xref: /linux/fs/nfs/inode.c (revision 4603f53a1dc3c76dfba841d123db9fa6204934f5)
1 /*
2  *  linux/fs/nfs/inode.c
3  *
4  *  Copyright (C) 1992  Rick Sladkey
5  *
6  *  nfs inode and superblock handling functions
7  *
8  *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9  *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
10  *
11  *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12  *  J.S.Peatfield@damtp.cam.ac.uk
13  *
14  */
15 
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/time.h>
20 #include <linux/kernel.h>
21 #include <linux/mm.h>
22 #include <linux/string.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/unistd.h>
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/metrics.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/vfs.h>
36 #include <linux/inet.h>
37 #include <linux/nfs_xdr.h>
38 #include <linux/slab.h>
39 #include <linux/compat.h>
40 #include <linux/freezer.h>
41 #include <linux/crc32.h>
42 
43 #include <asm/uaccess.h>
44 
45 #include "nfs4_fs.h"
46 #include "callback.h"
47 #include "delegation.h"
48 #include "iostat.h"
49 #include "internal.h"
50 #include "fscache.h"
51 #include "pnfs.h"
52 #include "nfs.h"
53 #include "netns.h"
54 
55 #define NFSDBG_FACILITY		NFSDBG_VFS
56 
57 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
58 
59 /* Default is to see 64-bit inode numbers */
60 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
61 
62 static void nfs_invalidate_inode(struct inode *);
63 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
64 
65 static struct kmem_cache * nfs_inode_cachep;
66 
67 static inline unsigned long
68 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
69 {
70 	return nfs_fileid_to_ino_t(fattr->fileid);
71 }
72 
73 /**
74  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
75  * @word: long word containing the bit lock
76  */
77 int nfs_wait_bit_killable(void *word)
78 {
79 	if (fatal_signal_pending(current))
80 		return -ERESTARTSYS;
81 	freezable_schedule_unsafe();
82 	return 0;
83 }
84 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
85 
86 /**
87  * nfs_compat_user_ino64 - returns the user-visible inode number
88  * @fileid: 64-bit fileid
89  *
90  * This function returns a 32-bit inode number if the boot parameter
91  * nfs.enable_ino64 is zero.
92  */
93 u64 nfs_compat_user_ino64(u64 fileid)
94 {
95 #ifdef CONFIG_COMPAT
96 	compat_ulong_t ino;
97 #else
98 	unsigned long ino;
99 #endif
100 
101 	if (enable_ino64)
102 		return fileid;
103 	ino = fileid;
104 	if (sizeof(ino) < sizeof(fileid))
105 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
106 	return ino;
107 }
108 
109 int nfs_drop_inode(struct inode *inode)
110 {
111 	return NFS_STALE(inode) || generic_drop_inode(inode);
112 }
113 EXPORT_SYMBOL_GPL(nfs_drop_inode);
114 
115 void nfs_clear_inode(struct inode *inode)
116 {
117 	/*
118 	 * The following should never happen...
119 	 */
120 	WARN_ON_ONCE(nfs_have_writebacks(inode));
121 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
122 	nfs_zap_acl_cache(inode);
123 	nfs_access_zap_cache(inode);
124 	nfs_fscache_release_inode_cookie(inode);
125 }
126 EXPORT_SYMBOL_GPL(nfs_clear_inode);
127 
128 void nfs_evict_inode(struct inode *inode)
129 {
130 	truncate_inode_pages(&inode->i_data, 0);
131 	clear_inode(inode);
132 	nfs_clear_inode(inode);
133 }
134 
135 /**
136  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
137  */
138 int nfs_sync_mapping(struct address_space *mapping)
139 {
140 	int ret = 0;
141 
142 	if (mapping->nrpages != 0) {
143 		unmap_mapping_range(mapping, 0, 0, 0);
144 		ret = nfs_wb_all(mapping->host);
145 	}
146 	return ret;
147 }
148 
149 /*
150  * Invalidate the local caches
151  */
152 static void nfs_zap_caches_locked(struct inode *inode)
153 {
154 	struct nfs_inode *nfsi = NFS_I(inode);
155 	int mode = inode->i_mode;
156 
157 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
158 
159 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
160 	nfsi->attrtimeo_timestamp = jiffies;
161 
162 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
163 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
164 		nfs_fscache_invalidate(inode);
165 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR
166 					| NFS_INO_INVALID_LABEL
167 					| NFS_INO_INVALID_DATA
168 					| NFS_INO_INVALID_ACCESS
169 					| NFS_INO_INVALID_ACL
170 					| NFS_INO_REVAL_PAGECACHE;
171 	} else
172 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR
173 					| NFS_INO_INVALID_LABEL
174 					| NFS_INO_INVALID_ACCESS
175 					| NFS_INO_INVALID_ACL
176 					| NFS_INO_REVAL_PAGECACHE;
177 }
178 
179 void nfs_zap_caches(struct inode *inode)
180 {
181 	spin_lock(&inode->i_lock);
182 	nfs_zap_caches_locked(inode);
183 	spin_unlock(&inode->i_lock);
184 }
185 
186 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
187 {
188 	if (mapping->nrpages != 0) {
189 		spin_lock(&inode->i_lock);
190 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
191 		nfs_fscache_invalidate(inode);
192 		spin_unlock(&inode->i_lock);
193 	}
194 }
195 
196 void nfs_zap_acl_cache(struct inode *inode)
197 {
198 	void (*clear_acl_cache)(struct inode *);
199 
200 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
201 	if (clear_acl_cache != NULL)
202 		clear_acl_cache(inode);
203 	spin_lock(&inode->i_lock);
204 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
205 	spin_unlock(&inode->i_lock);
206 }
207 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
208 
209 void nfs_invalidate_atime(struct inode *inode)
210 {
211 	spin_lock(&inode->i_lock);
212 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
213 	spin_unlock(&inode->i_lock);
214 }
215 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
216 
217 /*
218  * Invalidate, but do not unhash, the inode.
219  * NB: must be called with inode->i_lock held!
220  */
221 static void nfs_invalidate_inode(struct inode *inode)
222 {
223 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
224 	nfs_zap_caches_locked(inode);
225 }
226 
227 struct nfs_find_desc {
228 	struct nfs_fh		*fh;
229 	struct nfs_fattr	*fattr;
230 };
231 
232 /*
233  * In NFSv3 we can have 64bit inode numbers. In order to support
234  * this, and re-exported directories (also seen in NFSv2)
235  * we are forced to allow 2 different inodes to have the same
236  * i_ino.
237  */
238 static int
239 nfs_find_actor(struct inode *inode, void *opaque)
240 {
241 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
242 	struct nfs_fh		*fh = desc->fh;
243 	struct nfs_fattr	*fattr = desc->fattr;
244 
245 	if (NFS_FILEID(inode) != fattr->fileid)
246 		return 0;
247 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
248 		return 0;
249 	if (nfs_compare_fh(NFS_FH(inode), fh))
250 		return 0;
251 	if (is_bad_inode(inode) || NFS_STALE(inode))
252 		return 0;
253 	return 1;
254 }
255 
256 static int
257 nfs_init_locked(struct inode *inode, void *opaque)
258 {
259 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
260 	struct nfs_fattr	*fattr = desc->fattr;
261 
262 	set_nfs_fileid(inode, fattr->fileid);
263 	nfs_copy_fh(NFS_FH(inode), desc->fh);
264 	return 0;
265 }
266 
267 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
268 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
269 					struct nfs4_label *label)
270 {
271 	int error;
272 
273 	if (label == NULL)
274 		return;
275 
276 	if (nfs_server_capable(inode, NFS_CAP_SECURITY_LABEL) == 0)
277 		return;
278 
279 	if (NFS_SERVER(inode)->nfs_client->cl_minorversion < 2)
280 		return;
281 
282 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
283 		error = security_inode_notifysecctx(inode, label->label,
284 				label->len);
285 		if (error)
286 			printk(KERN_ERR "%s() %s %d "
287 					"security_inode_notifysecctx() %d\n",
288 					__func__,
289 					(char *)label->label,
290 					label->len, error);
291 	}
292 }
293 
294 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
295 {
296 	struct nfs4_label *label = NULL;
297 	int minor_version = server->nfs_client->cl_minorversion;
298 
299 	if (minor_version < 2)
300 		return label;
301 
302 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
303 		return label;
304 
305 	label = kzalloc(sizeof(struct nfs4_label), flags);
306 	if (label == NULL)
307 		return ERR_PTR(-ENOMEM);
308 
309 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
310 	if (label->label == NULL) {
311 		kfree(label);
312 		return ERR_PTR(-ENOMEM);
313 	}
314 	label->len = NFS4_MAXLABELLEN;
315 
316 	return label;
317 }
318 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
319 #else
320 void inline nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
321 					struct nfs4_label *label)
322 {
323 }
324 #endif
325 EXPORT_SYMBOL_GPL(nfs_setsecurity);
326 
327 /*
328  * This is our front-end to iget that looks up inodes by file handle
329  * instead of inode number.
330  */
331 struct inode *
332 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
333 {
334 	struct nfs_find_desc desc = {
335 		.fh	= fh,
336 		.fattr	= fattr
337 	};
338 	struct inode *inode = ERR_PTR(-ENOENT);
339 	unsigned long hash;
340 
341 	nfs_attr_check_mountpoint(sb, fattr);
342 
343 	if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
344 	    !nfs_attr_use_mounted_on_fileid(fattr))
345 		goto out_no_inode;
346 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
347 		goto out_no_inode;
348 
349 	hash = nfs_fattr_to_ino_t(fattr);
350 
351 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
352 	if (inode == NULL) {
353 		inode = ERR_PTR(-ENOMEM);
354 		goto out_no_inode;
355 	}
356 
357 	if (inode->i_state & I_NEW) {
358 		struct nfs_inode *nfsi = NFS_I(inode);
359 		unsigned long now = jiffies;
360 
361 		/* We set i_ino for the few things that still rely on it,
362 		 * such as stat(2) */
363 		inode->i_ino = hash;
364 
365 		/* We can't support update_atime(), since the server will reset it */
366 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
367 		inode->i_mode = fattr->mode;
368 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
369 				&& nfs_server_capable(inode, NFS_CAP_MODE))
370 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
371 		/* Why so? Because we want revalidate for devices/FIFOs, and
372 		 * that's precisely what we have in nfs_file_inode_operations.
373 		 */
374 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
375 		if (S_ISREG(inode->i_mode)) {
376 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
377 			inode->i_data.a_ops = &nfs_file_aops;
378 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
379 		} else if (S_ISDIR(inode->i_mode)) {
380 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
381 			inode->i_fop = &nfs_dir_operations;
382 			inode->i_data.a_ops = &nfs_dir_aops;
383 			/* Deal with crossing mountpoints */
384 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
385 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
386 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
387 					inode->i_op = &nfs_referral_inode_operations;
388 				else
389 					inode->i_op = &nfs_mountpoint_inode_operations;
390 				inode->i_fop = NULL;
391 				inode->i_flags |= S_AUTOMOUNT;
392 			}
393 		} else if (S_ISLNK(inode->i_mode))
394 			inode->i_op = &nfs_symlink_inode_operations;
395 		else
396 			init_special_inode(inode, inode->i_mode, fattr->rdev);
397 
398 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
399 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
400 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
401 		inode->i_version = 0;
402 		inode->i_size = 0;
403 		clear_nlink(inode);
404 		inode->i_uid = make_kuid(&init_user_ns, -2);
405 		inode->i_gid = make_kgid(&init_user_ns, -2);
406 		inode->i_blocks = 0;
407 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
408 		nfsi->write_io = 0;
409 		nfsi->read_io = 0;
410 
411 		nfsi->read_cache_jiffies = fattr->time_start;
412 		nfsi->attr_gencount = fattr->gencount;
413 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
414 			inode->i_atime = fattr->atime;
415 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
416 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
417 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
418 			inode->i_mtime = fattr->mtime;
419 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
420 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
421 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
422 			inode->i_ctime = fattr->ctime;
423 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
424 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
425 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
426 			inode->i_version = fattr->change_attr;
427 		else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
428 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
429 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
430 			inode->i_size = nfs_size_to_loff_t(fattr->size);
431 		else
432 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
433 				| NFS_INO_REVAL_PAGECACHE;
434 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
435 			set_nlink(inode, fattr->nlink);
436 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
437 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
438 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
439 			inode->i_uid = fattr->uid;
440 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
441 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
442 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
443 			inode->i_gid = fattr->gid;
444 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
445 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
446 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
447 			inode->i_blocks = fattr->du.nfs2.blocks;
448 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
449 			/*
450 			 * report the blocks in 512byte units
451 			 */
452 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
453 		}
454 
455 		nfs_setsecurity(inode, fattr, label);
456 
457 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
458 		nfsi->attrtimeo_timestamp = now;
459 		nfsi->access_cache = RB_ROOT;
460 
461 		nfs_fscache_init_inode_cookie(inode);
462 
463 		unlock_new_inode(inode);
464 	} else
465 		nfs_refresh_inode(inode, fattr);
466 		nfs_setsecurity(inode, fattr, label);
467 	dprintk("NFS: nfs_fhget(%s/%Ld fh_crc=0x%08x ct=%d)\n",
468 		inode->i_sb->s_id,
469 		(long long)NFS_FILEID(inode),
470 		nfs_display_fhandle_hash(fh),
471 		atomic_read(&inode->i_count));
472 
473 out:
474 	return inode;
475 
476 out_no_inode:
477 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
478 	goto out;
479 }
480 EXPORT_SYMBOL_GPL(nfs_fhget);
481 
482 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN)
483 
484 int
485 nfs_setattr(struct dentry *dentry, struct iattr *attr)
486 {
487 	struct inode *inode = dentry->d_inode;
488 	struct nfs_fattr *fattr;
489 	int error = -ENOMEM;
490 
491 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
492 
493 	/* skip mode change if it's just for clearing setuid/setgid */
494 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
495 		attr->ia_valid &= ~ATTR_MODE;
496 
497 	if (attr->ia_valid & ATTR_SIZE) {
498 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
499 			attr->ia_valid &= ~ATTR_SIZE;
500 	}
501 
502 	/* Optimization: if the end result is no change, don't RPC */
503 	attr->ia_valid &= NFS_VALID_ATTRS;
504 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
505 		return 0;
506 
507 	/* Write all dirty data */
508 	if (S_ISREG(inode->i_mode)) {
509 		nfs_inode_dio_wait(inode);
510 		nfs_wb_all(inode);
511 	}
512 
513 	fattr = nfs_alloc_fattr();
514 	if (fattr == NULL)
515 		goto out;
516 	/*
517 	 * Return any delegations if we're going to change ACLs
518 	 */
519 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
520 		NFS_PROTO(inode)->return_delegation(inode);
521 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
522 	if (error == 0)
523 		error = nfs_refresh_inode(inode, fattr);
524 	nfs_free_fattr(fattr);
525 out:
526 	return error;
527 }
528 EXPORT_SYMBOL_GPL(nfs_setattr);
529 
530 /**
531  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
532  * @inode: inode of the file used
533  * @offset: file offset to start truncating
534  *
535  * This is a copy of the common vmtruncate, but with the locking
536  * corrected to take into account the fact that NFS requires
537  * inode->i_size to be updated under the inode->i_lock.
538  */
539 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
540 {
541 	loff_t oldsize;
542 	int err;
543 
544 	err = inode_newsize_ok(inode, offset);
545 	if (err)
546 		goto out;
547 
548 	spin_lock(&inode->i_lock);
549 	oldsize = inode->i_size;
550 	i_size_write(inode, offset);
551 	spin_unlock(&inode->i_lock);
552 
553 	truncate_pagecache(inode, oldsize, offset);
554 out:
555 	return err;
556 }
557 
558 /**
559  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
560  * @inode: pointer to struct inode
561  * @attr: pointer to struct iattr
562  *
563  * Note: we do this in the *proc.c in order to ensure that
564  *       it works for things like exclusive creates too.
565  */
566 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
567 {
568 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
569 		spin_lock(&inode->i_lock);
570 		if ((attr->ia_valid & ATTR_MODE) != 0) {
571 			int mode = attr->ia_mode & S_IALLUGO;
572 			mode |= inode->i_mode & ~S_IALLUGO;
573 			inode->i_mode = mode;
574 		}
575 		if ((attr->ia_valid & ATTR_UID) != 0)
576 			inode->i_uid = attr->ia_uid;
577 		if ((attr->ia_valid & ATTR_GID) != 0)
578 			inode->i_gid = attr->ia_gid;
579 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
580 		spin_unlock(&inode->i_lock);
581 	}
582 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
583 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
584 		nfs_vmtruncate(inode, attr->ia_size);
585 	}
586 }
587 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
588 
589 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
590 {
591 	struct inode *inode = dentry->d_inode;
592 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
593 	int err;
594 
595 	/* Flush out writes to the server in order to update c/mtime.  */
596 	if (S_ISREG(inode->i_mode)) {
597 		nfs_inode_dio_wait(inode);
598 		err = filemap_write_and_wait(inode->i_mapping);
599 		if (err)
600 			goto out;
601 	}
602 
603 	/*
604 	 * We may force a getattr if the user cares about atime.
605 	 *
606 	 * Note that we only have to check the vfsmount flags here:
607 	 *  - NFS always sets S_NOATIME by so checking it would give a
608 	 *    bogus result
609 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
610 	 *    no point in checking those.
611 	 */
612  	if ((mnt->mnt_flags & MNT_NOATIME) ||
613  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
614 		need_atime = 0;
615 
616 	if (need_atime)
617 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
618 	else
619 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
620 	if (!err) {
621 		generic_fillattr(inode, stat);
622 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
623 	}
624 out:
625 	return err;
626 }
627 EXPORT_SYMBOL_GPL(nfs_getattr);
628 
629 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
630 {
631 	atomic_set(&l_ctx->count, 1);
632 	l_ctx->lockowner.l_owner = current->files;
633 	l_ctx->lockowner.l_pid = current->tgid;
634 	INIT_LIST_HEAD(&l_ctx->list);
635 	nfs_iocounter_init(&l_ctx->io_count);
636 }
637 
638 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
639 {
640 	struct nfs_lock_context *head = &ctx->lock_context;
641 	struct nfs_lock_context *pos = head;
642 
643 	do {
644 		if (pos->lockowner.l_owner != current->files)
645 			continue;
646 		if (pos->lockowner.l_pid != current->tgid)
647 			continue;
648 		atomic_inc(&pos->count);
649 		return pos;
650 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
651 	return NULL;
652 }
653 
654 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
655 {
656 	struct nfs_lock_context *res, *new = NULL;
657 	struct inode *inode = ctx->dentry->d_inode;
658 
659 	spin_lock(&inode->i_lock);
660 	res = __nfs_find_lock_context(ctx);
661 	if (res == NULL) {
662 		spin_unlock(&inode->i_lock);
663 		new = kmalloc(sizeof(*new), GFP_KERNEL);
664 		if (new == NULL)
665 			return ERR_PTR(-ENOMEM);
666 		nfs_init_lock_context(new);
667 		spin_lock(&inode->i_lock);
668 		res = __nfs_find_lock_context(ctx);
669 		if (res == NULL) {
670 			list_add_tail(&new->list, &ctx->lock_context.list);
671 			new->open_context = ctx;
672 			res = new;
673 			new = NULL;
674 		}
675 	}
676 	spin_unlock(&inode->i_lock);
677 	kfree(new);
678 	return res;
679 }
680 
681 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
682 {
683 	struct nfs_open_context *ctx = l_ctx->open_context;
684 	struct inode *inode = ctx->dentry->d_inode;
685 
686 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
687 		return;
688 	list_del(&l_ctx->list);
689 	spin_unlock(&inode->i_lock);
690 	kfree(l_ctx);
691 }
692 
693 /**
694  * nfs_close_context - Common close_context() routine NFSv2/v3
695  * @ctx: pointer to context
696  * @is_sync: is this a synchronous close
697  *
698  * always ensure that the attributes are up to date if we're mounted
699  * with close-to-open semantics
700  */
701 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
702 {
703 	struct inode *inode;
704 	struct nfs_server *server;
705 
706 	if (!(ctx->mode & FMODE_WRITE))
707 		return;
708 	if (!is_sync)
709 		return;
710 	inode = ctx->dentry->d_inode;
711 	if (!list_empty(&NFS_I(inode)->open_files))
712 		return;
713 	server = NFS_SERVER(inode);
714 	if (server->flags & NFS_MOUNT_NOCTO)
715 		return;
716 	nfs_revalidate_inode(server, inode);
717 }
718 EXPORT_SYMBOL_GPL(nfs_close_context);
719 
720 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
721 {
722 	struct nfs_open_context *ctx;
723 	struct rpc_cred *cred = rpc_lookup_cred();
724 	if (IS_ERR(cred))
725 		return ERR_CAST(cred);
726 
727 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
728 	if (!ctx) {
729 		put_rpccred(cred);
730 		return ERR_PTR(-ENOMEM);
731 	}
732 	nfs_sb_active(dentry->d_sb);
733 	ctx->dentry = dget(dentry);
734 	ctx->cred = cred;
735 	ctx->state = NULL;
736 	ctx->mode = f_mode;
737 	ctx->flags = 0;
738 	ctx->error = 0;
739 	nfs_init_lock_context(&ctx->lock_context);
740 	ctx->lock_context.open_context = ctx;
741 	INIT_LIST_HEAD(&ctx->list);
742 	ctx->mdsthreshold = NULL;
743 	return ctx;
744 }
745 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
746 
747 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
748 {
749 	if (ctx != NULL)
750 		atomic_inc(&ctx->lock_context.count);
751 	return ctx;
752 }
753 EXPORT_SYMBOL_GPL(get_nfs_open_context);
754 
755 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
756 {
757 	struct inode *inode = ctx->dentry->d_inode;
758 	struct super_block *sb = ctx->dentry->d_sb;
759 
760 	if (!list_empty(&ctx->list)) {
761 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
762 			return;
763 		list_del(&ctx->list);
764 		spin_unlock(&inode->i_lock);
765 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
766 		return;
767 	if (inode != NULL)
768 		NFS_PROTO(inode)->close_context(ctx, is_sync);
769 	if (ctx->cred != NULL)
770 		put_rpccred(ctx->cred);
771 	dput(ctx->dentry);
772 	nfs_sb_deactive(sb);
773 	kfree(ctx->mdsthreshold);
774 	kfree(ctx);
775 }
776 
777 void put_nfs_open_context(struct nfs_open_context *ctx)
778 {
779 	__put_nfs_open_context(ctx, 0);
780 }
781 EXPORT_SYMBOL_GPL(put_nfs_open_context);
782 
783 /*
784  * Ensure that mmap has a recent RPC credential for use when writing out
785  * shared pages
786  */
787 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
788 {
789 	struct inode *inode = ctx->dentry->d_inode;
790 	struct nfs_inode *nfsi = NFS_I(inode);
791 
792 	spin_lock(&inode->i_lock);
793 	list_add(&ctx->list, &nfsi->open_files);
794 	spin_unlock(&inode->i_lock);
795 }
796 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
797 
798 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
799 {
800 	filp->private_data = get_nfs_open_context(ctx);
801 	if (list_empty(&ctx->list))
802 		nfs_inode_attach_open_context(ctx);
803 }
804 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
805 
806 /*
807  * Given an inode, search for an open context with the desired characteristics
808  */
809 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
810 {
811 	struct nfs_inode *nfsi = NFS_I(inode);
812 	struct nfs_open_context *pos, *ctx = NULL;
813 
814 	spin_lock(&inode->i_lock);
815 	list_for_each_entry(pos, &nfsi->open_files, list) {
816 		if (cred != NULL && pos->cred != cred)
817 			continue;
818 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
819 			continue;
820 		ctx = get_nfs_open_context(pos);
821 		break;
822 	}
823 	spin_unlock(&inode->i_lock);
824 	return ctx;
825 }
826 
827 static void nfs_file_clear_open_context(struct file *filp)
828 {
829 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
830 
831 	if (ctx) {
832 		struct inode *inode = ctx->dentry->d_inode;
833 
834 		filp->private_data = NULL;
835 		spin_lock(&inode->i_lock);
836 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
837 		spin_unlock(&inode->i_lock);
838 		__put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
839 	}
840 }
841 
842 /*
843  * These allocate and release file read/write context information.
844  */
845 int nfs_open(struct inode *inode, struct file *filp)
846 {
847 	struct nfs_open_context *ctx;
848 
849 	ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
850 	if (IS_ERR(ctx))
851 		return PTR_ERR(ctx);
852 	nfs_file_set_open_context(filp, ctx);
853 	put_nfs_open_context(ctx);
854 	nfs_fscache_set_inode_cookie(inode, filp);
855 	return 0;
856 }
857 
858 int nfs_release(struct inode *inode, struct file *filp)
859 {
860 	nfs_file_clear_open_context(filp);
861 	return 0;
862 }
863 
864 /*
865  * This function is called whenever some part of NFS notices that
866  * the cached attributes have to be refreshed.
867  */
868 int
869 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
870 {
871 	int		 status = -ESTALE;
872 	struct nfs4_label *label = NULL;
873 	struct nfs_fattr *fattr = NULL;
874 	struct nfs_inode *nfsi = NFS_I(inode);
875 
876 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
877 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
878 
879 	if (is_bad_inode(inode))
880 		goto out;
881 	if (NFS_STALE(inode))
882 		goto out;
883 
884 	status = -ENOMEM;
885 	fattr = nfs_alloc_fattr();
886 	if (fattr == NULL)
887 		goto out;
888 
889 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
890 
891 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
892 	if (IS_ERR(label)) {
893 		status = PTR_ERR(label);
894 		goto out;
895 	}
896 
897 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
898 	if (status != 0) {
899 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
900 			 inode->i_sb->s_id,
901 			 (long long)NFS_FILEID(inode), status);
902 		if (status == -ESTALE) {
903 			nfs_zap_caches(inode);
904 			if (!S_ISDIR(inode->i_mode))
905 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
906 		}
907 		goto err_out;
908 	}
909 
910 	status = nfs_refresh_inode(inode, fattr);
911 	if (status) {
912 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
913 			 inode->i_sb->s_id,
914 			 (long long)NFS_FILEID(inode), status);
915 		goto err_out;
916 	}
917 
918 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
919 		nfs_zap_acl_cache(inode);
920 
921 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
922 		inode->i_sb->s_id,
923 		(long long)NFS_FILEID(inode));
924 
925 err_out:
926 	nfs4_label_free(label);
927 out:
928 	nfs_free_fattr(fattr);
929 	return status;
930 }
931 
932 int nfs_attribute_timeout(struct inode *inode)
933 {
934 	struct nfs_inode *nfsi = NFS_I(inode);
935 
936 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
937 }
938 
939 int nfs_attribute_cache_expired(struct inode *inode)
940 {
941 	if (nfs_have_delegated_attributes(inode))
942 		return 0;
943 	return nfs_attribute_timeout(inode);
944 }
945 
946 /**
947  * nfs_revalidate_inode - Revalidate the inode attributes
948  * @server - pointer to nfs_server struct
949  * @inode - pointer to inode struct
950  *
951  * Updates inode attribute information by retrieving the data from the server.
952  */
953 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
954 {
955 	if (!(NFS_I(inode)->cache_validity &
956 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
957 			&& !nfs_attribute_cache_expired(inode))
958 		return NFS_STALE(inode) ? -ESTALE : 0;
959 	return __nfs_revalidate_inode(server, inode);
960 }
961 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
962 
963 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
964 {
965 	struct nfs_inode *nfsi = NFS_I(inode);
966 
967 	if (mapping->nrpages != 0) {
968 		int ret = invalidate_inode_pages2(mapping);
969 		if (ret < 0)
970 			return ret;
971 	}
972 	spin_lock(&inode->i_lock);
973 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
974 	if (S_ISDIR(inode->i_mode))
975 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
976 	spin_unlock(&inode->i_lock);
977 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
978 	nfs_fscache_wait_on_invalidate(inode);
979 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
980 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
981 	return 0;
982 }
983 
984 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
985 {
986 	if (nfs_have_delegated_attributes(inode))
987 		return false;
988 	return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
989 		|| nfs_attribute_timeout(inode)
990 		|| NFS_STALE(inode);
991 }
992 
993 /**
994  * nfs_revalidate_mapping - Revalidate the pagecache
995  * @inode - pointer to host inode
996  * @mapping - pointer to mapping
997  */
998 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
999 {
1000 	struct nfs_inode *nfsi = NFS_I(inode);
1001 	int ret = 0;
1002 
1003 	/* swapfiles are not supposed to be shared. */
1004 	if (IS_SWAPFILE(inode))
1005 		goto out;
1006 
1007 	if (nfs_mapping_need_revalidate_inode(inode)) {
1008 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1009 		if (ret < 0)
1010 			goto out;
1011 	}
1012 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1013 		ret = nfs_invalidate_mapping(inode, mapping);
1014 out:
1015 	return ret;
1016 }
1017 
1018 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1019 {
1020 	struct nfs_inode *nfsi = NFS_I(inode);
1021 	unsigned long ret = 0;
1022 
1023 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1024 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1025 			&& inode->i_version == fattr->pre_change_attr) {
1026 		inode->i_version = fattr->change_attr;
1027 		if (S_ISDIR(inode->i_mode))
1028 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1029 		ret |= NFS_INO_INVALID_ATTR;
1030 	}
1031 	/* If we have atomic WCC data, we may update some attributes */
1032 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1033 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1034 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1035 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1036 		ret |= NFS_INO_INVALID_ATTR;
1037 	}
1038 
1039 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1040 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1041 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1042 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1043 		if (S_ISDIR(inode->i_mode))
1044 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1045 		ret |= NFS_INO_INVALID_ATTR;
1046 	}
1047 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1048 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1049 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1050 			&& nfsi->npages == 0) {
1051 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1052 		ret |= NFS_INO_INVALID_ATTR;
1053 	}
1054 
1055 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1056 		nfs_fscache_invalidate(inode);
1057 
1058 	return ret;
1059 }
1060 
1061 /**
1062  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1063  * @inode - pointer to inode
1064  * @fattr - updated attributes
1065  *
1066  * Verifies the attribute cache. If we have just changed the attributes,
1067  * so that fattr carries weak cache consistency data, then it may
1068  * also update the ctime/mtime/change_attribute.
1069  */
1070 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1071 {
1072 	struct nfs_inode *nfsi = NFS_I(inode);
1073 	loff_t cur_size, new_isize;
1074 	unsigned long invalid = 0;
1075 
1076 
1077 	if (nfs_have_delegated_attributes(inode))
1078 		return 0;
1079 	/* Has the inode gone and changed behind our back? */
1080 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1081 		return -EIO;
1082 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1083 		return -EIO;
1084 
1085 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1086 			inode->i_version != fattr->change_attr)
1087 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1088 
1089 	/* Verify a few of the more important attributes */
1090 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1091 		invalid |= NFS_INO_INVALID_ATTR;
1092 
1093 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1094 		cur_size = i_size_read(inode);
1095 		new_isize = nfs_size_to_loff_t(fattr->size);
1096 		if (cur_size != new_isize && nfsi->npages == 0)
1097 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1098 	}
1099 
1100 	/* Have any file permissions changed? */
1101 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1102 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1103 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1104 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1105 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1106 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1107 
1108 	/* Has the link count changed? */
1109 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1110 		invalid |= NFS_INO_INVALID_ATTR;
1111 
1112 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1113 		invalid |= NFS_INO_INVALID_ATIME;
1114 
1115 	if (invalid != 0)
1116 		nfsi->cache_validity |= invalid;
1117 
1118 	nfsi->read_cache_jiffies = fattr->time_start;
1119 	return 0;
1120 }
1121 
1122 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1123 {
1124 	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1125 		return 0;
1126 	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1127 }
1128 
1129 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1130 {
1131 	if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1132 		return 0;
1133 	return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1134 }
1135 
1136 static atomic_long_t nfs_attr_generation_counter;
1137 
1138 static unsigned long nfs_read_attr_generation_counter(void)
1139 {
1140 	return atomic_long_read(&nfs_attr_generation_counter);
1141 }
1142 
1143 unsigned long nfs_inc_attr_generation_counter(void)
1144 {
1145 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1146 }
1147 
1148 void nfs_fattr_init(struct nfs_fattr *fattr)
1149 {
1150 	fattr->valid = 0;
1151 	fattr->time_start = jiffies;
1152 	fattr->gencount = nfs_inc_attr_generation_counter();
1153 	fattr->owner_name = NULL;
1154 	fattr->group_name = NULL;
1155 }
1156 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1157 
1158 struct nfs_fattr *nfs_alloc_fattr(void)
1159 {
1160 	struct nfs_fattr *fattr;
1161 
1162 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1163 	if (fattr != NULL)
1164 		nfs_fattr_init(fattr);
1165 	return fattr;
1166 }
1167 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1168 
1169 struct nfs_fh *nfs_alloc_fhandle(void)
1170 {
1171 	struct nfs_fh *fh;
1172 
1173 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1174 	if (fh != NULL)
1175 		fh->size = 0;
1176 	return fh;
1177 }
1178 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1179 
1180 #ifdef NFS_DEBUG
1181 /*
1182  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1183  *                             in the same way that wireshark does
1184  *
1185  * @fh: file handle
1186  *
1187  * For debugging only.
1188  */
1189 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1190 {
1191 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1192 	 * not on the result */
1193 	return ~crc32(0xFFFFFFFF, &fh->data[0], fh->size);
1194 }
1195 
1196 /*
1197  * _nfs_display_fhandle - display an NFS file handle on the console
1198  *
1199  * @fh: file handle to display
1200  * @caption: display caption
1201  *
1202  * For debugging only.
1203  */
1204 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1205 {
1206 	unsigned short i;
1207 
1208 	if (fh == NULL || fh->size == 0) {
1209 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1210 		return;
1211 	}
1212 
1213 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1214 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1215 	for (i = 0; i < fh->size; i += 16) {
1216 		__be32 *pos = (__be32 *)&fh->data[i];
1217 
1218 		switch ((fh->size - i - 1) >> 2) {
1219 		case 0:
1220 			printk(KERN_DEFAULT " %08x\n",
1221 				be32_to_cpup(pos));
1222 			break;
1223 		case 1:
1224 			printk(KERN_DEFAULT " %08x %08x\n",
1225 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1226 			break;
1227 		case 2:
1228 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1229 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1230 				be32_to_cpup(pos + 2));
1231 			break;
1232 		default:
1233 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1234 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1235 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1236 		}
1237 	}
1238 }
1239 #endif
1240 
1241 /**
1242  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1243  * @inode - pointer to inode
1244  * @fattr - attributes
1245  *
1246  * Attempt to divine whether or not an RPC call reply carrying stale
1247  * attributes got scheduled after another call carrying updated ones.
1248  *
1249  * To do so, the function first assumes that a more recent ctime means
1250  * that the attributes in fattr are newer, however it also attempt to
1251  * catch the case where ctime either didn't change, or went backwards
1252  * (if someone reset the clock on the server) by looking at whether
1253  * or not this RPC call was started after the inode was last updated.
1254  * Note also the check for wraparound of 'attr_gencount'
1255  *
1256  * The function returns 'true' if it thinks the attributes in 'fattr' are
1257  * more recent than the ones cached in the inode.
1258  *
1259  */
1260 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1261 {
1262 	const struct nfs_inode *nfsi = NFS_I(inode);
1263 
1264 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1265 		nfs_ctime_need_update(inode, fattr) ||
1266 		nfs_size_need_update(inode, fattr) ||
1267 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1268 }
1269 
1270 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1271 {
1272 	if (nfs_inode_attrs_need_update(inode, fattr))
1273 		return nfs_update_inode(inode, fattr);
1274 	return nfs_check_inode_attributes(inode, fattr);
1275 }
1276 
1277 /**
1278  * nfs_refresh_inode - try to update the inode attribute cache
1279  * @inode - pointer to inode
1280  * @fattr - updated attributes
1281  *
1282  * Check that an RPC call that returned attributes has not overlapped with
1283  * other recent updates of the inode metadata, then decide whether it is
1284  * safe to do a full update of the inode attributes, or whether just to
1285  * call nfs_check_inode_attributes.
1286  */
1287 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1288 {
1289 	int status;
1290 
1291 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1292 		return 0;
1293 	spin_lock(&inode->i_lock);
1294 	status = nfs_refresh_inode_locked(inode, fattr);
1295 	spin_unlock(&inode->i_lock);
1296 
1297 	return status;
1298 }
1299 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1300 
1301 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1302 {
1303 	struct nfs_inode *nfsi = NFS_I(inode);
1304 
1305 	nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1306 	if (S_ISDIR(inode->i_mode)) {
1307 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1308 		nfs_fscache_invalidate(inode);
1309 	}
1310 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1311 		return 0;
1312 	return nfs_refresh_inode_locked(inode, fattr);
1313 }
1314 
1315 /**
1316  * nfs_post_op_update_inode - try to update the inode attribute cache
1317  * @inode - pointer to inode
1318  * @fattr - updated attributes
1319  *
1320  * After an operation that has changed the inode metadata, mark the
1321  * attribute cache as being invalid, then try to update it.
1322  *
1323  * NB: if the server didn't return any post op attributes, this
1324  * function will force the retrieval of attributes before the next
1325  * NFS request.  Thus it should be used only for operations that
1326  * are expected to change one or more attributes, to avoid
1327  * unnecessary NFS requests and trips through nfs_update_inode().
1328  */
1329 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1330 {
1331 	int status;
1332 
1333 	spin_lock(&inode->i_lock);
1334 	status = nfs_post_op_update_inode_locked(inode, fattr);
1335 	spin_unlock(&inode->i_lock);
1336 
1337 	return status;
1338 }
1339 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1340 
1341 /**
1342  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1343  * @inode - pointer to inode
1344  * @fattr - updated attributes
1345  *
1346  * After an operation that has changed the inode metadata, mark the
1347  * attribute cache as being invalid, then try to update it. Fake up
1348  * weak cache consistency data, if none exist.
1349  *
1350  * This function is mainly designed to be used by the ->write_done() functions.
1351  */
1352 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1353 {
1354 	int status;
1355 
1356 	spin_lock(&inode->i_lock);
1357 	/* Don't do a WCC update if these attributes are already stale */
1358 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1359 			!nfs_inode_attrs_need_update(inode, fattr)) {
1360 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1361 				| NFS_ATTR_FATTR_PRESIZE
1362 				| NFS_ATTR_FATTR_PREMTIME
1363 				| NFS_ATTR_FATTR_PRECTIME);
1364 		goto out_noforce;
1365 	}
1366 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1367 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1368 		fattr->pre_change_attr = inode->i_version;
1369 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1370 	}
1371 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1372 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1373 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1374 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1375 	}
1376 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1377 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1378 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1379 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1380 	}
1381 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1382 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1383 		fattr->pre_size = i_size_read(inode);
1384 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1385 	}
1386 out_noforce:
1387 	status = nfs_post_op_update_inode_locked(inode, fattr);
1388 	spin_unlock(&inode->i_lock);
1389 	return status;
1390 }
1391 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1392 
1393 /*
1394  * Many nfs protocol calls return the new file attributes after
1395  * an operation.  Here we update the inode to reflect the state
1396  * of the server's inode.
1397  *
1398  * This is a bit tricky because we have to make sure all dirty pages
1399  * have been sent off to the server before calling invalidate_inode_pages.
1400  * To make sure no other process adds more write requests while we try
1401  * our best to flush them, we make them sleep during the attribute refresh.
1402  *
1403  * A very similar scenario holds for the dir cache.
1404  */
1405 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1406 {
1407 	struct nfs_server *server;
1408 	struct nfs_inode *nfsi = NFS_I(inode);
1409 	loff_t cur_isize, new_isize;
1410 	unsigned long invalid = 0;
1411 	unsigned long now = jiffies;
1412 	unsigned long save_cache_validity;
1413 
1414 	dfprintk(VFS, "NFS: %s(%s/%ld fh_crc=0x%08x ct=%d info=0x%x)\n",
1415 			__func__, inode->i_sb->s_id, inode->i_ino,
1416 			nfs_display_fhandle_hash(NFS_FH(inode)),
1417 			atomic_read(&inode->i_count), fattr->valid);
1418 
1419 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1420 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1421 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1422 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1423 			inode->i_sb->s_id, (long long)nfsi->fileid,
1424 			(long long)fattr->fileid);
1425 		goto out_err;
1426 	}
1427 
1428 	/*
1429 	 * Make sure the inode's type hasn't changed.
1430 	 */
1431 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1432 		/*
1433 		* Big trouble! The inode has become a different object.
1434 		*/
1435 		printk(KERN_DEBUG "NFS: %s: inode %ld mode changed, %07o to %07o\n",
1436 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1437 		goto out_err;
1438 	}
1439 
1440 	server = NFS_SERVER(inode);
1441 	/* Update the fsid? */
1442 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1443 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1444 			!IS_AUTOMOUNT(inode))
1445 		server->fsid = fattr->fsid;
1446 
1447 	/*
1448 	 * Update the read time so we don't revalidate too often.
1449 	 */
1450 	nfsi->read_cache_jiffies = fattr->time_start;
1451 
1452 	save_cache_validity = nfsi->cache_validity;
1453 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1454 			| NFS_INO_INVALID_ATIME
1455 			| NFS_INO_REVAL_FORCED
1456 			| NFS_INO_REVAL_PAGECACHE);
1457 
1458 	/* Do atomic weak cache consistency updates */
1459 	invalid |= nfs_wcc_update_inode(inode, fattr);
1460 
1461 	/* More cache consistency checks */
1462 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1463 		if (inode->i_version != fattr->change_attr) {
1464 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1465 					inode->i_sb->s_id, inode->i_ino);
1466 			invalid |= NFS_INO_INVALID_ATTR
1467 				| NFS_INO_INVALID_DATA
1468 				| NFS_INO_INVALID_ACCESS
1469 				| NFS_INO_INVALID_ACL
1470 				| NFS_INO_REVAL_PAGECACHE;
1471 			if (S_ISDIR(inode->i_mode))
1472 				nfs_force_lookup_revalidate(inode);
1473 			inode->i_version = fattr->change_attr;
1474 		}
1475 	} else if (server->caps & NFS_CAP_CHANGE_ATTR)
1476 		invalid |= save_cache_validity;
1477 
1478 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1479 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1480 	} else if (server->caps & NFS_CAP_MTIME)
1481 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1482 				| NFS_INO_REVAL_FORCED);
1483 
1484 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1485 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1486 	} else if (server->caps & NFS_CAP_CTIME)
1487 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1488 				| NFS_INO_REVAL_FORCED);
1489 
1490 	/* Check if our cached file size is stale */
1491 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1492 		new_isize = nfs_size_to_loff_t(fattr->size);
1493 		cur_isize = i_size_read(inode);
1494 		if (new_isize != cur_isize) {
1495 			/* Do we perhaps have any outstanding writes, or has
1496 			 * the file grown beyond our last write? */
1497 			if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) ||
1498 			     new_isize > cur_isize) {
1499 				i_size_write(inode, new_isize);
1500 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1501 			}
1502 			dprintk("NFS: isize change on server for file %s/%ld "
1503 					"(%Ld to %Ld)\n",
1504 					inode->i_sb->s_id,
1505 					inode->i_ino,
1506 					(long long)cur_isize,
1507 					(long long)new_isize);
1508 		}
1509 	} else
1510 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1511 				| NFS_INO_REVAL_PAGECACHE
1512 				| NFS_INO_REVAL_FORCED);
1513 
1514 
1515 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1516 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1517 	else if (server->caps & NFS_CAP_ATIME)
1518 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1519 				| NFS_INO_REVAL_FORCED);
1520 
1521 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1522 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1523 			umode_t newmode = inode->i_mode & S_IFMT;
1524 			newmode |= fattr->mode & S_IALLUGO;
1525 			inode->i_mode = newmode;
1526 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1527 		}
1528 	} else if (server->caps & NFS_CAP_MODE)
1529 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1530 				| NFS_INO_INVALID_ACCESS
1531 				| NFS_INO_INVALID_ACL
1532 				| NFS_INO_REVAL_FORCED);
1533 
1534 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1535 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1536 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1537 			inode->i_uid = fattr->uid;
1538 		}
1539 	} else if (server->caps & NFS_CAP_OWNER)
1540 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1541 				| NFS_INO_INVALID_ACCESS
1542 				| NFS_INO_INVALID_ACL
1543 				| NFS_INO_REVAL_FORCED);
1544 
1545 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1546 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1547 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1548 			inode->i_gid = fattr->gid;
1549 		}
1550 	} else if (server->caps & NFS_CAP_OWNER_GROUP)
1551 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1552 				| NFS_INO_INVALID_ACCESS
1553 				| NFS_INO_INVALID_ACL
1554 				| NFS_INO_REVAL_FORCED);
1555 
1556 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1557 		if (inode->i_nlink != fattr->nlink) {
1558 			invalid |= NFS_INO_INVALID_ATTR;
1559 			if (S_ISDIR(inode->i_mode))
1560 				invalid |= NFS_INO_INVALID_DATA;
1561 			set_nlink(inode, fattr->nlink);
1562 		}
1563 	} else if (server->caps & NFS_CAP_NLINK)
1564 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1565 				| NFS_INO_REVAL_FORCED);
1566 
1567 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1568 		/*
1569 		 * report the blocks in 512byte units
1570 		 */
1571 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1572  	}
1573 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1574 		inode->i_blocks = fattr->du.nfs2.blocks;
1575 
1576 	/* Update attrtimeo value if we're out of the unstable period */
1577 	if (invalid & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL)) {
1578 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1579 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1580 		nfsi->attrtimeo_timestamp = now;
1581 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1582 	} else {
1583 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1584 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1585 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1586 			nfsi->attrtimeo_timestamp = now;
1587 		}
1588 	}
1589 	invalid &= ~NFS_INO_INVALID_ATTR;
1590 	invalid &= ~NFS_INO_INVALID_LABEL;
1591 	/* Don't invalidate the data if we were to blame */
1592 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1593 				|| S_ISLNK(inode->i_mode)))
1594 		invalid &= ~NFS_INO_INVALID_DATA;
1595 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1596 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1597 		nfsi->cache_validity |= invalid;
1598 
1599 	if (invalid & NFS_INO_INVALID_DATA)
1600 		nfs_fscache_invalidate(inode);
1601 
1602 	return 0;
1603  out_err:
1604 	/*
1605 	 * No need to worry about unhashing the dentry, as the
1606 	 * lookup validation will know that the inode is bad.
1607 	 * (But we fall through to invalidate the caches.)
1608 	 */
1609 	nfs_invalidate_inode(inode);
1610 	return -ESTALE;
1611 }
1612 
1613 struct inode *nfs_alloc_inode(struct super_block *sb)
1614 {
1615 	struct nfs_inode *nfsi;
1616 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1617 	if (!nfsi)
1618 		return NULL;
1619 	nfsi->flags = 0UL;
1620 	nfsi->cache_validity = 0UL;
1621 #ifdef CONFIG_NFS_V3_ACL
1622 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1623 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1624 #endif
1625 #if IS_ENABLED(CONFIG_NFS_V4)
1626 	nfsi->nfs4_acl = NULL;
1627 #endif /* CONFIG_NFS_V4 */
1628 	return &nfsi->vfs_inode;
1629 }
1630 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1631 
1632 static void nfs_i_callback(struct rcu_head *head)
1633 {
1634 	struct inode *inode = container_of(head, struct inode, i_rcu);
1635 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1636 }
1637 
1638 void nfs_destroy_inode(struct inode *inode)
1639 {
1640 	call_rcu(&inode->i_rcu, nfs_i_callback);
1641 }
1642 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1643 
1644 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1645 {
1646 #if IS_ENABLED(CONFIG_NFS_V4)
1647 	INIT_LIST_HEAD(&nfsi->open_states);
1648 	nfsi->delegation = NULL;
1649 	nfsi->delegation_state = 0;
1650 	init_rwsem(&nfsi->rwsem);
1651 	nfsi->layout = NULL;
1652 #endif
1653 }
1654 
1655 static void init_once(void *foo)
1656 {
1657 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1658 
1659 	inode_init_once(&nfsi->vfs_inode);
1660 	INIT_LIST_HEAD(&nfsi->open_files);
1661 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1662 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1663 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1664 	nfsi->npages = 0;
1665 	nfsi->commit_info.ncommit = 0;
1666 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1667 	atomic_set(&nfsi->silly_count, 1);
1668 	INIT_HLIST_HEAD(&nfsi->silly_list);
1669 	init_waitqueue_head(&nfsi->waitqueue);
1670 	nfs4_init_once(nfsi);
1671 }
1672 
1673 static int __init nfs_init_inodecache(void)
1674 {
1675 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1676 					     sizeof(struct nfs_inode),
1677 					     0, (SLAB_RECLAIM_ACCOUNT|
1678 						SLAB_MEM_SPREAD),
1679 					     init_once);
1680 	if (nfs_inode_cachep == NULL)
1681 		return -ENOMEM;
1682 
1683 	return 0;
1684 }
1685 
1686 static void nfs_destroy_inodecache(void)
1687 {
1688 	/*
1689 	 * Make sure all delayed rcu free inodes are flushed before we
1690 	 * destroy cache.
1691 	 */
1692 	rcu_barrier();
1693 	kmem_cache_destroy(nfs_inode_cachep);
1694 }
1695 
1696 struct workqueue_struct *nfsiod_workqueue;
1697 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1698 
1699 /*
1700  * start up the nfsiod workqueue
1701  */
1702 static int nfsiod_start(void)
1703 {
1704 	struct workqueue_struct *wq;
1705 	dprintk("RPC:       creating workqueue nfsiod\n");
1706 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1707 	if (wq == NULL)
1708 		return -ENOMEM;
1709 	nfsiod_workqueue = wq;
1710 	return 0;
1711 }
1712 
1713 /*
1714  * Destroy the nfsiod workqueue
1715  */
1716 static void nfsiod_stop(void)
1717 {
1718 	struct workqueue_struct *wq;
1719 
1720 	wq = nfsiod_workqueue;
1721 	if (wq == NULL)
1722 		return;
1723 	nfsiod_workqueue = NULL;
1724 	destroy_workqueue(wq);
1725 }
1726 
1727 int nfs_net_id;
1728 EXPORT_SYMBOL_GPL(nfs_net_id);
1729 
1730 static int nfs_net_init(struct net *net)
1731 {
1732 	nfs_clients_init(net);
1733 	return 0;
1734 }
1735 
1736 static void nfs_net_exit(struct net *net)
1737 {
1738 	nfs_cleanup_cb_ident_idr(net);
1739 }
1740 
1741 static struct pernet_operations nfs_net_ops = {
1742 	.init = nfs_net_init,
1743 	.exit = nfs_net_exit,
1744 	.id   = &nfs_net_id,
1745 	.size = sizeof(struct nfs_net),
1746 };
1747 
1748 /*
1749  * Initialize NFS
1750  */
1751 static int __init init_nfs_fs(void)
1752 {
1753 	int err;
1754 
1755 	err = register_pernet_subsys(&nfs_net_ops);
1756 	if (err < 0)
1757 		goto out9;
1758 
1759 	err = nfs_fscache_register();
1760 	if (err < 0)
1761 		goto out8;
1762 
1763 	err = nfsiod_start();
1764 	if (err)
1765 		goto out7;
1766 
1767 	err = nfs_fs_proc_init();
1768 	if (err)
1769 		goto out6;
1770 
1771 	err = nfs_init_nfspagecache();
1772 	if (err)
1773 		goto out5;
1774 
1775 	err = nfs_init_inodecache();
1776 	if (err)
1777 		goto out4;
1778 
1779 	err = nfs_init_readpagecache();
1780 	if (err)
1781 		goto out3;
1782 
1783 	err = nfs_init_writepagecache();
1784 	if (err)
1785 		goto out2;
1786 
1787 	err = nfs_init_directcache();
1788 	if (err)
1789 		goto out1;
1790 
1791 #ifdef CONFIG_PROC_FS
1792 	rpc_proc_register(&init_net, &nfs_rpcstat);
1793 #endif
1794 	if ((err = register_nfs_fs()) != 0)
1795 		goto out0;
1796 
1797 	return 0;
1798 out0:
1799 #ifdef CONFIG_PROC_FS
1800 	rpc_proc_unregister(&init_net, "nfs");
1801 #endif
1802 	nfs_destroy_directcache();
1803 out1:
1804 	nfs_destroy_writepagecache();
1805 out2:
1806 	nfs_destroy_readpagecache();
1807 out3:
1808 	nfs_destroy_inodecache();
1809 out4:
1810 	nfs_destroy_nfspagecache();
1811 out5:
1812 	nfs_fs_proc_exit();
1813 out6:
1814 	nfsiod_stop();
1815 out7:
1816 	nfs_fscache_unregister();
1817 out8:
1818 	unregister_pernet_subsys(&nfs_net_ops);
1819 out9:
1820 	return err;
1821 }
1822 
1823 static void __exit exit_nfs_fs(void)
1824 {
1825 	nfs_destroy_directcache();
1826 	nfs_destroy_writepagecache();
1827 	nfs_destroy_readpagecache();
1828 	nfs_destroy_inodecache();
1829 	nfs_destroy_nfspagecache();
1830 	nfs_fscache_unregister();
1831 	unregister_pernet_subsys(&nfs_net_ops);
1832 #ifdef CONFIG_PROC_FS
1833 	rpc_proc_unregister(&init_net, "nfs");
1834 #endif
1835 	unregister_nfs_fs();
1836 	nfs_fs_proc_exit();
1837 	nfsiod_stop();
1838 }
1839 
1840 /* Not quite true; I just maintain it */
1841 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1842 MODULE_LICENSE("GPL");
1843 module_param(enable_ino64, bool, 0644);
1844 
1845 module_init(init_nfs_fs)
1846 module_exit(exit_nfs_fs)
1847