xref: /linux/fs/nfs/inode.c (revision 98366c20a275e957416e9516db5dcb7195b4e101)
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 #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/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 
52 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
53 
54 /* Default is to see 64-bit inode numbers */
55 static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
56 
57 static void nfs_invalidate_inode(struct inode *);
58 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
59 
60 static void nfs_zap_acl_cache(struct inode *);
61 
62 static struct kmem_cache * nfs_inode_cachep;
63 
64 static inline unsigned long
65 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
66 {
67 	return nfs_fileid_to_ino_t(fattr->fileid);
68 }
69 
70 /**
71  * nfs_compat_user_ino64 - returns the user-visible inode number
72  * @fileid: 64-bit fileid
73  *
74  * This function returns a 32-bit inode number if the boot parameter
75  * nfs.enable_ino64 is zero.
76  */
77 u64 nfs_compat_user_ino64(u64 fileid)
78 {
79 	int ino;
80 
81 	if (enable_ino64)
82 		return fileid;
83 	ino = fileid;
84 	if (sizeof(ino) < sizeof(fileid))
85 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
86 	return ino;
87 }
88 
89 int nfs_write_inode(struct inode *inode, int sync)
90 {
91 	int ret;
92 
93 	if (sync) {
94 		ret = filemap_fdatawait(inode->i_mapping);
95 		if (ret == 0)
96 			ret = nfs_commit_inode(inode, FLUSH_SYNC);
97 	} else
98 		ret = nfs_commit_inode(inode, 0);
99 	if (ret >= 0)
100 		return 0;
101 	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
102 	return ret;
103 }
104 
105 void nfs_clear_inode(struct inode *inode)
106 {
107 	/*
108 	 * The following should never happen...
109 	 */
110 	BUG_ON(nfs_have_writebacks(inode));
111 	BUG_ON(!list_empty(&NFS_I(inode)->open_files));
112 	nfs_zap_acl_cache(inode);
113 	nfs_access_zap_cache(inode);
114 }
115 
116 /**
117  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
118  */
119 int nfs_sync_mapping(struct address_space *mapping)
120 {
121 	int ret;
122 
123 	if (mapping->nrpages == 0)
124 		return 0;
125 	unmap_mapping_range(mapping, 0, 0, 0);
126 	ret = filemap_write_and_wait(mapping);
127 	if (ret != 0)
128 		goto out;
129 	ret = nfs_wb_all(mapping->host);
130 out:
131 	return ret;
132 }
133 
134 /*
135  * Invalidate the local caches
136  */
137 static void nfs_zap_caches_locked(struct inode *inode)
138 {
139 	struct nfs_inode *nfsi = NFS_I(inode);
140 	int mode = inode->i_mode;
141 
142 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
143 
144 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
145 	nfsi->attrtimeo_timestamp = jiffies;
146 
147 	memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
148 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
149 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
150 	else
151 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
152 }
153 
154 void nfs_zap_caches(struct inode *inode)
155 {
156 	spin_lock(&inode->i_lock);
157 	nfs_zap_caches_locked(inode);
158 	spin_unlock(&inode->i_lock);
159 }
160 
161 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
162 {
163 	if (mapping->nrpages != 0) {
164 		spin_lock(&inode->i_lock);
165 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
166 		spin_unlock(&inode->i_lock);
167 	}
168 }
169 
170 static void nfs_zap_acl_cache(struct inode *inode)
171 {
172 	void (*clear_acl_cache)(struct inode *);
173 
174 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
175 	if (clear_acl_cache != NULL)
176 		clear_acl_cache(inode);
177 	spin_lock(&inode->i_lock);
178 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
179 	spin_unlock(&inode->i_lock);
180 }
181 
182 void nfs_invalidate_atime(struct inode *inode)
183 {
184 	spin_lock(&inode->i_lock);
185 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
186 	spin_unlock(&inode->i_lock);
187 }
188 
189 /*
190  * Invalidate, but do not unhash, the inode.
191  * NB: must be called with inode->i_lock held!
192  */
193 static void nfs_invalidate_inode(struct inode *inode)
194 {
195 	set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
196 	nfs_zap_caches_locked(inode);
197 }
198 
199 struct nfs_find_desc {
200 	struct nfs_fh		*fh;
201 	struct nfs_fattr	*fattr;
202 };
203 
204 /*
205  * In NFSv3 we can have 64bit inode numbers. In order to support
206  * this, and re-exported directories (also seen in NFSv2)
207  * we are forced to allow 2 different inodes to have the same
208  * i_ino.
209  */
210 static int
211 nfs_find_actor(struct inode *inode, void *opaque)
212 {
213 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
214 	struct nfs_fh		*fh = desc->fh;
215 	struct nfs_fattr	*fattr = desc->fattr;
216 
217 	if (NFS_FILEID(inode) != fattr->fileid)
218 		return 0;
219 	if (nfs_compare_fh(NFS_FH(inode), fh))
220 		return 0;
221 	if (is_bad_inode(inode) || NFS_STALE(inode))
222 		return 0;
223 	return 1;
224 }
225 
226 static int
227 nfs_init_locked(struct inode *inode, void *opaque)
228 {
229 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
230 	struct nfs_fattr	*fattr = desc->fattr;
231 
232 	NFS_FILEID(inode) = fattr->fileid;
233 	nfs_copy_fh(NFS_FH(inode), desc->fh);
234 	return 0;
235 }
236 
237 /* Don't use READDIRPLUS on directories that we believe are too large */
238 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
239 
240 /*
241  * This is our front-end to iget that looks up inodes by file handle
242  * instead of inode number.
243  */
244 struct inode *
245 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
246 {
247 	struct nfs_find_desc desc = {
248 		.fh	= fh,
249 		.fattr	= fattr
250 	};
251 	struct inode *inode = ERR_PTR(-ENOENT);
252 	unsigned long hash;
253 
254 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
255 		goto out_no_inode;
256 
257 	if (!fattr->nlink) {
258 		printk("NFS: Buggy server - nlink == 0!\n");
259 		goto out_no_inode;
260 	}
261 
262 	hash = nfs_fattr_to_ino_t(fattr);
263 
264 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
265 	if (inode == NULL) {
266 		inode = ERR_PTR(-ENOMEM);
267 		goto out_no_inode;
268 	}
269 
270 	if (inode->i_state & I_NEW) {
271 		struct nfs_inode *nfsi = NFS_I(inode);
272 		unsigned long now = jiffies;
273 
274 		/* We set i_ino for the few things that still rely on it,
275 		 * such as stat(2) */
276 		inode->i_ino = hash;
277 
278 		/* We can't support update_atime(), since the server will reset it */
279 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
280 		inode->i_mode = fattr->mode;
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 			    && fattr->size <= NFS_LIMIT_READDIRPLUS)
294 				set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
295 			/* Deal with crossing mountpoints */
296 			if (!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 			}
303 		} else if (S_ISLNK(inode->i_mode))
304 			inode->i_op = &nfs_symlink_inode_operations;
305 		else
306 			init_special_inode(inode, inode->i_mode, fattr->rdev);
307 
308 		nfsi->read_cache_jiffies = fattr->time_start;
309 		nfsi->last_updated = now;
310 		nfsi->cache_change_attribute = now;
311 		inode->i_atime = fattr->atime;
312 		inode->i_mtime = fattr->mtime;
313 		inode->i_ctime = fattr->ctime;
314 		if (fattr->valid & NFS_ATTR_FATTR_V4)
315 			nfsi->change_attr = fattr->change_attr;
316 		inode->i_size = nfs_size_to_loff_t(fattr->size);
317 		inode->i_nlink = fattr->nlink;
318 		inode->i_uid = fattr->uid;
319 		inode->i_gid = fattr->gid;
320 		if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
321 			/*
322 			 * report the blocks in 512byte units
323 			 */
324 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
325 		} else {
326 			inode->i_blocks = fattr->du.nfs2.blocks;
327 		}
328 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
329 		nfsi->attrtimeo_timestamp = now;
330 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
331 		nfsi->access_cache = RB_ROOT;
332 
333 		unlock_new_inode(inode);
334 	} else
335 		nfs_refresh_inode(inode, fattr);
336 	dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
337 		inode->i_sb->s_id,
338 		(long long)NFS_FILEID(inode),
339 		atomic_read(&inode->i_count));
340 
341 out:
342 	return inode;
343 
344 out_no_inode:
345 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
346 	goto out;
347 }
348 
349 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
350 
351 int
352 nfs_setattr(struct dentry *dentry, struct iattr *attr)
353 {
354 	struct inode *inode = dentry->d_inode;
355 	struct nfs_fattr fattr;
356 	int error;
357 
358 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
359 
360 	/* skip mode change if it's just for clearing setuid/setgid */
361 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
362 		attr->ia_valid &= ~ATTR_MODE;
363 
364 	if (attr->ia_valid & ATTR_SIZE) {
365 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
366 			attr->ia_valid &= ~ATTR_SIZE;
367 	}
368 
369 	/* Optimization: if the end result is no change, don't RPC */
370 	attr->ia_valid &= NFS_VALID_ATTRS;
371 	if (attr->ia_valid == 0)
372 		return 0;
373 
374 	lock_kernel();
375 	/* Write all dirty data */
376 	if (S_ISREG(inode->i_mode)) {
377 		filemap_write_and_wait(inode->i_mapping);
378 		nfs_wb_all(inode);
379 	}
380 	/*
381 	 * Return any delegations if we're going to change ACLs
382 	 */
383 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
384 		nfs_inode_return_delegation(inode);
385 	error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
386 	if (error == 0)
387 		nfs_refresh_inode(inode, &fattr);
388 	unlock_kernel();
389 	return error;
390 }
391 
392 /**
393  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
394  * @inode: pointer to struct inode
395  * @attr: pointer to struct iattr
396  *
397  * Note: we do this in the *proc.c in order to ensure that
398  *       it works for things like exclusive creates too.
399  */
400 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
401 {
402 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
403 		if ((attr->ia_valid & ATTR_MODE) != 0) {
404 			int mode = attr->ia_mode & S_IALLUGO;
405 			mode |= inode->i_mode & ~S_IALLUGO;
406 			inode->i_mode = mode;
407 		}
408 		if ((attr->ia_valid & ATTR_UID) != 0)
409 			inode->i_uid = attr->ia_uid;
410 		if ((attr->ia_valid & ATTR_GID) != 0)
411 			inode->i_gid = attr->ia_gid;
412 		spin_lock(&inode->i_lock);
413 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
414 		spin_unlock(&inode->i_lock);
415 	}
416 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
417 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
418 		inode->i_size = attr->ia_size;
419 		vmtruncate(inode, attr->ia_size);
420 	}
421 }
422 
423 static int nfs_wait_schedule(void *word)
424 {
425 	if (signal_pending(current))
426 		return -ERESTARTSYS;
427 	schedule();
428 	return 0;
429 }
430 
431 /*
432  * Wait for the inode to get unlocked.
433  */
434 static int nfs_wait_on_inode(struct inode *inode)
435 {
436 	struct rpc_clnt	*clnt = NFS_CLIENT(inode);
437 	struct nfs_inode *nfsi = NFS_I(inode);
438 	sigset_t oldmask;
439 	int error;
440 
441 	rpc_clnt_sigmask(clnt, &oldmask);
442 	error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
443 					nfs_wait_schedule, TASK_INTERRUPTIBLE);
444 	rpc_clnt_sigunmask(clnt, &oldmask);
445 
446 	return error;
447 }
448 
449 static void nfs_wake_up_inode(struct inode *inode)
450 {
451 	struct nfs_inode *nfsi = NFS_I(inode);
452 
453 	clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
454 	smp_mb__after_clear_bit();
455 	wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
456 }
457 
458 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
459 {
460 	struct inode *inode = dentry->d_inode;
461 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
462 	int err;
463 
464 	/* Flush out writes to the server in order to update c/mtime */
465 	if (S_ISREG(inode->i_mode))
466 		nfs_wb_nocommit(inode);
467 
468 	/*
469 	 * We may force a getattr if the user cares about atime.
470 	 *
471 	 * Note that we only have to check the vfsmount flags here:
472 	 *  - NFS always sets S_NOATIME by so checking it would give a
473 	 *    bogus result
474 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
475 	 *    no point in checking those.
476 	 */
477  	if ((mnt->mnt_flags & MNT_NOATIME) ||
478  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
479 		need_atime = 0;
480 
481 	if (need_atime)
482 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
483 	else
484 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
485 	if (!err) {
486 		generic_fillattr(inode, stat);
487 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
488 	}
489 	return err;
490 }
491 
492 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
493 {
494 	struct nfs_open_context *ctx;
495 
496 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
497 	if (ctx != NULL) {
498 		ctx->path.dentry = dget(dentry);
499 		ctx->path.mnt = mntget(mnt);
500 		ctx->cred = get_rpccred(cred);
501 		ctx->state = NULL;
502 		ctx->lockowner = current->files;
503 		ctx->error = 0;
504 		ctx->dir_cookie = 0;
505 		atomic_set(&ctx->count, 1);
506 	}
507 	return ctx;
508 }
509 
510 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
511 {
512 	if (ctx != NULL)
513 		atomic_inc(&ctx->count);
514 	return ctx;
515 }
516 
517 static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait)
518 {
519 	struct inode *inode = ctx->path.dentry->d_inode;
520 
521 	if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
522 		return;
523 	list_del(&ctx->list);
524 	spin_unlock(&inode->i_lock);
525 	if (ctx->state != NULL) {
526 		if (wait)
527 			nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
528 		else
529 			nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
530 	}
531 	if (ctx->cred != NULL)
532 		put_rpccred(ctx->cred);
533 	dput(ctx->path.dentry);
534 	mntput(ctx->path.mnt);
535 	kfree(ctx);
536 }
537 
538 void put_nfs_open_context(struct nfs_open_context *ctx)
539 {
540 	__put_nfs_open_context(ctx, 0);
541 }
542 
543 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
544 {
545 	__put_nfs_open_context(ctx, 1);
546 }
547 
548 /*
549  * Ensure that mmap has a recent RPC credential for use when writing out
550  * shared pages
551  */
552 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
553 {
554 	struct inode *inode = filp->f_path.dentry->d_inode;
555 	struct nfs_inode *nfsi = NFS_I(inode);
556 
557 	filp->private_data = get_nfs_open_context(ctx);
558 	spin_lock(&inode->i_lock);
559 	list_add(&ctx->list, &nfsi->open_files);
560 	spin_unlock(&inode->i_lock);
561 }
562 
563 /*
564  * Given an inode, search for an open context with the desired characteristics
565  */
566 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
567 {
568 	struct nfs_inode *nfsi = NFS_I(inode);
569 	struct nfs_open_context *pos, *ctx = NULL;
570 
571 	spin_lock(&inode->i_lock);
572 	list_for_each_entry(pos, &nfsi->open_files, list) {
573 		if (cred != NULL && pos->cred != cred)
574 			continue;
575 		if ((pos->mode & mode) == mode) {
576 			ctx = get_nfs_open_context(pos);
577 			break;
578 		}
579 	}
580 	spin_unlock(&inode->i_lock);
581 	return ctx;
582 }
583 
584 static void nfs_file_clear_open_context(struct file *filp)
585 {
586 	struct inode *inode = filp->f_path.dentry->d_inode;
587 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
588 
589 	if (ctx) {
590 		filp->private_data = NULL;
591 		spin_lock(&inode->i_lock);
592 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
593 		spin_unlock(&inode->i_lock);
594 		put_nfs_open_context_sync(ctx);
595 	}
596 }
597 
598 /*
599  * These allocate and release file read/write context information.
600  */
601 int nfs_open(struct inode *inode, struct file *filp)
602 {
603 	struct nfs_open_context *ctx;
604 	struct rpc_cred *cred;
605 
606 	cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
607 	if (IS_ERR(cred))
608 		return PTR_ERR(cred);
609 	ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
610 	put_rpccred(cred);
611 	if (ctx == NULL)
612 		return -ENOMEM;
613 	ctx->mode = filp->f_mode;
614 	nfs_file_set_open_context(filp, ctx);
615 	put_nfs_open_context(ctx);
616 	return 0;
617 }
618 
619 int nfs_release(struct inode *inode, struct file *filp)
620 {
621 	nfs_file_clear_open_context(filp);
622 	return 0;
623 }
624 
625 /*
626  * This function is called whenever some part of NFS notices that
627  * the cached attributes have to be refreshed.
628  */
629 int
630 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
631 {
632 	int		 status = -ESTALE;
633 	struct nfs_fattr fattr;
634 	struct nfs_inode *nfsi = NFS_I(inode);
635 
636 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
637 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
638 
639 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
640 	lock_kernel();
641 	if (is_bad_inode(inode))
642  		goto out_nowait;
643 	if (NFS_STALE(inode))
644  		goto out_nowait;
645 
646 	status = nfs_wait_on_inode(inode);
647 	if (status < 0)
648 		goto out;
649 
650 	status = -ESTALE;
651 	if (NFS_STALE(inode))
652 		goto out;
653 
654 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
655 	if (status != 0) {
656 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
657 			 inode->i_sb->s_id,
658 			 (long long)NFS_FILEID(inode), status);
659 		if (status == -ESTALE) {
660 			nfs_zap_caches(inode);
661 			if (!S_ISDIR(inode->i_mode))
662 				set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
663 		}
664 		goto out;
665 	}
666 
667 	spin_lock(&inode->i_lock);
668 	status = nfs_update_inode(inode, &fattr);
669 	if (status) {
670 		spin_unlock(&inode->i_lock);
671 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
672 			 inode->i_sb->s_id,
673 			 (long long)NFS_FILEID(inode), status);
674 		goto out;
675 	}
676 	spin_unlock(&inode->i_lock);
677 
678 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
679 		nfs_zap_acl_cache(inode);
680 
681 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
682 		inode->i_sb->s_id,
683 		(long long)NFS_FILEID(inode));
684 
685  out:
686 	nfs_wake_up_inode(inode);
687 
688  out_nowait:
689 	unlock_kernel();
690 	return status;
691 }
692 
693 int nfs_attribute_timeout(struct inode *inode)
694 {
695 	struct nfs_inode *nfsi = NFS_I(inode);
696 
697 	if (nfs_have_delegation(inode, FMODE_READ))
698 		return 0;
699 	return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
700 }
701 
702 /**
703  * nfs_revalidate_inode - Revalidate the inode attributes
704  * @server - pointer to nfs_server struct
705  * @inode - pointer to inode struct
706  *
707  * Updates inode attribute information by retrieving the data from the server.
708  */
709 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
710 {
711 	if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
712 			&& !nfs_attribute_timeout(inode))
713 		return NFS_STALE(inode) ? -ESTALE : 0;
714 	return __nfs_revalidate_inode(server, inode);
715 }
716 
717 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
718 {
719 	struct nfs_inode *nfsi = NFS_I(inode);
720 
721 	if (mapping->nrpages != 0) {
722 		int ret = invalidate_inode_pages2(mapping);
723 		if (ret < 0)
724 			return ret;
725 	}
726 	spin_lock(&inode->i_lock);
727 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
728 	if (S_ISDIR(inode->i_mode))
729 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
730 	spin_unlock(&inode->i_lock);
731 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
732 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
733 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
734 	return 0;
735 }
736 
737 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
738 {
739 	int ret = 0;
740 
741 	mutex_lock(&inode->i_mutex);
742 	if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
743 		ret = nfs_sync_mapping(mapping);
744 		if (ret == 0)
745 			ret = nfs_invalidate_mapping_nolock(inode, mapping);
746 	}
747 	mutex_unlock(&inode->i_mutex);
748 	return ret;
749 }
750 
751 /**
752  * nfs_revalidate_mapping_nolock - Revalidate the pagecache
753  * @inode - pointer to host inode
754  * @mapping - pointer to mapping
755  */
756 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
757 {
758 	struct nfs_inode *nfsi = NFS_I(inode);
759 	int ret = 0;
760 
761 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
762 			|| nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
763 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
764 		if (ret < 0)
765 			goto out;
766 	}
767 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
768 		ret = nfs_invalidate_mapping_nolock(inode, mapping);
769 out:
770 	return ret;
771 }
772 
773 /**
774  * nfs_revalidate_mapping - Revalidate the pagecache
775  * @inode - pointer to host inode
776  * @mapping - pointer to mapping
777  *
778  * This version of the function will take the inode->i_mutex and attempt to
779  * flush out all dirty data if it needs to invalidate the page cache.
780  */
781 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
782 {
783 	struct nfs_inode *nfsi = NFS_I(inode);
784 	int ret = 0;
785 
786 	if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
787 			|| nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
788 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
789 		if (ret < 0)
790 			goto out;
791 	}
792 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
793 		ret = nfs_invalidate_mapping(inode, mapping);
794 out:
795 	return ret;
796 }
797 
798 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
799 {
800 	struct nfs_inode *nfsi = NFS_I(inode);
801 
802 	if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
803 			nfsi->change_attr == fattr->pre_change_attr) {
804 		nfsi->change_attr = fattr->change_attr;
805 		if (S_ISDIR(inode->i_mode))
806 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
807 	}
808 	/* If we have atomic WCC data, we may update some attributes */
809 	if ((fattr->valid & NFS_ATTR_WCC) != 0) {
810 		if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
811 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
812 		if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
813 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
814 			if (S_ISDIR(inode->i_mode))
815 				nfsi->cache_validity |= NFS_INO_INVALID_DATA;
816 		}
817 		if (inode->i_size == fattr->pre_size && nfsi->npages == 0)
818 			inode->i_size = fattr->size;
819 	}
820 }
821 
822 /**
823  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
824  * @inode - pointer to inode
825  * @fattr - updated attributes
826  *
827  * Verifies the attribute cache. If we have just changed the attributes,
828  * so that fattr carries weak cache consistency data, then it may
829  * also update the ctime/mtime/change_attribute.
830  */
831 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
832 {
833 	struct nfs_inode *nfsi = NFS_I(inode);
834 	loff_t cur_size, new_isize;
835 	unsigned long invalid = 0;
836 
837 
838 	/* Has the inode gone and changed behind our back? */
839 	if (nfsi->fileid != fattr->fileid
840 			|| (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
841 		return -EIO;
842 	}
843 
844 	/* Do atomic weak cache consistency updates */
845 	nfs_wcc_update_inode(inode, fattr);
846 
847 	if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
848 			nfsi->change_attr != fattr->change_attr)
849 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
850 
851 	/* Verify a few of the more important attributes */
852 	if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
853 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
854 
855 	cur_size = i_size_read(inode);
856  	new_isize = nfs_size_to_loff_t(fattr->size);
857 	if (cur_size != new_isize && nfsi->npages == 0)
858 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
859 
860 	/* Have any file permissions changed? */
861 	if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
862 			|| inode->i_uid != fattr->uid
863 			|| inode->i_gid != fattr->gid)
864 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
865 
866 	/* Has the link count changed? */
867 	if (inode->i_nlink != fattr->nlink)
868 		invalid |= NFS_INO_INVALID_ATTR;
869 
870 	if (!timespec_equal(&inode->i_atime, &fattr->atime))
871 		invalid |= NFS_INO_INVALID_ATIME;
872 
873 	if (invalid != 0)
874 		nfsi->cache_validity |= invalid;
875 	else
876 		nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
877 				| NFS_INO_INVALID_ATIME
878 				| NFS_INO_REVAL_PAGECACHE);
879 
880 	nfsi->read_cache_jiffies = fattr->time_start;
881 	return 0;
882 }
883 
884 /**
885  * nfs_refresh_inode - try to update the inode attribute cache
886  * @inode - pointer to inode
887  * @fattr - updated attributes
888  *
889  * Check that an RPC call that returned attributes has not overlapped with
890  * other recent updates of the inode metadata, then decide whether it is
891  * safe to do a full update of the inode attributes, or whether just to
892  * call nfs_check_inode_attributes.
893  */
894 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
895 {
896 	struct nfs_inode *nfsi = NFS_I(inode);
897 	int status;
898 
899 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
900 		return 0;
901 	spin_lock(&inode->i_lock);
902 	if (time_after(fattr->time_start, nfsi->last_updated))
903 		status = nfs_update_inode(inode, fattr);
904 	else
905 		status = nfs_check_inode_attributes(inode, fattr);
906 
907 	spin_unlock(&inode->i_lock);
908 	return status;
909 }
910 
911 /**
912  * nfs_post_op_update_inode - try to update the inode attribute cache
913  * @inode - pointer to inode
914  * @fattr - updated attributes
915  *
916  * After an operation that has changed the inode metadata, mark the
917  * attribute cache as being invalid, then try to update it.
918  *
919  * NB: if the server didn't return any post op attributes, this
920  * function will force the retrieval of attributes before the next
921  * NFS request.  Thus it should be used only for operations that
922  * are expected to change one or more attributes, to avoid
923  * unnecessary NFS requests and trips through nfs_update_inode().
924  */
925 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
926 {
927 	struct nfs_inode *nfsi = NFS_I(inode);
928 
929 	spin_lock(&inode->i_lock);
930 	nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
931 	if (S_ISDIR(inode->i_mode))
932 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
933 	spin_unlock(&inode->i_lock);
934 	return nfs_refresh_inode(inode, fattr);
935 }
936 
937 /**
938  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
939  * @inode - pointer to inode
940  * @fattr - updated attributes
941  *
942  * After an operation that has changed the inode metadata, mark the
943  * attribute cache as being invalid, then try to update it. Fake up
944  * weak cache consistency data, if none exist.
945  *
946  * This function is mainly designed to be used by the ->write_done() functions.
947  */
948 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
949 {
950 	if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
951 			(fattr->valid & NFS_ATTR_WCC_V4) == 0) {
952 		fattr->pre_change_attr = NFS_I(inode)->change_attr;
953 		fattr->valid |= NFS_ATTR_WCC_V4;
954 	}
955 	if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
956 			(fattr->valid & NFS_ATTR_WCC) == 0) {
957 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
958 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
959 		fattr->pre_size = inode->i_size;
960 		fattr->valid |= NFS_ATTR_WCC;
961 	}
962 	return nfs_post_op_update_inode(inode, fattr);
963 }
964 
965 /*
966  * Many nfs protocol calls return the new file attributes after
967  * an operation.  Here we update the inode to reflect the state
968  * of the server's inode.
969  *
970  * This is a bit tricky because we have to make sure all dirty pages
971  * have been sent off to the server before calling invalidate_inode_pages.
972  * To make sure no other process adds more write requests while we try
973  * our best to flush them, we make them sleep during the attribute refresh.
974  *
975  * A very similar scenario holds for the dir cache.
976  */
977 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
978 {
979 	struct nfs_server *server;
980 	struct nfs_inode *nfsi = NFS_I(inode);
981 	loff_t cur_isize, new_isize;
982 	unsigned long invalid = 0;
983 	unsigned long now = jiffies;
984 
985 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
986 			__FUNCTION__, inode->i_sb->s_id, inode->i_ino,
987 			atomic_read(&inode->i_count), fattr->valid);
988 
989 	if (nfsi->fileid != fattr->fileid)
990 		goto out_fileid;
991 
992 	/*
993 	 * Make sure the inode's type hasn't changed.
994 	 */
995 	if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
996 		goto out_changed;
997 
998 	server = NFS_SERVER(inode);
999 	/* Update the fsid? */
1000 	if (S_ISDIR(inode->i_mode)
1001 			&& !nfs_fsid_equal(&server->fsid, &fattr->fsid))
1002 		server->fsid = fattr->fsid;
1003 
1004 	/*
1005 	 * Update the read time so we don't revalidate too often.
1006 	 */
1007 	nfsi->read_cache_jiffies = fattr->time_start;
1008 
1009 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
1010 			| NFS_INO_REVAL_PAGECACHE);
1011 
1012 	/* Do atomic weak cache consistency updates */
1013 	nfs_wcc_update_inode(inode, fattr);
1014 
1015 	/* More cache consistency checks */
1016 	if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
1017 		/* NFSv2/v3: Check if the mtime agrees */
1018 		if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1019 			dprintk("NFS: mtime change on server for file %s/%ld\n",
1020 					inode->i_sb->s_id, inode->i_ino);
1021 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1022 			nfsi->cache_change_attribute = now;
1023 		}
1024 		/* If ctime has changed we should definitely clear access+acl caches */
1025 		if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
1026 			invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1027 	} else if (nfsi->change_attr != fattr->change_attr) {
1028 		dprintk("NFS: change_attr change on server for file %s/%ld\n",
1029 				inode->i_sb->s_id, inode->i_ino);
1030 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1031 		nfsi->cache_change_attribute = now;
1032 	}
1033 
1034 	/* Check if our cached file size is stale */
1035  	new_isize = nfs_size_to_loff_t(fattr->size);
1036 	cur_isize = i_size_read(inode);
1037 	if (new_isize != cur_isize) {
1038 		/* Do we perhaps have any outstanding writes, or has
1039 		 * the file grown beyond our last write? */
1040 		if (nfsi->npages == 0 || new_isize > cur_isize) {
1041 			inode->i_size = new_isize;
1042 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1043 		}
1044 		dprintk("NFS: isize change on server for file %s/%ld\n",
1045 				inode->i_sb->s_id, inode->i_ino);
1046 	}
1047 
1048 
1049 	memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1050 	memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1051 	memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1052 	nfsi->change_attr = fattr->change_attr;
1053 
1054 	if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1055 	    inode->i_uid != fattr->uid ||
1056 	    inode->i_gid != fattr->gid)
1057 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1058 
1059 	inode->i_mode = fattr->mode;
1060 	inode->i_nlink = fattr->nlink;
1061 	inode->i_uid = fattr->uid;
1062 	inode->i_gid = fattr->gid;
1063 
1064 	if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1065 		/*
1066 		 * report the blocks in 512byte units
1067 		 */
1068 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1069  	} else {
1070  		inode->i_blocks = fattr->du.nfs2.blocks;
1071  	}
1072 
1073 	/* Update attrtimeo value if we're out of the unstable period */
1074 	if (invalid & NFS_INO_INVALID_ATTR) {
1075 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1076 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1077 		nfsi->attrtimeo_timestamp = now;
1078 		nfsi->last_updated = now;
1079 	} else {
1080 		if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1081 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1082 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1083 			nfsi->attrtimeo_timestamp = now;
1084 		}
1085 		/*
1086 		 * Avoid jiffy wraparound issues with nfsi->last_updated
1087 		 */
1088 		if (!time_in_range(nfsi->last_updated, nfsi->read_cache_jiffies, now))
1089 			nfsi->last_updated = nfsi->read_cache_jiffies;
1090 	}
1091 	invalid &= ~NFS_INO_INVALID_ATTR;
1092 	/* Don't invalidate the data if we were to blame */
1093 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1094 				|| S_ISLNK(inode->i_mode)))
1095 		invalid &= ~NFS_INO_INVALID_DATA;
1096 	if (!nfs_have_delegation(inode, FMODE_READ) ||
1097 			(nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1098 		nfsi->cache_validity |= invalid;
1099 	nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1100 
1101 	return 0;
1102  out_changed:
1103 	/*
1104 	 * Big trouble! The inode has become a different object.
1105 	 */
1106 	printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1107 			__FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1108  out_err:
1109 	/*
1110 	 * No need to worry about unhashing the dentry, as the
1111 	 * lookup validation will know that the inode is bad.
1112 	 * (But we fall through to invalidate the caches.)
1113 	 */
1114 	nfs_invalidate_inode(inode);
1115 	return -ESTALE;
1116 
1117  out_fileid:
1118 	printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1119 		"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1120 		NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1121 		(long long)nfsi->fileid, (long long)fattr->fileid);
1122 	goto out_err;
1123 }
1124 
1125 
1126 #ifdef CONFIG_NFS_V4
1127 
1128 /*
1129  * Clean out any remaining NFSv4 state that might be left over due
1130  * to open() calls that passed nfs_atomic_lookup, but failed to call
1131  * nfs_open().
1132  */
1133 void nfs4_clear_inode(struct inode *inode)
1134 {
1135 	/* If we are holding a delegation, return it! */
1136 	nfs_inode_return_delegation(inode);
1137 	/* First call standard NFS clear_inode() code */
1138 	nfs_clear_inode(inode);
1139 }
1140 #endif
1141 
1142 struct inode *nfs_alloc_inode(struct super_block *sb)
1143 {
1144 	struct nfs_inode *nfsi;
1145 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1146 	if (!nfsi)
1147 		return NULL;
1148 	nfsi->flags = 0UL;
1149 	nfsi->cache_validity = 0UL;
1150 #ifdef CONFIG_NFS_V3_ACL
1151 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1152 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1153 #endif
1154 #ifdef CONFIG_NFS_V4
1155 	nfsi->nfs4_acl = NULL;
1156 #endif /* CONFIG_NFS_V4 */
1157 	return &nfsi->vfs_inode;
1158 }
1159 
1160 void nfs_destroy_inode(struct inode *inode)
1161 {
1162 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1163 }
1164 
1165 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1166 {
1167 #ifdef CONFIG_NFS_V4
1168 	INIT_LIST_HEAD(&nfsi->open_states);
1169 	nfsi->delegation = NULL;
1170 	nfsi->delegation_state = 0;
1171 	init_rwsem(&nfsi->rwsem);
1172 #endif
1173 }
1174 
1175 static void init_once(struct kmem_cache * cachep, void *foo)
1176 {
1177 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1178 
1179 	inode_init_once(&nfsi->vfs_inode);
1180 	INIT_LIST_HEAD(&nfsi->open_files);
1181 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1182 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1183 	INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1184 	nfsi->ncommit = 0;
1185 	nfsi->npages = 0;
1186 	atomic_set(&nfsi->silly_count, 1);
1187 	INIT_HLIST_HEAD(&nfsi->silly_list);
1188 	init_waitqueue_head(&nfsi->waitqueue);
1189 	nfs4_init_once(nfsi);
1190 }
1191 
1192 static int __init nfs_init_inodecache(void)
1193 {
1194 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1195 					     sizeof(struct nfs_inode),
1196 					     0, (SLAB_RECLAIM_ACCOUNT|
1197 						SLAB_MEM_SPREAD),
1198 					     init_once);
1199 	if (nfs_inode_cachep == NULL)
1200 		return -ENOMEM;
1201 
1202 	return 0;
1203 }
1204 
1205 static void nfs_destroy_inodecache(void)
1206 {
1207 	kmem_cache_destroy(nfs_inode_cachep);
1208 }
1209 
1210 /*
1211  * Initialize NFS
1212  */
1213 static int __init init_nfs_fs(void)
1214 {
1215 	int err;
1216 
1217 	err = nfs_fs_proc_init();
1218 	if (err)
1219 		goto out5;
1220 
1221 	err = nfs_init_nfspagecache();
1222 	if (err)
1223 		goto out4;
1224 
1225 	err = nfs_init_inodecache();
1226 	if (err)
1227 		goto out3;
1228 
1229 	err = nfs_init_readpagecache();
1230 	if (err)
1231 		goto out2;
1232 
1233 	err = nfs_init_writepagecache();
1234 	if (err)
1235 		goto out1;
1236 
1237 	err = nfs_init_directcache();
1238 	if (err)
1239 		goto out0;
1240 
1241 #ifdef CONFIG_PROC_FS
1242 	rpc_proc_register(&nfs_rpcstat);
1243 #endif
1244 	if ((err = register_nfs_fs()) != 0)
1245 		goto out;
1246 	return 0;
1247 out:
1248 #ifdef CONFIG_PROC_FS
1249 	rpc_proc_unregister("nfs");
1250 #endif
1251 	nfs_destroy_directcache();
1252 out0:
1253 	nfs_destroy_writepagecache();
1254 out1:
1255 	nfs_destroy_readpagecache();
1256 out2:
1257 	nfs_destroy_inodecache();
1258 out3:
1259 	nfs_destroy_nfspagecache();
1260 out4:
1261 	nfs_fs_proc_exit();
1262 out5:
1263 	return err;
1264 }
1265 
1266 static void __exit exit_nfs_fs(void)
1267 {
1268 	nfs_destroy_directcache();
1269 	nfs_destroy_writepagecache();
1270 	nfs_destroy_readpagecache();
1271 	nfs_destroy_inodecache();
1272 	nfs_destroy_nfspagecache();
1273 #ifdef CONFIG_PROC_FS
1274 	rpc_proc_unregister("nfs");
1275 #endif
1276 	unregister_nfs_fs();
1277 	nfs_fs_proc_exit();
1278 }
1279 
1280 /* Not quite true; I just maintain it */
1281 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1282 MODULE_LICENSE("GPL");
1283 module_param(enable_ino64, bool, 0644);
1284 
1285 module_init(init_nfs_fs)
1286 module_exit(exit_nfs_fs)
1287