xref: /openbmc/linux/fs/nfs/inode.c (revision e6c81cce)
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 (nfs_attr_use_mounted_on_fileid(fattr))
356 		fattr->fileid = fattr->mounted_on_fileid;
357 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
358 		goto out_no_inode;
359 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
360 		goto out_no_inode;
361 
362 	hash = nfs_fattr_to_ino_t(fattr);
363 
364 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
365 	if (inode == NULL) {
366 		inode = ERR_PTR(-ENOMEM);
367 		goto out_no_inode;
368 	}
369 
370 	if (inode->i_state & I_NEW) {
371 		struct nfs_inode *nfsi = NFS_I(inode);
372 		unsigned long now = jiffies;
373 
374 		/* We set i_ino for the few things that still rely on it,
375 		 * such as stat(2) */
376 		inode->i_ino = hash;
377 
378 		/* We can't support update_atime(), since the server will reset it */
379 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
380 		inode->i_mode = fattr->mode;
381 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
382 				&& nfs_server_capable(inode, NFS_CAP_MODE))
383 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
384 		/* Why so? Because we want revalidate for devices/FIFOs, and
385 		 * that's precisely what we have in nfs_file_inode_operations.
386 		 */
387 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
388 		if (S_ISREG(inode->i_mode)) {
389 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
390 			inode->i_data.a_ops = &nfs_file_aops;
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 		loff_t i_size;
510 
511 		BUG_ON(!S_ISREG(inode->i_mode));
512 
513 		i_size = i_size_read(inode);
514 		if (attr->ia_size == i_size)
515 			attr->ia_valid &= ~ATTR_SIZE;
516 		else if (attr->ia_size < i_size && IS_SWAPFILE(inode))
517 			return -ETXTBSY;
518 	}
519 
520 	/* Optimization: if the end result is no change, don't RPC */
521 	attr->ia_valid &= NFS_VALID_ATTRS;
522 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
523 		return 0;
524 
525 	trace_nfs_setattr_enter(inode);
526 
527 	/* Write all dirty data */
528 	if (S_ISREG(inode->i_mode)) {
529 		nfs_inode_dio_wait(inode);
530 		nfs_wb_all(inode);
531 	}
532 
533 	fattr = nfs_alloc_fattr();
534 	if (fattr == NULL)
535 		goto out;
536 	/*
537 	 * Return any delegations if we're going to change ACLs
538 	 */
539 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
540 		NFS_PROTO(inode)->return_delegation(inode);
541 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
542 	if (error == 0)
543 		error = nfs_refresh_inode(inode, fattr);
544 	nfs_free_fattr(fattr);
545 out:
546 	trace_nfs_setattr_exit(inode, error);
547 	return error;
548 }
549 EXPORT_SYMBOL_GPL(nfs_setattr);
550 
551 /**
552  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
553  * @inode: inode of the file used
554  * @offset: file offset to start truncating
555  *
556  * This is a copy of the common vmtruncate, but with the locking
557  * corrected to take into account the fact that NFS requires
558  * inode->i_size to be updated under the inode->i_lock.
559  * Note: must be called with inode->i_lock held!
560  */
561 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
562 {
563 	int err;
564 
565 	err = inode_newsize_ok(inode, offset);
566 	if (err)
567 		goto out;
568 
569 	i_size_write(inode, offset);
570 	/* Optimisation */
571 	if (offset == 0)
572 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
573 
574 	spin_unlock(&inode->i_lock);
575 	truncate_pagecache(inode, offset);
576 	spin_lock(&inode->i_lock);
577 out:
578 	return err;
579 }
580 
581 /**
582  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
583  * @inode: pointer to struct inode
584  * @attr: pointer to struct iattr
585  *
586  * Note: we do this in the *proc.c in order to ensure that
587  *       it works for things like exclusive creates too.
588  */
589 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
590 		struct nfs_fattr *fattr)
591 {
592 	/* Barrier: bump the attribute generation count. */
593 	nfs_fattr_set_barrier(fattr);
594 
595 	spin_lock(&inode->i_lock);
596 	NFS_I(inode)->attr_gencount = fattr->gencount;
597 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
598 		if ((attr->ia_valid & ATTR_MODE) != 0) {
599 			int mode = attr->ia_mode & S_IALLUGO;
600 			mode |= inode->i_mode & ~S_IALLUGO;
601 			inode->i_mode = mode;
602 		}
603 		if ((attr->ia_valid & ATTR_UID) != 0)
604 			inode->i_uid = attr->ia_uid;
605 		if ((attr->ia_valid & ATTR_GID) != 0)
606 			inode->i_gid = attr->ia_gid;
607 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
608 				| NFS_INO_INVALID_ACL);
609 	}
610 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
611 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
612 		nfs_vmtruncate(inode, attr->ia_size);
613 	}
614 	nfs_update_inode(inode, fattr);
615 	spin_unlock(&inode->i_lock);
616 }
617 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
618 
619 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
620 {
621 	struct dentry *parent;
622 
623 	parent = dget_parent(dentry);
624 	nfs_force_use_readdirplus(parent->d_inode);
625 	dput(parent);
626 }
627 
628 static bool nfs_need_revalidate_inode(struct inode *inode)
629 {
630 	if (NFS_I(inode)->cache_validity &
631 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
632 		return true;
633 	if (nfs_attribute_cache_expired(inode))
634 		return true;
635 	return false;
636 }
637 
638 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
639 {
640 	struct inode *inode = dentry->d_inode;
641 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
642 	int err = 0;
643 
644 	trace_nfs_getattr_enter(inode);
645 	/* Flush out writes to the server in order to update c/mtime.  */
646 	if (S_ISREG(inode->i_mode)) {
647 		nfs_inode_dio_wait(inode);
648 		err = filemap_write_and_wait(inode->i_mapping);
649 		if (err)
650 			goto out;
651 	}
652 
653 	/*
654 	 * We may force a getattr if the user cares about atime.
655 	 *
656 	 * Note that we only have to check the vfsmount flags here:
657 	 *  - NFS always sets S_NOATIME by so checking it would give a
658 	 *    bogus result
659 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
660 	 *    no point in checking those.
661 	 */
662  	if ((mnt->mnt_flags & MNT_NOATIME) ||
663  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
664 		need_atime = 0;
665 
666 	if (need_atime || nfs_need_revalidate_inode(inode)) {
667 		struct nfs_server *server = NFS_SERVER(inode);
668 
669 		if (server->caps & NFS_CAP_READDIRPLUS)
670 			nfs_request_parent_use_readdirplus(dentry);
671 		err = __nfs_revalidate_inode(server, inode);
672 	}
673 	if (!err) {
674 		generic_fillattr(inode, stat);
675 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
676 	}
677 out:
678 	trace_nfs_getattr_exit(inode, err);
679 	return err;
680 }
681 EXPORT_SYMBOL_GPL(nfs_getattr);
682 
683 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
684 {
685 	atomic_set(&l_ctx->count, 1);
686 	l_ctx->lockowner.l_owner = current->files;
687 	l_ctx->lockowner.l_pid = current->tgid;
688 	INIT_LIST_HEAD(&l_ctx->list);
689 	nfs_iocounter_init(&l_ctx->io_count);
690 }
691 
692 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
693 {
694 	struct nfs_lock_context *head = &ctx->lock_context;
695 	struct nfs_lock_context *pos = head;
696 
697 	do {
698 		if (pos->lockowner.l_owner != current->files)
699 			continue;
700 		if (pos->lockowner.l_pid != current->tgid)
701 			continue;
702 		atomic_inc(&pos->count);
703 		return pos;
704 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
705 	return NULL;
706 }
707 
708 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
709 {
710 	struct nfs_lock_context *res, *new = NULL;
711 	struct inode *inode = ctx->dentry->d_inode;
712 
713 	spin_lock(&inode->i_lock);
714 	res = __nfs_find_lock_context(ctx);
715 	if (res == NULL) {
716 		spin_unlock(&inode->i_lock);
717 		new = kmalloc(sizeof(*new), GFP_KERNEL);
718 		if (new == NULL)
719 			return ERR_PTR(-ENOMEM);
720 		nfs_init_lock_context(new);
721 		spin_lock(&inode->i_lock);
722 		res = __nfs_find_lock_context(ctx);
723 		if (res == NULL) {
724 			list_add_tail(&new->list, &ctx->lock_context.list);
725 			new->open_context = ctx;
726 			res = new;
727 			new = NULL;
728 		}
729 	}
730 	spin_unlock(&inode->i_lock);
731 	kfree(new);
732 	return res;
733 }
734 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
735 
736 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
737 {
738 	struct nfs_open_context *ctx = l_ctx->open_context;
739 	struct inode *inode = ctx->dentry->d_inode;
740 
741 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
742 		return;
743 	list_del(&l_ctx->list);
744 	spin_unlock(&inode->i_lock);
745 	kfree(l_ctx);
746 }
747 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
748 
749 /**
750  * nfs_close_context - Common close_context() routine NFSv2/v3
751  * @ctx: pointer to context
752  * @is_sync: is this a synchronous close
753  *
754  * always ensure that the attributes are up to date if we're mounted
755  * with close-to-open semantics
756  */
757 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
758 {
759 	struct inode *inode;
760 	struct nfs_server *server;
761 
762 	if (!(ctx->mode & FMODE_WRITE))
763 		return;
764 	if (!is_sync)
765 		return;
766 	inode = ctx->dentry->d_inode;
767 	if (!list_empty(&NFS_I(inode)->open_files))
768 		return;
769 	server = NFS_SERVER(inode);
770 	if (server->flags & NFS_MOUNT_NOCTO)
771 		return;
772 	nfs_revalidate_inode(server, inode);
773 }
774 EXPORT_SYMBOL_GPL(nfs_close_context);
775 
776 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
777 {
778 	struct nfs_open_context *ctx;
779 	struct rpc_cred *cred = rpc_lookup_cred();
780 	if (IS_ERR(cred))
781 		return ERR_CAST(cred);
782 
783 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
784 	if (!ctx) {
785 		put_rpccred(cred);
786 		return ERR_PTR(-ENOMEM);
787 	}
788 	nfs_sb_active(dentry->d_sb);
789 	ctx->dentry = dget(dentry);
790 	ctx->cred = cred;
791 	ctx->state = NULL;
792 	ctx->mode = f_mode;
793 	ctx->flags = 0;
794 	ctx->error = 0;
795 	nfs_init_lock_context(&ctx->lock_context);
796 	ctx->lock_context.open_context = ctx;
797 	INIT_LIST_HEAD(&ctx->list);
798 	ctx->mdsthreshold = NULL;
799 	return ctx;
800 }
801 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
802 
803 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
804 {
805 	if (ctx != NULL)
806 		atomic_inc(&ctx->lock_context.count);
807 	return ctx;
808 }
809 EXPORT_SYMBOL_GPL(get_nfs_open_context);
810 
811 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
812 {
813 	struct inode *inode = ctx->dentry->d_inode;
814 	struct super_block *sb = ctx->dentry->d_sb;
815 
816 	if (!list_empty(&ctx->list)) {
817 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
818 			return;
819 		list_del(&ctx->list);
820 		spin_unlock(&inode->i_lock);
821 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
822 		return;
823 	if (inode != NULL)
824 		NFS_PROTO(inode)->close_context(ctx, is_sync);
825 	if (ctx->cred != NULL)
826 		put_rpccred(ctx->cred);
827 	dput(ctx->dentry);
828 	nfs_sb_deactive(sb);
829 	kfree(ctx->mdsthreshold);
830 	kfree(ctx);
831 }
832 
833 void put_nfs_open_context(struct nfs_open_context *ctx)
834 {
835 	__put_nfs_open_context(ctx, 0);
836 }
837 EXPORT_SYMBOL_GPL(put_nfs_open_context);
838 
839 /*
840  * Ensure that mmap has a recent RPC credential for use when writing out
841  * shared pages
842  */
843 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
844 {
845 	struct inode *inode = ctx->dentry->d_inode;
846 	struct nfs_inode *nfsi = NFS_I(inode);
847 
848 	spin_lock(&inode->i_lock);
849 	list_add(&ctx->list, &nfsi->open_files);
850 	spin_unlock(&inode->i_lock);
851 }
852 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
853 
854 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
855 {
856 	filp->private_data = get_nfs_open_context(ctx);
857 	if (list_empty(&ctx->list))
858 		nfs_inode_attach_open_context(ctx);
859 }
860 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
861 
862 /*
863  * Given an inode, search for an open context with the desired characteristics
864  */
865 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
866 {
867 	struct nfs_inode *nfsi = NFS_I(inode);
868 	struct nfs_open_context *pos, *ctx = NULL;
869 
870 	spin_lock(&inode->i_lock);
871 	list_for_each_entry(pos, &nfsi->open_files, list) {
872 		if (cred != NULL && pos->cred != cred)
873 			continue;
874 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
875 			continue;
876 		ctx = get_nfs_open_context(pos);
877 		break;
878 	}
879 	spin_unlock(&inode->i_lock);
880 	return ctx;
881 }
882 
883 static void nfs_file_clear_open_context(struct file *filp)
884 {
885 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
886 
887 	if (ctx) {
888 		struct inode *inode = ctx->dentry->d_inode;
889 
890 		filp->private_data = NULL;
891 		spin_lock(&inode->i_lock);
892 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
893 		spin_unlock(&inode->i_lock);
894 		__put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
895 	}
896 }
897 
898 /*
899  * These allocate and release file read/write context information.
900  */
901 int nfs_open(struct inode *inode, struct file *filp)
902 {
903 	struct nfs_open_context *ctx;
904 
905 	ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
906 	if (IS_ERR(ctx))
907 		return PTR_ERR(ctx);
908 	nfs_file_set_open_context(filp, ctx);
909 	put_nfs_open_context(ctx);
910 	nfs_fscache_open_file(inode, filp);
911 	return 0;
912 }
913 
914 int nfs_release(struct inode *inode, struct file *filp)
915 {
916 	nfs_file_clear_open_context(filp);
917 	return 0;
918 }
919 
920 /*
921  * This function is called whenever some part of NFS notices that
922  * the cached attributes have to be refreshed.
923  */
924 int
925 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
926 {
927 	int		 status = -ESTALE;
928 	struct nfs4_label *label = NULL;
929 	struct nfs_fattr *fattr = NULL;
930 	struct nfs_inode *nfsi = NFS_I(inode);
931 
932 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
933 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
934 
935 	trace_nfs_revalidate_inode_enter(inode);
936 
937 	if (is_bad_inode(inode))
938 		goto out;
939 	if (NFS_STALE(inode))
940 		goto out;
941 
942 	status = -ENOMEM;
943 	fattr = nfs_alloc_fattr();
944 	if (fattr == NULL)
945 		goto out;
946 
947 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
948 
949 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
950 	if (IS_ERR(label)) {
951 		status = PTR_ERR(label);
952 		goto out;
953 	}
954 
955 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
956 	if (status != 0) {
957 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
958 			 inode->i_sb->s_id,
959 			 (unsigned long long)NFS_FILEID(inode), status);
960 		if (status == -ESTALE) {
961 			nfs_zap_caches(inode);
962 			if (!S_ISDIR(inode->i_mode))
963 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
964 		}
965 		goto err_out;
966 	}
967 
968 	status = nfs_refresh_inode(inode, fattr);
969 	if (status) {
970 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
971 			 inode->i_sb->s_id,
972 			 (unsigned long long)NFS_FILEID(inode), status);
973 		goto err_out;
974 	}
975 
976 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
977 		nfs_zap_acl_cache(inode);
978 
979 	nfs_setsecurity(inode, fattr, label);
980 
981 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
982 		inode->i_sb->s_id,
983 		(unsigned long long)NFS_FILEID(inode));
984 
985 err_out:
986 	nfs4_label_free(label);
987 out:
988 	nfs_free_fattr(fattr);
989 	trace_nfs_revalidate_inode_exit(inode, status);
990 	return status;
991 }
992 
993 int nfs_attribute_timeout(struct inode *inode)
994 {
995 	struct nfs_inode *nfsi = NFS_I(inode);
996 
997 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
998 }
999 
1000 int nfs_attribute_cache_expired(struct inode *inode)
1001 {
1002 	if (nfs_have_delegated_attributes(inode))
1003 		return 0;
1004 	return nfs_attribute_timeout(inode);
1005 }
1006 
1007 /**
1008  * nfs_revalidate_inode - Revalidate the inode attributes
1009  * @server - pointer to nfs_server struct
1010  * @inode - pointer to inode struct
1011  *
1012  * Updates inode attribute information by retrieving the data from the server.
1013  */
1014 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1015 {
1016 	if (!nfs_need_revalidate_inode(inode))
1017 		return NFS_STALE(inode) ? -ESTALE : 0;
1018 	return __nfs_revalidate_inode(server, inode);
1019 }
1020 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1021 
1022 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1023 {
1024 	if (!(NFS_I(inode)->cache_validity &
1025 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1026 			&& !nfs_attribute_cache_expired(inode))
1027 		return NFS_STALE(inode) ? -ESTALE : 0;
1028 	return -ECHILD;
1029 }
1030 
1031 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1032 {
1033 	struct nfs_inode *nfsi = NFS_I(inode);
1034 	int ret;
1035 
1036 	if (mapping->nrpages != 0) {
1037 		if (S_ISREG(inode->i_mode)) {
1038 			unmap_mapping_range(mapping, 0, 0, 0);
1039 			ret = nfs_sync_mapping(mapping);
1040 			if (ret < 0)
1041 				return ret;
1042 		}
1043 		ret = invalidate_inode_pages2(mapping);
1044 		if (ret < 0)
1045 			return ret;
1046 	}
1047 	if (S_ISDIR(inode->i_mode)) {
1048 		spin_lock(&inode->i_lock);
1049 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1050 		spin_unlock(&inode->i_lock);
1051 	}
1052 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1053 	nfs_fscache_wait_on_invalidate(inode);
1054 
1055 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1056 			inode->i_sb->s_id,
1057 			(unsigned long long)NFS_FILEID(inode));
1058 	return 0;
1059 }
1060 
1061 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1062 {
1063 	if (nfs_have_delegated_attributes(inode))
1064 		return false;
1065 	return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1066 		|| nfs_attribute_timeout(inode)
1067 		|| NFS_STALE(inode);
1068 }
1069 
1070 /**
1071  * __nfs_revalidate_mapping - Revalidate the pagecache
1072  * @inode - pointer to host inode
1073  * @mapping - pointer to mapping
1074  * @may_lock - take inode->i_mutex?
1075  */
1076 static int __nfs_revalidate_mapping(struct inode *inode,
1077 		struct address_space *mapping,
1078 		bool may_lock)
1079 {
1080 	struct nfs_inode *nfsi = NFS_I(inode);
1081 	unsigned long *bitlock = &nfsi->flags;
1082 	int ret = 0;
1083 
1084 	/* swapfiles are not supposed to be shared. */
1085 	if (IS_SWAPFILE(inode))
1086 		goto out;
1087 
1088 	if (nfs_mapping_need_revalidate_inode(inode)) {
1089 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1090 		if (ret < 0)
1091 			goto out;
1092 	}
1093 
1094 	/*
1095 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1096 	 * invalidations that come in while we're shooting down the mappings
1097 	 * are respected. But, that leaves a race window where one revalidator
1098 	 * can clear the flag, and then another checks it before the mapping
1099 	 * gets invalidated. Fix that by serializing access to this part of
1100 	 * the function.
1101 	 *
1102 	 * At the same time, we need to allow other tasks to see whether we
1103 	 * might be in the middle of invalidating the pages, so we only set
1104 	 * the bit lock here if it looks like we're going to be doing that.
1105 	 */
1106 	for (;;) {
1107 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1108 					 nfs_wait_bit_killable, TASK_KILLABLE);
1109 		if (ret)
1110 			goto out;
1111 		spin_lock(&inode->i_lock);
1112 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1113 			spin_unlock(&inode->i_lock);
1114 			continue;
1115 		}
1116 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1117 			break;
1118 		spin_unlock(&inode->i_lock);
1119 		goto out;
1120 	}
1121 
1122 	set_bit(NFS_INO_INVALIDATING, bitlock);
1123 	smp_wmb();
1124 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1125 	spin_unlock(&inode->i_lock);
1126 	trace_nfs_invalidate_mapping_enter(inode);
1127 	if (may_lock) {
1128 		mutex_lock(&inode->i_mutex);
1129 		ret = nfs_invalidate_mapping(inode, mapping);
1130 		mutex_unlock(&inode->i_mutex);
1131 	} else
1132 		ret = nfs_invalidate_mapping(inode, mapping);
1133 	trace_nfs_invalidate_mapping_exit(inode, ret);
1134 
1135 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1136 	smp_mb__after_atomic();
1137 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1138 out:
1139 	return ret;
1140 }
1141 
1142 /**
1143  * nfs_revalidate_mapping - Revalidate the pagecache
1144  * @inode - pointer to host inode
1145  * @mapping - pointer to mapping
1146  */
1147 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1148 {
1149 	return __nfs_revalidate_mapping(inode, mapping, false);
1150 }
1151 
1152 /**
1153  * nfs_revalidate_mapping_protected - Revalidate the pagecache
1154  * @inode - pointer to host inode
1155  * @mapping - pointer to mapping
1156  *
1157  * Differs from nfs_revalidate_mapping() in that it grabs the inode->i_mutex
1158  * while invalidating the mapping.
1159  */
1160 int nfs_revalidate_mapping_protected(struct inode *inode, struct address_space *mapping)
1161 {
1162 	return __nfs_revalidate_mapping(inode, mapping, true);
1163 }
1164 
1165 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1166 {
1167 	struct nfs_inode *nfsi = NFS_I(inode);
1168 	unsigned long ret = 0;
1169 
1170 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1171 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1172 			&& inode->i_version == fattr->pre_change_attr) {
1173 		inode->i_version = fattr->change_attr;
1174 		if (S_ISDIR(inode->i_mode))
1175 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1176 		ret |= NFS_INO_INVALID_ATTR;
1177 	}
1178 	/* If we have atomic WCC data, we may update some attributes */
1179 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1180 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1181 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1182 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1183 		ret |= NFS_INO_INVALID_ATTR;
1184 	}
1185 
1186 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1187 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1188 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1189 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
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 ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1195 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1196 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1197 			&& nfsi->nrequests == 0) {
1198 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1199 		ret |= NFS_INO_INVALID_ATTR;
1200 	}
1201 
1202 	return ret;
1203 }
1204 
1205 /**
1206  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1207  * @inode - pointer to inode
1208  * @fattr - updated attributes
1209  *
1210  * Verifies the attribute cache. If we have just changed the attributes,
1211  * so that fattr carries weak cache consistency data, then it may
1212  * also update the ctime/mtime/change_attribute.
1213  */
1214 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1215 {
1216 	struct nfs_inode *nfsi = NFS_I(inode);
1217 	loff_t cur_size, new_isize;
1218 	unsigned long invalid = 0;
1219 
1220 
1221 	if (nfs_have_delegated_attributes(inode))
1222 		return 0;
1223 	/* Has the inode gone and changed behind our back? */
1224 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1225 		return -EIO;
1226 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1227 		return -EIO;
1228 
1229 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1230 			inode->i_version != fattr->change_attr)
1231 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1232 
1233 	/* Verify a few of the more important attributes */
1234 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1235 		invalid |= NFS_INO_INVALID_ATTR;
1236 
1237 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1238 		cur_size = i_size_read(inode);
1239 		new_isize = nfs_size_to_loff_t(fattr->size);
1240 		if (cur_size != new_isize && nfsi->nrequests == 0)
1241 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1242 	}
1243 
1244 	/* Have any file permissions changed? */
1245 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1246 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1247 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1248 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1249 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1250 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1251 
1252 	/* Has the link count changed? */
1253 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1254 		invalid |= NFS_INO_INVALID_ATTR;
1255 
1256 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1257 		invalid |= NFS_INO_INVALID_ATIME;
1258 
1259 	if (invalid != 0)
1260 		nfs_set_cache_invalid(inode, invalid);
1261 
1262 	nfsi->read_cache_jiffies = fattr->time_start;
1263 	return 0;
1264 }
1265 
1266 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1267 {
1268 	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1269 		return 0;
1270 	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1271 }
1272 
1273 static atomic_long_t nfs_attr_generation_counter;
1274 
1275 static unsigned long nfs_read_attr_generation_counter(void)
1276 {
1277 	return atomic_long_read(&nfs_attr_generation_counter);
1278 }
1279 
1280 unsigned long nfs_inc_attr_generation_counter(void)
1281 {
1282 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1283 }
1284 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1285 
1286 void nfs_fattr_init(struct nfs_fattr *fattr)
1287 {
1288 	fattr->valid = 0;
1289 	fattr->time_start = jiffies;
1290 	fattr->gencount = nfs_inc_attr_generation_counter();
1291 	fattr->owner_name = NULL;
1292 	fattr->group_name = NULL;
1293 }
1294 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1295 
1296 /**
1297  * nfs_fattr_set_barrier
1298  * @fattr: attributes
1299  *
1300  * Used to set a barrier after an attribute was updated. This
1301  * barrier ensures that older attributes from RPC calls that may
1302  * have raced with our update cannot clobber these new values.
1303  * Note that you are still responsible for ensuring that other
1304  * operations which change the attribute on the server do not
1305  * collide.
1306  */
1307 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1308 {
1309 	fattr->gencount = nfs_inc_attr_generation_counter();
1310 }
1311 
1312 struct nfs_fattr *nfs_alloc_fattr(void)
1313 {
1314 	struct nfs_fattr *fattr;
1315 
1316 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1317 	if (fattr != NULL)
1318 		nfs_fattr_init(fattr);
1319 	return fattr;
1320 }
1321 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1322 
1323 struct nfs_fh *nfs_alloc_fhandle(void)
1324 {
1325 	struct nfs_fh *fh;
1326 
1327 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1328 	if (fh != NULL)
1329 		fh->size = 0;
1330 	return fh;
1331 }
1332 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1333 
1334 #ifdef NFS_DEBUG
1335 /*
1336  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1337  *                             in the same way that wireshark does
1338  *
1339  * @fh: file handle
1340  *
1341  * For debugging only.
1342  */
1343 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1344 {
1345 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1346 	 * not on the result */
1347 	return nfs_fhandle_hash(fh);
1348 }
1349 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1350 
1351 /*
1352  * _nfs_display_fhandle - display an NFS file handle on the console
1353  *
1354  * @fh: file handle to display
1355  * @caption: display caption
1356  *
1357  * For debugging only.
1358  */
1359 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1360 {
1361 	unsigned short i;
1362 
1363 	if (fh == NULL || fh->size == 0) {
1364 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1365 		return;
1366 	}
1367 
1368 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1369 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1370 	for (i = 0; i < fh->size; i += 16) {
1371 		__be32 *pos = (__be32 *)&fh->data[i];
1372 
1373 		switch ((fh->size - i - 1) >> 2) {
1374 		case 0:
1375 			printk(KERN_DEFAULT " %08x\n",
1376 				be32_to_cpup(pos));
1377 			break;
1378 		case 1:
1379 			printk(KERN_DEFAULT " %08x %08x\n",
1380 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1381 			break;
1382 		case 2:
1383 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1384 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1385 				be32_to_cpup(pos + 2));
1386 			break;
1387 		default:
1388 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1389 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1390 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1391 		}
1392 	}
1393 }
1394 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1395 #endif
1396 
1397 /**
1398  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1399  * @inode - pointer to inode
1400  * @fattr - attributes
1401  *
1402  * Attempt to divine whether or not an RPC call reply carrying stale
1403  * attributes got scheduled after another call carrying updated ones.
1404  *
1405  * To do so, the function first assumes that a more recent ctime means
1406  * that the attributes in fattr are newer, however it also attempt to
1407  * catch the case where ctime either didn't change, or went backwards
1408  * (if someone reset the clock on the server) by looking at whether
1409  * or not this RPC call was started after the inode was last updated.
1410  * Note also the check for wraparound of 'attr_gencount'
1411  *
1412  * The function returns 'true' if it thinks the attributes in 'fattr' are
1413  * more recent than the ones cached in the inode.
1414  *
1415  */
1416 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1417 {
1418 	const struct nfs_inode *nfsi = NFS_I(inode);
1419 
1420 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1421 		nfs_ctime_need_update(inode, fattr) ||
1422 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1423 }
1424 
1425 /*
1426  * Don't trust the change_attribute, mtime, ctime or size if
1427  * a pnfs LAYOUTCOMMIT is outstanding
1428  */
1429 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1430 		struct nfs_fattr *fattr)
1431 {
1432 	if (pnfs_layoutcommit_outstanding(inode))
1433 		fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1434 				NFS_ATTR_FATTR_MTIME |
1435 				NFS_ATTR_FATTR_CTIME |
1436 				NFS_ATTR_FATTR_SIZE);
1437 }
1438 
1439 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1440 {
1441 	int ret;
1442 
1443 	trace_nfs_refresh_inode_enter(inode);
1444 
1445 	nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1446 
1447 	if (nfs_inode_attrs_need_update(inode, fattr))
1448 		ret = nfs_update_inode(inode, fattr);
1449 	else
1450 		ret = nfs_check_inode_attributes(inode, fattr);
1451 
1452 	trace_nfs_refresh_inode_exit(inode, ret);
1453 	return ret;
1454 }
1455 
1456 /**
1457  * nfs_refresh_inode - try to update the inode attribute cache
1458  * @inode - pointer to inode
1459  * @fattr - updated attributes
1460  *
1461  * Check that an RPC call that returned attributes has not overlapped with
1462  * other recent updates of the inode metadata, then decide whether it is
1463  * safe to do a full update of the inode attributes, or whether just to
1464  * call nfs_check_inode_attributes.
1465  */
1466 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1467 {
1468 	int status;
1469 
1470 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1471 		return 0;
1472 	spin_lock(&inode->i_lock);
1473 	status = nfs_refresh_inode_locked(inode, fattr);
1474 	spin_unlock(&inode->i_lock);
1475 
1476 	return status;
1477 }
1478 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1479 
1480 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1481 {
1482 	unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1483 
1484 	if (S_ISDIR(inode->i_mode))
1485 		invalid |= NFS_INO_INVALID_DATA;
1486 	nfs_set_cache_invalid(inode, invalid);
1487 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1488 		return 0;
1489 	return nfs_refresh_inode_locked(inode, fattr);
1490 }
1491 
1492 /**
1493  * nfs_post_op_update_inode - try to update the inode attribute cache
1494  * @inode - pointer to inode
1495  * @fattr - updated attributes
1496  *
1497  * After an operation that has changed the inode metadata, mark the
1498  * attribute cache as being invalid, then try to update it.
1499  *
1500  * NB: if the server didn't return any post op attributes, this
1501  * function will force the retrieval of attributes before the next
1502  * NFS request.  Thus it should be used only for operations that
1503  * are expected to change one or more attributes, to avoid
1504  * unnecessary NFS requests and trips through nfs_update_inode().
1505  */
1506 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1507 {
1508 	int status;
1509 
1510 	spin_lock(&inode->i_lock);
1511 	nfs_fattr_set_barrier(fattr);
1512 	status = nfs_post_op_update_inode_locked(inode, fattr);
1513 	spin_unlock(&inode->i_lock);
1514 
1515 	return status;
1516 }
1517 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1518 
1519 /**
1520  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1521  * @inode - pointer to inode
1522  * @fattr - updated attributes
1523  *
1524  * After an operation that has changed the inode metadata, mark the
1525  * attribute cache as being invalid, then try to update it. Fake up
1526  * weak cache consistency data, if none exist.
1527  *
1528  * This function is mainly designed to be used by the ->write_done() functions.
1529  */
1530 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1531 {
1532 	int status;
1533 
1534 	/* Don't do a WCC update if these attributes are already stale */
1535 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1536 			!nfs_inode_attrs_need_update(inode, fattr)) {
1537 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1538 				| NFS_ATTR_FATTR_PRESIZE
1539 				| NFS_ATTR_FATTR_PREMTIME
1540 				| NFS_ATTR_FATTR_PRECTIME);
1541 		goto out_noforce;
1542 	}
1543 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1544 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1545 		fattr->pre_change_attr = inode->i_version;
1546 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1547 	}
1548 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1549 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1550 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1551 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1552 	}
1553 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1554 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1555 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1556 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1557 	}
1558 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1559 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1560 		fattr->pre_size = i_size_read(inode);
1561 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1562 	}
1563 out_noforce:
1564 	status = nfs_post_op_update_inode_locked(inode, fattr);
1565 	return status;
1566 }
1567 
1568 /**
1569  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1570  * @inode - pointer to inode
1571  * @fattr - updated attributes
1572  *
1573  * After an operation that has changed the inode metadata, mark the
1574  * attribute cache as being invalid, then try to update it. Fake up
1575  * weak cache consistency data, if none exist.
1576  *
1577  * This function is mainly designed to be used by the ->write_done() functions.
1578  */
1579 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1580 {
1581 	int status;
1582 
1583 	spin_lock(&inode->i_lock);
1584 	nfs_fattr_set_barrier(fattr);
1585 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1586 	spin_unlock(&inode->i_lock);
1587 	return status;
1588 }
1589 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1590 
1591 /*
1592  * Many nfs protocol calls return the new file attributes after
1593  * an operation.  Here we update the inode to reflect the state
1594  * of the server's inode.
1595  *
1596  * This is a bit tricky because we have to make sure all dirty pages
1597  * have been sent off to the server before calling invalidate_inode_pages.
1598  * To make sure no other process adds more write requests while we try
1599  * our best to flush them, we make them sleep during the attribute refresh.
1600  *
1601  * A very similar scenario holds for the dir cache.
1602  */
1603 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1604 {
1605 	struct nfs_server *server;
1606 	struct nfs_inode *nfsi = NFS_I(inode);
1607 	loff_t cur_isize, new_isize;
1608 	unsigned long invalid = 0;
1609 	unsigned long now = jiffies;
1610 	unsigned long save_cache_validity;
1611 
1612 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1613 			__func__, inode->i_sb->s_id, inode->i_ino,
1614 			nfs_display_fhandle_hash(NFS_FH(inode)),
1615 			atomic_read(&inode->i_count), fattr->valid);
1616 
1617 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1618 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1619 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1620 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1621 			inode->i_sb->s_id, (long long)nfsi->fileid,
1622 			(long long)fattr->fileid);
1623 		goto out_err;
1624 	}
1625 
1626 	/*
1627 	 * Make sure the inode's type hasn't changed.
1628 	 */
1629 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1630 		/*
1631 		* Big trouble! The inode has become a different object.
1632 		*/
1633 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1634 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1635 		goto out_err;
1636 	}
1637 
1638 	server = NFS_SERVER(inode);
1639 	/* Update the fsid? */
1640 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1641 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1642 			!IS_AUTOMOUNT(inode))
1643 		server->fsid = fattr->fsid;
1644 
1645 	/*
1646 	 * Update the read time so we don't revalidate too often.
1647 	 */
1648 	nfsi->read_cache_jiffies = fattr->time_start;
1649 
1650 	save_cache_validity = nfsi->cache_validity;
1651 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1652 			| NFS_INO_INVALID_ATIME
1653 			| NFS_INO_REVAL_FORCED
1654 			| NFS_INO_REVAL_PAGECACHE);
1655 
1656 	/* Do atomic weak cache consistency updates */
1657 	invalid |= nfs_wcc_update_inode(inode, fattr);
1658 
1659 	/* More cache consistency checks */
1660 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1661 		if (inode->i_version != fattr->change_attr) {
1662 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1663 					inode->i_sb->s_id, inode->i_ino);
1664 			invalid |= NFS_INO_INVALID_ATTR
1665 				| NFS_INO_INVALID_DATA
1666 				| NFS_INO_INVALID_ACCESS
1667 				| NFS_INO_INVALID_ACL
1668 				| NFS_INO_REVAL_PAGECACHE;
1669 			if (S_ISDIR(inode->i_mode))
1670 				nfs_force_lookup_revalidate(inode);
1671 			inode->i_version = fattr->change_attr;
1672 		}
1673 	} else if (server->caps & NFS_CAP_CHANGE_ATTR)
1674 		nfsi->cache_validity |= save_cache_validity;
1675 
1676 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1677 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1678 	} else if (server->caps & NFS_CAP_MTIME)
1679 		nfsi->cache_validity |= save_cache_validity &
1680 				(NFS_INO_INVALID_ATTR
1681 				| NFS_INO_REVAL_FORCED);
1682 
1683 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1684 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1685 	} else if (server->caps & NFS_CAP_CTIME)
1686 		nfsi->cache_validity |= save_cache_validity &
1687 				(NFS_INO_INVALID_ATTR
1688 				| NFS_INO_REVAL_FORCED);
1689 
1690 	/* Check if our cached file size is stale */
1691 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1692 		new_isize = nfs_size_to_loff_t(fattr->size);
1693 		cur_isize = i_size_read(inode);
1694 		if (new_isize != cur_isize) {
1695 			/* Do we perhaps have any outstanding writes, or has
1696 			 * the file grown beyond our last write? */
1697 			if ((nfsi->nrequests == 0) || new_isize > cur_isize) {
1698 				i_size_write(inode, new_isize);
1699 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1700 				invalid &= ~NFS_INO_REVAL_PAGECACHE;
1701 			}
1702 			dprintk("NFS: isize change on server for file %s/%ld "
1703 					"(%Ld to %Ld)\n",
1704 					inode->i_sb->s_id,
1705 					inode->i_ino,
1706 					(long long)cur_isize,
1707 					(long long)new_isize);
1708 		}
1709 	} else
1710 		nfsi->cache_validity |= save_cache_validity &
1711 				(NFS_INO_INVALID_ATTR
1712 				| NFS_INO_REVAL_PAGECACHE
1713 				| NFS_INO_REVAL_FORCED);
1714 
1715 
1716 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1717 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1718 	else if (server->caps & NFS_CAP_ATIME)
1719 		nfsi->cache_validity |= save_cache_validity &
1720 				(NFS_INO_INVALID_ATIME
1721 				| NFS_INO_REVAL_FORCED);
1722 
1723 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1724 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1725 			umode_t newmode = inode->i_mode & S_IFMT;
1726 			newmode |= fattr->mode & S_IALLUGO;
1727 			inode->i_mode = newmode;
1728 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1729 		}
1730 	} else if (server->caps & NFS_CAP_MODE)
1731 		nfsi->cache_validity |= save_cache_validity &
1732 				(NFS_INO_INVALID_ATTR
1733 				| NFS_INO_INVALID_ACCESS
1734 				| NFS_INO_INVALID_ACL
1735 				| NFS_INO_REVAL_FORCED);
1736 
1737 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1738 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1739 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1740 			inode->i_uid = fattr->uid;
1741 		}
1742 	} else if (server->caps & NFS_CAP_OWNER)
1743 		nfsi->cache_validity |= save_cache_validity &
1744 				(NFS_INO_INVALID_ATTR
1745 				| NFS_INO_INVALID_ACCESS
1746 				| NFS_INO_INVALID_ACL
1747 				| NFS_INO_REVAL_FORCED);
1748 
1749 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1750 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1751 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1752 			inode->i_gid = fattr->gid;
1753 		}
1754 	} else if (server->caps & NFS_CAP_OWNER_GROUP)
1755 		nfsi->cache_validity |= save_cache_validity &
1756 				(NFS_INO_INVALID_ATTR
1757 				| NFS_INO_INVALID_ACCESS
1758 				| NFS_INO_INVALID_ACL
1759 				| NFS_INO_REVAL_FORCED);
1760 
1761 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1762 		if (inode->i_nlink != fattr->nlink) {
1763 			invalid |= NFS_INO_INVALID_ATTR;
1764 			if (S_ISDIR(inode->i_mode))
1765 				invalid |= NFS_INO_INVALID_DATA;
1766 			set_nlink(inode, fattr->nlink);
1767 		}
1768 	} else if (server->caps & NFS_CAP_NLINK)
1769 		nfsi->cache_validity |= save_cache_validity &
1770 				(NFS_INO_INVALID_ATTR
1771 				| NFS_INO_REVAL_FORCED);
1772 
1773 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1774 		/*
1775 		 * report the blocks in 512byte units
1776 		 */
1777 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1778  	}
1779 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1780 		inode->i_blocks = fattr->du.nfs2.blocks;
1781 
1782 	/* Update attrtimeo value if we're out of the unstable period */
1783 	if (invalid & NFS_INO_INVALID_ATTR) {
1784 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1785 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1786 		nfsi->attrtimeo_timestamp = now;
1787 		/* Set barrier to be more recent than all outstanding updates */
1788 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1789 	} else {
1790 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1791 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1792 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1793 			nfsi->attrtimeo_timestamp = now;
1794 		}
1795 		/* Set the barrier to be more recent than this fattr */
1796 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1797 			nfsi->attr_gencount = fattr->gencount;
1798 	}
1799 	invalid &= ~NFS_INO_INVALID_ATTR;
1800 	/* Don't invalidate the data if we were to blame */
1801 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1802 				|| S_ISLNK(inode->i_mode)))
1803 		invalid &= ~NFS_INO_INVALID_DATA;
1804 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1805 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1806 		nfs_set_cache_invalid(inode, invalid);
1807 
1808 	return 0;
1809  out_err:
1810 	/*
1811 	 * No need to worry about unhashing the dentry, as the
1812 	 * lookup validation will know that the inode is bad.
1813 	 * (But we fall through to invalidate the caches.)
1814 	 */
1815 	nfs_invalidate_inode(inode);
1816 	return -ESTALE;
1817 }
1818 
1819 struct inode *nfs_alloc_inode(struct super_block *sb)
1820 {
1821 	struct nfs_inode *nfsi;
1822 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1823 	if (!nfsi)
1824 		return NULL;
1825 	nfsi->flags = 0UL;
1826 	nfsi->cache_validity = 0UL;
1827 #if IS_ENABLED(CONFIG_NFS_V4)
1828 	nfsi->nfs4_acl = NULL;
1829 #endif /* CONFIG_NFS_V4 */
1830 	return &nfsi->vfs_inode;
1831 }
1832 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1833 
1834 static void nfs_i_callback(struct rcu_head *head)
1835 {
1836 	struct inode *inode = container_of(head, struct inode, i_rcu);
1837 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1838 }
1839 
1840 void nfs_destroy_inode(struct inode *inode)
1841 {
1842 	call_rcu(&inode->i_rcu, nfs_i_callback);
1843 }
1844 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1845 
1846 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1847 {
1848 #if IS_ENABLED(CONFIG_NFS_V4)
1849 	INIT_LIST_HEAD(&nfsi->open_states);
1850 	nfsi->delegation = NULL;
1851 	init_rwsem(&nfsi->rwsem);
1852 	nfsi->layout = NULL;
1853 #endif
1854 }
1855 
1856 static void init_once(void *foo)
1857 {
1858 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1859 
1860 	inode_init_once(&nfsi->vfs_inode);
1861 	INIT_LIST_HEAD(&nfsi->open_files);
1862 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1863 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1864 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1865 	nfsi->nrequests = 0;
1866 	nfsi->commit_info.ncommit = 0;
1867 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1868 	atomic_set(&nfsi->silly_count, 1);
1869 	INIT_HLIST_HEAD(&nfsi->silly_list);
1870 	init_waitqueue_head(&nfsi->waitqueue);
1871 	nfs4_init_once(nfsi);
1872 }
1873 
1874 static int __init nfs_init_inodecache(void)
1875 {
1876 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1877 					     sizeof(struct nfs_inode),
1878 					     0, (SLAB_RECLAIM_ACCOUNT|
1879 						SLAB_MEM_SPREAD),
1880 					     init_once);
1881 	if (nfs_inode_cachep == NULL)
1882 		return -ENOMEM;
1883 
1884 	return 0;
1885 }
1886 
1887 static void nfs_destroy_inodecache(void)
1888 {
1889 	/*
1890 	 * Make sure all delayed rcu free inodes are flushed before we
1891 	 * destroy cache.
1892 	 */
1893 	rcu_barrier();
1894 	kmem_cache_destroy(nfs_inode_cachep);
1895 }
1896 
1897 struct workqueue_struct *nfsiod_workqueue;
1898 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1899 
1900 /*
1901  * start up the nfsiod workqueue
1902  */
1903 static int nfsiod_start(void)
1904 {
1905 	struct workqueue_struct *wq;
1906 	dprintk("RPC:       creating workqueue nfsiod\n");
1907 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1908 	if (wq == NULL)
1909 		return -ENOMEM;
1910 	nfsiod_workqueue = wq;
1911 	return 0;
1912 }
1913 
1914 /*
1915  * Destroy the nfsiod workqueue
1916  */
1917 static void nfsiod_stop(void)
1918 {
1919 	struct workqueue_struct *wq;
1920 
1921 	wq = nfsiod_workqueue;
1922 	if (wq == NULL)
1923 		return;
1924 	nfsiod_workqueue = NULL;
1925 	destroy_workqueue(wq);
1926 }
1927 
1928 int nfs_net_id;
1929 EXPORT_SYMBOL_GPL(nfs_net_id);
1930 
1931 static int nfs_net_init(struct net *net)
1932 {
1933 	nfs_clients_init(net);
1934 	return nfs_fs_proc_net_init(net);
1935 }
1936 
1937 static void nfs_net_exit(struct net *net)
1938 {
1939 	nfs_fs_proc_net_exit(net);
1940 	nfs_cleanup_cb_ident_idr(net);
1941 }
1942 
1943 static struct pernet_operations nfs_net_ops = {
1944 	.init = nfs_net_init,
1945 	.exit = nfs_net_exit,
1946 	.id   = &nfs_net_id,
1947 	.size = sizeof(struct nfs_net),
1948 };
1949 
1950 /*
1951  * Initialize NFS
1952  */
1953 static int __init init_nfs_fs(void)
1954 {
1955 	int err;
1956 
1957 	err = register_pernet_subsys(&nfs_net_ops);
1958 	if (err < 0)
1959 		goto out9;
1960 
1961 	err = nfs_fscache_register();
1962 	if (err < 0)
1963 		goto out8;
1964 
1965 	err = nfsiod_start();
1966 	if (err)
1967 		goto out7;
1968 
1969 	err = nfs_fs_proc_init();
1970 	if (err)
1971 		goto out6;
1972 
1973 	err = nfs_init_nfspagecache();
1974 	if (err)
1975 		goto out5;
1976 
1977 	err = nfs_init_inodecache();
1978 	if (err)
1979 		goto out4;
1980 
1981 	err = nfs_init_readpagecache();
1982 	if (err)
1983 		goto out3;
1984 
1985 	err = nfs_init_writepagecache();
1986 	if (err)
1987 		goto out2;
1988 
1989 	err = nfs_init_directcache();
1990 	if (err)
1991 		goto out1;
1992 
1993 #ifdef CONFIG_PROC_FS
1994 	rpc_proc_register(&init_net, &nfs_rpcstat);
1995 #endif
1996 	if ((err = register_nfs_fs()) != 0)
1997 		goto out0;
1998 
1999 	return 0;
2000 out0:
2001 #ifdef CONFIG_PROC_FS
2002 	rpc_proc_unregister(&init_net, "nfs");
2003 #endif
2004 	nfs_destroy_directcache();
2005 out1:
2006 	nfs_destroy_writepagecache();
2007 out2:
2008 	nfs_destroy_readpagecache();
2009 out3:
2010 	nfs_destroy_inodecache();
2011 out4:
2012 	nfs_destroy_nfspagecache();
2013 out5:
2014 	nfs_fs_proc_exit();
2015 out6:
2016 	nfsiod_stop();
2017 out7:
2018 	nfs_fscache_unregister();
2019 out8:
2020 	unregister_pernet_subsys(&nfs_net_ops);
2021 out9:
2022 	return err;
2023 }
2024 
2025 static void __exit exit_nfs_fs(void)
2026 {
2027 	nfs_destroy_directcache();
2028 	nfs_destroy_writepagecache();
2029 	nfs_destroy_readpagecache();
2030 	nfs_destroy_inodecache();
2031 	nfs_destroy_nfspagecache();
2032 	nfs_fscache_unregister();
2033 	unregister_pernet_subsys(&nfs_net_ops);
2034 #ifdef CONFIG_PROC_FS
2035 	rpc_proc_unregister(&init_net, "nfs");
2036 #endif
2037 	unregister_nfs_fs();
2038 	nfs_fs_proc_exit();
2039 	nfsiod_stop();
2040 }
2041 
2042 /* Not quite true; I just maintain it */
2043 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2044 MODULE_LICENSE("GPL");
2045 module_param(enable_ino64, bool, 0644);
2046 
2047 module_init(init_nfs_fs)
2048 module_exit(exit_nfs_fs)
2049