xref: /linux/fs/nfs/inode.c (revision eb2bce7f5e7ac1ca6da434461217fadf3c688d2c)
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.Cox@linux.org>, 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 
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/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
38 #include <linux/inet.h>
39 #include <linux/nfs_xdr.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 
50 #define NFSDBG_FACILITY		NFSDBG_VFS
51 #define NFS_PARANOIA 1
52 
53 static void nfs_invalidate_inode(struct inode *);
54 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
55 
56 static void nfs_zap_acl_cache(struct inode *);
57 
58 static struct kmem_cache * nfs_inode_cachep;
59 
60 static inline unsigned long
61 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
62 {
63 	return nfs_fileid_to_ino_t(fattr->fileid);
64 }
65 
66 int nfs_write_inode(struct inode *inode, int sync)
67 {
68 	int ret;
69 
70 	if (sync) {
71 		ret = filemap_fdatawait(inode->i_mapping);
72 		if (ret == 0)
73 			ret = nfs_commit_inode(inode, FLUSH_SYNC);
74 	} else
75 		ret = nfs_commit_inode(inode, 0);
76 	if (ret >= 0)
77 		return 0;
78 	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
79 	return ret;
80 }
81 
82 void nfs_clear_inode(struct inode *inode)
83 {
84 	/*
85 	 * The following should never happen...
86 	 */
87 	BUG_ON(nfs_have_writebacks(inode));
88 	BUG_ON(!list_empty(&NFS_I(inode)->open_files));
89 	BUG_ON(atomic_read(&NFS_I(inode)->data_updates) != 0);
90 	nfs_zap_acl_cache(inode);
91 	nfs_access_zap_cache(inode);
92 }
93 
94 /**
95  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
96  */
97 int nfs_sync_mapping(struct address_space *mapping)
98 {
99 	int ret;
100 
101 	if (mapping->nrpages == 0)
102 		return 0;
103 	unmap_mapping_range(mapping, 0, 0, 0);
104 	ret = filemap_write_and_wait(mapping);
105 	if (ret != 0)
106 		goto out;
107 	ret = nfs_wb_all(mapping->host);
108 out:
109 	return ret;
110 }
111 
112 /*
113  * Invalidate the local caches
114  */
115 static void nfs_zap_caches_locked(struct inode *inode)
116 {
117 	struct nfs_inode *nfsi = NFS_I(inode);
118 	int mode = inode->i_mode;
119 
120 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
121 
122 	NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
123 	NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
124 
125 	memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
126 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
127 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
128 	else
129 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
130 }
131 
132 void nfs_zap_caches(struct inode *inode)
133 {
134 	spin_lock(&inode->i_lock);
135 	nfs_zap_caches_locked(inode);
136 	spin_unlock(&inode->i_lock);
137 }
138 
139 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
140 {
141 	if (mapping->nrpages != 0) {
142 		spin_lock(&inode->i_lock);
143 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
144 		spin_unlock(&inode->i_lock);
145 	}
146 }
147 
148 static void nfs_zap_acl_cache(struct inode *inode)
149 {
150 	void (*clear_acl_cache)(struct inode *);
151 
152 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
153 	if (clear_acl_cache != NULL)
154 		clear_acl_cache(inode);
155 	spin_lock(&inode->i_lock);
156 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
157 	spin_unlock(&inode->i_lock);
158 }
159 
160 /*
161  * Invalidate, but do not unhash, the inode.
162  * NB: must be called with inode->i_lock held!
163  */
164 static void nfs_invalidate_inode(struct inode *inode)
165 {
166 	set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
167 	nfs_zap_caches_locked(inode);
168 }
169 
170 struct nfs_find_desc {
171 	struct nfs_fh		*fh;
172 	struct nfs_fattr	*fattr;
173 };
174 
175 /*
176  * In NFSv3 we can have 64bit inode numbers. In order to support
177  * this, and re-exported directories (also seen in NFSv2)
178  * we are forced to allow 2 different inodes to have the same
179  * i_ino.
180  */
181 static int
182 nfs_find_actor(struct inode *inode, void *opaque)
183 {
184 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
185 	struct nfs_fh		*fh = desc->fh;
186 	struct nfs_fattr	*fattr = desc->fattr;
187 
188 	if (NFS_FILEID(inode) != fattr->fileid)
189 		return 0;
190 	if (nfs_compare_fh(NFS_FH(inode), fh))
191 		return 0;
192 	if (is_bad_inode(inode) || NFS_STALE(inode))
193 		return 0;
194 	return 1;
195 }
196 
197 static int
198 nfs_init_locked(struct inode *inode, void *opaque)
199 {
200 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
201 	struct nfs_fattr	*fattr = desc->fattr;
202 
203 	NFS_FILEID(inode) = fattr->fileid;
204 	nfs_copy_fh(NFS_FH(inode), desc->fh);
205 	return 0;
206 }
207 
208 /* Don't use READDIRPLUS on directories that we believe are too large */
209 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
210 
211 /*
212  * This is our front-end to iget that looks up inodes by file handle
213  * instead of inode number.
214  */
215 struct inode *
216 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
217 {
218 	struct nfs_find_desc desc = {
219 		.fh	= fh,
220 		.fattr	= fattr
221 	};
222 	struct inode *inode = ERR_PTR(-ENOENT);
223 	unsigned long hash;
224 
225 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
226 		goto out_no_inode;
227 
228 	if (!fattr->nlink) {
229 		printk("NFS: Buggy server - nlink == 0!\n");
230 		goto out_no_inode;
231 	}
232 
233 	hash = nfs_fattr_to_ino_t(fattr);
234 
235 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
236 	if (inode == NULL) {
237 		inode = ERR_PTR(-ENOMEM);
238 		goto out_no_inode;
239 	}
240 
241 	if (inode->i_state & I_NEW) {
242 		struct nfs_inode *nfsi = NFS_I(inode);
243 		unsigned long now = jiffies;
244 
245 		/* We set i_ino for the few things that still rely on it,
246 		 * such as stat(2) */
247 		inode->i_ino = hash;
248 
249 		/* We can't support update_atime(), since the server will reset it */
250 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
251 		inode->i_mode = fattr->mode;
252 		/* Why so? Because we want revalidate for devices/FIFOs, and
253 		 * that's precisely what we have in nfs_file_inode_operations.
254 		 */
255 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
256 		if (S_ISREG(inode->i_mode)) {
257 			inode->i_fop = &nfs_file_operations;
258 			inode->i_data.a_ops = &nfs_file_aops;
259 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
260 		} else if (S_ISDIR(inode->i_mode)) {
261 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
262 			inode->i_fop = &nfs_dir_operations;
263 			if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
264 			    && fattr->size <= NFS_LIMIT_READDIRPLUS)
265 				set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
266 			/* Deal with crossing mountpoints */
267 			if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
268 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
269 					inode->i_op = &nfs_referral_inode_operations;
270 				else
271 					inode->i_op = &nfs_mountpoint_inode_operations;
272 				inode->i_fop = NULL;
273 			}
274 		} else if (S_ISLNK(inode->i_mode))
275 			inode->i_op = &nfs_symlink_inode_operations;
276 		else
277 			init_special_inode(inode, inode->i_mode, fattr->rdev);
278 
279 		nfsi->read_cache_jiffies = fattr->time_start;
280 		nfsi->last_updated = now;
281 		nfsi->cache_change_attribute = now;
282 		inode->i_atime = fattr->atime;
283 		inode->i_mtime = fattr->mtime;
284 		inode->i_ctime = fattr->ctime;
285 		if (fattr->valid & NFS_ATTR_FATTR_V4)
286 			nfsi->change_attr = fattr->change_attr;
287 		inode->i_size = nfs_size_to_loff_t(fattr->size);
288 		inode->i_nlink = fattr->nlink;
289 		inode->i_uid = fattr->uid;
290 		inode->i_gid = fattr->gid;
291 		if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
292 			/*
293 			 * report the blocks in 512byte units
294 			 */
295 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
296 		} else {
297 			inode->i_blocks = fattr->du.nfs2.blocks;
298 		}
299 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
300 		nfsi->attrtimeo_timestamp = now;
301 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
302 		nfsi->access_cache = RB_ROOT;
303 
304 		unlock_new_inode(inode);
305 	} else
306 		nfs_refresh_inode(inode, fattr);
307 	dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
308 		inode->i_sb->s_id,
309 		(long long)NFS_FILEID(inode),
310 		atomic_read(&inode->i_count));
311 
312 out:
313 	return inode;
314 
315 out_no_inode:
316 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
317 	goto out;
318 }
319 
320 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
321 
322 int
323 nfs_setattr(struct dentry *dentry, struct iattr *attr)
324 {
325 	struct inode *inode = dentry->d_inode;
326 	struct nfs_fattr fattr;
327 	int error;
328 
329 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
330 
331 	if (attr->ia_valid & ATTR_SIZE) {
332 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
333 			attr->ia_valid &= ~ATTR_SIZE;
334 	}
335 
336 	/* Optimization: if the end result is no change, don't RPC */
337 	attr->ia_valid &= NFS_VALID_ATTRS;
338 	if (attr->ia_valid == 0)
339 		return 0;
340 
341 	lock_kernel();
342 	nfs_begin_data_update(inode);
343 	/* Write all dirty data */
344 	if (S_ISREG(inode->i_mode)) {
345 		filemap_write_and_wait(inode->i_mapping);
346 		nfs_wb_all(inode);
347 	}
348 	/*
349 	 * Return any delegations if we're going to change ACLs
350 	 */
351 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
352 		nfs_inode_return_delegation(inode);
353 	error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
354 	if (error == 0)
355 		nfs_refresh_inode(inode, &fattr);
356 	nfs_end_data_update(inode);
357 	unlock_kernel();
358 	return error;
359 }
360 
361 /**
362  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
363  * @inode: pointer to struct inode
364  * @attr: pointer to struct iattr
365  *
366  * Note: we do this in the *proc.c in order to ensure that
367  *       it works for things like exclusive creates too.
368  */
369 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
370 {
371 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
372 		if ((attr->ia_valid & ATTR_MODE) != 0) {
373 			int mode = attr->ia_mode & S_IALLUGO;
374 			mode |= inode->i_mode & ~S_IALLUGO;
375 			inode->i_mode = mode;
376 		}
377 		if ((attr->ia_valid & ATTR_UID) != 0)
378 			inode->i_uid = attr->ia_uid;
379 		if ((attr->ia_valid & ATTR_GID) != 0)
380 			inode->i_gid = attr->ia_gid;
381 		spin_lock(&inode->i_lock);
382 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
383 		spin_unlock(&inode->i_lock);
384 	}
385 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
386 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
387 		inode->i_size = attr->ia_size;
388 		vmtruncate(inode, attr->ia_size);
389 	}
390 }
391 
392 static int nfs_wait_schedule(void *word)
393 {
394 	if (signal_pending(current))
395 		return -ERESTARTSYS;
396 	schedule();
397 	return 0;
398 }
399 
400 /*
401  * Wait for the inode to get unlocked.
402  */
403 static int nfs_wait_on_inode(struct inode *inode)
404 {
405 	struct rpc_clnt	*clnt = NFS_CLIENT(inode);
406 	struct nfs_inode *nfsi = NFS_I(inode);
407 	sigset_t oldmask;
408 	int error;
409 
410 	rpc_clnt_sigmask(clnt, &oldmask);
411 	error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
412 					nfs_wait_schedule, TASK_INTERRUPTIBLE);
413 	rpc_clnt_sigunmask(clnt, &oldmask);
414 
415 	return error;
416 }
417 
418 static void nfs_wake_up_inode(struct inode *inode)
419 {
420 	struct nfs_inode *nfsi = NFS_I(inode);
421 
422 	clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
423 	smp_mb__after_clear_bit();
424 	wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
425 }
426 
427 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
428 {
429 	struct inode *inode = dentry->d_inode;
430 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
431 	int err;
432 
433 	/* Flush out writes to the server in order to update c/mtime */
434 	if (S_ISREG(inode->i_mode))
435 		nfs_sync_mapping_range(inode->i_mapping, 0, 0, FLUSH_NOCOMMIT);
436 
437 	/*
438 	 * We may force a getattr if the user cares about atime.
439 	 *
440 	 * Note that we only have to check the vfsmount flags here:
441 	 *  - NFS always sets S_NOATIME by so checking it would give a
442 	 *    bogus result
443 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
444 	 *    no point in checking those.
445 	 */
446  	if ((mnt->mnt_flags & MNT_NOATIME) ||
447  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
448 		need_atime = 0;
449 
450 	if (need_atime)
451 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
452 	else
453 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
454 	if (!err)
455 		generic_fillattr(inode, stat);
456 	return err;
457 }
458 
459 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
460 {
461 	struct nfs_open_context *ctx;
462 
463 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
464 	if (ctx != NULL) {
465 		atomic_set(&ctx->count, 1);
466 		ctx->dentry = dget(dentry);
467 		ctx->vfsmnt = mntget(mnt);
468 		ctx->cred = get_rpccred(cred);
469 		ctx->state = NULL;
470 		ctx->lockowner = current->files;
471 		ctx->error = 0;
472 		ctx->dir_cookie = 0;
473 	}
474 	return ctx;
475 }
476 
477 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
478 {
479 	if (ctx != NULL)
480 		atomic_inc(&ctx->count);
481 	return ctx;
482 }
483 
484 void put_nfs_open_context(struct nfs_open_context *ctx)
485 {
486 	if (atomic_dec_and_test(&ctx->count)) {
487 		if (!list_empty(&ctx->list)) {
488 			struct inode *inode = ctx->dentry->d_inode;
489 			spin_lock(&inode->i_lock);
490 			list_del(&ctx->list);
491 			spin_unlock(&inode->i_lock);
492 		}
493 		if (ctx->state != NULL)
494 			nfs4_close_state(ctx->state, ctx->mode);
495 		if (ctx->cred != NULL)
496 			put_rpccred(ctx->cred);
497 		dput(ctx->dentry);
498 		mntput(ctx->vfsmnt);
499 		kfree(ctx);
500 	}
501 }
502 
503 /*
504  * Ensure that mmap has a recent RPC credential for use when writing out
505  * shared pages
506  */
507 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
508 {
509 	struct inode *inode = filp->f_path.dentry->d_inode;
510 	struct nfs_inode *nfsi = NFS_I(inode);
511 
512 	filp->private_data = get_nfs_open_context(ctx);
513 	spin_lock(&inode->i_lock);
514 	list_add(&ctx->list, &nfsi->open_files);
515 	spin_unlock(&inode->i_lock);
516 }
517 
518 /*
519  * Given an inode, search for an open context with the desired characteristics
520  */
521 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
522 {
523 	struct nfs_inode *nfsi = NFS_I(inode);
524 	struct nfs_open_context *pos, *ctx = NULL;
525 
526 	spin_lock(&inode->i_lock);
527 	list_for_each_entry(pos, &nfsi->open_files, list) {
528 		if (cred != NULL && pos->cred != cred)
529 			continue;
530 		if ((pos->mode & mode) == mode) {
531 			ctx = get_nfs_open_context(pos);
532 			break;
533 		}
534 	}
535 	spin_unlock(&inode->i_lock);
536 	return ctx;
537 }
538 
539 static void nfs_file_clear_open_context(struct file *filp)
540 {
541 	struct inode *inode = filp->f_path.dentry->d_inode;
542 	struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
543 
544 	if (ctx) {
545 		filp->private_data = NULL;
546 		spin_lock(&inode->i_lock);
547 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
548 		spin_unlock(&inode->i_lock);
549 		put_nfs_open_context(ctx);
550 	}
551 }
552 
553 /*
554  * These allocate and release file read/write context information.
555  */
556 int nfs_open(struct inode *inode, struct file *filp)
557 {
558 	struct nfs_open_context *ctx;
559 	struct rpc_cred *cred;
560 
561 	cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
562 	if (IS_ERR(cred))
563 		return PTR_ERR(cred);
564 	ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
565 	put_rpccred(cred);
566 	if (ctx == NULL)
567 		return -ENOMEM;
568 	ctx->mode = filp->f_mode;
569 	nfs_file_set_open_context(filp, ctx);
570 	put_nfs_open_context(ctx);
571 	return 0;
572 }
573 
574 int nfs_release(struct inode *inode, struct file *filp)
575 {
576 	nfs_file_clear_open_context(filp);
577 	return 0;
578 }
579 
580 /*
581  * This function is called whenever some part of NFS notices that
582  * the cached attributes have to be refreshed.
583  */
584 int
585 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
586 {
587 	int		 status = -ESTALE;
588 	struct nfs_fattr fattr;
589 	struct nfs_inode *nfsi = NFS_I(inode);
590 
591 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
592 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
593 
594 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
595 	lock_kernel();
596 	if (is_bad_inode(inode))
597  		goto out_nowait;
598 	if (NFS_STALE(inode))
599  		goto out_nowait;
600 
601 	status = nfs_wait_on_inode(inode);
602 	if (status < 0)
603 		goto out;
604 	if (NFS_STALE(inode)) {
605 		status = -ESTALE;
606 		/* Do we trust the cached ESTALE? */
607 		if (NFS_ATTRTIMEO(inode) != 0) {
608 			if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME)) {
609 				/* no */
610 			} else
611 				goto out;
612 		}
613 	}
614 
615 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
616 	if (status != 0) {
617 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
618 			 inode->i_sb->s_id,
619 			 (long long)NFS_FILEID(inode), status);
620 		if (status == -ESTALE) {
621 			nfs_zap_caches(inode);
622 			if (!S_ISDIR(inode->i_mode))
623 				set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
624 		}
625 		goto out;
626 	}
627 
628 	spin_lock(&inode->i_lock);
629 	status = nfs_update_inode(inode, &fattr);
630 	if (status) {
631 		spin_unlock(&inode->i_lock);
632 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
633 			 inode->i_sb->s_id,
634 			 (long long)NFS_FILEID(inode), status);
635 		goto out;
636 	}
637 	spin_unlock(&inode->i_lock);
638 
639 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
640 		nfs_zap_acl_cache(inode);
641 
642 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
643 		inode->i_sb->s_id,
644 		(long long)NFS_FILEID(inode));
645 
646  out:
647 	nfs_wake_up_inode(inode);
648 
649  out_nowait:
650 	unlock_kernel();
651 	return status;
652 }
653 
654 int nfs_attribute_timeout(struct inode *inode)
655 {
656 	struct nfs_inode *nfsi = NFS_I(inode);
657 
658 	if (nfs_have_delegation(inode, FMODE_READ))
659 		return 0;
660 	return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
661 }
662 
663 /**
664  * nfs_revalidate_inode - Revalidate the inode attributes
665  * @server - pointer to nfs_server struct
666  * @inode - pointer to inode struct
667  *
668  * Updates inode attribute information by retrieving the data from the server.
669  */
670 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
671 {
672 	if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
673 			&& !nfs_attribute_timeout(inode))
674 		return NFS_STALE(inode) ? -ESTALE : 0;
675 	return __nfs_revalidate_inode(server, inode);
676 }
677 
678 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
679 {
680 	struct nfs_inode *nfsi = NFS_I(inode);
681 
682 	if (mapping->nrpages != 0) {
683 		int ret = invalidate_inode_pages2(mapping);
684 		if (ret < 0)
685 			return ret;
686 	}
687 	spin_lock(&inode->i_lock);
688 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
689 	if (S_ISDIR(inode->i_mode)) {
690 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
691 		/* This ensures we revalidate child dentries */
692 		nfsi->cache_change_attribute = jiffies;
693 	}
694 	spin_unlock(&inode->i_lock);
695 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
696 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
697 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
698 	return 0;
699 }
700 
701 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
702 {
703 	int ret = 0;
704 
705 	mutex_lock(&inode->i_mutex);
706 	if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
707 		ret = nfs_sync_mapping(mapping);
708 		if (ret == 0)
709 			ret = nfs_invalidate_mapping_nolock(inode, mapping);
710 	}
711 	mutex_unlock(&inode->i_mutex);
712 	return ret;
713 }
714 
715 /**
716  * nfs_revalidate_mapping_nolock - Revalidate the pagecache
717  * @inode - pointer to host inode
718  * @mapping - pointer to mapping
719  */
720 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
721 {
722 	struct nfs_inode *nfsi = NFS_I(inode);
723 	int ret = 0;
724 
725 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
726 			|| nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
727 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
728 		if (ret < 0)
729 			goto out;
730 	}
731 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
732 		ret = nfs_invalidate_mapping_nolock(inode, mapping);
733 out:
734 	return ret;
735 }
736 
737 /**
738  * nfs_revalidate_mapping - Revalidate the pagecache
739  * @inode - pointer to host inode
740  * @mapping - pointer to mapping
741  *
742  * This version of the function will take the inode->i_mutex and attempt to
743  * flush out all dirty data if it needs to invalidate the page cache.
744  */
745 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
746 {
747 	struct nfs_inode *nfsi = NFS_I(inode);
748 	int ret = 0;
749 
750 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
751 			|| nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
752 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
753 		if (ret < 0)
754 			goto out;
755 	}
756 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
757 		ret = nfs_invalidate_mapping(inode, mapping);
758 out:
759 	return ret;
760 }
761 
762 /**
763  * nfs_begin_data_update
764  * @inode - pointer to inode
765  * Declare that a set of operations will update file data on the server
766  */
767 void nfs_begin_data_update(struct inode *inode)
768 {
769 	atomic_inc(&NFS_I(inode)->data_updates);
770 }
771 
772 /**
773  * nfs_end_data_update
774  * @inode - pointer to inode
775  * Declare end of the operations that will update file data
776  * This will mark the inode as immediately needing revalidation
777  * of its attribute cache.
778  */
779 void nfs_end_data_update(struct inode *inode)
780 {
781 	struct nfs_inode *nfsi = NFS_I(inode);
782 
783 	/* Directories: invalidate page cache */
784 	if (S_ISDIR(inode->i_mode)) {
785 		spin_lock(&inode->i_lock);
786 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
787 		spin_unlock(&inode->i_lock);
788 	}
789 	nfsi->cache_change_attribute = jiffies;
790 	atomic_dec(&nfsi->data_updates);
791 }
792 
793 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
794 {
795 	struct nfs_inode *nfsi = NFS_I(inode);
796 	unsigned long now = jiffies;
797 
798 	/* If we have atomic WCC data, we may update some attributes */
799 	if ((fattr->valid & NFS_ATTR_WCC) != 0) {
800 		if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
801 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
802 			nfsi->cache_change_attribute = now;
803 		}
804 		if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
805 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
806 			nfsi->cache_change_attribute = now;
807 		}
808 		if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
809 			inode->i_size = fattr->size;
810 			nfsi->cache_change_attribute = now;
811 		}
812 	}
813 }
814 
815 /**
816  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
817  * @inode - pointer to inode
818  * @fattr - updated attributes
819  *
820  * Verifies the attribute cache. If we have just changed the attributes,
821  * so that fattr carries weak cache consistency data, then it may
822  * also update the ctime/mtime/change_attribute.
823  */
824 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
825 {
826 	struct nfs_inode *nfsi = NFS_I(inode);
827 	loff_t cur_size, new_isize;
828 	int data_unstable;
829 
830 
831 	/* Has the inode gone and changed behind our back? */
832 	if (nfsi->fileid != fattr->fileid
833 			|| (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
834 		return -EIO;
835 	}
836 
837 	/* Are we in the process of updating data on the server? */
838 	data_unstable = nfs_caches_unstable(inode);
839 
840 	/* Do atomic weak cache consistency updates */
841 	nfs_wcc_update_inode(inode, fattr);
842 
843 	if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
844 			nfsi->change_attr != fattr->change_attr)
845 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
846 
847 	/* Verify a few of the more important attributes */
848 	if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
849 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
850 
851 	cur_size = i_size_read(inode);
852  	new_isize = nfs_size_to_loff_t(fattr->size);
853 	if (cur_size != new_isize && nfsi->npages == 0)
854 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
855 
856 	/* Have any file permissions changed? */
857 	if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
858 			|| inode->i_uid != fattr->uid
859 			|| inode->i_gid != fattr->gid)
860 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
861 
862 	/* Has the link count changed? */
863 	if (inode->i_nlink != fattr->nlink)
864 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
865 
866 	if (!timespec_equal(&inode->i_atime, &fattr->atime))
867 		nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
868 
869 	nfsi->read_cache_jiffies = fattr->time_start;
870 	return 0;
871 }
872 
873 /**
874  * nfs_refresh_inode - try to update the inode attribute cache
875  * @inode - pointer to inode
876  * @fattr - updated attributes
877  *
878  * Check that an RPC call that returned attributes has not overlapped with
879  * other recent updates of the inode metadata, then decide whether it is
880  * safe to do a full update of the inode attributes, or whether just to
881  * call nfs_check_inode_attributes.
882  */
883 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
884 {
885 	struct nfs_inode *nfsi = NFS_I(inode);
886 	int status;
887 
888 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
889 		return 0;
890 	spin_lock(&inode->i_lock);
891 	if (time_after(fattr->time_start, nfsi->last_updated))
892 		status = nfs_update_inode(inode, fattr);
893 	else
894 		status = nfs_check_inode_attributes(inode, fattr);
895 
896 	spin_unlock(&inode->i_lock);
897 	return status;
898 }
899 
900 /**
901  * nfs_post_op_update_inode - try to update the inode attribute cache
902  * @inode - pointer to inode
903  * @fattr - updated attributes
904  *
905  * After an operation that has changed the inode metadata, mark the
906  * attribute cache as being invalid, then try to update it.
907  *
908  * NB: if the server didn't return any post op attributes, this
909  * function will force the retrieval of attributes before the next
910  * NFS request.  Thus it should be used only for operations that
911  * are expected to change one or more attributes, to avoid
912  * unnecessary NFS requests and trips through nfs_update_inode().
913  */
914 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
915 {
916 	struct nfs_inode *nfsi = NFS_I(inode);
917 	int status = 0;
918 
919 	spin_lock(&inode->i_lock);
920 	if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
921 		nfsi->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
922 		goto out;
923 	}
924 	status = nfs_update_inode(inode, fattr);
925 out:
926 	spin_unlock(&inode->i_lock);
927 	return status;
928 }
929 
930 /*
931  * Many nfs protocol calls return the new file attributes after
932  * an operation.  Here we update the inode to reflect the state
933  * of the server's inode.
934  *
935  * This is a bit tricky because we have to make sure all dirty pages
936  * have been sent off to the server before calling invalidate_inode_pages.
937  * To make sure no other process adds more write requests while we try
938  * our best to flush them, we make them sleep during the attribute refresh.
939  *
940  * A very similar scenario holds for the dir cache.
941  */
942 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
943 {
944 	struct nfs_server *server;
945 	struct nfs_inode *nfsi = NFS_I(inode);
946 	loff_t cur_isize, new_isize;
947 	unsigned int	invalid = 0;
948 	unsigned long now = jiffies;
949 	int data_stable;
950 
951 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
952 			__FUNCTION__, inode->i_sb->s_id, inode->i_ino,
953 			atomic_read(&inode->i_count), fattr->valid);
954 
955 	if (nfsi->fileid != fattr->fileid)
956 		goto out_fileid;
957 
958 	/*
959 	 * Make sure the inode's type hasn't changed.
960 	 */
961 	if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
962 		goto out_changed;
963 
964 	server = NFS_SERVER(inode);
965 	/* Update the fsid if and only if this is the root directory */
966 	if (inode == inode->i_sb->s_root->d_inode
967 			&& !nfs_fsid_equal(&server->fsid, &fattr->fsid))
968 		server->fsid = fattr->fsid;
969 
970 	/*
971 	 * Update the read time so we don't revalidate too often.
972 	 */
973 	nfsi->read_cache_jiffies = fattr->time_start;
974 	nfsi->last_updated = now;
975 
976 	/* Fix a wraparound issue with nfsi->cache_change_attribute */
977 	if (time_before(now, nfsi->cache_change_attribute))
978 		nfsi->cache_change_attribute = now - 600*HZ;
979 
980 	/* Are we racing with known updates of the metadata on the server? */
981 	data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
982 	if (data_stable)
983 		nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATIME);
984 
985 	/* Do atomic weak cache consistency updates */
986 	nfs_wcc_update_inode(inode, fattr);
987 
988 	/* Check if our cached file size is stale */
989  	new_isize = nfs_size_to_loff_t(fattr->size);
990 	cur_isize = i_size_read(inode);
991 	if (new_isize != cur_isize) {
992 		/* Do we perhaps have any outstanding writes? */
993 		if (nfsi->npages == 0) {
994 			/* No, but did we race with nfs_end_data_update()? */
995 			if (data_stable) {
996 				inode->i_size = new_isize;
997 				invalid |= NFS_INO_INVALID_DATA;
998 			}
999 			invalid |= NFS_INO_INVALID_ATTR;
1000 		} else if (new_isize > cur_isize) {
1001 			inode->i_size = new_isize;
1002 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1003 		}
1004 		nfsi->cache_change_attribute = now;
1005 		dprintk("NFS: isize change on server for file %s/%ld\n",
1006 				inode->i_sb->s_id, inode->i_ino);
1007 	}
1008 
1009 	/* Check if the mtime agrees */
1010 	if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1011 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1012 		dprintk("NFS: mtime change on server for file %s/%ld\n",
1013 				inode->i_sb->s_id, inode->i_ino);
1014 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1015 		nfsi->cache_change_attribute = now;
1016 	}
1017 
1018 	/* If ctime has changed we should definitely clear access+acl caches */
1019 	if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1020 		invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1021 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1022 		nfsi->cache_change_attribute = now;
1023 	}
1024 	memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1025 
1026 	if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1027 	    inode->i_uid != fattr->uid ||
1028 	    inode->i_gid != fattr->gid)
1029 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1030 
1031 	inode->i_mode = fattr->mode;
1032 	inode->i_nlink = fattr->nlink;
1033 	inode->i_uid = fattr->uid;
1034 	inode->i_gid = fattr->gid;
1035 
1036 	if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1037 		/*
1038 		 * report the blocks in 512byte units
1039 		 */
1040 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1041  	} else {
1042  		inode->i_blocks = fattr->du.nfs2.blocks;
1043  	}
1044 
1045 	if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1046 			nfsi->change_attr != fattr->change_attr) {
1047 		dprintk("NFS: change_attr change on server for file %s/%ld\n",
1048 				inode->i_sb->s_id, inode->i_ino);
1049 		nfsi->change_attr = fattr->change_attr;
1050 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1051 		nfsi->cache_change_attribute = now;
1052 	}
1053 
1054 	/* Update attrtimeo value if we're out of the unstable period */
1055 	if (invalid & NFS_INO_INVALID_ATTR) {
1056 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1057 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1058 		nfsi->attrtimeo_timestamp = now;
1059 	} else if (time_after(now, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1060 		if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1061 			nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1062 		nfsi->attrtimeo_timestamp = now;
1063 	}
1064 	/* Don't invalidate the data if we were to blame */
1065 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1066 				|| S_ISLNK(inode->i_mode)))
1067 		invalid &= ~NFS_INO_INVALID_DATA;
1068 	if (data_stable)
1069 		invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1070 	if (!nfs_have_delegation(inode, FMODE_READ))
1071 		nfsi->cache_validity |= invalid;
1072 
1073 	return 0;
1074  out_changed:
1075 	/*
1076 	 * Big trouble! The inode has become a different object.
1077 	 */
1078 #ifdef NFS_PARANOIA
1079 	printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1080 			__FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1081 #endif
1082  out_err:
1083 	/*
1084 	 * No need to worry about unhashing the dentry, as the
1085 	 * lookup validation will know that the inode is bad.
1086 	 * (But we fall through to invalidate the caches.)
1087 	 */
1088 	nfs_invalidate_inode(inode);
1089 	return -ESTALE;
1090 
1091  out_fileid:
1092 	printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1093 		"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1094 		NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1095 		(long long)nfsi->fileid, (long long)fattr->fileid);
1096 	goto out_err;
1097 }
1098 
1099 
1100 #ifdef CONFIG_NFS_V4
1101 
1102 /*
1103  * Clean out any remaining NFSv4 state that might be left over due
1104  * to open() calls that passed nfs_atomic_lookup, but failed to call
1105  * nfs_open().
1106  */
1107 void nfs4_clear_inode(struct inode *inode)
1108 {
1109 	struct nfs_inode *nfsi = NFS_I(inode);
1110 
1111 	/* If we are holding a delegation, return it! */
1112 	nfs_inode_return_delegation(inode);
1113 	/* First call standard NFS clear_inode() code */
1114 	nfs_clear_inode(inode);
1115 	/* Now clear out any remaining state */
1116 	while (!list_empty(&nfsi->open_states)) {
1117 		struct nfs4_state *state;
1118 
1119 		state = list_entry(nfsi->open_states.next,
1120 				struct nfs4_state,
1121 				inode_states);
1122 		dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1123 				__FUNCTION__,
1124 				inode->i_sb->s_id,
1125 				(long long)NFS_FILEID(inode),
1126 				state);
1127 		BUG_ON(atomic_read(&state->count) != 1);
1128 		nfs4_close_state(state, state->state);
1129 	}
1130 }
1131 #endif
1132 
1133 struct inode *nfs_alloc_inode(struct super_block *sb)
1134 {
1135 	struct nfs_inode *nfsi;
1136 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1137 	if (!nfsi)
1138 		return NULL;
1139 	nfsi->flags = 0UL;
1140 	nfsi->cache_validity = 0UL;
1141 #ifdef CONFIG_NFS_V3_ACL
1142 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1143 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1144 #endif
1145 #ifdef CONFIG_NFS_V4
1146 	nfsi->nfs4_acl = NULL;
1147 #endif /* CONFIG_NFS_V4 */
1148 	return &nfsi->vfs_inode;
1149 }
1150 
1151 void nfs_destroy_inode(struct inode *inode)
1152 {
1153 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1154 }
1155 
1156 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1157 {
1158 #ifdef CONFIG_NFS_V4
1159 	INIT_LIST_HEAD(&nfsi->open_states);
1160 	nfsi->delegation = NULL;
1161 	nfsi->delegation_state = 0;
1162 	init_rwsem(&nfsi->rwsem);
1163 #endif
1164 }
1165 
1166 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
1167 {
1168 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1169 
1170 	if (flags & SLAB_CTOR_CONSTRUCTOR) {
1171 		inode_init_once(&nfsi->vfs_inode);
1172 		spin_lock_init(&nfsi->req_lock);
1173 		INIT_LIST_HEAD(&nfsi->dirty);
1174 		INIT_LIST_HEAD(&nfsi->commit);
1175 		INIT_LIST_HEAD(&nfsi->open_files);
1176 		INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1177 		INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1178 		INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1179 		atomic_set(&nfsi->data_updates, 0);
1180 		nfsi->ndirty = 0;
1181 		nfsi->ncommit = 0;
1182 		nfsi->npages = 0;
1183 		nfs4_init_once(nfsi);
1184 	}
1185 }
1186 
1187 static int __init nfs_init_inodecache(void)
1188 {
1189 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1190 					     sizeof(struct nfs_inode),
1191 					     0, (SLAB_RECLAIM_ACCOUNT|
1192 						SLAB_MEM_SPREAD),
1193 					     init_once, NULL);
1194 	if (nfs_inode_cachep == NULL)
1195 		return -ENOMEM;
1196 
1197 	return 0;
1198 }
1199 
1200 static void nfs_destroy_inodecache(void)
1201 {
1202 	kmem_cache_destroy(nfs_inode_cachep);
1203 }
1204 
1205 /*
1206  * Initialize NFS
1207  */
1208 static int __init init_nfs_fs(void)
1209 {
1210 	int err;
1211 
1212 	err = nfs_fs_proc_init();
1213 	if (err)
1214 		goto out5;
1215 
1216 	err = nfs_init_nfspagecache();
1217 	if (err)
1218 		goto out4;
1219 
1220 	err = nfs_init_inodecache();
1221 	if (err)
1222 		goto out3;
1223 
1224 	err = nfs_init_readpagecache();
1225 	if (err)
1226 		goto out2;
1227 
1228 	err = nfs_init_writepagecache();
1229 	if (err)
1230 		goto out1;
1231 
1232 	err = nfs_init_directcache();
1233 	if (err)
1234 		goto out0;
1235 
1236 #ifdef CONFIG_PROC_FS
1237 	rpc_proc_register(&nfs_rpcstat);
1238 #endif
1239 	if ((err = register_nfs_fs()) != 0)
1240 		goto out;
1241 	return 0;
1242 out:
1243 #ifdef CONFIG_PROC_FS
1244 	rpc_proc_unregister("nfs");
1245 #endif
1246 	nfs_destroy_directcache();
1247 out0:
1248 	nfs_destroy_writepagecache();
1249 out1:
1250 	nfs_destroy_readpagecache();
1251 out2:
1252 	nfs_destroy_inodecache();
1253 out3:
1254 	nfs_destroy_nfspagecache();
1255 out4:
1256 	nfs_fs_proc_exit();
1257 out5:
1258 	return err;
1259 }
1260 
1261 static void __exit exit_nfs_fs(void)
1262 {
1263 	nfs_destroy_directcache();
1264 	nfs_destroy_writepagecache();
1265 	nfs_destroy_readpagecache();
1266 	nfs_destroy_inodecache();
1267 	nfs_destroy_nfspagecache();
1268 #ifdef CONFIG_PROC_FS
1269 	rpc_proc_unregister("nfs");
1270 #endif
1271 	unregister_nfs_fs();
1272 	nfs_fs_proc_exit();
1273 }
1274 
1275 /* Not quite true; I just maintain it */
1276 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1277 MODULE_LICENSE("GPL");
1278 
1279 module_init(init_nfs_fs)
1280 module_exit(exit_nfs_fs)
1281