xref: /linux/fs/nfs/inode.c (revision a1087ef6abedf0bfd60e5e3fddf33192cb2c1325)
1 /*
2  *  linux/fs/nfs/inode.c
3  *
4  *  Copyright (C) 1992  Rick Sladkey
5  *
6  *  nfs inode and superblock handling functions
7  *
8  *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9  *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
10  *
11  *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12  *  J.S.Peatfield@damtp.cam.ac.uk
13  *
14  */
15 
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/time.h>
20 #include <linux/kernel.h>
21 #include <linux/mm.h>
22 #include <linux/string.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/unistd.h>
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/metrics.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/nfs_idmap.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 
41 #include <asm/system.h>
42 #include <asm/uaccess.h>
43 
44 #include "nfs4_fs.h"
45 #include "callback.h"
46 #include "delegation.h"
47 #include "iostat.h"
48 #include "internal.h"
49 #include "fscache.h"
50 #include "dns_resolve.h"
51 #include "pnfs.h"
52 
53 #define NFSDBG_FACILITY		NFSDBG_VFS
54 
55 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
56 
57 /* Default is to see 64-bit inode numbers */
58 static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
59 
60 static void nfs_invalidate_inode(struct inode *);
61 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
62 
63 static struct kmem_cache * nfs_inode_cachep;
64 
65 static inline unsigned long
66 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
67 {
68 	return nfs_fileid_to_ino_t(fattr->fileid);
69 }
70 
71 /**
72  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
73  * @word: long word containing the bit lock
74  */
75 int nfs_wait_bit_killable(void *word)
76 {
77 	if (fatal_signal_pending(current))
78 		return -ERESTARTSYS;
79 	schedule();
80 	return 0;
81 }
82 
83 /**
84  * nfs_compat_user_ino64 - returns the user-visible inode number
85  * @fileid: 64-bit fileid
86  *
87  * This function returns a 32-bit inode number if the boot parameter
88  * nfs.enable_ino64 is zero.
89  */
90 u64 nfs_compat_user_ino64(u64 fileid)
91 {
92 	int ino;
93 
94 	if (enable_ino64)
95 		return fileid;
96 	ino = fileid;
97 	if (sizeof(ino) < sizeof(fileid))
98 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
99 	return ino;
100 }
101 
102 static void nfs_clear_inode(struct inode *inode)
103 {
104 	/*
105 	 * The following should never happen...
106 	 */
107 	BUG_ON(nfs_have_writebacks(inode));
108 	BUG_ON(!list_empty(&NFS_I(inode)->open_files));
109 	nfs_zap_acl_cache(inode);
110 	nfs_access_zap_cache(inode);
111 	nfs_fscache_release_inode_cookie(inode);
112 }
113 
114 void nfs_evict_inode(struct inode *inode)
115 {
116 	truncate_inode_pages(&inode->i_data, 0);
117 	end_writeback(inode);
118 	nfs_clear_inode(inode);
119 }
120 
121 /**
122  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
123  */
124 int nfs_sync_mapping(struct address_space *mapping)
125 {
126 	int ret = 0;
127 
128 	if (mapping->nrpages != 0) {
129 		unmap_mapping_range(mapping, 0, 0, 0);
130 		ret = nfs_wb_all(mapping->host);
131 	}
132 	return ret;
133 }
134 
135 /*
136  * Invalidate the local caches
137  */
138 static void nfs_zap_caches_locked(struct inode *inode)
139 {
140 	struct nfs_inode *nfsi = NFS_I(inode);
141 	int mode = inode->i_mode;
142 
143 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
144 
145 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
146 	nfsi->attrtimeo_timestamp = jiffies;
147 
148 	memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
149 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
150 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
151 	else
152 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
153 }
154 
155 void nfs_zap_caches(struct inode *inode)
156 {
157 	spin_lock(&inode->i_lock);
158 	nfs_zap_caches_locked(inode);
159 	spin_unlock(&inode->i_lock);
160 }
161 
162 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
163 {
164 	if (mapping->nrpages != 0) {
165 		spin_lock(&inode->i_lock);
166 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
167 		spin_unlock(&inode->i_lock);
168 	}
169 }
170 
171 void nfs_zap_acl_cache(struct inode *inode)
172 {
173 	void (*clear_acl_cache)(struct inode *);
174 
175 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
176 	if (clear_acl_cache != NULL)
177 		clear_acl_cache(inode);
178 	spin_lock(&inode->i_lock);
179 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
180 	spin_unlock(&inode->i_lock);
181 }
182 
183 void nfs_invalidate_atime(struct inode *inode)
184 {
185 	spin_lock(&inode->i_lock);
186 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
187 	spin_unlock(&inode->i_lock);
188 }
189 
190 /*
191  * Invalidate, but do not unhash, the inode.
192  * NB: must be called with inode->i_lock held!
193  */
194 static void nfs_invalidate_inode(struct inode *inode)
195 {
196 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
197 	nfs_zap_caches_locked(inode);
198 }
199 
200 struct nfs_find_desc {
201 	struct nfs_fh		*fh;
202 	struct nfs_fattr	*fattr;
203 };
204 
205 /*
206  * In NFSv3 we can have 64bit inode numbers. In order to support
207  * this, and re-exported directories (also seen in NFSv2)
208  * we are forced to allow 2 different inodes to have the same
209  * i_ino.
210  */
211 static int
212 nfs_find_actor(struct inode *inode, void *opaque)
213 {
214 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
215 	struct nfs_fh		*fh = desc->fh;
216 	struct nfs_fattr	*fattr = desc->fattr;
217 
218 	if (NFS_FILEID(inode) != fattr->fileid)
219 		return 0;
220 	if (nfs_compare_fh(NFS_FH(inode), fh))
221 		return 0;
222 	if (is_bad_inode(inode) || NFS_STALE(inode))
223 		return 0;
224 	return 1;
225 }
226 
227 static int
228 nfs_init_locked(struct inode *inode, void *opaque)
229 {
230 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
231 	struct nfs_fattr	*fattr = desc->fattr;
232 
233 	set_nfs_fileid(inode, fattr->fileid);
234 	nfs_copy_fh(NFS_FH(inode), desc->fh);
235 	return 0;
236 }
237 
238 /*
239  * This is our front-end to iget that looks up inodes by file handle
240  * instead of inode number.
241  */
242 struct inode *
243 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
244 {
245 	struct nfs_find_desc desc = {
246 		.fh	= fh,
247 		.fattr	= fattr
248 	};
249 	struct inode *inode = ERR_PTR(-ENOENT);
250 	unsigned long hash;
251 
252 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
253 		goto out_no_inode;
254 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
255 		goto out_no_inode;
256 
257 	hash = nfs_fattr_to_ino_t(fattr);
258 
259 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
260 	if (inode == NULL) {
261 		inode = ERR_PTR(-ENOMEM);
262 		goto out_no_inode;
263 	}
264 
265 	if (inode->i_state & I_NEW) {
266 		struct nfs_inode *nfsi = NFS_I(inode);
267 		unsigned long now = jiffies;
268 
269 		/* We set i_ino for the few things that still rely on it,
270 		 * such as stat(2) */
271 		inode->i_ino = hash;
272 
273 		/* We can't support update_atime(), since the server will reset it */
274 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
275 		inode->i_mode = fattr->mode;
276 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
277 				&& nfs_server_capable(inode, NFS_CAP_MODE))
278 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
279 				| NFS_INO_INVALID_ACCESS
280 				| NFS_INO_INVALID_ACL;
281 		/* Why so? Because we want revalidate for devices/FIFOs, and
282 		 * that's precisely what we have in nfs_file_inode_operations.
283 		 */
284 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
285 		if (S_ISREG(inode->i_mode)) {
286 			inode->i_fop = &nfs_file_operations;
287 			inode->i_data.a_ops = &nfs_file_aops;
288 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
289 		} else if (S_ISDIR(inode->i_mode)) {
290 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
291 			inode->i_fop = &nfs_dir_operations;
292 			if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS))
293 				set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
294 			/* Deal with crossing mountpoints */
295 			if ((fattr->valid & NFS_ATTR_FATTR_FSID)
296 					&& !nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
297 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
298 					inode->i_op = &nfs_referral_inode_operations;
299 				else
300 					inode->i_op = &nfs_mountpoint_inode_operations;
301 				inode->i_fop = NULL;
302 				set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
303 			}
304 		} else if (S_ISLNK(inode->i_mode))
305 			inode->i_op = &nfs_symlink_inode_operations;
306 		else
307 			init_special_inode(inode, inode->i_mode, fattr->rdev);
308 
309 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
310 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
311 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
312 		nfsi->change_attr = 0;
313 		inode->i_size = 0;
314 		inode->i_nlink = 0;
315 		inode->i_uid = -2;
316 		inode->i_gid = -2;
317 		inode->i_blocks = 0;
318 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
319 
320 		nfsi->read_cache_jiffies = fattr->time_start;
321 		nfsi->attr_gencount = fattr->gencount;
322 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
323 			inode->i_atime = fattr->atime;
324 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
325 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
326 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
327 			inode->i_mtime = fattr->mtime;
328 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
329 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
330 				| NFS_INO_INVALID_DATA;
331 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
332 			inode->i_ctime = fattr->ctime;
333 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
334 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
335 				| NFS_INO_INVALID_ACCESS
336 				| NFS_INO_INVALID_ACL;
337 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
338 			nfsi->change_attr = fattr->change_attr;
339 		else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
340 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
341 				| NFS_INO_INVALID_DATA;
342 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
343 			inode->i_size = nfs_size_to_loff_t(fattr->size);
344 		else
345 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
346 				| NFS_INO_INVALID_DATA
347 				| NFS_INO_REVAL_PAGECACHE;
348 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
349 			inode->i_nlink = fattr->nlink;
350 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
351 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
352 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
353 			inode->i_uid = fattr->uid;
354 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
355 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
356 				| NFS_INO_INVALID_ACCESS
357 				| NFS_INO_INVALID_ACL;
358 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
359 			inode->i_gid = fattr->gid;
360 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
361 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
362 				| NFS_INO_INVALID_ACCESS
363 				| NFS_INO_INVALID_ACL;
364 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
365 			inode->i_blocks = fattr->du.nfs2.blocks;
366 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
367 			/*
368 			 * report the blocks in 512byte units
369 			 */
370 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
371 		}
372 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
373 		nfsi->attrtimeo_timestamp = now;
374 		nfsi->access_cache = RB_ROOT;
375 
376 		nfs_fscache_init_inode_cookie(inode);
377 
378 		unlock_new_inode(inode);
379 	} else
380 		nfs_refresh_inode(inode, fattr);
381 	dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
382 		inode->i_sb->s_id,
383 		(long long)NFS_FILEID(inode),
384 		atomic_read(&inode->i_count));
385 
386 out:
387 	return inode;
388 
389 out_no_inode:
390 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
391 	goto out;
392 }
393 
394 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
395 
396 int
397 nfs_setattr(struct dentry *dentry, struct iattr *attr)
398 {
399 	struct inode *inode = dentry->d_inode;
400 	struct nfs_fattr *fattr;
401 	int error = -ENOMEM;
402 
403 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
404 
405 	/* skip mode change if it's just for clearing setuid/setgid */
406 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
407 		attr->ia_valid &= ~ATTR_MODE;
408 
409 	if (attr->ia_valid & ATTR_SIZE) {
410 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
411 			attr->ia_valid &= ~ATTR_SIZE;
412 	}
413 
414 	/* Optimization: if the end result is no change, don't RPC */
415 	attr->ia_valid &= NFS_VALID_ATTRS;
416 	if ((attr->ia_valid & ~ATTR_FILE) == 0)
417 		return 0;
418 
419 	/* Write all dirty data */
420 	if (S_ISREG(inode->i_mode))
421 		nfs_wb_all(inode);
422 
423 	fattr = nfs_alloc_fattr();
424 	if (fattr == NULL)
425 		goto out;
426 	/*
427 	 * Return any delegations if we're going to change ACLs
428 	 */
429 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
430 		nfs_inode_return_delegation(inode);
431 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
432 	if (error == 0)
433 		nfs_refresh_inode(inode, fattr);
434 	nfs_free_fattr(fattr);
435 out:
436 	return error;
437 }
438 
439 /**
440  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
441  * @inode: inode of the file used
442  * @offset: file offset to start truncating
443  *
444  * This is a copy of the common vmtruncate, but with the locking
445  * corrected to take into account the fact that NFS requires
446  * inode->i_size to be updated under the inode->i_lock.
447  */
448 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
449 {
450 	loff_t oldsize;
451 	int err;
452 
453 	err = inode_newsize_ok(inode, offset);
454 	if (err)
455 		goto out;
456 
457 	spin_lock(&inode->i_lock);
458 	oldsize = inode->i_size;
459 	i_size_write(inode, offset);
460 	spin_unlock(&inode->i_lock);
461 
462 	truncate_pagecache(inode, oldsize, offset);
463 out:
464 	return err;
465 }
466 
467 /**
468  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
469  * @inode: pointer to struct inode
470  * @attr: pointer to struct iattr
471  *
472  * Note: we do this in the *proc.c in order to ensure that
473  *       it works for things like exclusive creates too.
474  */
475 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
476 {
477 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
478 		spin_lock(&inode->i_lock);
479 		if ((attr->ia_valid & ATTR_MODE) != 0) {
480 			int mode = attr->ia_mode & S_IALLUGO;
481 			mode |= inode->i_mode & ~S_IALLUGO;
482 			inode->i_mode = mode;
483 		}
484 		if ((attr->ia_valid & ATTR_UID) != 0)
485 			inode->i_uid = attr->ia_uid;
486 		if ((attr->ia_valid & ATTR_GID) != 0)
487 			inode->i_gid = attr->ia_gid;
488 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
489 		spin_unlock(&inode->i_lock);
490 	}
491 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
492 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
493 		nfs_vmtruncate(inode, attr->ia_size);
494 	}
495 }
496 
497 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
498 {
499 	struct inode *inode = dentry->d_inode;
500 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
501 	int err;
502 
503 	/* Flush out writes to the server in order to update c/mtime.  */
504 	if (S_ISREG(inode->i_mode)) {
505 		err = filemap_write_and_wait(inode->i_mapping);
506 		if (err)
507 			goto out;
508 	}
509 
510 	/*
511 	 * We may force a getattr if the user cares about atime.
512 	 *
513 	 * Note that we only have to check the vfsmount flags here:
514 	 *  - NFS always sets S_NOATIME by so checking it would give a
515 	 *    bogus result
516 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
517 	 *    no point in checking those.
518 	 */
519  	if ((mnt->mnt_flags & MNT_NOATIME) ||
520  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
521 		need_atime = 0;
522 
523 	if (need_atime)
524 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
525 	else
526 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
527 	if (!err) {
528 		generic_fillattr(inode, stat);
529 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
530 	}
531 out:
532 	return err;
533 }
534 
535 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
536 {
537 	atomic_set(&l_ctx->count, 1);
538 	l_ctx->lockowner = current->files;
539 	l_ctx->pid = current->tgid;
540 	INIT_LIST_HEAD(&l_ctx->list);
541 }
542 
543 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
544 {
545 	struct nfs_lock_context *pos;
546 
547 	list_for_each_entry(pos, &ctx->lock_context.list, list) {
548 		if (pos->lockowner != current->files)
549 			continue;
550 		if (pos->pid != current->tgid)
551 			continue;
552 		atomic_inc(&pos->count);
553 		return pos;
554 	}
555 	return NULL;
556 }
557 
558 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
559 {
560 	struct nfs_lock_context *res, *new = NULL;
561 	struct inode *inode = ctx->path.dentry->d_inode;
562 
563 	spin_lock(&inode->i_lock);
564 	res = __nfs_find_lock_context(ctx);
565 	if (res == NULL) {
566 		spin_unlock(&inode->i_lock);
567 		new = kmalloc(sizeof(*new), GFP_KERNEL);
568 		if (new == NULL)
569 			return NULL;
570 		nfs_init_lock_context(new);
571 		spin_lock(&inode->i_lock);
572 		res = __nfs_find_lock_context(ctx);
573 		if (res == NULL) {
574 			list_add_tail(&new->list, &ctx->lock_context.list);
575 			new->open_context = ctx;
576 			res = new;
577 			new = NULL;
578 		}
579 	}
580 	spin_unlock(&inode->i_lock);
581 	kfree(new);
582 	return res;
583 }
584 
585 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
586 {
587 	struct nfs_open_context *ctx = l_ctx->open_context;
588 	struct inode *inode = ctx->path.dentry->d_inode;
589 
590 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
591 		return;
592 	list_del(&l_ctx->list);
593 	spin_unlock(&inode->i_lock);
594 	kfree(l_ctx);
595 }
596 
597 /**
598  * nfs_close_context - Common close_context() routine NFSv2/v3
599  * @ctx: pointer to context
600  * @is_sync: is this a synchronous close
601  *
602  * always ensure that the attributes are up to date if we're mounted
603  * with close-to-open semantics
604  */
605 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
606 {
607 	struct inode *inode;
608 	struct nfs_server *server;
609 
610 	if (!(ctx->mode & FMODE_WRITE))
611 		return;
612 	if (!is_sync)
613 		return;
614 	inode = ctx->path.dentry->d_inode;
615 	if (!list_empty(&NFS_I(inode)->open_files))
616 		return;
617 	server = NFS_SERVER(inode);
618 	if (server->flags & NFS_MOUNT_NOCTO)
619 		return;
620 	nfs_revalidate_inode(server, inode);
621 }
622 
623 struct nfs_open_context *alloc_nfs_open_context(struct path *path, struct rpc_cred *cred, fmode_t f_mode)
624 {
625 	struct nfs_open_context *ctx;
626 
627 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
628 	if (ctx != NULL) {
629 		ctx->path = *path;
630 		path_get(&ctx->path);
631 		ctx->cred = get_rpccred(cred);
632 		ctx->state = NULL;
633 		ctx->mode = f_mode;
634 		ctx->flags = 0;
635 		ctx->error = 0;
636 		ctx->dir_cookie = 0;
637 		nfs_init_lock_context(&ctx->lock_context);
638 		ctx->lock_context.open_context = ctx;
639 		INIT_LIST_HEAD(&ctx->list);
640 	}
641 	return ctx;
642 }
643 
644 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
645 {
646 	if (ctx != NULL)
647 		atomic_inc(&ctx->lock_context.count);
648 	return ctx;
649 }
650 
651 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
652 {
653 	struct inode *inode = ctx->path.dentry->d_inode;
654 
655 	if (!list_empty(&ctx->list)) {
656 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
657 			return;
658 		list_del(&ctx->list);
659 		spin_unlock(&inode->i_lock);
660 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
661 		return;
662 	if (inode != NULL)
663 		NFS_PROTO(inode)->close_context(ctx, is_sync);
664 	if (ctx->cred != NULL)
665 		put_rpccred(ctx->cred);
666 	path_put(&ctx->path);
667 	kfree(ctx);
668 }
669 
670 void put_nfs_open_context(struct nfs_open_context *ctx)
671 {
672 	__put_nfs_open_context(ctx, 0);
673 }
674 
675 /*
676  * Ensure that mmap has a recent RPC credential for use when writing out
677  * shared pages
678  */
679 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
680 {
681 	struct inode *inode = filp->f_path.dentry->d_inode;
682 	struct nfs_inode *nfsi = NFS_I(inode);
683 
684 	filp->private_data = get_nfs_open_context(ctx);
685 	spin_lock(&inode->i_lock);
686 	list_add(&ctx->list, &nfsi->open_files);
687 	spin_unlock(&inode->i_lock);
688 }
689 
690 /*
691  * Given an inode, search for an open context with the desired characteristics
692  */
693 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
694 {
695 	struct nfs_inode *nfsi = NFS_I(inode);
696 	struct nfs_open_context *pos, *ctx = NULL;
697 
698 	spin_lock(&inode->i_lock);
699 	list_for_each_entry(pos, &nfsi->open_files, list) {
700 		if (cred != NULL && pos->cred != cred)
701 			continue;
702 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
703 			continue;
704 		ctx = get_nfs_open_context(pos);
705 		break;
706 	}
707 	spin_unlock(&inode->i_lock);
708 	return ctx;
709 }
710 
711 static void nfs_file_clear_open_context(struct file *filp)
712 {
713 	struct inode *inode = filp->f_path.dentry->d_inode;
714 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
715 
716 	if (ctx) {
717 		filp->private_data = NULL;
718 		spin_lock(&inode->i_lock);
719 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
720 		spin_unlock(&inode->i_lock);
721 		__put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
722 	}
723 }
724 
725 /*
726  * These allocate and release file read/write context information.
727  */
728 int nfs_open(struct inode *inode, struct file *filp)
729 {
730 	struct nfs_open_context *ctx;
731 	struct rpc_cred *cred;
732 
733 	cred = rpc_lookup_cred();
734 	if (IS_ERR(cred))
735 		return PTR_ERR(cred);
736 	ctx = alloc_nfs_open_context(&filp->f_path, cred, filp->f_mode);
737 	put_rpccred(cred);
738 	if (ctx == NULL)
739 		return -ENOMEM;
740 	nfs_file_set_open_context(filp, ctx);
741 	put_nfs_open_context(ctx);
742 	nfs_fscache_set_inode_cookie(inode, filp);
743 	return 0;
744 }
745 
746 int nfs_release(struct inode *inode, struct file *filp)
747 {
748 	nfs_file_clear_open_context(filp);
749 	return 0;
750 }
751 
752 /*
753  * This function is called whenever some part of NFS notices that
754  * the cached attributes have to be refreshed.
755  */
756 int
757 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
758 {
759 	int		 status = -ESTALE;
760 	struct nfs_fattr *fattr = NULL;
761 	struct nfs_inode *nfsi = NFS_I(inode);
762 
763 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
764 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
765 
766 	if (is_bad_inode(inode))
767 		goto out;
768 	if (NFS_STALE(inode))
769 		goto out;
770 
771 	status = -ENOMEM;
772 	fattr = nfs_alloc_fattr();
773 	if (fattr == NULL)
774 		goto out;
775 
776 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
777 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr);
778 	if (status != 0) {
779 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
780 			 inode->i_sb->s_id,
781 			 (long long)NFS_FILEID(inode), status);
782 		if (status == -ESTALE) {
783 			nfs_zap_caches(inode);
784 			if (!S_ISDIR(inode->i_mode))
785 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
786 		}
787 		goto out;
788 	}
789 
790 	status = nfs_refresh_inode(inode, fattr);
791 	if (status) {
792 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
793 			 inode->i_sb->s_id,
794 			 (long long)NFS_FILEID(inode), status);
795 		goto out;
796 	}
797 
798 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
799 		nfs_zap_acl_cache(inode);
800 
801 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
802 		inode->i_sb->s_id,
803 		(long long)NFS_FILEID(inode));
804 
805  out:
806 	nfs_free_fattr(fattr);
807 	return status;
808 }
809 
810 int nfs_attribute_timeout(struct inode *inode)
811 {
812 	struct nfs_inode *nfsi = NFS_I(inode);
813 
814 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
815 }
816 
817 static int nfs_attribute_cache_expired(struct inode *inode)
818 {
819 	if (nfs_have_delegated_attributes(inode))
820 		return 0;
821 	return nfs_attribute_timeout(inode);
822 }
823 
824 /**
825  * nfs_revalidate_inode - Revalidate the inode attributes
826  * @server - pointer to nfs_server struct
827  * @inode - pointer to inode struct
828  *
829  * Updates inode attribute information by retrieving the data from the server.
830  */
831 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
832 {
833 	if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
834 			&& !nfs_attribute_cache_expired(inode))
835 		return NFS_STALE(inode) ? -ESTALE : 0;
836 	return __nfs_revalidate_inode(server, inode);
837 }
838 
839 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
840 {
841 	struct nfs_inode *nfsi = NFS_I(inode);
842 
843 	if (mapping->nrpages != 0) {
844 		int ret = invalidate_inode_pages2(mapping);
845 		if (ret < 0)
846 			return ret;
847 	}
848 	spin_lock(&inode->i_lock);
849 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
850 	if (S_ISDIR(inode->i_mode))
851 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
852 	spin_unlock(&inode->i_lock);
853 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
854 	nfs_fscache_reset_inode_cookie(inode);
855 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
856 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
857 	return 0;
858 }
859 
860 /**
861  * nfs_revalidate_mapping - Revalidate the pagecache
862  * @inode - pointer to host inode
863  * @mapping - pointer to mapping
864  */
865 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
866 {
867 	struct nfs_inode *nfsi = NFS_I(inode);
868 	int ret = 0;
869 
870 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
871 			|| nfs_attribute_cache_expired(inode)
872 			|| NFS_STALE(inode)) {
873 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
874 		if (ret < 0)
875 			goto out;
876 	}
877 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
878 		ret = nfs_invalidate_mapping(inode, mapping);
879 out:
880 	return ret;
881 }
882 
883 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
884 {
885 	struct nfs_inode *nfsi = NFS_I(inode);
886 
887 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
888 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
889 			&& nfsi->change_attr == fattr->pre_change_attr) {
890 		nfsi->change_attr = fattr->change_attr;
891 		if (S_ISDIR(inode->i_mode))
892 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
893 	}
894 	/* If we have atomic WCC data, we may update some attributes */
895 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
896 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
897 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
898 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
899 
900 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
901 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
902 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
903 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
904 			if (S_ISDIR(inode->i_mode))
905 				nfsi->cache_validity |= NFS_INO_INVALID_DATA;
906 	}
907 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
908 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
909 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
910 			&& nfsi->npages == 0)
911 			i_size_write(inode, nfs_size_to_loff_t(fattr->size));
912 }
913 
914 /**
915  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
916  * @inode - pointer to inode
917  * @fattr - updated attributes
918  *
919  * Verifies the attribute cache. If we have just changed the attributes,
920  * so that fattr carries weak cache consistency data, then it may
921  * also update the ctime/mtime/change_attribute.
922  */
923 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
924 {
925 	struct nfs_inode *nfsi = NFS_I(inode);
926 	loff_t cur_size, new_isize;
927 	unsigned long invalid = 0;
928 
929 
930 	/* Has the inode gone and changed behind our back? */
931 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
932 		return -EIO;
933 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
934 		return -EIO;
935 
936 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
937 			nfsi->change_attr != fattr->change_attr)
938 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
939 
940 	/* Verify a few of the more important attributes */
941 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
942 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
943 
944 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
945 		cur_size = i_size_read(inode);
946 		new_isize = nfs_size_to_loff_t(fattr->size);
947 		if (cur_size != new_isize && nfsi->npages == 0)
948 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
949 	}
950 
951 	/* Have any file permissions changed? */
952 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
953 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
954 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
955 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
956 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
957 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
958 
959 	/* Has the link count changed? */
960 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
961 		invalid |= NFS_INO_INVALID_ATTR;
962 
963 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
964 		invalid |= NFS_INO_INVALID_ATIME;
965 
966 	if (invalid != 0)
967 		nfsi->cache_validity |= invalid;
968 
969 	nfsi->read_cache_jiffies = fattr->time_start;
970 	return 0;
971 }
972 
973 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
974 {
975 	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
976 		return 0;
977 	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
978 }
979 
980 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
981 {
982 	if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
983 		return 0;
984 	return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
985 }
986 
987 static atomic_long_t nfs_attr_generation_counter;
988 
989 static unsigned long nfs_read_attr_generation_counter(void)
990 {
991 	return atomic_long_read(&nfs_attr_generation_counter);
992 }
993 
994 unsigned long nfs_inc_attr_generation_counter(void)
995 {
996 	return atomic_long_inc_return(&nfs_attr_generation_counter);
997 }
998 
999 void nfs_fattr_init(struct nfs_fattr *fattr)
1000 {
1001 	fattr->valid = 0;
1002 	fattr->time_start = jiffies;
1003 	fattr->gencount = nfs_inc_attr_generation_counter();
1004 }
1005 
1006 struct nfs_fattr *nfs_alloc_fattr(void)
1007 {
1008 	struct nfs_fattr *fattr;
1009 
1010 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1011 	if (fattr != NULL)
1012 		nfs_fattr_init(fattr);
1013 	return fattr;
1014 }
1015 
1016 struct nfs_fh *nfs_alloc_fhandle(void)
1017 {
1018 	struct nfs_fh *fh;
1019 
1020 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1021 	if (fh != NULL)
1022 		fh->size = 0;
1023 	return fh;
1024 }
1025 
1026 /**
1027  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1028  * @inode - pointer to inode
1029  * @fattr - attributes
1030  *
1031  * Attempt to divine whether or not an RPC call reply carrying stale
1032  * attributes got scheduled after another call carrying updated ones.
1033  *
1034  * To do so, the function first assumes that a more recent ctime means
1035  * that the attributes in fattr are newer, however it also attempt to
1036  * catch the case where ctime either didn't change, or went backwards
1037  * (if someone reset the clock on the server) by looking at whether
1038  * or not this RPC call was started after the inode was last updated.
1039  * Note also the check for wraparound of 'attr_gencount'
1040  *
1041  * The function returns 'true' if it thinks the attributes in 'fattr' are
1042  * more recent than the ones cached in the inode.
1043  *
1044  */
1045 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1046 {
1047 	const struct nfs_inode *nfsi = NFS_I(inode);
1048 
1049 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1050 		nfs_ctime_need_update(inode, fattr) ||
1051 		nfs_size_need_update(inode, fattr) ||
1052 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1053 }
1054 
1055 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1056 {
1057 	if (nfs_inode_attrs_need_update(inode, fattr))
1058 		return nfs_update_inode(inode, fattr);
1059 	return nfs_check_inode_attributes(inode, fattr);
1060 }
1061 
1062 /**
1063  * nfs_refresh_inode - try to update the inode attribute cache
1064  * @inode - pointer to inode
1065  * @fattr - updated attributes
1066  *
1067  * Check that an RPC call that returned attributes has not overlapped with
1068  * other recent updates of the inode metadata, then decide whether it is
1069  * safe to do a full update of the inode attributes, or whether just to
1070  * call nfs_check_inode_attributes.
1071  */
1072 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1073 {
1074 	int status;
1075 
1076 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1077 		return 0;
1078 	spin_lock(&inode->i_lock);
1079 	status = nfs_refresh_inode_locked(inode, fattr);
1080 	spin_unlock(&inode->i_lock);
1081 
1082 	return status;
1083 }
1084 
1085 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1086 {
1087 	struct nfs_inode *nfsi = NFS_I(inode);
1088 
1089 	nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1090 	if (S_ISDIR(inode->i_mode))
1091 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1092 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1093 		return 0;
1094 	return nfs_refresh_inode_locked(inode, fattr);
1095 }
1096 
1097 /**
1098  * nfs_post_op_update_inode - try to update the inode attribute cache
1099  * @inode - pointer to inode
1100  * @fattr - updated attributes
1101  *
1102  * After an operation that has changed the inode metadata, mark the
1103  * attribute cache as being invalid, then try to update it.
1104  *
1105  * NB: if the server didn't return any post op attributes, this
1106  * function will force the retrieval of attributes before the next
1107  * NFS request.  Thus it should be used only for operations that
1108  * are expected to change one or more attributes, to avoid
1109  * unnecessary NFS requests and trips through nfs_update_inode().
1110  */
1111 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1112 {
1113 	int status;
1114 
1115 	spin_lock(&inode->i_lock);
1116 	status = nfs_post_op_update_inode_locked(inode, fattr);
1117 	spin_unlock(&inode->i_lock);
1118 	return status;
1119 }
1120 
1121 /**
1122  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1123  * @inode - pointer to inode
1124  * @fattr - updated attributes
1125  *
1126  * After an operation that has changed the inode metadata, mark the
1127  * attribute cache as being invalid, then try to update it. Fake up
1128  * weak cache consistency data, if none exist.
1129  *
1130  * This function is mainly designed to be used by the ->write_done() functions.
1131  */
1132 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1133 {
1134 	int status;
1135 
1136 	spin_lock(&inode->i_lock);
1137 	/* Don't do a WCC update if these attributes are already stale */
1138 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1139 			!nfs_inode_attrs_need_update(inode, fattr)) {
1140 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1141 				| NFS_ATTR_FATTR_PRESIZE
1142 				| NFS_ATTR_FATTR_PREMTIME
1143 				| NFS_ATTR_FATTR_PRECTIME);
1144 		goto out_noforce;
1145 	}
1146 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1147 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1148 		fattr->pre_change_attr = NFS_I(inode)->change_attr;
1149 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1150 	}
1151 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1152 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1153 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1154 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1155 	}
1156 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1157 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1158 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1159 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1160 	}
1161 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1162 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1163 		fattr->pre_size = i_size_read(inode);
1164 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1165 	}
1166 out_noforce:
1167 	status = nfs_post_op_update_inode_locked(inode, fattr);
1168 	spin_unlock(&inode->i_lock);
1169 	return status;
1170 }
1171 
1172 /*
1173  * Many nfs protocol calls return the new file attributes after
1174  * an operation.  Here we update the inode to reflect the state
1175  * of the server's inode.
1176  *
1177  * This is a bit tricky because we have to make sure all dirty pages
1178  * have been sent off to the server before calling invalidate_inode_pages.
1179  * To make sure no other process adds more write requests while we try
1180  * our best to flush them, we make them sleep during the attribute refresh.
1181  *
1182  * A very similar scenario holds for the dir cache.
1183  */
1184 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1185 {
1186 	struct nfs_server *server;
1187 	struct nfs_inode *nfsi = NFS_I(inode);
1188 	loff_t cur_isize, new_isize;
1189 	unsigned long invalid = 0;
1190 	unsigned long now = jiffies;
1191 	unsigned long save_cache_validity;
1192 
1193 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1194 			__func__, inode->i_sb->s_id, inode->i_ino,
1195 			atomic_read(&inode->i_count), fattr->valid);
1196 
1197 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1198 		goto out_fileid;
1199 
1200 	/*
1201 	 * Make sure the inode's type hasn't changed.
1202 	 */
1203 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1204 		goto out_changed;
1205 
1206 	server = NFS_SERVER(inode);
1207 	/* Update the fsid? */
1208 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1209 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1210 			!test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
1211 		server->fsid = fattr->fsid;
1212 
1213 	/*
1214 	 * Update the read time so we don't revalidate too often.
1215 	 */
1216 	nfsi->read_cache_jiffies = fattr->time_start;
1217 
1218 	save_cache_validity = nfsi->cache_validity;
1219 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1220 			| NFS_INO_INVALID_ATIME
1221 			| NFS_INO_REVAL_FORCED
1222 			| NFS_INO_REVAL_PAGECACHE);
1223 
1224 	/* Do atomic weak cache consistency updates */
1225 	nfs_wcc_update_inode(inode, fattr);
1226 
1227 	/* More cache consistency checks */
1228 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1229 		if (nfsi->change_attr != fattr->change_attr) {
1230 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1231 					inode->i_sb->s_id, inode->i_ino);
1232 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1233 			if (S_ISDIR(inode->i_mode))
1234 				nfs_force_lookup_revalidate(inode);
1235 			nfsi->change_attr = fattr->change_attr;
1236 		}
1237 	} else if (server->caps & NFS_CAP_CHANGE_ATTR)
1238 		invalid |= save_cache_validity;
1239 
1240 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1241 		/* NFSv2/v3: Check if the mtime agrees */
1242 		if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1243 			dprintk("NFS: mtime change on server for file %s/%ld\n",
1244 					inode->i_sb->s_id, inode->i_ino);
1245 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1246 			if (S_ISDIR(inode->i_mode))
1247 				nfs_force_lookup_revalidate(inode);
1248 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1249 		}
1250 	} else if (server->caps & NFS_CAP_MTIME)
1251 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1252 				| NFS_INO_INVALID_DATA
1253 				| NFS_INO_REVAL_PAGECACHE
1254 				| NFS_INO_REVAL_FORCED);
1255 
1256 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1257 		/* If ctime has changed we should definitely clear access+acl caches */
1258 		if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1259 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1260 			/* and probably clear data for a directory too as utimes can cause
1261 			 * havoc with our cache.
1262 			 */
1263 			if (S_ISDIR(inode->i_mode)) {
1264 				invalid |= NFS_INO_INVALID_DATA;
1265 				nfs_force_lookup_revalidate(inode);
1266 			}
1267 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1268 		}
1269 	} else if (server->caps & NFS_CAP_CTIME)
1270 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1271 				| NFS_INO_INVALID_ACCESS
1272 				| NFS_INO_INVALID_ACL
1273 				| NFS_INO_REVAL_FORCED);
1274 
1275 	/* Check if our cached file size is stale */
1276 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1277 		new_isize = nfs_size_to_loff_t(fattr->size);
1278 		cur_isize = i_size_read(inode);
1279 		if (new_isize != cur_isize) {
1280 			/* Do we perhaps have any outstanding writes, or has
1281 			 * the file grown beyond our last write? */
1282 			if (nfsi->npages == 0 || new_isize > cur_isize) {
1283 				i_size_write(inode, new_isize);
1284 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1285 			}
1286 			dprintk("NFS: isize change on server for file %s/%ld\n",
1287 					inode->i_sb->s_id, inode->i_ino);
1288 		}
1289 	} else
1290 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1291 				| NFS_INO_REVAL_PAGECACHE
1292 				| NFS_INO_REVAL_FORCED);
1293 
1294 
1295 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1296 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1297 	else if (server->caps & NFS_CAP_ATIME)
1298 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1299 				| NFS_INO_REVAL_FORCED);
1300 
1301 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1302 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1303 			umode_t newmode = inode->i_mode & S_IFMT;
1304 			newmode |= fattr->mode & S_IALLUGO;
1305 			inode->i_mode = newmode;
1306 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1307 		}
1308 	} else if (server->caps & NFS_CAP_MODE)
1309 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1310 				| NFS_INO_INVALID_ACCESS
1311 				| NFS_INO_INVALID_ACL
1312 				| NFS_INO_REVAL_FORCED);
1313 
1314 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1315 		if (inode->i_uid != fattr->uid) {
1316 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1317 			inode->i_uid = fattr->uid;
1318 		}
1319 	} else if (server->caps & NFS_CAP_OWNER)
1320 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1321 				| NFS_INO_INVALID_ACCESS
1322 				| NFS_INO_INVALID_ACL
1323 				| NFS_INO_REVAL_FORCED);
1324 
1325 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1326 		if (inode->i_gid != fattr->gid) {
1327 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1328 			inode->i_gid = fattr->gid;
1329 		}
1330 	} else if (server->caps & NFS_CAP_OWNER_GROUP)
1331 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1332 				| NFS_INO_INVALID_ACCESS
1333 				| NFS_INO_INVALID_ACL
1334 				| NFS_INO_REVAL_FORCED);
1335 
1336 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1337 		if (inode->i_nlink != fattr->nlink) {
1338 			invalid |= NFS_INO_INVALID_ATTR;
1339 			if (S_ISDIR(inode->i_mode))
1340 				invalid |= NFS_INO_INVALID_DATA;
1341 			inode->i_nlink = fattr->nlink;
1342 		}
1343 	} else if (server->caps & NFS_CAP_NLINK)
1344 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1345 				| NFS_INO_REVAL_FORCED);
1346 
1347 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1348 		/*
1349 		 * report the blocks in 512byte units
1350 		 */
1351 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1352  	}
1353 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1354 		inode->i_blocks = fattr->du.nfs2.blocks;
1355 
1356 	/* Update attrtimeo value if we're out of the unstable period */
1357 	if (invalid & NFS_INO_INVALID_ATTR) {
1358 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1359 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1360 		nfsi->attrtimeo_timestamp = now;
1361 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1362 	} else {
1363 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1364 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1365 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1366 			nfsi->attrtimeo_timestamp = now;
1367 		}
1368 	}
1369 	invalid &= ~NFS_INO_INVALID_ATTR;
1370 	/* Don't invalidate the data if we were to blame */
1371 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1372 				|| S_ISLNK(inode->i_mode)))
1373 		invalid &= ~NFS_INO_INVALID_DATA;
1374 	if (!nfs_have_delegation(inode, FMODE_READ) ||
1375 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1376 		nfsi->cache_validity |= invalid;
1377 
1378 	return 0;
1379  out_changed:
1380 	/*
1381 	 * Big trouble! The inode has become a different object.
1382 	 */
1383 	printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1384 			__func__, inode->i_ino, inode->i_mode, fattr->mode);
1385  out_err:
1386 	/*
1387 	 * No need to worry about unhashing the dentry, as the
1388 	 * lookup validation will know that the inode is bad.
1389 	 * (But we fall through to invalidate the caches.)
1390 	 */
1391 	nfs_invalidate_inode(inode);
1392 	return -ESTALE;
1393 
1394  out_fileid:
1395 	printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1396 		"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1397 		NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1398 		(long long)nfsi->fileid, (long long)fattr->fileid);
1399 	goto out_err;
1400 }
1401 
1402 
1403 #ifdef CONFIG_NFS_V4
1404 
1405 /*
1406  * Clean out any remaining NFSv4 state that might be left over due
1407  * to open() calls that passed nfs_atomic_lookup, but failed to call
1408  * nfs_open().
1409  */
1410 void nfs4_evict_inode(struct inode *inode)
1411 {
1412 	truncate_inode_pages(&inode->i_data, 0);
1413 	end_writeback(inode);
1414 	pnfs_destroy_layout(NFS_I(inode));
1415 	/* If we are holding a delegation, return it! */
1416 	nfs_inode_return_delegation_noreclaim(inode);
1417 	/* First call standard NFS clear_inode() code */
1418 	nfs_clear_inode(inode);
1419 }
1420 #endif
1421 
1422 struct inode *nfs_alloc_inode(struct super_block *sb)
1423 {
1424 	struct nfs_inode *nfsi;
1425 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1426 	if (!nfsi)
1427 		return NULL;
1428 	nfsi->flags = 0UL;
1429 	nfsi->cache_validity = 0UL;
1430 #ifdef CONFIG_NFS_V3_ACL
1431 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1432 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1433 #endif
1434 #ifdef CONFIG_NFS_V4
1435 	nfsi->nfs4_acl = NULL;
1436 #endif /* CONFIG_NFS_V4 */
1437 	return &nfsi->vfs_inode;
1438 }
1439 
1440 void nfs_destroy_inode(struct inode *inode)
1441 {
1442 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1443 }
1444 
1445 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1446 {
1447 #ifdef CONFIG_NFS_V4
1448 	INIT_LIST_HEAD(&nfsi->open_states);
1449 	nfsi->delegation = NULL;
1450 	nfsi->delegation_state = 0;
1451 	init_rwsem(&nfsi->rwsem);
1452 	nfsi->layout = NULL;
1453 #endif
1454 }
1455 
1456 static void init_once(void *foo)
1457 {
1458 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1459 
1460 	inode_init_once(&nfsi->vfs_inode);
1461 	INIT_LIST_HEAD(&nfsi->open_files);
1462 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1463 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1464 	INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1465 	nfsi->npages = 0;
1466 	nfsi->ncommit = 0;
1467 	atomic_set(&nfsi->silly_count, 1);
1468 	INIT_HLIST_HEAD(&nfsi->silly_list);
1469 	init_waitqueue_head(&nfsi->waitqueue);
1470 	nfs4_init_once(nfsi);
1471 }
1472 
1473 static int __init nfs_init_inodecache(void)
1474 {
1475 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1476 					     sizeof(struct nfs_inode),
1477 					     0, (SLAB_RECLAIM_ACCOUNT|
1478 						SLAB_MEM_SPREAD),
1479 					     init_once);
1480 	if (nfs_inode_cachep == NULL)
1481 		return -ENOMEM;
1482 
1483 	return 0;
1484 }
1485 
1486 static void nfs_destroy_inodecache(void)
1487 {
1488 	kmem_cache_destroy(nfs_inode_cachep);
1489 }
1490 
1491 struct workqueue_struct *nfsiod_workqueue;
1492 
1493 /*
1494  * start up the nfsiod workqueue
1495  */
1496 static int nfsiod_start(void)
1497 {
1498 	struct workqueue_struct *wq;
1499 	dprintk("RPC:       creating workqueue nfsiod\n");
1500 	wq = alloc_workqueue("nfsiod", WQ_RESCUER, 0);
1501 	if (wq == NULL)
1502 		return -ENOMEM;
1503 	nfsiod_workqueue = wq;
1504 	return 0;
1505 }
1506 
1507 /*
1508  * Destroy the nfsiod workqueue
1509  */
1510 static void nfsiod_stop(void)
1511 {
1512 	struct workqueue_struct *wq;
1513 
1514 	wq = nfsiod_workqueue;
1515 	if (wq == NULL)
1516 		return;
1517 	nfsiod_workqueue = NULL;
1518 	destroy_workqueue(wq);
1519 }
1520 
1521 /*
1522  * Initialize NFS
1523  */
1524 static int __init init_nfs_fs(void)
1525 {
1526 	int err;
1527 
1528 	err = nfs_idmap_init();
1529 	if (err < 0)
1530 		goto out9;
1531 
1532 	err = nfs_dns_resolver_init();
1533 	if (err < 0)
1534 		goto out8;
1535 
1536 	err = nfs_fscache_register();
1537 	if (err < 0)
1538 		goto out7;
1539 
1540 	err = nfsiod_start();
1541 	if (err)
1542 		goto out6;
1543 
1544 	err = nfs_fs_proc_init();
1545 	if (err)
1546 		goto out5;
1547 
1548 	err = nfs_init_nfspagecache();
1549 	if (err)
1550 		goto out4;
1551 
1552 	err = nfs_init_inodecache();
1553 	if (err)
1554 		goto out3;
1555 
1556 	err = nfs_init_readpagecache();
1557 	if (err)
1558 		goto out2;
1559 
1560 	err = nfs_init_writepagecache();
1561 	if (err)
1562 		goto out1;
1563 
1564 	err = nfs_init_directcache();
1565 	if (err)
1566 		goto out0;
1567 
1568 #ifdef CONFIG_PROC_FS
1569 	rpc_proc_register(&nfs_rpcstat);
1570 #endif
1571 	if ((err = register_nfs_fs()) != 0)
1572 		goto out;
1573 	return 0;
1574 out:
1575 #ifdef CONFIG_PROC_FS
1576 	rpc_proc_unregister("nfs");
1577 #endif
1578 	nfs_destroy_directcache();
1579 out0:
1580 	nfs_destroy_writepagecache();
1581 out1:
1582 	nfs_destroy_readpagecache();
1583 out2:
1584 	nfs_destroy_inodecache();
1585 out3:
1586 	nfs_destroy_nfspagecache();
1587 out4:
1588 	nfs_fs_proc_exit();
1589 out5:
1590 	nfsiod_stop();
1591 out6:
1592 	nfs_fscache_unregister();
1593 out7:
1594 	nfs_dns_resolver_destroy();
1595 out8:
1596 	nfs_idmap_quit();
1597 out9:
1598 	return err;
1599 }
1600 
1601 static void __exit exit_nfs_fs(void)
1602 {
1603 	nfs_destroy_directcache();
1604 	nfs_destroy_writepagecache();
1605 	nfs_destroy_readpagecache();
1606 	nfs_destroy_inodecache();
1607 	nfs_destroy_nfspagecache();
1608 	nfs_fscache_unregister();
1609 	nfs_dns_resolver_destroy();
1610 	nfs_idmap_quit();
1611 #ifdef CONFIG_PROC_FS
1612 	rpc_proc_unregister("nfs");
1613 #endif
1614 	unregister_nfs_fs();
1615 	nfs_fs_proc_exit();
1616 	nfsiod_stop();
1617 }
1618 
1619 /* Not quite true; I just maintain it */
1620 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1621 MODULE_LICENSE("GPL");
1622 module_param(enable_ino64, bool, 0644);
1623 
1624 module_init(init_nfs_fs)
1625 module_exit(exit_nfs_fs)
1626