xref: /openbmc/linux/fs/nfs/inode.c (revision 8730046c)
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 <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(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
707 {
708 	struct inode *inode = d_inode(dentry);
709 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
710 	int err = 0;
711 
712 	trace_nfs_getattr_enter(inode);
713 	/* Flush out writes to the server in order to update c/mtime.  */
714 	if (S_ISREG(inode->i_mode)) {
715 		err = filemap_write_and_wait(inode->i_mapping);
716 		if (err)
717 			goto out;
718 	}
719 
720 	/*
721 	 * We may force a getattr if the user cares about atime.
722 	 *
723 	 * Note that we only have to check the vfsmount flags here:
724 	 *  - NFS always sets S_NOATIME by so checking it would give a
725 	 *    bogus result
726 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
727 	 *    no point in checking those.
728 	 */
729  	if ((mnt->mnt_flags & MNT_NOATIME) ||
730  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
731 		need_atime = 0;
732 
733 	if (need_atime || nfs_need_revalidate_inode(inode)) {
734 		struct nfs_server *server = NFS_SERVER(inode);
735 
736 		nfs_readdirplus_parent_cache_miss(dentry);
737 		err = __nfs_revalidate_inode(server, inode);
738 	} else
739 		nfs_readdirplus_parent_cache_hit(dentry);
740 	if (!err) {
741 		generic_fillattr(inode, stat);
742 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
743 		if (S_ISDIR(inode->i_mode))
744 			stat->blksize = NFS_SERVER(inode)->dtsize;
745 	}
746 out:
747 	trace_nfs_getattr_exit(inode, err);
748 	return err;
749 }
750 EXPORT_SYMBOL_GPL(nfs_getattr);
751 
752 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
753 {
754 	atomic_set(&l_ctx->count, 1);
755 	l_ctx->lockowner = current->files;
756 	INIT_LIST_HEAD(&l_ctx->list);
757 	atomic_set(&l_ctx->io_count, 0);
758 }
759 
760 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
761 {
762 	struct nfs_lock_context *head = &ctx->lock_context;
763 	struct nfs_lock_context *pos = head;
764 
765 	do {
766 		if (pos->lockowner != current->files)
767 			continue;
768 		atomic_inc(&pos->count);
769 		return pos;
770 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
771 	return NULL;
772 }
773 
774 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
775 {
776 	struct nfs_lock_context *res, *new = NULL;
777 	struct inode *inode = d_inode(ctx->dentry);
778 
779 	spin_lock(&inode->i_lock);
780 	res = __nfs_find_lock_context(ctx);
781 	if (res == NULL) {
782 		spin_unlock(&inode->i_lock);
783 		new = kmalloc(sizeof(*new), GFP_KERNEL);
784 		if (new == NULL)
785 			return ERR_PTR(-ENOMEM);
786 		nfs_init_lock_context(new);
787 		spin_lock(&inode->i_lock);
788 		res = __nfs_find_lock_context(ctx);
789 		if (res == NULL) {
790 			list_add_tail(&new->list, &ctx->lock_context.list);
791 			new->open_context = ctx;
792 			res = new;
793 			new = NULL;
794 		}
795 	}
796 	spin_unlock(&inode->i_lock);
797 	kfree(new);
798 	return res;
799 }
800 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
801 
802 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
803 {
804 	struct nfs_open_context *ctx = l_ctx->open_context;
805 	struct inode *inode = d_inode(ctx->dentry);
806 
807 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
808 		return;
809 	list_del(&l_ctx->list);
810 	spin_unlock(&inode->i_lock);
811 	kfree(l_ctx);
812 }
813 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
814 
815 /**
816  * nfs_close_context - Common close_context() routine NFSv2/v3
817  * @ctx: pointer to context
818  * @is_sync: is this a synchronous close
819  *
820  * Ensure that the attributes are up to date if we're mounted
821  * with close-to-open semantics and we have cached data that will
822  * need to be revalidated on open.
823  */
824 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
825 {
826 	struct nfs_inode *nfsi;
827 	struct inode *inode;
828 	struct nfs_server *server;
829 
830 	if (!(ctx->mode & FMODE_WRITE))
831 		return;
832 	if (!is_sync)
833 		return;
834 	inode = d_inode(ctx->dentry);
835 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
836 		return;
837 	nfsi = NFS_I(inode);
838 	if (inode->i_mapping->nrpages == 0)
839 		return;
840 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
841 		return;
842 	if (!list_empty(&nfsi->open_files))
843 		return;
844 	server = NFS_SERVER(inode);
845 	if (server->flags & NFS_MOUNT_NOCTO)
846 		return;
847 	nfs_revalidate_inode(server, inode);
848 }
849 EXPORT_SYMBOL_GPL(nfs_close_context);
850 
851 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
852 						fmode_t f_mode,
853 						struct file *filp)
854 {
855 	struct nfs_open_context *ctx;
856 	struct rpc_cred *cred = rpc_lookup_cred();
857 	if (IS_ERR(cred))
858 		return ERR_CAST(cred);
859 
860 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
861 	if (!ctx) {
862 		put_rpccred(cred);
863 		return ERR_PTR(-ENOMEM);
864 	}
865 	nfs_sb_active(dentry->d_sb);
866 	ctx->dentry = dget(dentry);
867 	ctx->cred = cred;
868 	ctx->state = NULL;
869 	ctx->mode = f_mode;
870 	ctx->flags = 0;
871 	ctx->error = 0;
872 	ctx->flock_owner = (fl_owner_t)filp;
873 	nfs_init_lock_context(&ctx->lock_context);
874 	ctx->lock_context.open_context = ctx;
875 	INIT_LIST_HEAD(&ctx->list);
876 	ctx->mdsthreshold = NULL;
877 	return ctx;
878 }
879 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
880 
881 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
882 {
883 	if (ctx != NULL)
884 		atomic_inc(&ctx->lock_context.count);
885 	return ctx;
886 }
887 EXPORT_SYMBOL_GPL(get_nfs_open_context);
888 
889 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
890 {
891 	struct inode *inode = d_inode(ctx->dentry);
892 	struct super_block *sb = ctx->dentry->d_sb;
893 
894 	if (!list_empty(&ctx->list)) {
895 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
896 			return;
897 		list_del(&ctx->list);
898 		spin_unlock(&inode->i_lock);
899 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
900 		return;
901 	if (inode != NULL)
902 		NFS_PROTO(inode)->close_context(ctx, is_sync);
903 	if (ctx->cred != NULL)
904 		put_rpccred(ctx->cred);
905 	dput(ctx->dentry);
906 	nfs_sb_deactive(sb);
907 	kfree(ctx->mdsthreshold);
908 	kfree(ctx);
909 }
910 
911 void put_nfs_open_context(struct nfs_open_context *ctx)
912 {
913 	__put_nfs_open_context(ctx, 0);
914 }
915 EXPORT_SYMBOL_GPL(put_nfs_open_context);
916 
917 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
918 {
919 	__put_nfs_open_context(ctx, 1);
920 }
921 
922 /*
923  * Ensure that mmap has a recent RPC credential for use when writing out
924  * shared pages
925  */
926 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
927 {
928 	struct inode *inode = d_inode(ctx->dentry);
929 	struct nfs_inode *nfsi = NFS_I(inode);
930 
931 	spin_lock(&inode->i_lock);
932 	if (ctx->mode & FMODE_WRITE)
933 		list_add(&ctx->list, &nfsi->open_files);
934 	else
935 		list_add_tail(&ctx->list, &nfsi->open_files);
936 	spin_unlock(&inode->i_lock);
937 }
938 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
939 
940 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
941 {
942 	filp->private_data = get_nfs_open_context(ctx);
943 	if (list_empty(&ctx->list))
944 		nfs_inode_attach_open_context(ctx);
945 }
946 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
947 
948 /*
949  * Given an inode, search for an open context with the desired characteristics
950  */
951 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
952 {
953 	struct nfs_inode *nfsi = NFS_I(inode);
954 	struct nfs_open_context *pos, *ctx = NULL;
955 
956 	spin_lock(&inode->i_lock);
957 	list_for_each_entry(pos, &nfsi->open_files, list) {
958 		if (cred != NULL && pos->cred != cred)
959 			continue;
960 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
961 			continue;
962 		ctx = get_nfs_open_context(pos);
963 		break;
964 	}
965 	spin_unlock(&inode->i_lock);
966 	return ctx;
967 }
968 
969 void nfs_file_clear_open_context(struct file *filp)
970 {
971 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
972 
973 	if (ctx) {
974 		struct inode *inode = d_inode(ctx->dentry);
975 
976 		/*
977 		 * We fatal error on write before. Try to writeback
978 		 * every page again.
979 		 */
980 		if (ctx->error < 0)
981 			invalidate_inode_pages2(inode->i_mapping);
982 		filp->private_data = NULL;
983 		spin_lock(&inode->i_lock);
984 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
985 		spin_unlock(&inode->i_lock);
986 		put_nfs_open_context_sync(ctx);
987 	}
988 }
989 
990 /*
991  * These allocate and release file read/write context information.
992  */
993 int nfs_open(struct inode *inode, struct file *filp)
994 {
995 	struct nfs_open_context *ctx;
996 
997 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
998 	if (IS_ERR(ctx))
999 		return PTR_ERR(ctx);
1000 	nfs_file_set_open_context(filp, ctx);
1001 	put_nfs_open_context(ctx);
1002 	nfs_fscache_open_file(inode, filp);
1003 	return 0;
1004 }
1005 
1006 /*
1007  * This function is called whenever some part of NFS notices that
1008  * the cached attributes have to be refreshed.
1009  */
1010 int
1011 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1012 {
1013 	int		 status = -ESTALE;
1014 	struct nfs4_label *label = NULL;
1015 	struct nfs_fattr *fattr = NULL;
1016 	struct nfs_inode *nfsi = NFS_I(inode);
1017 
1018 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1019 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1020 
1021 	trace_nfs_revalidate_inode_enter(inode);
1022 
1023 	if (is_bad_inode(inode))
1024 		goto out;
1025 	if (NFS_STALE(inode))
1026 		goto out;
1027 
1028 	/* pNFS: Attributes aren't updated until we layoutcommit */
1029 	if (S_ISREG(inode->i_mode)) {
1030 		status = pnfs_sync_inode(inode, false);
1031 		if (status)
1032 			goto out;
1033 	}
1034 
1035 	status = -ENOMEM;
1036 	fattr = nfs_alloc_fattr();
1037 	if (fattr == NULL)
1038 		goto out;
1039 
1040 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1041 
1042 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1043 	if (IS_ERR(label)) {
1044 		status = PTR_ERR(label);
1045 		goto out;
1046 	}
1047 
1048 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
1049 	if (status != 0) {
1050 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1051 			 inode->i_sb->s_id,
1052 			 (unsigned long long)NFS_FILEID(inode), status);
1053 		if (status == -ESTALE) {
1054 			nfs_zap_caches(inode);
1055 			if (!S_ISDIR(inode->i_mode))
1056 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1057 		}
1058 		goto err_out;
1059 	}
1060 
1061 	status = nfs_refresh_inode(inode, fattr);
1062 	if (status) {
1063 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1064 			 inode->i_sb->s_id,
1065 			 (unsigned long long)NFS_FILEID(inode), status);
1066 		goto err_out;
1067 	}
1068 
1069 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1070 		nfs_zap_acl_cache(inode);
1071 
1072 	nfs_setsecurity(inode, fattr, label);
1073 
1074 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1075 		inode->i_sb->s_id,
1076 		(unsigned long long)NFS_FILEID(inode));
1077 
1078 err_out:
1079 	nfs4_label_free(label);
1080 out:
1081 	nfs_free_fattr(fattr);
1082 	trace_nfs_revalidate_inode_exit(inode, status);
1083 	return status;
1084 }
1085 
1086 int nfs_attribute_cache_expired(struct inode *inode)
1087 {
1088 	if (nfs_have_delegated_attributes(inode))
1089 		return 0;
1090 	return nfs_attribute_timeout(inode);
1091 }
1092 
1093 /**
1094  * nfs_revalidate_inode - Revalidate the inode attributes
1095  * @server - pointer to nfs_server struct
1096  * @inode - pointer to inode struct
1097  *
1098  * Updates inode attribute information by retrieving the data from the server.
1099  */
1100 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1101 {
1102 	if (!nfs_need_revalidate_inode(inode))
1103 		return NFS_STALE(inode) ? -ESTALE : 0;
1104 	return __nfs_revalidate_inode(server, inode);
1105 }
1106 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1107 
1108 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1109 {
1110 	struct nfs_inode *nfsi = NFS_I(inode);
1111 	int ret;
1112 
1113 	if (mapping->nrpages != 0) {
1114 		if (S_ISREG(inode->i_mode)) {
1115 			unmap_mapping_range(mapping, 0, 0, 0);
1116 			ret = nfs_sync_mapping(mapping);
1117 			if (ret < 0)
1118 				return ret;
1119 		}
1120 		ret = invalidate_inode_pages2(mapping);
1121 		if (ret < 0)
1122 			return ret;
1123 	}
1124 	if (S_ISDIR(inode->i_mode)) {
1125 		spin_lock(&inode->i_lock);
1126 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1127 		spin_unlock(&inode->i_lock);
1128 	}
1129 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1130 	nfs_fscache_wait_on_invalidate(inode);
1131 
1132 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1133 			inode->i_sb->s_id,
1134 			(unsigned long long)NFS_FILEID(inode));
1135 	return 0;
1136 }
1137 
1138 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1139 {
1140 	return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1141 		NFS_STALE(inode);
1142 }
1143 
1144 int nfs_revalidate_mapping_rcu(struct inode *inode)
1145 {
1146 	struct nfs_inode *nfsi = NFS_I(inode);
1147 	unsigned long *bitlock = &nfsi->flags;
1148 	int ret = 0;
1149 
1150 	if (IS_SWAPFILE(inode))
1151 		goto out;
1152 	if (nfs_mapping_need_revalidate_inode(inode)) {
1153 		ret = -ECHILD;
1154 		goto out;
1155 	}
1156 	spin_lock(&inode->i_lock);
1157 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1158 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1159 		ret = -ECHILD;
1160 	spin_unlock(&inode->i_lock);
1161 out:
1162 	return ret;
1163 }
1164 
1165 /**
1166  * nfs_revalidate_mapping - Revalidate the pagecache
1167  * @inode - pointer to host inode
1168  * @mapping - pointer to mapping
1169  */
1170 int nfs_revalidate_mapping(struct inode *inode,
1171 		struct address_space *mapping)
1172 {
1173 	struct nfs_inode *nfsi = NFS_I(inode);
1174 	unsigned long *bitlock = &nfsi->flags;
1175 	int ret = 0;
1176 
1177 	/* swapfiles are not supposed to be shared. */
1178 	if (IS_SWAPFILE(inode))
1179 		goto out;
1180 
1181 	if (nfs_mapping_need_revalidate_inode(inode)) {
1182 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1183 		if (ret < 0)
1184 			goto out;
1185 	}
1186 
1187 	/*
1188 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1189 	 * invalidations that come in while we're shooting down the mappings
1190 	 * are respected. But, that leaves a race window where one revalidator
1191 	 * can clear the flag, and then another checks it before the mapping
1192 	 * gets invalidated. Fix that by serializing access to this part of
1193 	 * the function.
1194 	 *
1195 	 * At the same time, we need to allow other tasks to see whether we
1196 	 * might be in the middle of invalidating the pages, so we only set
1197 	 * the bit lock here if it looks like we're going to be doing that.
1198 	 */
1199 	for (;;) {
1200 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1201 					 nfs_wait_bit_killable, TASK_KILLABLE);
1202 		if (ret)
1203 			goto out;
1204 		spin_lock(&inode->i_lock);
1205 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1206 			spin_unlock(&inode->i_lock);
1207 			continue;
1208 		}
1209 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1210 			break;
1211 		spin_unlock(&inode->i_lock);
1212 		goto out;
1213 	}
1214 
1215 	set_bit(NFS_INO_INVALIDATING, bitlock);
1216 	smp_wmb();
1217 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1218 	spin_unlock(&inode->i_lock);
1219 	trace_nfs_invalidate_mapping_enter(inode);
1220 	ret = nfs_invalidate_mapping(inode, mapping);
1221 	trace_nfs_invalidate_mapping_exit(inode, ret);
1222 
1223 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1224 	smp_mb__after_atomic();
1225 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1226 out:
1227 	return ret;
1228 }
1229 
1230 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1231 {
1232 	struct inode *inode = &nfsi->vfs_inode;
1233 
1234 	assert_spin_locked(&inode->i_lock);
1235 
1236 	if (!S_ISREG(inode->i_mode))
1237 		return false;
1238 	if (list_empty(&nfsi->open_files))
1239 		return false;
1240 	/* Note: This relies on nfsi->open_files being ordered with writers
1241 	 *       being placed at the head of the list.
1242 	 *       See nfs_inode_attach_open_context()
1243 	 */
1244 	return (list_first_entry(&nfsi->open_files,
1245 			struct nfs_open_context,
1246 			list)->mode & FMODE_WRITE) == FMODE_WRITE;
1247 }
1248 
1249 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1250 {
1251 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1252 }
1253 
1254 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1255 {
1256 	struct nfs_inode *nfsi = NFS_I(inode);
1257 	unsigned long ret = 0;
1258 
1259 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1260 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1261 			&& inode->i_version == fattr->pre_change_attr) {
1262 		inode->i_version = fattr->change_attr;
1263 		if (S_ISDIR(inode->i_mode))
1264 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1265 		ret |= NFS_INO_INVALID_ATTR;
1266 	}
1267 	/* If we have atomic WCC data, we may update some attributes */
1268 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1269 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1270 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1271 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1272 		ret |= NFS_INO_INVALID_ATTR;
1273 	}
1274 
1275 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1276 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1277 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1278 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1279 		if (S_ISDIR(inode->i_mode))
1280 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1281 		ret |= NFS_INO_INVALID_ATTR;
1282 	}
1283 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1284 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1285 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1286 			&& nfsi->nrequests == 0) {
1287 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1288 		ret |= NFS_INO_INVALID_ATTR;
1289 	}
1290 
1291 	return ret;
1292 }
1293 
1294 /**
1295  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1296  * @inode - pointer to inode
1297  * @fattr - updated attributes
1298  *
1299  * Verifies the attribute cache. If we have just changed the attributes,
1300  * so that fattr carries weak cache consistency data, then it may
1301  * also update the ctime/mtime/change_attribute.
1302  */
1303 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1304 {
1305 	struct nfs_inode *nfsi = NFS_I(inode);
1306 	loff_t cur_size, new_isize;
1307 	unsigned long invalid = 0;
1308 
1309 
1310 	if (nfs_have_delegated_attributes(inode))
1311 		return 0;
1312 	/* Has the inode gone and changed behind our back? */
1313 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1314 		return -EIO;
1315 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1316 		return -EIO;
1317 
1318 	if (!nfs_file_has_buffered_writers(nfsi)) {
1319 		/* Verify a few of the more important attributes */
1320 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && inode->i_version != fattr->change_attr)
1321 			invalid |= NFS_INO_INVALID_ATTR | NFS_INO_REVAL_PAGECACHE;
1322 
1323 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1324 			invalid |= NFS_INO_INVALID_ATTR;
1325 
1326 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&inode->i_ctime, &fattr->ctime))
1327 			invalid |= NFS_INO_INVALID_ATTR;
1328 
1329 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1330 			cur_size = i_size_read(inode);
1331 			new_isize = nfs_size_to_loff_t(fattr->size);
1332 			if (cur_size != new_isize)
1333 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1334 		}
1335 	}
1336 
1337 	/* Have any file permissions changed? */
1338 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1339 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1340 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1341 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1342 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1343 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1344 
1345 	/* Has the link count changed? */
1346 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1347 		invalid |= NFS_INO_INVALID_ATTR;
1348 
1349 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1350 		invalid |= NFS_INO_INVALID_ATIME;
1351 
1352 	if (invalid != 0)
1353 		nfs_set_cache_invalid(inode, invalid | NFS_INO_REVAL_FORCED);
1354 
1355 	nfsi->read_cache_jiffies = fattr->time_start;
1356 	return 0;
1357 }
1358 
1359 static atomic_long_t nfs_attr_generation_counter;
1360 
1361 static unsigned long nfs_read_attr_generation_counter(void)
1362 {
1363 	return atomic_long_read(&nfs_attr_generation_counter);
1364 }
1365 
1366 unsigned long nfs_inc_attr_generation_counter(void)
1367 {
1368 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1369 }
1370 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1371 
1372 void nfs_fattr_init(struct nfs_fattr *fattr)
1373 {
1374 	fattr->valid = 0;
1375 	fattr->time_start = jiffies;
1376 	fattr->gencount = nfs_inc_attr_generation_counter();
1377 	fattr->owner_name = NULL;
1378 	fattr->group_name = NULL;
1379 }
1380 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1381 
1382 /**
1383  * nfs_fattr_set_barrier
1384  * @fattr: attributes
1385  *
1386  * Used to set a barrier after an attribute was updated. This
1387  * barrier ensures that older attributes from RPC calls that may
1388  * have raced with our update cannot clobber these new values.
1389  * Note that you are still responsible for ensuring that other
1390  * operations which change the attribute on the server do not
1391  * collide.
1392  */
1393 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1394 {
1395 	fattr->gencount = nfs_inc_attr_generation_counter();
1396 }
1397 
1398 struct nfs_fattr *nfs_alloc_fattr(void)
1399 {
1400 	struct nfs_fattr *fattr;
1401 
1402 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1403 	if (fattr != NULL)
1404 		nfs_fattr_init(fattr);
1405 	return fattr;
1406 }
1407 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1408 
1409 struct nfs_fh *nfs_alloc_fhandle(void)
1410 {
1411 	struct nfs_fh *fh;
1412 
1413 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1414 	if (fh != NULL)
1415 		fh->size = 0;
1416 	return fh;
1417 }
1418 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1419 
1420 #ifdef NFS_DEBUG
1421 /*
1422  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1423  *                             in the same way that wireshark does
1424  *
1425  * @fh: file handle
1426  *
1427  * For debugging only.
1428  */
1429 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1430 {
1431 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1432 	 * not on the result */
1433 	return nfs_fhandle_hash(fh);
1434 }
1435 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1436 
1437 /*
1438  * _nfs_display_fhandle - display an NFS file handle on the console
1439  *
1440  * @fh: file handle to display
1441  * @caption: display caption
1442  *
1443  * For debugging only.
1444  */
1445 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1446 {
1447 	unsigned short i;
1448 
1449 	if (fh == NULL || fh->size == 0) {
1450 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1451 		return;
1452 	}
1453 
1454 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1455 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1456 	for (i = 0; i < fh->size; i += 16) {
1457 		__be32 *pos = (__be32 *)&fh->data[i];
1458 
1459 		switch ((fh->size - i - 1) >> 2) {
1460 		case 0:
1461 			printk(KERN_DEFAULT " %08x\n",
1462 				be32_to_cpup(pos));
1463 			break;
1464 		case 1:
1465 			printk(KERN_DEFAULT " %08x %08x\n",
1466 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1467 			break;
1468 		case 2:
1469 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1470 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1471 				be32_to_cpup(pos + 2));
1472 			break;
1473 		default:
1474 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1475 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1476 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1477 		}
1478 	}
1479 }
1480 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1481 #endif
1482 
1483 /**
1484  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1485  * @inode - pointer to inode
1486  * @fattr - attributes
1487  *
1488  * Attempt to divine whether or not an RPC call reply carrying stale
1489  * attributes got scheduled after another call carrying updated ones.
1490  *
1491  * To do so, the function first assumes that a more recent ctime means
1492  * that the attributes in fattr are newer, however it also attempt to
1493  * catch the case where ctime either didn't change, or went backwards
1494  * (if someone reset the clock on the server) by looking at whether
1495  * or not this RPC call was started after the inode was last updated.
1496  * Note also the check for wraparound of 'attr_gencount'
1497  *
1498  * The function returns 'true' if it thinks the attributes in 'fattr' are
1499  * more recent than the ones cached in the inode.
1500  *
1501  */
1502 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1503 {
1504 	const struct nfs_inode *nfsi = NFS_I(inode);
1505 
1506 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1507 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1508 }
1509 
1510 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1511 {
1512 	int ret;
1513 
1514 	trace_nfs_refresh_inode_enter(inode);
1515 
1516 	if (nfs_inode_attrs_need_update(inode, fattr))
1517 		ret = nfs_update_inode(inode, fattr);
1518 	else
1519 		ret = nfs_check_inode_attributes(inode, fattr);
1520 
1521 	trace_nfs_refresh_inode_exit(inode, ret);
1522 	return ret;
1523 }
1524 
1525 /**
1526  * nfs_refresh_inode - try to update the inode attribute cache
1527  * @inode - pointer to inode
1528  * @fattr - updated attributes
1529  *
1530  * Check that an RPC call that returned attributes has not overlapped with
1531  * other recent updates of the inode metadata, then decide whether it is
1532  * safe to do a full update of the inode attributes, or whether just to
1533  * call nfs_check_inode_attributes.
1534  */
1535 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1536 {
1537 	int status;
1538 
1539 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1540 		return 0;
1541 	spin_lock(&inode->i_lock);
1542 	status = nfs_refresh_inode_locked(inode, fattr);
1543 	spin_unlock(&inode->i_lock);
1544 
1545 	return status;
1546 }
1547 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1548 
1549 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1550 {
1551 	unsigned long invalid = NFS_INO_INVALID_ATTR;
1552 
1553 	if (S_ISDIR(inode->i_mode))
1554 		invalid |= NFS_INO_INVALID_DATA;
1555 	nfs_set_cache_invalid(inode, invalid);
1556 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1557 		return 0;
1558 	return nfs_refresh_inode_locked(inode, fattr);
1559 }
1560 
1561 /**
1562  * nfs_post_op_update_inode - try to update the inode attribute cache
1563  * @inode - pointer to inode
1564  * @fattr - updated attributes
1565  *
1566  * After an operation that has changed the inode metadata, mark the
1567  * attribute cache as being invalid, then try to update it.
1568  *
1569  * NB: if the server didn't return any post op attributes, this
1570  * function will force the retrieval of attributes before the next
1571  * NFS request.  Thus it should be used only for operations that
1572  * are expected to change one or more attributes, to avoid
1573  * unnecessary NFS requests and trips through nfs_update_inode().
1574  */
1575 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1576 {
1577 	int status;
1578 
1579 	spin_lock(&inode->i_lock);
1580 	nfs_fattr_set_barrier(fattr);
1581 	status = nfs_post_op_update_inode_locked(inode, fattr);
1582 	spin_unlock(&inode->i_lock);
1583 
1584 	return status;
1585 }
1586 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1587 
1588 /**
1589  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1590  * @inode - pointer to inode
1591  * @fattr - updated attributes
1592  *
1593  * After an operation that has changed the inode metadata, mark the
1594  * attribute cache as being invalid, then try to update it. Fake up
1595  * weak cache consistency data, if none exist.
1596  *
1597  * This function is mainly designed to be used by the ->write_done() functions.
1598  */
1599 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1600 {
1601 	int status;
1602 
1603 	/* Don't do a WCC update if these attributes are already stale */
1604 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1605 			!nfs_inode_attrs_need_update(inode, fattr)) {
1606 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1607 				| NFS_ATTR_FATTR_PRESIZE
1608 				| NFS_ATTR_FATTR_PREMTIME
1609 				| NFS_ATTR_FATTR_PRECTIME);
1610 		goto out_noforce;
1611 	}
1612 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1613 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1614 		fattr->pre_change_attr = inode->i_version;
1615 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1616 	}
1617 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1618 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1619 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1620 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1621 	}
1622 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1623 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1624 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1625 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1626 	}
1627 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1628 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1629 		fattr->pre_size = i_size_read(inode);
1630 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1631 	}
1632 out_noforce:
1633 	status = nfs_post_op_update_inode_locked(inode, fattr);
1634 	return status;
1635 }
1636 
1637 /**
1638  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1639  * @inode - pointer to inode
1640  * @fattr - updated attributes
1641  *
1642  * After an operation that has changed the inode metadata, mark the
1643  * attribute cache as being invalid, then try to update it. Fake up
1644  * weak cache consistency data, if none exist.
1645  *
1646  * This function is mainly designed to be used by the ->write_done() functions.
1647  */
1648 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1649 {
1650 	int status;
1651 
1652 	spin_lock(&inode->i_lock);
1653 	nfs_fattr_set_barrier(fattr);
1654 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1655 	spin_unlock(&inode->i_lock);
1656 	return status;
1657 }
1658 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1659 
1660 
1661 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1662 				    struct nfs_fattr *fattr)
1663 {
1664 	bool ret1 = true, ret2 = true;
1665 
1666 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1667 		ret1 = (nfsi->fileid == fattr->fileid);
1668 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1669 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1670 	return ret1 || ret2;
1671 }
1672 
1673 /*
1674  * Many nfs protocol calls return the new file attributes after
1675  * an operation.  Here we update the inode to reflect the state
1676  * of the server's inode.
1677  *
1678  * This is a bit tricky because we have to make sure all dirty pages
1679  * have been sent off to the server before calling invalidate_inode_pages.
1680  * To make sure no other process adds more write requests while we try
1681  * our best to flush them, we make them sleep during the attribute refresh.
1682  *
1683  * A very similar scenario holds for the dir cache.
1684  */
1685 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1686 {
1687 	struct nfs_server *server;
1688 	struct nfs_inode *nfsi = NFS_I(inode);
1689 	loff_t cur_isize, new_isize;
1690 	unsigned long invalid = 0;
1691 	unsigned long now = jiffies;
1692 	unsigned long save_cache_validity;
1693 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
1694 	bool cache_revalidated = true;
1695 
1696 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1697 			__func__, inode->i_sb->s_id, inode->i_ino,
1698 			nfs_display_fhandle_hash(NFS_FH(inode)),
1699 			atomic_read(&inode->i_count), fattr->valid);
1700 
1701 	if (!nfs_fileid_valid(nfsi, fattr)) {
1702 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1703 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1704 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1705 			inode->i_sb->s_id, (long long)nfsi->fileid,
1706 			(long long)fattr->fileid);
1707 		goto out_err;
1708 	}
1709 
1710 	/*
1711 	 * Make sure the inode's type hasn't changed.
1712 	 */
1713 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1714 		/*
1715 		* Big trouble! The inode has become a different object.
1716 		*/
1717 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1718 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1719 		goto out_err;
1720 	}
1721 
1722 	server = NFS_SERVER(inode);
1723 	/* Update the fsid? */
1724 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1725 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1726 			!IS_AUTOMOUNT(inode))
1727 		server->fsid = fattr->fsid;
1728 
1729 	/*
1730 	 * Update the read time so we don't revalidate too often.
1731 	 */
1732 	nfsi->read_cache_jiffies = fattr->time_start;
1733 
1734 	save_cache_validity = nfsi->cache_validity;
1735 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1736 			| NFS_INO_INVALID_ATIME
1737 			| NFS_INO_REVAL_FORCED
1738 			| NFS_INO_REVAL_PAGECACHE);
1739 
1740 	/* Do atomic weak cache consistency updates */
1741 	invalid |= nfs_wcc_update_inode(inode, fattr);
1742 
1743 	if (pnfs_layoutcommit_outstanding(inode)) {
1744 		nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1745 		cache_revalidated = false;
1746 	}
1747 
1748 	/* More cache consistency checks */
1749 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1750 		if (inode->i_version != fattr->change_attr) {
1751 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1752 					inode->i_sb->s_id, inode->i_ino);
1753 			/* Could it be a race with writeback? */
1754 			if (!have_writers) {
1755 				invalid |= NFS_INO_INVALID_ATTR
1756 					| NFS_INO_INVALID_DATA
1757 					| NFS_INO_INVALID_ACCESS
1758 					| NFS_INO_INVALID_ACL;
1759 				if (S_ISDIR(inode->i_mode))
1760 					nfs_force_lookup_revalidate(inode);
1761 			}
1762 			inode->i_version = fattr->change_attr;
1763 		}
1764 	} else {
1765 		nfsi->cache_validity |= save_cache_validity;
1766 		cache_revalidated = false;
1767 	}
1768 
1769 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1770 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1771 	} else if (server->caps & NFS_CAP_MTIME) {
1772 		nfsi->cache_validity |= save_cache_validity &
1773 				(NFS_INO_INVALID_ATTR
1774 				| NFS_INO_REVAL_FORCED);
1775 		cache_revalidated = false;
1776 	}
1777 
1778 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1779 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1780 	} else if (server->caps & NFS_CAP_CTIME) {
1781 		nfsi->cache_validity |= save_cache_validity &
1782 				(NFS_INO_INVALID_ATTR
1783 				| NFS_INO_REVAL_FORCED);
1784 		cache_revalidated = false;
1785 	}
1786 
1787 	/* Check if our cached file size is stale */
1788 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1789 		new_isize = nfs_size_to_loff_t(fattr->size);
1790 		cur_isize = i_size_read(inode);
1791 		if (new_isize != cur_isize) {
1792 			/* Do we perhaps have any outstanding writes, or has
1793 			 * the file grown beyond our last write? */
1794 			if (nfsi->nrequests == 0 || new_isize > cur_isize) {
1795 				i_size_write(inode, new_isize);
1796 				if (!have_writers)
1797 					invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1798 			}
1799 			dprintk("NFS: isize change on server for file %s/%ld "
1800 					"(%Ld to %Ld)\n",
1801 					inode->i_sb->s_id,
1802 					inode->i_ino,
1803 					(long long)cur_isize,
1804 					(long long)new_isize);
1805 		}
1806 	} else {
1807 		nfsi->cache_validity |= save_cache_validity &
1808 				(NFS_INO_INVALID_ATTR
1809 				| NFS_INO_REVAL_PAGECACHE
1810 				| NFS_INO_REVAL_FORCED);
1811 		cache_revalidated = false;
1812 	}
1813 
1814 
1815 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1816 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1817 	else if (server->caps & NFS_CAP_ATIME) {
1818 		nfsi->cache_validity |= save_cache_validity &
1819 				(NFS_INO_INVALID_ATIME
1820 				| NFS_INO_REVAL_FORCED);
1821 		cache_revalidated = false;
1822 	}
1823 
1824 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1825 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1826 			umode_t newmode = inode->i_mode & S_IFMT;
1827 			newmode |= fattr->mode & S_IALLUGO;
1828 			inode->i_mode = newmode;
1829 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1830 		}
1831 	} else if (server->caps & NFS_CAP_MODE) {
1832 		nfsi->cache_validity |= save_cache_validity &
1833 				(NFS_INO_INVALID_ATTR
1834 				| NFS_INO_INVALID_ACCESS
1835 				| NFS_INO_INVALID_ACL
1836 				| NFS_INO_REVAL_FORCED);
1837 		cache_revalidated = false;
1838 	}
1839 
1840 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1841 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1842 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1843 			inode->i_uid = fattr->uid;
1844 		}
1845 	} else if (server->caps & NFS_CAP_OWNER) {
1846 		nfsi->cache_validity |= save_cache_validity &
1847 				(NFS_INO_INVALID_ATTR
1848 				| NFS_INO_INVALID_ACCESS
1849 				| NFS_INO_INVALID_ACL
1850 				| NFS_INO_REVAL_FORCED);
1851 		cache_revalidated = false;
1852 	}
1853 
1854 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1855 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1856 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1857 			inode->i_gid = fattr->gid;
1858 		}
1859 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
1860 		nfsi->cache_validity |= save_cache_validity &
1861 				(NFS_INO_INVALID_ATTR
1862 				| NFS_INO_INVALID_ACCESS
1863 				| NFS_INO_INVALID_ACL
1864 				| NFS_INO_REVAL_FORCED);
1865 		cache_revalidated = false;
1866 	}
1867 
1868 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1869 		if (inode->i_nlink != fattr->nlink) {
1870 			invalid |= NFS_INO_INVALID_ATTR;
1871 			if (S_ISDIR(inode->i_mode))
1872 				invalid |= NFS_INO_INVALID_DATA;
1873 			set_nlink(inode, fattr->nlink);
1874 		}
1875 	} else if (server->caps & NFS_CAP_NLINK) {
1876 		nfsi->cache_validity |= save_cache_validity &
1877 				(NFS_INO_INVALID_ATTR
1878 				| NFS_INO_REVAL_FORCED);
1879 		cache_revalidated = false;
1880 	}
1881 
1882 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1883 		/*
1884 		 * report the blocks in 512byte units
1885 		 */
1886 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1887 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1888 		inode->i_blocks = fattr->du.nfs2.blocks;
1889 	else
1890 		cache_revalidated = false;
1891 
1892 	/* Update attrtimeo value if we're out of the unstable period */
1893 	if (invalid & NFS_INO_INVALID_ATTR) {
1894 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1895 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1896 		nfsi->attrtimeo_timestamp = now;
1897 		/* Set barrier to be more recent than all outstanding updates */
1898 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1899 	} else {
1900 		if (cache_revalidated) {
1901 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1902 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1903 				nfsi->attrtimeo <<= 1;
1904 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1905 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1906 			}
1907 			nfsi->attrtimeo_timestamp = now;
1908 		}
1909 		/* Set the barrier to be more recent than this fattr */
1910 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1911 			nfsi->attr_gencount = fattr->gencount;
1912 	}
1913 
1914 	/* Don't declare attrcache up to date if there were no attrs! */
1915 	if (cache_revalidated)
1916 		invalid &= ~NFS_INO_INVALID_ATTR;
1917 
1918 	/* Don't invalidate the data if we were to blame */
1919 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1920 				|| S_ISLNK(inode->i_mode)))
1921 		invalid &= ~NFS_INO_INVALID_DATA;
1922 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1923 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1924 		nfs_set_cache_invalid(inode, invalid);
1925 
1926 	return 0;
1927  out_err:
1928 	/*
1929 	 * No need to worry about unhashing the dentry, as the
1930 	 * lookup validation will know that the inode is bad.
1931 	 * (But we fall through to invalidate the caches.)
1932 	 */
1933 	nfs_invalidate_inode(inode);
1934 	return -ESTALE;
1935 }
1936 
1937 struct inode *nfs_alloc_inode(struct super_block *sb)
1938 {
1939 	struct nfs_inode *nfsi;
1940 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1941 	if (!nfsi)
1942 		return NULL;
1943 	nfsi->flags = 0UL;
1944 	nfsi->cache_validity = 0UL;
1945 #if IS_ENABLED(CONFIG_NFS_V4)
1946 	nfsi->nfs4_acl = NULL;
1947 #endif /* CONFIG_NFS_V4 */
1948 	return &nfsi->vfs_inode;
1949 }
1950 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1951 
1952 static void nfs_i_callback(struct rcu_head *head)
1953 {
1954 	struct inode *inode = container_of(head, struct inode, i_rcu);
1955 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1956 }
1957 
1958 void nfs_destroy_inode(struct inode *inode)
1959 {
1960 	call_rcu(&inode->i_rcu, nfs_i_callback);
1961 }
1962 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1963 
1964 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1965 {
1966 #if IS_ENABLED(CONFIG_NFS_V4)
1967 	INIT_LIST_HEAD(&nfsi->open_states);
1968 	nfsi->delegation = NULL;
1969 	init_rwsem(&nfsi->rwsem);
1970 	nfsi->layout = NULL;
1971 #endif
1972 }
1973 
1974 static void init_once(void *foo)
1975 {
1976 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1977 
1978 	inode_init_once(&nfsi->vfs_inode);
1979 	INIT_LIST_HEAD(&nfsi->open_files);
1980 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1981 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1982 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1983 	nfsi->nrequests = 0;
1984 	nfsi->commit_info.ncommit = 0;
1985 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1986 	init_rwsem(&nfsi->rmdir_sem);
1987 	nfs4_init_once(nfsi);
1988 }
1989 
1990 static int __init nfs_init_inodecache(void)
1991 {
1992 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1993 					     sizeof(struct nfs_inode),
1994 					     0, (SLAB_RECLAIM_ACCOUNT|
1995 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1996 					     init_once);
1997 	if (nfs_inode_cachep == NULL)
1998 		return -ENOMEM;
1999 
2000 	return 0;
2001 }
2002 
2003 static void nfs_destroy_inodecache(void)
2004 {
2005 	/*
2006 	 * Make sure all delayed rcu free inodes are flushed before we
2007 	 * destroy cache.
2008 	 */
2009 	rcu_barrier();
2010 	kmem_cache_destroy(nfs_inode_cachep);
2011 }
2012 
2013 struct workqueue_struct *nfsiod_workqueue;
2014 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2015 
2016 /*
2017  * start up the nfsiod workqueue
2018  */
2019 static int nfsiod_start(void)
2020 {
2021 	struct workqueue_struct *wq;
2022 	dprintk("RPC:       creating workqueue nfsiod\n");
2023 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2024 	if (wq == NULL)
2025 		return -ENOMEM;
2026 	nfsiod_workqueue = wq;
2027 	return 0;
2028 }
2029 
2030 /*
2031  * Destroy the nfsiod workqueue
2032  */
2033 static void nfsiod_stop(void)
2034 {
2035 	struct workqueue_struct *wq;
2036 
2037 	wq = nfsiod_workqueue;
2038 	if (wq == NULL)
2039 		return;
2040 	nfsiod_workqueue = NULL;
2041 	destroy_workqueue(wq);
2042 }
2043 
2044 unsigned int nfs_net_id;
2045 EXPORT_SYMBOL_GPL(nfs_net_id);
2046 
2047 static int nfs_net_init(struct net *net)
2048 {
2049 	nfs_clients_init(net);
2050 	return nfs_fs_proc_net_init(net);
2051 }
2052 
2053 static void nfs_net_exit(struct net *net)
2054 {
2055 	nfs_fs_proc_net_exit(net);
2056 	nfs_cleanup_cb_ident_idr(net);
2057 }
2058 
2059 static struct pernet_operations nfs_net_ops = {
2060 	.init = nfs_net_init,
2061 	.exit = nfs_net_exit,
2062 	.id   = &nfs_net_id,
2063 	.size = sizeof(struct nfs_net),
2064 };
2065 
2066 /*
2067  * Initialize NFS
2068  */
2069 static int __init init_nfs_fs(void)
2070 {
2071 	int err;
2072 
2073 	err = register_pernet_subsys(&nfs_net_ops);
2074 	if (err < 0)
2075 		goto out9;
2076 
2077 	err = nfs_fscache_register();
2078 	if (err < 0)
2079 		goto out8;
2080 
2081 	err = nfsiod_start();
2082 	if (err)
2083 		goto out7;
2084 
2085 	err = nfs_fs_proc_init();
2086 	if (err)
2087 		goto out6;
2088 
2089 	err = nfs_init_nfspagecache();
2090 	if (err)
2091 		goto out5;
2092 
2093 	err = nfs_init_inodecache();
2094 	if (err)
2095 		goto out4;
2096 
2097 	err = nfs_init_readpagecache();
2098 	if (err)
2099 		goto out3;
2100 
2101 	err = nfs_init_writepagecache();
2102 	if (err)
2103 		goto out2;
2104 
2105 	err = nfs_init_directcache();
2106 	if (err)
2107 		goto out1;
2108 
2109 	rpc_proc_register(&init_net, &nfs_rpcstat);
2110 
2111 	err = register_nfs_fs();
2112 	if (err)
2113 		goto out0;
2114 
2115 	return 0;
2116 out0:
2117 	rpc_proc_unregister(&init_net, "nfs");
2118 	nfs_destroy_directcache();
2119 out1:
2120 	nfs_destroy_writepagecache();
2121 out2:
2122 	nfs_destroy_readpagecache();
2123 out3:
2124 	nfs_destroy_inodecache();
2125 out4:
2126 	nfs_destroy_nfspagecache();
2127 out5:
2128 	nfs_fs_proc_exit();
2129 out6:
2130 	nfsiod_stop();
2131 out7:
2132 	nfs_fscache_unregister();
2133 out8:
2134 	unregister_pernet_subsys(&nfs_net_ops);
2135 out9:
2136 	return err;
2137 }
2138 
2139 static void __exit exit_nfs_fs(void)
2140 {
2141 	nfs_destroy_directcache();
2142 	nfs_destroy_writepagecache();
2143 	nfs_destroy_readpagecache();
2144 	nfs_destroy_inodecache();
2145 	nfs_destroy_nfspagecache();
2146 	nfs_fscache_unregister();
2147 	unregister_pernet_subsys(&nfs_net_ops);
2148 	rpc_proc_unregister(&init_net, "nfs");
2149 	unregister_nfs_fs();
2150 	nfs_fs_proc_exit();
2151 	nfsiod_stop();
2152 }
2153 
2154 /* Not quite true; I just maintain it */
2155 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2156 MODULE_LICENSE("GPL");
2157 module_param(enable_ino64, bool, 0644);
2158 
2159 module_init(init_nfs_fs)
2160 module_exit(exit_nfs_fs)
2161