xref: /linux/fs/nfs/inode.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
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, unsigned int mode)
89 {
90 	return nfs_wait_killable(mode);
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 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
390 struct inode *
391 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
392 {
393 	struct nfs_find_desc desc = {
394 		.fh	= fh,
395 		.fattr	= fattr,
396 	};
397 	struct inode *inode;
398 	unsigned long hash;
399 
400 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
401 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
402 		return NULL;
403 
404 	hash = nfs_fattr_to_ino_t(fattr);
405 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
406 
407 	dprintk("%s: returning %p\n", __func__, inode);
408 	return inode;
409 }
410 
411 /*
412  * This is our front-end to iget that looks up inodes by file handle
413  * instead of inode number.
414  */
415 struct inode *
416 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
417 {
418 	struct nfs_find_desc desc = {
419 		.fh	= fh,
420 		.fattr	= fattr
421 	};
422 	struct inode *inode = ERR_PTR(-ENOENT);
423 	unsigned long hash;
424 
425 	nfs_attr_check_mountpoint(sb, fattr);
426 
427 	if (nfs_attr_use_mounted_on_fileid(fattr))
428 		fattr->fileid = fattr->mounted_on_fileid;
429 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
430 		goto out_no_inode;
431 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
432 		goto out_no_inode;
433 
434 	hash = nfs_fattr_to_ino_t(fattr);
435 
436 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
437 	if (inode == NULL) {
438 		inode = ERR_PTR(-ENOMEM);
439 		goto out_no_inode;
440 	}
441 
442 	if (inode->i_state & I_NEW) {
443 		struct nfs_inode *nfsi = NFS_I(inode);
444 		unsigned long now = jiffies;
445 
446 		/* We set i_ino for the few things that still rely on it,
447 		 * such as stat(2) */
448 		inode->i_ino = hash;
449 
450 		/* We can't support update_atime(), since the server will reset it */
451 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
452 		inode->i_mode = fattr->mode;
453 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
454 				&& nfs_server_capable(inode, NFS_CAP_MODE))
455 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
456 		/* Why so? Because we want revalidate for devices/FIFOs, and
457 		 * that's precisely what we have in nfs_file_inode_operations.
458 		 */
459 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
460 		if (S_ISREG(inode->i_mode)) {
461 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
462 			inode->i_data.a_ops = &nfs_file_aops;
463 		} else if (S_ISDIR(inode->i_mode)) {
464 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
465 			inode->i_fop = &nfs_dir_operations;
466 			inode->i_data.a_ops = &nfs_dir_aops;
467 			/* Deal with crossing mountpoints */
468 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
469 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
470 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
471 					inode->i_op = &nfs_referral_inode_operations;
472 				else
473 					inode->i_op = &nfs_mountpoint_inode_operations;
474 				inode->i_fop = NULL;
475 				inode->i_flags |= S_AUTOMOUNT;
476 			}
477 		} else if (S_ISLNK(inode->i_mode)) {
478 			inode->i_op = &nfs_symlink_inode_operations;
479 			inode_nohighmem(inode);
480 		} else
481 			init_special_inode(inode, inode->i_mode, fattr->rdev);
482 
483 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
484 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
485 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
486 		inode->i_version = 0;
487 		inode->i_size = 0;
488 		clear_nlink(inode);
489 		inode->i_uid = make_kuid(&init_user_ns, -2);
490 		inode->i_gid = make_kgid(&init_user_ns, -2);
491 		inode->i_blocks = 0;
492 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
493 		nfsi->write_io = 0;
494 		nfsi->read_io = 0;
495 
496 		nfsi->read_cache_jiffies = fattr->time_start;
497 		nfsi->attr_gencount = fattr->gencount;
498 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
499 			inode->i_atime = fattr->atime;
500 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
501 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
502 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
503 			inode->i_mtime = fattr->mtime;
504 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
505 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
506 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
507 			inode->i_ctime = fattr->ctime;
508 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
509 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
510 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
511 			inode->i_version = fattr->change_attr;
512 		else
513 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
514 				| NFS_INO_REVAL_PAGECACHE);
515 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
516 			inode->i_size = nfs_size_to_loff_t(fattr->size);
517 		else
518 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
519 				| NFS_INO_REVAL_PAGECACHE);
520 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
521 			set_nlink(inode, fattr->nlink);
522 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
523 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
524 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
525 			inode->i_uid = fattr->uid;
526 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
527 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
528 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
529 			inode->i_gid = fattr->gid;
530 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
531 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
532 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
533 			inode->i_blocks = fattr->du.nfs2.blocks;
534 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
535 			/*
536 			 * report the blocks in 512byte units
537 			 */
538 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
539 		}
540 
541 		nfs_setsecurity(inode, fattr, label);
542 
543 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
544 		nfsi->attrtimeo_timestamp = now;
545 		nfsi->access_cache = RB_ROOT;
546 
547 		nfs_fscache_init_inode(inode);
548 
549 		unlock_new_inode(inode);
550 	} else {
551 		int err = nfs_refresh_inode(inode, fattr);
552 		if (err < 0) {
553 			iput(inode);
554 			inode = ERR_PTR(err);
555 			goto out_no_inode;
556 		}
557 	}
558 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
559 		inode->i_sb->s_id,
560 		(unsigned long long)NFS_FILEID(inode),
561 		nfs_display_fhandle_hash(fh),
562 		atomic_read(&inode->i_count));
563 
564 out:
565 	return inode;
566 
567 out_no_inode:
568 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
569 	goto out;
570 }
571 EXPORT_SYMBOL_GPL(nfs_fhget);
572 
573 #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)
574 
575 int
576 nfs_setattr(struct dentry *dentry, struct iattr *attr)
577 {
578 	struct inode *inode = d_inode(dentry);
579 	struct nfs_fattr *fattr;
580 	int error = 0;
581 
582 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
583 
584 	/* skip mode change if it's just for clearing setuid/setgid */
585 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
586 		attr->ia_valid &= ~ATTR_MODE;
587 
588 	if (attr->ia_valid & ATTR_SIZE) {
589 		BUG_ON(!S_ISREG(inode->i_mode));
590 
591 		error = inode_newsize_ok(inode, attr->ia_size);
592 		if (error)
593 			return error;
594 
595 		if (attr->ia_size == i_size_read(inode))
596 			attr->ia_valid &= ~ATTR_SIZE;
597 	}
598 
599 	/* Optimization: if the end result is no change, don't RPC */
600 	attr->ia_valid &= NFS_VALID_ATTRS;
601 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
602 		return 0;
603 
604 	trace_nfs_setattr_enter(inode);
605 
606 	/* Write all dirty data */
607 	if (S_ISREG(inode->i_mode))
608 		nfs_sync_inode(inode);
609 
610 	fattr = nfs_alloc_fattr();
611 	if (fattr == NULL) {
612 		error = -ENOMEM;
613 		goto out;
614 	}
615 
616 	/*
617 	 * Return any delegations if we're going to change ACLs
618 	 */
619 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
620 		NFS_PROTO(inode)->return_delegation(inode);
621 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
622 	if (error == 0)
623 		error = nfs_refresh_inode(inode, fattr);
624 	nfs_free_fattr(fattr);
625 out:
626 	trace_nfs_setattr_exit(inode, error);
627 	return error;
628 }
629 EXPORT_SYMBOL_GPL(nfs_setattr);
630 
631 /**
632  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
633  * @inode: inode of the file used
634  * @offset: file offset to start truncating
635  *
636  * This is a copy of the common vmtruncate, but with the locking
637  * corrected to take into account the fact that NFS requires
638  * inode->i_size to be updated under the inode->i_lock.
639  * Note: must be called with inode->i_lock held!
640  */
641 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
642 {
643 	int err;
644 
645 	err = inode_newsize_ok(inode, offset);
646 	if (err)
647 		goto out;
648 
649 	i_size_write(inode, offset);
650 	/* Optimisation */
651 	if (offset == 0)
652 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
653 
654 	spin_unlock(&inode->i_lock);
655 	truncate_pagecache(inode, offset);
656 	spin_lock(&inode->i_lock);
657 out:
658 	return err;
659 }
660 
661 /**
662  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
663  * @inode: pointer to struct inode
664  * @attr: pointer to struct iattr
665  *
666  * Note: we do this in the *proc.c in order to ensure that
667  *       it works for things like exclusive creates too.
668  */
669 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
670 		struct nfs_fattr *fattr)
671 {
672 	/* Barrier: bump the attribute generation count. */
673 	nfs_fattr_set_barrier(fattr);
674 
675 	spin_lock(&inode->i_lock);
676 	NFS_I(inode)->attr_gencount = fattr->gencount;
677 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
678 		if ((attr->ia_valid & ATTR_MODE) != 0) {
679 			int mode = attr->ia_mode & S_IALLUGO;
680 			mode |= inode->i_mode & ~S_IALLUGO;
681 			inode->i_mode = mode;
682 		}
683 		if ((attr->ia_valid & ATTR_UID) != 0)
684 			inode->i_uid = attr->ia_uid;
685 		if ((attr->ia_valid & ATTR_GID) != 0)
686 			inode->i_gid = attr->ia_gid;
687 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
688 				| NFS_INO_INVALID_ACL);
689 	}
690 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
691 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
692 		nfs_vmtruncate(inode, attr->ia_size);
693 	}
694 	if (fattr->valid)
695 		nfs_update_inode(inode, fattr);
696 	else
697 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
698 	spin_unlock(&inode->i_lock);
699 }
700 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
701 
702 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
703 {
704 	struct dentry *parent;
705 
706 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
707 		return;
708 	parent = dget_parent(dentry);
709 	nfs_force_use_readdirplus(d_inode(parent));
710 	dput(parent);
711 }
712 
713 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
714 {
715 	struct dentry *parent;
716 
717 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
718 		return;
719 	parent = dget_parent(dentry);
720 	nfs_advise_use_readdirplus(d_inode(parent));
721 	dput(parent);
722 }
723 
724 static bool nfs_need_revalidate_inode(struct inode *inode)
725 {
726 	if (NFS_I(inode)->cache_validity &
727 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
728 		return true;
729 	if (nfs_attribute_cache_expired(inode))
730 		return true;
731 	return false;
732 }
733 
734 int nfs_getattr(const struct path *path, struct kstat *stat,
735 		u32 request_mask, unsigned int query_flags)
736 {
737 	struct inode *inode = d_inode(path->dentry);
738 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
739 	int err = 0;
740 
741 	trace_nfs_getattr_enter(inode);
742 	/* Flush out writes to the server in order to update c/mtime.  */
743 	if (S_ISREG(inode->i_mode)) {
744 		err = filemap_write_and_wait(inode->i_mapping);
745 		if (err)
746 			goto out;
747 	}
748 
749 	/*
750 	 * We may force a getattr if the user cares about atime.
751 	 *
752 	 * Note that we only have to check the vfsmount flags here:
753 	 *  - NFS always sets S_NOATIME by so checking it would give a
754 	 *    bogus result
755 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
756 	 *    no point in checking those.
757 	 */
758 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
759 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
760 		need_atime = 0;
761 
762 	if (need_atime || nfs_need_revalidate_inode(inode)) {
763 		struct nfs_server *server = NFS_SERVER(inode);
764 
765 		if (!(server->flags & NFS_MOUNT_NOAC))
766 			nfs_readdirplus_parent_cache_miss(path->dentry);
767 		else
768 			nfs_readdirplus_parent_cache_hit(path->dentry);
769 		err = __nfs_revalidate_inode(server, inode);
770 	} else
771 		nfs_readdirplus_parent_cache_hit(path->dentry);
772 	if (!err) {
773 		generic_fillattr(inode, stat);
774 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
775 		if (S_ISDIR(inode->i_mode))
776 			stat->blksize = NFS_SERVER(inode)->dtsize;
777 	}
778 out:
779 	trace_nfs_getattr_exit(inode, err);
780 	return err;
781 }
782 EXPORT_SYMBOL_GPL(nfs_getattr);
783 
784 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
785 {
786 	refcount_set(&l_ctx->count, 1);
787 	l_ctx->lockowner = current->files;
788 	INIT_LIST_HEAD(&l_ctx->list);
789 	atomic_set(&l_ctx->io_count, 0);
790 }
791 
792 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
793 {
794 	struct nfs_lock_context *head = &ctx->lock_context;
795 	struct nfs_lock_context *pos = head;
796 
797 	do {
798 		if (pos->lockowner != current->files)
799 			continue;
800 		refcount_inc(&pos->count);
801 		return pos;
802 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
803 	return NULL;
804 }
805 
806 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
807 {
808 	struct nfs_lock_context *res, *new = NULL;
809 	struct inode *inode = d_inode(ctx->dentry);
810 
811 	spin_lock(&inode->i_lock);
812 	res = __nfs_find_lock_context(ctx);
813 	if (res == NULL) {
814 		spin_unlock(&inode->i_lock);
815 		new = kmalloc(sizeof(*new), GFP_KERNEL);
816 		if (new == NULL)
817 			return ERR_PTR(-ENOMEM);
818 		nfs_init_lock_context(new);
819 		spin_lock(&inode->i_lock);
820 		res = __nfs_find_lock_context(ctx);
821 		if (res == NULL) {
822 			list_add_tail(&new->list, &ctx->lock_context.list);
823 			new->open_context = ctx;
824 			res = new;
825 			new = NULL;
826 		}
827 	}
828 	spin_unlock(&inode->i_lock);
829 	kfree(new);
830 	return res;
831 }
832 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
833 
834 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
835 {
836 	struct nfs_open_context *ctx = l_ctx->open_context;
837 	struct inode *inode = d_inode(ctx->dentry);
838 
839 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
840 		return;
841 	list_del(&l_ctx->list);
842 	spin_unlock(&inode->i_lock);
843 	kfree(l_ctx);
844 }
845 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
846 
847 /**
848  * nfs_close_context - Common close_context() routine NFSv2/v3
849  * @ctx: pointer to context
850  * @is_sync: is this a synchronous close
851  *
852  * Ensure that the attributes are up to date if we're mounted
853  * with close-to-open semantics and we have cached data that will
854  * need to be revalidated on open.
855  */
856 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
857 {
858 	struct nfs_inode *nfsi;
859 	struct inode *inode;
860 	struct nfs_server *server;
861 
862 	if (!(ctx->mode & FMODE_WRITE))
863 		return;
864 	if (!is_sync)
865 		return;
866 	inode = d_inode(ctx->dentry);
867 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
868 		return;
869 	nfsi = NFS_I(inode);
870 	if (inode->i_mapping->nrpages == 0)
871 		return;
872 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
873 		return;
874 	if (!list_empty(&nfsi->open_files))
875 		return;
876 	server = NFS_SERVER(inode);
877 	if (server->flags & NFS_MOUNT_NOCTO)
878 		return;
879 	nfs_revalidate_inode(server, inode);
880 }
881 EXPORT_SYMBOL_GPL(nfs_close_context);
882 
883 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
884 						fmode_t f_mode,
885 						struct file *filp)
886 {
887 	struct nfs_open_context *ctx;
888 	struct rpc_cred *cred = rpc_lookup_cred();
889 	if (IS_ERR(cred))
890 		return ERR_CAST(cred);
891 
892 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
893 	if (!ctx) {
894 		put_rpccred(cred);
895 		return ERR_PTR(-ENOMEM);
896 	}
897 	nfs_sb_active(dentry->d_sb);
898 	ctx->dentry = dget(dentry);
899 	ctx->cred = cred;
900 	ctx->state = NULL;
901 	ctx->mode = f_mode;
902 	ctx->flags = 0;
903 	ctx->error = 0;
904 	ctx->flock_owner = (fl_owner_t)filp;
905 	nfs_init_lock_context(&ctx->lock_context);
906 	ctx->lock_context.open_context = ctx;
907 	INIT_LIST_HEAD(&ctx->list);
908 	ctx->mdsthreshold = NULL;
909 	return ctx;
910 }
911 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
912 
913 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
914 {
915 	if (ctx != NULL)
916 		refcount_inc(&ctx->lock_context.count);
917 	return ctx;
918 }
919 EXPORT_SYMBOL_GPL(get_nfs_open_context);
920 
921 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
922 {
923 	struct inode *inode = d_inode(ctx->dentry);
924 	struct super_block *sb = ctx->dentry->d_sb;
925 
926 	if (!list_empty(&ctx->list)) {
927 		if (!refcount_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
928 			return;
929 		list_del(&ctx->list);
930 		spin_unlock(&inode->i_lock);
931 	} else if (!refcount_dec_and_test(&ctx->lock_context.count))
932 		return;
933 	if (inode != NULL)
934 		NFS_PROTO(inode)->close_context(ctx, is_sync);
935 	if (ctx->cred != NULL)
936 		put_rpccred(ctx->cred);
937 	dput(ctx->dentry);
938 	nfs_sb_deactive(sb);
939 	kfree(ctx->mdsthreshold);
940 	kfree(ctx);
941 }
942 
943 void put_nfs_open_context(struct nfs_open_context *ctx)
944 {
945 	__put_nfs_open_context(ctx, 0);
946 }
947 EXPORT_SYMBOL_GPL(put_nfs_open_context);
948 
949 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
950 {
951 	__put_nfs_open_context(ctx, 1);
952 }
953 
954 /*
955  * Ensure that mmap has a recent RPC credential for use when writing out
956  * shared pages
957  */
958 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
959 {
960 	struct inode *inode = d_inode(ctx->dentry);
961 	struct nfs_inode *nfsi = NFS_I(inode);
962 
963 	spin_lock(&inode->i_lock);
964 	if (ctx->mode & FMODE_WRITE)
965 		list_add(&ctx->list, &nfsi->open_files);
966 	else
967 		list_add_tail(&ctx->list, &nfsi->open_files);
968 	spin_unlock(&inode->i_lock);
969 }
970 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
971 
972 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
973 {
974 	filp->private_data = get_nfs_open_context(ctx);
975 	if (list_empty(&ctx->list))
976 		nfs_inode_attach_open_context(ctx);
977 }
978 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
979 
980 /*
981  * Given an inode, search for an open context with the desired characteristics
982  */
983 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
984 {
985 	struct nfs_inode *nfsi = NFS_I(inode);
986 	struct nfs_open_context *pos, *ctx = NULL;
987 
988 	spin_lock(&inode->i_lock);
989 	list_for_each_entry(pos, &nfsi->open_files, list) {
990 		if (cred != NULL && pos->cred != cred)
991 			continue;
992 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
993 			continue;
994 		ctx = get_nfs_open_context(pos);
995 		break;
996 	}
997 	spin_unlock(&inode->i_lock);
998 	return ctx;
999 }
1000 
1001 void nfs_file_clear_open_context(struct file *filp)
1002 {
1003 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1004 
1005 	if (ctx) {
1006 		struct inode *inode = d_inode(ctx->dentry);
1007 
1008 		/*
1009 		 * We fatal error on write before. Try to writeback
1010 		 * every page again.
1011 		 */
1012 		if (ctx->error < 0)
1013 			invalidate_inode_pages2(inode->i_mapping);
1014 		filp->private_data = NULL;
1015 		spin_lock(&inode->i_lock);
1016 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1017 		spin_unlock(&inode->i_lock);
1018 		put_nfs_open_context_sync(ctx);
1019 	}
1020 }
1021 
1022 /*
1023  * These allocate and release file read/write context information.
1024  */
1025 int nfs_open(struct inode *inode, struct file *filp)
1026 {
1027 	struct nfs_open_context *ctx;
1028 
1029 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1030 	if (IS_ERR(ctx))
1031 		return PTR_ERR(ctx);
1032 	nfs_file_set_open_context(filp, ctx);
1033 	put_nfs_open_context(ctx);
1034 	nfs_fscache_open_file(inode, filp);
1035 	return 0;
1036 }
1037 
1038 /*
1039  * This function is called whenever some part of NFS notices that
1040  * the cached attributes have to be refreshed.
1041  */
1042 int
1043 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1044 {
1045 	int		 status = -ESTALE;
1046 	struct nfs4_label *label = NULL;
1047 	struct nfs_fattr *fattr = NULL;
1048 	struct nfs_inode *nfsi = NFS_I(inode);
1049 
1050 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1051 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1052 
1053 	trace_nfs_revalidate_inode_enter(inode);
1054 
1055 	if (is_bad_inode(inode))
1056 		goto out;
1057 	if (NFS_STALE(inode))
1058 		goto out;
1059 
1060 	/* pNFS: Attributes aren't updated until we layoutcommit */
1061 	if (S_ISREG(inode->i_mode)) {
1062 		status = pnfs_sync_inode(inode, false);
1063 		if (status)
1064 			goto out;
1065 	}
1066 
1067 	status = -ENOMEM;
1068 	fattr = nfs_alloc_fattr();
1069 	if (fattr == NULL)
1070 		goto out;
1071 
1072 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1073 
1074 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1075 	if (IS_ERR(label)) {
1076 		status = PTR_ERR(label);
1077 		goto out;
1078 	}
1079 
1080 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
1081 	if (status != 0) {
1082 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1083 			 inode->i_sb->s_id,
1084 			 (unsigned long long)NFS_FILEID(inode), status);
1085 		if (status == -ESTALE) {
1086 			nfs_zap_caches(inode);
1087 			if (!S_ISDIR(inode->i_mode))
1088 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1089 		}
1090 		goto err_out;
1091 	}
1092 
1093 	status = nfs_refresh_inode(inode, fattr);
1094 	if (status) {
1095 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1096 			 inode->i_sb->s_id,
1097 			 (unsigned long long)NFS_FILEID(inode), status);
1098 		goto err_out;
1099 	}
1100 
1101 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1102 		nfs_zap_acl_cache(inode);
1103 
1104 	nfs_setsecurity(inode, fattr, label);
1105 
1106 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1107 		inode->i_sb->s_id,
1108 		(unsigned long long)NFS_FILEID(inode));
1109 
1110 err_out:
1111 	nfs4_label_free(label);
1112 out:
1113 	nfs_free_fattr(fattr);
1114 	trace_nfs_revalidate_inode_exit(inode, status);
1115 	return status;
1116 }
1117 
1118 int nfs_attribute_cache_expired(struct inode *inode)
1119 {
1120 	if (nfs_have_delegated_attributes(inode))
1121 		return 0;
1122 	return nfs_attribute_timeout(inode);
1123 }
1124 
1125 /**
1126  * nfs_revalidate_inode - Revalidate the inode attributes
1127  * @server - pointer to nfs_server struct
1128  * @inode - pointer to inode struct
1129  *
1130  * Updates inode attribute information by retrieving the data from the server.
1131  */
1132 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1133 {
1134 	if (!nfs_need_revalidate_inode(inode))
1135 		return NFS_STALE(inode) ? -ESTALE : 0;
1136 	return __nfs_revalidate_inode(server, inode);
1137 }
1138 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1139 
1140 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1141 {
1142 	struct nfs_inode *nfsi = NFS_I(inode);
1143 	int ret;
1144 
1145 	if (mapping->nrpages != 0) {
1146 		if (S_ISREG(inode->i_mode)) {
1147 			unmap_mapping_range(mapping, 0, 0, 0);
1148 			ret = nfs_sync_mapping(mapping);
1149 			if (ret < 0)
1150 				return ret;
1151 		}
1152 		ret = invalidate_inode_pages2(mapping);
1153 		if (ret < 0)
1154 			return ret;
1155 	}
1156 	if (S_ISDIR(inode->i_mode)) {
1157 		spin_lock(&inode->i_lock);
1158 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1159 		spin_unlock(&inode->i_lock);
1160 	}
1161 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1162 	nfs_fscache_wait_on_invalidate(inode);
1163 
1164 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1165 			inode->i_sb->s_id,
1166 			(unsigned long long)NFS_FILEID(inode));
1167 	return 0;
1168 }
1169 
1170 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1171 {
1172 	return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1173 		NFS_STALE(inode);
1174 }
1175 
1176 int nfs_revalidate_mapping_rcu(struct inode *inode)
1177 {
1178 	struct nfs_inode *nfsi = NFS_I(inode);
1179 	unsigned long *bitlock = &nfsi->flags;
1180 	int ret = 0;
1181 
1182 	if (IS_SWAPFILE(inode))
1183 		goto out;
1184 	if (nfs_mapping_need_revalidate_inode(inode)) {
1185 		ret = -ECHILD;
1186 		goto out;
1187 	}
1188 	spin_lock(&inode->i_lock);
1189 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1190 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1191 		ret = -ECHILD;
1192 	spin_unlock(&inode->i_lock);
1193 out:
1194 	return ret;
1195 }
1196 
1197 /**
1198  * nfs_revalidate_mapping - Revalidate the pagecache
1199  * @inode - pointer to host inode
1200  * @mapping - pointer to mapping
1201  */
1202 int nfs_revalidate_mapping(struct inode *inode,
1203 		struct address_space *mapping)
1204 {
1205 	struct nfs_inode *nfsi = NFS_I(inode);
1206 	unsigned long *bitlock = &nfsi->flags;
1207 	int ret = 0;
1208 
1209 	/* swapfiles are not supposed to be shared. */
1210 	if (IS_SWAPFILE(inode))
1211 		goto out;
1212 
1213 	if (nfs_mapping_need_revalidate_inode(inode)) {
1214 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1215 		if (ret < 0)
1216 			goto out;
1217 	}
1218 
1219 	/*
1220 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1221 	 * invalidations that come in while we're shooting down the mappings
1222 	 * are respected. But, that leaves a race window where one revalidator
1223 	 * can clear the flag, and then another checks it before the mapping
1224 	 * gets invalidated. Fix that by serializing access to this part of
1225 	 * the function.
1226 	 *
1227 	 * At the same time, we need to allow other tasks to see whether we
1228 	 * might be in the middle of invalidating the pages, so we only set
1229 	 * the bit lock here if it looks like we're going to be doing that.
1230 	 */
1231 	for (;;) {
1232 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1233 					 nfs_wait_bit_killable, TASK_KILLABLE);
1234 		if (ret)
1235 			goto out;
1236 		spin_lock(&inode->i_lock);
1237 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1238 			spin_unlock(&inode->i_lock);
1239 			continue;
1240 		}
1241 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1242 			break;
1243 		spin_unlock(&inode->i_lock);
1244 		goto out;
1245 	}
1246 
1247 	set_bit(NFS_INO_INVALIDATING, bitlock);
1248 	smp_wmb();
1249 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1250 	spin_unlock(&inode->i_lock);
1251 	trace_nfs_invalidate_mapping_enter(inode);
1252 	ret = nfs_invalidate_mapping(inode, mapping);
1253 	trace_nfs_invalidate_mapping_exit(inode, ret);
1254 
1255 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1256 	smp_mb__after_atomic();
1257 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1258 out:
1259 	return ret;
1260 }
1261 
1262 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1263 {
1264 	struct inode *inode = &nfsi->vfs_inode;
1265 
1266 	assert_spin_locked(&inode->i_lock);
1267 
1268 	if (!S_ISREG(inode->i_mode))
1269 		return false;
1270 	if (list_empty(&nfsi->open_files))
1271 		return false;
1272 	/* Note: This relies on nfsi->open_files being ordered with writers
1273 	 *       being placed at the head of the list.
1274 	 *       See nfs_inode_attach_open_context()
1275 	 */
1276 	return (list_first_entry(&nfsi->open_files,
1277 			struct nfs_open_context,
1278 			list)->mode & FMODE_WRITE) == FMODE_WRITE;
1279 }
1280 
1281 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1282 {
1283 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1284 }
1285 
1286 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1287 {
1288 	unsigned long ret = 0;
1289 
1290 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1291 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1292 			&& inode->i_version == fattr->pre_change_attr) {
1293 		inode->i_version = fattr->change_attr;
1294 		if (S_ISDIR(inode->i_mode))
1295 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1296 		ret |= NFS_INO_INVALID_ATTR;
1297 	}
1298 	/* If we have atomic WCC data, we may update some attributes */
1299 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1300 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1301 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1302 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1303 		ret |= NFS_INO_INVALID_ATTR;
1304 	}
1305 
1306 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1307 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1308 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1309 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1310 		if (S_ISDIR(inode->i_mode))
1311 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1312 		ret |= NFS_INO_INVALID_ATTR;
1313 	}
1314 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1315 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1316 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1317 			&& !nfs_have_writebacks(inode)) {
1318 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1319 		ret |= NFS_INO_INVALID_ATTR;
1320 	}
1321 
1322 	return ret;
1323 }
1324 
1325 /**
1326  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1327  * @inode - pointer to inode
1328  * @fattr - updated attributes
1329  *
1330  * Verifies the attribute cache. If we have just changed the attributes,
1331  * so that fattr carries weak cache consistency data, then it may
1332  * also update the ctime/mtime/change_attribute.
1333  */
1334 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1335 {
1336 	struct nfs_inode *nfsi = NFS_I(inode);
1337 	loff_t cur_size, new_isize;
1338 	unsigned long invalid = 0;
1339 
1340 
1341 	if (nfs_have_delegated_attributes(inode))
1342 		return 0;
1343 	/* Has the inode gone and changed behind our back? */
1344 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1345 		return -ESTALE;
1346 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1347 		return -ESTALE;
1348 
1349 	if (!nfs_file_has_buffered_writers(nfsi)) {
1350 		/* Verify a few of the more important attributes */
1351 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && inode->i_version != fattr->change_attr)
1352 			invalid |= NFS_INO_INVALID_ATTR | NFS_INO_REVAL_PAGECACHE;
1353 
1354 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1355 			invalid |= NFS_INO_INVALID_ATTR;
1356 
1357 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&inode->i_ctime, &fattr->ctime))
1358 			invalid |= NFS_INO_INVALID_ATTR;
1359 
1360 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1361 			cur_size = i_size_read(inode);
1362 			new_isize = nfs_size_to_loff_t(fattr->size);
1363 			if (cur_size != new_isize)
1364 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1365 		}
1366 	}
1367 
1368 	/* Have any file permissions changed? */
1369 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1370 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1371 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1372 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1373 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1374 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1375 
1376 	/* Has the link count changed? */
1377 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1378 		invalid |= NFS_INO_INVALID_ATTR;
1379 
1380 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1381 		invalid |= NFS_INO_INVALID_ATIME;
1382 
1383 	if (invalid != 0)
1384 		nfs_set_cache_invalid(inode, invalid | NFS_INO_REVAL_FORCED);
1385 
1386 	nfsi->read_cache_jiffies = fattr->time_start;
1387 	return 0;
1388 }
1389 
1390 static atomic_long_t nfs_attr_generation_counter;
1391 
1392 static unsigned long nfs_read_attr_generation_counter(void)
1393 {
1394 	return atomic_long_read(&nfs_attr_generation_counter);
1395 }
1396 
1397 unsigned long nfs_inc_attr_generation_counter(void)
1398 {
1399 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1400 }
1401 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1402 
1403 void nfs_fattr_init(struct nfs_fattr *fattr)
1404 {
1405 	fattr->valid = 0;
1406 	fattr->time_start = jiffies;
1407 	fattr->gencount = nfs_inc_attr_generation_counter();
1408 	fattr->owner_name = NULL;
1409 	fattr->group_name = NULL;
1410 }
1411 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1412 
1413 /**
1414  * nfs_fattr_set_barrier
1415  * @fattr: attributes
1416  *
1417  * Used to set a barrier after an attribute was updated. This
1418  * barrier ensures that older attributes from RPC calls that may
1419  * have raced with our update cannot clobber these new values.
1420  * Note that you are still responsible for ensuring that other
1421  * operations which change the attribute on the server do not
1422  * collide.
1423  */
1424 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1425 {
1426 	fattr->gencount = nfs_inc_attr_generation_counter();
1427 }
1428 
1429 struct nfs_fattr *nfs_alloc_fattr(void)
1430 {
1431 	struct nfs_fattr *fattr;
1432 
1433 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1434 	if (fattr != NULL)
1435 		nfs_fattr_init(fattr);
1436 	return fattr;
1437 }
1438 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1439 
1440 struct nfs_fh *nfs_alloc_fhandle(void)
1441 {
1442 	struct nfs_fh *fh;
1443 
1444 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1445 	if (fh != NULL)
1446 		fh->size = 0;
1447 	return fh;
1448 }
1449 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1450 
1451 #ifdef NFS_DEBUG
1452 /*
1453  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1454  *                             in the same way that wireshark does
1455  *
1456  * @fh: file handle
1457  *
1458  * For debugging only.
1459  */
1460 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1461 {
1462 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1463 	 * not on the result */
1464 	return nfs_fhandle_hash(fh);
1465 }
1466 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1467 
1468 /*
1469  * _nfs_display_fhandle - display an NFS file handle on the console
1470  *
1471  * @fh: file handle to display
1472  * @caption: display caption
1473  *
1474  * For debugging only.
1475  */
1476 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1477 {
1478 	unsigned short i;
1479 
1480 	if (fh == NULL || fh->size == 0) {
1481 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1482 		return;
1483 	}
1484 
1485 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1486 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1487 	for (i = 0; i < fh->size; i += 16) {
1488 		__be32 *pos = (__be32 *)&fh->data[i];
1489 
1490 		switch ((fh->size - i - 1) >> 2) {
1491 		case 0:
1492 			printk(KERN_DEFAULT " %08x\n",
1493 				be32_to_cpup(pos));
1494 			break;
1495 		case 1:
1496 			printk(KERN_DEFAULT " %08x %08x\n",
1497 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1498 			break;
1499 		case 2:
1500 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1501 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1502 				be32_to_cpup(pos + 2));
1503 			break;
1504 		default:
1505 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1506 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1507 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1508 		}
1509 	}
1510 }
1511 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1512 #endif
1513 
1514 /**
1515  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1516  * @inode - pointer to inode
1517  * @fattr - attributes
1518  *
1519  * Attempt to divine whether or not an RPC call reply carrying stale
1520  * attributes got scheduled after another call carrying updated ones.
1521  *
1522  * To do so, the function first assumes that a more recent ctime means
1523  * that the attributes in fattr are newer, however it also attempt to
1524  * catch the case where ctime either didn't change, or went backwards
1525  * (if someone reset the clock on the server) by looking at whether
1526  * or not this RPC call was started after the inode was last updated.
1527  * Note also the check for wraparound of 'attr_gencount'
1528  *
1529  * The function returns 'true' if it thinks the attributes in 'fattr' are
1530  * more recent than the ones cached in the inode.
1531  *
1532  */
1533 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1534 {
1535 	const struct nfs_inode *nfsi = NFS_I(inode);
1536 
1537 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1538 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1539 }
1540 
1541 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1542 {
1543 	int ret;
1544 
1545 	trace_nfs_refresh_inode_enter(inode);
1546 
1547 	if (nfs_inode_attrs_need_update(inode, fattr))
1548 		ret = nfs_update_inode(inode, fattr);
1549 	else
1550 		ret = nfs_check_inode_attributes(inode, fattr);
1551 
1552 	trace_nfs_refresh_inode_exit(inode, ret);
1553 	return ret;
1554 }
1555 
1556 /**
1557  * nfs_refresh_inode - try to update the inode attribute cache
1558  * @inode - pointer to inode
1559  * @fattr - updated attributes
1560  *
1561  * Check that an RPC call that returned attributes has not overlapped with
1562  * other recent updates of the inode metadata, then decide whether it is
1563  * safe to do a full update of the inode attributes, or whether just to
1564  * call nfs_check_inode_attributes.
1565  */
1566 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1567 {
1568 	int status;
1569 
1570 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1571 		return 0;
1572 	spin_lock(&inode->i_lock);
1573 	status = nfs_refresh_inode_locked(inode, fattr);
1574 	spin_unlock(&inode->i_lock);
1575 
1576 	return status;
1577 }
1578 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1579 
1580 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1581 {
1582 	unsigned long invalid = NFS_INO_INVALID_ATTR;
1583 
1584 	if (S_ISDIR(inode->i_mode))
1585 		invalid |= NFS_INO_INVALID_DATA;
1586 	nfs_set_cache_invalid(inode, invalid);
1587 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1588 		return 0;
1589 	return nfs_refresh_inode_locked(inode, fattr);
1590 }
1591 
1592 /**
1593  * nfs_post_op_update_inode - try to update the inode attribute cache
1594  * @inode - pointer to inode
1595  * @fattr - updated attributes
1596  *
1597  * After an operation that has changed the inode metadata, mark the
1598  * attribute cache as being invalid, then try to update it.
1599  *
1600  * NB: if the server didn't return any post op attributes, this
1601  * function will force the retrieval of attributes before the next
1602  * NFS request.  Thus it should be used only for operations that
1603  * are expected to change one or more attributes, to avoid
1604  * unnecessary NFS requests and trips through nfs_update_inode().
1605  */
1606 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1607 {
1608 	int status;
1609 
1610 	spin_lock(&inode->i_lock);
1611 	nfs_fattr_set_barrier(fattr);
1612 	status = nfs_post_op_update_inode_locked(inode, fattr);
1613 	spin_unlock(&inode->i_lock);
1614 
1615 	return status;
1616 }
1617 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1618 
1619 /**
1620  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1621  * @inode - pointer to inode
1622  * @fattr - updated attributes
1623  *
1624  * After an operation that has changed the inode metadata, mark the
1625  * attribute cache as being invalid, then try to update it. Fake up
1626  * weak cache consistency data, if none exist.
1627  *
1628  * This function is mainly designed to be used by the ->write_done() functions.
1629  */
1630 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1631 {
1632 	int status;
1633 
1634 	/* Don't do a WCC update if these attributes are already stale */
1635 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1636 			!nfs_inode_attrs_need_update(inode, fattr)) {
1637 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1638 				| NFS_ATTR_FATTR_PRESIZE
1639 				| NFS_ATTR_FATTR_PREMTIME
1640 				| NFS_ATTR_FATTR_PRECTIME);
1641 		goto out_noforce;
1642 	}
1643 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1644 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1645 		fattr->pre_change_attr = inode->i_version;
1646 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1647 	}
1648 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1649 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1650 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1651 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1652 	}
1653 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1654 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1655 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1656 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1657 	}
1658 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1659 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1660 		fattr->pre_size = i_size_read(inode);
1661 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1662 	}
1663 out_noforce:
1664 	status = nfs_post_op_update_inode_locked(inode, fattr);
1665 	return status;
1666 }
1667 
1668 /**
1669  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1670  * @inode - pointer to inode
1671  * @fattr - updated attributes
1672  *
1673  * After an operation that has changed the inode metadata, mark the
1674  * attribute cache as being invalid, then try to update it. Fake up
1675  * weak cache consistency data, if none exist.
1676  *
1677  * This function is mainly designed to be used by the ->write_done() functions.
1678  */
1679 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1680 {
1681 	int status;
1682 
1683 	spin_lock(&inode->i_lock);
1684 	nfs_fattr_set_barrier(fattr);
1685 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1686 	spin_unlock(&inode->i_lock);
1687 	return status;
1688 }
1689 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1690 
1691 
1692 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1693 				    struct nfs_fattr *fattr)
1694 {
1695 	bool ret1 = true, ret2 = true;
1696 
1697 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1698 		ret1 = (nfsi->fileid == fattr->fileid);
1699 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1700 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1701 	return ret1 || ret2;
1702 }
1703 
1704 /*
1705  * Many nfs protocol calls return the new file attributes after
1706  * an operation.  Here we update the inode to reflect the state
1707  * of the server's inode.
1708  *
1709  * This is a bit tricky because we have to make sure all dirty pages
1710  * have been sent off to the server before calling invalidate_inode_pages.
1711  * To make sure no other process adds more write requests while we try
1712  * our best to flush them, we make them sleep during the attribute refresh.
1713  *
1714  * A very similar scenario holds for the dir cache.
1715  */
1716 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1717 {
1718 	struct nfs_server *server;
1719 	struct nfs_inode *nfsi = NFS_I(inode);
1720 	loff_t cur_isize, new_isize;
1721 	unsigned long invalid = 0;
1722 	unsigned long now = jiffies;
1723 	unsigned long save_cache_validity;
1724 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
1725 	bool cache_revalidated = true;
1726 
1727 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1728 			__func__, inode->i_sb->s_id, inode->i_ino,
1729 			nfs_display_fhandle_hash(NFS_FH(inode)),
1730 			atomic_read(&inode->i_count), fattr->valid);
1731 
1732 	if (!nfs_fileid_valid(nfsi, fattr)) {
1733 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1734 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1735 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1736 			inode->i_sb->s_id, (long long)nfsi->fileid,
1737 			(long long)fattr->fileid);
1738 		goto out_err;
1739 	}
1740 
1741 	/*
1742 	 * Make sure the inode's type hasn't changed.
1743 	 */
1744 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1745 		/*
1746 		* Big trouble! The inode has become a different object.
1747 		*/
1748 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1749 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1750 		goto out_err;
1751 	}
1752 
1753 	server = NFS_SERVER(inode);
1754 	/* Update the fsid? */
1755 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1756 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1757 			!IS_AUTOMOUNT(inode))
1758 		server->fsid = fattr->fsid;
1759 
1760 	/*
1761 	 * Update the read time so we don't revalidate too often.
1762 	 */
1763 	nfsi->read_cache_jiffies = fattr->time_start;
1764 
1765 	save_cache_validity = nfsi->cache_validity;
1766 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1767 			| NFS_INO_INVALID_ATIME
1768 			| NFS_INO_REVAL_FORCED
1769 			| NFS_INO_REVAL_PAGECACHE);
1770 
1771 	/* Do atomic weak cache consistency updates */
1772 	invalid |= nfs_wcc_update_inode(inode, fattr);
1773 
1774 	if (pnfs_layoutcommit_outstanding(inode)) {
1775 		nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1776 		cache_revalidated = false;
1777 	}
1778 
1779 	/* More cache consistency checks */
1780 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1781 		if (inode->i_version != fattr->change_attr) {
1782 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1783 					inode->i_sb->s_id, inode->i_ino);
1784 			/* Could it be a race with writeback? */
1785 			if (!have_writers) {
1786 				invalid |= NFS_INO_INVALID_ATTR
1787 					| NFS_INO_INVALID_DATA
1788 					| NFS_INO_INVALID_ACCESS
1789 					| NFS_INO_INVALID_ACL;
1790 				if (S_ISDIR(inode->i_mode))
1791 					nfs_force_lookup_revalidate(inode);
1792 			}
1793 			inode->i_version = fattr->change_attr;
1794 		}
1795 	} else {
1796 		nfsi->cache_validity |= save_cache_validity;
1797 		cache_revalidated = false;
1798 	}
1799 
1800 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1801 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1802 	} else if (server->caps & NFS_CAP_MTIME) {
1803 		nfsi->cache_validity |= save_cache_validity &
1804 				(NFS_INO_INVALID_ATTR
1805 				| NFS_INO_REVAL_FORCED);
1806 		cache_revalidated = false;
1807 	}
1808 
1809 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1810 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1811 	} else if (server->caps & NFS_CAP_CTIME) {
1812 		nfsi->cache_validity |= save_cache_validity &
1813 				(NFS_INO_INVALID_ATTR
1814 				| NFS_INO_REVAL_FORCED);
1815 		cache_revalidated = false;
1816 	}
1817 
1818 	/* Check if our cached file size is stale */
1819 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1820 		new_isize = nfs_size_to_loff_t(fattr->size);
1821 		cur_isize = i_size_read(inode);
1822 		if (new_isize != cur_isize) {
1823 			/* Do we perhaps have any outstanding writes, or has
1824 			 * the file grown beyond our last write? */
1825 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1826 				i_size_write(inode, new_isize);
1827 				if (!have_writers)
1828 					invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1829 			}
1830 			dprintk("NFS: isize change on server for file %s/%ld "
1831 					"(%Ld to %Ld)\n",
1832 					inode->i_sb->s_id,
1833 					inode->i_ino,
1834 					(long long)cur_isize,
1835 					(long long)new_isize);
1836 		}
1837 	} else {
1838 		nfsi->cache_validity |= save_cache_validity &
1839 				(NFS_INO_INVALID_ATTR
1840 				| NFS_INO_REVAL_PAGECACHE
1841 				| NFS_INO_REVAL_FORCED);
1842 		cache_revalidated = false;
1843 	}
1844 
1845 
1846 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1847 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1848 	else if (server->caps & NFS_CAP_ATIME) {
1849 		nfsi->cache_validity |= save_cache_validity &
1850 				(NFS_INO_INVALID_ATIME
1851 				| NFS_INO_REVAL_FORCED);
1852 		cache_revalidated = false;
1853 	}
1854 
1855 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1856 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1857 			umode_t newmode = inode->i_mode & S_IFMT;
1858 			newmode |= fattr->mode & S_IALLUGO;
1859 			inode->i_mode = newmode;
1860 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1861 		}
1862 	} else if (server->caps & NFS_CAP_MODE) {
1863 		nfsi->cache_validity |= save_cache_validity &
1864 				(NFS_INO_INVALID_ATTR
1865 				| NFS_INO_INVALID_ACCESS
1866 				| NFS_INO_INVALID_ACL
1867 				| NFS_INO_REVAL_FORCED);
1868 		cache_revalidated = false;
1869 	}
1870 
1871 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1872 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1873 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1874 			inode->i_uid = fattr->uid;
1875 		}
1876 	} else if (server->caps & NFS_CAP_OWNER) {
1877 		nfsi->cache_validity |= save_cache_validity &
1878 				(NFS_INO_INVALID_ATTR
1879 				| NFS_INO_INVALID_ACCESS
1880 				| NFS_INO_INVALID_ACL
1881 				| NFS_INO_REVAL_FORCED);
1882 		cache_revalidated = false;
1883 	}
1884 
1885 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1886 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1887 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1888 			inode->i_gid = fattr->gid;
1889 		}
1890 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
1891 		nfsi->cache_validity |= save_cache_validity &
1892 				(NFS_INO_INVALID_ATTR
1893 				| NFS_INO_INVALID_ACCESS
1894 				| NFS_INO_INVALID_ACL
1895 				| NFS_INO_REVAL_FORCED);
1896 		cache_revalidated = false;
1897 	}
1898 
1899 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1900 		if (inode->i_nlink != fattr->nlink) {
1901 			invalid |= NFS_INO_INVALID_ATTR;
1902 			if (S_ISDIR(inode->i_mode))
1903 				invalid |= NFS_INO_INVALID_DATA;
1904 			set_nlink(inode, fattr->nlink);
1905 		}
1906 	} else if (server->caps & NFS_CAP_NLINK) {
1907 		nfsi->cache_validity |= save_cache_validity &
1908 				(NFS_INO_INVALID_ATTR
1909 				| NFS_INO_REVAL_FORCED);
1910 		cache_revalidated = false;
1911 	}
1912 
1913 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1914 		/*
1915 		 * report the blocks in 512byte units
1916 		 */
1917 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1918 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1919 		inode->i_blocks = fattr->du.nfs2.blocks;
1920 	else
1921 		cache_revalidated = false;
1922 
1923 	/* Update attrtimeo value if we're out of the unstable period */
1924 	if (invalid & NFS_INO_INVALID_ATTR) {
1925 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1926 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1927 		nfsi->attrtimeo_timestamp = now;
1928 		/* Set barrier to be more recent than all outstanding updates */
1929 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1930 	} else {
1931 		if (cache_revalidated) {
1932 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1933 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1934 				nfsi->attrtimeo <<= 1;
1935 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1936 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1937 			}
1938 			nfsi->attrtimeo_timestamp = now;
1939 		}
1940 		/* Set the barrier to be more recent than this fattr */
1941 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1942 			nfsi->attr_gencount = fattr->gencount;
1943 	}
1944 
1945 	/* Don't declare attrcache up to date if there were no attrs! */
1946 	if (cache_revalidated)
1947 		invalid &= ~NFS_INO_INVALID_ATTR;
1948 
1949 	/* Don't invalidate the data if we were to blame */
1950 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1951 				|| S_ISLNK(inode->i_mode)))
1952 		invalid &= ~NFS_INO_INVALID_DATA;
1953 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1954 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1955 		nfs_set_cache_invalid(inode, invalid);
1956 
1957 	return 0;
1958  out_err:
1959 	/*
1960 	 * No need to worry about unhashing the dentry, as the
1961 	 * lookup validation will know that the inode is bad.
1962 	 * (But we fall through to invalidate the caches.)
1963 	 */
1964 	nfs_invalidate_inode(inode);
1965 	return -ESTALE;
1966 }
1967 
1968 struct inode *nfs_alloc_inode(struct super_block *sb)
1969 {
1970 	struct nfs_inode *nfsi;
1971 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1972 	if (!nfsi)
1973 		return NULL;
1974 	nfsi->flags = 0UL;
1975 	nfsi->cache_validity = 0UL;
1976 #if IS_ENABLED(CONFIG_NFS_V4)
1977 	nfsi->nfs4_acl = NULL;
1978 #endif /* CONFIG_NFS_V4 */
1979 	return &nfsi->vfs_inode;
1980 }
1981 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1982 
1983 static void nfs_i_callback(struct rcu_head *head)
1984 {
1985 	struct inode *inode = container_of(head, struct inode, i_rcu);
1986 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1987 }
1988 
1989 void nfs_destroy_inode(struct inode *inode)
1990 {
1991 	call_rcu(&inode->i_rcu, nfs_i_callback);
1992 }
1993 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1994 
1995 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1996 {
1997 #if IS_ENABLED(CONFIG_NFS_V4)
1998 	INIT_LIST_HEAD(&nfsi->open_states);
1999 	nfsi->delegation = NULL;
2000 	init_rwsem(&nfsi->rwsem);
2001 	nfsi->layout = NULL;
2002 #endif
2003 }
2004 
2005 static void init_once(void *foo)
2006 {
2007 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2008 
2009 	inode_init_once(&nfsi->vfs_inode);
2010 	INIT_LIST_HEAD(&nfsi->open_files);
2011 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2012 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2013 	INIT_LIST_HEAD(&nfsi->commit_info.list);
2014 	atomic_long_set(&nfsi->nrequests, 0);
2015 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
2016 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
2017 	init_rwsem(&nfsi->rmdir_sem);
2018 	mutex_init(&nfsi->commit_mutex);
2019 	nfs4_init_once(nfsi);
2020 }
2021 
2022 static int __init nfs_init_inodecache(void)
2023 {
2024 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2025 					     sizeof(struct nfs_inode),
2026 					     0, (SLAB_RECLAIM_ACCOUNT|
2027 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2028 					     init_once);
2029 	if (nfs_inode_cachep == NULL)
2030 		return -ENOMEM;
2031 
2032 	return 0;
2033 }
2034 
2035 static void nfs_destroy_inodecache(void)
2036 {
2037 	/*
2038 	 * Make sure all delayed rcu free inodes are flushed before we
2039 	 * destroy cache.
2040 	 */
2041 	rcu_barrier();
2042 	kmem_cache_destroy(nfs_inode_cachep);
2043 }
2044 
2045 struct workqueue_struct *nfsiod_workqueue;
2046 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2047 
2048 /*
2049  * start up the nfsiod workqueue
2050  */
2051 static int nfsiod_start(void)
2052 {
2053 	struct workqueue_struct *wq;
2054 	dprintk("RPC:       creating workqueue nfsiod\n");
2055 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2056 	if (wq == NULL)
2057 		return -ENOMEM;
2058 	nfsiod_workqueue = wq;
2059 	return 0;
2060 }
2061 
2062 /*
2063  * Destroy the nfsiod workqueue
2064  */
2065 static void nfsiod_stop(void)
2066 {
2067 	struct workqueue_struct *wq;
2068 
2069 	wq = nfsiod_workqueue;
2070 	if (wq == NULL)
2071 		return;
2072 	nfsiod_workqueue = NULL;
2073 	destroy_workqueue(wq);
2074 }
2075 
2076 unsigned int nfs_net_id;
2077 EXPORT_SYMBOL_GPL(nfs_net_id);
2078 
2079 static int nfs_net_init(struct net *net)
2080 {
2081 	nfs_clients_init(net);
2082 	return nfs_fs_proc_net_init(net);
2083 }
2084 
2085 static void nfs_net_exit(struct net *net)
2086 {
2087 	struct nfs_net *nn = net_generic(net, nfs_net_id);
2088 
2089 	nfs_fs_proc_net_exit(net);
2090 	nfs_cleanup_cb_ident_idr(net);
2091 	WARN_ON_ONCE(!list_empty(&nn->nfs_client_list));
2092 	WARN_ON_ONCE(!list_empty(&nn->nfs_volume_list));
2093 }
2094 
2095 static struct pernet_operations nfs_net_ops = {
2096 	.init = nfs_net_init,
2097 	.exit = nfs_net_exit,
2098 	.id   = &nfs_net_id,
2099 	.size = sizeof(struct nfs_net),
2100 };
2101 
2102 /*
2103  * Initialize NFS
2104  */
2105 static int __init init_nfs_fs(void)
2106 {
2107 	int err;
2108 
2109 	err = register_pernet_subsys(&nfs_net_ops);
2110 	if (err < 0)
2111 		goto out9;
2112 
2113 	err = nfs_fscache_register();
2114 	if (err < 0)
2115 		goto out8;
2116 
2117 	err = nfsiod_start();
2118 	if (err)
2119 		goto out7;
2120 
2121 	err = nfs_fs_proc_init();
2122 	if (err)
2123 		goto out6;
2124 
2125 	err = nfs_init_nfspagecache();
2126 	if (err)
2127 		goto out5;
2128 
2129 	err = nfs_init_inodecache();
2130 	if (err)
2131 		goto out4;
2132 
2133 	err = nfs_init_readpagecache();
2134 	if (err)
2135 		goto out3;
2136 
2137 	err = nfs_init_writepagecache();
2138 	if (err)
2139 		goto out2;
2140 
2141 	err = nfs_init_directcache();
2142 	if (err)
2143 		goto out1;
2144 
2145 	rpc_proc_register(&init_net, &nfs_rpcstat);
2146 
2147 	err = register_nfs_fs();
2148 	if (err)
2149 		goto out0;
2150 
2151 	return 0;
2152 out0:
2153 	rpc_proc_unregister(&init_net, "nfs");
2154 	nfs_destroy_directcache();
2155 out1:
2156 	nfs_destroy_writepagecache();
2157 out2:
2158 	nfs_destroy_readpagecache();
2159 out3:
2160 	nfs_destroy_inodecache();
2161 out4:
2162 	nfs_destroy_nfspagecache();
2163 out5:
2164 	nfs_fs_proc_exit();
2165 out6:
2166 	nfsiod_stop();
2167 out7:
2168 	nfs_fscache_unregister();
2169 out8:
2170 	unregister_pernet_subsys(&nfs_net_ops);
2171 out9:
2172 	return err;
2173 }
2174 
2175 static void __exit exit_nfs_fs(void)
2176 {
2177 	nfs_destroy_directcache();
2178 	nfs_destroy_writepagecache();
2179 	nfs_destroy_readpagecache();
2180 	nfs_destroy_inodecache();
2181 	nfs_destroy_nfspagecache();
2182 	nfs_fscache_unregister();
2183 	unregister_pernet_subsys(&nfs_net_ops);
2184 	rpc_proc_unregister(&init_net, "nfs");
2185 	unregister_nfs_fs();
2186 	nfs_fs_proc_exit();
2187 	nfsiod_stop();
2188 }
2189 
2190 /* Not quite true; I just maintain it */
2191 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2192 MODULE_LICENSE("GPL");
2193 module_param(enable_ino64, bool, 0644);
2194 
2195 module_init(init_nfs_fs)
2196 module_exit(exit_nfs_fs)
2197