xref: /linux/fs/nfs/inode.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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(struct wait_bit_key *key)
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_final(&inode->i_data);
132 	clear_inode(inode);
133 	nfs_clear_inode(inode);
134 }
135 
136 int nfs_sync_inode(struct inode *inode)
137 {
138 	nfs_inode_dio_wait(inode);
139 	return nfs_wb_all(inode);
140 }
141 EXPORT_SYMBOL_GPL(nfs_sync_inode);
142 
143 /**
144  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
145  */
146 int nfs_sync_mapping(struct address_space *mapping)
147 {
148 	int ret = 0;
149 
150 	if (mapping->nrpages != 0) {
151 		unmap_mapping_range(mapping, 0, 0, 0);
152 		ret = nfs_wb_all(mapping->host);
153 	}
154 	return ret;
155 }
156 
157 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
158 {
159 	struct nfs_inode *nfsi = NFS_I(inode);
160 
161 	if (inode->i_mapping->nrpages == 0)
162 		flags &= ~NFS_INO_INVALID_DATA;
163 	nfsi->cache_validity |= flags;
164 	if (flags & NFS_INO_INVALID_DATA)
165 		nfs_fscache_invalidate(inode);
166 }
167 
168 /*
169  * Invalidate the local caches
170  */
171 static void nfs_zap_caches_locked(struct inode *inode)
172 {
173 	struct nfs_inode *nfsi = NFS_I(inode);
174 	int mode = inode->i_mode;
175 
176 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
177 
178 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
179 	nfsi->attrtimeo_timestamp = jiffies;
180 
181 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
182 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
183 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
184 					| NFS_INO_INVALID_DATA
185 					| NFS_INO_INVALID_ACCESS
186 					| NFS_INO_INVALID_ACL
187 					| NFS_INO_REVAL_PAGECACHE);
188 	} else
189 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
190 					| NFS_INO_INVALID_ACCESS
191 					| NFS_INO_INVALID_ACL
192 					| NFS_INO_REVAL_PAGECACHE);
193 	nfs_zap_label_cache_locked(nfsi);
194 }
195 
196 void nfs_zap_caches(struct inode *inode)
197 {
198 	spin_lock(&inode->i_lock);
199 	nfs_zap_caches_locked(inode);
200 	spin_unlock(&inode->i_lock);
201 }
202 
203 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
204 {
205 	if (mapping->nrpages != 0) {
206 		spin_lock(&inode->i_lock);
207 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
208 		spin_unlock(&inode->i_lock);
209 	}
210 }
211 
212 void nfs_zap_acl_cache(struct inode *inode)
213 {
214 	void (*clear_acl_cache)(struct inode *);
215 
216 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
217 	if (clear_acl_cache != NULL)
218 		clear_acl_cache(inode);
219 	spin_lock(&inode->i_lock);
220 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
221 	spin_unlock(&inode->i_lock);
222 }
223 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
224 
225 void nfs_invalidate_atime(struct inode *inode)
226 {
227 	spin_lock(&inode->i_lock);
228 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
229 	spin_unlock(&inode->i_lock);
230 }
231 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
232 
233 /*
234  * Invalidate, but do not unhash, the inode.
235  * NB: must be called with inode->i_lock held!
236  */
237 static void nfs_invalidate_inode(struct inode *inode)
238 {
239 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
240 	nfs_zap_caches_locked(inode);
241 }
242 
243 struct nfs_find_desc {
244 	struct nfs_fh		*fh;
245 	struct nfs_fattr	*fattr;
246 };
247 
248 /*
249  * In NFSv3 we can have 64bit inode numbers. In order to support
250  * this, and re-exported directories (also seen in NFSv2)
251  * we are forced to allow 2 different inodes to have the same
252  * i_ino.
253  */
254 static int
255 nfs_find_actor(struct inode *inode, void *opaque)
256 {
257 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
258 	struct nfs_fh		*fh = desc->fh;
259 	struct nfs_fattr	*fattr = desc->fattr;
260 
261 	if (NFS_FILEID(inode) != fattr->fileid)
262 		return 0;
263 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
264 		return 0;
265 	if (nfs_compare_fh(NFS_FH(inode), fh))
266 		return 0;
267 	if (is_bad_inode(inode) || NFS_STALE(inode))
268 		return 0;
269 	return 1;
270 }
271 
272 static int
273 nfs_init_locked(struct inode *inode, void *opaque)
274 {
275 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
276 	struct nfs_fattr	*fattr = desc->fattr;
277 
278 	set_nfs_fileid(inode, fattr->fileid);
279 	nfs_copy_fh(NFS_FH(inode), desc->fh);
280 	return 0;
281 }
282 
283 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
284 static void nfs_clear_label_invalid(struct inode *inode)
285 {
286 	spin_lock(&inode->i_lock);
287 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
288 	spin_unlock(&inode->i_lock);
289 }
290 
291 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
292 					struct nfs4_label *label)
293 {
294 	int error;
295 
296 	if (label == NULL)
297 		return;
298 
299 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
300 		error = security_inode_notifysecctx(inode, label->label,
301 				label->len);
302 		if (error)
303 			printk(KERN_ERR "%s() %s %d "
304 					"security_inode_notifysecctx() %d\n",
305 					__func__,
306 					(char *)label->label,
307 					label->len, error);
308 		nfs_clear_label_invalid(inode);
309 	}
310 }
311 
312 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
313 {
314 	struct nfs4_label *label = NULL;
315 	int minor_version = server->nfs_client->cl_minorversion;
316 
317 	if (minor_version < 2)
318 		return label;
319 
320 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
321 		return label;
322 
323 	label = kzalloc(sizeof(struct nfs4_label), flags);
324 	if (label == NULL)
325 		return ERR_PTR(-ENOMEM);
326 
327 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
328 	if (label->label == NULL) {
329 		kfree(label);
330 		return ERR_PTR(-ENOMEM);
331 	}
332 	label->len = NFS4_MAXLABELLEN;
333 
334 	return label;
335 }
336 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
337 #else
338 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
339 					struct nfs4_label *label)
340 {
341 }
342 #endif
343 EXPORT_SYMBOL_GPL(nfs_setsecurity);
344 
345 /*
346  * This is our front-end to iget that looks up inodes by file handle
347  * instead of inode number.
348  */
349 struct inode *
350 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
351 {
352 	struct nfs_find_desc desc = {
353 		.fh	= fh,
354 		.fattr	= fattr
355 	};
356 	struct inode *inode = ERR_PTR(-ENOENT);
357 	unsigned long hash;
358 
359 	nfs_attr_check_mountpoint(sb, fattr);
360 
361 	if (nfs_attr_use_mounted_on_fileid(fattr))
362 		fattr->fileid = fattr->mounted_on_fileid;
363 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
364 		goto out_no_inode;
365 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
366 		goto out_no_inode;
367 
368 	hash = nfs_fattr_to_ino_t(fattr);
369 
370 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
371 	if (inode == NULL) {
372 		inode = ERR_PTR(-ENOMEM);
373 		goto out_no_inode;
374 	}
375 
376 	if (inode->i_state & I_NEW) {
377 		struct nfs_inode *nfsi = NFS_I(inode);
378 		unsigned long now = jiffies;
379 
380 		/* We set i_ino for the few things that still rely on it,
381 		 * such as stat(2) */
382 		inode->i_ino = hash;
383 
384 		/* We can't support update_atime(), since the server will reset it */
385 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
386 		inode->i_mode = fattr->mode;
387 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
388 				&& nfs_server_capable(inode, NFS_CAP_MODE))
389 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
390 		/* Why so? Because we want revalidate for devices/FIFOs, and
391 		 * that's precisely what we have in nfs_file_inode_operations.
392 		 */
393 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
394 		if (S_ISREG(inode->i_mode)) {
395 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
396 			inode->i_data.a_ops = &nfs_file_aops;
397 		} else if (S_ISDIR(inode->i_mode)) {
398 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
399 			inode->i_fop = &nfs_dir_operations;
400 			inode->i_data.a_ops = &nfs_dir_aops;
401 			/* Deal with crossing mountpoints */
402 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
403 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
404 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
405 					inode->i_op = &nfs_referral_inode_operations;
406 				else
407 					inode->i_op = &nfs_mountpoint_inode_operations;
408 				inode->i_fop = NULL;
409 				inode->i_flags |= S_AUTOMOUNT;
410 			}
411 		} else if (S_ISLNK(inode->i_mode))
412 			inode->i_op = &nfs_symlink_inode_operations;
413 		else
414 			init_special_inode(inode, inode->i_mode, fattr->rdev);
415 
416 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
417 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
418 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
419 		inode->i_version = 0;
420 		inode->i_size = 0;
421 		clear_nlink(inode);
422 		inode->i_uid = make_kuid(&init_user_ns, -2);
423 		inode->i_gid = make_kgid(&init_user_ns, -2);
424 		inode->i_blocks = 0;
425 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
426 		nfsi->write_io = 0;
427 		nfsi->read_io = 0;
428 
429 		nfsi->read_cache_jiffies = fattr->time_start;
430 		nfsi->attr_gencount = fattr->gencount;
431 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
432 			inode->i_atime = fattr->atime;
433 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
434 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
435 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
436 			inode->i_mtime = fattr->mtime;
437 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
438 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
439 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
440 			inode->i_ctime = fattr->ctime;
441 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
442 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
443 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
444 			inode->i_version = fattr->change_attr;
445 		else
446 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
447 				| NFS_INO_REVAL_PAGECACHE);
448 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
449 			inode->i_size = nfs_size_to_loff_t(fattr->size);
450 		else
451 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
452 				| NFS_INO_REVAL_PAGECACHE);
453 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
454 			set_nlink(inode, fattr->nlink);
455 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
456 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
457 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
458 			inode->i_uid = fattr->uid;
459 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
460 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
461 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
462 			inode->i_gid = fattr->gid;
463 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
464 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
465 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
466 			inode->i_blocks = fattr->du.nfs2.blocks;
467 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
468 			/*
469 			 * report the blocks in 512byte units
470 			 */
471 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
472 		}
473 
474 		nfs_setsecurity(inode, fattr, label);
475 
476 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
477 		nfsi->attrtimeo_timestamp = now;
478 		nfsi->access_cache = RB_ROOT;
479 
480 		nfs_fscache_init_inode(inode);
481 
482 		unlock_new_inode(inode);
483 	} else
484 		nfs_refresh_inode(inode, fattr);
485 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
486 		inode->i_sb->s_id,
487 		(unsigned long long)NFS_FILEID(inode),
488 		nfs_display_fhandle_hash(fh),
489 		atomic_read(&inode->i_count));
490 
491 out:
492 	return inode;
493 
494 out_no_inode:
495 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
496 	goto out;
497 }
498 EXPORT_SYMBOL_GPL(nfs_fhget);
499 
500 #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)
501 
502 int
503 nfs_setattr(struct dentry *dentry, struct iattr *attr)
504 {
505 	struct inode *inode = d_inode(dentry);
506 	struct nfs_fattr *fattr;
507 	int error = 0;
508 
509 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
510 
511 	/* skip mode change if it's just for clearing setuid/setgid */
512 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
513 		attr->ia_valid &= ~ATTR_MODE;
514 
515 	if (attr->ia_valid & ATTR_SIZE) {
516 		BUG_ON(!S_ISREG(inode->i_mode));
517 
518 		error = inode_newsize_ok(inode, attr->ia_size);
519 		if (error)
520 			return error;
521 
522 		if (attr->ia_size == i_size_read(inode))
523 			attr->ia_valid &= ~ATTR_SIZE;
524 	}
525 
526 	/* Optimization: if the end result is no change, don't RPC */
527 	attr->ia_valid &= NFS_VALID_ATTRS;
528 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
529 		return 0;
530 
531 	trace_nfs_setattr_enter(inode);
532 
533 	/* Write all dirty data */
534 	if (S_ISREG(inode->i_mode))
535 		nfs_sync_inode(inode);
536 
537 	fattr = nfs_alloc_fattr();
538 	if (fattr == NULL) {
539 		error = -ENOMEM;
540 		goto out;
541 	}
542 
543 	/*
544 	 * Return any delegations if we're going to change ACLs
545 	 */
546 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
547 		NFS_PROTO(inode)->return_delegation(inode);
548 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
549 	if (error == 0)
550 		error = nfs_refresh_inode(inode, fattr);
551 	nfs_free_fattr(fattr);
552 out:
553 	trace_nfs_setattr_exit(inode, error);
554 	return error;
555 }
556 EXPORT_SYMBOL_GPL(nfs_setattr);
557 
558 /**
559  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
560  * @inode: inode of the file used
561  * @offset: file offset to start truncating
562  *
563  * This is a copy of the common vmtruncate, but with the locking
564  * corrected to take into account the fact that NFS requires
565  * inode->i_size to be updated under the inode->i_lock.
566  * Note: must be called with inode->i_lock held!
567  */
568 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
569 {
570 	int err;
571 
572 	err = inode_newsize_ok(inode, offset);
573 	if (err)
574 		goto out;
575 
576 	i_size_write(inode, offset);
577 	/* Optimisation */
578 	if (offset == 0)
579 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
580 
581 	spin_unlock(&inode->i_lock);
582 	truncate_pagecache(inode, offset);
583 	spin_lock(&inode->i_lock);
584 out:
585 	return err;
586 }
587 
588 /**
589  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
590  * @inode: pointer to struct inode
591  * @attr: pointer to struct iattr
592  *
593  * Note: we do this in the *proc.c in order to ensure that
594  *       it works for things like exclusive creates too.
595  */
596 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
597 		struct nfs_fattr *fattr)
598 {
599 	/* Barrier: bump the attribute generation count. */
600 	nfs_fattr_set_barrier(fattr);
601 
602 	spin_lock(&inode->i_lock);
603 	NFS_I(inode)->attr_gencount = fattr->gencount;
604 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
605 		if ((attr->ia_valid & ATTR_MODE) != 0) {
606 			int mode = attr->ia_mode & S_IALLUGO;
607 			mode |= inode->i_mode & ~S_IALLUGO;
608 			inode->i_mode = mode;
609 		}
610 		if ((attr->ia_valid & ATTR_UID) != 0)
611 			inode->i_uid = attr->ia_uid;
612 		if ((attr->ia_valid & ATTR_GID) != 0)
613 			inode->i_gid = attr->ia_gid;
614 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
615 				| NFS_INO_INVALID_ACL);
616 	}
617 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
618 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
619 		nfs_vmtruncate(inode, attr->ia_size);
620 	}
621 	nfs_update_inode(inode, fattr);
622 	spin_unlock(&inode->i_lock);
623 }
624 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
625 
626 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
627 {
628 	struct dentry *parent;
629 
630 	parent = dget_parent(dentry);
631 	nfs_force_use_readdirplus(d_inode(parent));
632 	dput(parent);
633 }
634 
635 static bool nfs_need_revalidate_inode(struct inode *inode)
636 {
637 	if (NFS_I(inode)->cache_validity &
638 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
639 		return true;
640 	if (nfs_attribute_cache_expired(inode))
641 		return true;
642 	return false;
643 }
644 
645 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
646 {
647 	struct inode *inode = d_inode(dentry);
648 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
649 	int err = 0;
650 
651 	trace_nfs_getattr_enter(inode);
652 	/* Flush out writes to the server in order to update c/mtime.  */
653 	if (S_ISREG(inode->i_mode)) {
654 		mutex_lock(&inode->i_mutex);
655 		err = nfs_sync_inode(inode);
656 		mutex_unlock(&inode->i_mutex);
657 		if (err)
658 			goto out;
659 	}
660 
661 	/*
662 	 * We may force a getattr if the user cares about atime.
663 	 *
664 	 * Note that we only have to check the vfsmount flags here:
665 	 *  - NFS always sets S_NOATIME by so checking it would give a
666 	 *    bogus result
667 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
668 	 *    no point in checking those.
669 	 */
670  	if ((mnt->mnt_flags & MNT_NOATIME) ||
671  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
672 		need_atime = 0;
673 
674 	if (need_atime || nfs_need_revalidate_inode(inode)) {
675 		struct nfs_server *server = NFS_SERVER(inode);
676 
677 		if (server->caps & NFS_CAP_READDIRPLUS)
678 			nfs_request_parent_use_readdirplus(dentry);
679 		err = __nfs_revalidate_inode(server, inode);
680 	}
681 	if (!err) {
682 		generic_fillattr(inode, stat);
683 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
684 		if (S_ISDIR(inode->i_mode))
685 			stat->blksize = NFS_SERVER(inode)->dtsize;
686 	}
687 out:
688 	trace_nfs_getattr_exit(inode, err);
689 	return err;
690 }
691 EXPORT_SYMBOL_GPL(nfs_getattr);
692 
693 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
694 {
695 	atomic_set(&l_ctx->count, 1);
696 	l_ctx->lockowner.l_owner = current->files;
697 	l_ctx->lockowner.l_pid = current->tgid;
698 	INIT_LIST_HEAD(&l_ctx->list);
699 	nfs_iocounter_init(&l_ctx->io_count);
700 }
701 
702 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
703 {
704 	struct nfs_lock_context *head = &ctx->lock_context;
705 	struct nfs_lock_context *pos = head;
706 
707 	do {
708 		if (pos->lockowner.l_owner != current->files)
709 			continue;
710 		if (pos->lockowner.l_pid != current->tgid)
711 			continue;
712 		atomic_inc(&pos->count);
713 		return pos;
714 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
715 	return NULL;
716 }
717 
718 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
719 {
720 	struct nfs_lock_context *res, *new = NULL;
721 	struct inode *inode = d_inode(ctx->dentry);
722 
723 	spin_lock(&inode->i_lock);
724 	res = __nfs_find_lock_context(ctx);
725 	if (res == NULL) {
726 		spin_unlock(&inode->i_lock);
727 		new = kmalloc(sizeof(*new), GFP_KERNEL);
728 		if (new == NULL)
729 			return ERR_PTR(-ENOMEM);
730 		nfs_init_lock_context(new);
731 		spin_lock(&inode->i_lock);
732 		res = __nfs_find_lock_context(ctx);
733 		if (res == NULL) {
734 			list_add_tail(&new->list, &ctx->lock_context.list);
735 			new->open_context = ctx;
736 			res = new;
737 			new = NULL;
738 		}
739 	}
740 	spin_unlock(&inode->i_lock);
741 	kfree(new);
742 	return res;
743 }
744 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
745 
746 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
747 {
748 	struct nfs_open_context *ctx = l_ctx->open_context;
749 	struct inode *inode = d_inode(ctx->dentry);
750 
751 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
752 		return;
753 	list_del(&l_ctx->list);
754 	spin_unlock(&inode->i_lock);
755 	kfree(l_ctx);
756 }
757 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
758 
759 /**
760  * nfs_close_context - Common close_context() routine NFSv2/v3
761  * @ctx: pointer to context
762  * @is_sync: is this a synchronous close
763  *
764  * Ensure that the attributes are up to date if we're mounted
765  * with close-to-open semantics and we have cached data that will
766  * need to be revalidated on open.
767  */
768 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
769 {
770 	struct nfs_inode *nfsi;
771 	struct inode *inode;
772 	struct nfs_server *server;
773 
774 	if (!(ctx->mode & FMODE_WRITE))
775 		return;
776 	if (!is_sync)
777 		return;
778 	inode = d_inode(ctx->dentry);
779 	nfsi = NFS_I(inode);
780 	if (inode->i_mapping->nrpages == 0)
781 		return;
782 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
783 		return;
784 	if (!list_empty(&nfsi->open_files))
785 		return;
786 	server = NFS_SERVER(inode);
787 	if (server->flags & NFS_MOUNT_NOCTO)
788 		return;
789 	nfs_revalidate_inode(server, inode);
790 }
791 EXPORT_SYMBOL_GPL(nfs_close_context);
792 
793 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
794 {
795 	struct nfs_open_context *ctx;
796 	struct rpc_cred *cred = rpc_lookup_cred();
797 	if (IS_ERR(cred))
798 		return ERR_CAST(cred);
799 
800 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
801 	if (!ctx) {
802 		put_rpccred(cred);
803 		return ERR_PTR(-ENOMEM);
804 	}
805 	nfs_sb_active(dentry->d_sb);
806 	ctx->dentry = dget(dentry);
807 	ctx->cred = cred;
808 	ctx->state = NULL;
809 	ctx->mode = f_mode;
810 	ctx->flags = 0;
811 	ctx->error = 0;
812 	nfs_init_lock_context(&ctx->lock_context);
813 	ctx->lock_context.open_context = ctx;
814 	INIT_LIST_HEAD(&ctx->list);
815 	ctx->mdsthreshold = NULL;
816 	return ctx;
817 }
818 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
819 
820 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
821 {
822 	if (ctx != NULL)
823 		atomic_inc(&ctx->lock_context.count);
824 	return ctx;
825 }
826 EXPORT_SYMBOL_GPL(get_nfs_open_context);
827 
828 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
829 {
830 	struct inode *inode = d_inode(ctx->dentry);
831 	struct super_block *sb = ctx->dentry->d_sb;
832 
833 	if (!list_empty(&ctx->list)) {
834 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
835 			return;
836 		list_del(&ctx->list);
837 		spin_unlock(&inode->i_lock);
838 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
839 		return;
840 	if (inode != NULL)
841 		NFS_PROTO(inode)->close_context(ctx, is_sync);
842 	if (ctx->cred != NULL)
843 		put_rpccred(ctx->cred);
844 	dput(ctx->dentry);
845 	nfs_sb_deactive(sb);
846 	kfree(ctx->mdsthreshold);
847 	kfree(ctx);
848 }
849 
850 void put_nfs_open_context(struct nfs_open_context *ctx)
851 {
852 	__put_nfs_open_context(ctx, 0);
853 }
854 EXPORT_SYMBOL_GPL(put_nfs_open_context);
855 
856 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
857 {
858 	__put_nfs_open_context(ctx, 1);
859 }
860 
861 /*
862  * Ensure that mmap has a recent RPC credential for use when writing out
863  * shared pages
864  */
865 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
866 {
867 	struct inode *inode = d_inode(ctx->dentry);
868 	struct nfs_inode *nfsi = NFS_I(inode);
869 
870 	spin_lock(&inode->i_lock);
871 	list_add(&ctx->list, &nfsi->open_files);
872 	spin_unlock(&inode->i_lock);
873 }
874 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
875 
876 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
877 {
878 	filp->private_data = get_nfs_open_context(ctx);
879 	if (list_empty(&ctx->list))
880 		nfs_inode_attach_open_context(ctx);
881 }
882 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
883 
884 /*
885  * Given an inode, search for an open context with the desired characteristics
886  */
887 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
888 {
889 	struct nfs_inode *nfsi = NFS_I(inode);
890 	struct nfs_open_context *pos, *ctx = NULL;
891 
892 	spin_lock(&inode->i_lock);
893 	list_for_each_entry(pos, &nfsi->open_files, list) {
894 		if (cred != NULL && pos->cred != cred)
895 			continue;
896 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
897 			continue;
898 		ctx = get_nfs_open_context(pos);
899 		break;
900 	}
901 	spin_unlock(&inode->i_lock);
902 	return ctx;
903 }
904 
905 void nfs_file_clear_open_context(struct file *filp)
906 {
907 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
908 
909 	if (ctx) {
910 		struct inode *inode = d_inode(ctx->dentry);
911 
912 		filp->private_data = NULL;
913 		spin_lock(&inode->i_lock);
914 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
915 		spin_unlock(&inode->i_lock);
916 		put_nfs_open_context_sync(ctx);
917 	}
918 }
919 
920 /*
921  * These allocate and release file read/write context information.
922  */
923 int nfs_open(struct inode *inode, struct file *filp)
924 {
925 	struct nfs_open_context *ctx;
926 
927 	ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
928 	if (IS_ERR(ctx))
929 		return PTR_ERR(ctx);
930 	nfs_file_set_open_context(filp, ctx);
931 	put_nfs_open_context(ctx);
932 	nfs_fscache_open_file(inode, filp);
933 	return 0;
934 }
935 
936 /*
937  * This function is called whenever some part of NFS notices that
938  * the cached attributes have to be refreshed.
939  */
940 int
941 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
942 {
943 	int		 status = -ESTALE;
944 	struct nfs4_label *label = NULL;
945 	struct nfs_fattr *fattr = NULL;
946 	struct nfs_inode *nfsi = NFS_I(inode);
947 
948 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
949 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
950 
951 	trace_nfs_revalidate_inode_enter(inode);
952 
953 	if (is_bad_inode(inode))
954 		goto out;
955 	if (NFS_STALE(inode))
956 		goto out;
957 
958 	status = -ENOMEM;
959 	fattr = nfs_alloc_fattr();
960 	if (fattr == NULL)
961 		goto out;
962 
963 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
964 
965 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
966 	if (IS_ERR(label)) {
967 		status = PTR_ERR(label);
968 		goto out;
969 	}
970 
971 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
972 	if (status != 0) {
973 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
974 			 inode->i_sb->s_id,
975 			 (unsigned long long)NFS_FILEID(inode), status);
976 		if (status == -ESTALE) {
977 			nfs_zap_caches(inode);
978 			if (!S_ISDIR(inode->i_mode))
979 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
980 		}
981 		goto err_out;
982 	}
983 
984 	status = nfs_refresh_inode(inode, fattr);
985 	if (status) {
986 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
987 			 inode->i_sb->s_id,
988 			 (unsigned long long)NFS_FILEID(inode), status);
989 		goto err_out;
990 	}
991 
992 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
993 		nfs_zap_acl_cache(inode);
994 
995 	nfs_setsecurity(inode, fattr, label);
996 
997 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
998 		inode->i_sb->s_id,
999 		(unsigned long long)NFS_FILEID(inode));
1000 
1001 err_out:
1002 	nfs4_label_free(label);
1003 out:
1004 	nfs_free_fattr(fattr);
1005 	trace_nfs_revalidate_inode_exit(inode, status);
1006 	return status;
1007 }
1008 
1009 int nfs_attribute_timeout(struct inode *inode)
1010 {
1011 	struct nfs_inode *nfsi = NFS_I(inode);
1012 
1013 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1014 }
1015 
1016 int nfs_attribute_cache_expired(struct inode *inode)
1017 {
1018 	if (nfs_have_delegated_attributes(inode))
1019 		return 0;
1020 	return nfs_attribute_timeout(inode);
1021 }
1022 
1023 /**
1024  * nfs_revalidate_inode - Revalidate the inode attributes
1025  * @server - pointer to nfs_server struct
1026  * @inode - pointer to inode struct
1027  *
1028  * Updates inode attribute information by retrieving the data from the server.
1029  */
1030 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1031 {
1032 	if (!nfs_need_revalidate_inode(inode))
1033 		return NFS_STALE(inode) ? -ESTALE : 0;
1034 	return __nfs_revalidate_inode(server, inode);
1035 }
1036 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1037 
1038 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1039 {
1040 	if (!(NFS_I(inode)->cache_validity &
1041 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1042 			&& !nfs_attribute_cache_expired(inode))
1043 		return NFS_STALE(inode) ? -ESTALE : 0;
1044 	return -ECHILD;
1045 }
1046 
1047 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1048 {
1049 	struct nfs_inode *nfsi = NFS_I(inode);
1050 	int ret;
1051 
1052 	if (mapping->nrpages != 0) {
1053 		if (S_ISREG(inode->i_mode)) {
1054 			unmap_mapping_range(mapping, 0, 0, 0);
1055 			ret = nfs_sync_mapping(mapping);
1056 			if (ret < 0)
1057 				return ret;
1058 		}
1059 		ret = invalidate_inode_pages2(mapping);
1060 		if (ret < 0)
1061 			return ret;
1062 	}
1063 	if (S_ISDIR(inode->i_mode)) {
1064 		spin_lock(&inode->i_lock);
1065 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1066 		spin_unlock(&inode->i_lock);
1067 	}
1068 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1069 	nfs_fscache_wait_on_invalidate(inode);
1070 
1071 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1072 			inode->i_sb->s_id,
1073 			(unsigned long long)NFS_FILEID(inode));
1074 	return 0;
1075 }
1076 
1077 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1078 {
1079 	if (nfs_have_delegated_attributes(inode))
1080 		return false;
1081 	return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1082 		|| nfs_attribute_timeout(inode)
1083 		|| NFS_STALE(inode);
1084 }
1085 
1086 /**
1087  * __nfs_revalidate_mapping - Revalidate the pagecache
1088  * @inode - pointer to host inode
1089  * @mapping - pointer to mapping
1090  * @may_lock - take inode->i_mutex?
1091  */
1092 static int __nfs_revalidate_mapping(struct inode *inode,
1093 		struct address_space *mapping,
1094 		bool may_lock)
1095 {
1096 	struct nfs_inode *nfsi = NFS_I(inode);
1097 	unsigned long *bitlock = &nfsi->flags;
1098 	int ret = 0;
1099 
1100 	/* swapfiles are not supposed to be shared. */
1101 	if (IS_SWAPFILE(inode))
1102 		goto out;
1103 
1104 	if (nfs_mapping_need_revalidate_inode(inode)) {
1105 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1106 		if (ret < 0)
1107 			goto out;
1108 	}
1109 
1110 	/*
1111 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1112 	 * invalidations that come in while we're shooting down the mappings
1113 	 * are respected. But, that leaves a race window where one revalidator
1114 	 * can clear the flag, and then another checks it before the mapping
1115 	 * gets invalidated. Fix that by serializing access to this part of
1116 	 * the function.
1117 	 *
1118 	 * At the same time, we need to allow other tasks to see whether we
1119 	 * might be in the middle of invalidating the pages, so we only set
1120 	 * the bit lock here if it looks like we're going to be doing that.
1121 	 */
1122 	for (;;) {
1123 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1124 					 nfs_wait_bit_killable, TASK_KILLABLE);
1125 		if (ret)
1126 			goto out;
1127 		spin_lock(&inode->i_lock);
1128 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1129 			spin_unlock(&inode->i_lock);
1130 			continue;
1131 		}
1132 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1133 			break;
1134 		spin_unlock(&inode->i_lock);
1135 		goto out;
1136 	}
1137 
1138 	set_bit(NFS_INO_INVALIDATING, bitlock);
1139 	smp_wmb();
1140 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1141 	spin_unlock(&inode->i_lock);
1142 	trace_nfs_invalidate_mapping_enter(inode);
1143 	if (may_lock) {
1144 		mutex_lock(&inode->i_mutex);
1145 		ret = nfs_invalidate_mapping(inode, mapping);
1146 		mutex_unlock(&inode->i_mutex);
1147 	} else
1148 		ret = nfs_invalidate_mapping(inode, mapping);
1149 	trace_nfs_invalidate_mapping_exit(inode, ret);
1150 
1151 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1152 	smp_mb__after_atomic();
1153 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1154 out:
1155 	return ret;
1156 }
1157 
1158 /**
1159  * nfs_revalidate_mapping - Revalidate the pagecache
1160  * @inode - pointer to host inode
1161  * @mapping - pointer to mapping
1162  */
1163 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1164 {
1165 	return __nfs_revalidate_mapping(inode, mapping, false);
1166 }
1167 
1168 /**
1169  * nfs_revalidate_mapping_protected - Revalidate the pagecache
1170  * @inode - pointer to host inode
1171  * @mapping - pointer to mapping
1172  *
1173  * Differs from nfs_revalidate_mapping() in that it grabs the inode->i_mutex
1174  * while invalidating the mapping.
1175  */
1176 int nfs_revalidate_mapping_protected(struct inode *inode, struct address_space *mapping)
1177 {
1178 	return __nfs_revalidate_mapping(inode, mapping, true);
1179 }
1180 
1181 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1182 {
1183 	struct nfs_inode *nfsi = NFS_I(inode);
1184 	unsigned long ret = 0;
1185 
1186 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1187 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1188 			&& inode->i_version == fattr->pre_change_attr) {
1189 		inode->i_version = fattr->change_attr;
1190 		if (S_ISDIR(inode->i_mode))
1191 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1192 		ret |= NFS_INO_INVALID_ATTR;
1193 	}
1194 	/* If we have atomic WCC data, we may update some attributes */
1195 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1196 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1197 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1198 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1199 		ret |= NFS_INO_INVALID_ATTR;
1200 	}
1201 
1202 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1203 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1204 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1205 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1206 		if (S_ISDIR(inode->i_mode))
1207 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1208 		ret |= NFS_INO_INVALID_ATTR;
1209 	}
1210 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1211 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1212 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1213 			&& nfsi->nrequests == 0) {
1214 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1215 		ret |= NFS_INO_INVALID_ATTR;
1216 	}
1217 
1218 	return ret;
1219 }
1220 
1221 /**
1222  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1223  * @inode - pointer to inode
1224  * @fattr - updated attributes
1225  *
1226  * Verifies the attribute cache. If we have just changed the attributes,
1227  * so that fattr carries weak cache consistency data, then it may
1228  * also update the ctime/mtime/change_attribute.
1229  */
1230 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1231 {
1232 	struct nfs_inode *nfsi = NFS_I(inode);
1233 	loff_t cur_size, new_isize;
1234 	unsigned long invalid = 0;
1235 
1236 
1237 	if (nfs_have_delegated_attributes(inode))
1238 		return 0;
1239 	/* Has the inode gone and changed behind our back? */
1240 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1241 		return -EIO;
1242 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1243 		return -EIO;
1244 
1245 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1246 			inode->i_version != fattr->change_attr)
1247 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1248 
1249 	/* Verify a few of the more important attributes */
1250 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1251 		invalid |= NFS_INO_INVALID_ATTR;
1252 
1253 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1254 		cur_size = i_size_read(inode);
1255 		new_isize = nfs_size_to_loff_t(fattr->size);
1256 		if (cur_size != new_isize)
1257 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1258 	}
1259 	if (nfsi->nrequests != 0)
1260 		invalid &= ~NFS_INO_REVAL_PAGECACHE;
1261 
1262 	/* Have any file permissions changed? */
1263 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1264 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1265 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1266 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1267 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1268 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1269 
1270 	/* Has the link count changed? */
1271 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1272 		invalid |= NFS_INO_INVALID_ATTR;
1273 
1274 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1275 		invalid |= NFS_INO_INVALID_ATIME;
1276 
1277 	if (invalid != 0)
1278 		nfs_set_cache_invalid(inode, invalid);
1279 
1280 	nfsi->read_cache_jiffies = fattr->time_start;
1281 	return 0;
1282 }
1283 
1284 static atomic_long_t nfs_attr_generation_counter;
1285 
1286 static unsigned long nfs_read_attr_generation_counter(void)
1287 {
1288 	return atomic_long_read(&nfs_attr_generation_counter);
1289 }
1290 
1291 unsigned long nfs_inc_attr_generation_counter(void)
1292 {
1293 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1294 }
1295 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1296 
1297 void nfs_fattr_init(struct nfs_fattr *fattr)
1298 {
1299 	fattr->valid = 0;
1300 	fattr->time_start = jiffies;
1301 	fattr->gencount = nfs_inc_attr_generation_counter();
1302 	fattr->owner_name = NULL;
1303 	fattr->group_name = NULL;
1304 }
1305 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1306 
1307 /**
1308  * nfs_fattr_set_barrier
1309  * @fattr: attributes
1310  *
1311  * Used to set a barrier after an attribute was updated. This
1312  * barrier ensures that older attributes from RPC calls that may
1313  * have raced with our update cannot clobber these new values.
1314  * Note that you are still responsible for ensuring that other
1315  * operations which change the attribute on the server do not
1316  * collide.
1317  */
1318 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1319 {
1320 	fattr->gencount = nfs_inc_attr_generation_counter();
1321 }
1322 
1323 struct nfs_fattr *nfs_alloc_fattr(void)
1324 {
1325 	struct nfs_fattr *fattr;
1326 
1327 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1328 	if (fattr != NULL)
1329 		nfs_fattr_init(fattr);
1330 	return fattr;
1331 }
1332 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1333 
1334 struct nfs_fh *nfs_alloc_fhandle(void)
1335 {
1336 	struct nfs_fh *fh;
1337 
1338 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1339 	if (fh != NULL)
1340 		fh->size = 0;
1341 	return fh;
1342 }
1343 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1344 
1345 #ifdef NFS_DEBUG
1346 /*
1347  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1348  *                             in the same way that wireshark does
1349  *
1350  * @fh: file handle
1351  *
1352  * For debugging only.
1353  */
1354 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1355 {
1356 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1357 	 * not on the result */
1358 	return nfs_fhandle_hash(fh);
1359 }
1360 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1361 
1362 /*
1363  * _nfs_display_fhandle - display an NFS file handle on the console
1364  *
1365  * @fh: file handle to display
1366  * @caption: display caption
1367  *
1368  * For debugging only.
1369  */
1370 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1371 {
1372 	unsigned short i;
1373 
1374 	if (fh == NULL || fh->size == 0) {
1375 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1376 		return;
1377 	}
1378 
1379 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1380 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1381 	for (i = 0; i < fh->size; i += 16) {
1382 		__be32 *pos = (__be32 *)&fh->data[i];
1383 
1384 		switch ((fh->size - i - 1) >> 2) {
1385 		case 0:
1386 			printk(KERN_DEFAULT " %08x\n",
1387 				be32_to_cpup(pos));
1388 			break;
1389 		case 1:
1390 			printk(KERN_DEFAULT " %08x %08x\n",
1391 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1392 			break;
1393 		case 2:
1394 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1395 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1396 				be32_to_cpup(pos + 2));
1397 			break;
1398 		default:
1399 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1400 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1401 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1402 		}
1403 	}
1404 }
1405 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1406 #endif
1407 
1408 /**
1409  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1410  * @inode - pointer to inode
1411  * @fattr - attributes
1412  *
1413  * Attempt to divine whether or not an RPC call reply carrying stale
1414  * attributes got scheduled after another call carrying updated ones.
1415  *
1416  * To do so, the function first assumes that a more recent ctime means
1417  * that the attributes in fattr are newer, however it also attempt to
1418  * catch the case where ctime either didn't change, or went backwards
1419  * (if someone reset the clock on the server) by looking at whether
1420  * or not this RPC call was started after the inode was last updated.
1421  * Note also the check for wraparound of 'attr_gencount'
1422  *
1423  * The function returns 'true' if it thinks the attributes in 'fattr' are
1424  * more recent than the ones cached in the inode.
1425  *
1426  */
1427 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1428 {
1429 	const struct nfs_inode *nfsi = NFS_I(inode);
1430 
1431 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1432 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1433 }
1434 
1435 /*
1436  * Don't trust the change_attribute, mtime, ctime or size if
1437  * a pnfs LAYOUTCOMMIT is outstanding
1438  */
1439 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1440 		struct nfs_fattr *fattr)
1441 {
1442 	if (pnfs_layoutcommit_outstanding(inode))
1443 		fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1444 				NFS_ATTR_FATTR_MTIME |
1445 				NFS_ATTR_FATTR_CTIME |
1446 				NFS_ATTR_FATTR_SIZE);
1447 }
1448 
1449 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1450 {
1451 	int ret;
1452 
1453 	trace_nfs_refresh_inode_enter(inode);
1454 
1455 	nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1456 
1457 	if (nfs_inode_attrs_need_update(inode, fattr))
1458 		ret = nfs_update_inode(inode, fattr);
1459 	else
1460 		ret = nfs_check_inode_attributes(inode, fattr);
1461 
1462 	trace_nfs_refresh_inode_exit(inode, ret);
1463 	return ret;
1464 }
1465 
1466 /**
1467  * nfs_refresh_inode - try to update the inode attribute cache
1468  * @inode - pointer to inode
1469  * @fattr - updated attributes
1470  *
1471  * Check that an RPC call that returned attributes has not overlapped with
1472  * other recent updates of the inode metadata, then decide whether it is
1473  * safe to do a full update of the inode attributes, or whether just to
1474  * call nfs_check_inode_attributes.
1475  */
1476 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1477 {
1478 	int status;
1479 
1480 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1481 		return 0;
1482 	spin_lock(&inode->i_lock);
1483 	status = nfs_refresh_inode_locked(inode, fattr);
1484 	spin_unlock(&inode->i_lock);
1485 
1486 	return status;
1487 }
1488 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1489 
1490 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1491 {
1492 	unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1493 
1494 	/*
1495 	 * Don't revalidate the pagecache if we hold a delegation, but do
1496 	 * force an attribute update
1497 	 */
1498 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1499 		invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_FORCED;
1500 
1501 	if (S_ISDIR(inode->i_mode))
1502 		invalid |= NFS_INO_INVALID_DATA;
1503 	nfs_set_cache_invalid(inode, invalid);
1504 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1505 		return 0;
1506 	return nfs_refresh_inode_locked(inode, fattr);
1507 }
1508 
1509 /**
1510  * nfs_post_op_update_inode - try to update the inode attribute cache
1511  * @inode - pointer to inode
1512  * @fattr - updated attributes
1513  *
1514  * After an operation that has changed the inode metadata, mark the
1515  * attribute cache as being invalid, then try to update it.
1516  *
1517  * NB: if the server didn't return any post op attributes, this
1518  * function will force the retrieval of attributes before the next
1519  * NFS request.  Thus it should be used only for operations that
1520  * are expected to change one or more attributes, to avoid
1521  * unnecessary NFS requests and trips through nfs_update_inode().
1522  */
1523 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1524 {
1525 	int status;
1526 
1527 	spin_lock(&inode->i_lock);
1528 	nfs_fattr_set_barrier(fattr);
1529 	status = nfs_post_op_update_inode_locked(inode, fattr);
1530 	spin_unlock(&inode->i_lock);
1531 
1532 	return status;
1533 }
1534 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1535 
1536 /**
1537  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1538  * @inode - pointer to inode
1539  * @fattr - updated attributes
1540  *
1541  * After an operation that has changed the inode metadata, mark the
1542  * attribute cache as being invalid, then try to update it. Fake up
1543  * weak cache consistency data, if none exist.
1544  *
1545  * This function is mainly designed to be used by the ->write_done() functions.
1546  */
1547 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1548 {
1549 	int status;
1550 
1551 	/* Don't do a WCC update if these attributes are already stale */
1552 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1553 			!nfs_inode_attrs_need_update(inode, fattr)) {
1554 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1555 				| NFS_ATTR_FATTR_PRESIZE
1556 				| NFS_ATTR_FATTR_PREMTIME
1557 				| NFS_ATTR_FATTR_PRECTIME);
1558 		goto out_noforce;
1559 	}
1560 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1561 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1562 		fattr->pre_change_attr = inode->i_version;
1563 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1564 	}
1565 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1566 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1567 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1568 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1569 	}
1570 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1571 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1572 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1573 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1574 	}
1575 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1576 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1577 		fattr->pre_size = i_size_read(inode);
1578 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1579 	}
1580 out_noforce:
1581 	status = nfs_post_op_update_inode_locked(inode, fattr);
1582 	return status;
1583 }
1584 
1585 /**
1586  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1587  * @inode - pointer to inode
1588  * @fattr - updated attributes
1589  *
1590  * After an operation that has changed the inode metadata, mark the
1591  * attribute cache as being invalid, then try to update it. Fake up
1592  * weak cache consistency data, if none exist.
1593  *
1594  * This function is mainly designed to be used by the ->write_done() functions.
1595  */
1596 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1597 {
1598 	int status;
1599 
1600 	spin_lock(&inode->i_lock);
1601 	nfs_fattr_set_barrier(fattr);
1602 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1603 	spin_unlock(&inode->i_lock);
1604 	return status;
1605 }
1606 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1607 
1608 
1609 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1610 				    struct nfs_fattr *fattr)
1611 {
1612 	bool ret1 = true, ret2 = true;
1613 
1614 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1615 		ret1 = (nfsi->fileid == fattr->fileid);
1616 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1617 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1618 	return ret1 || ret2;
1619 }
1620 
1621 /*
1622  * Many nfs protocol calls return the new file attributes after
1623  * an operation.  Here we update the inode to reflect the state
1624  * of the server's inode.
1625  *
1626  * This is a bit tricky because we have to make sure all dirty pages
1627  * have been sent off to the server before calling invalidate_inode_pages.
1628  * To make sure no other process adds more write requests while we try
1629  * our best to flush them, we make them sleep during the attribute refresh.
1630  *
1631  * A very similar scenario holds for the dir cache.
1632  */
1633 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1634 {
1635 	struct nfs_server *server;
1636 	struct nfs_inode *nfsi = NFS_I(inode);
1637 	loff_t cur_isize, new_isize;
1638 	unsigned long invalid = 0;
1639 	unsigned long now = jiffies;
1640 	unsigned long save_cache_validity;
1641 
1642 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1643 			__func__, inode->i_sb->s_id, inode->i_ino,
1644 			nfs_display_fhandle_hash(NFS_FH(inode)),
1645 			atomic_read(&inode->i_count), fattr->valid);
1646 
1647 	if (!nfs_fileid_valid(nfsi, fattr)) {
1648 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1649 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1650 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1651 			inode->i_sb->s_id, (long long)nfsi->fileid,
1652 			(long long)fattr->fileid);
1653 		goto out_err;
1654 	}
1655 
1656 	/*
1657 	 * Make sure the inode's type hasn't changed.
1658 	 */
1659 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1660 		/*
1661 		* Big trouble! The inode has become a different object.
1662 		*/
1663 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1664 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1665 		goto out_err;
1666 	}
1667 
1668 	server = NFS_SERVER(inode);
1669 	/* Update the fsid? */
1670 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1671 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1672 			!IS_AUTOMOUNT(inode))
1673 		server->fsid = fattr->fsid;
1674 
1675 	/*
1676 	 * Update the read time so we don't revalidate too often.
1677 	 */
1678 	nfsi->read_cache_jiffies = fattr->time_start;
1679 
1680 	save_cache_validity = nfsi->cache_validity;
1681 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1682 			| NFS_INO_INVALID_ATIME
1683 			| NFS_INO_REVAL_FORCED
1684 			| NFS_INO_REVAL_PAGECACHE);
1685 
1686 	/* Do atomic weak cache consistency updates */
1687 	invalid |= nfs_wcc_update_inode(inode, fattr);
1688 
1689 	/* More cache consistency checks */
1690 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1691 		if (inode->i_version != fattr->change_attr) {
1692 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1693 					inode->i_sb->s_id, inode->i_ino);
1694 			invalid |= NFS_INO_INVALID_ATTR
1695 				| NFS_INO_INVALID_DATA
1696 				| NFS_INO_INVALID_ACCESS
1697 				| NFS_INO_INVALID_ACL;
1698 			if (S_ISDIR(inode->i_mode))
1699 				nfs_force_lookup_revalidate(inode);
1700 			inode->i_version = fattr->change_attr;
1701 		}
1702 	} else
1703 		nfsi->cache_validity |= save_cache_validity;
1704 
1705 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1706 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1707 	} else if (server->caps & NFS_CAP_MTIME)
1708 		nfsi->cache_validity |= save_cache_validity &
1709 				(NFS_INO_INVALID_ATTR
1710 				| NFS_INO_REVAL_FORCED);
1711 
1712 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1713 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1714 	} else if (server->caps & NFS_CAP_CTIME)
1715 		nfsi->cache_validity |= save_cache_validity &
1716 				(NFS_INO_INVALID_ATTR
1717 				| NFS_INO_REVAL_FORCED);
1718 
1719 	/* Check if our cached file size is stale */
1720 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1721 		new_isize = nfs_size_to_loff_t(fattr->size);
1722 		cur_isize = i_size_read(inode);
1723 		if (new_isize != cur_isize) {
1724 			/* Do we perhaps have any outstanding writes, or has
1725 			 * the file grown beyond our last write? */
1726 			if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1727 				i_size_write(inode, new_isize);
1728 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1729 			}
1730 			dprintk("NFS: isize change on server for file %s/%ld "
1731 					"(%Ld to %Ld)\n",
1732 					inode->i_sb->s_id,
1733 					inode->i_ino,
1734 					(long long)cur_isize,
1735 					(long long)new_isize);
1736 		}
1737 	} else
1738 		nfsi->cache_validity |= save_cache_validity &
1739 				(NFS_INO_INVALID_ATTR
1740 				| NFS_INO_REVAL_PAGECACHE
1741 				| NFS_INO_REVAL_FORCED);
1742 
1743 
1744 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1745 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1746 	else if (server->caps & NFS_CAP_ATIME)
1747 		nfsi->cache_validity |= save_cache_validity &
1748 				(NFS_INO_INVALID_ATIME
1749 				| NFS_INO_REVAL_FORCED);
1750 
1751 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1752 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1753 			umode_t newmode = inode->i_mode & S_IFMT;
1754 			newmode |= fattr->mode & S_IALLUGO;
1755 			inode->i_mode = newmode;
1756 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1757 		}
1758 	} else if (server->caps & NFS_CAP_MODE)
1759 		nfsi->cache_validity |= save_cache_validity &
1760 				(NFS_INO_INVALID_ATTR
1761 				| NFS_INO_INVALID_ACCESS
1762 				| NFS_INO_INVALID_ACL
1763 				| NFS_INO_REVAL_FORCED);
1764 
1765 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1766 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1767 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1768 			inode->i_uid = fattr->uid;
1769 		}
1770 	} else if (server->caps & NFS_CAP_OWNER)
1771 		nfsi->cache_validity |= save_cache_validity &
1772 				(NFS_INO_INVALID_ATTR
1773 				| NFS_INO_INVALID_ACCESS
1774 				| NFS_INO_INVALID_ACL
1775 				| NFS_INO_REVAL_FORCED);
1776 
1777 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1778 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1779 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1780 			inode->i_gid = fattr->gid;
1781 		}
1782 	} else if (server->caps & NFS_CAP_OWNER_GROUP)
1783 		nfsi->cache_validity |= save_cache_validity &
1784 				(NFS_INO_INVALID_ATTR
1785 				| NFS_INO_INVALID_ACCESS
1786 				| NFS_INO_INVALID_ACL
1787 				| NFS_INO_REVAL_FORCED);
1788 
1789 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1790 		if (inode->i_nlink != fattr->nlink) {
1791 			invalid |= NFS_INO_INVALID_ATTR;
1792 			if (S_ISDIR(inode->i_mode))
1793 				invalid |= NFS_INO_INVALID_DATA;
1794 			set_nlink(inode, fattr->nlink);
1795 		}
1796 	} else if (server->caps & NFS_CAP_NLINK)
1797 		nfsi->cache_validity |= save_cache_validity &
1798 				(NFS_INO_INVALID_ATTR
1799 				| NFS_INO_REVAL_FORCED);
1800 
1801 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1802 		/*
1803 		 * report the blocks in 512byte units
1804 		 */
1805 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1806  	}
1807 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1808 		inode->i_blocks = fattr->du.nfs2.blocks;
1809 
1810 	/* Update attrtimeo value if we're out of the unstable period */
1811 	if (invalid & NFS_INO_INVALID_ATTR) {
1812 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1813 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1814 		nfsi->attrtimeo_timestamp = now;
1815 		/* Set barrier to be more recent than all outstanding updates */
1816 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1817 	} else {
1818 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1819 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1820 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1821 			nfsi->attrtimeo_timestamp = now;
1822 		}
1823 		/* Set the barrier to be more recent than this fattr */
1824 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1825 			nfsi->attr_gencount = fattr->gencount;
1826 	}
1827 	invalid &= ~NFS_INO_INVALID_ATTR;
1828 	/* Don't invalidate the data if we were to blame */
1829 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1830 				|| S_ISLNK(inode->i_mode)))
1831 		invalid &= ~NFS_INO_INVALID_DATA;
1832 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1833 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1834 		nfs_set_cache_invalid(inode, invalid);
1835 
1836 	return 0;
1837  out_err:
1838 	/*
1839 	 * No need to worry about unhashing the dentry, as the
1840 	 * lookup validation will know that the inode is bad.
1841 	 * (But we fall through to invalidate the caches.)
1842 	 */
1843 	nfs_invalidate_inode(inode);
1844 	return -ESTALE;
1845 }
1846 
1847 struct inode *nfs_alloc_inode(struct super_block *sb)
1848 {
1849 	struct nfs_inode *nfsi;
1850 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1851 	if (!nfsi)
1852 		return NULL;
1853 	nfsi->flags = 0UL;
1854 	nfsi->cache_validity = 0UL;
1855 #if IS_ENABLED(CONFIG_NFS_V4)
1856 	nfsi->nfs4_acl = NULL;
1857 #endif /* CONFIG_NFS_V4 */
1858 	return &nfsi->vfs_inode;
1859 }
1860 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1861 
1862 static void nfs_i_callback(struct rcu_head *head)
1863 {
1864 	struct inode *inode = container_of(head, struct inode, i_rcu);
1865 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1866 }
1867 
1868 void nfs_destroy_inode(struct inode *inode)
1869 {
1870 	call_rcu(&inode->i_rcu, nfs_i_callback);
1871 }
1872 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1873 
1874 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1875 {
1876 #if IS_ENABLED(CONFIG_NFS_V4)
1877 	INIT_LIST_HEAD(&nfsi->open_states);
1878 	nfsi->delegation = NULL;
1879 	init_rwsem(&nfsi->rwsem);
1880 	nfsi->layout = NULL;
1881 #endif
1882 }
1883 
1884 static void init_once(void *foo)
1885 {
1886 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1887 
1888 	inode_init_once(&nfsi->vfs_inode);
1889 	INIT_LIST_HEAD(&nfsi->open_files);
1890 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1891 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1892 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1893 	nfsi->nrequests = 0;
1894 	nfsi->commit_info.ncommit = 0;
1895 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1896 	atomic_set(&nfsi->silly_count, 1);
1897 	INIT_HLIST_HEAD(&nfsi->silly_list);
1898 	init_waitqueue_head(&nfsi->waitqueue);
1899 	nfs4_init_once(nfsi);
1900 }
1901 
1902 static int __init nfs_init_inodecache(void)
1903 {
1904 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1905 					     sizeof(struct nfs_inode),
1906 					     0, (SLAB_RECLAIM_ACCOUNT|
1907 						SLAB_MEM_SPREAD),
1908 					     init_once);
1909 	if (nfs_inode_cachep == NULL)
1910 		return -ENOMEM;
1911 
1912 	return 0;
1913 }
1914 
1915 static void nfs_destroy_inodecache(void)
1916 {
1917 	/*
1918 	 * Make sure all delayed rcu free inodes are flushed before we
1919 	 * destroy cache.
1920 	 */
1921 	rcu_barrier();
1922 	kmem_cache_destroy(nfs_inode_cachep);
1923 }
1924 
1925 struct workqueue_struct *nfsiod_workqueue;
1926 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1927 
1928 /*
1929  * start up the nfsiod workqueue
1930  */
1931 static int nfsiod_start(void)
1932 {
1933 	struct workqueue_struct *wq;
1934 	dprintk("RPC:       creating workqueue nfsiod\n");
1935 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1936 	if (wq == NULL)
1937 		return -ENOMEM;
1938 	nfsiod_workqueue = wq;
1939 	return 0;
1940 }
1941 
1942 /*
1943  * Destroy the nfsiod workqueue
1944  */
1945 static void nfsiod_stop(void)
1946 {
1947 	struct workqueue_struct *wq;
1948 
1949 	wq = nfsiod_workqueue;
1950 	if (wq == NULL)
1951 		return;
1952 	nfsiod_workqueue = NULL;
1953 	destroy_workqueue(wq);
1954 }
1955 
1956 int nfs_net_id;
1957 EXPORT_SYMBOL_GPL(nfs_net_id);
1958 
1959 static int nfs_net_init(struct net *net)
1960 {
1961 	nfs_clients_init(net);
1962 	return nfs_fs_proc_net_init(net);
1963 }
1964 
1965 static void nfs_net_exit(struct net *net)
1966 {
1967 	nfs_fs_proc_net_exit(net);
1968 	nfs_cleanup_cb_ident_idr(net);
1969 }
1970 
1971 static struct pernet_operations nfs_net_ops = {
1972 	.init = nfs_net_init,
1973 	.exit = nfs_net_exit,
1974 	.id   = &nfs_net_id,
1975 	.size = sizeof(struct nfs_net),
1976 };
1977 
1978 /*
1979  * Initialize NFS
1980  */
1981 static int __init init_nfs_fs(void)
1982 {
1983 	int err;
1984 
1985 	err = register_pernet_subsys(&nfs_net_ops);
1986 	if (err < 0)
1987 		goto out9;
1988 
1989 	err = nfs_fscache_register();
1990 	if (err < 0)
1991 		goto out8;
1992 
1993 	err = nfsiod_start();
1994 	if (err)
1995 		goto out7;
1996 
1997 	err = nfs_fs_proc_init();
1998 	if (err)
1999 		goto out6;
2000 
2001 	err = nfs_init_nfspagecache();
2002 	if (err)
2003 		goto out5;
2004 
2005 	err = nfs_init_inodecache();
2006 	if (err)
2007 		goto out4;
2008 
2009 	err = nfs_init_readpagecache();
2010 	if (err)
2011 		goto out3;
2012 
2013 	err = nfs_init_writepagecache();
2014 	if (err)
2015 		goto out2;
2016 
2017 	err = nfs_init_directcache();
2018 	if (err)
2019 		goto out1;
2020 
2021 	rpc_proc_register(&init_net, &nfs_rpcstat);
2022 
2023 	err = register_nfs_fs();
2024 	if (err)
2025 		goto out0;
2026 
2027 	return 0;
2028 out0:
2029 	rpc_proc_unregister(&init_net, "nfs");
2030 	nfs_destroy_directcache();
2031 out1:
2032 	nfs_destroy_writepagecache();
2033 out2:
2034 	nfs_destroy_readpagecache();
2035 out3:
2036 	nfs_destroy_inodecache();
2037 out4:
2038 	nfs_destroy_nfspagecache();
2039 out5:
2040 	nfs_fs_proc_exit();
2041 out6:
2042 	nfsiod_stop();
2043 out7:
2044 	nfs_fscache_unregister();
2045 out8:
2046 	unregister_pernet_subsys(&nfs_net_ops);
2047 out9:
2048 	return err;
2049 }
2050 
2051 static void __exit exit_nfs_fs(void)
2052 {
2053 	nfs_destroy_directcache();
2054 	nfs_destroy_writepagecache();
2055 	nfs_destroy_readpagecache();
2056 	nfs_destroy_inodecache();
2057 	nfs_destroy_nfspagecache();
2058 	nfs_fscache_unregister();
2059 	unregister_pernet_subsys(&nfs_net_ops);
2060 	rpc_proc_unregister(&init_net, "nfs");
2061 	unregister_nfs_fs();
2062 	nfs_fs_proc_exit();
2063 	nfsiod_stop();
2064 }
2065 
2066 /* Not quite true; I just maintain it */
2067 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2068 MODULE_LICENSE("GPL");
2069 module_param(enable_ino64, bool, 0644);
2070 
2071 module_init(init_nfs_fs)
2072 module_exit(exit_nfs_fs)
2073