xref: /linux/fs/nfs/inode.c (revision d2912cb15bdda8ba4a5dd73396ad62641af2f520)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/nfs/inode.c
4  *
5  *  Copyright (C) 1992  Rick Sladkey
6  *
7  *  nfs inode and superblock handling functions
8  *
9  *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
10  *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
11  *
12  *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
13  *  J.S.Peatfield@damtp.cam.ac.uk
14  *
15  */
16 
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/sched/signal.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/sunrpc/metrics.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/nfs4_mount.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 #include <linux/compat.h>
41 #include <linux/freezer.h>
42 #include <linux/uaccess.h>
43 #include <linux/iversion.h>
44 
45 #include "nfs4_fs.h"
46 #include "callback.h"
47 #include "delegation.h"
48 #include "iostat.h"
49 #include "internal.h"
50 #include "fscache.h"
51 #include "pnfs.h"
52 #include "nfs.h"
53 #include "netns.h"
54 
55 #include "nfstrace.h"
56 
57 #define NFSDBG_FACILITY		NFSDBG_VFS
58 
59 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
60 
61 /* Default is to see 64-bit inode numbers */
62 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
63 
64 static void nfs_invalidate_inode(struct inode *);
65 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
66 
67 static struct kmem_cache * nfs_inode_cachep;
68 
69 static inline unsigned long
70 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71 {
72 	return nfs_fileid_to_ino_t(fattr->fileid);
73 }
74 
75 static int nfs_wait_killable(int mode)
76 {
77 	freezable_schedule_unsafe();
78 	if (signal_pending_state(mode, current))
79 		return -ERESTARTSYS;
80 	return 0;
81 }
82 
83 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
84 {
85 	return nfs_wait_killable(mode);
86 }
87 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
88 
89 /**
90  * nfs_compat_user_ino64 - returns the user-visible inode number
91  * @fileid: 64-bit fileid
92  *
93  * This function returns a 32-bit inode number if the boot parameter
94  * nfs.enable_ino64 is zero.
95  */
96 u64 nfs_compat_user_ino64(u64 fileid)
97 {
98 #ifdef CONFIG_COMPAT
99 	compat_ulong_t ino;
100 #else
101 	unsigned long ino;
102 #endif
103 
104 	if (enable_ino64)
105 		return fileid;
106 	ino = fileid;
107 	if (sizeof(ino) < sizeof(fileid))
108 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
109 	return ino;
110 }
111 
112 int nfs_drop_inode(struct inode *inode)
113 {
114 	return NFS_STALE(inode) || generic_drop_inode(inode);
115 }
116 EXPORT_SYMBOL_GPL(nfs_drop_inode);
117 
118 void nfs_clear_inode(struct inode *inode)
119 {
120 	/*
121 	 * The following should never happen...
122 	 */
123 	WARN_ON_ONCE(nfs_have_writebacks(inode));
124 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
125 	nfs_zap_acl_cache(inode);
126 	nfs_access_zap_cache(inode);
127 	nfs_fscache_clear_inode(inode);
128 }
129 EXPORT_SYMBOL_GPL(nfs_clear_inode);
130 
131 void nfs_evict_inode(struct inode *inode)
132 {
133 	truncate_inode_pages_final(&inode->i_data);
134 	clear_inode(inode);
135 	nfs_clear_inode(inode);
136 }
137 
138 int nfs_sync_inode(struct inode *inode)
139 {
140 	inode_dio_wait(inode);
141 	return nfs_wb_all(inode);
142 }
143 EXPORT_SYMBOL_GPL(nfs_sync_inode);
144 
145 /**
146  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
147  * @mapping: pointer to struct address_space
148  */
149 int nfs_sync_mapping(struct address_space *mapping)
150 {
151 	int ret = 0;
152 
153 	if (mapping->nrpages != 0) {
154 		unmap_mapping_range(mapping, 0, 0, 0);
155 		ret = nfs_wb_all(mapping->host);
156 	}
157 	return ret;
158 }
159 
160 static int nfs_attribute_timeout(struct inode *inode)
161 {
162 	struct nfs_inode *nfsi = NFS_I(inode);
163 
164 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
165 }
166 
167 static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
168 {
169 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
170 
171 	/* Special case for the pagecache or access cache */
172 	if (flags == NFS_INO_REVAL_PAGECACHE &&
173 	    !(cache_validity & NFS_INO_REVAL_FORCED))
174 		return false;
175 	return (cache_validity & flags) != 0;
176 }
177 
178 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
179 {
180 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
181 
182 	if ((cache_validity & flags) != 0)
183 		return true;
184 	if (nfs_attribute_timeout(inode))
185 		return true;
186 	return false;
187 }
188 
189 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
190 {
191 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
192 		return nfs_check_cache_invalid_delegated(inode, flags);
193 
194 	return nfs_check_cache_invalid_not_delegated(inode, flags);
195 }
196 
197 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
198 {
199 	struct nfs_inode *nfsi = NFS_I(inode);
200 	bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
201 
202 	if (have_delegation) {
203 		if (!(flags & NFS_INO_REVAL_FORCED))
204 			flags &= ~NFS_INO_INVALID_OTHER;
205 		flags &= ~(NFS_INO_INVALID_CHANGE
206 				| NFS_INO_INVALID_SIZE
207 				| NFS_INO_REVAL_PAGECACHE);
208 	}
209 
210 	if (inode->i_mapping->nrpages == 0)
211 		flags &= ~NFS_INO_INVALID_DATA;
212 	nfsi->cache_validity |= flags;
213 	if (flags & NFS_INO_INVALID_DATA)
214 		nfs_fscache_invalidate(inode);
215 }
216 
217 /*
218  * Invalidate the local caches
219  */
220 static void nfs_zap_caches_locked(struct inode *inode)
221 {
222 	struct nfs_inode *nfsi = NFS_I(inode);
223 	int mode = inode->i_mode;
224 
225 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
226 
227 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
228 	nfsi->attrtimeo_timestamp = jiffies;
229 
230 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
231 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
232 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
233 					| NFS_INO_INVALID_DATA
234 					| NFS_INO_INVALID_ACCESS
235 					| NFS_INO_INVALID_ACL
236 					| NFS_INO_REVAL_PAGECACHE);
237 	} else
238 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
239 					| NFS_INO_INVALID_ACCESS
240 					| NFS_INO_INVALID_ACL
241 					| NFS_INO_REVAL_PAGECACHE);
242 	nfs_zap_label_cache_locked(nfsi);
243 }
244 
245 void nfs_zap_caches(struct inode *inode)
246 {
247 	spin_lock(&inode->i_lock);
248 	nfs_zap_caches_locked(inode);
249 	spin_unlock(&inode->i_lock);
250 }
251 
252 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
253 {
254 	if (mapping->nrpages != 0) {
255 		spin_lock(&inode->i_lock);
256 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
257 		spin_unlock(&inode->i_lock);
258 	}
259 }
260 
261 void nfs_zap_acl_cache(struct inode *inode)
262 {
263 	void (*clear_acl_cache)(struct inode *);
264 
265 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
266 	if (clear_acl_cache != NULL)
267 		clear_acl_cache(inode);
268 	spin_lock(&inode->i_lock);
269 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
270 	spin_unlock(&inode->i_lock);
271 }
272 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
273 
274 void nfs_invalidate_atime(struct inode *inode)
275 {
276 	spin_lock(&inode->i_lock);
277 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
278 	spin_unlock(&inode->i_lock);
279 }
280 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
281 
282 /*
283  * Invalidate, but do not unhash, the inode.
284  * NB: must be called with inode->i_lock held!
285  */
286 static void nfs_invalidate_inode(struct inode *inode)
287 {
288 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
289 	nfs_zap_caches_locked(inode);
290 }
291 
292 struct nfs_find_desc {
293 	struct nfs_fh		*fh;
294 	struct nfs_fattr	*fattr;
295 };
296 
297 /*
298  * In NFSv3 we can have 64bit inode numbers. In order to support
299  * this, and re-exported directories (also seen in NFSv2)
300  * we are forced to allow 2 different inodes to have the same
301  * i_ino.
302  */
303 static int
304 nfs_find_actor(struct inode *inode, void *opaque)
305 {
306 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
307 	struct nfs_fh		*fh = desc->fh;
308 	struct nfs_fattr	*fattr = desc->fattr;
309 
310 	if (NFS_FILEID(inode) != fattr->fileid)
311 		return 0;
312 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
313 		return 0;
314 	if (nfs_compare_fh(NFS_FH(inode), fh))
315 		return 0;
316 	if (is_bad_inode(inode) || NFS_STALE(inode))
317 		return 0;
318 	return 1;
319 }
320 
321 static int
322 nfs_init_locked(struct inode *inode, void *opaque)
323 {
324 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
325 	struct nfs_fattr	*fattr = desc->fattr;
326 
327 	set_nfs_fileid(inode, fattr->fileid);
328 	inode->i_mode = fattr->mode;
329 	nfs_copy_fh(NFS_FH(inode), desc->fh);
330 	return 0;
331 }
332 
333 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
334 static void nfs_clear_label_invalid(struct inode *inode)
335 {
336 	spin_lock(&inode->i_lock);
337 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
338 	spin_unlock(&inode->i_lock);
339 }
340 
341 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
342 					struct nfs4_label *label)
343 {
344 	int error;
345 
346 	if (label == NULL)
347 		return;
348 
349 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
350 		error = security_inode_notifysecctx(inode, label->label,
351 				label->len);
352 		if (error)
353 			printk(KERN_ERR "%s() %s %d "
354 					"security_inode_notifysecctx() %d\n",
355 					__func__,
356 					(char *)label->label,
357 					label->len, error);
358 		nfs_clear_label_invalid(inode);
359 	}
360 }
361 
362 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
363 {
364 	struct nfs4_label *label = NULL;
365 	int minor_version = server->nfs_client->cl_minorversion;
366 
367 	if (minor_version < 2)
368 		return label;
369 
370 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
371 		return label;
372 
373 	label = kzalloc(sizeof(struct nfs4_label), flags);
374 	if (label == NULL)
375 		return ERR_PTR(-ENOMEM);
376 
377 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
378 	if (label->label == NULL) {
379 		kfree(label);
380 		return ERR_PTR(-ENOMEM);
381 	}
382 	label->len = NFS4_MAXLABELLEN;
383 
384 	return label;
385 }
386 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
387 #else
388 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
389 					struct nfs4_label *label)
390 {
391 }
392 #endif
393 EXPORT_SYMBOL_GPL(nfs_setsecurity);
394 
395 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
396 struct inode *
397 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
398 {
399 	struct nfs_find_desc desc = {
400 		.fh	= fh,
401 		.fattr	= fattr,
402 	};
403 	struct inode *inode;
404 	unsigned long hash;
405 
406 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
407 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
408 		return NULL;
409 
410 	hash = nfs_fattr_to_ino_t(fattr);
411 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
412 
413 	dprintk("%s: returning %p\n", __func__, inode);
414 	return inode;
415 }
416 
417 /*
418  * This is our front-end to iget that looks up inodes by file handle
419  * instead of inode number.
420  */
421 struct inode *
422 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
423 {
424 	struct nfs_find_desc desc = {
425 		.fh	= fh,
426 		.fattr	= fattr
427 	};
428 	struct inode *inode = ERR_PTR(-ENOENT);
429 	unsigned long hash;
430 
431 	nfs_attr_check_mountpoint(sb, fattr);
432 
433 	if (nfs_attr_use_mounted_on_fileid(fattr))
434 		fattr->fileid = fattr->mounted_on_fileid;
435 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
436 		goto out_no_inode;
437 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
438 		goto out_no_inode;
439 
440 	hash = nfs_fattr_to_ino_t(fattr);
441 
442 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
443 	if (inode == NULL) {
444 		inode = ERR_PTR(-ENOMEM);
445 		goto out_no_inode;
446 	}
447 
448 	if (inode->i_state & I_NEW) {
449 		struct nfs_inode *nfsi = NFS_I(inode);
450 		unsigned long now = jiffies;
451 
452 		/* We set i_ino for the few things that still rely on it,
453 		 * such as stat(2) */
454 		inode->i_ino = hash;
455 
456 		/* We can't support update_atime(), since the server will reset it */
457 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
458 		inode->i_mode = fattr->mode;
459 		nfsi->cache_validity = 0;
460 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
461 				&& nfs_server_capable(inode, NFS_CAP_MODE))
462 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
463 		/* Why so? Because we want revalidate for devices/FIFOs, and
464 		 * that's precisely what we have in nfs_file_inode_operations.
465 		 */
466 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
467 		if (S_ISREG(inode->i_mode)) {
468 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
469 			inode->i_data.a_ops = &nfs_file_aops;
470 		} else if (S_ISDIR(inode->i_mode)) {
471 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
472 			inode->i_fop = &nfs_dir_operations;
473 			inode->i_data.a_ops = &nfs_dir_aops;
474 			/* Deal with crossing mountpoints */
475 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
476 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
477 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
478 					inode->i_op = &nfs_referral_inode_operations;
479 				else
480 					inode->i_op = &nfs_mountpoint_inode_operations;
481 				inode->i_fop = NULL;
482 				inode->i_flags |= S_AUTOMOUNT;
483 			}
484 		} else if (S_ISLNK(inode->i_mode)) {
485 			inode->i_op = &nfs_symlink_inode_operations;
486 			inode_nohighmem(inode);
487 		} else
488 			init_special_inode(inode, inode->i_mode, fattr->rdev);
489 
490 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
491 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
492 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
493 		inode_set_iversion_raw(inode, 0);
494 		inode->i_size = 0;
495 		clear_nlink(inode);
496 		inode->i_uid = make_kuid(&init_user_ns, -2);
497 		inode->i_gid = make_kgid(&init_user_ns, -2);
498 		inode->i_blocks = 0;
499 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
500 		nfsi->write_io = 0;
501 		nfsi->read_io = 0;
502 
503 		nfsi->read_cache_jiffies = fattr->time_start;
504 		nfsi->attr_gencount = fattr->gencount;
505 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
506 			inode->i_atime = timespec_to_timespec64(fattr->atime);
507 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
508 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
509 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
510 			inode->i_mtime = timespec_to_timespec64(fattr->mtime);
511 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
512 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
513 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
514 			inode->i_ctime = timespec_to_timespec64(fattr->ctime);
515 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
516 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
517 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
518 			inode_set_iversion_raw(inode, fattr->change_attr);
519 		else
520 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
521 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
522 			inode->i_size = nfs_size_to_loff_t(fattr->size);
523 		else
524 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
525 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
526 			set_nlink(inode, fattr->nlink);
527 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
528 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
529 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
530 			inode->i_uid = fattr->uid;
531 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
532 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
533 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
534 			inode->i_gid = fattr->gid;
535 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
536 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
537 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
538 			inode->i_blocks = fattr->du.nfs2.blocks;
539 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
540 			/*
541 			 * report the blocks in 512byte units
542 			 */
543 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
544 		}
545 
546 		if (nfsi->cache_validity != 0)
547 			nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
548 
549 		nfs_setsecurity(inode, fattr, label);
550 
551 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
552 		nfsi->attrtimeo_timestamp = now;
553 		nfsi->access_cache = RB_ROOT;
554 
555 		nfs_fscache_init_inode(inode);
556 
557 		unlock_new_inode(inode);
558 	} else {
559 		int err = nfs_refresh_inode(inode, fattr);
560 		if (err < 0) {
561 			iput(inode);
562 			inode = ERR_PTR(err);
563 			goto out_no_inode;
564 		}
565 	}
566 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
567 		inode->i_sb->s_id,
568 		(unsigned long long)NFS_FILEID(inode),
569 		nfs_display_fhandle_hash(fh),
570 		atomic_read(&inode->i_count));
571 
572 out:
573 	return inode;
574 
575 out_no_inode:
576 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
577 	goto out;
578 }
579 EXPORT_SYMBOL_GPL(nfs_fhget);
580 
581 #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)
582 
583 int
584 nfs_setattr(struct dentry *dentry, struct iattr *attr)
585 {
586 	struct inode *inode = d_inode(dentry);
587 	struct nfs_fattr *fattr;
588 	int error = 0;
589 
590 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
591 
592 	/* skip mode change if it's just for clearing setuid/setgid */
593 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
594 		attr->ia_valid &= ~ATTR_MODE;
595 
596 	if (attr->ia_valid & ATTR_SIZE) {
597 		BUG_ON(!S_ISREG(inode->i_mode));
598 
599 		error = inode_newsize_ok(inode, attr->ia_size);
600 		if (error)
601 			return error;
602 
603 		if (attr->ia_size == i_size_read(inode))
604 			attr->ia_valid &= ~ATTR_SIZE;
605 	}
606 
607 	/* Optimization: if the end result is no change, don't RPC */
608 	attr->ia_valid &= NFS_VALID_ATTRS;
609 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
610 		return 0;
611 
612 	trace_nfs_setattr_enter(inode);
613 
614 	/* Write all dirty data */
615 	if (S_ISREG(inode->i_mode))
616 		nfs_sync_inode(inode);
617 
618 	fattr = nfs_alloc_fattr();
619 	if (fattr == NULL) {
620 		error = -ENOMEM;
621 		goto out;
622 	}
623 
624 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
625 	if (error == 0)
626 		error = nfs_refresh_inode(inode, fattr);
627 	nfs_free_fattr(fattr);
628 out:
629 	trace_nfs_setattr_exit(inode, error);
630 	return error;
631 }
632 EXPORT_SYMBOL_GPL(nfs_setattr);
633 
634 /**
635  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
636  * @inode: inode of the file used
637  * @offset: file offset to start truncating
638  *
639  * This is a copy of the common vmtruncate, but with the locking
640  * corrected to take into account the fact that NFS requires
641  * inode->i_size to be updated under the inode->i_lock.
642  * Note: must be called with inode->i_lock held!
643  */
644 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
645 {
646 	int err;
647 
648 	err = inode_newsize_ok(inode, offset);
649 	if (err)
650 		goto out;
651 
652 	i_size_write(inode, offset);
653 	/* Optimisation */
654 	if (offset == 0)
655 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
656 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
657 
658 	spin_unlock(&inode->i_lock);
659 	truncate_pagecache(inode, offset);
660 	spin_lock(&inode->i_lock);
661 out:
662 	return err;
663 }
664 
665 /**
666  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
667  * @inode: pointer to struct inode
668  * @attr: pointer to struct iattr
669  * @fattr: pointer to struct nfs_fattr
670  *
671  * Note: we do this in the *proc.c in order to ensure that
672  *       it works for things like exclusive creates too.
673  */
674 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
675 		struct nfs_fattr *fattr)
676 {
677 	/* Barrier: bump the attribute generation count. */
678 	nfs_fattr_set_barrier(fattr);
679 
680 	spin_lock(&inode->i_lock);
681 	NFS_I(inode)->attr_gencount = fattr->gencount;
682 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
683 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
684 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
685 		nfs_vmtruncate(inode, attr->ia_size);
686 	}
687 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
688 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
689 		if ((attr->ia_valid & ATTR_MODE) != 0) {
690 			int mode = attr->ia_mode & S_IALLUGO;
691 			mode |= inode->i_mode & ~S_IALLUGO;
692 			inode->i_mode = mode;
693 		}
694 		if ((attr->ia_valid & ATTR_UID) != 0)
695 			inode->i_uid = attr->ia_uid;
696 		if ((attr->ia_valid & ATTR_GID) != 0)
697 			inode->i_gid = attr->ia_gid;
698 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
699 			inode->i_ctime = timespec_to_timespec64(fattr->ctime);
700 		else
701 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
702 					| NFS_INO_INVALID_CTIME);
703 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
704 				| NFS_INO_INVALID_ACL);
705 	}
706 	if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
707 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
708 				| NFS_INO_INVALID_CTIME);
709 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
710 			inode->i_atime = timespec_to_timespec64(fattr->atime);
711 		else if (attr->ia_valid & ATTR_ATIME_SET)
712 			inode->i_atime = attr->ia_atime;
713 		else
714 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
715 
716 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
717 			inode->i_ctime = timespec_to_timespec64(fattr->ctime);
718 		else
719 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
720 					| NFS_INO_INVALID_CTIME);
721 	}
722 	if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
723 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
724 				| NFS_INO_INVALID_CTIME);
725 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
726 			inode->i_mtime = timespec_to_timespec64(fattr->mtime);
727 		else if (attr->ia_valid & ATTR_MTIME_SET)
728 			inode->i_mtime = attr->ia_mtime;
729 		else
730 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
731 
732 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
733 			inode->i_ctime = timespec_to_timespec64(fattr->ctime);
734 		else
735 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
736 					| NFS_INO_INVALID_CTIME);
737 	}
738 	if (fattr->valid)
739 		nfs_update_inode(inode, fattr);
740 	spin_unlock(&inode->i_lock);
741 }
742 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
743 
744 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
745 {
746 	struct dentry *parent;
747 
748 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
749 		return;
750 	parent = dget_parent(dentry);
751 	nfs_force_use_readdirplus(d_inode(parent));
752 	dput(parent);
753 }
754 
755 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
756 {
757 	struct dentry *parent;
758 
759 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
760 		return;
761 	parent = dget_parent(dentry);
762 	nfs_advise_use_readdirplus(d_inode(parent));
763 	dput(parent);
764 }
765 
766 static bool nfs_need_revalidate_inode(struct inode *inode)
767 {
768 	if (NFS_I(inode)->cache_validity &
769 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
770 		return true;
771 	if (nfs_attribute_cache_expired(inode))
772 		return true;
773 	return false;
774 }
775 
776 int nfs_getattr(const struct path *path, struct kstat *stat,
777 		u32 request_mask, unsigned int query_flags)
778 {
779 	struct inode *inode = d_inode(path->dentry);
780 	struct nfs_server *server = NFS_SERVER(inode);
781 	unsigned long cache_validity;
782 	int err = 0;
783 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
784 	bool do_update = false;
785 
786 	trace_nfs_getattr_enter(inode);
787 
788 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync)
789 		goto out_no_update;
790 
791 	/* Flush out writes to the server in order to update c/mtime.  */
792 	if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
793 			S_ISREG(inode->i_mode)) {
794 		err = filemap_write_and_wait(inode->i_mapping);
795 		if (err)
796 			goto out;
797 	}
798 
799 	/*
800 	 * We may force a getattr if the user cares about atime.
801 	 *
802 	 * Note that we only have to check the vfsmount flags here:
803 	 *  - NFS always sets S_NOATIME by so checking it would give a
804 	 *    bogus result
805 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
806 	 *    no point in checking those.
807 	 */
808 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
809 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
810 		request_mask &= ~STATX_ATIME;
811 
812 	/* Is the user requesting attributes that might need revalidation? */
813 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
814 					STATX_MTIME|STATX_UID|STATX_GID|
815 					STATX_SIZE|STATX_BLOCKS)))
816 		goto out_no_revalidate;
817 
818 	/* Check whether the cached attributes are stale */
819 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
820 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
821 	do_update |= cache_validity &
822 		(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
823 	if (request_mask & STATX_ATIME)
824 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
825 	if (request_mask & (STATX_CTIME|STATX_MTIME))
826 		do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
827 	if (do_update) {
828 		/* Update the attribute cache */
829 		if (!(server->flags & NFS_MOUNT_NOAC))
830 			nfs_readdirplus_parent_cache_miss(path->dentry);
831 		else
832 			nfs_readdirplus_parent_cache_hit(path->dentry);
833 		err = __nfs_revalidate_inode(server, inode);
834 		if (err)
835 			goto out;
836 	} else
837 		nfs_readdirplus_parent_cache_hit(path->dentry);
838 out_no_revalidate:
839 	/* Only return attributes that were revalidated. */
840 	stat->result_mask &= request_mask;
841 out_no_update:
842 	generic_fillattr(inode, stat);
843 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
844 	if (S_ISDIR(inode->i_mode))
845 		stat->blksize = NFS_SERVER(inode)->dtsize;
846 out:
847 	trace_nfs_getattr_exit(inode, err);
848 	return err;
849 }
850 EXPORT_SYMBOL_GPL(nfs_getattr);
851 
852 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
853 {
854 	refcount_set(&l_ctx->count, 1);
855 	l_ctx->lockowner = current->files;
856 	INIT_LIST_HEAD(&l_ctx->list);
857 	atomic_set(&l_ctx->io_count, 0);
858 }
859 
860 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
861 {
862 	struct nfs_lock_context *pos;
863 
864 	list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
865 		if (pos->lockowner != current->files)
866 			continue;
867 		if (refcount_inc_not_zero(&pos->count))
868 			return pos;
869 	}
870 	return NULL;
871 }
872 
873 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
874 {
875 	struct nfs_lock_context *res, *new = NULL;
876 	struct inode *inode = d_inode(ctx->dentry);
877 
878 	rcu_read_lock();
879 	res = __nfs_find_lock_context(ctx);
880 	rcu_read_unlock();
881 	if (res == NULL) {
882 		new = kmalloc(sizeof(*new), GFP_KERNEL);
883 		if (new == NULL)
884 			return ERR_PTR(-ENOMEM);
885 		nfs_init_lock_context(new);
886 		spin_lock(&inode->i_lock);
887 		res = __nfs_find_lock_context(ctx);
888 		if (res == NULL) {
889 			new->open_context = get_nfs_open_context(ctx);
890 			if (new->open_context) {
891 				list_add_tail_rcu(&new->list,
892 						&ctx->lock_context.list);
893 				res = new;
894 				new = NULL;
895 			} else
896 				res = ERR_PTR(-EBADF);
897 		}
898 		spin_unlock(&inode->i_lock);
899 		kfree(new);
900 	}
901 	return res;
902 }
903 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
904 
905 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
906 {
907 	struct nfs_open_context *ctx = l_ctx->open_context;
908 	struct inode *inode = d_inode(ctx->dentry);
909 
910 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
911 		return;
912 	list_del_rcu(&l_ctx->list);
913 	spin_unlock(&inode->i_lock);
914 	put_nfs_open_context(ctx);
915 	kfree_rcu(l_ctx, rcu_head);
916 }
917 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
918 
919 /**
920  * nfs_close_context - Common close_context() routine NFSv2/v3
921  * @ctx: pointer to context
922  * @is_sync: is this a synchronous close
923  *
924  * Ensure that the attributes are up to date if we're mounted
925  * with close-to-open semantics and we have cached data that will
926  * need to be revalidated on open.
927  */
928 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
929 {
930 	struct nfs_inode *nfsi;
931 	struct inode *inode;
932 	struct nfs_server *server;
933 
934 	if (!(ctx->mode & FMODE_WRITE))
935 		return;
936 	if (!is_sync)
937 		return;
938 	inode = d_inode(ctx->dentry);
939 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
940 		return;
941 	nfsi = NFS_I(inode);
942 	if (inode->i_mapping->nrpages == 0)
943 		return;
944 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
945 		return;
946 	if (!list_empty(&nfsi->open_files))
947 		return;
948 	server = NFS_SERVER(inode);
949 	if (server->flags & NFS_MOUNT_NOCTO)
950 		return;
951 	nfs_revalidate_inode(server, inode);
952 }
953 EXPORT_SYMBOL_GPL(nfs_close_context);
954 
955 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
956 						fmode_t f_mode,
957 						struct file *filp)
958 {
959 	struct nfs_open_context *ctx;
960 	const struct cred *cred = get_current_cred();
961 
962 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
963 	if (!ctx) {
964 		put_cred(cred);
965 		return ERR_PTR(-ENOMEM);
966 	}
967 	nfs_sb_active(dentry->d_sb);
968 	ctx->dentry = dget(dentry);
969 	ctx->cred = cred;
970 	ctx->ll_cred = NULL;
971 	ctx->state = NULL;
972 	ctx->mode = f_mode;
973 	ctx->flags = 0;
974 	ctx->error = 0;
975 	ctx->flock_owner = (fl_owner_t)filp;
976 	nfs_init_lock_context(&ctx->lock_context);
977 	ctx->lock_context.open_context = ctx;
978 	INIT_LIST_HEAD(&ctx->list);
979 	ctx->mdsthreshold = NULL;
980 	return ctx;
981 }
982 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
983 
984 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
985 {
986 	if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
987 		return ctx;
988 	return NULL;
989 }
990 EXPORT_SYMBOL_GPL(get_nfs_open_context);
991 
992 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
993 {
994 	struct inode *inode = d_inode(ctx->dentry);
995 	struct super_block *sb = ctx->dentry->d_sb;
996 
997 	if (!refcount_dec_and_test(&ctx->lock_context.count))
998 		return;
999 	if (!list_empty(&ctx->list)) {
1000 		spin_lock(&inode->i_lock);
1001 		list_del_rcu(&ctx->list);
1002 		spin_unlock(&inode->i_lock);
1003 	}
1004 	if (inode != NULL)
1005 		NFS_PROTO(inode)->close_context(ctx, is_sync);
1006 	put_cred(ctx->cred);
1007 	dput(ctx->dentry);
1008 	nfs_sb_deactive(sb);
1009 	put_rpccred(ctx->ll_cred);
1010 	kfree(ctx->mdsthreshold);
1011 	kfree_rcu(ctx, rcu_head);
1012 }
1013 
1014 void put_nfs_open_context(struct nfs_open_context *ctx)
1015 {
1016 	__put_nfs_open_context(ctx, 0);
1017 }
1018 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1019 
1020 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1021 {
1022 	__put_nfs_open_context(ctx, 1);
1023 }
1024 
1025 /*
1026  * Ensure that mmap has a recent RPC credential for use when writing out
1027  * shared pages
1028  */
1029 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1030 {
1031 	struct inode *inode = d_inode(ctx->dentry);
1032 	struct nfs_inode *nfsi = NFS_I(inode);
1033 
1034 	spin_lock(&inode->i_lock);
1035 	list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1036 	spin_unlock(&inode->i_lock);
1037 }
1038 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1039 
1040 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1041 {
1042 	filp->private_data = get_nfs_open_context(ctx);
1043 	if (list_empty(&ctx->list))
1044 		nfs_inode_attach_open_context(ctx);
1045 }
1046 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1047 
1048 /*
1049  * Given an inode, search for an open context with the desired characteristics
1050  */
1051 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1052 {
1053 	struct nfs_inode *nfsi = NFS_I(inode);
1054 	struct nfs_open_context *pos, *ctx = NULL;
1055 
1056 	rcu_read_lock();
1057 	list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1058 		if (cred != NULL && pos->cred != cred)
1059 			continue;
1060 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1061 			continue;
1062 		ctx = get_nfs_open_context(pos);
1063 		if (ctx)
1064 			break;
1065 	}
1066 	rcu_read_unlock();
1067 	return ctx;
1068 }
1069 
1070 void nfs_file_clear_open_context(struct file *filp)
1071 {
1072 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1073 
1074 	if (ctx) {
1075 		struct inode *inode = d_inode(ctx->dentry);
1076 
1077 		/*
1078 		 * We fatal error on write before. Try to writeback
1079 		 * every page again.
1080 		 */
1081 		if (ctx->error < 0)
1082 			invalidate_inode_pages2(inode->i_mapping);
1083 		filp->private_data = NULL;
1084 		put_nfs_open_context_sync(ctx);
1085 	}
1086 }
1087 
1088 /*
1089  * These allocate and release file read/write context information.
1090  */
1091 int nfs_open(struct inode *inode, struct file *filp)
1092 {
1093 	struct nfs_open_context *ctx;
1094 
1095 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1096 	if (IS_ERR(ctx))
1097 		return PTR_ERR(ctx);
1098 	nfs_file_set_open_context(filp, ctx);
1099 	put_nfs_open_context(ctx);
1100 	nfs_fscache_open_file(inode, filp);
1101 	return 0;
1102 }
1103 
1104 /*
1105  * This function is called whenever some part of NFS notices that
1106  * the cached attributes have to be refreshed.
1107  */
1108 int
1109 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1110 {
1111 	int		 status = -ESTALE;
1112 	struct nfs4_label *label = NULL;
1113 	struct nfs_fattr *fattr = NULL;
1114 	struct nfs_inode *nfsi = NFS_I(inode);
1115 
1116 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1117 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1118 
1119 	trace_nfs_revalidate_inode_enter(inode);
1120 
1121 	if (is_bad_inode(inode))
1122 		goto out;
1123 	if (NFS_STALE(inode))
1124 		goto out;
1125 
1126 	/* pNFS: Attributes aren't updated until we layoutcommit */
1127 	if (S_ISREG(inode->i_mode)) {
1128 		status = pnfs_sync_inode(inode, false);
1129 		if (status)
1130 			goto out;
1131 	}
1132 
1133 	status = -ENOMEM;
1134 	fattr = nfs_alloc_fattr();
1135 	if (fattr == NULL)
1136 		goto out;
1137 
1138 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1139 
1140 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1141 	if (IS_ERR(label)) {
1142 		status = PTR_ERR(label);
1143 		goto out;
1144 	}
1145 
1146 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1147 			label, inode);
1148 	if (status != 0) {
1149 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1150 			 inode->i_sb->s_id,
1151 			 (unsigned long long)NFS_FILEID(inode), status);
1152 		if (status == -ESTALE) {
1153 			nfs_zap_caches(inode);
1154 			if (!S_ISDIR(inode->i_mode))
1155 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1156 		}
1157 		goto err_out;
1158 	}
1159 
1160 	status = nfs_refresh_inode(inode, fattr);
1161 	if (status) {
1162 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1163 			 inode->i_sb->s_id,
1164 			 (unsigned long long)NFS_FILEID(inode), status);
1165 		goto err_out;
1166 	}
1167 
1168 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1169 		nfs_zap_acl_cache(inode);
1170 
1171 	nfs_setsecurity(inode, fattr, label);
1172 
1173 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1174 		inode->i_sb->s_id,
1175 		(unsigned long long)NFS_FILEID(inode));
1176 
1177 err_out:
1178 	nfs4_label_free(label);
1179 out:
1180 	nfs_free_fattr(fattr);
1181 	trace_nfs_revalidate_inode_exit(inode, status);
1182 	return status;
1183 }
1184 
1185 int nfs_attribute_cache_expired(struct inode *inode)
1186 {
1187 	if (nfs_have_delegated_attributes(inode))
1188 		return 0;
1189 	return nfs_attribute_timeout(inode);
1190 }
1191 
1192 /**
1193  * nfs_revalidate_inode - Revalidate the inode attributes
1194  * @server: pointer to nfs_server struct
1195  * @inode: pointer to inode struct
1196  *
1197  * Updates inode attribute information by retrieving the data from the server.
1198  */
1199 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1200 {
1201 	if (!nfs_need_revalidate_inode(inode))
1202 		return NFS_STALE(inode) ? -ESTALE : 0;
1203 	return __nfs_revalidate_inode(server, inode);
1204 }
1205 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1206 
1207 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1208 {
1209 	struct nfs_inode *nfsi = NFS_I(inode);
1210 	int ret;
1211 
1212 	if (mapping->nrpages != 0) {
1213 		if (S_ISREG(inode->i_mode)) {
1214 			ret = nfs_sync_mapping(mapping);
1215 			if (ret < 0)
1216 				return ret;
1217 		}
1218 		ret = invalidate_inode_pages2(mapping);
1219 		if (ret < 0)
1220 			return ret;
1221 	}
1222 	if (S_ISDIR(inode->i_mode)) {
1223 		spin_lock(&inode->i_lock);
1224 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1225 		spin_unlock(&inode->i_lock);
1226 	}
1227 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1228 	nfs_fscache_wait_on_invalidate(inode);
1229 
1230 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1231 			inode->i_sb->s_id,
1232 			(unsigned long long)NFS_FILEID(inode));
1233 	return 0;
1234 }
1235 
1236 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1237 {
1238 	return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1239 		NFS_STALE(inode);
1240 }
1241 
1242 int nfs_revalidate_mapping_rcu(struct inode *inode)
1243 {
1244 	struct nfs_inode *nfsi = NFS_I(inode);
1245 	unsigned long *bitlock = &nfsi->flags;
1246 	int ret = 0;
1247 
1248 	if (IS_SWAPFILE(inode))
1249 		goto out;
1250 	if (nfs_mapping_need_revalidate_inode(inode)) {
1251 		ret = -ECHILD;
1252 		goto out;
1253 	}
1254 	spin_lock(&inode->i_lock);
1255 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1256 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1257 		ret = -ECHILD;
1258 	spin_unlock(&inode->i_lock);
1259 out:
1260 	return ret;
1261 }
1262 
1263 /**
1264  * nfs_revalidate_mapping - Revalidate the pagecache
1265  * @inode: pointer to host inode
1266  * @mapping: pointer to mapping
1267  */
1268 int nfs_revalidate_mapping(struct inode *inode,
1269 		struct address_space *mapping)
1270 {
1271 	struct nfs_inode *nfsi = NFS_I(inode);
1272 	unsigned long *bitlock = &nfsi->flags;
1273 	int ret = 0;
1274 
1275 	/* swapfiles are not supposed to be shared. */
1276 	if (IS_SWAPFILE(inode))
1277 		goto out;
1278 
1279 	if (nfs_mapping_need_revalidate_inode(inode)) {
1280 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1281 		if (ret < 0)
1282 			goto out;
1283 	}
1284 
1285 	/*
1286 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1287 	 * invalidations that come in while we're shooting down the mappings
1288 	 * are respected. But, that leaves a race window where one revalidator
1289 	 * can clear the flag, and then another checks it before the mapping
1290 	 * gets invalidated. Fix that by serializing access to this part of
1291 	 * the function.
1292 	 *
1293 	 * At the same time, we need to allow other tasks to see whether we
1294 	 * might be in the middle of invalidating the pages, so we only set
1295 	 * the bit lock here if it looks like we're going to be doing that.
1296 	 */
1297 	for (;;) {
1298 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1299 					 nfs_wait_bit_killable, TASK_KILLABLE);
1300 		if (ret)
1301 			goto out;
1302 		spin_lock(&inode->i_lock);
1303 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1304 			spin_unlock(&inode->i_lock);
1305 			continue;
1306 		}
1307 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1308 			break;
1309 		spin_unlock(&inode->i_lock);
1310 		goto out;
1311 	}
1312 
1313 	set_bit(NFS_INO_INVALIDATING, bitlock);
1314 	smp_wmb();
1315 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1316 	spin_unlock(&inode->i_lock);
1317 	trace_nfs_invalidate_mapping_enter(inode);
1318 	ret = nfs_invalidate_mapping(inode, mapping);
1319 	trace_nfs_invalidate_mapping_exit(inode, ret);
1320 
1321 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1322 	smp_mb__after_atomic();
1323 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1324 out:
1325 	return ret;
1326 }
1327 
1328 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1329 {
1330 	struct inode *inode = &nfsi->vfs_inode;
1331 
1332 	if (!S_ISREG(inode->i_mode))
1333 		return false;
1334 	if (list_empty(&nfsi->open_files))
1335 		return false;
1336 	return inode_is_open_for_write(inode);
1337 }
1338 
1339 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1340 {
1341 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1342 }
1343 
1344 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1345 {
1346 	struct timespec ts;
1347 
1348 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1349 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1350 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1351 		inode_set_iversion_raw(inode, fattr->change_attr);
1352 		if (S_ISDIR(inode->i_mode))
1353 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1354 	}
1355 	/* If we have atomic WCC data, we may update some attributes */
1356 	ts = timespec64_to_timespec(inode->i_ctime);
1357 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1358 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1359 			&& timespec_equal(&ts, &fattr->pre_ctime)) {
1360 		inode->i_ctime = timespec_to_timespec64(fattr->ctime);
1361 	}
1362 
1363 	ts = timespec64_to_timespec(inode->i_mtime);
1364 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1365 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1366 			&& timespec_equal(&ts, &fattr->pre_mtime)) {
1367 		inode->i_mtime = timespec_to_timespec64(fattr->mtime);
1368 		if (S_ISDIR(inode->i_mode))
1369 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1370 	}
1371 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1372 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1373 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1374 			&& !nfs_have_writebacks(inode)) {
1375 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1376 	}
1377 }
1378 
1379 /**
1380  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1381  * @inode: pointer to inode
1382  * @fattr: updated attributes
1383  *
1384  * Verifies the attribute cache. If we have just changed the attributes,
1385  * so that fattr carries weak cache consistency data, then it may
1386  * also update the ctime/mtime/change_attribute.
1387  */
1388 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1389 {
1390 	struct nfs_inode *nfsi = NFS_I(inode);
1391 	loff_t cur_size, new_isize;
1392 	unsigned long invalid = 0;
1393 	struct timespec ts;
1394 
1395 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1396 		return 0;
1397 
1398 	/* Has the inode gone and changed behind our back? */
1399 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1400 		return -ESTALE;
1401 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1402 		return -ESTALE;
1403 
1404 	if (!nfs_file_has_buffered_writers(nfsi)) {
1405 		/* Verify a few of the more important attributes */
1406 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1407 			invalid |= NFS_INO_INVALID_CHANGE
1408 				| NFS_INO_REVAL_PAGECACHE;
1409 
1410 		ts = timespec64_to_timespec(inode->i_mtime);
1411 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&ts, &fattr->mtime))
1412 			invalid |= NFS_INO_INVALID_MTIME;
1413 
1414 		ts = timespec64_to_timespec(inode->i_ctime);
1415 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&ts, &fattr->ctime))
1416 			invalid |= NFS_INO_INVALID_CTIME;
1417 
1418 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1419 			cur_size = i_size_read(inode);
1420 			new_isize = nfs_size_to_loff_t(fattr->size);
1421 			if (cur_size != new_isize)
1422 				invalid |= NFS_INO_INVALID_SIZE
1423 					| NFS_INO_REVAL_PAGECACHE;
1424 		}
1425 	}
1426 
1427 	/* Have any file permissions changed? */
1428 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1429 		invalid |= NFS_INO_INVALID_ACCESS
1430 			| NFS_INO_INVALID_ACL
1431 			| NFS_INO_INVALID_OTHER;
1432 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1433 		invalid |= NFS_INO_INVALID_ACCESS
1434 			| NFS_INO_INVALID_ACL
1435 			| NFS_INO_INVALID_OTHER;
1436 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1437 		invalid |= NFS_INO_INVALID_ACCESS
1438 			| NFS_INO_INVALID_ACL
1439 			| NFS_INO_INVALID_OTHER;
1440 
1441 	/* Has the link count changed? */
1442 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1443 		invalid |= NFS_INO_INVALID_OTHER;
1444 
1445 	ts = timespec64_to_timespec(inode->i_atime);
1446 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&ts, &fattr->atime))
1447 		invalid |= NFS_INO_INVALID_ATIME;
1448 
1449 	if (invalid != 0)
1450 		nfs_set_cache_invalid(inode, invalid);
1451 
1452 	nfsi->read_cache_jiffies = fattr->time_start;
1453 	return 0;
1454 }
1455 
1456 static atomic_long_t nfs_attr_generation_counter;
1457 
1458 static unsigned long nfs_read_attr_generation_counter(void)
1459 {
1460 	return atomic_long_read(&nfs_attr_generation_counter);
1461 }
1462 
1463 unsigned long nfs_inc_attr_generation_counter(void)
1464 {
1465 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1466 }
1467 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1468 
1469 void nfs_fattr_init(struct nfs_fattr *fattr)
1470 {
1471 	fattr->valid = 0;
1472 	fattr->time_start = jiffies;
1473 	fattr->gencount = nfs_inc_attr_generation_counter();
1474 	fattr->owner_name = NULL;
1475 	fattr->group_name = NULL;
1476 }
1477 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1478 
1479 /**
1480  * nfs_fattr_set_barrier
1481  * @fattr: attributes
1482  *
1483  * Used to set a barrier after an attribute was updated. This
1484  * barrier ensures that older attributes from RPC calls that may
1485  * have raced with our update cannot clobber these new values.
1486  * Note that you are still responsible for ensuring that other
1487  * operations which change the attribute on the server do not
1488  * collide.
1489  */
1490 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1491 {
1492 	fattr->gencount = nfs_inc_attr_generation_counter();
1493 }
1494 
1495 struct nfs_fattr *nfs_alloc_fattr(void)
1496 {
1497 	struct nfs_fattr *fattr;
1498 
1499 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1500 	if (fattr != NULL)
1501 		nfs_fattr_init(fattr);
1502 	return fattr;
1503 }
1504 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1505 
1506 struct nfs_fh *nfs_alloc_fhandle(void)
1507 {
1508 	struct nfs_fh *fh;
1509 
1510 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1511 	if (fh != NULL)
1512 		fh->size = 0;
1513 	return fh;
1514 }
1515 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1516 
1517 #ifdef NFS_DEBUG
1518 /*
1519  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1520  *                             in the same way that wireshark does
1521  *
1522  * @fh: file handle
1523  *
1524  * For debugging only.
1525  */
1526 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1527 {
1528 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1529 	 * not on the result */
1530 	return nfs_fhandle_hash(fh);
1531 }
1532 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1533 
1534 /*
1535  * _nfs_display_fhandle - display an NFS file handle on the console
1536  *
1537  * @fh: file handle to display
1538  * @caption: display caption
1539  *
1540  * For debugging only.
1541  */
1542 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1543 {
1544 	unsigned short i;
1545 
1546 	if (fh == NULL || fh->size == 0) {
1547 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1548 		return;
1549 	}
1550 
1551 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1552 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1553 	for (i = 0; i < fh->size; i += 16) {
1554 		__be32 *pos = (__be32 *)&fh->data[i];
1555 
1556 		switch ((fh->size - i - 1) >> 2) {
1557 		case 0:
1558 			printk(KERN_DEFAULT " %08x\n",
1559 				be32_to_cpup(pos));
1560 			break;
1561 		case 1:
1562 			printk(KERN_DEFAULT " %08x %08x\n",
1563 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1564 			break;
1565 		case 2:
1566 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1567 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1568 				be32_to_cpup(pos + 2));
1569 			break;
1570 		default:
1571 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1572 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1573 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1574 		}
1575 	}
1576 }
1577 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1578 #endif
1579 
1580 /**
1581  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1582  * @inode: pointer to inode
1583  * @fattr: attributes
1584  *
1585  * Attempt to divine whether or not an RPC call reply carrying stale
1586  * attributes got scheduled after another call carrying updated ones.
1587  *
1588  * To do so, the function first assumes that a more recent ctime means
1589  * that the attributes in fattr are newer, however it also attempt to
1590  * catch the case where ctime either didn't change, or went backwards
1591  * (if someone reset the clock on the server) by looking at whether
1592  * or not this RPC call was started after the inode was last updated.
1593  * Note also the check for wraparound of 'attr_gencount'
1594  *
1595  * The function returns 'true' if it thinks the attributes in 'fattr' are
1596  * more recent than the ones cached in the inode.
1597  *
1598  */
1599 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1600 {
1601 	const struct nfs_inode *nfsi = NFS_I(inode);
1602 
1603 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1604 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1605 }
1606 
1607 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1608 {
1609 	int ret;
1610 
1611 	trace_nfs_refresh_inode_enter(inode);
1612 
1613 	if (nfs_inode_attrs_need_update(inode, fattr))
1614 		ret = nfs_update_inode(inode, fattr);
1615 	else
1616 		ret = nfs_check_inode_attributes(inode, fattr);
1617 
1618 	trace_nfs_refresh_inode_exit(inode, ret);
1619 	return ret;
1620 }
1621 
1622 /**
1623  * nfs_refresh_inode - try to update the inode attribute cache
1624  * @inode: pointer to inode
1625  * @fattr: updated attributes
1626  *
1627  * Check that an RPC call that returned attributes has not overlapped with
1628  * other recent updates of the inode metadata, then decide whether it is
1629  * safe to do a full update of the inode attributes, or whether just to
1630  * call nfs_check_inode_attributes.
1631  */
1632 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1633 {
1634 	int status;
1635 
1636 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1637 		return 0;
1638 	spin_lock(&inode->i_lock);
1639 	status = nfs_refresh_inode_locked(inode, fattr);
1640 	spin_unlock(&inode->i_lock);
1641 
1642 	return status;
1643 }
1644 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1645 
1646 static int nfs_post_op_update_inode_locked(struct inode *inode,
1647 		struct nfs_fattr *fattr, unsigned int invalid)
1648 {
1649 	if (S_ISDIR(inode->i_mode))
1650 		invalid |= NFS_INO_INVALID_DATA;
1651 	nfs_set_cache_invalid(inode, invalid);
1652 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1653 		return 0;
1654 	return nfs_refresh_inode_locked(inode, fattr);
1655 }
1656 
1657 /**
1658  * nfs_post_op_update_inode - try to update the inode attribute cache
1659  * @inode: pointer to inode
1660  * @fattr: updated attributes
1661  *
1662  * After an operation that has changed the inode metadata, mark the
1663  * attribute cache as being invalid, then try to update it.
1664  *
1665  * NB: if the server didn't return any post op attributes, this
1666  * function will force the retrieval of attributes before the next
1667  * NFS request.  Thus it should be used only for operations that
1668  * are expected to change one or more attributes, to avoid
1669  * unnecessary NFS requests and trips through nfs_update_inode().
1670  */
1671 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1672 {
1673 	int status;
1674 
1675 	spin_lock(&inode->i_lock);
1676 	nfs_fattr_set_barrier(fattr);
1677 	status = nfs_post_op_update_inode_locked(inode, fattr,
1678 			NFS_INO_INVALID_CHANGE
1679 			| NFS_INO_INVALID_CTIME
1680 			| NFS_INO_REVAL_FORCED);
1681 	spin_unlock(&inode->i_lock);
1682 
1683 	return status;
1684 }
1685 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1686 
1687 /**
1688  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1689  * @inode: pointer to inode
1690  * @fattr: updated attributes
1691  *
1692  * After an operation that has changed the inode metadata, mark the
1693  * attribute cache as being invalid, then try to update it. Fake up
1694  * weak cache consistency data, if none exist.
1695  *
1696  * This function is mainly designed to be used by the ->write_done() functions.
1697  */
1698 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1699 {
1700 	int status;
1701 
1702 	/* Don't do a WCC update if these attributes are already stale */
1703 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1704 			!nfs_inode_attrs_need_update(inode, fattr)) {
1705 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1706 				| NFS_ATTR_FATTR_PRESIZE
1707 				| NFS_ATTR_FATTR_PREMTIME
1708 				| NFS_ATTR_FATTR_PRECTIME);
1709 		goto out_noforce;
1710 	}
1711 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1712 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1713 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1714 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1715 	}
1716 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1717 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1718 		fattr->pre_ctime = timespec64_to_timespec(inode->i_ctime);
1719 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1720 	}
1721 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1722 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1723 		fattr->pre_mtime = timespec64_to_timespec(inode->i_mtime);
1724 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1725 	}
1726 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1727 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1728 		fattr->pre_size = i_size_read(inode);
1729 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1730 	}
1731 out_noforce:
1732 	status = nfs_post_op_update_inode_locked(inode, fattr,
1733 			NFS_INO_INVALID_CHANGE
1734 			| NFS_INO_INVALID_CTIME
1735 			| NFS_INO_INVALID_MTIME);
1736 	return status;
1737 }
1738 
1739 /**
1740  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1741  * @inode: pointer to inode
1742  * @fattr: updated attributes
1743  *
1744  * After an operation that has changed the inode metadata, mark the
1745  * attribute cache as being invalid, then try to update it. Fake up
1746  * weak cache consistency data, if none exist.
1747  *
1748  * This function is mainly designed to be used by the ->write_done() functions.
1749  */
1750 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1751 {
1752 	int status;
1753 
1754 	spin_lock(&inode->i_lock);
1755 	nfs_fattr_set_barrier(fattr);
1756 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1757 	spin_unlock(&inode->i_lock);
1758 	return status;
1759 }
1760 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1761 
1762 
1763 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1764 				    struct nfs_fattr *fattr)
1765 {
1766 	bool ret1 = true, ret2 = true;
1767 
1768 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1769 		ret1 = (nfsi->fileid == fattr->fileid);
1770 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1771 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1772 	return ret1 || ret2;
1773 }
1774 
1775 /*
1776  * Many nfs protocol calls return the new file attributes after
1777  * an operation.  Here we update the inode to reflect the state
1778  * of the server's inode.
1779  *
1780  * This is a bit tricky because we have to make sure all dirty pages
1781  * have been sent off to the server before calling invalidate_inode_pages.
1782  * To make sure no other process adds more write requests while we try
1783  * our best to flush them, we make them sleep during the attribute refresh.
1784  *
1785  * A very similar scenario holds for the dir cache.
1786  */
1787 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1788 {
1789 	struct nfs_server *server;
1790 	struct nfs_inode *nfsi = NFS_I(inode);
1791 	loff_t cur_isize, new_isize;
1792 	unsigned long invalid = 0;
1793 	unsigned long now = jiffies;
1794 	unsigned long save_cache_validity;
1795 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
1796 	bool cache_revalidated = true;
1797 	bool attr_changed = false;
1798 	bool have_delegation;
1799 
1800 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1801 			__func__, inode->i_sb->s_id, inode->i_ino,
1802 			nfs_display_fhandle_hash(NFS_FH(inode)),
1803 			atomic_read(&inode->i_count), fattr->valid);
1804 
1805 	if (!nfs_fileid_valid(nfsi, fattr)) {
1806 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1807 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1808 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1809 			inode->i_sb->s_id, (long long)nfsi->fileid,
1810 			(long long)fattr->fileid);
1811 		goto out_err;
1812 	}
1813 
1814 	/*
1815 	 * Make sure the inode's type hasn't changed.
1816 	 */
1817 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1818 		/*
1819 		* Big trouble! The inode has become a different object.
1820 		*/
1821 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1822 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1823 		goto out_err;
1824 	}
1825 
1826 	server = NFS_SERVER(inode);
1827 	/* Update the fsid? */
1828 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1829 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1830 			!IS_AUTOMOUNT(inode))
1831 		server->fsid = fattr->fsid;
1832 
1833 	/* Save the delegation state before clearing cache_validity */
1834 	have_delegation = nfs_have_delegated_attributes(inode);
1835 
1836 	/*
1837 	 * Update the read time so we don't revalidate too often.
1838 	 */
1839 	nfsi->read_cache_jiffies = fattr->time_start;
1840 
1841 	save_cache_validity = nfsi->cache_validity;
1842 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1843 			| NFS_INO_INVALID_ATIME
1844 			| NFS_INO_REVAL_FORCED
1845 			| NFS_INO_REVAL_PAGECACHE);
1846 
1847 	/* Do atomic weak cache consistency updates */
1848 	nfs_wcc_update_inode(inode, fattr);
1849 
1850 	if (pnfs_layoutcommit_outstanding(inode)) {
1851 		nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1852 		cache_revalidated = false;
1853 	}
1854 
1855 	/* More cache consistency checks */
1856 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1857 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1858 			/* Could it be a race with writeback? */
1859 			if (!(have_writers || have_delegation)) {
1860 				invalid |= NFS_INO_INVALID_DATA
1861 					| NFS_INO_INVALID_ACCESS
1862 					| NFS_INO_INVALID_ACL;
1863 				/* Force revalidate of all attributes */
1864 				save_cache_validity |= NFS_INO_INVALID_CTIME
1865 					| NFS_INO_INVALID_MTIME
1866 					| NFS_INO_INVALID_SIZE
1867 					| NFS_INO_INVALID_OTHER;
1868 				if (S_ISDIR(inode->i_mode))
1869 					nfs_force_lookup_revalidate(inode);
1870 				dprintk("NFS: change_attr change on server for file %s/%ld\n",
1871 						inode->i_sb->s_id,
1872 						inode->i_ino);
1873 			}
1874 			inode_set_iversion_raw(inode, fattr->change_attr);
1875 			attr_changed = true;
1876 		}
1877 	} else {
1878 		nfsi->cache_validity |= save_cache_validity &
1879 				(NFS_INO_INVALID_CHANGE
1880 				| NFS_INO_REVAL_PAGECACHE
1881 				| NFS_INO_REVAL_FORCED);
1882 		cache_revalidated = false;
1883 	}
1884 
1885 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1886 		inode->i_mtime = timespec_to_timespec64(fattr->mtime);
1887 	} else if (server->caps & NFS_CAP_MTIME) {
1888 		nfsi->cache_validity |= save_cache_validity &
1889 				(NFS_INO_INVALID_MTIME
1890 				| NFS_INO_REVAL_FORCED);
1891 		cache_revalidated = false;
1892 	}
1893 
1894 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1895 		inode->i_ctime = timespec_to_timespec64(fattr->ctime);
1896 	} else if (server->caps & NFS_CAP_CTIME) {
1897 		nfsi->cache_validity |= save_cache_validity &
1898 				(NFS_INO_INVALID_CTIME
1899 				| NFS_INO_REVAL_FORCED);
1900 		cache_revalidated = false;
1901 	}
1902 
1903 	/* Check if our cached file size is stale */
1904 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1905 		new_isize = nfs_size_to_loff_t(fattr->size);
1906 		cur_isize = i_size_read(inode);
1907 		if (new_isize != cur_isize && !have_delegation) {
1908 			/* Do we perhaps have any outstanding writes, or has
1909 			 * the file grown beyond our last write? */
1910 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1911 				i_size_write(inode, new_isize);
1912 				if (!have_writers)
1913 					invalid |= NFS_INO_INVALID_DATA;
1914 				attr_changed = true;
1915 			}
1916 			dprintk("NFS: isize change on server for file %s/%ld "
1917 					"(%Ld to %Ld)\n",
1918 					inode->i_sb->s_id,
1919 					inode->i_ino,
1920 					(long long)cur_isize,
1921 					(long long)new_isize);
1922 		}
1923 	} else {
1924 		nfsi->cache_validity |= save_cache_validity &
1925 				(NFS_INO_INVALID_SIZE
1926 				| NFS_INO_REVAL_PAGECACHE
1927 				| NFS_INO_REVAL_FORCED);
1928 		cache_revalidated = false;
1929 	}
1930 
1931 
1932 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1933 		inode->i_atime = timespec_to_timespec64(fattr->atime);
1934 	else if (server->caps & NFS_CAP_ATIME) {
1935 		nfsi->cache_validity |= save_cache_validity &
1936 				(NFS_INO_INVALID_ATIME
1937 				| NFS_INO_REVAL_FORCED);
1938 		cache_revalidated = false;
1939 	}
1940 
1941 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1942 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1943 			umode_t newmode = inode->i_mode & S_IFMT;
1944 			newmode |= fattr->mode & S_IALLUGO;
1945 			inode->i_mode = newmode;
1946 			invalid |= NFS_INO_INVALID_ACCESS
1947 				| NFS_INO_INVALID_ACL;
1948 			attr_changed = true;
1949 		}
1950 	} else if (server->caps & NFS_CAP_MODE) {
1951 		nfsi->cache_validity |= save_cache_validity &
1952 				(NFS_INO_INVALID_OTHER
1953 				| NFS_INO_REVAL_FORCED);
1954 		cache_revalidated = false;
1955 	}
1956 
1957 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1958 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1959 			invalid |= NFS_INO_INVALID_ACCESS
1960 				| NFS_INO_INVALID_ACL;
1961 			inode->i_uid = fattr->uid;
1962 			attr_changed = true;
1963 		}
1964 	} else if (server->caps & NFS_CAP_OWNER) {
1965 		nfsi->cache_validity |= save_cache_validity &
1966 				(NFS_INO_INVALID_OTHER
1967 				| NFS_INO_REVAL_FORCED);
1968 		cache_revalidated = false;
1969 	}
1970 
1971 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1972 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1973 			invalid |= NFS_INO_INVALID_ACCESS
1974 				| NFS_INO_INVALID_ACL;
1975 			inode->i_gid = fattr->gid;
1976 			attr_changed = true;
1977 		}
1978 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
1979 		nfsi->cache_validity |= save_cache_validity &
1980 				(NFS_INO_INVALID_OTHER
1981 				| NFS_INO_REVAL_FORCED);
1982 		cache_revalidated = false;
1983 	}
1984 
1985 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1986 		if (inode->i_nlink != fattr->nlink) {
1987 			if (S_ISDIR(inode->i_mode))
1988 				invalid |= NFS_INO_INVALID_DATA;
1989 			set_nlink(inode, fattr->nlink);
1990 			attr_changed = true;
1991 		}
1992 	} else if (server->caps & NFS_CAP_NLINK) {
1993 		nfsi->cache_validity |= save_cache_validity &
1994 				(NFS_INO_INVALID_OTHER
1995 				| NFS_INO_REVAL_FORCED);
1996 		cache_revalidated = false;
1997 	}
1998 
1999 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2000 		/*
2001 		 * report the blocks in 512byte units
2002 		 */
2003 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2004 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2005 		inode->i_blocks = fattr->du.nfs2.blocks;
2006 	else
2007 		cache_revalidated = false;
2008 
2009 	/* Update attrtimeo value if we're out of the unstable period */
2010 	if (attr_changed) {
2011 		invalid &= ~NFS_INO_INVALID_ATTR;
2012 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2013 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2014 		nfsi->attrtimeo_timestamp = now;
2015 		/* Set barrier to be more recent than all outstanding updates */
2016 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2017 	} else {
2018 		if (cache_revalidated) {
2019 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2020 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2021 				nfsi->attrtimeo <<= 1;
2022 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2023 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2024 			}
2025 			nfsi->attrtimeo_timestamp = now;
2026 		}
2027 		/* Set the barrier to be more recent than this fattr */
2028 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
2029 			nfsi->attr_gencount = fattr->gencount;
2030 	}
2031 
2032 	/* Don't invalidate the data if we were to blame */
2033 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2034 				|| S_ISLNK(inode->i_mode)))
2035 		invalid &= ~NFS_INO_INVALID_DATA;
2036 	nfs_set_cache_invalid(inode, invalid);
2037 
2038 	return 0;
2039  out_err:
2040 	/*
2041 	 * No need to worry about unhashing the dentry, as the
2042 	 * lookup validation will know that the inode is bad.
2043 	 * (But we fall through to invalidate the caches.)
2044 	 */
2045 	nfs_invalidate_inode(inode);
2046 	return -ESTALE;
2047 }
2048 
2049 struct inode *nfs_alloc_inode(struct super_block *sb)
2050 {
2051 	struct nfs_inode *nfsi;
2052 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2053 	if (!nfsi)
2054 		return NULL;
2055 	nfsi->flags = 0UL;
2056 	nfsi->cache_validity = 0UL;
2057 #if IS_ENABLED(CONFIG_NFS_V4)
2058 	nfsi->nfs4_acl = NULL;
2059 #endif /* CONFIG_NFS_V4 */
2060 	return &nfsi->vfs_inode;
2061 }
2062 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2063 
2064 void nfs_free_inode(struct inode *inode)
2065 {
2066 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2067 }
2068 EXPORT_SYMBOL_GPL(nfs_free_inode);
2069 
2070 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2071 {
2072 #if IS_ENABLED(CONFIG_NFS_V4)
2073 	INIT_LIST_HEAD(&nfsi->open_states);
2074 	nfsi->delegation = NULL;
2075 	init_rwsem(&nfsi->rwsem);
2076 	nfsi->layout = NULL;
2077 #endif
2078 }
2079 
2080 static void init_once(void *foo)
2081 {
2082 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2083 
2084 	inode_init_once(&nfsi->vfs_inode);
2085 	INIT_LIST_HEAD(&nfsi->open_files);
2086 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2087 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2088 	INIT_LIST_HEAD(&nfsi->commit_info.list);
2089 	atomic_long_set(&nfsi->nrequests, 0);
2090 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
2091 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
2092 	init_rwsem(&nfsi->rmdir_sem);
2093 	mutex_init(&nfsi->commit_mutex);
2094 	nfs4_init_once(nfsi);
2095 }
2096 
2097 static int __init nfs_init_inodecache(void)
2098 {
2099 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2100 					     sizeof(struct nfs_inode),
2101 					     0, (SLAB_RECLAIM_ACCOUNT|
2102 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2103 					     init_once);
2104 	if (nfs_inode_cachep == NULL)
2105 		return -ENOMEM;
2106 
2107 	return 0;
2108 }
2109 
2110 static void nfs_destroy_inodecache(void)
2111 {
2112 	/*
2113 	 * Make sure all delayed rcu free inodes are flushed before we
2114 	 * destroy cache.
2115 	 */
2116 	rcu_barrier();
2117 	kmem_cache_destroy(nfs_inode_cachep);
2118 }
2119 
2120 struct workqueue_struct *nfsiod_workqueue;
2121 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2122 
2123 /*
2124  * start up the nfsiod workqueue
2125  */
2126 static int nfsiod_start(void)
2127 {
2128 	struct workqueue_struct *wq;
2129 	dprintk("RPC:       creating workqueue nfsiod\n");
2130 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2131 	if (wq == NULL)
2132 		return -ENOMEM;
2133 	nfsiod_workqueue = wq;
2134 	return 0;
2135 }
2136 
2137 /*
2138  * Destroy the nfsiod workqueue
2139  */
2140 static void nfsiod_stop(void)
2141 {
2142 	struct workqueue_struct *wq;
2143 
2144 	wq = nfsiod_workqueue;
2145 	if (wq == NULL)
2146 		return;
2147 	nfsiod_workqueue = NULL;
2148 	destroy_workqueue(wq);
2149 }
2150 
2151 unsigned int nfs_net_id;
2152 EXPORT_SYMBOL_GPL(nfs_net_id);
2153 
2154 static int nfs_net_init(struct net *net)
2155 {
2156 	nfs_clients_init(net);
2157 	return nfs_fs_proc_net_init(net);
2158 }
2159 
2160 static void nfs_net_exit(struct net *net)
2161 {
2162 	struct nfs_net *nn = net_generic(net, nfs_net_id);
2163 
2164 	nfs_fs_proc_net_exit(net);
2165 	nfs_cleanup_cb_ident_idr(net);
2166 	WARN_ON_ONCE(!list_empty(&nn->nfs_client_list));
2167 	WARN_ON_ONCE(!list_empty(&nn->nfs_volume_list));
2168 }
2169 
2170 static struct pernet_operations nfs_net_ops = {
2171 	.init = nfs_net_init,
2172 	.exit = nfs_net_exit,
2173 	.id   = &nfs_net_id,
2174 	.size = sizeof(struct nfs_net),
2175 };
2176 
2177 /*
2178  * Initialize NFS
2179  */
2180 static int __init init_nfs_fs(void)
2181 {
2182 	int err;
2183 
2184 	err = register_pernet_subsys(&nfs_net_ops);
2185 	if (err < 0)
2186 		goto out9;
2187 
2188 	err = nfs_fscache_register();
2189 	if (err < 0)
2190 		goto out8;
2191 
2192 	err = nfsiod_start();
2193 	if (err)
2194 		goto out7;
2195 
2196 	err = nfs_fs_proc_init();
2197 	if (err)
2198 		goto out6;
2199 
2200 	err = nfs_init_nfspagecache();
2201 	if (err)
2202 		goto out5;
2203 
2204 	err = nfs_init_inodecache();
2205 	if (err)
2206 		goto out4;
2207 
2208 	err = nfs_init_readpagecache();
2209 	if (err)
2210 		goto out3;
2211 
2212 	err = nfs_init_writepagecache();
2213 	if (err)
2214 		goto out2;
2215 
2216 	err = nfs_init_directcache();
2217 	if (err)
2218 		goto out1;
2219 
2220 	rpc_proc_register(&init_net, &nfs_rpcstat);
2221 
2222 	err = register_nfs_fs();
2223 	if (err)
2224 		goto out0;
2225 
2226 	return 0;
2227 out0:
2228 	rpc_proc_unregister(&init_net, "nfs");
2229 	nfs_destroy_directcache();
2230 out1:
2231 	nfs_destroy_writepagecache();
2232 out2:
2233 	nfs_destroy_readpagecache();
2234 out3:
2235 	nfs_destroy_inodecache();
2236 out4:
2237 	nfs_destroy_nfspagecache();
2238 out5:
2239 	nfs_fs_proc_exit();
2240 out6:
2241 	nfsiod_stop();
2242 out7:
2243 	nfs_fscache_unregister();
2244 out8:
2245 	unregister_pernet_subsys(&nfs_net_ops);
2246 out9:
2247 	return err;
2248 }
2249 
2250 static void __exit exit_nfs_fs(void)
2251 {
2252 	nfs_destroy_directcache();
2253 	nfs_destroy_writepagecache();
2254 	nfs_destroy_readpagecache();
2255 	nfs_destroy_inodecache();
2256 	nfs_destroy_nfspagecache();
2257 	nfs_fscache_unregister();
2258 	unregister_pernet_subsys(&nfs_net_ops);
2259 	rpc_proc_unregister(&init_net, "nfs");
2260 	unregister_nfs_fs();
2261 	nfs_fs_proc_exit();
2262 	nfsiod_stop();
2263 }
2264 
2265 /* Not quite true; I just maintain it */
2266 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2267 MODULE_LICENSE("GPL");
2268 module_param(enable_ino64, bool, 0644);
2269 
2270 module_init(init_nfs_fs)
2271 module_exit(exit_nfs_fs)
2272