xref: /linux/fs/nfs/inode.c (revision a7edd0e676d51145ae634a2acf7a447e319200fa)
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 	filemap_write_and_wait(inode->i_mapping);
345 	nfs_wb_all(inode);
346 	/*
347 	 * Return any delegations if we're going to change ACLs
348 	 */
349 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
350 		nfs_inode_return_delegation(inode);
351 	error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
352 	if (error == 0)
353 		nfs_refresh_inode(inode, &fattr);
354 	nfs_end_data_update(inode);
355 	unlock_kernel();
356 	return error;
357 }
358 
359 /**
360  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
361  * @inode: pointer to struct inode
362  * @attr: pointer to struct iattr
363  *
364  * Note: we do this in the *proc.c in order to ensure that
365  *       it works for things like exclusive creates too.
366  */
367 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
368 {
369 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
370 		if ((attr->ia_valid & ATTR_MODE) != 0) {
371 			int mode = attr->ia_mode & S_IALLUGO;
372 			mode |= inode->i_mode & ~S_IALLUGO;
373 			inode->i_mode = mode;
374 		}
375 		if ((attr->ia_valid & ATTR_UID) != 0)
376 			inode->i_uid = attr->ia_uid;
377 		if ((attr->ia_valid & ATTR_GID) != 0)
378 			inode->i_gid = attr->ia_gid;
379 		spin_lock(&inode->i_lock);
380 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
381 		spin_unlock(&inode->i_lock);
382 	}
383 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
384 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
385 		inode->i_size = attr->ia_size;
386 		vmtruncate(inode, attr->ia_size);
387 	}
388 }
389 
390 static int nfs_wait_schedule(void *word)
391 {
392 	if (signal_pending(current))
393 		return -ERESTARTSYS;
394 	schedule();
395 	return 0;
396 }
397 
398 /*
399  * Wait for the inode to get unlocked.
400  */
401 static int nfs_wait_on_inode(struct inode *inode)
402 {
403 	struct rpc_clnt	*clnt = NFS_CLIENT(inode);
404 	struct nfs_inode *nfsi = NFS_I(inode);
405 	sigset_t oldmask;
406 	int error;
407 
408 	rpc_clnt_sigmask(clnt, &oldmask);
409 	error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
410 					nfs_wait_schedule, TASK_INTERRUPTIBLE);
411 	rpc_clnt_sigunmask(clnt, &oldmask);
412 
413 	return error;
414 }
415 
416 static void nfs_wake_up_inode(struct inode *inode)
417 {
418 	struct nfs_inode *nfsi = NFS_I(inode);
419 
420 	clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
421 	smp_mb__after_clear_bit();
422 	wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
423 }
424 
425 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
426 {
427 	struct inode *inode = dentry->d_inode;
428 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
429 	int err;
430 
431 	/* Flush out writes to the server in order to update c/mtime */
432 	if (S_ISREG(inode->i_mode))
433 		nfs_sync_mapping_range(inode->i_mapping, 0, 0, FLUSH_NOCOMMIT);
434 
435 	/*
436 	 * We may force a getattr if the user cares about atime.
437 	 *
438 	 * Note that we only have to check the vfsmount flags here:
439 	 *  - NFS always sets S_NOATIME by so checking it would give a
440 	 *    bogus result
441 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
442 	 *    no point in checking those.
443 	 */
444  	if ((mnt->mnt_flags & MNT_NOATIME) ||
445  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
446 		need_atime = 0;
447 
448 	if (need_atime)
449 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
450 	else
451 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
452 	if (!err)
453 		generic_fillattr(inode, stat);
454 	return err;
455 }
456 
457 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
458 {
459 	struct nfs_open_context *ctx;
460 
461 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
462 	if (ctx != NULL) {
463 		atomic_set(&ctx->count, 1);
464 		ctx->dentry = dget(dentry);
465 		ctx->vfsmnt = mntget(mnt);
466 		ctx->cred = get_rpccred(cred);
467 		ctx->state = NULL;
468 		ctx->lockowner = current->files;
469 		ctx->error = 0;
470 		ctx->dir_cookie = 0;
471 	}
472 	return ctx;
473 }
474 
475 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
476 {
477 	if (ctx != NULL)
478 		atomic_inc(&ctx->count);
479 	return ctx;
480 }
481 
482 void put_nfs_open_context(struct nfs_open_context *ctx)
483 {
484 	if (atomic_dec_and_test(&ctx->count)) {
485 		if (!list_empty(&ctx->list)) {
486 			struct inode *inode = ctx->dentry->d_inode;
487 			spin_lock(&inode->i_lock);
488 			list_del(&ctx->list);
489 			spin_unlock(&inode->i_lock);
490 		}
491 		if (ctx->state != NULL)
492 			nfs4_close_state(ctx->state, ctx->mode);
493 		if (ctx->cred != NULL)
494 			put_rpccred(ctx->cred);
495 		dput(ctx->dentry);
496 		mntput(ctx->vfsmnt);
497 		kfree(ctx);
498 	}
499 }
500 
501 /*
502  * Ensure that mmap has a recent RPC credential for use when writing out
503  * shared pages
504  */
505 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
506 {
507 	struct inode *inode = filp->f_path.dentry->d_inode;
508 	struct nfs_inode *nfsi = NFS_I(inode);
509 
510 	filp->private_data = get_nfs_open_context(ctx);
511 	spin_lock(&inode->i_lock);
512 	list_add(&ctx->list, &nfsi->open_files);
513 	spin_unlock(&inode->i_lock);
514 }
515 
516 /*
517  * Given an inode, search for an open context with the desired characteristics
518  */
519 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
520 {
521 	struct nfs_inode *nfsi = NFS_I(inode);
522 	struct nfs_open_context *pos, *ctx = NULL;
523 
524 	spin_lock(&inode->i_lock);
525 	list_for_each_entry(pos, &nfsi->open_files, list) {
526 		if (cred != NULL && pos->cred != cred)
527 			continue;
528 		if ((pos->mode & mode) == mode) {
529 			ctx = get_nfs_open_context(pos);
530 			break;
531 		}
532 	}
533 	spin_unlock(&inode->i_lock);
534 	return ctx;
535 }
536 
537 static void nfs_file_clear_open_context(struct file *filp)
538 {
539 	struct inode *inode = filp->f_path.dentry->d_inode;
540 	struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
541 
542 	if (ctx) {
543 		filp->private_data = NULL;
544 		spin_lock(&inode->i_lock);
545 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
546 		spin_unlock(&inode->i_lock);
547 		put_nfs_open_context(ctx);
548 	}
549 }
550 
551 /*
552  * These allocate and release file read/write context information.
553  */
554 int nfs_open(struct inode *inode, struct file *filp)
555 {
556 	struct nfs_open_context *ctx;
557 	struct rpc_cred *cred;
558 
559 	cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
560 	if (IS_ERR(cred))
561 		return PTR_ERR(cred);
562 	ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
563 	put_rpccred(cred);
564 	if (ctx == NULL)
565 		return -ENOMEM;
566 	ctx->mode = filp->f_mode;
567 	nfs_file_set_open_context(filp, ctx);
568 	put_nfs_open_context(ctx);
569 	return 0;
570 }
571 
572 int nfs_release(struct inode *inode, struct file *filp)
573 {
574 	nfs_file_clear_open_context(filp);
575 	return 0;
576 }
577 
578 /*
579  * This function is called whenever some part of NFS notices that
580  * the cached attributes have to be refreshed.
581  */
582 int
583 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
584 {
585 	int		 status = -ESTALE;
586 	struct nfs_fattr fattr;
587 	struct nfs_inode *nfsi = NFS_I(inode);
588 
589 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
590 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
591 
592 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
593 	lock_kernel();
594 	if (is_bad_inode(inode))
595  		goto out_nowait;
596 	if (NFS_STALE(inode))
597  		goto out_nowait;
598 
599 	status = nfs_wait_on_inode(inode);
600 	if (status < 0)
601 		goto out;
602 	if (NFS_STALE(inode)) {
603 		status = -ESTALE;
604 		/* Do we trust the cached ESTALE? */
605 		if (NFS_ATTRTIMEO(inode) != 0) {
606 			if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME)) {
607 				/* no */
608 			} else
609 				goto out;
610 		}
611 	}
612 
613 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
614 	if (status != 0) {
615 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
616 			 inode->i_sb->s_id,
617 			 (long long)NFS_FILEID(inode), status);
618 		if (status == -ESTALE) {
619 			nfs_zap_caches(inode);
620 			if (!S_ISDIR(inode->i_mode))
621 				set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
622 		}
623 		goto out;
624 	}
625 
626 	spin_lock(&inode->i_lock);
627 	status = nfs_update_inode(inode, &fattr);
628 	if (status) {
629 		spin_unlock(&inode->i_lock);
630 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
631 			 inode->i_sb->s_id,
632 			 (long long)NFS_FILEID(inode), status);
633 		goto out;
634 	}
635 	spin_unlock(&inode->i_lock);
636 
637 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
638 		nfs_zap_acl_cache(inode);
639 
640 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
641 		inode->i_sb->s_id,
642 		(long long)NFS_FILEID(inode));
643 
644  out:
645 	nfs_wake_up_inode(inode);
646 
647  out_nowait:
648 	unlock_kernel();
649 	return status;
650 }
651 
652 int nfs_attribute_timeout(struct inode *inode)
653 {
654 	struct nfs_inode *nfsi = NFS_I(inode);
655 
656 	if (nfs_have_delegation(inode, FMODE_READ))
657 		return 0;
658 	return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
659 }
660 
661 /**
662  * nfs_revalidate_inode - Revalidate the inode attributes
663  * @server - pointer to nfs_server struct
664  * @inode - pointer to inode struct
665  *
666  * Updates inode attribute information by retrieving the data from the server.
667  */
668 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
669 {
670 	if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
671 			&& !nfs_attribute_timeout(inode))
672 		return NFS_STALE(inode) ? -ESTALE : 0;
673 	return __nfs_revalidate_inode(server, inode);
674 }
675 
676 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
677 {
678 	struct nfs_inode *nfsi = NFS_I(inode);
679 
680 	if (mapping->nrpages != 0) {
681 		int ret = invalidate_inode_pages2(mapping);
682 		if (ret < 0)
683 			return ret;
684 	}
685 	spin_lock(&inode->i_lock);
686 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
687 	if (S_ISDIR(inode->i_mode)) {
688 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
689 		/* This ensures we revalidate child dentries */
690 		nfsi->cache_change_attribute = jiffies;
691 	}
692 	spin_unlock(&inode->i_lock);
693 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
694 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
695 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
696 	return 0;
697 }
698 
699 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
700 {
701 	int ret = 0;
702 
703 	mutex_lock(&inode->i_mutex);
704 	if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
705 		ret = nfs_sync_mapping(mapping);
706 		if (ret == 0)
707 			ret = nfs_invalidate_mapping_nolock(inode, mapping);
708 	}
709 	mutex_unlock(&inode->i_mutex);
710 	return ret;
711 }
712 
713 /**
714  * nfs_revalidate_mapping_nolock - Revalidate the pagecache
715  * @inode - pointer to host inode
716  * @mapping - pointer to mapping
717  */
718 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
719 {
720 	struct nfs_inode *nfsi = NFS_I(inode);
721 	int ret = 0;
722 
723 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
724 			|| nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
725 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
726 		if (ret < 0)
727 			goto out;
728 	}
729 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
730 		ret = nfs_invalidate_mapping_nolock(inode, mapping);
731 out:
732 	return ret;
733 }
734 
735 /**
736  * nfs_revalidate_mapping - Revalidate the pagecache
737  * @inode - pointer to host inode
738  * @mapping - pointer to mapping
739  *
740  * This version of the function will take the inode->i_mutex and attempt to
741  * flush out all dirty data if it needs to invalidate the page cache.
742  */
743 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
744 {
745 	struct nfs_inode *nfsi = NFS_I(inode);
746 	int ret = 0;
747 
748 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
749 			|| nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
750 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
751 		if (ret < 0)
752 			goto out;
753 	}
754 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
755 		ret = nfs_invalidate_mapping(inode, mapping);
756 out:
757 	return ret;
758 }
759 
760 /**
761  * nfs_begin_data_update
762  * @inode - pointer to inode
763  * Declare that a set of operations will update file data on the server
764  */
765 void nfs_begin_data_update(struct inode *inode)
766 {
767 	atomic_inc(&NFS_I(inode)->data_updates);
768 }
769 
770 /**
771  * nfs_end_data_update
772  * @inode - pointer to inode
773  * Declare end of the operations that will update file data
774  * This will mark the inode as immediately needing revalidation
775  * of its attribute cache.
776  */
777 void nfs_end_data_update(struct inode *inode)
778 {
779 	struct nfs_inode *nfsi = NFS_I(inode);
780 
781 	/* Directories: invalidate page cache */
782 	if (S_ISDIR(inode->i_mode)) {
783 		spin_lock(&inode->i_lock);
784 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
785 		spin_unlock(&inode->i_lock);
786 	}
787 	nfsi->cache_change_attribute = jiffies;
788 	atomic_dec(&nfsi->data_updates);
789 }
790 
791 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
792 {
793 	struct nfs_inode *nfsi = NFS_I(inode);
794 	unsigned long now = jiffies;
795 
796 	/* If we have atomic WCC data, we may update some attributes */
797 	if ((fattr->valid & NFS_ATTR_WCC) != 0) {
798 		if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
799 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
800 			nfsi->cache_change_attribute = now;
801 		}
802 		if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
803 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
804 			nfsi->cache_change_attribute = now;
805 		}
806 		if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
807 			inode->i_size = fattr->size;
808 			nfsi->cache_change_attribute = now;
809 		}
810 	}
811 }
812 
813 /**
814  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
815  * @inode - pointer to inode
816  * @fattr - updated attributes
817  *
818  * Verifies the attribute cache. If we have just changed the attributes,
819  * so that fattr carries weak cache consistency data, then it may
820  * also update the ctime/mtime/change_attribute.
821  */
822 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
823 {
824 	struct nfs_inode *nfsi = NFS_I(inode);
825 	loff_t cur_size, new_isize;
826 	int data_unstable;
827 
828 
829 	/* Has the inode gone and changed behind our back? */
830 	if (nfsi->fileid != fattr->fileid
831 			|| (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
832 		return -EIO;
833 	}
834 
835 	/* Are we in the process of updating data on the server? */
836 	data_unstable = nfs_caches_unstable(inode);
837 
838 	/* Do atomic weak cache consistency updates */
839 	nfs_wcc_update_inode(inode, fattr);
840 
841 	if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
842 			nfsi->change_attr != fattr->change_attr)
843 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
844 
845 	/* Verify a few of the more important attributes */
846 	if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
847 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
848 
849 	cur_size = i_size_read(inode);
850  	new_isize = nfs_size_to_loff_t(fattr->size);
851 	if (cur_size != new_isize && nfsi->npages == 0)
852 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
853 
854 	/* Have any file permissions changed? */
855 	if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
856 			|| inode->i_uid != fattr->uid
857 			|| inode->i_gid != fattr->gid)
858 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
859 
860 	/* Has the link count changed? */
861 	if (inode->i_nlink != fattr->nlink)
862 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
863 
864 	if (!timespec_equal(&inode->i_atime, &fattr->atime))
865 		nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
866 
867 	nfsi->read_cache_jiffies = fattr->time_start;
868 	return 0;
869 }
870 
871 /**
872  * nfs_refresh_inode - try to update the inode attribute cache
873  * @inode - pointer to inode
874  * @fattr - updated attributes
875  *
876  * Check that an RPC call that returned attributes has not overlapped with
877  * other recent updates of the inode metadata, then decide whether it is
878  * safe to do a full update of the inode attributes, or whether just to
879  * call nfs_check_inode_attributes.
880  */
881 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
882 {
883 	struct nfs_inode *nfsi = NFS_I(inode);
884 	int status;
885 
886 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
887 		return 0;
888 	spin_lock(&inode->i_lock);
889 	if (time_after(fattr->time_start, nfsi->last_updated))
890 		status = nfs_update_inode(inode, fattr);
891 	else
892 		status = nfs_check_inode_attributes(inode, fattr);
893 
894 	spin_unlock(&inode->i_lock);
895 	return status;
896 }
897 
898 /**
899  * nfs_post_op_update_inode - try to update the inode attribute cache
900  * @inode - pointer to inode
901  * @fattr - updated attributes
902  *
903  * After an operation that has changed the inode metadata, mark the
904  * attribute cache as being invalid, then try to update it.
905  *
906  * NB: if the server didn't return any post op attributes, this
907  * function will force the retrieval of attributes before the next
908  * NFS request.  Thus it should be used only for operations that
909  * are expected to change one or more attributes, to avoid
910  * unnecessary NFS requests and trips through nfs_update_inode().
911  */
912 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
913 {
914 	struct nfs_inode *nfsi = NFS_I(inode);
915 	int status = 0;
916 
917 	spin_lock(&inode->i_lock);
918 	if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
919 		nfsi->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
920 		goto out;
921 	}
922 	status = nfs_update_inode(inode, fattr);
923 out:
924 	spin_unlock(&inode->i_lock);
925 	return status;
926 }
927 
928 /*
929  * Many nfs protocol calls return the new file attributes after
930  * an operation.  Here we update the inode to reflect the state
931  * of the server's inode.
932  *
933  * This is a bit tricky because we have to make sure all dirty pages
934  * have been sent off to the server before calling invalidate_inode_pages.
935  * To make sure no other process adds more write requests while we try
936  * our best to flush them, we make them sleep during the attribute refresh.
937  *
938  * A very similar scenario holds for the dir cache.
939  */
940 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
941 {
942 	struct nfs_server *server;
943 	struct nfs_inode *nfsi = NFS_I(inode);
944 	loff_t cur_isize, new_isize;
945 	unsigned int	invalid = 0;
946 	unsigned long now = jiffies;
947 	int data_stable;
948 
949 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
950 			__FUNCTION__, inode->i_sb->s_id, inode->i_ino,
951 			atomic_read(&inode->i_count), fattr->valid);
952 
953 	if (nfsi->fileid != fattr->fileid)
954 		goto out_fileid;
955 
956 	/*
957 	 * Make sure the inode's type hasn't changed.
958 	 */
959 	if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
960 		goto out_changed;
961 
962 	server = NFS_SERVER(inode);
963 	/* Update the fsid if and only if this is the root directory */
964 	if (inode == inode->i_sb->s_root->d_inode
965 			&& !nfs_fsid_equal(&server->fsid, &fattr->fsid))
966 		server->fsid = fattr->fsid;
967 
968 	/*
969 	 * Update the read time so we don't revalidate too often.
970 	 */
971 	nfsi->read_cache_jiffies = fattr->time_start;
972 	nfsi->last_updated = now;
973 
974 	/* Fix a wraparound issue with nfsi->cache_change_attribute */
975 	if (time_before(now, nfsi->cache_change_attribute))
976 		nfsi->cache_change_attribute = now - 600*HZ;
977 
978 	/* Are we racing with known updates of the metadata on the server? */
979 	data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
980 	if (data_stable)
981 		nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATIME);
982 
983 	/* Do atomic weak cache consistency updates */
984 	nfs_wcc_update_inode(inode, fattr);
985 
986 	/* Check if our cached file size is stale */
987  	new_isize = nfs_size_to_loff_t(fattr->size);
988 	cur_isize = i_size_read(inode);
989 	if (new_isize != cur_isize) {
990 		/* Do we perhaps have any outstanding writes? */
991 		if (nfsi->npages == 0) {
992 			/* No, but did we race with nfs_end_data_update()? */
993 			if (data_stable) {
994 				inode->i_size = new_isize;
995 				invalid |= NFS_INO_INVALID_DATA;
996 			}
997 			invalid |= NFS_INO_INVALID_ATTR;
998 		} else if (new_isize > cur_isize) {
999 			inode->i_size = new_isize;
1000 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1001 		}
1002 		nfsi->cache_change_attribute = now;
1003 		dprintk("NFS: isize change on server for file %s/%ld\n",
1004 				inode->i_sb->s_id, inode->i_ino);
1005 	}
1006 
1007 	/* Check if the mtime agrees */
1008 	if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1009 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1010 		dprintk("NFS: mtime change on server for file %s/%ld\n",
1011 				inode->i_sb->s_id, inode->i_ino);
1012 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1013 		nfsi->cache_change_attribute = now;
1014 	}
1015 
1016 	/* If ctime has changed we should definitely clear access+acl caches */
1017 	if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1018 		invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1019 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1020 		nfsi->cache_change_attribute = now;
1021 	}
1022 	memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1023 
1024 	if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1025 	    inode->i_uid != fattr->uid ||
1026 	    inode->i_gid != fattr->gid)
1027 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1028 
1029 	inode->i_mode = fattr->mode;
1030 	inode->i_nlink = fattr->nlink;
1031 	inode->i_uid = fattr->uid;
1032 	inode->i_gid = fattr->gid;
1033 
1034 	if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1035 		/*
1036 		 * report the blocks in 512byte units
1037 		 */
1038 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1039  	} else {
1040  		inode->i_blocks = fattr->du.nfs2.blocks;
1041  	}
1042 
1043 	if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1044 			nfsi->change_attr != fattr->change_attr) {
1045 		dprintk("NFS: change_attr change on server for file %s/%ld\n",
1046 				inode->i_sb->s_id, inode->i_ino);
1047 		nfsi->change_attr = fattr->change_attr;
1048 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1049 		nfsi->cache_change_attribute = now;
1050 	}
1051 
1052 	/* Update attrtimeo value if we're out of the unstable period */
1053 	if (invalid & NFS_INO_INVALID_ATTR) {
1054 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1055 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1056 		nfsi->attrtimeo_timestamp = now;
1057 	} else if (time_after(now, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1058 		if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1059 			nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1060 		nfsi->attrtimeo_timestamp = now;
1061 	}
1062 	/* Don't invalidate the data if we were to blame */
1063 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1064 				|| S_ISLNK(inode->i_mode)))
1065 		invalid &= ~NFS_INO_INVALID_DATA;
1066 	if (data_stable)
1067 		invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1068 	if (!nfs_have_delegation(inode, FMODE_READ))
1069 		nfsi->cache_validity |= invalid;
1070 
1071 	return 0;
1072  out_changed:
1073 	/*
1074 	 * Big trouble! The inode has become a different object.
1075 	 */
1076 #ifdef NFS_PARANOIA
1077 	printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1078 			__FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1079 #endif
1080  out_err:
1081 	/*
1082 	 * No need to worry about unhashing the dentry, as the
1083 	 * lookup validation will know that the inode is bad.
1084 	 * (But we fall through to invalidate the caches.)
1085 	 */
1086 	nfs_invalidate_inode(inode);
1087 	return -ESTALE;
1088 
1089  out_fileid:
1090 	printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1091 		"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1092 		NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1093 		(long long)nfsi->fileid, (long long)fattr->fileid);
1094 	goto out_err;
1095 }
1096 
1097 
1098 #ifdef CONFIG_NFS_V4
1099 
1100 /*
1101  * Clean out any remaining NFSv4 state that might be left over due
1102  * to open() calls that passed nfs_atomic_lookup, but failed to call
1103  * nfs_open().
1104  */
1105 void nfs4_clear_inode(struct inode *inode)
1106 {
1107 	struct nfs_inode *nfsi = NFS_I(inode);
1108 
1109 	/* If we are holding a delegation, return it! */
1110 	nfs_inode_return_delegation(inode);
1111 	/* First call standard NFS clear_inode() code */
1112 	nfs_clear_inode(inode);
1113 	/* Now clear out any remaining state */
1114 	while (!list_empty(&nfsi->open_states)) {
1115 		struct nfs4_state *state;
1116 
1117 		state = list_entry(nfsi->open_states.next,
1118 				struct nfs4_state,
1119 				inode_states);
1120 		dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1121 				__FUNCTION__,
1122 				inode->i_sb->s_id,
1123 				(long long)NFS_FILEID(inode),
1124 				state);
1125 		BUG_ON(atomic_read(&state->count) != 1);
1126 		nfs4_close_state(state, state->state);
1127 	}
1128 }
1129 #endif
1130 
1131 struct inode *nfs_alloc_inode(struct super_block *sb)
1132 {
1133 	struct nfs_inode *nfsi;
1134 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1135 	if (!nfsi)
1136 		return NULL;
1137 	nfsi->flags = 0UL;
1138 	nfsi->cache_validity = 0UL;
1139 #ifdef CONFIG_NFS_V3_ACL
1140 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1141 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1142 #endif
1143 #ifdef CONFIG_NFS_V4
1144 	nfsi->nfs4_acl = NULL;
1145 #endif /* CONFIG_NFS_V4 */
1146 	return &nfsi->vfs_inode;
1147 }
1148 
1149 void nfs_destroy_inode(struct inode *inode)
1150 {
1151 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1152 }
1153 
1154 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1155 {
1156 #ifdef CONFIG_NFS_V4
1157 	INIT_LIST_HEAD(&nfsi->open_states);
1158 	nfsi->delegation = NULL;
1159 	nfsi->delegation_state = 0;
1160 	init_rwsem(&nfsi->rwsem);
1161 #endif
1162 }
1163 
1164 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
1165 {
1166 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1167 
1168 	if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1169 	    SLAB_CTOR_CONSTRUCTOR) {
1170 		inode_init_once(&nfsi->vfs_inode);
1171 		spin_lock_init(&nfsi->req_lock);
1172 		INIT_LIST_HEAD(&nfsi->dirty);
1173 		INIT_LIST_HEAD(&nfsi->commit);
1174 		INIT_LIST_HEAD(&nfsi->open_files);
1175 		INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1176 		INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1177 		INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1178 		atomic_set(&nfsi->data_updates, 0);
1179 		nfsi->ndirty = 0;
1180 		nfsi->ncommit = 0;
1181 		nfsi->npages = 0;
1182 		nfs4_init_once(nfsi);
1183 	}
1184 }
1185 
1186 static int __init nfs_init_inodecache(void)
1187 {
1188 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1189 					     sizeof(struct nfs_inode),
1190 					     0, (SLAB_RECLAIM_ACCOUNT|
1191 						SLAB_MEM_SPREAD),
1192 					     init_once, NULL);
1193 	if (nfs_inode_cachep == NULL)
1194 		return -ENOMEM;
1195 
1196 	return 0;
1197 }
1198 
1199 static void nfs_destroy_inodecache(void)
1200 {
1201 	kmem_cache_destroy(nfs_inode_cachep);
1202 }
1203 
1204 /*
1205  * Initialize NFS
1206  */
1207 static int __init init_nfs_fs(void)
1208 {
1209 	int err;
1210 
1211 	err = nfs_fs_proc_init();
1212 	if (err)
1213 		goto out5;
1214 
1215 	err = nfs_init_nfspagecache();
1216 	if (err)
1217 		goto out4;
1218 
1219 	err = nfs_init_inodecache();
1220 	if (err)
1221 		goto out3;
1222 
1223 	err = nfs_init_readpagecache();
1224 	if (err)
1225 		goto out2;
1226 
1227 	err = nfs_init_writepagecache();
1228 	if (err)
1229 		goto out1;
1230 
1231 	err = nfs_init_directcache();
1232 	if (err)
1233 		goto out0;
1234 
1235 #ifdef CONFIG_PROC_FS
1236 	rpc_proc_register(&nfs_rpcstat);
1237 #endif
1238 	if ((err = register_nfs_fs()) != 0)
1239 		goto out;
1240 	return 0;
1241 out:
1242 #ifdef CONFIG_PROC_FS
1243 	rpc_proc_unregister("nfs");
1244 #endif
1245 	nfs_destroy_directcache();
1246 out0:
1247 	nfs_destroy_writepagecache();
1248 out1:
1249 	nfs_destroy_readpagecache();
1250 out2:
1251 	nfs_destroy_inodecache();
1252 out3:
1253 	nfs_destroy_nfspagecache();
1254 out4:
1255 	nfs_fs_proc_exit();
1256 out5:
1257 	return err;
1258 }
1259 
1260 static void __exit exit_nfs_fs(void)
1261 {
1262 	nfs_destroy_directcache();
1263 	nfs_destroy_writepagecache();
1264 	nfs_destroy_readpagecache();
1265 	nfs_destroy_inodecache();
1266 	nfs_destroy_nfspagecache();
1267 #ifdef CONFIG_PROC_FS
1268 	rpc_proc_unregister("nfs");
1269 #endif
1270 	unregister_nfs_fs();
1271 	nfs_fs_proc_exit();
1272 }
1273 
1274 /* Not quite true; I just maintain it */
1275 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1276 MODULE_LICENSE("GPL");
1277 
1278 module_init(init_nfs_fs)
1279 module_exit(exit_nfs_fs)
1280