xref: /linux/fs/nfs/inode.c (revision 93d90ad708b8da6efc0e487b66111aa9db7f70c7)
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 /**
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 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
151 {
152 	struct nfs_inode *nfsi = NFS_I(inode);
153 
154 	if (inode->i_mapping->nrpages == 0)
155 		flags &= ~NFS_INO_INVALID_DATA;
156 	nfsi->cache_validity |= flags;
157 	if (flags & NFS_INO_INVALID_DATA)
158 		nfs_fscache_invalidate(inode);
159 }
160 
161 /*
162  * Invalidate the local caches
163  */
164 static void nfs_zap_caches_locked(struct inode *inode)
165 {
166 	struct nfs_inode *nfsi = NFS_I(inode);
167 	int mode = inode->i_mode;
168 
169 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
170 
171 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
172 	nfsi->attrtimeo_timestamp = jiffies;
173 
174 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
175 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
176 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
177 					| NFS_INO_INVALID_DATA
178 					| NFS_INO_INVALID_ACCESS
179 					| NFS_INO_INVALID_ACL
180 					| NFS_INO_REVAL_PAGECACHE);
181 	} else
182 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
183 					| NFS_INO_INVALID_ACCESS
184 					| NFS_INO_INVALID_ACL
185 					| NFS_INO_REVAL_PAGECACHE);
186 	nfs_zap_label_cache_locked(nfsi);
187 }
188 
189 void nfs_zap_caches(struct inode *inode)
190 {
191 	spin_lock(&inode->i_lock);
192 	nfs_zap_caches_locked(inode);
193 	spin_unlock(&inode->i_lock);
194 }
195 EXPORT_SYMBOL_GPL(nfs_zap_caches);
196 
197 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
198 {
199 	if (mapping->nrpages != 0) {
200 		spin_lock(&inode->i_lock);
201 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
202 		spin_unlock(&inode->i_lock);
203 	}
204 }
205 
206 void nfs_zap_acl_cache(struct inode *inode)
207 {
208 	void (*clear_acl_cache)(struct inode *);
209 
210 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
211 	if (clear_acl_cache != NULL)
212 		clear_acl_cache(inode);
213 	spin_lock(&inode->i_lock);
214 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
215 	spin_unlock(&inode->i_lock);
216 }
217 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
218 
219 void nfs_invalidate_atime(struct inode *inode)
220 {
221 	spin_lock(&inode->i_lock);
222 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
223 	spin_unlock(&inode->i_lock);
224 }
225 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
226 
227 /*
228  * Invalidate, but do not unhash, the inode.
229  * NB: must be called with inode->i_lock held!
230  */
231 static void nfs_invalidate_inode(struct inode *inode)
232 {
233 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
234 	nfs_zap_caches_locked(inode);
235 }
236 
237 struct nfs_find_desc {
238 	struct nfs_fh		*fh;
239 	struct nfs_fattr	*fattr;
240 };
241 
242 /*
243  * In NFSv3 we can have 64bit inode numbers. In order to support
244  * this, and re-exported directories (also seen in NFSv2)
245  * we are forced to allow 2 different inodes to have the same
246  * i_ino.
247  */
248 static int
249 nfs_find_actor(struct inode *inode, void *opaque)
250 {
251 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
252 	struct nfs_fh		*fh = desc->fh;
253 	struct nfs_fattr	*fattr = desc->fattr;
254 
255 	if (NFS_FILEID(inode) != fattr->fileid)
256 		return 0;
257 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
258 		return 0;
259 	if (nfs_compare_fh(NFS_FH(inode), fh))
260 		return 0;
261 	if (is_bad_inode(inode) || NFS_STALE(inode))
262 		return 0;
263 	return 1;
264 }
265 
266 static int
267 nfs_init_locked(struct inode *inode, void *opaque)
268 {
269 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
270 	struct nfs_fattr	*fattr = desc->fattr;
271 
272 	set_nfs_fileid(inode, fattr->fileid);
273 	nfs_copy_fh(NFS_FH(inode), desc->fh);
274 	return 0;
275 }
276 
277 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
278 static void nfs_clear_label_invalid(struct inode *inode)
279 {
280 	spin_lock(&inode->i_lock);
281 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
282 	spin_unlock(&inode->i_lock);
283 }
284 
285 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
286 					struct nfs4_label *label)
287 {
288 	int error;
289 
290 	if (label == NULL)
291 		return;
292 
293 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
294 		error = security_inode_notifysecctx(inode, label->label,
295 				label->len);
296 		if (error)
297 			printk(KERN_ERR "%s() %s %d "
298 					"security_inode_notifysecctx() %d\n",
299 					__func__,
300 					(char *)label->label,
301 					label->len, error);
302 		nfs_clear_label_invalid(inode);
303 	}
304 }
305 
306 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
307 {
308 	struct nfs4_label *label = NULL;
309 	int minor_version = server->nfs_client->cl_minorversion;
310 
311 	if (minor_version < 2)
312 		return label;
313 
314 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
315 		return label;
316 
317 	label = kzalloc(sizeof(struct nfs4_label), flags);
318 	if (label == NULL)
319 		return ERR_PTR(-ENOMEM);
320 
321 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
322 	if (label->label == NULL) {
323 		kfree(label);
324 		return ERR_PTR(-ENOMEM);
325 	}
326 	label->len = NFS4_MAXLABELLEN;
327 
328 	return label;
329 }
330 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
331 #else
332 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
333 					struct nfs4_label *label)
334 {
335 }
336 #endif
337 EXPORT_SYMBOL_GPL(nfs_setsecurity);
338 
339 /*
340  * This is our front-end to iget that looks up inodes by file handle
341  * instead of inode number.
342  */
343 struct inode *
344 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
345 {
346 	struct nfs_find_desc desc = {
347 		.fh	= fh,
348 		.fattr	= fattr
349 	};
350 	struct inode *inode = ERR_PTR(-ENOENT);
351 	unsigned long hash;
352 
353 	nfs_attr_check_mountpoint(sb, fattr);
354 
355 	if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
356 	    !nfs_attr_use_mounted_on_fileid(fattr))
357 		goto out_no_inode;
358 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
359 		goto out_no_inode;
360 
361 	hash = nfs_fattr_to_ino_t(fattr);
362 
363 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
364 	if (inode == NULL) {
365 		inode = ERR_PTR(-ENOMEM);
366 		goto out_no_inode;
367 	}
368 
369 	if (inode->i_state & I_NEW) {
370 		struct nfs_inode *nfsi = NFS_I(inode);
371 		unsigned long now = jiffies;
372 
373 		/* We set i_ino for the few things that still rely on it,
374 		 * such as stat(2) */
375 		inode->i_ino = hash;
376 
377 		/* We can't support update_atime(), since the server will reset it */
378 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
379 		inode->i_mode = fattr->mode;
380 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
381 				&& nfs_server_capable(inode, NFS_CAP_MODE))
382 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
383 		/* Why so? Because we want revalidate for devices/FIFOs, and
384 		 * that's precisely what we have in nfs_file_inode_operations.
385 		 */
386 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
387 		if (S_ISREG(inode->i_mode)) {
388 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
389 			inode->i_data.a_ops = &nfs_file_aops;
390 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
391 		} else if (S_ISDIR(inode->i_mode)) {
392 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
393 			inode->i_fop = &nfs_dir_operations;
394 			inode->i_data.a_ops = &nfs_dir_aops;
395 			/* Deal with crossing mountpoints */
396 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
397 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
398 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
399 					inode->i_op = &nfs_referral_inode_operations;
400 				else
401 					inode->i_op = &nfs_mountpoint_inode_operations;
402 				inode->i_fop = NULL;
403 				inode->i_flags |= S_AUTOMOUNT;
404 			}
405 		} else if (S_ISLNK(inode->i_mode))
406 			inode->i_op = &nfs_symlink_inode_operations;
407 		else
408 			init_special_inode(inode, inode->i_mode, fattr->rdev);
409 
410 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
411 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
412 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
413 		inode->i_version = 0;
414 		inode->i_size = 0;
415 		clear_nlink(inode);
416 		inode->i_uid = make_kuid(&init_user_ns, -2);
417 		inode->i_gid = make_kgid(&init_user_ns, -2);
418 		inode->i_blocks = 0;
419 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
420 		nfsi->write_io = 0;
421 		nfsi->read_io = 0;
422 
423 		nfsi->read_cache_jiffies = fattr->time_start;
424 		nfsi->attr_gencount = fattr->gencount;
425 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
426 			inode->i_atime = fattr->atime;
427 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
428 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
429 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
430 			inode->i_mtime = fattr->mtime;
431 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
432 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
433 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
434 			inode->i_ctime = fattr->ctime;
435 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
436 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
437 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
438 			inode->i_version = fattr->change_attr;
439 		else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
440 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
441 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
442 			inode->i_size = nfs_size_to_loff_t(fattr->size);
443 		else
444 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
445 				| NFS_INO_REVAL_PAGECACHE);
446 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
447 			set_nlink(inode, fattr->nlink);
448 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
449 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
450 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
451 			inode->i_uid = fattr->uid;
452 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
453 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
454 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
455 			inode->i_gid = fattr->gid;
456 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
457 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
458 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
459 			inode->i_blocks = fattr->du.nfs2.blocks;
460 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
461 			/*
462 			 * report the blocks in 512byte units
463 			 */
464 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
465 		}
466 
467 		nfs_setsecurity(inode, fattr, label);
468 
469 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
470 		nfsi->attrtimeo_timestamp = now;
471 		nfsi->access_cache = RB_ROOT;
472 
473 		nfs_fscache_init_inode(inode);
474 
475 		unlock_new_inode(inode);
476 	} else
477 		nfs_refresh_inode(inode, fattr);
478 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
479 		inode->i_sb->s_id,
480 		(unsigned long long)NFS_FILEID(inode),
481 		nfs_display_fhandle_hash(fh),
482 		atomic_read(&inode->i_count));
483 
484 out:
485 	return inode;
486 
487 out_no_inode:
488 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
489 	goto out;
490 }
491 EXPORT_SYMBOL_GPL(nfs_fhget);
492 
493 #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)
494 
495 int
496 nfs_setattr(struct dentry *dentry, struct iattr *attr)
497 {
498 	struct inode *inode = dentry->d_inode;
499 	struct nfs_fattr *fattr;
500 	int error = -ENOMEM;
501 
502 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
503 
504 	/* skip mode change if it's just for clearing setuid/setgid */
505 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
506 		attr->ia_valid &= ~ATTR_MODE;
507 
508 	if (attr->ia_valid & ATTR_SIZE) {
509 		BUG_ON(!S_ISREG(inode->i_mode));
510 
511 		if (attr->ia_size == i_size_read(inode))
512 			attr->ia_valid &= ~ATTR_SIZE;
513 	}
514 
515 	/* Optimization: if the end result is no change, don't RPC */
516 	attr->ia_valid &= NFS_VALID_ATTRS;
517 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
518 		return 0;
519 
520 	trace_nfs_setattr_enter(inode);
521 
522 	/* Write all dirty data */
523 	if (S_ISREG(inode->i_mode)) {
524 		nfs_inode_dio_wait(inode);
525 		nfs_wb_all(inode);
526 	}
527 
528 	fattr = nfs_alloc_fattr();
529 	if (fattr == NULL)
530 		goto out;
531 	/*
532 	 * Return any delegations if we're going to change ACLs
533 	 */
534 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
535 		NFS_PROTO(inode)->return_delegation(inode);
536 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
537 	if (error == 0)
538 		error = nfs_refresh_inode(inode, fattr);
539 	nfs_free_fattr(fattr);
540 out:
541 	trace_nfs_setattr_exit(inode, error);
542 	return error;
543 }
544 EXPORT_SYMBOL_GPL(nfs_setattr);
545 
546 /**
547  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
548  * @inode: inode of the file used
549  * @offset: file offset to start truncating
550  *
551  * This is a copy of the common vmtruncate, but with the locking
552  * corrected to take into account the fact that NFS requires
553  * inode->i_size to be updated under the inode->i_lock.
554  */
555 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
556 {
557 	int err;
558 
559 	err = inode_newsize_ok(inode, offset);
560 	if (err)
561 		goto out;
562 
563 	spin_lock(&inode->i_lock);
564 	i_size_write(inode, offset);
565 	/* Optimisation */
566 	if (offset == 0)
567 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
568 	spin_unlock(&inode->i_lock);
569 
570 	truncate_pagecache(inode, offset);
571 out:
572 	return err;
573 }
574 
575 /**
576  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
577  * @inode: pointer to struct inode
578  * @attr: pointer to struct iattr
579  *
580  * Note: we do this in the *proc.c in order to ensure that
581  *       it works for things like exclusive creates too.
582  */
583 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
584 {
585 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
586 		spin_lock(&inode->i_lock);
587 		if ((attr->ia_valid & ATTR_MODE) != 0) {
588 			int mode = attr->ia_mode & S_IALLUGO;
589 			mode |= inode->i_mode & ~S_IALLUGO;
590 			inode->i_mode = mode;
591 		}
592 		if ((attr->ia_valid & ATTR_UID) != 0)
593 			inode->i_uid = attr->ia_uid;
594 		if ((attr->ia_valid & ATTR_GID) != 0)
595 			inode->i_gid = attr->ia_gid;
596 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
597 				| NFS_INO_INVALID_ACL);
598 		spin_unlock(&inode->i_lock);
599 	}
600 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
601 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
602 		nfs_vmtruncate(inode, attr->ia_size);
603 	}
604 }
605 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
606 
607 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
608 {
609 	struct dentry *parent;
610 
611 	parent = dget_parent(dentry);
612 	nfs_force_use_readdirplus(parent->d_inode);
613 	dput(parent);
614 }
615 
616 static bool nfs_need_revalidate_inode(struct inode *inode)
617 {
618 	if (NFS_I(inode)->cache_validity &
619 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
620 		return true;
621 	if (nfs_attribute_cache_expired(inode))
622 		return true;
623 	return false;
624 }
625 
626 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
627 {
628 	struct inode *inode = dentry->d_inode;
629 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
630 	int err = 0;
631 
632 	trace_nfs_getattr_enter(inode);
633 	/* Flush out writes to the server in order to update c/mtime.  */
634 	if (S_ISREG(inode->i_mode)) {
635 		nfs_inode_dio_wait(inode);
636 		err = filemap_write_and_wait(inode->i_mapping);
637 		if (err)
638 			goto out;
639 	}
640 
641 	/*
642 	 * We may force a getattr if the user cares about atime.
643 	 *
644 	 * Note that we only have to check the vfsmount flags here:
645 	 *  - NFS always sets S_NOATIME by so checking it would give a
646 	 *    bogus result
647 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
648 	 *    no point in checking those.
649 	 */
650  	if ((mnt->mnt_flags & MNT_NOATIME) ||
651  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
652 		need_atime = 0;
653 
654 	if (need_atime || nfs_need_revalidate_inode(inode)) {
655 		struct nfs_server *server = NFS_SERVER(inode);
656 
657 		if (server->caps & NFS_CAP_READDIRPLUS)
658 			nfs_request_parent_use_readdirplus(dentry);
659 		err = __nfs_revalidate_inode(server, inode);
660 	}
661 	if (!err) {
662 		generic_fillattr(inode, stat);
663 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
664 	}
665 out:
666 	trace_nfs_getattr_exit(inode, err);
667 	return err;
668 }
669 EXPORT_SYMBOL_GPL(nfs_getattr);
670 
671 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
672 {
673 	atomic_set(&l_ctx->count, 1);
674 	l_ctx->lockowner.l_owner = current->files;
675 	l_ctx->lockowner.l_pid = current->tgid;
676 	INIT_LIST_HEAD(&l_ctx->list);
677 	nfs_iocounter_init(&l_ctx->io_count);
678 }
679 
680 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
681 {
682 	struct nfs_lock_context *head = &ctx->lock_context;
683 	struct nfs_lock_context *pos = head;
684 
685 	do {
686 		if (pos->lockowner.l_owner != current->files)
687 			continue;
688 		if (pos->lockowner.l_pid != current->tgid)
689 			continue;
690 		atomic_inc(&pos->count);
691 		return pos;
692 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
693 	return NULL;
694 }
695 
696 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
697 {
698 	struct nfs_lock_context *res, *new = NULL;
699 	struct inode *inode = ctx->dentry->d_inode;
700 
701 	spin_lock(&inode->i_lock);
702 	res = __nfs_find_lock_context(ctx);
703 	if (res == NULL) {
704 		spin_unlock(&inode->i_lock);
705 		new = kmalloc(sizeof(*new), GFP_KERNEL);
706 		if (new == NULL)
707 			return ERR_PTR(-ENOMEM);
708 		nfs_init_lock_context(new);
709 		spin_lock(&inode->i_lock);
710 		res = __nfs_find_lock_context(ctx);
711 		if (res == NULL) {
712 			list_add_tail(&new->list, &ctx->lock_context.list);
713 			new->open_context = ctx;
714 			res = new;
715 			new = NULL;
716 		}
717 	}
718 	spin_unlock(&inode->i_lock);
719 	kfree(new);
720 	return res;
721 }
722 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
723 
724 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
725 {
726 	struct nfs_open_context *ctx = l_ctx->open_context;
727 	struct inode *inode = ctx->dentry->d_inode;
728 
729 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
730 		return;
731 	list_del(&l_ctx->list);
732 	spin_unlock(&inode->i_lock);
733 	kfree(l_ctx);
734 }
735 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
736 
737 /**
738  * nfs_close_context - Common close_context() routine NFSv2/v3
739  * @ctx: pointer to context
740  * @is_sync: is this a synchronous close
741  *
742  * always ensure that the attributes are up to date if we're mounted
743  * with close-to-open semantics
744  */
745 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
746 {
747 	struct inode *inode;
748 	struct nfs_server *server;
749 
750 	if (!(ctx->mode & FMODE_WRITE))
751 		return;
752 	if (!is_sync)
753 		return;
754 	inode = ctx->dentry->d_inode;
755 	if (!list_empty(&NFS_I(inode)->open_files))
756 		return;
757 	server = NFS_SERVER(inode);
758 	if (server->flags & NFS_MOUNT_NOCTO)
759 		return;
760 	nfs_revalidate_inode(server, inode);
761 }
762 EXPORT_SYMBOL_GPL(nfs_close_context);
763 
764 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
765 {
766 	struct nfs_open_context *ctx;
767 	struct rpc_cred *cred = rpc_lookup_cred();
768 	if (IS_ERR(cred))
769 		return ERR_CAST(cred);
770 
771 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
772 	if (!ctx) {
773 		put_rpccred(cred);
774 		return ERR_PTR(-ENOMEM);
775 	}
776 	nfs_sb_active(dentry->d_sb);
777 	ctx->dentry = dget(dentry);
778 	ctx->cred = cred;
779 	ctx->state = NULL;
780 	ctx->mode = f_mode;
781 	ctx->flags = 0;
782 	ctx->error = 0;
783 	nfs_init_lock_context(&ctx->lock_context);
784 	ctx->lock_context.open_context = ctx;
785 	INIT_LIST_HEAD(&ctx->list);
786 	ctx->mdsthreshold = NULL;
787 	return ctx;
788 }
789 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
790 
791 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
792 {
793 	if (ctx != NULL)
794 		atomic_inc(&ctx->lock_context.count);
795 	return ctx;
796 }
797 EXPORT_SYMBOL_GPL(get_nfs_open_context);
798 
799 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
800 {
801 	struct inode *inode = ctx->dentry->d_inode;
802 	struct super_block *sb = ctx->dentry->d_sb;
803 
804 	if (!list_empty(&ctx->list)) {
805 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
806 			return;
807 		list_del(&ctx->list);
808 		spin_unlock(&inode->i_lock);
809 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
810 		return;
811 	if (inode != NULL)
812 		NFS_PROTO(inode)->close_context(ctx, is_sync);
813 	if (ctx->cred != NULL)
814 		put_rpccred(ctx->cred);
815 	dput(ctx->dentry);
816 	nfs_sb_deactive(sb);
817 	kfree(ctx->mdsthreshold);
818 	kfree(ctx);
819 }
820 
821 void put_nfs_open_context(struct nfs_open_context *ctx)
822 {
823 	__put_nfs_open_context(ctx, 0);
824 }
825 EXPORT_SYMBOL_GPL(put_nfs_open_context);
826 
827 /*
828  * Ensure that mmap has a recent RPC credential for use when writing out
829  * shared pages
830  */
831 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
832 {
833 	struct inode *inode = ctx->dentry->d_inode;
834 	struct nfs_inode *nfsi = NFS_I(inode);
835 
836 	spin_lock(&inode->i_lock);
837 	list_add(&ctx->list, &nfsi->open_files);
838 	spin_unlock(&inode->i_lock);
839 }
840 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
841 
842 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
843 {
844 	filp->private_data = get_nfs_open_context(ctx);
845 	if (list_empty(&ctx->list))
846 		nfs_inode_attach_open_context(ctx);
847 }
848 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
849 
850 /*
851  * Given an inode, search for an open context with the desired characteristics
852  */
853 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
854 {
855 	struct nfs_inode *nfsi = NFS_I(inode);
856 	struct nfs_open_context *pos, *ctx = NULL;
857 
858 	spin_lock(&inode->i_lock);
859 	list_for_each_entry(pos, &nfsi->open_files, list) {
860 		if (cred != NULL && pos->cred != cred)
861 			continue;
862 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
863 			continue;
864 		ctx = get_nfs_open_context(pos);
865 		break;
866 	}
867 	spin_unlock(&inode->i_lock);
868 	return ctx;
869 }
870 
871 static void nfs_file_clear_open_context(struct file *filp)
872 {
873 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
874 
875 	if (ctx) {
876 		struct inode *inode = ctx->dentry->d_inode;
877 
878 		filp->private_data = NULL;
879 		spin_lock(&inode->i_lock);
880 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
881 		spin_unlock(&inode->i_lock);
882 		__put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
883 	}
884 }
885 
886 /*
887  * These allocate and release file read/write context information.
888  */
889 int nfs_open(struct inode *inode, struct file *filp)
890 {
891 	struct nfs_open_context *ctx;
892 
893 	ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
894 	if (IS_ERR(ctx))
895 		return PTR_ERR(ctx);
896 	nfs_file_set_open_context(filp, ctx);
897 	put_nfs_open_context(ctx);
898 	nfs_fscache_open_file(inode, filp);
899 	return 0;
900 }
901 
902 int nfs_release(struct inode *inode, struct file *filp)
903 {
904 	nfs_file_clear_open_context(filp);
905 	return 0;
906 }
907 
908 /*
909  * This function is called whenever some part of NFS notices that
910  * the cached attributes have to be refreshed.
911  */
912 int
913 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
914 {
915 	int		 status = -ESTALE;
916 	struct nfs4_label *label = NULL;
917 	struct nfs_fattr *fattr = NULL;
918 	struct nfs_inode *nfsi = NFS_I(inode);
919 
920 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
921 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
922 
923 	trace_nfs_revalidate_inode_enter(inode);
924 
925 	if (is_bad_inode(inode))
926 		goto out;
927 	if (NFS_STALE(inode))
928 		goto out;
929 
930 	status = -ENOMEM;
931 	fattr = nfs_alloc_fattr();
932 	if (fattr == NULL)
933 		goto out;
934 
935 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
936 
937 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
938 	if (IS_ERR(label)) {
939 		status = PTR_ERR(label);
940 		goto out;
941 	}
942 
943 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
944 	if (status != 0) {
945 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
946 			 inode->i_sb->s_id,
947 			 (unsigned long long)NFS_FILEID(inode), status);
948 		if (status == -ESTALE) {
949 			nfs_zap_caches(inode);
950 			if (!S_ISDIR(inode->i_mode))
951 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
952 		}
953 		goto err_out;
954 	}
955 
956 	status = nfs_refresh_inode(inode, fattr);
957 	if (status) {
958 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
959 			 inode->i_sb->s_id,
960 			 (unsigned long long)NFS_FILEID(inode), status);
961 		goto err_out;
962 	}
963 
964 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
965 		nfs_zap_acl_cache(inode);
966 
967 	nfs_setsecurity(inode, fattr, label);
968 
969 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
970 		inode->i_sb->s_id,
971 		(unsigned long long)NFS_FILEID(inode));
972 
973 err_out:
974 	nfs4_label_free(label);
975 out:
976 	nfs_free_fattr(fattr);
977 	trace_nfs_revalidate_inode_exit(inode, status);
978 	return status;
979 }
980 
981 int nfs_attribute_timeout(struct inode *inode)
982 {
983 	struct nfs_inode *nfsi = NFS_I(inode);
984 
985 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
986 }
987 
988 int nfs_attribute_cache_expired(struct inode *inode)
989 {
990 	if (nfs_have_delegated_attributes(inode))
991 		return 0;
992 	return nfs_attribute_timeout(inode);
993 }
994 
995 /**
996  * nfs_revalidate_inode - Revalidate the inode attributes
997  * @server - pointer to nfs_server struct
998  * @inode - pointer to inode struct
999  *
1000  * Updates inode attribute information by retrieving the data from the server.
1001  */
1002 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1003 {
1004 	if (!nfs_need_revalidate_inode(inode))
1005 		return NFS_STALE(inode) ? -ESTALE : 0;
1006 	return __nfs_revalidate_inode(server, inode);
1007 }
1008 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1009 
1010 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1011 {
1012 	if (!(NFS_I(inode)->cache_validity &
1013 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1014 			&& !nfs_attribute_cache_expired(inode))
1015 		return NFS_STALE(inode) ? -ESTALE : 0;
1016 	return -ECHILD;
1017 }
1018 
1019 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1020 {
1021 	struct nfs_inode *nfsi = NFS_I(inode);
1022 	int ret;
1023 
1024 	if (mapping->nrpages != 0) {
1025 		if (S_ISREG(inode->i_mode)) {
1026 			ret = nfs_sync_mapping(mapping);
1027 			if (ret < 0)
1028 				return ret;
1029 		}
1030 		ret = invalidate_inode_pages2(mapping);
1031 		if (ret < 0)
1032 			return ret;
1033 	}
1034 	if (S_ISDIR(inode->i_mode)) {
1035 		spin_lock(&inode->i_lock);
1036 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1037 		spin_unlock(&inode->i_lock);
1038 	}
1039 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1040 	nfs_fscache_wait_on_invalidate(inode);
1041 
1042 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1043 			inode->i_sb->s_id,
1044 			(unsigned long long)NFS_FILEID(inode));
1045 	return 0;
1046 }
1047 
1048 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1049 {
1050 	if (nfs_have_delegated_attributes(inode))
1051 		return false;
1052 	return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1053 		|| nfs_attribute_timeout(inode)
1054 		|| NFS_STALE(inode);
1055 }
1056 
1057 /**
1058  * nfs_revalidate_mapping - Revalidate the pagecache
1059  * @inode - pointer to host inode
1060  * @mapping - pointer to mapping
1061  */
1062 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1063 {
1064 	struct nfs_inode *nfsi = NFS_I(inode);
1065 	unsigned long *bitlock = &nfsi->flags;
1066 	int ret = 0;
1067 
1068 	/* swapfiles are not supposed to be shared. */
1069 	if (IS_SWAPFILE(inode))
1070 		goto out;
1071 
1072 	if (nfs_mapping_need_revalidate_inode(inode)) {
1073 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1074 		if (ret < 0)
1075 			goto out;
1076 	}
1077 
1078 	/*
1079 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1080 	 * invalidations that come in while we're shooting down the mappings
1081 	 * are respected. But, that leaves a race window where one revalidator
1082 	 * can clear the flag, and then another checks it before the mapping
1083 	 * gets invalidated. Fix that by serializing access to this part of
1084 	 * the function.
1085 	 *
1086 	 * At the same time, we need to allow other tasks to see whether we
1087 	 * might be in the middle of invalidating the pages, so we only set
1088 	 * the bit lock here if it looks like we're going to be doing that.
1089 	 */
1090 	for (;;) {
1091 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1092 					 nfs_wait_bit_killable, TASK_KILLABLE);
1093 		if (ret)
1094 			goto out;
1095 		spin_lock(&inode->i_lock);
1096 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1097 			spin_unlock(&inode->i_lock);
1098 			continue;
1099 		}
1100 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1101 			break;
1102 		spin_unlock(&inode->i_lock);
1103 		goto out;
1104 	}
1105 
1106 	set_bit(NFS_INO_INVALIDATING, bitlock);
1107 	smp_wmb();
1108 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1109 	spin_unlock(&inode->i_lock);
1110 	trace_nfs_invalidate_mapping_enter(inode);
1111 	ret = nfs_invalidate_mapping(inode, mapping);
1112 	trace_nfs_invalidate_mapping_exit(inode, ret);
1113 
1114 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1115 	smp_mb__after_atomic();
1116 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1117 out:
1118 	return ret;
1119 }
1120 
1121 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1122 {
1123 	struct nfs_inode *nfsi = NFS_I(inode);
1124 	unsigned long ret = 0;
1125 
1126 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1127 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1128 			&& inode->i_version == fattr->pre_change_attr) {
1129 		inode->i_version = fattr->change_attr;
1130 		if (S_ISDIR(inode->i_mode))
1131 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1132 		ret |= NFS_INO_INVALID_ATTR;
1133 	}
1134 	/* If we have atomic WCC data, we may update some attributes */
1135 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1136 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1137 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1138 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1139 		ret |= NFS_INO_INVALID_ATTR;
1140 	}
1141 
1142 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1143 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1144 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1145 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1146 		if (S_ISDIR(inode->i_mode))
1147 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1148 		ret |= NFS_INO_INVALID_ATTR;
1149 	}
1150 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1151 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1152 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1153 			&& nfsi->nrequests == 0) {
1154 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1155 		ret |= NFS_INO_INVALID_ATTR;
1156 	}
1157 
1158 	return ret;
1159 }
1160 
1161 /**
1162  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1163  * @inode - pointer to inode
1164  * @fattr - updated attributes
1165  *
1166  * Verifies the attribute cache. If we have just changed the attributes,
1167  * so that fattr carries weak cache consistency data, then it may
1168  * also update the ctime/mtime/change_attribute.
1169  */
1170 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1171 {
1172 	struct nfs_inode *nfsi = NFS_I(inode);
1173 	loff_t cur_size, new_isize;
1174 	unsigned long invalid = 0;
1175 
1176 
1177 	if (nfs_have_delegated_attributes(inode))
1178 		return 0;
1179 	/* Has the inode gone and changed behind our back? */
1180 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1181 		return -EIO;
1182 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1183 		return -EIO;
1184 
1185 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1186 			inode->i_version != fattr->change_attr)
1187 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1188 
1189 	/* Verify a few of the more important attributes */
1190 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1191 		invalid |= NFS_INO_INVALID_ATTR;
1192 
1193 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1194 		cur_size = i_size_read(inode);
1195 		new_isize = nfs_size_to_loff_t(fattr->size);
1196 		if (cur_size != new_isize && nfsi->nrequests == 0)
1197 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1198 	}
1199 
1200 	/* Have any file permissions changed? */
1201 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1202 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1203 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1204 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1205 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1206 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1207 
1208 	/* Has the link count changed? */
1209 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1210 		invalid |= NFS_INO_INVALID_ATTR;
1211 
1212 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1213 		invalid |= NFS_INO_INVALID_ATIME;
1214 
1215 	if (invalid != 0)
1216 		nfs_set_cache_invalid(inode, invalid);
1217 
1218 	nfsi->read_cache_jiffies = fattr->time_start;
1219 	return 0;
1220 }
1221 
1222 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1223 {
1224 	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1225 		return 0;
1226 	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1227 }
1228 
1229 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1230 {
1231 	if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1232 		return 0;
1233 	return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1234 }
1235 
1236 static atomic_long_t nfs_attr_generation_counter;
1237 
1238 static unsigned long nfs_read_attr_generation_counter(void)
1239 {
1240 	return atomic_long_read(&nfs_attr_generation_counter);
1241 }
1242 
1243 unsigned long nfs_inc_attr_generation_counter(void)
1244 {
1245 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1246 }
1247 
1248 void nfs_fattr_init(struct nfs_fattr *fattr)
1249 {
1250 	fattr->valid = 0;
1251 	fattr->time_start = jiffies;
1252 	fattr->gencount = nfs_inc_attr_generation_counter();
1253 	fattr->owner_name = NULL;
1254 	fattr->group_name = NULL;
1255 }
1256 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1257 
1258 struct nfs_fattr *nfs_alloc_fattr(void)
1259 {
1260 	struct nfs_fattr *fattr;
1261 
1262 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1263 	if (fattr != NULL)
1264 		nfs_fattr_init(fattr);
1265 	return fattr;
1266 }
1267 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1268 
1269 struct nfs_fh *nfs_alloc_fhandle(void)
1270 {
1271 	struct nfs_fh *fh;
1272 
1273 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1274 	if (fh != NULL)
1275 		fh->size = 0;
1276 	return fh;
1277 }
1278 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1279 
1280 #ifdef NFS_DEBUG
1281 /*
1282  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1283  *                             in the same way that wireshark does
1284  *
1285  * @fh: file handle
1286  *
1287  * For debugging only.
1288  */
1289 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1290 {
1291 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1292 	 * not on the result */
1293 	return nfs_fhandle_hash(fh);
1294 }
1295 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1296 
1297 /*
1298  * _nfs_display_fhandle - display an NFS file handle on the console
1299  *
1300  * @fh: file handle to display
1301  * @caption: display caption
1302  *
1303  * For debugging only.
1304  */
1305 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1306 {
1307 	unsigned short i;
1308 
1309 	if (fh == NULL || fh->size == 0) {
1310 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1311 		return;
1312 	}
1313 
1314 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1315 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1316 	for (i = 0; i < fh->size; i += 16) {
1317 		__be32 *pos = (__be32 *)&fh->data[i];
1318 
1319 		switch ((fh->size - i - 1) >> 2) {
1320 		case 0:
1321 			printk(KERN_DEFAULT " %08x\n",
1322 				be32_to_cpup(pos));
1323 			break;
1324 		case 1:
1325 			printk(KERN_DEFAULT " %08x %08x\n",
1326 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1327 			break;
1328 		case 2:
1329 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1330 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1331 				be32_to_cpup(pos + 2));
1332 			break;
1333 		default:
1334 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1335 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1336 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1337 		}
1338 	}
1339 }
1340 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1341 #endif
1342 
1343 /**
1344  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1345  * @inode - pointer to inode
1346  * @fattr - attributes
1347  *
1348  * Attempt to divine whether or not an RPC call reply carrying stale
1349  * attributes got scheduled after another call carrying updated ones.
1350  *
1351  * To do so, the function first assumes that a more recent ctime means
1352  * that the attributes in fattr are newer, however it also attempt to
1353  * catch the case where ctime either didn't change, or went backwards
1354  * (if someone reset the clock on the server) by looking at whether
1355  * or not this RPC call was started after the inode was last updated.
1356  * Note also the check for wraparound of 'attr_gencount'
1357  *
1358  * The function returns 'true' if it thinks the attributes in 'fattr' are
1359  * more recent than the ones cached in the inode.
1360  *
1361  */
1362 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1363 {
1364 	const struct nfs_inode *nfsi = NFS_I(inode);
1365 
1366 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1367 		nfs_ctime_need_update(inode, fattr) ||
1368 		nfs_size_need_update(inode, fattr) ||
1369 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1370 }
1371 
1372 /*
1373  * Don't trust the change_attribute, mtime, ctime or size if
1374  * a pnfs LAYOUTCOMMIT is outstanding
1375  */
1376 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1377 		struct nfs_fattr *fattr)
1378 {
1379 	if (pnfs_layoutcommit_outstanding(inode))
1380 		fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1381 				NFS_ATTR_FATTR_MTIME |
1382 				NFS_ATTR_FATTR_CTIME |
1383 				NFS_ATTR_FATTR_SIZE);
1384 }
1385 
1386 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1387 {
1388 	int ret;
1389 
1390 	trace_nfs_refresh_inode_enter(inode);
1391 
1392 	nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1393 
1394 	if (nfs_inode_attrs_need_update(inode, fattr))
1395 		ret = nfs_update_inode(inode, fattr);
1396 	else
1397 		ret = nfs_check_inode_attributes(inode, fattr);
1398 
1399 	trace_nfs_refresh_inode_exit(inode, ret);
1400 	return ret;
1401 }
1402 
1403 /**
1404  * nfs_refresh_inode - try to update the inode attribute cache
1405  * @inode - pointer to inode
1406  * @fattr - updated attributes
1407  *
1408  * Check that an RPC call that returned attributes has not overlapped with
1409  * other recent updates of the inode metadata, then decide whether it is
1410  * safe to do a full update of the inode attributes, or whether just to
1411  * call nfs_check_inode_attributes.
1412  */
1413 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1414 {
1415 	int status;
1416 
1417 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1418 		return 0;
1419 	spin_lock(&inode->i_lock);
1420 	status = nfs_refresh_inode_locked(inode, fattr);
1421 	spin_unlock(&inode->i_lock);
1422 
1423 	return status;
1424 }
1425 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1426 
1427 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1428 {
1429 	unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1430 
1431 	if (S_ISDIR(inode->i_mode))
1432 		invalid |= NFS_INO_INVALID_DATA;
1433 	nfs_set_cache_invalid(inode, invalid);
1434 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1435 		return 0;
1436 	return nfs_refresh_inode_locked(inode, fattr);
1437 }
1438 
1439 /**
1440  * nfs_post_op_update_inode - try to update the inode attribute cache
1441  * @inode - pointer to inode
1442  * @fattr - updated attributes
1443  *
1444  * After an operation that has changed the inode metadata, mark the
1445  * attribute cache as being invalid, then try to update it.
1446  *
1447  * NB: if the server didn't return any post op attributes, this
1448  * function will force the retrieval of attributes before the next
1449  * NFS request.  Thus it should be used only for operations that
1450  * are expected to change one or more attributes, to avoid
1451  * unnecessary NFS requests and trips through nfs_update_inode().
1452  */
1453 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1454 {
1455 	int status;
1456 
1457 	spin_lock(&inode->i_lock);
1458 	status = nfs_post_op_update_inode_locked(inode, fattr);
1459 	spin_unlock(&inode->i_lock);
1460 
1461 	return status;
1462 }
1463 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1464 
1465 /**
1466  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1467  * @inode - pointer to inode
1468  * @fattr - updated attributes
1469  *
1470  * After an operation that has changed the inode metadata, mark the
1471  * attribute cache as being invalid, then try to update it. Fake up
1472  * weak cache consistency data, if none exist.
1473  *
1474  * This function is mainly designed to be used by the ->write_done() functions.
1475  */
1476 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1477 {
1478 	int status;
1479 
1480 	spin_lock(&inode->i_lock);
1481 	/* Don't do a WCC update if these attributes are already stale */
1482 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1483 			!nfs_inode_attrs_need_update(inode, fattr)) {
1484 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1485 				| NFS_ATTR_FATTR_PRESIZE
1486 				| NFS_ATTR_FATTR_PREMTIME
1487 				| NFS_ATTR_FATTR_PRECTIME);
1488 		goto out_noforce;
1489 	}
1490 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1491 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1492 		fattr->pre_change_attr = inode->i_version;
1493 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1494 	}
1495 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1496 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1497 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1498 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1499 	}
1500 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1501 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1502 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1503 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1504 	}
1505 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1506 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1507 		fattr->pre_size = i_size_read(inode);
1508 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1509 	}
1510 out_noforce:
1511 	status = nfs_post_op_update_inode_locked(inode, fattr);
1512 	spin_unlock(&inode->i_lock);
1513 	return status;
1514 }
1515 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1516 
1517 /*
1518  * Many nfs protocol calls return the new file attributes after
1519  * an operation.  Here we update the inode to reflect the state
1520  * of the server's inode.
1521  *
1522  * This is a bit tricky because we have to make sure all dirty pages
1523  * have been sent off to the server before calling invalidate_inode_pages.
1524  * To make sure no other process adds more write requests while we try
1525  * our best to flush them, we make them sleep during the attribute refresh.
1526  *
1527  * A very similar scenario holds for the dir cache.
1528  */
1529 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1530 {
1531 	struct nfs_server *server;
1532 	struct nfs_inode *nfsi = NFS_I(inode);
1533 	loff_t cur_isize, new_isize;
1534 	unsigned long invalid = 0;
1535 	unsigned long now = jiffies;
1536 	unsigned long save_cache_validity;
1537 
1538 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1539 			__func__, inode->i_sb->s_id, inode->i_ino,
1540 			nfs_display_fhandle_hash(NFS_FH(inode)),
1541 			atomic_read(&inode->i_count), fattr->valid);
1542 
1543 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1544 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1545 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1546 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1547 			inode->i_sb->s_id, (long long)nfsi->fileid,
1548 			(long long)fattr->fileid);
1549 		goto out_err;
1550 	}
1551 
1552 	/*
1553 	 * Make sure the inode's type hasn't changed.
1554 	 */
1555 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1556 		/*
1557 		* Big trouble! The inode has become a different object.
1558 		*/
1559 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1560 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1561 		goto out_err;
1562 	}
1563 
1564 	server = NFS_SERVER(inode);
1565 	/* Update the fsid? */
1566 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1567 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1568 			!IS_AUTOMOUNT(inode))
1569 		server->fsid = fattr->fsid;
1570 
1571 	/*
1572 	 * Update the read time so we don't revalidate too often.
1573 	 */
1574 	nfsi->read_cache_jiffies = fattr->time_start;
1575 
1576 	save_cache_validity = nfsi->cache_validity;
1577 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1578 			| NFS_INO_INVALID_ATIME
1579 			| NFS_INO_REVAL_FORCED
1580 			| NFS_INO_REVAL_PAGECACHE);
1581 
1582 	/* Do atomic weak cache consistency updates */
1583 	invalid |= nfs_wcc_update_inode(inode, fattr);
1584 
1585 	/* More cache consistency checks */
1586 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1587 		if (inode->i_version != fattr->change_attr) {
1588 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1589 					inode->i_sb->s_id, inode->i_ino);
1590 			invalid |= NFS_INO_INVALID_ATTR
1591 				| NFS_INO_INVALID_DATA
1592 				| NFS_INO_INVALID_ACCESS
1593 				| NFS_INO_INVALID_ACL
1594 				| NFS_INO_REVAL_PAGECACHE;
1595 			if (S_ISDIR(inode->i_mode))
1596 				nfs_force_lookup_revalidate(inode);
1597 			inode->i_version = fattr->change_attr;
1598 		}
1599 	} else if (server->caps & NFS_CAP_CHANGE_ATTR)
1600 		nfsi->cache_validity |= save_cache_validity;
1601 
1602 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1603 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1604 	} else if (server->caps & NFS_CAP_MTIME)
1605 		nfsi->cache_validity |= save_cache_validity &
1606 				(NFS_INO_INVALID_ATTR
1607 				| NFS_INO_REVAL_FORCED);
1608 
1609 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1610 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1611 	} else if (server->caps & NFS_CAP_CTIME)
1612 		nfsi->cache_validity |= save_cache_validity &
1613 				(NFS_INO_INVALID_ATTR
1614 				| NFS_INO_REVAL_FORCED);
1615 
1616 	/* Check if our cached file size is stale */
1617 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1618 		new_isize = nfs_size_to_loff_t(fattr->size);
1619 		cur_isize = i_size_read(inode);
1620 		if (new_isize != cur_isize) {
1621 			/* Do we perhaps have any outstanding writes, or has
1622 			 * the file grown beyond our last write? */
1623 			if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1624 				i_size_write(inode, new_isize);
1625 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1626 				invalid &= ~NFS_INO_REVAL_PAGECACHE;
1627 			}
1628 			dprintk("NFS: isize change on server for file %s/%ld "
1629 					"(%Ld to %Ld)\n",
1630 					inode->i_sb->s_id,
1631 					inode->i_ino,
1632 					(long long)cur_isize,
1633 					(long long)new_isize);
1634 		}
1635 	} else
1636 		nfsi->cache_validity |= save_cache_validity &
1637 				(NFS_INO_INVALID_ATTR
1638 				| NFS_INO_REVAL_PAGECACHE
1639 				| NFS_INO_REVAL_FORCED);
1640 
1641 
1642 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1643 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1644 	else if (server->caps & NFS_CAP_ATIME)
1645 		nfsi->cache_validity |= save_cache_validity &
1646 				(NFS_INO_INVALID_ATIME
1647 				| NFS_INO_REVAL_FORCED);
1648 
1649 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1650 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1651 			umode_t newmode = inode->i_mode & S_IFMT;
1652 			newmode |= fattr->mode & S_IALLUGO;
1653 			inode->i_mode = newmode;
1654 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1655 		}
1656 	} else if (server->caps & NFS_CAP_MODE)
1657 		nfsi->cache_validity |= save_cache_validity &
1658 				(NFS_INO_INVALID_ATTR
1659 				| NFS_INO_INVALID_ACCESS
1660 				| NFS_INO_INVALID_ACL
1661 				| NFS_INO_REVAL_FORCED);
1662 
1663 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1664 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1665 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1666 			inode->i_uid = fattr->uid;
1667 		}
1668 	} else if (server->caps & NFS_CAP_OWNER)
1669 		nfsi->cache_validity |= save_cache_validity &
1670 				(NFS_INO_INVALID_ATTR
1671 				| NFS_INO_INVALID_ACCESS
1672 				| NFS_INO_INVALID_ACL
1673 				| NFS_INO_REVAL_FORCED);
1674 
1675 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1676 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1677 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1678 			inode->i_gid = fattr->gid;
1679 		}
1680 	} else if (server->caps & NFS_CAP_OWNER_GROUP)
1681 		nfsi->cache_validity |= save_cache_validity &
1682 				(NFS_INO_INVALID_ATTR
1683 				| NFS_INO_INVALID_ACCESS
1684 				| NFS_INO_INVALID_ACL
1685 				| NFS_INO_REVAL_FORCED);
1686 
1687 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1688 		if (inode->i_nlink != fattr->nlink) {
1689 			invalid |= NFS_INO_INVALID_ATTR;
1690 			if (S_ISDIR(inode->i_mode))
1691 				invalid |= NFS_INO_INVALID_DATA;
1692 			set_nlink(inode, fattr->nlink);
1693 		}
1694 	} else if (server->caps & NFS_CAP_NLINK)
1695 		nfsi->cache_validity |= save_cache_validity &
1696 				(NFS_INO_INVALID_ATTR
1697 				| NFS_INO_REVAL_FORCED);
1698 
1699 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1700 		/*
1701 		 * report the blocks in 512byte units
1702 		 */
1703 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1704  	}
1705 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1706 		inode->i_blocks = fattr->du.nfs2.blocks;
1707 
1708 	/* Update attrtimeo value if we're out of the unstable period */
1709 	if (invalid & NFS_INO_INVALID_ATTR) {
1710 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1711 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1712 		nfsi->attrtimeo_timestamp = now;
1713 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1714 	} else {
1715 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1716 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1717 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1718 			nfsi->attrtimeo_timestamp = now;
1719 		}
1720 	}
1721 	invalid &= ~NFS_INO_INVALID_ATTR;
1722 	/* Don't invalidate the data if we were to blame */
1723 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1724 				|| S_ISLNK(inode->i_mode)))
1725 		invalid &= ~NFS_INO_INVALID_DATA;
1726 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1727 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1728 		nfs_set_cache_invalid(inode, invalid);
1729 
1730 	return 0;
1731  out_err:
1732 	/*
1733 	 * No need to worry about unhashing the dentry, as the
1734 	 * lookup validation will know that the inode is bad.
1735 	 * (But we fall through to invalidate the caches.)
1736 	 */
1737 	nfs_invalidate_inode(inode);
1738 	return -ESTALE;
1739 }
1740 
1741 struct inode *nfs_alloc_inode(struct super_block *sb)
1742 {
1743 	struct nfs_inode *nfsi;
1744 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1745 	if (!nfsi)
1746 		return NULL;
1747 	nfsi->flags = 0UL;
1748 	nfsi->cache_validity = 0UL;
1749 #if IS_ENABLED(CONFIG_NFS_V4)
1750 	nfsi->nfs4_acl = NULL;
1751 #endif /* CONFIG_NFS_V4 */
1752 	return &nfsi->vfs_inode;
1753 }
1754 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1755 
1756 static void nfs_i_callback(struct rcu_head *head)
1757 {
1758 	struct inode *inode = container_of(head, struct inode, i_rcu);
1759 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1760 }
1761 
1762 void nfs_destroy_inode(struct inode *inode)
1763 {
1764 	call_rcu(&inode->i_rcu, nfs_i_callback);
1765 }
1766 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1767 
1768 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1769 {
1770 #if IS_ENABLED(CONFIG_NFS_V4)
1771 	INIT_LIST_HEAD(&nfsi->open_states);
1772 	nfsi->delegation = NULL;
1773 	nfsi->delegation_state = 0;
1774 	init_rwsem(&nfsi->rwsem);
1775 	nfsi->layout = NULL;
1776 #endif
1777 }
1778 
1779 static void init_once(void *foo)
1780 {
1781 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1782 
1783 	inode_init_once(&nfsi->vfs_inode);
1784 	INIT_LIST_HEAD(&nfsi->open_files);
1785 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1786 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1787 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1788 	nfsi->nrequests = 0;
1789 	nfsi->commit_info.ncommit = 0;
1790 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1791 	atomic_set(&nfsi->silly_count, 1);
1792 	INIT_HLIST_HEAD(&nfsi->silly_list);
1793 	init_waitqueue_head(&nfsi->waitqueue);
1794 	nfs4_init_once(nfsi);
1795 }
1796 
1797 static int __init nfs_init_inodecache(void)
1798 {
1799 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1800 					     sizeof(struct nfs_inode),
1801 					     0, (SLAB_RECLAIM_ACCOUNT|
1802 						SLAB_MEM_SPREAD),
1803 					     init_once);
1804 	if (nfs_inode_cachep == NULL)
1805 		return -ENOMEM;
1806 
1807 	return 0;
1808 }
1809 
1810 static void nfs_destroy_inodecache(void)
1811 {
1812 	/*
1813 	 * Make sure all delayed rcu free inodes are flushed before we
1814 	 * destroy cache.
1815 	 */
1816 	rcu_barrier();
1817 	kmem_cache_destroy(nfs_inode_cachep);
1818 }
1819 
1820 struct workqueue_struct *nfsiod_workqueue;
1821 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1822 
1823 /*
1824  * start up the nfsiod workqueue
1825  */
1826 static int nfsiod_start(void)
1827 {
1828 	struct workqueue_struct *wq;
1829 	dprintk("RPC:       creating workqueue nfsiod\n");
1830 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1831 	if (wq == NULL)
1832 		return -ENOMEM;
1833 	nfsiod_workqueue = wq;
1834 	return 0;
1835 }
1836 
1837 /*
1838  * Destroy the nfsiod workqueue
1839  */
1840 static void nfsiod_stop(void)
1841 {
1842 	struct workqueue_struct *wq;
1843 
1844 	wq = nfsiod_workqueue;
1845 	if (wq == NULL)
1846 		return;
1847 	nfsiod_workqueue = NULL;
1848 	destroy_workqueue(wq);
1849 }
1850 
1851 int nfs_net_id;
1852 EXPORT_SYMBOL_GPL(nfs_net_id);
1853 
1854 static int nfs_net_init(struct net *net)
1855 {
1856 	nfs_clients_init(net);
1857 	return nfs_fs_proc_net_init(net);
1858 }
1859 
1860 static void nfs_net_exit(struct net *net)
1861 {
1862 	nfs_fs_proc_net_exit(net);
1863 	nfs_cleanup_cb_ident_idr(net);
1864 }
1865 
1866 static struct pernet_operations nfs_net_ops = {
1867 	.init = nfs_net_init,
1868 	.exit = nfs_net_exit,
1869 	.id   = &nfs_net_id,
1870 	.size = sizeof(struct nfs_net),
1871 };
1872 
1873 /*
1874  * Initialize NFS
1875  */
1876 static int __init init_nfs_fs(void)
1877 {
1878 	int err;
1879 
1880 	err = register_pernet_subsys(&nfs_net_ops);
1881 	if (err < 0)
1882 		goto out9;
1883 
1884 	err = nfs_fscache_register();
1885 	if (err < 0)
1886 		goto out8;
1887 
1888 	err = nfsiod_start();
1889 	if (err)
1890 		goto out7;
1891 
1892 	err = nfs_fs_proc_init();
1893 	if (err)
1894 		goto out6;
1895 
1896 	err = nfs_init_nfspagecache();
1897 	if (err)
1898 		goto out5;
1899 
1900 	err = nfs_init_inodecache();
1901 	if (err)
1902 		goto out4;
1903 
1904 	err = nfs_init_readpagecache();
1905 	if (err)
1906 		goto out3;
1907 
1908 	err = nfs_init_writepagecache();
1909 	if (err)
1910 		goto out2;
1911 
1912 	err = nfs_init_directcache();
1913 	if (err)
1914 		goto out1;
1915 
1916 #ifdef CONFIG_PROC_FS
1917 	rpc_proc_register(&init_net, &nfs_rpcstat);
1918 #endif
1919 	if ((err = register_nfs_fs()) != 0)
1920 		goto out0;
1921 
1922 	return 0;
1923 out0:
1924 #ifdef CONFIG_PROC_FS
1925 	rpc_proc_unregister(&init_net, "nfs");
1926 #endif
1927 	nfs_destroy_directcache();
1928 out1:
1929 	nfs_destroy_writepagecache();
1930 out2:
1931 	nfs_destroy_readpagecache();
1932 out3:
1933 	nfs_destroy_inodecache();
1934 out4:
1935 	nfs_destroy_nfspagecache();
1936 out5:
1937 	nfs_fs_proc_exit();
1938 out6:
1939 	nfsiod_stop();
1940 out7:
1941 	nfs_fscache_unregister();
1942 out8:
1943 	unregister_pernet_subsys(&nfs_net_ops);
1944 out9:
1945 	return err;
1946 }
1947 
1948 static void __exit exit_nfs_fs(void)
1949 {
1950 	nfs_destroy_directcache();
1951 	nfs_destroy_writepagecache();
1952 	nfs_destroy_readpagecache();
1953 	nfs_destroy_inodecache();
1954 	nfs_destroy_nfspagecache();
1955 	nfs_fscache_unregister();
1956 	unregister_pernet_subsys(&nfs_net_ops);
1957 #ifdef CONFIG_PROC_FS
1958 	rpc_proc_unregister(&init_net, "nfs");
1959 #endif
1960 	unregister_nfs_fs();
1961 	nfs_fs_proc_exit();
1962 	nfsiod_stop();
1963 }
1964 
1965 /* Not quite true; I just maintain it */
1966 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1967 MODULE_LICENSE("GPL");
1968 module_param(enable_ino64, bool, 0644);
1969 
1970 module_init(init_nfs_fs)
1971 module_exit(exit_nfs_fs)
1972