xref: /openbmc/linux/fs/nfs/inode.c (revision aac5987a)
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/signal.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 <linux/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 static int nfs_wait_killable(int mode)
75 {
76 	freezable_schedule_unsafe();
77 	if (signal_pending_state(mode, current))
78 		return -ERESTARTSYS;
79 	return 0;
80 }
81 
82 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
83 {
84 	return nfs_wait_killable(mode);
85 }
86 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
87 
88 int nfs_wait_atomic_killable(atomic_t *p)
89 {
90 	return nfs_wait_killable(TASK_KILLABLE);
91 }
92 
93 /**
94  * nfs_compat_user_ino64 - returns the user-visible inode number
95  * @fileid: 64-bit fileid
96  *
97  * This function returns a 32-bit inode number if the boot parameter
98  * nfs.enable_ino64 is zero.
99  */
100 u64 nfs_compat_user_ino64(u64 fileid)
101 {
102 #ifdef CONFIG_COMPAT
103 	compat_ulong_t ino;
104 #else
105 	unsigned long ino;
106 #endif
107 
108 	if (enable_ino64)
109 		return fileid;
110 	ino = fileid;
111 	if (sizeof(ino) < sizeof(fileid))
112 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
113 	return ino;
114 }
115 
116 int nfs_drop_inode(struct inode *inode)
117 {
118 	return NFS_STALE(inode) || generic_drop_inode(inode);
119 }
120 EXPORT_SYMBOL_GPL(nfs_drop_inode);
121 
122 void nfs_clear_inode(struct inode *inode)
123 {
124 	/*
125 	 * The following should never happen...
126 	 */
127 	WARN_ON_ONCE(nfs_have_writebacks(inode));
128 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
129 	nfs_zap_acl_cache(inode);
130 	nfs_access_zap_cache(inode);
131 	nfs_fscache_clear_inode(inode);
132 }
133 EXPORT_SYMBOL_GPL(nfs_clear_inode);
134 
135 void nfs_evict_inode(struct inode *inode)
136 {
137 	truncate_inode_pages_final(&inode->i_data);
138 	clear_inode(inode);
139 	nfs_clear_inode(inode);
140 }
141 
142 int nfs_sync_inode(struct inode *inode)
143 {
144 	inode_dio_wait(inode);
145 	return nfs_wb_all(inode);
146 }
147 EXPORT_SYMBOL_GPL(nfs_sync_inode);
148 
149 /**
150  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
151  */
152 int nfs_sync_mapping(struct address_space *mapping)
153 {
154 	int ret = 0;
155 
156 	if (mapping->nrpages != 0) {
157 		unmap_mapping_range(mapping, 0, 0, 0);
158 		ret = nfs_wb_all(mapping->host);
159 	}
160 	return ret;
161 }
162 
163 static int nfs_attribute_timeout(struct inode *inode)
164 {
165 	struct nfs_inode *nfsi = NFS_I(inode);
166 
167 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
168 }
169 
170 static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
171 {
172 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
173 
174 	/* Special case for the pagecache or access cache */
175 	if (flags == NFS_INO_REVAL_PAGECACHE &&
176 	    !(cache_validity & NFS_INO_REVAL_FORCED))
177 		return false;
178 	return (cache_validity & flags) != 0;
179 }
180 
181 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
182 {
183 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
184 
185 	if ((cache_validity & flags) != 0)
186 		return true;
187 	if (nfs_attribute_timeout(inode))
188 		return true;
189 	return false;
190 }
191 
192 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
193 {
194 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
195 		return nfs_check_cache_invalid_delegated(inode, flags);
196 
197 	return nfs_check_cache_invalid_not_delegated(inode, flags);
198 }
199 
200 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
201 {
202 	struct nfs_inode *nfsi = NFS_I(inode);
203 
204 	if (inode->i_mapping->nrpages == 0)
205 		flags &= ~NFS_INO_INVALID_DATA;
206 	nfsi->cache_validity |= flags;
207 	if (flags & NFS_INO_INVALID_DATA)
208 		nfs_fscache_invalidate(inode);
209 }
210 
211 /*
212  * Invalidate the local caches
213  */
214 static void nfs_zap_caches_locked(struct inode *inode)
215 {
216 	struct nfs_inode *nfsi = NFS_I(inode);
217 	int mode = inode->i_mode;
218 
219 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
220 
221 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
222 	nfsi->attrtimeo_timestamp = jiffies;
223 
224 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
225 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
226 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
227 					| NFS_INO_INVALID_DATA
228 					| NFS_INO_INVALID_ACCESS
229 					| NFS_INO_INVALID_ACL
230 					| NFS_INO_REVAL_PAGECACHE);
231 	} else
232 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
233 					| NFS_INO_INVALID_ACCESS
234 					| NFS_INO_INVALID_ACL
235 					| NFS_INO_REVAL_PAGECACHE);
236 	nfs_zap_label_cache_locked(nfsi);
237 }
238 
239 void nfs_zap_caches(struct inode *inode)
240 {
241 	spin_lock(&inode->i_lock);
242 	nfs_zap_caches_locked(inode);
243 	spin_unlock(&inode->i_lock);
244 }
245 
246 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
247 {
248 	if (mapping->nrpages != 0) {
249 		spin_lock(&inode->i_lock);
250 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
251 		spin_unlock(&inode->i_lock);
252 	}
253 }
254 
255 void nfs_zap_acl_cache(struct inode *inode)
256 {
257 	void (*clear_acl_cache)(struct inode *);
258 
259 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
260 	if (clear_acl_cache != NULL)
261 		clear_acl_cache(inode);
262 	spin_lock(&inode->i_lock);
263 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
264 	spin_unlock(&inode->i_lock);
265 }
266 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
267 
268 void nfs_invalidate_atime(struct inode *inode)
269 {
270 	spin_lock(&inode->i_lock);
271 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
272 	spin_unlock(&inode->i_lock);
273 }
274 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
275 
276 /*
277  * Invalidate, but do not unhash, the inode.
278  * NB: must be called with inode->i_lock held!
279  */
280 static void nfs_invalidate_inode(struct inode *inode)
281 {
282 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
283 	nfs_zap_caches_locked(inode);
284 }
285 
286 struct nfs_find_desc {
287 	struct nfs_fh		*fh;
288 	struct nfs_fattr	*fattr;
289 };
290 
291 /*
292  * In NFSv3 we can have 64bit inode numbers. In order to support
293  * this, and re-exported directories (also seen in NFSv2)
294  * we are forced to allow 2 different inodes to have the same
295  * i_ino.
296  */
297 static int
298 nfs_find_actor(struct inode *inode, void *opaque)
299 {
300 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
301 	struct nfs_fh		*fh = desc->fh;
302 	struct nfs_fattr	*fattr = desc->fattr;
303 
304 	if (NFS_FILEID(inode) != fattr->fileid)
305 		return 0;
306 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
307 		return 0;
308 	if (nfs_compare_fh(NFS_FH(inode), fh))
309 		return 0;
310 	if (is_bad_inode(inode) || NFS_STALE(inode))
311 		return 0;
312 	return 1;
313 }
314 
315 static int
316 nfs_init_locked(struct inode *inode, void *opaque)
317 {
318 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
319 	struct nfs_fattr	*fattr = desc->fattr;
320 
321 	set_nfs_fileid(inode, fattr->fileid);
322 	inode->i_mode = fattr->mode;
323 	nfs_copy_fh(NFS_FH(inode), desc->fh);
324 	return 0;
325 }
326 
327 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
328 static void nfs_clear_label_invalid(struct inode *inode)
329 {
330 	spin_lock(&inode->i_lock);
331 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
332 	spin_unlock(&inode->i_lock);
333 }
334 
335 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
336 					struct nfs4_label *label)
337 {
338 	int error;
339 
340 	if (label == NULL)
341 		return;
342 
343 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
344 		error = security_inode_notifysecctx(inode, label->label,
345 				label->len);
346 		if (error)
347 			printk(KERN_ERR "%s() %s %d "
348 					"security_inode_notifysecctx() %d\n",
349 					__func__,
350 					(char *)label->label,
351 					label->len, error);
352 		nfs_clear_label_invalid(inode);
353 	}
354 }
355 
356 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
357 {
358 	struct nfs4_label *label = NULL;
359 	int minor_version = server->nfs_client->cl_minorversion;
360 
361 	if (minor_version < 2)
362 		return label;
363 
364 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
365 		return label;
366 
367 	label = kzalloc(sizeof(struct nfs4_label), flags);
368 	if (label == NULL)
369 		return ERR_PTR(-ENOMEM);
370 
371 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
372 	if (label->label == NULL) {
373 		kfree(label);
374 		return ERR_PTR(-ENOMEM);
375 	}
376 	label->len = NFS4_MAXLABELLEN;
377 
378 	return label;
379 }
380 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
381 #else
382 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
383 					struct nfs4_label *label)
384 {
385 }
386 #endif
387 EXPORT_SYMBOL_GPL(nfs_setsecurity);
388 
389 /*
390  * This is our front-end to iget that looks up inodes by file handle
391  * instead of inode number.
392  */
393 struct inode *
394 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
395 {
396 	struct nfs_find_desc desc = {
397 		.fh	= fh,
398 		.fattr	= fattr
399 	};
400 	struct inode *inode = ERR_PTR(-ENOENT);
401 	unsigned long hash;
402 
403 	nfs_attr_check_mountpoint(sb, fattr);
404 
405 	if (nfs_attr_use_mounted_on_fileid(fattr))
406 		fattr->fileid = fattr->mounted_on_fileid;
407 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
408 		goto out_no_inode;
409 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
410 		goto out_no_inode;
411 
412 	hash = nfs_fattr_to_ino_t(fattr);
413 
414 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
415 	if (inode == NULL) {
416 		inode = ERR_PTR(-ENOMEM);
417 		goto out_no_inode;
418 	}
419 
420 	if (inode->i_state & I_NEW) {
421 		struct nfs_inode *nfsi = NFS_I(inode);
422 		unsigned long now = jiffies;
423 
424 		/* We set i_ino for the few things that still rely on it,
425 		 * such as stat(2) */
426 		inode->i_ino = hash;
427 
428 		/* We can't support update_atime(), since the server will reset it */
429 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
430 		inode->i_mode = fattr->mode;
431 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
432 				&& nfs_server_capable(inode, NFS_CAP_MODE))
433 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
434 		/* Why so? Because we want revalidate for devices/FIFOs, and
435 		 * that's precisely what we have in nfs_file_inode_operations.
436 		 */
437 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
438 		if (S_ISREG(inode->i_mode)) {
439 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
440 			inode->i_data.a_ops = &nfs_file_aops;
441 		} else if (S_ISDIR(inode->i_mode)) {
442 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
443 			inode->i_fop = &nfs_dir_operations;
444 			inode->i_data.a_ops = &nfs_dir_aops;
445 			/* Deal with crossing mountpoints */
446 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
447 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
448 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
449 					inode->i_op = &nfs_referral_inode_operations;
450 				else
451 					inode->i_op = &nfs_mountpoint_inode_operations;
452 				inode->i_fop = NULL;
453 				inode->i_flags |= S_AUTOMOUNT;
454 			}
455 		} else if (S_ISLNK(inode->i_mode)) {
456 			inode->i_op = &nfs_symlink_inode_operations;
457 			inode_nohighmem(inode);
458 		} else
459 			init_special_inode(inode, inode->i_mode, fattr->rdev);
460 
461 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
462 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
463 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
464 		inode->i_version = 0;
465 		inode->i_size = 0;
466 		clear_nlink(inode);
467 		inode->i_uid = make_kuid(&init_user_ns, -2);
468 		inode->i_gid = make_kgid(&init_user_ns, -2);
469 		inode->i_blocks = 0;
470 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
471 		nfsi->write_io = 0;
472 		nfsi->read_io = 0;
473 
474 		nfsi->read_cache_jiffies = fattr->time_start;
475 		nfsi->attr_gencount = fattr->gencount;
476 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
477 			inode->i_atime = fattr->atime;
478 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
479 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
480 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
481 			inode->i_mtime = fattr->mtime;
482 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
483 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
484 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
485 			inode->i_ctime = fattr->ctime;
486 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
487 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
488 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
489 			inode->i_version = fattr->change_attr;
490 		else
491 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
492 				| NFS_INO_REVAL_PAGECACHE);
493 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
494 			inode->i_size = nfs_size_to_loff_t(fattr->size);
495 		else
496 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
497 				| NFS_INO_REVAL_PAGECACHE);
498 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
499 			set_nlink(inode, fattr->nlink);
500 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
501 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
502 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
503 			inode->i_uid = fattr->uid;
504 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
505 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
506 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
507 			inode->i_gid = fattr->gid;
508 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
509 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
510 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
511 			inode->i_blocks = fattr->du.nfs2.blocks;
512 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
513 			/*
514 			 * report the blocks in 512byte units
515 			 */
516 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
517 		}
518 
519 		nfs_setsecurity(inode, fattr, label);
520 
521 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
522 		nfsi->attrtimeo_timestamp = now;
523 		nfsi->access_cache = RB_ROOT;
524 
525 		nfs_fscache_init_inode(inode);
526 
527 		unlock_new_inode(inode);
528 	} else
529 		nfs_refresh_inode(inode, fattr);
530 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
531 		inode->i_sb->s_id,
532 		(unsigned long long)NFS_FILEID(inode),
533 		nfs_display_fhandle_hash(fh),
534 		atomic_read(&inode->i_count));
535 
536 out:
537 	return inode;
538 
539 out_no_inode:
540 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
541 	goto out;
542 }
543 EXPORT_SYMBOL_GPL(nfs_fhget);
544 
545 #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)
546 
547 int
548 nfs_setattr(struct dentry *dentry, struct iattr *attr)
549 {
550 	struct inode *inode = d_inode(dentry);
551 	struct nfs_fattr *fattr;
552 	int error = 0;
553 
554 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
555 
556 	/* skip mode change if it's just for clearing setuid/setgid */
557 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
558 		attr->ia_valid &= ~ATTR_MODE;
559 
560 	if (attr->ia_valid & ATTR_SIZE) {
561 		BUG_ON(!S_ISREG(inode->i_mode));
562 
563 		error = inode_newsize_ok(inode, attr->ia_size);
564 		if (error)
565 			return error;
566 
567 		if (attr->ia_size == i_size_read(inode))
568 			attr->ia_valid &= ~ATTR_SIZE;
569 	}
570 
571 	/* Optimization: if the end result is no change, don't RPC */
572 	attr->ia_valid &= NFS_VALID_ATTRS;
573 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
574 		return 0;
575 
576 	trace_nfs_setattr_enter(inode);
577 
578 	/* Write all dirty data */
579 	if (S_ISREG(inode->i_mode))
580 		nfs_sync_inode(inode);
581 
582 	fattr = nfs_alloc_fattr();
583 	if (fattr == NULL) {
584 		error = -ENOMEM;
585 		goto out;
586 	}
587 
588 	/*
589 	 * Return any delegations if we're going to change ACLs
590 	 */
591 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
592 		NFS_PROTO(inode)->return_delegation(inode);
593 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
594 	if (error == 0)
595 		error = nfs_refresh_inode(inode, fattr);
596 	nfs_free_fattr(fattr);
597 out:
598 	trace_nfs_setattr_exit(inode, error);
599 	return error;
600 }
601 EXPORT_SYMBOL_GPL(nfs_setattr);
602 
603 /**
604  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
605  * @inode: inode of the file used
606  * @offset: file offset to start truncating
607  *
608  * This is a copy of the common vmtruncate, but with the locking
609  * corrected to take into account the fact that NFS requires
610  * inode->i_size to be updated under the inode->i_lock.
611  * Note: must be called with inode->i_lock held!
612  */
613 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
614 {
615 	int err;
616 
617 	err = inode_newsize_ok(inode, offset);
618 	if (err)
619 		goto out;
620 
621 	i_size_write(inode, offset);
622 	/* Optimisation */
623 	if (offset == 0)
624 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
625 
626 	spin_unlock(&inode->i_lock);
627 	truncate_pagecache(inode, offset);
628 	spin_lock(&inode->i_lock);
629 out:
630 	return err;
631 }
632 
633 /**
634  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
635  * @inode: pointer to struct inode
636  * @attr: pointer to struct iattr
637  *
638  * Note: we do this in the *proc.c in order to ensure that
639  *       it works for things like exclusive creates too.
640  */
641 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
642 		struct nfs_fattr *fattr)
643 {
644 	/* Barrier: bump the attribute generation count. */
645 	nfs_fattr_set_barrier(fattr);
646 
647 	spin_lock(&inode->i_lock);
648 	NFS_I(inode)->attr_gencount = fattr->gencount;
649 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
650 		if ((attr->ia_valid & ATTR_MODE) != 0) {
651 			int mode = attr->ia_mode & S_IALLUGO;
652 			mode |= inode->i_mode & ~S_IALLUGO;
653 			inode->i_mode = mode;
654 		}
655 		if ((attr->ia_valid & ATTR_UID) != 0)
656 			inode->i_uid = attr->ia_uid;
657 		if ((attr->ia_valid & ATTR_GID) != 0)
658 			inode->i_gid = attr->ia_gid;
659 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
660 				| NFS_INO_INVALID_ACL);
661 	}
662 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
663 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
664 		nfs_vmtruncate(inode, attr->ia_size);
665 	}
666 	if (fattr->valid)
667 		nfs_update_inode(inode, fattr);
668 	else
669 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
670 	spin_unlock(&inode->i_lock);
671 }
672 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
673 
674 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
675 {
676 	struct dentry *parent;
677 
678 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
679 		return;
680 	parent = dget_parent(dentry);
681 	nfs_force_use_readdirplus(d_inode(parent));
682 	dput(parent);
683 }
684 
685 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
686 {
687 	struct dentry *parent;
688 
689 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
690 		return;
691 	parent = dget_parent(dentry);
692 	nfs_advise_use_readdirplus(d_inode(parent));
693 	dput(parent);
694 }
695 
696 static bool nfs_need_revalidate_inode(struct inode *inode)
697 {
698 	if (NFS_I(inode)->cache_validity &
699 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
700 		return true;
701 	if (nfs_attribute_cache_expired(inode))
702 		return true;
703 	return false;
704 }
705 
706 int nfs_getattr(const struct path *path, struct kstat *stat,
707 		u32 request_mask, unsigned int query_flags)
708 {
709 	struct inode *inode = d_inode(path->dentry);
710 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
711 	int err = 0;
712 
713 	trace_nfs_getattr_enter(inode);
714 	/* Flush out writes to the server in order to update c/mtime.  */
715 	if (S_ISREG(inode->i_mode)) {
716 		err = filemap_write_and_wait(inode->i_mapping);
717 		if (err)
718 			goto out;
719 	}
720 
721 	/*
722 	 * We may force a getattr if the user cares about atime.
723 	 *
724 	 * Note that we only have to check the vfsmount flags here:
725 	 *  - NFS always sets S_NOATIME by so checking it would give a
726 	 *    bogus result
727 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
728 	 *    no point in checking those.
729 	 */
730 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
731 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
732 		need_atime = 0;
733 
734 	if (need_atime || nfs_need_revalidate_inode(inode)) {
735 		struct nfs_server *server = NFS_SERVER(inode);
736 
737 		nfs_readdirplus_parent_cache_miss(path->dentry);
738 		err = __nfs_revalidate_inode(server, inode);
739 	} else
740 		nfs_readdirplus_parent_cache_hit(path->dentry);
741 	if (!err) {
742 		generic_fillattr(inode, stat);
743 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
744 		if (S_ISDIR(inode->i_mode))
745 			stat->blksize = NFS_SERVER(inode)->dtsize;
746 	}
747 out:
748 	trace_nfs_getattr_exit(inode, err);
749 	return err;
750 }
751 EXPORT_SYMBOL_GPL(nfs_getattr);
752 
753 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
754 {
755 	atomic_set(&l_ctx->count, 1);
756 	l_ctx->lockowner = current->files;
757 	INIT_LIST_HEAD(&l_ctx->list);
758 	atomic_set(&l_ctx->io_count, 0);
759 }
760 
761 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
762 {
763 	struct nfs_lock_context *head = &ctx->lock_context;
764 	struct nfs_lock_context *pos = head;
765 
766 	do {
767 		if (pos->lockowner != current->files)
768 			continue;
769 		atomic_inc(&pos->count);
770 		return pos;
771 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
772 	return NULL;
773 }
774 
775 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
776 {
777 	struct nfs_lock_context *res, *new = NULL;
778 	struct inode *inode = d_inode(ctx->dentry);
779 
780 	spin_lock(&inode->i_lock);
781 	res = __nfs_find_lock_context(ctx);
782 	if (res == NULL) {
783 		spin_unlock(&inode->i_lock);
784 		new = kmalloc(sizeof(*new), GFP_KERNEL);
785 		if (new == NULL)
786 			return ERR_PTR(-ENOMEM);
787 		nfs_init_lock_context(new);
788 		spin_lock(&inode->i_lock);
789 		res = __nfs_find_lock_context(ctx);
790 		if (res == NULL) {
791 			list_add_tail(&new->list, &ctx->lock_context.list);
792 			new->open_context = ctx;
793 			res = new;
794 			new = NULL;
795 		}
796 	}
797 	spin_unlock(&inode->i_lock);
798 	kfree(new);
799 	return res;
800 }
801 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
802 
803 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
804 {
805 	struct nfs_open_context *ctx = l_ctx->open_context;
806 	struct inode *inode = d_inode(ctx->dentry);
807 
808 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
809 		return;
810 	list_del(&l_ctx->list);
811 	spin_unlock(&inode->i_lock);
812 	kfree(l_ctx);
813 }
814 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
815 
816 /**
817  * nfs_close_context - Common close_context() routine NFSv2/v3
818  * @ctx: pointer to context
819  * @is_sync: is this a synchronous close
820  *
821  * Ensure that the attributes are up to date if we're mounted
822  * with close-to-open semantics and we have cached data that will
823  * need to be revalidated on open.
824  */
825 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
826 {
827 	struct nfs_inode *nfsi;
828 	struct inode *inode;
829 	struct nfs_server *server;
830 
831 	if (!(ctx->mode & FMODE_WRITE))
832 		return;
833 	if (!is_sync)
834 		return;
835 	inode = d_inode(ctx->dentry);
836 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
837 		return;
838 	nfsi = NFS_I(inode);
839 	if (inode->i_mapping->nrpages == 0)
840 		return;
841 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
842 		return;
843 	if (!list_empty(&nfsi->open_files))
844 		return;
845 	server = NFS_SERVER(inode);
846 	if (server->flags & NFS_MOUNT_NOCTO)
847 		return;
848 	nfs_revalidate_inode(server, inode);
849 }
850 EXPORT_SYMBOL_GPL(nfs_close_context);
851 
852 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
853 						fmode_t f_mode,
854 						struct file *filp)
855 {
856 	struct nfs_open_context *ctx;
857 	struct rpc_cred *cred = rpc_lookup_cred();
858 	if (IS_ERR(cred))
859 		return ERR_CAST(cred);
860 
861 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
862 	if (!ctx) {
863 		put_rpccred(cred);
864 		return ERR_PTR(-ENOMEM);
865 	}
866 	nfs_sb_active(dentry->d_sb);
867 	ctx->dentry = dget(dentry);
868 	ctx->cred = cred;
869 	ctx->state = NULL;
870 	ctx->mode = f_mode;
871 	ctx->flags = 0;
872 	ctx->error = 0;
873 	ctx->flock_owner = (fl_owner_t)filp;
874 	nfs_init_lock_context(&ctx->lock_context);
875 	ctx->lock_context.open_context = ctx;
876 	INIT_LIST_HEAD(&ctx->list);
877 	ctx->mdsthreshold = NULL;
878 	return ctx;
879 }
880 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
881 
882 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
883 {
884 	if (ctx != NULL)
885 		atomic_inc(&ctx->lock_context.count);
886 	return ctx;
887 }
888 EXPORT_SYMBOL_GPL(get_nfs_open_context);
889 
890 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
891 {
892 	struct inode *inode = d_inode(ctx->dentry);
893 	struct super_block *sb = ctx->dentry->d_sb;
894 
895 	if (!list_empty(&ctx->list)) {
896 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
897 			return;
898 		list_del(&ctx->list);
899 		spin_unlock(&inode->i_lock);
900 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
901 		return;
902 	if (inode != NULL)
903 		NFS_PROTO(inode)->close_context(ctx, is_sync);
904 	if (ctx->cred != NULL)
905 		put_rpccred(ctx->cred);
906 	dput(ctx->dentry);
907 	nfs_sb_deactive(sb);
908 	kfree(ctx->mdsthreshold);
909 	kfree(ctx);
910 }
911 
912 void put_nfs_open_context(struct nfs_open_context *ctx)
913 {
914 	__put_nfs_open_context(ctx, 0);
915 }
916 EXPORT_SYMBOL_GPL(put_nfs_open_context);
917 
918 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
919 {
920 	__put_nfs_open_context(ctx, 1);
921 }
922 
923 /*
924  * Ensure that mmap has a recent RPC credential for use when writing out
925  * shared pages
926  */
927 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
928 {
929 	struct inode *inode = d_inode(ctx->dentry);
930 	struct nfs_inode *nfsi = NFS_I(inode);
931 
932 	spin_lock(&inode->i_lock);
933 	if (ctx->mode & FMODE_WRITE)
934 		list_add(&ctx->list, &nfsi->open_files);
935 	else
936 		list_add_tail(&ctx->list, &nfsi->open_files);
937 	spin_unlock(&inode->i_lock);
938 }
939 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
940 
941 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
942 {
943 	filp->private_data = get_nfs_open_context(ctx);
944 	if (list_empty(&ctx->list))
945 		nfs_inode_attach_open_context(ctx);
946 }
947 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
948 
949 /*
950  * Given an inode, search for an open context with the desired characteristics
951  */
952 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
953 {
954 	struct nfs_inode *nfsi = NFS_I(inode);
955 	struct nfs_open_context *pos, *ctx = NULL;
956 
957 	spin_lock(&inode->i_lock);
958 	list_for_each_entry(pos, &nfsi->open_files, list) {
959 		if (cred != NULL && pos->cred != cred)
960 			continue;
961 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
962 			continue;
963 		ctx = get_nfs_open_context(pos);
964 		break;
965 	}
966 	spin_unlock(&inode->i_lock);
967 	return ctx;
968 }
969 
970 void nfs_file_clear_open_context(struct file *filp)
971 {
972 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
973 
974 	if (ctx) {
975 		struct inode *inode = d_inode(ctx->dentry);
976 
977 		/*
978 		 * We fatal error on write before. Try to writeback
979 		 * every page again.
980 		 */
981 		if (ctx->error < 0)
982 			invalidate_inode_pages2(inode->i_mapping);
983 		filp->private_data = NULL;
984 		spin_lock(&inode->i_lock);
985 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
986 		spin_unlock(&inode->i_lock);
987 		put_nfs_open_context_sync(ctx);
988 	}
989 }
990 
991 /*
992  * These allocate and release file read/write context information.
993  */
994 int nfs_open(struct inode *inode, struct file *filp)
995 {
996 	struct nfs_open_context *ctx;
997 
998 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
999 	if (IS_ERR(ctx))
1000 		return PTR_ERR(ctx);
1001 	nfs_file_set_open_context(filp, ctx);
1002 	put_nfs_open_context(ctx);
1003 	nfs_fscache_open_file(inode, filp);
1004 	return 0;
1005 }
1006 
1007 /*
1008  * This function is called whenever some part of NFS notices that
1009  * the cached attributes have to be refreshed.
1010  */
1011 int
1012 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1013 {
1014 	int		 status = -ESTALE;
1015 	struct nfs4_label *label = NULL;
1016 	struct nfs_fattr *fattr = NULL;
1017 	struct nfs_inode *nfsi = NFS_I(inode);
1018 
1019 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1020 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1021 
1022 	trace_nfs_revalidate_inode_enter(inode);
1023 
1024 	if (is_bad_inode(inode))
1025 		goto out;
1026 	if (NFS_STALE(inode))
1027 		goto out;
1028 
1029 	/* pNFS: Attributes aren't updated until we layoutcommit */
1030 	if (S_ISREG(inode->i_mode)) {
1031 		status = pnfs_sync_inode(inode, false);
1032 		if (status)
1033 			goto out;
1034 	}
1035 
1036 	status = -ENOMEM;
1037 	fattr = nfs_alloc_fattr();
1038 	if (fattr == NULL)
1039 		goto out;
1040 
1041 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1042 
1043 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1044 	if (IS_ERR(label)) {
1045 		status = PTR_ERR(label);
1046 		goto out;
1047 	}
1048 
1049 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
1050 	if (status != 0) {
1051 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1052 			 inode->i_sb->s_id,
1053 			 (unsigned long long)NFS_FILEID(inode), status);
1054 		if (status == -ESTALE) {
1055 			nfs_zap_caches(inode);
1056 			if (!S_ISDIR(inode->i_mode))
1057 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1058 		}
1059 		goto err_out;
1060 	}
1061 
1062 	status = nfs_refresh_inode(inode, fattr);
1063 	if (status) {
1064 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1065 			 inode->i_sb->s_id,
1066 			 (unsigned long long)NFS_FILEID(inode), status);
1067 		goto err_out;
1068 	}
1069 
1070 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1071 		nfs_zap_acl_cache(inode);
1072 
1073 	nfs_setsecurity(inode, fattr, label);
1074 
1075 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1076 		inode->i_sb->s_id,
1077 		(unsigned long long)NFS_FILEID(inode));
1078 
1079 err_out:
1080 	nfs4_label_free(label);
1081 out:
1082 	nfs_free_fattr(fattr);
1083 	trace_nfs_revalidate_inode_exit(inode, status);
1084 	return status;
1085 }
1086 
1087 int nfs_attribute_cache_expired(struct inode *inode)
1088 {
1089 	if (nfs_have_delegated_attributes(inode))
1090 		return 0;
1091 	return nfs_attribute_timeout(inode);
1092 }
1093 
1094 /**
1095  * nfs_revalidate_inode - Revalidate the inode attributes
1096  * @server - pointer to nfs_server struct
1097  * @inode - pointer to inode struct
1098  *
1099  * Updates inode attribute information by retrieving the data from the server.
1100  */
1101 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1102 {
1103 	if (!nfs_need_revalidate_inode(inode))
1104 		return NFS_STALE(inode) ? -ESTALE : 0;
1105 	return __nfs_revalidate_inode(server, inode);
1106 }
1107 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1108 
1109 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1110 {
1111 	struct nfs_inode *nfsi = NFS_I(inode);
1112 	int ret;
1113 
1114 	if (mapping->nrpages != 0) {
1115 		if (S_ISREG(inode->i_mode)) {
1116 			unmap_mapping_range(mapping, 0, 0, 0);
1117 			ret = nfs_sync_mapping(mapping);
1118 			if (ret < 0)
1119 				return ret;
1120 		}
1121 		ret = invalidate_inode_pages2(mapping);
1122 		if (ret < 0)
1123 			return ret;
1124 	}
1125 	if (S_ISDIR(inode->i_mode)) {
1126 		spin_lock(&inode->i_lock);
1127 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1128 		spin_unlock(&inode->i_lock);
1129 	}
1130 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1131 	nfs_fscache_wait_on_invalidate(inode);
1132 
1133 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1134 			inode->i_sb->s_id,
1135 			(unsigned long long)NFS_FILEID(inode));
1136 	return 0;
1137 }
1138 
1139 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1140 {
1141 	return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1142 		NFS_STALE(inode);
1143 }
1144 
1145 int nfs_revalidate_mapping_rcu(struct inode *inode)
1146 {
1147 	struct nfs_inode *nfsi = NFS_I(inode);
1148 	unsigned long *bitlock = &nfsi->flags;
1149 	int ret = 0;
1150 
1151 	if (IS_SWAPFILE(inode))
1152 		goto out;
1153 	if (nfs_mapping_need_revalidate_inode(inode)) {
1154 		ret = -ECHILD;
1155 		goto out;
1156 	}
1157 	spin_lock(&inode->i_lock);
1158 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1159 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1160 		ret = -ECHILD;
1161 	spin_unlock(&inode->i_lock);
1162 out:
1163 	return ret;
1164 }
1165 
1166 /**
1167  * nfs_revalidate_mapping - Revalidate the pagecache
1168  * @inode - pointer to host inode
1169  * @mapping - pointer to mapping
1170  */
1171 int nfs_revalidate_mapping(struct inode *inode,
1172 		struct address_space *mapping)
1173 {
1174 	struct nfs_inode *nfsi = NFS_I(inode);
1175 	unsigned long *bitlock = &nfsi->flags;
1176 	int ret = 0;
1177 
1178 	/* swapfiles are not supposed to be shared. */
1179 	if (IS_SWAPFILE(inode))
1180 		goto out;
1181 
1182 	if (nfs_mapping_need_revalidate_inode(inode)) {
1183 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1184 		if (ret < 0)
1185 			goto out;
1186 	}
1187 
1188 	/*
1189 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1190 	 * invalidations that come in while we're shooting down the mappings
1191 	 * are respected. But, that leaves a race window where one revalidator
1192 	 * can clear the flag, and then another checks it before the mapping
1193 	 * gets invalidated. Fix that by serializing access to this part of
1194 	 * the function.
1195 	 *
1196 	 * At the same time, we need to allow other tasks to see whether we
1197 	 * might be in the middle of invalidating the pages, so we only set
1198 	 * the bit lock here if it looks like we're going to be doing that.
1199 	 */
1200 	for (;;) {
1201 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1202 					 nfs_wait_bit_killable, TASK_KILLABLE);
1203 		if (ret)
1204 			goto out;
1205 		spin_lock(&inode->i_lock);
1206 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1207 			spin_unlock(&inode->i_lock);
1208 			continue;
1209 		}
1210 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1211 			break;
1212 		spin_unlock(&inode->i_lock);
1213 		goto out;
1214 	}
1215 
1216 	set_bit(NFS_INO_INVALIDATING, bitlock);
1217 	smp_wmb();
1218 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1219 	spin_unlock(&inode->i_lock);
1220 	trace_nfs_invalidate_mapping_enter(inode);
1221 	ret = nfs_invalidate_mapping(inode, mapping);
1222 	trace_nfs_invalidate_mapping_exit(inode, ret);
1223 
1224 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1225 	smp_mb__after_atomic();
1226 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1227 out:
1228 	return ret;
1229 }
1230 
1231 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1232 {
1233 	struct inode *inode = &nfsi->vfs_inode;
1234 
1235 	assert_spin_locked(&inode->i_lock);
1236 
1237 	if (!S_ISREG(inode->i_mode))
1238 		return false;
1239 	if (list_empty(&nfsi->open_files))
1240 		return false;
1241 	/* Note: This relies on nfsi->open_files being ordered with writers
1242 	 *       being placed at the head of the list.
1243 	 *       See nfs_inode_attach_open_context()
1244 	 */
1245 	return (list_first_entry(&nfsi->open_files,
1246 			struct nfs_open_context,
1247 			list)->mode & FMODE_WRITE) == FMODE_WRITE;
1248 }
1249 
1250 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1251 {
1252 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1253 }
1254 
1255 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1256 {
1257 	struct nfs_inode *nfsi = NFS_I(inode);
1258 	unsigned long ret = 0;
1259 
1260 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1261 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1262 			&& inode->i_version == fattr->pre_change_attr) {
1263 		inode->i_version = fattr->change_attr;
1264 		if (S_ISDIR(inode->i_mode))
1265 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1266 		ret |= NFS_INO_INVALID_ATTR;
1267 	}
1268 	/* If we have atomic WCC data, we may update some attributes */
1269 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1270 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1271 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1272 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1273 		ret |= NFS_INO_INVALID_ATTR;
1274 	}
1275 
1276 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1277 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1278 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1279 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1280 		if (S_ISDIR(inode->i_mode))
1281 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1282 		ret |= NFS_INO_INVALID_ATTR;
1283 	}
1284 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1285 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1286 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1287 			&& nfsi->nrequests == 0) {
1288 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1289 		ret |= NFS_INO_INVALID_ATTR;
1290 	}
1291 
1292 	return ret;
1293 }
1294 
1295 /**
1296  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1297  * @inode - pointer to inode
1298  * @fattr - updated attributes
1299  *
1300  * Verifies the attribute cache. If we have just changed the attributes,
1301  * so that fattr carries weak cache consistency data, then it may
1302  * also update the ctime/mtime/change_attribute.
1303  */
1304 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1305 {
1306 	struct nfs_inode *nfsi = NFS_I(inode);
1307 	loff_t cur_size, new_isize;
1308 	unsigned long invalid = 0;
1309 
1310 
1311 	if (nfs_have_delegated_attributes(inode))
1312 		return 0;
1313 	/* Has the inode gone and changed behind our back? */
1314 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1315 		return -EIO;
1316 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1317 		return -EIO;
1318 
1319 	if (!nfs_file_has_buffered_writers(nfsi)) {
1320 		/* Verify a few of the more important attributes */
1321 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && inode->i_version != fattr->change_attr)
1322 			invalid |= NFS_INO_INVALID_ATTR | NFS_INO_REVAL_PAGECACHE;
1323 
1324 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1325 			invalid |= NFS_INO_INVALID_ATTR;
1326 
1327 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&inode->i_ctime, &fattr->ctime))
1328 			invalid |= NFS_INO_INVALID_ATTR;
1329 
1330 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1331 			cur_size = i_size_read(inode);
1332 			new_isize = nfs_size_to_loff_t(fattr->size);
1333 			if (cur_size != new_isize)
1334 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1335 		}
1336 	}
1337 
1338 	/* Have any file permissions changed? */
1339 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1340 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1341 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1342 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1343 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1344 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1345 
1346 	/* Has the link count changed? */
1347 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1348 		invalid |= NFS_INO_INVALID_ATTR;
1349 
1350 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1351 		invalid |= NFS_INO_INVALID_ATIME;
1352 
1353 	if (invalid != 0)
1354 		nfs_set_cache_invalid(inode, invalid | NFS_INO_REVAL_FORCED);
1355 
1356 	nfsi->read_cache_jiffies = fattr->time_start;
1357 	return 0;
1358 }
1359 
1360 static atomic_long_t nfs_attr_generation_counter;
1361 
1362 static unsigned long nfs_read_attr_generation_counter(void)
1363 {
1364 	return atomic_long_read(&nfs_attr_generation_counter);
1365 }
1366 
1367 unsigned long nfs_inc_attr_generation_counter(void)
1368 {
1369 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1370 }
1371 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1372 
1373 void nfs_fattr_init(struct nfs_fattr *fattr)
1374 {
1375 	fattr->valid = 0;
1376 	fattr->time_start = jiffies;
1377 	fattr->gencount = nfs_inc_attr_generation_counter();
1378 	fattr->owner_name = NULL;
1379 	fattr->group_name = NULL;
1380 }
1381 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1382 
1383 /**
1384  * nfs_fattr_set_barrier
1385  * @fattr: attributes
1386  *
1387  * Used to set a barrier after an attribute was updated. This
1388  * barrier ensures that older attributes from RPC calls that may
1389  * have raced with our update cannot clobber these new values.
1390  * Note that you are still responsible for ensuring that other
1391  * operations which change the attribute on the server do not
1392  * collide.
1393  */
1394 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1395 {
1396 	fattr->gencount = nfs_inc_attr_generation_counter();
1397 }
1398 
1399 struct nfs_fattr *nfs_alloc_fattr(void)
1400 {
1401 	struct nfs_fattr *fattr;
1402 
1403 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1404 	if (fattr != NULL)
1405 		nfs_fattr_init(fattr);
1406 	return fattr;
1407 }
1408 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1409 
1410 struct nfs_fh *nfs_alloc_fhandle(void)
1411 {
1412 	struct nfs_fh *fh;
1413 
1414 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1415 	if (fh != NULL)
1416 		fh->size = 0;
1417 	return fh;
1418 }
1419 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1420 
1421 #ifdef NFS_DEBUG
1422 /*
1423  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1424  *                             in the same way that wireshark does
1425  *
1426  * @fh: file handle
1427  *
1428  * For debugging only.
1429  */
1430 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1431 {
1432 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1433 	 * not on the result */
1434 	return nfs_fhandle_hash(fh);
1435 }
1436 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1437 
1438 /*
1439  * _nfs_display_fhandle - display an NFS file handle on the console
1440  *
1441  * @fh: file handle to display
1442  * @caption: display caption
1443  *
1444  * For debugging only.
1445  */
1446 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1447 {
1448 	unsigned short i;
1449 
1450 	if (fh == NULL || fh->size == 0) {
1451 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1452 		return;
1453 	}
1454 
1455 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1456 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1457 	for (i = 0; i < fh->size; i += 16) {
1458 		__be32 *pos = (__be32 *)&fh->data[i];
1459 
1460 		switch ((fh->size - i - 1) >> 2) {
1461 		case 0:
1462 			printk(KERN_DEFAULT " %08x\n",
1463 				be32_to_cpup(pos));
1464 			break;
1465 		case 1:
1466 			printk(KERN_DEFAULT " %08x %08x\n",
1467 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1468 			break;
1469 		case 2:
1470 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1471 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1472 				be32_to_cpup(pos + 2));
1473 			break;
1474 		default:
1475 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1476 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1477 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1478 		}
1479 	}
1480 }
1481 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1482 #endif
1483 
1484 /**
1485  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1486  * @inode - pointer to inode
1487  * @fattr - attributes
1488  *
1489  * Attempt to divine whether or not an RPC call reply carrying stale
1490  * attributes got scheduled after another call carrying updated ones.
1491  *
1492  * To do so, the function first assumes that a more recent ctime means
1493  * that the attributes in fattr are newer, however it also attempt to
1494  * catch the case where ctime either didn't change, or went backwards
1495  * (if someone reset the clock on the server) by looking at whether
1496  * or not this RPC call was started after the inode was last updated.
1497  * Note also the check for wraparound of 'attr_gencount'
1498  *
1499  * The function returns 'true' if it thinks the attributes in 'fattr' are
1500  * more recent than the ones cached in the inode.
1501  *
1502  */
1503 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1504 {
1505 	const struct nfs_inode *nfsi = NFS_I(inode);
1506 
1507 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1508 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1509 }
1510 
1511 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1512 {
1513 	int ret;
1514 
1515 	trace_nfs_refresh_inode_enter(inode);
1516 
1517 	if (nfs_inode_attrs_need_update(inode, fattr))
1518 		ret = nfs_update_inode(inode, fattr);
1519 	else
1520 		ret = nfs_check_inode_attributes(inode, fattr);
1521 
1522 	trace_nfs_refresh_inode_exit(inode, ret);
1523 	return ret;
1524 }
1525 
1526 /**
1527  * nfs_refresh_inode - try to update the inode attribute cache
1528  * @inode - pointer to inode
1529  * @fattr - updated attributes
1530  *
1531  * Check that an RPC call that returned attributes has not overlapped with
1532  * other recent updates of the inode metadata, then decide whether it is
1533  * safe to do a full update of the inode attributes, or whether just to
1534  * call nfs_check_inode_attributes.
1535  */
1536 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1537 {
1538 	int status;
1539 
1540 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1541 		return 0;
1542 	spin_lock(&inode->i_lock);
1543 	status = nfs_refresh_inode_locked(inode, fattr);
1544 	spin_unlock(&inode->i_lock);
1545 
1546 	return status;
1547 }
1548 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1549 
1550 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1551 {
1552 	unsigned long invalid = NFS_INO_INVALID_ATTR;
1553 
1554 	if (S_ISDIR(inode->i_mode))
1555 		invalid |= NFS_INO_INVALID_DATA;
1556 	nfs_set_cache_invalid(inode, invalid);
1557 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1558 		return 0;
1559 	return nfs_refresh_inode_locked(inode, fattr);
1560 }
1561 
1562 /**
1563  * nfs_post_op_update_inode - try to update the inode attribute cache
1564  * @inode - pointer to inode
1565  * @fattr - updated attributes
1566  *
1567  * After an operation that has changed the inode metadata, mark the
1568  * attribute cache as being invalid, then try to update it.
1569  *
1570  * NB: if the server didn't return any post op attributes, this
1571  * function will force the retrieval of attributes before the next
1572  * NFS request.  Thus it should be used only for operations that
1573  * are expected to change one or more attributes, to avoid
1574  * unnecessary NFS requests and trips through nfs_update_inode().
1575  */
1576 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1577 {
1578 	int status;
1579 
1580 	spin_lock(&inode->i_lock);
1581 	nfs_fattr_set_barrier(fattr);
1582 	status = nfs_post_op_update_inode_locked(inode, fattr);
1583 	spin_unlock(&inode->i_lock);
1584 
1585 	return status;
1586 }
1587 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1588 
1589 /**
1590  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1591  * @inode - pointer to inode
1592  * @fattr - updated attributes
1593  *
1594  * After an operation that has changed the inode metadata, mark the
1595  * attribute cache as being invalid, then try to update it. Fake up
1596  * weak cache consistency data, if none exist.
1597  *
1598  * This function is mainly designed to be used by the ->write_done() functions.
1599  */
1600 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1601 {
1602 	int status;
1603 
1604 	/* Don't do a WCC update if these attributes are already stale */
1605 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1606 			!nfs_inode_attrs_need_update(inode, fattr)) {
1607 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1608 				| NFS_ATTR_FATTR_PRESIZE
1609 				| NFS_ATTR_FATTR_PREMTIME
1610 				| NFS_ATTR_FATTR_PRECTIME);
1611 		goto out_noforce;
1612 	}
1613 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1614 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1615 		fattr->pre_change_attr = inode->i_version;
1616 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1617 	}
1618 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1619 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1620 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1621 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1622 	}
1623 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1624 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1625 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1626 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1627 	}
1628 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1629 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1630 		fattr->pre_size = i_size_read(inode);
1631 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1632 	}
1633 out_noforce:
1634 	status = nfs_post_op_update_inode_locked(inode, fattr);
1635 	return status;
1636 }
1637 
1638 /**
1639  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1640  * @inode - pointer to inode
1641  * @fattr - updated attributes
1642  *
1643  * After an operation that has changed the inode metadata, mark the
1644  * attribute cache as being invalid, then try to update it. Fake up
1645  * weak cache consistency data, if none exist.
1646  *
1647  * This function is mainly designed to be used by the ->write_done() functions.
1648  */
1649 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1650 {
1651 	int status;
1652 
1653 	spin_lock(&inode->i_lock);
1654 	nfs_fattr_set_barrier(fattr);
1655 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1656 	spin_unlock(&inode->i_lock);
1657 	return status;
1658 }
1659 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1660 
1661 
1662 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1663 				    struct nfs_fattr *fattr)
1664 {
1665 	bool ret1 = true, ret2 = true;
1666 
1667 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1668 		ret1 = (nfsi->fileid == fattr->fileid);
1669 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1670 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1671 	return ret1 || ret2;
1672 }
1673 
1674 /*
1675  * Many nfs protocol calls return the new file attributes after
1676  * an operation.  Here we update the inode to reflect the state
1677  * of the server's inode.
1678  *
1679  * This is a bit tricky because we have to make sure all dirty pages
1680  * have been sent off to the server before calling invalidate_inode_pages.
1681  * To make sure no other process adds more write requests while we try
1682  * our best to flush them, we make them sleep during the attribute refresh.
1683  *
1684  * A very similar scenario holds for the dir cache.
1685  */
1686 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1687 {
1688 	struct nfs_server *server;
1689 	struct nfs_inode *nfsi = NFS_I(inode);
1690 	loff_t cur_isize, new_isize;
1691 	unsigned long invalid = 0;
1692 	unsigned long now = jiffies;
1693 	unsigned long save_cache_validity;
1694 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
1695 	bool cache_revalidated = true;
1696 
1697 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1698 			__func__, inode->i_sb->s_id, inode->i_ino,
1699 			nfs_display_fhandle_hash(NFS_FH(inode)),
1700 			atomic_read(&inode->i_count), fattr->valid);
1701 
1702 	if (!nfs_fileid_valid(nfsi, fattr)) {
1703 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1704 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1705 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1706 			inode->i_sb->s_id, (long long)nfsi->fileid,
1707 			(long long)fattr->fileid);
1708 		goto out_err;
1709 	}
1710 
1711 	/*
1712 	 * Make sure the inode's type hasn't changed.
1713 	 */
1714 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1715 		/*
1716 		* Big trouble! The inode has become a different object.
1717 		*/
1718 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1719 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1720 		goto out_err;
1721 	}
1722 
1723 	server = NFS_SERVER(inode);
1724 	/* Update the fsid? */
1725 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1726 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1727 			!IS_AUTOMOUNT(inode))
1728 		server->fsid = fattr->fsid;
1729 
1730 	/*
1731 	 * Update the read time so we don't revalidate too often.
1732 	 */
1733 	nfsi->read_cache_jiffies = fattr->time_start;
1734 
1735 	save_cache_validity = nfsi->cache_validity;
1736 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1737 			| NFS_INO_INVALID_ATIME
1738 			| NFS_INO_REVAL_FORCED
1739 			| NFS_INO_REVAL_PAGECACHE);
1740 
1741 	/* Do atomic weak cache consistency updates */
1742 	invalid |= nfs_wcc_update_inode(inode, fattr);
1743 
1744 	if (pnfs_layoutcommit_outstanding(inode)) {
1745 		nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1746 		cache_revalidated = false;
1747 	}
1748 
1749 	/* More cache consistency checks */
1750 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1751 		if (inode->i_version != fattr->change_attr) {
1752 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1753 					inode->i_sb->s_id, inode->i_ino);
1754 			/* Could it be a race with writeback? */
1755 			if (!have_writers) {
1756 				invalid |= NFS_INO_INVALID_ATTR
1757 					| NFS_INO_INVALID_DATA
1758 					| NFS_INO_INVALID_ACCESS
1759 					| NFS_INO_INVALID_ACL;
1760 				if (S_ISDIR(inode->i_mode))
1761 					nfs_force_lookup_revalidate(inode);
1762 			}
1763 			inode->i_version = fattr->change_attr;
1764 		}
1765 	} else {
1766 		nfsi->cache_validity |= save_cache_validity;
1767 		cache_revalidated = false;
1768 	}
1769 
1770 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1771 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1772 	} else if (server->caps & NFS_CAP_MTIME) {
1773 		nfsi->cache_validity |= save_cache_validity &
1774 				(NFS_INO_INVALID_ATTR
1775 				| NFS_INO_REVAL_FORCED);
1776 		cache_revalidated = false;
1777 	}
1778 
1779 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1780 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1781 	} else if (server->caps & NFS_CAP_CTIME) {
1782 		nfsi->cache_validity |= save_cache_validity &
1783 				(NFS_INO_INVALID_ATTR
1784 				| NFS_INO_REVAL_FORCED);
1785 		cache_revalidated = false;
1786 	}
1787 
1788 	/* Check if our cached file size is stale */
1789 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1790 		new_isize = nfs_size_to_loff_t(fattr->size);
1791 		cur_isize = i_size_read(inode);
1792 		if (new_isize != cur_isize) {
1793 			/* Do we perhaps have any outstanding writes, or has
1794 			 * the file grown beyond our last write? */
1795 			if (nfsi->nrequests == 0 || new_isize > cur_isize) {
1796 				i_size_write(inode, new_isize);
1797 				if (!have_writers)
1798 					invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1799 			}
1800 			dprintk("NFS: isize change on server for file %s/%ld "
1801 					"(%Ld to %Ld)\n",
1802 					inode->i_sb->s_id,
1803 					inode->i_ino,
1804 					(long long)cur_isize,
1805 					(long long)new_isize);
1806 		}
1807 	} else {
1808 		nfsi->cache_validity |= save_cache_validity &
1809 				(NFS_INO_INVALID_ATTR
1810 				| NFS_INO_REVAL_PAGECACHE
1811 				| NFS_INO_REVAL_FORCED);
1812 		cache_revalidated = false;
1813 	}
1814 
1815 
1816 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1817 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1818 	else if (server->caps & NFS_CAP_ATIME) {
1819 		nfsi->cache_validity |= save_cache_validity &
1820 				(NFS_INO_INVALID_ATIME
1821 				| NFS_INO_REVAL_FORCED);
1822 		cache_revalidated = false;
1823 	}
1824 
1825 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1826 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1827 			umode_t newmode = inode->i_mode & S_IFMT;
1828 			newmode |= fattr->mode & S_IALLUGO;
1829 			inode->i_mode = newmode;
1830 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1831 		}
1832 	} else if (server->caps & NFS_CAP_MODE) {
1833 		nfsi->cache_validity |= save_cache_validity &
1834 				(NFS_INO_INVALID_ATTR
1835 				| NFS_INO_INVALID_ACCESS
1836 				| NFS_INO_INVALID_ACL
1837 				| NFS_INO_REVAL_FORCED);
1838 		cache_revalidated = false;
1839 	}
1840 
1841 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1842 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1843 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1844 			inode->i_uid = fattr->uid;
1845 		}
1846 	} else if (server->caps & NFS_CAP_OWNER) {
1847 		nfsi->cache_validity |= save_cache_validity &
1848 				(NFS_INO_INVALID_ATTR
1849 				| NFS_INO_INVALID_ACCESS
1850 				| NFS_INO_INVALID_ACL
1851 				| NFS_INO_REVAL_FORCED);
1852 		cache_revalidated = false;
1853 	}
1854 
1855 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1856 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1857 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1858 			inode->i_gid = fattr->gid;
1859 		}
1860 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
1861 		nfsi->cache_validity |= save_cache_validity &
1862 				(NFS_INO_INVALID_ATTR
1863 				| NFS_INO_INVALID_ACCESS
1864 				| NFS_INO_INVALID_ACL
1865 				| NFS_INO_REVAL_FORCED);
1866 		cache_revalidated = false;
1867 	}
1868 
1869 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1870 		if (inode->i_nlink != fattr->nlink) {
1871 			invalid |= NFS_INO_INVALID_ATTR;
1872 			if (S_ISDIR(inode->i_mode))
1873 				invalid |= NFS_INO_INVALID_DATA;
1874 			set_nlink(inode, fattr->nlink);
1875 		}
1876 	} else if (server->caps & NFS_CAP_NLINK) {
1877 		nfsi->cache_validity |= save_cache_validity &
1878 				(NFS_INO_INVALID_ATTR
1879 				| NFS_INO_REVAL_FORCED);
1880 		cache_revalidated = false;
1881 	}
1882 
1883 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1884 		/*
1885 		 * report the blocks in 512byte units
1886 		 */
1887 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1888 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1889 		inode->i_blocks = fattr->du.nfs2.blocks;
1890 	else
1891 		cache_revalidated = false;
1892 
1893 	/* Update attrtimeo value if we're out of the unstable period */
1894 	if (invalid & NFS_INO_INVALID_ATTR) {
1895 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1896 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1897 		nfsi->attrtimeo_timestamp = now;
1898 		/* Set barrier to be more recent than all outstanding updates */
1899 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1900 	} else {
1901 		if (cache_revalidated) {
1902 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1903 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1904 				nfsi->attrtimeo <<= 1;
1905 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1906 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1907 			}
1908 			nfsi->attrtimeo_timestamp = now;
1909 		}
1910 		/* Set the barrier to be more recent than this fattr */
1911 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1912 			nfsi->attr_gencount = fattr->gencount;
1913 	}
1914 
1915 	/* Don't declare attrcache up to date if there were no attrs! */
1916 	if (cache_revalidated)
1917 		invalid &= ~NFS_INO_INVALID_ATTR;
1918 
1919 	/* Don't invalidate the data if we were to blame */
1920 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1921 				|| S_ISLNK(inode->i_mode)))
1922 		invalid &= ~NFS_INO_INVALID_DATA;
1923 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1924 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1925 		nfs_set_cache_invalid(inode, invalid);
1926 
1927 	return 0;
1928  out_err:
1929 	/*
1930 	 * No need to worry about unhashing the dentry, as the
1931 	 * lookup validation will know that the inode is bad.
1932 	 * (But we fall through to invalidate the caches.)
1933 	 */
1934 	nfs_invalidate_inode(inode);
1935 	return -ESTALE;
1936 }
1937 
1938 struct inode *nfs_alloc_inode(struct super_block *sb)
1939 {
1940 	struct nfs_inode *nfsi;
1941 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1942 	if (!nfsi)
1943 		return NULL;
1944 	nfsi->flags = 0UL;
1945 	nfsi->cache_validity = 0UL;
1946 #if IS_ENABLED(CONFIG_NFS_V4)
1947 	nfsi->nfs4_acl = NULL;
1948 #endif /* CONFIG_NFS_V4 */
1949 	return &nfsi->vfs_inode;
1950 }
1951 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1952 
1953 static void nfs_i_callback(struct rcu_head *head)
1954 {
1955 	struct inode *inode = container_of(head, struct inode, i_rcu);
1956 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1957 }
1958 
1959 void nfs_destroy_inode(struct inode *inode)
1960 {
1961 	call_rcu(&inode->i_rcu, nfs_i_callback);
1962 }
1963 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1964 
1965 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1966 {
1967 #if IS_ENABLED(CONFIG_NFS_V4)
1968 	INIT_LIST_HEAD(&nfsi->open_states);
1969 	nfsi->delegation = NULL;
1970 	init_rwsem(&nfsi->rwsem);
1971 	nfsi->layout = NULL;
1972 #endif
1973 }
1974 
1975 static void init_once(void *foo)
1976 {
1977 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1978 
1979 	inode_init_once(&nfsi->vfs_inode);
1980 	INIT_LIST_HEAD(&nfsi->open_files);
1981 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1982 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1983 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1984 	nfsi->nrequests = 0;
1985 	nfsi->commit_info.ncommit = 0;
1986 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1987 	init_rwsem(&nfsi->rmdir_sem);
1988 	nfs4_init_once(nfsi);
1989 }
1990 
1991 static int __init nfs_init_inodecache(void)
1992 {
1993 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1994 					     sizeof(struct nfs_inode),
1995 					     0, (SLAB_RECLAIM_ACCOUNT|
1996 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1997 					     init_once);
1998 	if (nfs_inode_cachep == NULL)
1999 		return -ENOMEM;
2000 
2001 	return 0;
2002 }
2003 
2004 static void nfs_destroy_inodecache(void)
2005 {
2006 	/*
2007 	 * Make sure all delayed rcu free inodes are flushed before we
2008 	 * destroy cache.
2009 	 */
2010 	rcu_barrier();
2011 	kmem_cache_destroy(nfs_inode_cachep);
2012 }
2013 
2014 struct workqueue_struct *nfsiod_workqueue;
2015 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2016 
2017 /*
2018  * start up the nfsiod workqueue
2019  */
2020 static int nfsiod_start(void)
2021 {
2022 	struct workqueue_struct *wq;
2023 	dprintk("RPC:       creating workqueue nfsiod\n");
2024 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2025 	if (wq == NULL)
2026 		return -ENOMEM;
2027 	nfsiod_workqueue = wq;
2028 	return 0;
2029 }
2030 
2031 /*
2032  * Destroy the nfsiod workqueue
2033  */
2034 static void nfsiod_stop(void)
2035 {
2036 	struct workqueue_struct *wq;
2037 
2038 	wq = nfsiod_workqueue;
2039 	if (wq == NULL)
2040 		return;
2041 	nfsiod_workqueue = NULL;
2042 	destroy_workqueue(wq);
2043 }
2044 
2045 unsigned int nfs_net_id;
2046 EXPORT_SYMBOL_GPL(nfs_net_id);
2047 
2048 static int nfs_net_init(struct net *net)
2049 {
2050 	nfs_clients_init(net);
2051 	return nfs_fs_proc_net_init(net);
2052 }
2053 
2054 static void nfs_net_exit(struct net *net)
2055 {
2056 	nfs_fs_proc_net_exit(net);
2057 	nfs_cleanup_cb_ident_idr(net);
2058 }
2059 
2060 static struct pernet_operations nfs_net_ops = {
2061 	.init = nfs_net_init,
2062 	.exit = nfs_net_exit,
2063 	.id   = &nfs_net_id,
2064 	.size = sizeof(struct nfs_net),
2065 };
2066 
2067 /*
2068  * Initialize NFS
2069  */
2070 static int __init init_nfs_fs(void)
2071 {
2072 	int err;
2073 
2074 	err = register_pernet_subsys(&nfs_net_ops);
2075 	if (err < 0)
2076 		goto out9;
2077 
2078 	err = nfs_fscache_register();
2079 	if (err < 0)
2080 		goto out8;
2081 
2082 	err = nfsiod_start();
2083 	if (err)
2084 		goto out7;
2085 
2086 	err = nfs_fs_proc_init();
2087 	if (err)
2088 		goto out6;
2089 
2090 	err = nfs_init_nfspagecache();
2091 	if (err)
2092 		goto out5;
2093 
2094 	err = nfs_init_inodecache();
2095 	if (err)
2096 		goto out4;
2097 
2098 	err = nfs_init_readpagecache();
2099 	if (err)
2100 		goto out3;
2101 
2102 	err = nfs_init_writepagecache();
2103 	if (err)
2104 		goto out2;
2105 
2106 	err = nfs_init_directcache();
2107 	if (err)
2108 		goto out1;
2109 
2110 	rpc_proc_register(&init_net, &nfs_rpcstat);
2111 
2112 	err = register_nfs_fs();
2113 	if (err)
2114 		goto out0;
2115 
2116 	return 0;
2117 out0:
2118 	rpc_proc_unregister(&init_net, "nfs");
2119 	nfs_destroy_directcache();
2120 out1:
2121 	nfs_destroy_writepagecache();
2122 out2:
2123 	nfs_destroy_readpagecache();
2124 out3:
2125 	nfs_destroy_inodecache();
2126 out4:
2127 	nfs_destroy_nfspagecache();
2128 out5:
2129 	nfs_fs_proc_exit();
2130 out6:
2131 	nfsiod_stop();
2132 out7:
2133 	nfs_fscache_unregister();
2134 out8:
2135 	unregister_pernet_subsys(&nfs_net_ops);
2136 out9:
2137 	return err;
2138 }
2139 
2140 static void __exit exit_nfs_fs(void)
2141 {
2142 	nfs_destroy_directcache();
2143 	nfs_destroy_writepagecache();
2144 	nfs_destroy_readpagecache();
2145 	nfs_destroy_inodecache();
2146 	nfs_destroy_nfspagecache();
2147 	nfs_fscache_unregister();
2148 	unregister_pernet_subsys(&nfs_net_ops);
2149 	rpc_proc_unregister(&init_net, "nfs");
2150 	unregister_nfs_fs();
2151 	nfs_fs_proc_exit();
2152 	nfsiod_stop();
2153 }
2154 
2155 /* Not quite true; I just maintain it */
2156 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2157 MODULE_LICENSE("GPL");
2158 module_param(enable_ino64, bool, 0644);
2159 
2160 module_init(init_nfs_fs)
2161 module_exit(exit_nfs_fs)
2162