	}
813 	mutex_unlock(&inode->i_mutex);
814 	return ret;
815 }
816 
817 /**
818  * nfs_revalidate_mapping_nolock - Revalidate the pagecache
819  * @inode - pointer to host inode
820  * @mapping - pointer to mapping
821  */
822 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
823 {
824 	struct nfs_inode *nfsi = NFS_I(inode);
825 	int ret = 0;
826 
827 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
828 			|| nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
829 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
830 		if (ret < 0)
831 			goto out;
832 	}
833 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
834 		ret = nfs_invalidate_mapping_nolock(inode, mapping);
835 out:
836 	return ret;
837 }
838 
839 /**
840  * nfs_revalidate_mapping - Revalidate the pagecache
841  * @inode - pointer to host inode
842  * @mapping - pointer to mapping
843  *
844  * This version of the function will take the inode->i_mutex and attempt to
845  * flush out all dirty data if it needs to invalidate the page cache.
846  */
847 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
848 {
849 	struct nfs_inode *nfsi = NFS_I(inode);
850 	int ret = 0;
851 
852 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
853 			|| nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
854 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
855 		if (ret < 0)
856 			goto out;
857 	}
858 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
859 		ret = nfs_invalidate_mapping(inode, mapping);
860 out:
861 	return ret;
862 }
863 
864 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
865 {
866 	struct nfs_inode *nfsi = NFS_I(inode);
867 
868 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
869 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
870 			&& nfsi->change_attr == fattr->pre_change_attr) {
871 		nfsi->change_attr = fattr->change_attr;
872 		if (S_ISDIR(inode->i_mode))
873 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
874 	}
875 	/* If we have atomic WCC data, we may update some attributes */
876 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
877 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
878 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
879 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
880 
881 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
882 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
883 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
884 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
885 			if (S_ISDIR(inode->i_mode))
886 				nfsi->cache_validity |= NFS_INO_INVALID_DATA;
887 	}
888 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
889 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
890 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
891 			&& nfsi->npages == 0)
892 			i_size_write(inode, nfs_size_to_loff_t(fattr->size));
893 }
894 
895 /**
896  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
897  * @inode - pointer to inode
898  * @fattr - updated attributes
899  *
900  * Verifies the attribute cache. If we have just changed the attributes,
901  * so that fattr carries weak cache consistency data, then it may
902  * also update the ctime/mtime/change_attribute.
903  */
904 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
905 {
906 	struct nfs_inode *nfsi = NFS_I(inode);
907 	loff_t cur_size, new_isize;
908 	unsigned long invalid = 0;
909 
910 
911 	/* Has the inode gone and changed behind our back? */
912 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
913 		return -EIO;
914 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
915 		return -EIO;
916 
917 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
918 			nfsi->change_attr != fattr->change_attr)
919 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
920 
921 	/* Verify a few of the more important attributes */
922 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
923 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
924 
925 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
926 		cur_size = i_size_read(inode);
927 		new_isize = nfs_size_to_loff_t(fattr->size);
928 		if (cur_size != new_isize && nfsi->npages == 0)
929 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
930 	}
931 
932 	/* Have any file permissions changed? */
933 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
934 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
935 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
936 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
937 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
938 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
939 
940 	/* Has the link count changed? */
941 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
942 		invalid |= NFS_INO_INVALID_ATTR;
943 
944 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
945 		invalid |= NFS_INO_INVALID_ATIME;
946 
947 	if (invalid != 0)
948 		nfsi->cache_validity |= invalid;
949 
950 	nfsi->read_cache_jiffies = fattr->time_start;
951 	return 0;
952 }
953 
954 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
955 {
956 	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
957 		return 0;
958 	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
959 }
960 
961 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
962 {
963 	if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
964 		return 0;
965 	return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
966 }
967 
968 static atomic_long_t nfs_attr_generation_counter;
969 
970 static unsigned long nfs_read_attr_generation_counter(void)
971 {
972 	return atomic_long_read(&nfs_attr_generation_counter);
973 }
974 
975 unsigned long nfs_inc_attr_generation_counter(void)
976 {
977 	return atomic_long_inc_return(&nfs_attr_generation_counter);
978 }
979 
980 void nfs_fattr_init(struct nfs_fattr *fattr)
981 {
982 	fattr->valid = 0;
983 	fattr->time_start = jiffies;
984 	fattr->gencount = nfs_inc_attr_generation_counter();
985 }
986 
987 /**
988  * nfs_inode_attrs_need_update - check if the inode attributes need updating
989  * @inode - pointer to inode
990  * @fattr - attributes
991  *
992  * Attempt to divine whether or not an RPC call reply carrying stale
993  * attributes got scheduled after another call carrying updated ones.
994  *
995  * To do so, the function first assumes that a more recent ctime means
996  * that the attributes in fattr are newer, however it also attempt to
997  * catch the case where ctime either didn't change, or went backwards
998  * (if someone reset the clock on the server) by looking at whether
999  * or not this RPC call was started after the inode was last updated.
1000  * Note also the check for wraparound of 'attr_gencount'
1001  *
1002  * The function returns 'true' if it thinks the attributes in 'fattr' are
1003  * more recent than the ones cached in the inode.
1004  *
1005  */
1006 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1007 {
1008 	const struct nfs_inode *nfsi = NFS_I(inode);
1009 
1010 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1011 		nfs_ctime_need_update(inode, fattr) ||
1012 		nfs_size_need_update(inode, fattr) ||
1013 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1014 }
1015 
1016 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1017 {
1018 	if (nfs_inode_attrs_need_update(inode, fattr))
1019 		return nfs_update_inode(inode, fattr);
1020 	return nfs_check_inode_attributes(inode, fattr);
1021 }
1022 
1023 /**
1024  * nfs_refresh_inode - try to update the inode attribute cache
1025  * @inode - pointer to inode
1026  * @fattr - updated attributes
1027  *
1028  * Check that an RPC call that returned attributes has not overlapped with
1029  * other recent updates of the inode metadata, then decide whether it is
1030  * safe to do a full update of the inode attributes, or whether just to
1031  * call nfs_check_inode_attributes.
1032  */
1033 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1034 {
1035 	int status;
1036 
1037 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1038 		return 0;
1039 	spin_lock(&inode->i_lock);
1040 	status = nfs_refresh_inode_locked(inode, fattr);
1041 	spin_unlock(&inode->i_lock);
1042 
1043 	return status;
1044 }
1045 
1046 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1047 {
1048 	struct nfs_inode *nfsi = NFS_I(inode);
1049 
1050 	nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1051 	if (S_ISDIR(inode->i_mode))
1052 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1053 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1054 		return 0;
1055 	return nfs_refresh_inode_locked(inode, fattr);
1056 }
1057 
1058 /**
1059  * nfs_post_op_update_inode - try to update the inode attribute cache
1060  * @inode - pointer to inode
1061  * @fattr - updated attributes
1062  *
1063  * After an operation that has changed the inode metadata, mark the
1064  * attribute cache as being invalid, then try to update it.
1065  *
1066  * NB: if the server didn't return any post op attributes, this
1067  * function will force the retrieval of attributes before the next
1068  * NFS request.  Thus it should be used only for operations that
1069  * are expected to change one or more attributes, to avoid
1070  * unnecessary NFS requests and trips through nfs_update_inode().
1071  */
1072 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1073 {
1074 	int status;
1075 
1076 	spin_lock(&inode->i_lock);
1077 	status = nfs_post_op_update_inode_locked(inode, fattr);
1078 	spin_unlock(&inode->i_lock);
1079 	return status;
1080 }
1081 
1082 /**
1083  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1084  * @inode - pointer to inode
1085  * @fattr - updated attributes
1086  *
1087  * After an operation that has changed the inode metadata, mark the
1088  * attribute cache as being invalid, then try to update it. Fake up
1089  * weak cache consistency data, if none exist.
1090  *
1091  * This function is mainly designed to be used by the ->write_done() functions.
1092  */
1093 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1094 {
1095 	int status;
1096 
1097 	spin_lock(&inode->i_lock);
1098 	/* Don't do a WCC update if these attributes are already stale */
1099 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1100 			!nfs_inode_attrs_need_update(inode, fattr)) {
1101 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1102 				| NFS_ATTR_FATTR_PRESIZE
1103 				| NFS_ATTR_FATTR_PREMTIME
1104 				| NFS_ATTR_FATTR_PRECTIME);
1105 		goto out_noforce;
1106 	}
1107 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1108 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1109 		fattr->pre_change_attr = NFS_I(inode)->change_attr;
1110 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1111 	}
1112 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1113 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1114 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1115 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1116 	}
1117 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1118 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1119 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1120 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1121 	}
1122 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1123 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1124 		fattr->pre_size = i_size_read(inode);
1125 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1126 	}
1127 out_noforce:
1128 	status = nfs_post_op_update_inode_locked(inode, fattr);
1129 	spin_unlock(&inode->i_lock);
1130 	return status;
1131 }
1132 
1133 /*
1134  * Many nfs protocol calls return the new file attributes after
1135  * an operation.  Here we update the inode to reflect the state
1136  * of the server's inode.
1137  *
1138  * This is a bit tricky because we have to make sure all dirty pages
1139  * have been sent off to the server before calling invalidate_inode_pages.
1140  * To make sure no other process adds more write requests while we try
1141  * our best to flush them, we make them sleep during the attribute refresh.
1142  *
1143  * A very similar scenario holds for the dir cache.
1144  */
1145 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1146 {
1147 	struct nfs_server *server;
1148 	struct nfs_inode *nfsi = NFS_I(inode);
1149 	loff_t cur_isize, new_isize;
1150 	unsigned long invalid = 0;
1151 	unsigned long now = jiffies;
1152 	unsigned long save_cache_validity;
1153 
1154 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1155 			__func__, inode->i_sb->s_id, inode->i_ino,
1156 			atomic_read(&inode->i_count), fattr->valid);
1157 
1158 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1159 		goto out_fileid;
1160 
1161 	/*
1162 	 * Make sure the inode's type hasn't changed.
1163 	 */
1164 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1165 		goto out_changed;
1166 
1167 	server = NFS_SERVER(inode);
1168 	/* Update the fsid? */
1169 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1170 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1171 			!test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
1172 		server->fsid = fattr->fsid;
1173 
1174 	/*
1175 	 * Update the read time so we don't revalidate too often.
1176 	 */
1177 	nfsi->read_cache_jiffies = fattr->time_start;
1178 
1179 	save_cache_validity = nfsi->cache_validity;
1180 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1181 			| NFS_INO_INVALID_ATIME
1182 			| NFS_INO_REVAL_FORCED
1183 			| NFS_INO_REVAL_PAGECACHE);
1184 
1185 	/* Do atomic weak cache consistency updates */
1186 	nfs_wcc_update_inode(inode, fattr);
1187 
1188 	/* More cache consistency checks */
1189 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1190 		if (nfsi->change_attr != fattr->change_attr) {
1191 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1192 					inode->i_sb->s_id, inode->i_ino);
1193 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1194 			if (S_ISDIR(inode->i_mode))
1195 				nfs_force_lookup_revalidate(inode);
1196 			nfsi->change_attr = fattr->change_attr;
1197 		}
1198 	} else if (server->caps & NFS_CAP_CHANGE_ATTR)
1199 		invalid |= save_cache_validity;
1200 
1201 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1202 		/* NFSv2/v3: Check if the mtime agrees */
1203 		if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1204 			dprintk("NFS: mtime change on server for file %s/%ld\n",
1205 					inode->i_sb->s_id, inode->i_ino);
1206 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1207 			if (S_ISDIR(inode->i_mode))
1208 				nfs_force_lookup_revalidate(inode);
1209 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1210 		}
1211 	} else if (server->caps & NFS_CAP_MTIME)
1212 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1213 				| NFS_INO_INVALID_DATA
1214 				| NFS_INO_REVAL_PAGECACHE
1215 				| NFS_INO_REVAL_FORCED);
1216 
1217 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1218 		/* If ctime has changed we should definitely clear access+acl caches */
1219 		if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1220 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1221 			/* and probably clear data for a directory too as utimes can cause
1222 			 * havoc with our cache.
1223 			 */
1224 			if (S_ISDIR(inode->i_mode)) {
1225 				invalid |= NFS_INO_INVALID_DATA;
1226 				nfs_force_lookup_revalidate(inode);
1227 			}
1228 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1229 		}
1230 	} else if (server->caps & NFS_CAP_CTIME)
1231 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1232 				| NFS_INO_INVALID_ACCESS
1233 				| NFS_INO_INVALID_ACL
1234 				| NFS_INO_REVAL_FORCED);
1235 
1236 	/* Check if our cached file size is stale */
1237 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1238 		new_isize = nfs_size_to_loff_t(fattr->size);
1239 		cur_isize = i_size_read(inode);
1240 		if (new_isize != cur_isize) {
1241 			/* Do we perhaps have any outstanding writes, or has
1242 			 * the file grown beyond our last write? */
1243 			if (nfsi->npages == 0 || new_isize > cur_isize) {
1244 				i_size_write(inode, new_isize);
1245 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1246 			}
1247 			dprintk("NFS: isize change on server for file %s/%ld\n",
1248 					inode->i_sb->s_id, inode->i_ino);
1249 		}
1250 	} else
1251 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1252 				| NFS_INO_REVAL_PAGECACHE
1253 				| NFS_INO_REVAL_FORCED);
1254 
1255 
1256 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1257 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1258 	else if (server->caps & NFS_CAP_ATIME)
1259 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1260 				| NFS_INO_REVAL_FORCED);
1261 
1262 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1263 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1264 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1265 			inode->i_mode = fattr->mode;
1266 		}
1267 	} else if (server->caps & NFS_CAP_MODE)
1268 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1269 				| NFS_INO_INVALID_ACCESS
1270 				| NFS_INO_INVALID_ACL
1271 				| NFS_INO_REVAL_FORCED);
1272 
1273 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1274 		if (inode->i_uid != fattr->uid) {
1275 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1276 			inode->i_uid = fattr->uid;
1277 		}
1278 	} else if (server->caps & NFS_CAP_OWNER)
1279 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1280 				| NFS_INO_INVALID_ACCESS
1281 				| NFS_INO_INVALID_ACL
1282 				| NFS_INO_REVAL_FORCED);
1283 
1284 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1285 		if (inode->i_gid != fattr->gid) {
1286 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1287 			inode->i_gid = fattr->gid;
1288 		}
1289 	} else if (server->caps & NFS_CAP_OWNER_GROUP)
1290 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1291 				| NFS_INO_INVALID_ACCESS
1292 				| NFS_INO_INVALID_ACL
1293 				| NFS_INO_REVAL_FORCED);
1294 
1295 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1296 		if (inode->i_nlink != fattr->nlink) {
1297 			invalid |= NFS_INO_INVALID_ATTR;
1298 			if (S_ISDIR(inode->i_mode))
1299 				invalid |= NFS_INO_INVALID_DATA;
1300 			inode->i_nlink = fattr->nlink;
1301 		}
1302 	} else if (server->caps & NFS_CAP_NLINK)
1303 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1304 				| NFS_INO_REVAL_FORCED);
1305 
1306 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1307 		/*
1308 		 * report the blocks in 512byte units
1309 		 */
1310 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1311  	}
1312 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1313 		inode->i_blocks = fattr->du.nfs2.blocks;
1314 
1315 	/* Update attrtimeo value if we're out of the unstable period */
1316 	if (invalid & NFS_INO_INVALID_ATTR) {
1317 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1318 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1319 		nfsi->attrtimeo_timestamp = now;
1320 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1321 	} else {
1322 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1323 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1324 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1325 			nfsi->attrtimeo_timestamp = now;
1326 		}
1327 	}
1328 	invalid &= ~NFS_INO_INVALID_ATTR;
1329 	/* Don't invalidate the data if we were to blame */
1330 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1331 				|| S_ISLNK(inode->i_mode)))
1332 		invalid &= ~NFS_INO_INVALID_DATA;
1333 	if (!nfs_have_delegation(inode, FMODE_READ) ||
1334 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1335 		nfsi->cache_validity |= invalid;
1336 
1337 	return 0;
1338  out_changed:
1339 	/*
1340 	 * Big trouble! The inode has become a different object.
1341 	 */
1342 	printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1343 			__func__, inode->i_ino, inode->i_mode, fattr->mode);
1344  out_err:
1345 	/*
1346 	 * No need to worry about unhashing the dentry, as the
1347 	 * lookup validation will know that the inode is bad.
1348 	 * (But we fall through to invalidate the caches.)
1349 	 */
1350 	nfs_invalidate_inode(inode);
1351 	return -ESTALE;
1352 
1353  out_fileid:
1354 	printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1355 		"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1356 		NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1357 		(long long)nfsi->fileid, (long long)fattr->fileid);
1358 	goto out_err;
1359 }
1360 
1361 
1362 #ifdef CONFIG_NFS_V4
1363 
1364 /*
1365  * Clean out any remaining NFSv4 state that might be left over due
1366  * to open() calls that passed nfs_atomic_lookup, but failed to call
1367  * nfs_open().
1368  */
1369 void nfs4_clear_inode(struct inode *inode)
1370 {
1371 	/* If we are holding a delegation, return it! */
1372 	nfs_inode_return_delegation_noreclaim(inode);
1373 	/* First call standard NFS clear_inode() code */
1374 	nfs_clear_inode(inode);
1375 }
1376 #endif
1377 
1378 struct inode *nfs_alloc_inode(struct super_block *sb)
1379 {
1380 	struct nfs_inode *nfsi;
1381 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1382 	if (!nfsi)
1383 		return NULL;
1384 	nfsi->flags = 0UL;
1385 	nfsi->cache_validity = 0UL;
1386 #ifdef CONFIG_NFS_V3_ACL
1387 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1388 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1389 #endif
1390 #ifdef CONFIG_NFS_V4
1391 	nfsi->nfs4_acl = NULL;
1392 #endif /* CONFIG_NFS_V4 */
1393 	return &nfsi->vfs_inode;
1394 }
1395 
1396 void nfs_destroy_inode(struct inode *inode)
1397 {
1398 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1399 }
1400 
1401 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1402 {
1403 #ifdef CONFIG_NFS_V4
1404 	INIT_LIST_HEAD(&nfsi->open_states);
1405 	nfsi->delegation = NULL;
1406 	nfsi->delegation_state = 0;
1407 	init_rwsem(&nfsi->rwsem);
1408 #endif
1409 }
1410 
1411 static void init_once(void *foo)
1412 {
1413 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1414 
1415 	inode_init_once(&nfsi->vfs_inode);
1416 	INIT_LIST_HEAD(&nfsi->open_files);
1417 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1418 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1419 	INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1420 	nfsi->npages = 0;
1421 	atomic_set(&nfsi->silly_count, 1);
1422 	INIT_HLIST_HEAD(&nfsi->silly_list);
1423 	init_waitqueue_head(&nfsi->waitqueue);
1424 	nfs4_init_once(nfsi);
1425 }
1426 
1427 static int __init nfs_init_inodecache(void)
1428 {
1429 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1430 					     sizeof(struct nfs_inode),
1431 					     0, (SLAB_RECLAIM_ACCOUNT|
1432 						SLAB_MEM_SPREAD),
1433 					     init_once);
1434 	if (nfs_inode_cachep == NULL)
1435 		return -ENOMEM;
1436 
1437 	return 0;
1438 }
1439 
1440 static void nfs_destroy_inodecache(void)
1441 {
1442 	kmem_cache_destroy(nfs_inode_cachep);
1443 }
1444 
1445 struct workqueue_struct *nfsiod_workqueue;
1446 
1447 /*
1448  * start up the nfsiod workqueue
1449  */
1450 static int nfsiod_start(void)
1451 {
1452 	struct workqueue_struct *wq;
1453 	dprintk("RPC:       creating workqueue nfsiod\n");
1454 	wq = create_singlethread_workqueue("nfsiod");
1455 	if (wq == NULL)
1456 		return -ENOMEM;
1457 	nfsiod_workqueue = wq;
1458 	return 0;
1459 }
1460 
1461 /*
1462  * Destroy the nfsiod workqueue
1463  */
1464 static void nfsiod_stop(void)
1465 {
1466 	struct workqueue_struct *wq;
1467 
1468 	wq = nfsiod_workqueue;
1469 	if (wq == NULL)
1470 		return;
1471 	nfsiod_workqueue = NULL;
1472 	destroy_workqueue(wq);
1473 }
1474 
1475 /*
1476  * Initialize NFS
1477  */
1478 static int __init init_nfs_fs(void)
1479 {
1480 	int err;
1481 
1482 	err = nfs_dns_resolver_init();
1483 	if (err < 0)
1484 		goto out8;
1485 
1486 	err = nfs_fscache_register();
1487 	if (err < 0)
1488 		goto out7;
1489 
1490 	err = nfsiod_start();
1491 	if (err)
1492 		goto out6;
1493 
1494 	err = nfs_fs_proc_init();
1495 	if (err)
1496 		goto out5;
1497 
1498 	err = nfs_init_nfspagecache();
1499 	if (err)
1500 		goto out4;
1501 
1502 	err = nfs_init_inodecache();
1503 	if (err)
1504 		goto out3;
1505 
1506 	err = nfs_init_readpagecache();
1507 	if (err)
1508 		goto out2;
1509 
1510 	err = nfs_init_writepagecache();
1511 	if (err)
1512 		goto out1;
1513 
1514 	err = nfs_init_directcache();
1515 	if (err)
1516 		goto out0;
1517 
1518 #ifdef CONFIG_PROC_FS
1519 	rpc_proc_register(&nfs_rpcstat);
1520 #endif
1521 	if ((err = register_nfs_fs()) != 0)
1522 		goto out;
1523 	return 0;
1524 out:
1525 #ifdef CONFIG_PROC_FS
1526 	rpc_proc_unregister("nfs");
1527 #endif
1528 	nfs_destroy_directcache();
1529 out0:
1530 	nfs_destroy_writepagecache();
1531 out1:
1532 	nfs_destroy_readpagecache();
1533 out2:
1534 	nfs_destroy_inodecache();
1535 out3:
1536 	nfs_destroy_nfspagecache();
1537 out4:
1538 	nfs_fs_proc_exit();
1539 out5:
1540 	nfsiod_stop();
1541 out6:
1542 	nfs_fscache_unregister();
1543 out7:
1544 	nfs_dns_resolver_destroy();
1545 out8:
1546 	return err;
1547 }
1548 
1549 static void __exit exit_nfs_fs(void)
1550 {
1551 	nfs_destroy_directcache();
1552 	nfs_destroy_writepagecache();
1553 	nfs_destroy_readpagecache();
1554 	nfs_destroy_inodecache();
1555 	nfs_destroy_nfspagecache();
1556 	nfs_fscache_unregister();
1557 	nfs_dns_resolver_destroy();
1558 #ifdef CONFIG_PROC_FS
1559 	rpc_proc_unregister("nfs");
1560 #endif
1561 	unregister_nfs_fs();
1562 	nfs_fs_proc_exit();
1563 	nfsiod_stop();
1564 }
1565 
1566 /* Not quite true; I just maintain it */
1567 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1568 MODULE_LICENSE("GPL");
1569 module_param(enable_ino64, bool, 0644);
1570 
1571 module_init(init_nfs_fs)
1572 module_exit(exit_nfs_fs)
1573