xref: /linux/fs/nfs/inode.c (revision c4ee0af3fa0dc65f690fc908f02b8355f9576ea0)
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 
42 #include <asm/uaccess.h>
43 
44 #include "nfs4_fs.h"
45 #include "callback.h"
46 #include "delegation.h"
47 #include "iostat.h"
48 #include "internal.h"
49 #include "fscache.h"
50 #include "pnfs.h"
51 #include "nfs.h"
52 #include "netns.h"
53 
54 #include "nfstrace.h"
55 
56 #define NFSDBG_FACILITY		NFSDBG_VFS
57 
58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
59 
60 /* Default is to see 64-bit inode numbers */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
62 
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
65 
66 static struct kmem_cache * nfs_inode_cachep;
67 
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
70 {
71 	return nfs_fileid_to_ino_t(fattr->fileid);
72 }
73 
74 /**
75  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76  * @word: long word containing the bit lock
77  */
78 int nfs_wait_bit_killable(void *word)
79 {
80 	if (fatal_signal_pending(current))
81 		return -ERESTARTSYS;
82 	freezable_schedule_unsafe();
83 	return 0;
84 }
85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
86 
87 /**
88  * nfs_compat_user_ino64 - returns the user-visible inode number
89  * @fileid: 64-bit fileid
90  *
91  * This function returns a 32-bit inode number if the boot parameter
92  * nfs.enable_ino64 is zero.
93  */
94 u64 nfs_compat_user_ino64(u64 fileid)
95 {
96 #ifdef CONFIG_COMPAT
97 	compat_ulong_t ino;
98 #else
99 	unsigned long ino;
100 #endif
101 
102 	if (enable_ino64)
103 		return fileid;
104 	ino = fileid;
105 	if (sizeof(ino) < sizeof(fileid))
106 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
107 	return ino;
108 }
109 
110 int nfs_drop_inode(struct inode *inode)
111 {
112 	return NFS_STALE(inode) || generic_drop_inode(inode);
113 }
114 EXPORT_SYMBOL_GPL(nfs_drop_inode);
115 
116 void nfs_clear_inode(struct inode *inode)
117 {
118 	/*
119 	 * The following should never happen...
120 	 */
121 	WARN_ON_ONCE(nfs_have_writebacks(inode));
122 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
123 	nfs_zap_acl_cache(inode);
124 	nfs_access_zap_cache(inode);
125 	nfs_fscache_clear_inode(inode);
126 }
127 EXPORT_SYMBOL_GPL(nfs_clear_inode);
128 
129 void nfs_evict_inode(struct inode *inode)
130 {
131 	truncate_inode_pages(&inode->i_data, 0);
132 	clear_inode(inode);
133 	nfs_clear_inode(inode);
134 }
135 
136 /**
137  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
138  */
139 int nfs_sync_mapping(struct address_space *mapping)
140 {
141 	int ret = 0;
142 
143 	if (mapping->nrpages != 0) {
144 		unmap_mapping_range(mapping, 0, 0, 0);
145 		ret = nfs_wb_all(mapping->host);
146 	}
147 	return ret;
148 }
149 
150 /*
151  * Invalidate the local caches
152  */
153 static void nfs_zap_caches_locked(struct inode *inode)
154 {
155 	struct nfs_inode *nfsi = NFS_I(inode);
156 	int mode = inode->i_mode;
157 
158 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
159 
160 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
161 	nfsi->attrtimeo_timestamp = jiffies;
162 
163 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
164 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
165 		nfs_fscache_invalidate(inode);
166 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR
167 					| NFS_INO_INVALID_LABEL
168 					| NFS_INO_INVALID_DATA
169 					| NFS_INO_INVALID_ACCESS
170 					| NFS_INO_INVALID_ACL
171 					| NFS_INO_REVAL_PAGECACHE;
172 	} else
173 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR
174 					| NFS_INO_INVALID_LABEL
175 					| NFS_INO_INVALID_ACCESS
176 					| NFS_INO_INVALID_ACL
177 					| NFS_INO_REVAL_PAGECACHE;
178 }
179 
180 void nfs_zap_caches(struct inode *inode)
181 {
182 	spin_lock(&inode->i_lock);
183 	nfs_zap_caches_locked(inode);
184 	spin_unlock(&inode->i_lock);
185 }
186 
187 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
188 {
189 	if (mapping->nrpages != 0) {
190 		spin_lock(&inode->i_lock);
191 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
192 		nfs_fscache_invalidate(inode);
193 		spin_unlock(&inode->i_lock);
194 	}
195 }
196 
197 void nfs_zap_acl_cache(struct inode *inode)
198 {
199 	void (*clear_acl_cache)(struct inode *);
200 
201 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
202 	if (clear_acl_cache != NULL)
203 		clear_acl_cache(inode);
204 	spin_lock(&inode->i_lock);
205 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
206 	spin_unlock(&inode->i_lock);
207 }
208 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
209 
210 void nfs_invalidate_atime(struct inode *inode)
211 {
212 	spin_lock(&inode->i_lock);
213 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
214 	spin_unlock(&inode->i_lock);
215 }
216 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
217 
218 /*
219  * Invalidate, but do not unhash, the inode.
220  * NB: must be called with inode->i_lock held!
221  */
222 static void nfs_invalidate_inode(struct inode *inode)
223 {
224 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
225 	nfs_zap_caches_locked(inode);
226 }
227 
228 struct nfs_find_desc {
229 	struct nfs_fh		*fh;
230 	struct nfs_fattr	*fattr;
231 };
232 
233 /*
234  * In NFSv3 we can have 64bit inode numbers. In order to support
235  * this, and re-exported directories (also seen in NFSv2)
236  * we are forced to allow 2 different inodes to have the same
237  * i_ino.
238  */
239 static int
240 nfs_find_actor(struct inode *inode, void *opaque)
241 {
242 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
243 	struct nfs_fh		*fh = desc->fh;
244 	struct nfs_fattr	*fattr = desc->fattr;
245 
246 	if (NFS_FILEID(inode) != fattr->fileid)
247 		return 0;
248 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
249 		return 0;
250 	if (nfs_compare_fh(NFS_FH(inode), fh))
251 		return 0;
252 	if (is_bad_inode(inode) || NFS_STALE(inode))
253 		return 0;
254 	return 1;
255 }
256 
257 static int
258 nfs_init_locked(struct inode *inode, void *opaque)
259 {
260 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
261 	struct nfs_fattr	*fattr = desc->fattr;
262 
263 	set_nfs_fileid(inode, fattr->fileid);
264 	nfs_copy_fh(NFS_FH(inode), desc->fh);
265 	return 0;
266 }
267 
268 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
269 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
270 					struct nfs4_label *label)
271 {
272 	int error;
273 
274 	if (label == NULL)
275 		return;
276 
277 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
278 		error = security_inode_notifysecctx(inode, label->label,
279 				label->len);
280 		if (error)
281 			printk(KERN_ERR "%s() %s %d "
282 					"security_inode_notifysecctx() %d\n",
283 					__func__,
284 					(char *)label->label,
285 					label->len, error);
286 	}
287 }
288 
289 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
290 {
291 	struct nfs4_label *label = NULL;
292 	int minor_version = server->nfs_client->cl_minorversion;
293 
294 	if (minor_version < 2)
295 		return label;
296 
297 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
298 		return label;
299 
300 	label = kzalloc(sizeof(struct nfs4_label), flags);
301 	if (label == NULL)
302 		return ERR_PTR(-ENOMEM);
303 
304 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
305 	if (label->label == NULL) {
306 		kfree(label);
307 		return ERR_PTR(-ENOMEM);
308 	}
309 	label->len = NFS4_MAXLABELLEN;
310 
311 	return label;
312 }
313 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
314 #else
315 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
316 					struct nfs4_label *label)
317 {
318 }
319 #endif
320 EXPORT_SYMBOL_GPL(nfs_setsecurity);
321 
322 /*
323  * This is our front-end to iget that looks up inodes by file handle
324  * instead of inode number.
325  */
326 struct inode *
327 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
328 {
329 	struct nfs_find_desc desc = {
330 		.fh	= fh,
331 		.fattr	= fattr
332 	};
333 	struct inode *inode = ERR_PTR(-ENOENT);
334 	unsigned long hash;
335 
336 	nfs_attr_check_mountpoint(sb, fattr);
337 
338 	if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
339 	    !nfs_attr_use_mounted_on_fileid(fattr))
340 		goto out_no_inode;
341 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
342 		goto out_no_inode;
343 
344 	hash = nfs_fattr_to_ino_t(fattr);
345 
346 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
347 	if (inode == NULL) {
348 		inode = ERR_PTR(-ENOMEM);
349 		goto out_no_inode;
350 	}
351 
352 	if (inode->i_state & I_NEW) {
353 		struct nfs_inode *nfsi = NFS_I(inode);
354 		unsigned long now = jiffies;
355 
356 		/* We set i_ino for the few things that still rely on it,
357 		 * such as stat(2) */
358 		inode->i_ino = hash;
359 
360 		/* We can't support update_atime(), since the server will reset it */
361 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
362 		inode->i_mode = fattr->mode;
363 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
364 				&& nfs_server_capable(inode, NFS_CAP_MODE))
365 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
366 		/* Why so? Because we want revalidate for devices/FIFOs, and
367 		 * that's precisely what we have in nfs_file_inode_operations.
368 		 */
369 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
370 		if (S_ISREG(inode->i_mode)) {
371 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
372 			inode->i_data.a_ops = &nfs_file_aops;
373 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
374 		} else if (S_ISDIR(inode->i_mode)) {
375 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
376 			inode->i_fop = &nfs_dir_operations;
377 			inode->i_data.a_ops = &nfs_dir_aops;
378 			/* Deal with crossing mountpoints */
379 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
380 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
381 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
382 					inode->i_op = &nfs_referral_inode_operations;
383 				else
384 					inode->i_op = &nfs_mountpoint_inode_operations;
385 				inode->i_fop = NULL;
386 				inode->i_flags |= S_AUTOMOUNT;
387 			}
388 		} else if (S_ISLNK(inode->i_mode))
389 			inode->i_op = &nfs_symlink_inode_operations;
390 		else
391 			init_special_inode(inode, inode->i_mode, fattr->rdev);
392 
393 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
394 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
395 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
396 		inode->i_version = 0;
397 		inode->i_size = 0;
398 		clear_nlink(inode);
399 		inode->i_uid = make_kuid(&init_user_ns, -2);
400 		inode->i_gid = make_kgid(&init_user_ns, -2);
401 		inode->i_blocks = 0;
402 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
403 		nfsi->write_io = 0;
404 		nfsi->read_io = 0;
405 
406 		nfsi->read_cache_jiffies = fattr->time_start;
407 		nfsi->attr_gencount = fattr->gencount;
408 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
409 			inode->i_atime = fattr->atime;
410 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
411 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
412 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
413 			inode->i_mtime = fattr->mtime;
414 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
415 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
416 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
417 			inode->i_ctime = fattr->ctime;
418 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
419 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
420 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
421 			inode->i_version = fattr->change_attr;
422 		else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
423 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
424 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
425 			inode->i_size = nfs_size_to_loff_t(fattr->size);
426 		else
427 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
428 				| NFS_INO_REVAL_PAGECACHE;
429 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
430 			set_nlink(inode, fattr->nlink);
431 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
432 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
433 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
434 			inode->i_uid = fattr->uid;
435 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
436 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
437 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
438 			inode->i_gid = fattr->gid;
439 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
440 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
441 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
442 			inode->i_blocks = fattr->du.nfs2.blocks;
443 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
444 			/*
445 			 * report the blocks in 512byte units
446 			 */
447 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
448 		}
449 
450 		nfs_setsecurity(inode, fattr, label);
451 
452 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
453 		nfsi->attrtimeo_timestamp = now;
454 		nfsi->access_cache = RB_ROOT;
455 
456 		nfs_fscache_init_inode(inode);
457 
458 		unlock_new_inode(inode);
459 	} else
460 		nfs_refresh_inode(inode, fattr);
461 	dprintk("NFS: nfs_fhget(%s/%Ld fh_crc=0x%08x ct=%d)\n",
462 		inode->i_sb->s_id,
463 		(long long)NFS_FILEID(inode),
464 		nfs_display_fhandle_hash(fh),
465 		atomic_read(&inode->i_count));
466 
467 out:
468 	return inode;
469 
470 out_no_inode:
471 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
472 	goto out;
473 }
474 EXPORT_SYMBOL_GPL(nfs_fhget);
475 
476 #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)
477 
478 int
479 nfs_setattr(struct dentry *dentry, struct iattr *attr)
480 {
481 	struct inode *inode = dentry->d_inode;
482 	struct nfs_fattr *fattr;
483 	int error = -ENOMEM;
484 
485 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
486 
487 	/* skip mode change if it's just for clearing setuid/setgid */
488 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
489 		attr->ia_valid &= ~ATTR_MODE;
490 
491 	if (attr->ia_valid & ATTR_SIZE) {
492 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
493 			attr->ia_valid &= ~ATTR_SIZE;
494 	}
495 
496 	/* Optimization: if the end result is no change, don't RPC */
497 	attr->ia_valid &= NFS_VALID_ATTRS;
498 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
499 		return 0;
500 
501 	trace_nfs_setattr_enter(inode);
502 
503 	/* Write all dirty data */
504 	if (S_ISREG(inode->i_mode)) {
505 		nfs_inode_dio_wait(inode);
506 		nfs_wb_all(inode);
507 	}
508 
509 	fattr = nfs_alloc_fattr();
510 	if (fattr == NULL)
511 		goto out;
512 	/*
513 	 * Return any delegations if we're going to change ACLs
514 	 */
515 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
516 		NFS_PROTO(inode)->return_delegation(inode);
517 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
518 	if (error == 0)
519 		error = nfs_refresh_inode(inode, fattr);
520 	nfs_free_fattr(fattr);
521 out:
522 	trace_nfs_setattr_exit(inode, error);
523 	return error;
524 }
525 EXPORT_SYMBOL_GPL(nfs_setattr);
526 
527 /**
528  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
529  * @inode: inode of the file used
530  * @offset: file offset to start truncating
531  *
532  * This is a copy of the common vmtruncate, but with the locking
533  * corrected to take into account the fact that NFS requires
534  * inode->i_size to be updated under the inode->i_lock.
535  */
536 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
537 {
538 	int err;
539 
540 	err = inode_newsize_ok(inode, offset);
541 	if (err)
542 		goto out;
543 
544 	spin_lock(&inode->i_lock);
545 	i_size_write(inode, offset);
546 	spin_unlock(&inode->i_lock);
547 
548 	truncate_pagecache(inode, offset);
549 out:
550 	return err;
551 }
552 
553 /**
554  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
555  * @inode: pointer to struct inode
556  * @attr: pointer to struct iattr
557  *
558  * Note: we do this in the *proc.c in order to ensure that
559  *       it works for things like exclusive creates too.
560  */
561 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
562 {
563 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
564 		spin_lock(&inode->i_lock);
565 		if ((attr->ia_valid & ATTR_MODE) != 0) {
566 			int mode = attr->ia_mode & S_IALLUGO;
567 			mode |= inode->i_mode & ~S_IALLUGO;
568 			inode->i_mode = mode;
569 		}
570 		if ((attr->ia_valid & ATTR_UID) != 0)
571 			inode->i_uid = attr->ia_uid;
572 		if ((attr->ia_valid & ATTR_GID) != 0)
573 			inode->i_gid = attr->ia_gid;
574 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
575 		spin_unlock(&inode->i_lock);
576 	}
577 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
578 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
579 		nfs_vmtruncate(inode, attr->ia_size);
580 	}
581 }
582 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
583 
584 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
585 {
586 	struct inode *inode = dentry->d_inode;
587 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
588 	int err;
589 
590 	trace_nfs_getattr_enter(inode);
591 	/* Flush out writes to the server in order to update c/mtime.  */
592 	if (S_ISREG(inode->i_mode)) {
593 		nfs_inode_dio_wait(inode);
594 		err = filemap_write_and_wait(inode->i_mapping);
595 		if (err)
596 			goto out;
597 	}
598 
599 	/*
600 	 * We may force a getattr if the user cares about atime.
601 	 *
602 	 * Note that we only have to check the vfsmount flags here:
603 	 *  - NFS always sets S_NOATIME by so checking it would give a
604 	 *    bogus result
605 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
606 	 *    no point in checking those.
607 	 */
608  	if ((mnt->mnt_flags & MNT_NOATIME) ||
609  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
610 		need_atime = 0;
611 
612 	if (need_atime)
613 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
614 	else
615 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
616 	if (!err) {
617 		generic_fillattr(inode, stat);
618 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
619 	}
620 out:
621 	trace_nfs_getattr_exit(inode, err);
622 	return err;
623 }
624 EXPORT_SYMBOL_GPL(nfs_getattr);
625 
626 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
627 {
628 	atomic_set(&l_ctx->count, 1);
629 	l_ctx->lockowner.l_owner = current->files;
630 	l_ctx->lockowner.l_pid = current->tgid;
631 	INIT_LIST_HEAD(&l_ctx->list);
632 	nfs_iocounter_init(&l_ctx->io_count);
633 }
634 
635 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
636 {
637 	struct nfs_lock_context *head = &ctx->lock_context;
638 	struct nfs_lock_context *pos = head;
639 
640 	do {
641 		if (pos->lockowner.l_owner != current->files)
642 			continue;
643 		if (pos->lockowner.l_pid != current->tgid)
644 			continue;
645 		atomic_inc(&pos->count);
646 		return pos;
647 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
648 	return NULL;
649 }
650 
651 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
652 {
653 	struct nfs_lock_context *res, *new = NULL;
654 	struct inode *inode = ctx->dentry->d_inode;
655 
656 	spin_lock(&inode->i_lock);
657 	res = __nfs_find_lock_context(ctx);
658 	if (res == NULL) {
659 		spin_unlock(&inode->i_lock);
660 		new = kmalloc(sizeof(*new), GFP_KERNEL);
661 		if (new == NULL)
662 			return ERR_PTR(-ENOMEM);
663 		nfs_init_lock_context(new);
664 		spin_lock(&inode->i_lock);
665 		res = __nfs_find_lock_context(ctx);
666 		if (res == NULL) {
667 			list_add_tail(&new->list, &ctx->lock_context.list);
668 			new->open_context = ctx;
669 			res = new;
670 			new = NULL;
671 		}
672 	}
673 	spin_unlock(&inode->i_lock);
674 	kfree(new);
675 	return res;
676 }
677 
678 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
679 {
680 	struct nfs_open_context *ctx = l_ctx->open_context;
681 	struct inode *inode = ctx->dentry->d_inode;
682 
683 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
684 		return;
685 	list_del(&l_ctx->list);
686 	spin_unlock(&inode->i_lock);
687 	kfree(l_ctx);
688 }
689 
690 /**
691  * nfs_close_context - Common close_context() routine NFSv2/v3
692  * @ctx: pointer to context
693  * @is_sync: is this a synchronous close
694  *
695  * always ensure that the attributes are up to date if we're mounted
696  * with close-to-open semantics
697  */
698 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
699 {
700 	struct inode *inode;
701 	struct nfs_server *server;
702 
703 	if (!(ctx->mode & FMODE_WRITE))
704 		return;
705 	if (!is_sync)
706 		return;
707 	inode = ctx->dentry->d_inode;
708 	if (!list_empty(&NFS_I(inode)->open_files))
709 		return;
710 	server = NFS_SERVER(inode);
711 	if (server->flags & NFS_MOUNT_NOCTO)
712 		return;
713 	nfs_revalidate_inode(server, inode);
714 }
715 EXPORT_SYMBOL_GPL(nfs_close_context);
716 
717 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
718 {
719 	struct nfs_open_context *ctx;
720 	struct rpc_cred *cred = rpc_lookup_cred();
721 	if (IS_ERR(cred))
722 		return ERR_CAST(cred);
723 
724 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
725 	if (!ctx) {
726 		put_rpccred(cred);
727 		return ERR_PTR(-ENOMEM);
728 	}
729 	nfs_sb_active(dentry->d_sb);
730 	ctx->dentry = dget(dentry);
731 	ctx->cred = cred;
732 	ctx->state = NULL;
733 	ctx->mode = f_mode;
734 	ctx->flags = 0;
735 	ctx->error = 0;
736 	nfs_init_lock_context(&ctx->lock_context);
737 	ctx->lock_context.open_context = ctx;
738 	INIT_LIST_HEAD(&ctx->list);
739 	ctx->mdsthreshold = NULL;
740 	return ctx;
741 }
742 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
743 
744 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
745 {
746 	if (ctx != NULL)
747 		atomic_inc(&ctx->lock_context.count);
748 	return ctx;
749 }
750 EXPORT_SYMBOL_GPL(get_nfs_open_context);
751 
752 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
753 {
754 	struct inode *inode = ctx->dentry->d_inode;
755 	struct super_block *sb = ctx->dentry->d_sb;
756 
757 	if (!list_empty(&ctx->list)) {
758 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
759 			return;
760 		list_del(&ctx->list);
761 		spin_unlock(&inode->i_lock);
762 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
763 		return;
764 	if (inode != NULL)
765 		NFS_PROTO(inode)->close_context(ctx, is_sync);
766 	if (ctx->cred != NULL)
767 		put_rpccred(ctx->cred);
768 	dput(ctx->dentry);
769 	nfs_sb_deactive(sb);
770 	kfree(ctx->mdsthreshold);
771 	kfree(ctx);
772 }
773 
774 void put_nfs_open_context(struct nfs_open_context *ctx)
775 {
776 	__put_nfs_open_context(ctx, 0);
777 }
778 EXPORT_SYMBOL_GPL(put_nfs_open_context);
779 
780 /*
781  * Ensure that mmap has a recent RPC credential for use when writing out
782  * shared pages
783  */
784 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
785 {
786 	struct inode *inode = ctx->dentry->d_inode;
787 	struct nfs_inode *nfsi = NFS_I(inode);
788 
789 	spin_lock(&inode->i_lock);
790 	list_add(&ctx->list, &nfsi->open_files);
791 	spin_unlock(&inode->i_lock);
792 }
793 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
794 
795 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
796 {
797 	filp->private_data = get_nfs_open_context(ctx);
798 	if (list_empty(&ctx->list))
799 		nfs_inode_attach_open_context(ctx);
800 }
801 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
802 
803 /*
804  * Given an inode, search for an open context with the desired characteristics
805  */
806 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
807 {
808 	struct nfs_inode *nfsi = NFS_I(inode);
809 	struct nfs_open_context *pos, *ctx = NULL;
810 
811 	spin_lock(&inode->i_lock);
812 	list_for_each_entry(pos, &nfsi->open_files, list) {
813 		if (cred != NULL && pos->cred != cred)
814 			continue;
815 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
816 			continue;
817 		ctx = get_nfs_open_context(pos);
818 		break;
819 	}
820 	spin_unlock(&inode->i_lock);
821 	return ctx;
822 }
823 
824 static void nfs_file_clear_open_context(struct file *filp)
825 {
826 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
827 
828 	if (ctx) {
829 		struct inode *inode = ctx->dentry->d_inode;
830 
831 		filp->private_data = NULL;
832 		spin_lock(&inode->i_lock);
833 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
834 		spin_unlock(&inode->i_lock);
835 		__put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
836 	}
837 }
838 
839 /*
840  * These allocate and release file read/write context information.
841  */
842 int nfs_open(struct inode *inode, struct file *filp)
843 {
844 	struct nfs_open_context *ctx;
845 
846 	ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
847 	if (IS_ERR(ctx))
848 		return PTR_ERR(ctx);
849 	nfs_file_set_open_context(filp, ctx);
850 	put_nfs_open_context(ctx);
851 	nfs_fscache_open_file(inode, filp);
852 	return 0;
853 }
854 
855 int nfs_release(struct inode *inode, struct file *filp)
856 {
857 	nfs_file_clear_open_context(filp);
858 	return 0;
859 }
860 
861 /*
862  * This function is called whenever some part of NFS notices that
863  * the cached attributes have to be refreshed.
864  */
865 int
866 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
867 {
868 	int		 status = -ESTALE;
869 	struct nfs4_label *label = NULL;
870 	struct nfs_fattr *fattr = NULL;
871 	struct nfs_inode *nfsi = NFS_I(inode);
872 
873 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
874 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
875 
876 	trace_nfs_revalidate_inode_enter(inode);
877 
878 	if (is_bad_inode(inode))
879 		goto out;
880 	if (NFS_STALE(inode))
881 		goto out;
882 
883 	status = -ENOMEM;
884 	fattr = nfs_alloc_fattr();
885 	if (fattr == NULL)
886 		goto out;
887 
888 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
889 
890 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
891 	if (IS_ERR(label)) {
892 		status = PTR_ERR(label);
893 		goto out;
894 	}
895 
896 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
897 	if (status != 0) {
898 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
899 			 inode->i_sb->s_id,
900 			 (long long)NFS_FILEID(inode), status);
901 		if (status == -ESTALE) {
902 			nfs_zap_caches(inode);
903 			if (!S_ISDIR(inode->i_mode))
904 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
905 		}
906 		goto err_out;
907 	}
908 
909 	status = nfs_refresh_inode(inode, fattr);
910 	if (status) {
911 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
912 			 inode->i_sb->s_id,
913 			 (long long)NFS_FILEID(inode), status);
914 		goto err_out;
915 	}
916 
917 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
918 		nfs_zap_acl_cache(inode);
919 
920 	nfs_setsecurity(inode, fattr, label);
921 
922 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
923 		inode->i_sb->s_id,
924 		(long long)NFS_FILEID(inode));
925 
926 err_out:
927 	nfs4_label_free(label);
928 out:
929 	nfs_free_fattr(fattr);
930 	trace_nfs_revalidate_inode_exit(inode, status);
931 	return status;
932 }
933 
934 int nfs_attribute_timeout(struct inode *inode)
935 {
936 	struct nfs_inode *nfsi = NFS_I(inode);
937 
938 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
939 }
940 
941 int nfs_attribute_cache_expired(struct inode *inode)
942 {
943 	if (nfs_have_delegated_attributes(inode))
944 		return 0;
945 	return nfs_attribute_timeout(inode);
946 }
947 
948 /**
949  * nfs_revalidate_inode - Revalidate the inode attributes
950  * @server - pointer to nfs_server struct
951  * @inode - pointer to inode struct
952  *
953  * Updates inode attribute information by retrieving the data from the server.
954  */
955 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
956 {
957 	if (!(NFS_I(inode)->cache_validity &
958 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
959 			&& !nfs_attribute_cache_expired(inode))
960 		return NFS_STALE(inode) ? -ESTALE : 0;
961 	return __nfs_revalidate_inode(server, inode);
962 }
963 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
964 
965 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
966 {
967 	struct nfs_inode *nfsi = NFS_I(inode);
968 	int ret;
969 
970 	if (mapping->nrpages != 0) {
971 		if (S_ISREG(inode->i_mode)) {
972 			ret = nfs_sync_mapping(mapping);
973 			if (ret < 0)
974 				return ret;
975 		}
976 		ret = invalidate_inode_pages2(mapping);
977 		if (ret < 0)
978 			return ret;
979 	}
980 	spin_lock(&inode->i_lock);
981 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
982 	if (S_ISDIR(inode->i_mode))
983 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
984 	spin_unlock(&inode->i_lock);
985 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
986 	nfs_fscache_wait_on_invalidate(inode);
987 
988 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
989 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
990 	return 0;
991 }
992 
993 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
994 {
995 	if (nfs_have_delegated_attributes(inode))
996 		return false;
997 	return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
998 		|| nfs_attribute_timeout(inode)
999 		|| NFS_STALE(inode);
1000 }
1001 
1002 /**
1003  * nfs_revalidate_mapping - Revalidate the pagecache
1004  * @inode - pointer to host inode
1005  * @mapping - pointer to mapping
1006  */
1007 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1008 {
1009 	struct nfs_inode *nfsi = NFS_I(inode);
1010 	int ret = 0;
1011 
1012 	/* swapfiles are not supposed to be shared. */
1013 	if (IS_SWAPFILE(inode))
1014 		goto out;
1015 
1016 	if (nfs_mapping_need_revalidate_inode(inode)) {
1017 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1018 		if (ret < 0)
1019 			goto out;
1020 	}
1021 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1022 		trace_nfs_invalidate_mapping_enter(inode);
1023 		ret = nfs_invalidate_mapping(inode, mapping);
1024 		trace_nfs_invalidate_mapping_exit(inode, ret);
1025 	}
1026 
1027 out:
1028 	return ret;
1029 }
1030 
1031 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1032 {
1033 	struct nfs_inode *nfsi = NFS_I(inode);
1034 	unsigned long ret = 0;
1035 
1036 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1037 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1038 			&& inode->i_version == fattr->pre_change_attr) {
1039 		inode->i_version = fattr->change_attr;
1040 		if (S_ISDIR(inode->i_mode))
1041 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1042 		ret |= NFS_INO_INVALID_ATTR;
1043 	}
1044 	/* If we have atomic WCC data, we may update some attributes */
1045 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1046 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1047 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1048 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1049 		ret |= NFS_INO_INVALID_ATTR;
1050 	}
1051 
1052 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1053 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1054 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1055 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1056 		if (S_ISDIR(inode->i_mode))
1057 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1058 		ret |= NFS_INO_INVALID_ATTR;
1059 	}
1060 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1061 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1062 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1063 			&& nfsi->npages == 0) {
1064 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1065 		ret |= NFS_INO_INVALID_ATTR;
1066 	}
1067 
1068 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1069 		nfs_fscache_invalidate(inode);
1070 
1071 	return ret;
1072 }
1073 
1074 /**
1075  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1076  * @inode - pointer to inode
1077  * @fattr - updated attributes
1078  *
1079  * Verifies the attribute cache. If we have just changed the attributes,
1080  * so that fattr carries weak cache consistency data, then it may
1081  * also update the ctime/mtime/change_attribute.
1082  */
1083 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1084 {
1085 	struct nfs_inode *nfsi = NFS_I(inode);
1086 	loff_t cur_size, new_isize;
1087 	unsigned long invalid = 0;
1088 
1089 
1090 	if (nfs_have_delegated_attributes(inode))
1091 		return 0;
1092 	/* Has the inode gone and changed behind our back? */
1093 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1094 		return -EIO;
1095 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1096 		return -EIO;
1097 
1098 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1099 			inode->i_version != fattr->change_attr)
1100 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1101 
1102 	/* Verify a few of the more important attributes */
1103 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1104 		invalid |= NFS_INO_INVALID_ATTR;
1105 
1106 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1107 		cur_size = i_size_read(inode);
1108 		new_isize = nfs_size_to_loff_t(fattr->size);
1109 		if (cur_size != new_isize && nfsi->npages == 0)
1110 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1111 	}
1112 
1113 	/* Have any file permissions changed? */
1114 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1115 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1116 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1117 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1118 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1119 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1120 
1121 	/* Has the link count changed? */
1122 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1123 		invalid |= NFS_INO_INVALID_ATTR;
1124 
1125 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1126 		invalid |= NFS_INO_INVALID_ATIME;
1127 
1128 	if (invalid != 0)
1129 		nfsi->cache_validity |= invalid;
1130 
1131 	nfsi->read_cache_jiffies = fattr->time_start;
1132 	return 0;
1133 }
1134 
1135 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1136 {
1137 	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1138 		return 0;
1139 	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1140 }
1141 
1142 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1143 {
1144 	if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1145 		return 0;
1146 	return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1147 }
1148 
1149 static atomic_long_t nfs_attr_generation_counter;
1150 
1151 static unsigned long nfs_read_attr_generation_counter(void)
1152 {
1153 	return atomic_long_read(&nfs_attr_generation_counter);
1154 }
1155 
1156 unsigned long nfs_inc_attr_generation_counter(void)
1157 {
1158 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1159 }
1160 
1161 void nfs_fattr_init(struct nfs_fattr *fattr)
1162 {
1163 	fattr->valid = 0;
1164 	fattr->time_start = jiffies;
1165 	fattr->gencount = nfs_inc_attr_generation_counter();
1166 	fattr->owner_name = NULL;
1167 	fattr->group_name = NULL;
1168 }
1169 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1170 
1171 struct nfs_fattr *nfs_alloc_fattr(void)
1172 {
1173 	struct nfs_fattr *fattr;
1174 
1175 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1176 	if (fattr != NULL)
1177 		nfs_fattr_init(fattr);
1178 	return fattr;
1179 }
1180 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1181 
1182 struct nfs_fh *nfs_alloc_fhandle(void)
1183 {
1184 	struct nfs_fh *fh;
1185 
1186 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1187 	if (fh != NULL)
1188 		fh->size = 0;
1189 	return fh;
1190 }
1191 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1192 
1193 #ifdef NFS_DEBUG
1194 /*
1195  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1196  *                             in the same way that wireshark does
1197  *
1198  * @fh: file handle
1199  *
1200  * For debugging only.
1201  */
1202 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1203 {
1204 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1205 	 * not on the result */
1206 	return nfs_fhandle_hash(fh);
1207 }
1208 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1209 
1210 /*
1211  * _nfs_display_fhandle - display an NFS file handle on the console
1212  *
1213  * @fh: file handle to display
1214  * @caption: display caption
1215  *
1216  * For debugging only.
1217  */
1218 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1219 {
1220 	unsigned short i;
1221 
1222 	if (fh == NULL || fh->size == 0) {
1223 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1224 		return;
1225 	}
1226 
1227 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1228 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1229 	for (i = 0; i < fh->size; i += 16) {
1230 		__be32 *pos = (__be32 *)&fh->data[i];
1231 
1232 		switch ((fh->size - i - 1) >> 2) {
1233 		case 0:
1234 			printk(KERN_DEFAULT " %08x\n",
1235 				be32_to_cpup(pos));
1236 			break;
1237 		case 1:
1238 			printk(KERN_DEFAULT " %08x %08x\n",
1239 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1240 			break;
1241 		case 2:
1242 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1243 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1244 				be32_to_cpup(pos + 2));
1245 			break;
1246 		default:
1247 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1248 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1249 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1250 		}
1251 	}
1252 }
1253 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1254 #endif
1255 
1256 /**
1257  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1258  * @inode - pointer to inode
1259  * @fattr - attributes
1260  *
1261  * Attempt to divine whether or not an RPC call reply carrying stale
1262  * attributes got scheduled after another call carrying updated ones.
1263  *
1264  * To do so, the function first assumes that a more recent ctime means
1265  * that the attributes in fattr are newer, however it also attempt to
1266  * catch the case where ctime either didn't change, or went backwards
1267  * (if someone reset the clock on the server) by looking at whether
1268  * or not this RPC call was started after the inode was last updated.
1269  * Note also the check for wraparound of 'attr_gencount'
1270  *
1271  * The function returns 'true' if it thinks the attributes in 'fattr' are
1272  * more recent than the ones cached in the inode.
1273  *
1274  */
1275 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1276 {
1277 	const struct nfs_inode *nfsi = NFS_I(inode);
1278 
1279 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1280 		nfs_ctime_need_update(inode, fattr) ||
1281 		nfs_size_need_update(inode, fattr) ||
1282 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1283 }
1284 
1285 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1286 {
1287 	int ret;
1288 
1289 	trace_nfs_refresh_inode_enter(inode);
1290 
1291 	if (nfs_inode_attrs_need_update(inode, fattr))
1292 		ret = nfs_update_inode(inode, fattr);
1293 	else
1294 		ret = nfs_check_inode_attributes(inode, fattr);
1295 
1296 	trace_nfs_refresh_inode_exit(inode, ret);
1297 	return ret;
1298 }
1299 
1300 /**
1301  * nfs_refresh_inode - try to update the inode attribute cache
1302  * @inode - pointer to inode
1303  * @fattr - updated attributes
1304  *
1305  * Check that an RPC call that returned attributes has not overlapped with
1306  * other recent updates of the inode metadata, then decide whether it is
1307  * safe to do a full update of the inode attributes, or whether just to
1308  * call nfs_check_inode_attributes.
1309  */
1310 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1311 {
1312 	int status;
1313 
1314 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1315 		return 0;
1316 	spin_lock(&inode->i_lock);
1317 	status = nfs_refresh_inode_locked(inode, fattr);
1318 	spin_unlock(&inode->i_lock);
1319 
1320 	return status;
1321 }
1322 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1323 
1324 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1325 {
1326 	struct nfs_inode *nfsi = NFS_I(inode);
1327 
1328 	nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1329 	if (S_ISDIR(inode->i_mode)) {
1330 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1331 		nfs_fscache_invalidate(inode);
1332 	}
1333 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1334 		return 0;
1335 	return nfs_refresh_inode_locked(inode, fattr);
1336 }
1337 
1338 /**
1339  * nfs_post_op_update_inode - try to update the inode attribute cache
1340  * @inode - pointer to inode
1341  * @fattr - updated attributes
1342  *
1343  * After an operation that has changed the inode metadata, mark the
1344  * attribute cache as being invalid, then try to update it.
1345  *
1346  * NB: if the server didn't return any post op attributes, this
1347  * function will force the retrieval of attributes before the next
1348  * NFS request.  Thus it should be used only for operations that
1349  * are expected to change one or more attributes, to avoid
1350  * unnecessary NFS requests and trips through nfs_update_inode().
1351  */
1352 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1353 {
1354 	int status;
1355 
1356 	spin_lock(&inode->i_lock);
1357 	status = nfs_post_op_update_inode_locked(inode, fattr);
1358 	spin_unlock(&inode->i_lock);
1359 
1360 	return status;
1361 }
1362 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1363 
1364 /**
1365  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1366  * @inode - pointer to inode
1367  * @fattr - updated attributes
1368  *
1369  * After an operation that has changed the inode metadata, mark the
1370  * attribute cache as being invalid, then try to update it. Fake up
1371  * weak cache consistency data, if none exist.
1372  *
1373  * This function is mainly designed to be used by the ->write_done() functions.
1374  */
1375 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1376 {
1377 	int status;
1378 
1379 	spin_lock(&inode->i_lock);
1380 	/* Don't do a WCC update if these attributes are already stale */
1381 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1382 			!nfs_inode_attrs_need_update(inode, fattr)) {
1383 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1384 				| NFS_ATTR_FATTR_PRESIZE
1385 				| NFS_ATTR_FATTR_PREMTIME
1386 				| NFS_ATTR_FATTR_PRECTIME);
1387 		goto out_noforce;
1388 	}
1389 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1390 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1391 		fattr->pre_change_attr = inode->i_version;
1392 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1393 	}
1394 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1395 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1396 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1397 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1398 	}
1399 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1400 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1401 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1402 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1403 	}
1404 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1405 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1406 		fattr->pre_size = i_size_read(inode);
1407 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1408 	}
1409 out_noforce:
1410 	status = nfs_post_op_update_inode_locked(inode, fattr);
1411 	spin_unlock(&inode->i_lock);
1412 	return status;
1413 }
1414 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1415 
1416 /*
1417  * Many nfs protocol calls return the new file attributes after
1418  * an operation.  Here we update the inode to reflect the state
1419  * of the server's inode.
1420  *
1421  * This is a bit tricky because we have to make sure all dirty pages
1422  * have been sent off to the server before calling invalidate_inode_pages.
1423  * To make sure no other process adds more write requests while we try
1424  * our best to flush them, we make them sleep during the attribute refresh.
1425  *
1426  * A very similar scenario holds for the dir cache.
1427  */
1428 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1429 {
1430 	struct nfs_server *server;
1431 	struct nfs_inode *nfsi = NFS_I(inode);
1432 	loff_t cur_isize, new_isize;
1433 	unsigned long invalid = 0;
1434 	unsigned long now = jiffies;
1435 	unsigned long save_cache_validity;
1436 
1437 	dfprintk(VFS, "NFS: %s(%s/%ld fh_crc=0x%08x ct=%d info=0x%x)\n",
1438 			__func__, inode->i_sb->s_id, inode->i_ino,
1439 			nfs_display_fhandle_hash(NFS_FH(inode)),
1440 			atomic_read(&inode->i_count), fattr->valid);
1441 
1442 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1443 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1444 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1445 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1446 			inode->i_sb->s_id, (long long)nfsi->fileid,
1447 			(long long)fattr->fileid);
1448 		goto out_err;
1449 	}
1450 
1451 	/*
1452 	 * Make sure the inode's type hasn't changed.
1453 	 */
1454 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1455 		/*
1456 		* Big trouble! The inode has become a different object.
1457 		*/
1458 		printk(KERN_DEBUG "NFS: %s: inode %ld mode changed, %07o to %07o\n",
1459 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1460 		goto out_err;
1461 	}
1462 
1463 	server = NFS_SERVER(inode);
1464 	/* Update the fsid? */
1465 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1466 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1467 			!IS_AUTOMOUNT(inode))
1468 		server->fsid = fattr->fsid;
1469 
1470 	/*
1471 	 * Update the read time so we don't revalidate too often.
1472 	 */
1473 	nfsi->read_cache_jiffies = fattr->time_start;
1474 
1475 	save_cache_validity = nfsi->cache_validity;
1476 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1477 			| NFS_INO_INVALID_ATIME
1478 			| NFS_INO_REVAL_FORCED
1479 			| NFS_INO_REVAL_PAGECACHE);
1480 
1481 	/* Do atomic weak cache consistency updates */
1482 	invalid |= nfs_wcc_update_inode(inode, fattr);
1483 
1484 	/* More cache consistency checks */
1485 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1486 		if (inode->i_version != fattr->change_attr) {
1487 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1488 					inode->i_sb->s_id, inode->i_ino);
1489 			invalid |= NFS_INO_INVALID_ATTR
1490 				| NFS_INO_INVALID_DATA
1491 				| NFS_INO_INVALID_ACCESS
1492 				| NFS_INO_INVALID_ACL
1493 				| NFS_INO_REVAL_PAGECACHE;
1494 			if (S_ISDIR(inode->i_mode))
1495 				nfs_force_lookup_revalidate(inode);
1496 			inode->i_version = fattr->change_attr;
1497 		}
1498 	} else if (server->caps & NFS_CAP_CHANGE_ATTR)
1499 		invalid |= save_cache_validity;
1500 
1501 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1502 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1503 	} else if (server->caps & NFS_CAP_MTIME)
1504 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1505 				| NFS_INO_REVAL_FORCED);
1506 
1507 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1508 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1509 	} else if (server->caps & NFS_CAP_CTIME)
1510 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1511 				| NFS_INO_REVAL_FORCED);
1512 
1513 	/* Check if our cached file size is stale */
1514 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1515 		new_isize = nfs_size_to_loff_t(fattr->size);
1516 		cur_isize = i_size_read(inode);
1517 		if (new_isize != cur_isize) {
1518 			/* Do we perhaps have any outstanding writes, or has
1519 			 * the file grown beyond our last write? */
1520 			if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) ||
1521 			     new_isize > cur_isize) {
1522 				i_size_write(inode, new_isize);
1523 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1524 			}
1525 			dprintk("NFS: isize change on server for file %s/%ld "
1526 					"(%Ld to %Ld)\n",
1527 					inode->i_sb->s_id,
1528 					inode->i_ino,
1529 					(long long)cur_isize,
1530 					(long long)new_isize);
1531 		}
1532 	} else
1533 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1534 				| NFS_INO_REVAL_PAGECACHE
1535 				| NFS_INO_REVAL_FORCED);
1536 
1537 
1538 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1539 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1540 	else if (server->caps & NFS_CAP_ATIME)
1541 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1542 				| NFS_INO_REVAL_FORCED);
1543 
1544 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1545 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1546 			umode_t newmode = inode->i_mode & S_IFMT;
1547 			newmode |= fattr->mode & S_IALLUGO;
1548 			inode->i_mode = newmode;
1549 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1550 		}
1551 	} else if (server->caps & NFS_CAP_MODE)
1552 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1553 				| NFS_INO_INVALID_ACCESS
1554 				| NFS_INO_INVALID_ACL
1555 				| NFS_INO_REVAL_FORCED);
1556 
1557 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1558 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1559 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1560 			inode->i_uid = fattr->uid;
1561 		}
1562 	} else if (server->caps & NFS_CAP_OWNER)
1563 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1564 				| NFS_INO_INVALID_ACCESS
1565 				| NFS_INO_INVALID_ACL
1566 				| NFS_INO_REVAL_FORCED);
1567 
1568 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1569 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1570 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1571 			inode->i_gid = fattr->gid;
1572 		}
1573 	} else if (server->caps & NFS_CAP_OWNER_GROUP)
1574 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1575 				| NFS_INO_INVALID_ACCESS
1576 				| NFS_INO_INVALID_ACL
1577 				| NFS_INO_REVAL_FORCED);
1578 
1579 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1580 		if (inode->i_nlink != fattr->nlink) {
1581 			invalid |= NFS_INO_INVALID_ATTR;
1582 			if (S_ISDIR(inode->i_mode))
1583 				invalid |= NFS_INO_INVALID_DATA;
1584 			set_nlink(inode, fattr->nlink);
1585 		}
1586 	} else if (server->caps & NFS_CAP_NLINK)
1587 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1588 				| NFS_INO_REVAL_FORCED);
1589 
1590 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1591 		/*
1592 		 * report the blocks in 512byte units
1593 		 */
1594 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1595  	}
1596 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1597 		inode->i_blocks = fattr->du.nfs2.blocks;
1598 
1599 	/* Update attrtimeo value if we're out of the unstable period */
1600 	if (invalid & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL)) {
1601 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1602 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1603 		nfsi->attrtimeo_timestamp = now;
1604 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1605 	} else {
1606 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1607 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1608 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1609 			nfsi->attrtimeo_timestamp = now;
1610 		}
1611 	}
1612 	invalid &= ~NFS_INO_INVALID_ATTR;
1613 	invalid &= ~NFS_INO_INVALID_LABEL;
1614 	/* Don't invalidate the data if we were to blame */
1615 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1616 				|| S_ISLNK(inode->i_mode)))
1617 		invalid &= ~NFS_INO_INVALID_DATA;
1618 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1619 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1620 		nfsi->cache_validity |= invalid;
1621 
1622 	if (invalid & NFS_INO_INVALID_DATA)
1623 		nfs_fscache_invalidate(inode);
1624 
1625 	return 0;
1626  out_err:
1627 	/*
1628 	 * No need to worry about unhashing the dentry, as the
1629 	 * lookup validation will know that the inode is bad.
1630 	 * (But we fall through to invalidate the caches.)
1631 	 */
1632 	nfs_invalidate_inode(inode);
1633 	return -ESTALE;
1634 }
1635 
1636 struct inode *nfs_alloc_inode(struct super_block *sb)
1637 {
1638 	struct nfs_inode *nfsi;
1639 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1640 	if (!nfsi)
1641 		return NULL;
1642 	nfsi->flags = 0UL;
1643 	nfsi->cache_validity = 0UL;
1644 #ifdef CONFIG_NFS_V3_ACL
1645 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1646 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1647 #endif
1648 #if IS_ENABLED(CONFIG_NFS_V4)
1649 	nfsi->nfs4_acl = NULL;
1650 #endif /* CONFIG_NFS_V4 */
1651 	return &nfsi->vfs_inode;
1652 }
1653 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1654 
1655 static void nfs_i_callback(struct rcu_head *head)
1656 {
1657 	struct inode *inode = container_of(head, struct inode, i_rcu);
1658 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1659 }
1660 
1661 void nfs_destroy_inode(struct inode *inode)
1662 {
1663 	call_rcu(&inode->i_rcu, nfs_i_callback);
1664 }
1665 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1666 
1667 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1668 {
1669 #if IS_ENABLED(CONFIG_NFS_V4)
1670 	INIT_LIST_HEAD(&nfsi->open_states);
1671 	nfsi->delegation = NULL;
1672 	nfsi->delegation_state = 0;
1673 	init_rwsem(&nfsi->rwsem);
1674 	nfsi->layout = NULL;
1675 #endif
1676 }
1677 
1678 static void init_once(void *foo)
1679 {
1680 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1681 
1682 	inode_init_once(&nfsi->vfs_inode);
1683 	INIT_LIST_HEAD(&nfsi->open_files);
1684 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1685 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1686 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1687 	nfsi->npages = 0;
1688 	nfsi->commit_info.ncommit = 0;
1689 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1690 	atomic_set(&nfsi->silly_count, 1);
1691 	INIT_HLIST_HEAD(&nfsi->silly_list);
1692 	init_waitqueue_head(&nfsi->waitqueue);
1693 	nfs4_init_once(nfsi);
1694 }
1695 
1696 static int __init nfs_init_inodecache(void)
1697 {
1698 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1699 					     sizeof(struct nfs_inode),
1700 					     0, (SLAB_RECLAIM_ACCOUNT|
1701 						SLAB_MEM_SPREAD),
1702 					     init_once);
1703 	if (nfs_inode_cachep == NULL)
1704 		return -ENOMEM;
1705 
1706 	return 0;
1707 }
1708 
1709 static void nfs_destroy_inodecache(void)
1710 {
1711 	/*
1712 	 * Make sure all delayed rcu free inodes are flushed before we
1713 	 * destroy cache.
1714 	 */
1715 	rcu_barrier();
1716 	kmem_cache_destroy(nfs_inode_cachep);
1717 }
1718 
1719 struct workqueue_struct *nfsiod_workqueue;
1720 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1721 
1722 /*
1723  * start up the nfsiod workqueue
1724  */
1725 static int nfsiod_start(void)
1726 {
1727 	struct workqueue_struct *wq;
1728 	dprintk("RPC:       creating workqueue nfsiod\n");
1729 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1730 	if (wq == NULL)
1731 		return -ENOMEM;
1732 	nfsiod_workqueue = wq;
1733 	return 0;
1734 }
1735 
1736 /*
1737  * Destroy the nfsiod workqueue
1738  */
1739 static void nfsiod_stop(void)
1740 {
1741 	struct workqueue_struct *wq;
1742 
1743 	wq = nfsiod_workqueue;
1744 	if (wq == NULL)
1745 		return;
1746 	nfsiod_workqueue = NULL;
1747 	destroy_workqueue(wq);
1748 }
1749 
1750 int nfs_net_id;
1751 EXPORT_SYMBOL_GPL(nfs_net_id);
1752 
1753 static int nfs_net_init(struct net *net)
1754 {
1755 	nfs_clients_init(net);
1756 	return 0;
1757 }
1758 
1759 static void nfs_net_exit(struct net *net)
1760 {
1761 	nfs_cleanup_cb_ident_idr(net);
1762 }
1763 
1764 static struct pernet_operations nfs_net_ops = {
1765 	.init = nfs_net_init,
1766 	.exit = nfs_net_exit,
1767 	.id   = &nfs_net_id,
1768 	.size = sizeof(struct nfs_net),
1769 };
1770 
1771 /*
1772  * Initialize NFS
1773  */
1774 static int __init init_nfs_fs(void)
1775 {
1776 	int err;
1777 
1778 	err = register_pernet_subsys(&nfs_net_ops);
1779 	if (err < 0)
1780 		goto out9;
1781 
1782 	err = nfs_fscache_register();
1783 	if (err < 0)
1784 		goto out8;
1785 
1786 	err = nfsiod_start();
1787 	if (err)
1788 		goto out7;
1789 
1790 	err = nfs_fs_proc_init();
1791 	if (err)
1792 		goto out6;
1793 
1794 	err = nfs_init_nfspagecache();
1795 	if (err)
1796 		goto out5;
1797 
1798 	err = nfs_init_inodecache();
1799 	if (err)
1800 		goto out4;
1801 
1802 	err = nfs_init_readpagecache();
1803 	if (err)
1804 		goto out3;
1805 
1806 	err = nfs_init_writepagecache();
1807 	if (err)
1808 		goto out2;
1809 
1810 	err = nfs_init_directcache();
1811 	if (err)
1812 		goto out1;
1813 
1814 #ifdef CONFIG_PROC_FS
1815 	rpc_proc_register(&init_net, &nfs_rpcstat);
1816 #endif
1817 	if ((err = register_nfs_fs()) != 0)
1818 		goto out0;
1819 
1820 	return 0;
1821 out0:
1822 #ifdef CONFIG_PROC_FS
1823 	rpc_proc_unregister(&init_net, "nfs");
1824 #endif
1825 	nfs_destroy_directcache();
1826 out1:
1827 	nfs_destroy_writepagecache();
1828 out2:
1829 	nfs_destroy_readpagecache();
1830 out3:
1831 	nfs_destroy_inodecache();
1832 out4:
1833 	nfs_destroy_nfspagecache();
1834 out5:
1835 	nfs_fs_proc_exit();
1836 out6:
1837 	nfsiod_stop();
1838 out7:
1839 	nfs_fscache_unregister();
1840 out8:
1841 	unregister_pernet_subsys(&nfs_net_ops);
1842 out9:
1843 	return err;
1844 }
1845 
1846 static void __exit exit_nfs_fs(void)
1847 {
1848 	nfs_destroy_directcache();
1849 	nfs_destroy_writepagecache();
1850 	nfs_destroy_readpagecache();
1851 	nfs_destroy_inodecache();
1852 	nfs_destroy_nfspagecache();
1853 	nfs_fscache_unregister();
1854 	unregister_pernet_subsys(&nfs_net_ops);
1855 #ifdef CONFIG_PROC_FS
1856 	rpc_proc_unregister(&init_net, "nfs");
1857 #endif
1858 	unregister_nfs_fs();
1859 	nfs_fs_proc_exit();
1860 	nfsiod_stop();
1861 }
1862 
1863 /* Not quite true; I just maintain it */
1864 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1865 MODULE_LICENSE("GPL");
1866 module_param(enable_ino64, bool, 0644);
1867 
1868 module_init(init_nfs_fs)
1869 module_exit(exit_nfs_fs)
1870