xref: /linux/fs/nfs/inode.c (revision 46f2dd5ce5723a2c07051d332f8f1c4c4ce548f3)
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 #include "sysfs.h"
55 
56 #include "nfstrace.h"
57 
58 #define NFSDBG_FACILITY		NFSDBG_VFS
59 
60 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
61 
62 /* Default is to see 64-bit inode numbers */
63 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
64 
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 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
76 {
77 	schedule();
78 	if (signal_pending_state(mode, current))
79 		return -ERESTARTSYS;
80 	return 0;
81 }
82 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
83 
84 /**
85  * nfs_compat_user_ino64 - returns the user-visible inode number
86  * @fileid: 64-bit fileid
87  *
88  * This function returns a 32-bit inode number if the boot parameter
89  * nfs.enable_ino64 is zero.
90  */
91 u64 nfs_compat_user_ino64(u64 fileid)
92 {
93 #ifdef CONFIG_COMPAT
94 	compat_ulong_t ino;
95 #else
96 	unsigned long ino;
97 #endif
98 
99 	if (enable_ino64)
100 		return fileid;
101 	ino = fileid;
102 	if (sizeof(ino) < sizeof(fileid))
103 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
104 	return ino;
105 }
106 
107 int nfs_drop_inode(struct inode *inode)
108 {
109 	return NFS_STALE(inode) || generic_drop_inode(inode);
110 }
111 EXPORT_SYMBOL_GPL(nfs_drop_inode);
112 
113 void nfs_clear_inode(struct inode *inode)
114 {
115 	/*
116 	 * The following should never happen...
117 	 */
118 	WARN_ON_ONCE(nfs_have_writebacks(inode));
119 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
120 	nfs_zap_acl_cache(inode);
121 	nfs_access_zap_cache(inode);
122 	nfs_fscache_clear_inode(inode);
123 }
124 EXPORT_SYMBOL_GPL(nfs_clear_inode);
125 
126 void nfs_evict_inode(struct inode *inode)
127 {
128 	truncate_inode_pages_final(&inode->i_data);
129 	clear_inode(inode);
130 	nfs_clear_inode(inode);
131 }
132 
133 int nfs_sync_inode(struct inode *inode)
134 {
135 	inode_dio_wait(inode);
136 	return nfs_wb_all(inode);
137 }
138 EXPORT_SYMBOL_GPL(nfs_sync_inode);
139 
140 /**
141  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
142  * @mapping: pointer to struct address_space
143  */
144 int nfs_sync_mapping(struct address_space *mapping)
145 {
146 	int ret = 0;
147 
148 	if (mapping->nrpages != 0) {
149 		unmap_mapping_range(mapping, 0, 0, 0);
150 		ret = nfs_wb_all(mapping->host);
151 	}
152 	return ret;
153 }
154 
155 static int nfs_attribute_timeout(struct inode *inode)
156 {
157 	struct nfs_inode *nfsi = NFS_I(inode);
158 
159 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
160 }
161 
162 static bool nfs_check_cache_flags_invalid(struct inode *inode,
163 					  unsigned long flags)
164 {
165 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
166 
167 	return (cache_validity & flags) != 0;
168 }
169 
170 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
171 {
172 	if (nfs_check_cache_flags_invalid(inode, flags))
173 		return true;
174 	return nfs_attribute_cache_expired(inode);
175 }
176 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
177 
178 #ifdef CONFIG_NFS_V4_2
179 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
180 {
181 	return nfsi->xattr_cache != NULL;
182 }
183 #else
184 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
185 {
186 	return false;
187 }
188 #endif
189 
190 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
191 {
192 	struct nfs_inode *nfsi = NFS_I(inode);
193 
194 	if (nfs_have_delegated_attributes(inode)) {
195 		if (!(flags & NFS_INO_REVAL_FORCED))
196 			flags &= ~(NFS_INO_INVALID_MODE |
197 				   NFS_INO_INVALID_OTHER |
198 				   NFS_INO_INVALID_XATTR);
199 		flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
200 	}
201 
202 	if (!nfs_has_xattr_cache(nfsi))
203 		flags &= ~NFS_INO_INVALID_XATTR;
204 	if (flags & NFS_INO_INVALID_DATA)
205 		nfs_fscache_invalidate(inode, 0);
206 	flags &= ~NFS_INO_REVAL_FORCED;
207 
208 	nfsi->cache_validity |= flags;
209 
210 	if (inode->i_mapping->nrpages == 0) {
211 		nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
212 		nfs_ooo_clear(nfsi);
213 	} else if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
214 		nfs_ooo_clear(nfsi);
215 	}
216 	trace_nfs_set_cache_invalid(inode, 0);
217 }
218 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
219 
220 /*
221  * Invalidate the local caches
222  */
223 static void nfs_zap_caches_locked(struct inode *inode)
224 {
225 	struct nfs_inode *nfsi = NFS_I(inode);
226 	int mode = inode->i_mode;
227 
228 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
229 
230 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
231 	nfsi->attrtimeo_timestamp = jiffies;
232 
233 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
234 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
235 						     NFS_INO_INVALID_DATA |
236 						     NFS_INO_INVALID_ACCESS |
237 						     NFS_INO_INVALID_ACL |
238 						     NFS_INO_INVALID_XATTR);
239 	else
240 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
241 						     NFS_INO_INVALID_ACCESS |
242 						     NFS_INO_INVALID_ACL |
243 						     NFS_INO_INVALID_XATTR);
244 	nfs_zap_label_cache_locked(nfsi);
245 }
246 
247 void nfs_zap_caches(struct inode *inode)
248 {
249 	spin_lock(&inode->i_lock);
250 	nfs_zap_caches_locked(inode);
251 	spin_unlock(&inode->i_lock);
252 }
253 
254 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
255 {
256 	if (mapping->nrpages != 0) {
257 		spin_lock(&inode->i_lock);
258 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
259 		spin_unlock(&inode->i_lock);
260 	}
261 }
262 
263 void nfs_zap_acl_cache(struct inode *inode)
264 {
265 	void (*clear_acl_cache)(struct inode *);
266 
267 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
268 	if (clear_acl_cache != NULL)
269 		clear_acl_cache(inode);
270 	spin_lock(&inode->i_lock);
271 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
272 	spin_unlock(&inode->i_lock);
273 }
274 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
275 
276 void nfs_invalidate_atime(struct inode *inode)
277 {
278 	if (nfs_have_delegated_atime(inode))
279 		return;
280 	spin_lock(&inode->i_lock);
281 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
282 	spin_unlock(&inode->i_lock);
283 }
284 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
285 
286 /*
287  * Invalidate, but do not unhash, the inode.
288  * NB: must be called with inode->i_lock held!
289  */
290 static void nfs_set_inode_stale_locked(struct inode *inode)
291 {
292 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
293 	nfs_zap_caches_locked(inode);
294 	trace_nfs_set_inode_stale(inode);
295 }
296 
297 void nfs_set_inode_stale(struct inode *inode)
298 {
299 	spin_lock(&inode->i_lock);
300 	nfs_set_inode_stale_locked(inode);
301 	spin_unlock(&inode->i_lock);
302 }
303 
304 struct nfs_find_desc {
305 	struct nfs_fh		*fh;
306 	struct nfs_fattr	*fattr;
307 };
308 
309 /*
310  * In NFSv3 we can have 64bit inode numbers. In order to support
311  * this, and re-exported directories (also seen in NFSv2)
312  * we are forced to allow 2 different inodes to have the same
313  * i_ino.
314  */
315 static int
316 nfs_find_actor(struct inode *inode, void *opaque)
317 {
318 	struct nfs_find_desc	*desc = opaque;
319 	struct nfs_fh		*fh = desc->fh;
320 	struct nfs_fattr	*fattr = desc->fattr;
321 
322 	if (NFS_FILEID(inode) != fattr->fileid)
323 		return 0;
324 	if (inode_wrong_type(inode, fattr->mode))
325 		return 0;
326 	if (nfs_compare_fh(NFS_FH(inode), fh))
327 		return 0;
328 	if (is_bad_inode(inode) || NFS_STALE(inode))
329 		return 0;
330 	return 1;
331 }
332 
333 static int
334 nfs_init_locked(struct inode *inode, void *opaque)
335 {
336 	struct nfs_find_desc	*desc = opaque;
337 	struct nfs_fattr	*fattr = desc->fattr;
338 
339 	set_nfs_fileid(inode, fattr->fileid);
340 	inode->i_mode = fattr->mode;
341 	nfs_copy_fh(NFS_FH(inode), desc->fh);
342 	return 0;
343 }
344 
345 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
346 static void nfs_clear_label_invalid(struct inode *inode)
347 {
348 	spin_lock(&inode->i_lock);
349 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
350 	spin_unlock(&inode->i_lock);
351 }
352 
353 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
354 {
355 	int error;
356 
357 	if (fattr->label == NULL)
358 		return;
359 
360 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
361 		error = security_inode_notifysecctx(inode, fattr->label->label,
362 				fattr->label->len);
363 		if (error)
364 			printk(KERN_ERR "%s() %s %d "
365 					"security_inode_notifysecctx() %d\n",
366 					__func__,
367 					(char *)fattr->label->label,
368 					fattr->label->len, error);
369 		nfs_clear_label_invalid(inode);
370 	}
371 }
372 
373 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
374 {
375 	struct nfs4_label *label;
376 
377 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
378 		return NULL;
379 
380 	label = kzalloc(sizeof(struct nfs4_label), flags);
381 	if (label == NULL)
382 		return ERR_PTR(-ENOMEM);
383 
384 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
385 	if (label->label == NULL) {
386 		kfree(label);
387 		return ERR_PTR(-ENOMEM);
388 	}
389 	label->len = NFS4_MAXLABELLEN;
390 
391 	return label;
392 }
393 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
394 #else
395 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
396 {
397 }
398 #endif
399 EXPORT_SYMBOL_GPL(nfs_setsecurity);
400 
401 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
402 struct inode *
403 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
404 {
405 	struct nfs_find_desc desc = {
406 		.fh	= fh,
407 		.fattr	= fattr,
408 	};
409 	struct inode *inode;
410 	unsigned long hash;
411 
412 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
413 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
414 		return NULL;
415 
416 	hash = nfs_fattr_to_ino_t(fattr);
417 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
418 
419 	dprintk("%s: returning %p\n", __func__, inode);
420 	return inode;
421 }
422 
423 static void nfs_inode_init_regular(struct nfs_inode *nfsi)
424 {
425 	atomic_long_set(&nfsi->nrequests, 0);
426 	atomic_long_set(&nfsi->redirtied_pages, 0);
427 	INIT_LIST_HEAD(&nfsi->commit_info.list);
428 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
429 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
430 	mutex_init(&nfsi->commit_mutex);
431 }
432 
433 static void nfs_inode_init_dir(struct nfs_inode *nfsi)
434 {
435 	nfsi->cache_change_attribute = 0;
436 	memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
437 	init_rwsem(&nfsi->rmdir_sem);
438 }
439 
440 /*
441  * This is our front-end to iget that looks up inodes by file handle
442  * instead of inode number.
443  */
444 struct inode *
445 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
446 {
447 	struct nfs_find_desc desc = {
448 		.fh	= fh,
449 		.fattr	= fattr
450 	};
451 	struct inode *inode = ERR_PTR(-ENOENT);
452 	u64 fattr_supported = NFS_SB(sb)->fattr_valid;
453 	unsigned long hash;
454 
455 	nfs_attr_check_mountpoint(sb, fattr);
456 
457 	if (nfs_attr_use_mounted_on_fileid(fattr))
458 		fattr->fileid = fattr->mounted_on_fileid;
459 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
460 		goto out_no_inode;
461 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
462 		goto out_no_inode;
463 
464 	hash = nfs_fattr_to_ino_t(fattr);
465 
466 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
467 	if (inode == NULL) {
468 		inode = ERR_PTR(-ENOMEM);
469 		goto out_no_inode;
470 	}
471 
472 	if (inode->i_state & I_NEW) {
473 		struct nfs_inode *nfsi = NFS_I(inode);
474 		unsigned long now = jiffies;
475 
476 		/* We set i_ino for the few things that still rely on it,
477 		 * such as stat(2) */
478 		inode->i_ino = hash;
479 
480 		/* We can't support update_atime(), since the server will reset it */
481 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
482 		inode->i_mode = fattr->mode;
483 		nfsi->cache_validity = 0;
484 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
485 				&& (fattr_supported & NFS_ATTR_FATTR_MODE))
486 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
487 		/* Why so? Because we want revalidate for devices/FIFOs, and
488 		 * that's precisely what we have in nfs_file_inode_operations.
489 		 */
490 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
491 		if (S_ISREG(inode->i_mode)) {
492 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
493 			inode->i_data.a_ops = &nfs_file_aops;
494 			nfs_inode_init_regular(nfsi);
495 			mapping_set_large_folios(inode->i_mapping);
496 		} else if (S_ISDIR(inode->i_mode)) {
497 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
498 			inode->i_fop = &nfs_dir_operations;
499 			inode->i_data.a_ops = &nfs_dir_aops;
500 			nfs_inode_init_dir(nfsi);
501 			/* Deal with crossing mountpoints */
502 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
503 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
504 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
505 					inode->i_op = &nfs_referral_inode_operations;
506 				else
507 					inode->i_op = &nfs_mountpoint_inode_operations;
508 				inode->i_fop = NULL;
509 				inode->i_flags |= S_AUTOMOUNT;
510 			}
511 		} else if (S_ISLNK(inode->i_mode)) {
512 			inode->i_op = &nfs_symlink_inode_operations;
513 			inode_nohighmem(inode);
514 		} else
515 			init_special_inode(inode, inode->i_mode, fattr->rdev);
516 
517 		inode_set_atime(inode, 0, 0);
518 		inode_set_mtime(inode, 0, 0);
519 		inode_set_ctime(inode, 0, 0);
520 		inode_set_iversion_raw(inode, 0);
521 		inode->i_size = 0;
522 		clear_nlink(inode);
523 		inode->i_uid = make_kuid(&init_user_ns, -2);
524 		inode->i_gid = make_kgid(&init_user_ns, -2);
525 		inode->i_blocks = 0;
526 		nfsi->write_io = 0;
527 		nfsi->read_io = 0;
528 
529 		nfsi->read_cache_jiffies = fattr->time_start;
530 		nfsi->attr_gencount = fattr->gencount;
531 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
532 			inode_set_atime_to_ts(inode, fattr->atime);
533 		else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
534 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
535 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
536 			inode_set_mtime_to_ts(inode, fattr->mtime);
537 		else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
538 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
539 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
540 			inode_set_ctime_to_ts(inode, fattr->ctime);
541 		else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
542 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
543 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
544 			inode_set_iversion_raw(inode, fattr->change_attr);
545 		else
546 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
547 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
548 			inode->i_size = nfs_size_to_loff_t(fattr->size);
549 		else
550 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
551 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
552 			set_nlink(inode, fattr->nlink);
553 		else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
554 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
555 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
556 			inode->i_uid = fattr->uid;
557 		else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
558 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
559 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
560 			inode->i_gid = fattr->gid;
561 		else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
562 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
563 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
564 			inode->i_blocks = fattr->du.nfs2.blocks;
565 		else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
566 			 fattr->size != 0)
567 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
568 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
569 			/*
570 			 * report the blocks in 512byte units
571 			 */
572 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
573 		} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
574 			   fattr->size != 0)
575 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
576 
577 		nfs_setsecurity(inode, fattr);
578 
579 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
580 		nfsi->attrtimeo_timestamp = now;
581 		nfsi->access_cache = RB_ROOT;
582 
583 		nfs_fscache_init_inode(inode);
584 
585 		unlock_new_inode(inode);
586 	} else {
587 		int err = nfs_refresh_inode(inode, fattr);
588 		if (err < 0) {
589 			iput(inode);
590 			inode = ERR_PTR(err);
591 			goto out_no_inode;
592 		}
593 	}
594 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
595 		inode->i_sb->s_id,
596 		(unsigned long long)NFS_FILEID(inode),
597 		nfs_display_fhandle_hash(fh),
598 		atomic_read(&inode->i_count));
599 
600 out:
601 	return inode;
602 
603 out_no_inode:
604 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
605 	goto out;
606 }
607 EXPORT_SYMBOL_GPL(nfs_fhget);
608 
609 static void
610 nfs_fattr_fixup_delegated(struct inode *inode, struct nfs_fattr *fattr)
611 {
612 	unsigned long cache_validity = NFS_I(inode)->cache_validity;
613 
614 	if (nfs_have_delegated_mtime(inode)) {
615 		if (!(cache_validity & NFS_INO_INVALID_CTIME))
616 			fattr->valid &= ~(NFS_ATTR_FATTR_PRECTIME |
617 					  NFS_ATTR_FATTR_CTIME);
618 
619 		if (!(cache_validity & NFS_INO_INVALID_MTIME))
620 			fattr->valid &= ~(NFS_ATTR_FATTR_PREMTIME |
621 					  NFS_ATTR_FATTR_MTIME);
622 
623 		if (!(cache_validity & NFS_INO_INVALID_ATIME))
624 			fattr->valid &= ~NFS_ATTR_FATTR_ATIME;
625 	} else if (nfs_have_delegated_atime(inode)) {
626 		if (!(cache_validity & NFS_INO_INVALID_ATIME))
627 			fattr->valid &= ~NFS_ATTR_FATTR_ATIME;
628 	}
629 }
630 
631 void nfs_update_delegated_atime(struct inode *inode)
632 {
633 	spin_lock(&inode->i_lock);
634 	if (nfs_have_delegated_atime(inode)) {
635 		inode_update_timestamps(inode, S_ATIME);
636 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ATIME;
637 	}
638 	spin_unlock(&inode->i_lock);
639 }
640 
641 void nfs_update_delegated_mtime_locked(struct inode *inode)
642 {
643 	if (nfs_have_delegated_mtime(inode)) {
644 		inode_update_timestamps(inode, S_CTIME | S_MTIME);
645 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_CTIME |
646 						  NFS_INO_INVALID_MTIME);
647 	}
648 }
649 
650 void nfs_update_delegated_mtime(struct inode *inode)
651 {
652 	spin_lock(&inode->i_lock);
653 	nfs_update_delegated_mtime_locked(inode);
654 	spin_unlock(&inode->i_lock);
655 }
656 EXPORT_SYMBOL_GPL(nfs_update_delegated_mtime);
657 
658 #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)
659 
660 int
661 nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
662 	    struct iattr *attr)
663 {
664 	struct inode *inode = d_inode(dentry);
665 	struct nfs_fattr *fattr;
666 	int error = 0;
667 
668 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
669 
670 	/* skip mode change if it's just for clearing setuid/setgid */
671 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
672 		attr->ia_valid &= ~ATTR_MODE;
673 
674 	if (attr->ia_valid & ATTR_SIZE) {
675 		BUG_ON(!S_ISREG(inode->i_mode));
676 
677 		error = inode_newsize_ok(inode, attr->ia_size);
678 		if (error)
679 			return error;
680 
681 		if (attr->ia_size == i_size_read(inode))
682 			attr->ia_valid &= ~ATTR_SIZE;
683 	}
684 
685 	if (nfs_have_delegated_mtime(inode)) {
686 		if (attr->ia_valid & ATTR_MTIME) {
687 			nfs_update_delegated_mtime(inode);
688 			attr->ia_valid &= ~ATTR_MTIME;
689 		}
690 		if (attr->ia_valid & ATTR_ATIME) {
691 			nfs_update_delegated_atime(inode);
692 			attr->ia_valid &= ~ATTR_ATIME;
693 		}
694 	}
695 
696 	/* Optimization: if the end result is no change, don't RPC */
697 	if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
698 		return 0;
699 
700 	trace_nfs_setattr_enter(inode);
701 
702 	/* Write all dirty data */
703 	if (S_ISREG(inode->i_mode))
704 		nfs_sync_inode(inode);
705 
706 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
707 	if (fattr == NULL) {
708 		error = -ENOMEM;
709 		goto out;
710 	}
711 
712 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
713 	if (error == 0)
714 		error = nfs_refresh_inode(inode, fattr);
715 	nfs_free_fattr(fattr);
716 out:
717 	trace_nfs_setattr_exit(inode, error);
718 	return error;
719 }
720 EXPORT_SYMBOL_GPL(nfs_setattr);
721 
722 /**
723  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
724  * @inode: inode of the file used
725  * @offset: file offset to start truncating
726  *
727  * This is a copy of the common vmtruncate, but with the locking
728  * corrected to take into account the fact that NFS requires
729  * inode->i_size to be updated under the inode->i_lock.
730  * Note: must be called with inode->i_lock held!
731  */
732 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
733 {
734 	int err;
735 
736 	err = inode_newsize_ok(inode, offset);
737 	if (err)
738 		goto out;
739 
740 	trace_nfs_size_truncate(inode, offset);
741 	i_size_write(inode, offset);
742 	/* Optimisation */
743 	if (offset == 0) {
744 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
745 		nfs_ooo_clear(NFS_I(inode));
746 	}
747 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
748 
749 	spin_unlock(&inode->i_lock);
750 	truncate_pagecache(inode, offset);
751 	nfs_update_delegated_mtime_locked(inode);
752 	spin_lock(&inode->i_lock);
753 out:
754 	return err;
755 }
756 
757 /**
758  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
759  * @inode: pointer to struct inode
760  * @attr: pointer to struct iattr
761  * @fattr: pointer to struct nfs_fattr
762  *
763  * Note: we do this in the *proc.c in order to ensure that
764  *       it works for things like exclusive creates too.
765  */
766 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
767 		struct nfs_fattr *fattr)
768 {
769 	/* Barrier: bump the attribute generation count. */
770 	nfs_fattr_set_barrier(fattr);
771 
772 	spin_lock(&inode->i_lock);
773 	NFS_I(inode)->attr_gencount = fattr->gencount;
774 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
775 		if (!nfs_have_delegated_mtime(inode))
776 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
777 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
778 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
779 		nfs_vmtruncate(inode, attr->ia_size);
780 	}
781 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
782 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
783 		if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
784 		    inode->i_mode & S_ISUID)
785 			inode->i_mode &= ~S_ISUID;
786 		if (setattr_should_drop_sgid(&nop_mnt_idmap, inode))
787 			inode->i_mode &= ~S_ISGID;
788 		if ((attr->ia_valid & ATTR_MODE) != 0) {
789 			int mode = attr->ia_mode & S_IALLUGO;
790 			mode |= inode->i_mode & ~S_IALLUGO;
791 			inode->i_mode = mode;
792 		}
793 		if ((attr->ia_valid & ATTR_UID) != 0)
794 			inode->i_uid = attr->ia_uid;
795 		if ((attr->ia_valid & ATTR_GID) != 0)
796 			inode->i_gid = attr->ia_gid;
797 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
798 			inode_set_ctime_to_ts(inode, fattr->ctime);
799 		else
800 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
801 					| NFS_INO_INVALID_CTIME);
802 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
803 				| NFS_INO_INVALID_ACL);
804 	}
805 	if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
806 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
807 				| NFS_INO_INVALID_CTIME);
808 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
809 			inode_set_atime_to_ts(inode, fattr->atime);
810 		else if (attr->ia_valid & ATTR_ATIME_SET)
811 			inode_set_atime_to_ts(inode, attr->ia_atime);
812 		else
813 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
814 
815 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
816 			inode_set_ctime_to_ts(inode, fattr->ctime);
817 		else
818 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
819 					| NFS_INO_INVALID_CTIME);
820 	}
821 	if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
822 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
823 				| NFS_INO_INVALID_CTIME);
824 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
825 			inode_set_mtime_to_ts(inode, fattr->mtime);
826 		else if (attr->ia_valid & ATTR_MTIME_SET)
827 			inode_set_mtime_to_ts(inode, attr->ia_mtime);
828 		else
829 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
830 
831 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
832 			inode_set_ctime_to_ts(inode, fattr->ctime);
833 		else
834 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
835 					| NFS_INO_INVALID_CTIME);
836 	}
837 	if (fattr->valid)
838 		nfs_update_inode(inode, fattr);
839 	spin_unlock(&inode->i_lock);
840 }
841 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
842 
843 /*
844  * Don't request help from readdirplus if the file is being written to,
845  * or if attribute caching is turned off
846  */
847 static bool nfs_getattr_readdirplus_enable(const struct inode *inode)
848 {
849 	return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) &&
850 	       !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ;
851 }
852 
853 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
854 {
855 	if (!IS_ROOT(dentry)) {
856 		struct dentry *parent = dget_parent(dentry);
857 		nfs_readdir_record_entry_cache_miss(d_inode(parent));
858 		dput(parent);
859 	}
860 }
861 
862 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
863 {
864 	if (!IS_ROOT(dentry)) {
865 		struct dentry *parent = dget_parent(dentry);
866 		nfs_readdir_record_entry_cache_hit(d_inode(parent));
867 		dput(parent);
868 	}
869 }
870 
871 static u32 nfs_get_valid_attrmask(struct inode *inode)
872 {
873 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
874 	u32 reply_mask = STATX_INO | STATX_TYPE;
875 
876 	if (!(cache_validity & NFS_INO_INVALID_ATIME))
877 		reply_mask |= STATX_ATIME;
878 	if (!(cache_validity & NFS_INO_INVALID_CTIME))
879 		reply_mask |= STATX_CTIME;
880 	if (!(cache_validity & NFS_INO_INVALID_MTIME))
881 		reply_mask |= STATX_MTIME;
882 	if (!(cache_validity & NFS_INO_INVALID_SIZE))
883 		reply_mask |= STATX_SIZE;
884 	if (!(cache_validity & NFS_INO_INVALID_NLINK))
885 		reply_mask |= STATX_NLINK;
886 	if (!(cache_validity & NFS_INO_INVALID_MODE))
887 		reply_mask |= STATX_MODE;
888 	if (!(cache_validity & NFS_INO_INVALID_OTHER))
889 		reply_mask |= STATX_UID | STATX_GID;
890 	if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
891 		reply_mask |= STATX_BLOCKS;
892 	if (!(cache_validity & NFS_INO_INVALID_CHANGE))
893 		reply_mask |= STATX_CHANGE_COOKIE;
894 	return reply_mask;
895 }
896 
897 int nfs_getattr(struct mnt_idmap *idmap, const struct path *path,
898 		struct kstat *stat, u32 request_mask, unsigned int query_flags)
899 {
900 	struct inode *inode = d_inode(path->dentry);
901 	struct nfs_server *server = NFS_SERVER(inode);
902 	unsigned long cache_validity;
903 	int err = 0;
904 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
905 	bool do_update = false;
906 	bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
907 
908 	trace_nfs_getattr_enter(inode);
909 
910 	request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
911 			STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
912 			STATX_INO | STATX_SIZE | STATX_BLOCKS |
913 			STATX_CHANGE_COOKIE;
914 
915 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
916 		if (readdirplus_enabled)
917 			nfs_readdirplus_parent_cache_hit(path->dentry);
918 		goto out_no_revalidate;
919 	}
920 
921 	/* Flush out writes to the server in order to update c/mtime/version.  */
922 	if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) &&
923 	    S_ISREG(inode->i_mode)) {
924 		if (nfs_have_delegated_mtime(inode))
925 			filemap_fdatawrite(inode->i_mapping);
926 		else
927 			filemap_write_and_wait(inode->i_mapping);
928 	}
929 
930 	/*
931 	 * We may force a getattr if the user cares about atime.
932 	 *
933 	 * Note that we only have to check the vfsmount flags here:
934 	 *  - NFS always sets S_NOATIME by so checking it would give a
935 	 *    bogus result
936 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
937 	 *    no point in checking those.
938 	 */
939 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
940 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
941 		request_mask &= ~STATX_ATIME;
942 
943 	/* Is the user requesting attributes that might need revalidation? */
944 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
945 					STATX_MTIME|STATX_UID|STATX_GID|
946 					STATX_SIZE|STATX_BLOCKS|
947 					STATX_CHANGE_COOKIE)))
948 		goto out_no_revalidate;
949 
950 	/* Check whether the cached attributes are stale */
951 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
952 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
953 	do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
954 	if (request_mask & STATX_ATIME)
955 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
956 	if (request_mask & STATX_CTIME)
957 		do_update |= cache_validity & NFS_INO_INVALID_CTIME;
958 	if (request_mask & STATX_MTIME)
959 		do_update |= cache_validity & NFS_INO_INVALID_MTIME;
960 	if (request_mask & STATX_SIZE)
961 		do_update |= cache_validity & NFS_INO_INVALID_SIZE;
962 	if (request_mask & STATX_NLINK)
963 		do_update |= cache_validity & NFS_INO_INVALID_NLINK;
964 	if (request_mask & STATX_MODE)
965 		do_update |= cache_validity & NFS_INO_INVALID_MODE;
966 	if (request_mask & (STATX_UID | STATX_GID))
967 		do_update |= cache_validity & NFS_INO_INVALID_OTHER;
968 	if (request_mask & STATX_BLOCKS)
969 		do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
970 
971 	if (do_update) {
972 		if (readdirplus_enabled)
973 			nfs_readdirplus_parent_cache_miss(path->dentry);
974 		err = __nfs_revalidate_inode(server, inode);
975 		if (err)
976 			goto out;
977 	} else if (readdirplus_enabled)
978 		nfs_readdirplus_parent_cache_hit(path->dentry);
979 out_no_revalidate:
980 	/* Only return attributes that were revalidated. */
981 	stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
982 
983 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
984 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
985 	stat->change_cookie = inode_peek_iversion_raw(inode);
986 	stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC;
987 	if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED)
988 		stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
989 	if (S_ISDIR(inode->i_mode))
990 		stat->blksize = NFS_SERVER(inode)->dtsize;
991 out:
992 	trace_nfs_getattr_exit(inode, err);
993 	return err;
994 }
995 EXPORT_SYMBOL_GPL(nfs_getattr);
996 
997 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
998 {
999 	refcount_set(&l_ctx->count, 1);
1000 	l_ctx->lockowner = current->files;
1001 	INIT_LIST_HEAD(&l_ctx->list);
1002 	atomic_set(&l_ctx->io_count, 0);
1003 }
1004 
1005 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
1006 {
1007 	struct nfs_lock_context *pos;
1008 
1009 	list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
1010 		if (pos->lockowner != current->files)
1011 			continue;
1012 		if (refcount_inc_not_zero(&pos->count))
1013 			return pos;
1014 	}
1015 	return NULL;
1016 }
1017 
1018 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
1019 {
1020 	struct nfs_lock_context *res, *new = NULL;
1021 	struct inode *inode = d_inode(ctx->dentry);
1022 
1023 	rcu_read_lock();
1024 	res = __nfs_find_lock_context(ctx);
1025 	rcu_read_unlock();
1026 	if (res == NULL) {
1027 		new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1028 		if (new == NULL)
1029 			return ERR_PTR(-ENOMEM);
1030 		nfs_init_lock_context(new);
1031 		spin_lock(&inode->i_lock);
1032 		res = __nfs_find_lock_context(ctx);
1033 		if (res == NULL) {
1034 			new->open_context = get_nfs_open_context(ctx);
1035 			if (new->open_context) {
1036 				list_add_tail_rcu(&new->list,
1037 						&ctx->lock_context.list);
1038 				res = new;
1039 				new = NULL;
1040 			} else
1041 				res = ERR_PTR(-EBADF);
1042 		}
1043 		spin_unlock(&inode->i_lock);
1044 		kfree(new);
1045 	}
1046 	return res;
1047 }
1048 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
1049 
1050 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
1051 {
1052 	struct nfs_open_context *ctx = l_ctx->open_context;
1053 	struct inode *inode = d_inode(ctx->dentry);
1054 
1055 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
1056 		return;
1057 	list_del_rcu(&l_ctx->list);
1058 	spin_unlock(&inode->i_lock);
1059 	put_nfs_open_context(ctx);
1060 	kfree_rcu(l_ctx, rcu_head);
1061 }
1062 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
1063 
1064 /**
1065  * nfs_close_context - Common close_context() routine NFSv2/v3
1066  * @ctx: pointer to context
1067  * @is_sync: is this a synchronous close
1068  *
1069  * Ensure that the attributes are up to date if we're mounted
1070  * with close-to-open semantics and we have cached data that will
1071  * need to be revalidated on open.
1072  */
1073 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1074 {
1075 	struct nfs_inode *nfsi;
1076 	struct inode *inode;
1077 
1078 	if (!(ctx->mode & FMODE_WRITE))
1079 		return;
1080 	if (!is_sync)
1081 		return;
1082 	inode = d_inode(ctx->dentry);
1083 	if (nfs_have_read_or_write_delegation(inode))
1084 		return;
1085 	nfsi = NFS_I(inode);
1086 	if (inode->i_mapping->nrpages == 0)
1087 		return;
1088 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1089 		return;
1090 	if (!list_empty(&nfsi->open_files))
1091 		return;
1092 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1093 		return;
1094 	nfs_revalidate_inode(inode,
1095 			     NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1096 }
1097 EXPORT_SYMBOL_GPL(nfs_close_context);
1098 
1099 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1100 						fmode_t f_mode,
1101 						struct file *filp)
1102 {
1103 	struct nfs_open_context *ctx;
1104 
1105 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1106 	if (!ctx)
1107 		return ERR_PTR(-ENOMEM);
1108 	nfs_sb_active(dentry->d_sb);
1109 	ctx->dentry = dget(dentry);
1110 	if (filp)
1111 		ctx->cred = get_cred(filp->f_cred);
1112 	else
1113 		ctx->cred = get_current_cred();
1114 	rcu_assign_pointer(ctx->ll_cred, NULL);
1115 	ctx->state = NULL;
1116 	ctx->mode = f_mode;
1117 	ctx->flags = 0;
1118 	ctx->error = 0;
1119 	ctx->flock_owner = (fl_owner_t)filp;
1120 	nfs_init_lock_context(&ctx->lock_context);
1121 	ctx->lock_context.open_context = ctx;
1122 	INIT_LIST_HEAD(&ctx->list);
1123 	ctx->mdsthreshold = NULL;
1124 	return ctx;
1125 }
1126 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1127 
1128 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1129 {
1130 	if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1131 		return ctx;
1132 	return NULL;
1133 }
1134 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1135 
1136 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1137 {
1138 	struct inode *inode = d_inode(ctx->dentry);
1139 	struct super_block *sb = ctx->dentry->d_sb;
1140 
1141 	if (!refcount_dec_and_test(&ctx->lock_context.count))
1142 		return;
1143 	if (!list_empty(&ctx->list)) {
1144 		spin_lock(&inode->i_lock);
1145 		list_del_rcu(&ctx->list);
1146 		spin_unlock(&inode->i_lock);
1147 	}
1148 	if (inode != NULL)
1149 		NFS_PROTO(inode)->close_context(ctx, is_sync);
1150 	put_cred(ctx->cred);
1151 	dput(ctx->dentry);
1152 	nfs_sb_deactive(sb);
1153 	put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1154 	kfree(ctx->mdsthreshold);
1155 	kfree_rcu(ctx, rcu_head);
1156 }
1157 
1158 void put_nfs_open_context(struct nfs_open_context *ctx)
1159 {
1160 	__put_nfs_open_context(ctx, 0);
1161 }
1162 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1163 
1164 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1165 {
1166 	__put_nfs_open_context(ctx, 1);
1167 }
1168 
1169 /*
1170  * Ensure that mmap has a recent RPC credential for use when writing out
1171  * shared pages
1172  */
1173 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1174 {
1175 	struct inode *inode = d_inode(ctx->dentry);
1176 	struct nfs_inode *nfsi = NFS_I(inode);
1177 
1178 	spin_lock(&inode->i_lock);
1179 	if (list_empty(&nfsi->open_files) &&
1180 	    nfs_ooo_test(nfsi))
1181 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1182 						     NFS_INO_REVAL_FORCED);
1183 	list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1184 	spin_unlock(&inode->i_lock);
1185 }
1186 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1187 
1188 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1189 {
1190 	filp->private_data = get_nfs_open_context(ctx);
1191 	set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1192 	if (list_empty(&ctx->list))
1193 		nfs_inode_attach_open_context(ctx);
1194 }
1195 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1196 
1197 /*
1198  * Given an inode, search for an open context with the desired characteristics
1199  */
1200 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1201 {
1202 	struct nfs_inode *nfsi = NFS_I(inode);
1203 	struct nfs_open_context *pos, *ctx = NULL;
1204 
1205 	rcu_read_lock();
1206 	list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1207 		if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1208 			continue;
1209 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1210 			continue;
1211 		if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1212 			continue;
1213 		ctx = get_nfs_open_context(pos);
1214 		if (ctx)
1215 			break;
1216 	}
1217 	rcu_read_unlock();
1218 	return ctx;
1219 }
1220 
1221 void nfs_file_clear_open_context(struct file *filp)
1222 {
1223 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1224 
1225 	if (ctx) {
1226 		struct inode *inode = d_inode(ctx->dentry);
1227 
1228 		clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1229 		/*
1230 		 * We fatal error on write before. Try to writeback
1231 		 * every page again.
1232 		 */
1233 		if (ctx->error < 0)
1234 			invalidate_inode_pages2(inode->i_mapping);
1235 		filp->private_data = NULL;
1236 		put_nfs_open_context_sync(ctx);
1237 	}
1238 }
1239 
1240 /*
1241  * These allocate and release file read/write context information.
1242  */
1243 int nfs_open(struct inode *inode, struct file *filp)
1244 {
1245 	struct nfs_open_context *ctx;
1246 
1247 	ctx = alloc_nfs_open_context(file_dentry(filp),
1248 				     flags_to_mode(filp->f_flags), filp);
1249 	if (IS_ERR(ctx))
1250 		return PTR_ERR(ctx);
1251 	nfs_file_set_open_context(filp, ctx);
1252 	put_nfs_open_context(ctx);
1253 	nfs_fscache_open_file(inode, filp);
1254 	return 0;
1255 }
1256 
1257 /*
1258  * This function is called whenever some part of NFS notices that
1259  * the cached attributes have to be refreshed.
1260  */
1261 int
1262 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1263 {
1264 	int		 status = -ESTALE;
1265 	struct nfs_fattr *fattr = NULL;
1266 	struct nfs_inode *nfsi = NFS_I(inode);
1267 
1268 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1269 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1270 
1271 	trace_nfs_revalidate_inode_enter(inode);
1272 
1273 	if (is_bad_inode(inode))
1274 		goto out;
1275 	if (NFS_STALE(inode))
1276 		goto out;
1277 
1278 	/* pNFS: Attributes aren't updated until we layoutcommit */
1279 	if (S_ISREG(inode->i_mode)) {
1280 		status = pnfs_sync_inode(inode, false);
1281 		if (status)
1282 			goto out;
1283 	}
1284 
1285 	status = -ENOMEM;
1286 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1287 	if (fattr == NULL)
1288 		goto out;
1289 
1290 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1291 
1292 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1293 	if (status != 0) {
1294 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1295 			 inode->i_sb->s_id,
1296 			 (unsigned long long)NFS_FILEID(inode), status);
1297 		switch (status) {
1298 		case -ETIMEDOUT:
1299 			/* A soft timeout occurred. Use cached information? */
1300 			if (server->flags & NFS_MOUNT_SOFTREVAL)
1301 				status = 0;
1302 			break;
1303 		case -ESTALE:
1304 			if (!S_ISDIR(inode->i_mode))
1305 				nfs_set_inode_stale(inode);
1306 			else
1307 				nfs_zap_caches(inode);
1308 		}
1309 		goto out;
1310 	}
1311 
1312 	status = nfs_refresh_inode(inode, fattr);
1313 	if (status) {
1314 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1315 			 inode->i_sb->s_id,
1316 			 (unsigned long long)NFS_FILEID(inode), status);
1317 		goto out;
1318 	}
1319 
1320 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1321 		nfs_zap_acl_cache(inode);
1322 
1323 	nfs_setsecurity(inode, fattr);
1324 
1325 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1326 		inode->i_sb->s_id,
1327 		(unsigned long long)NFS_FILEID(inode));
1328 
1329 out:
1330 	nfs_free_fattr(fattr);
1331 	trace_nfs_revalidate_inode_exit(inode, status);
1332 	return status;
1333 }
1334 
1335 int nfs_attribute_cache_expired(struct inode *inode)
1336 {
1337 	if (nfs_have_delegated_attributes(inode))
1338 		return 0;
1339 	return nfs_attribute_timeout(inode);
1340 }
1341 
1342 /**
1343  * nfs_revalidate_inode - Revalidate the inode attributes
1344  * @inode: pointer to inode struct
1345  * @flags: cache flags to check
1346  *
1347  * Updates inode attribute information by retrieving the data from the server.
1348  */
1349 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1350 {
1351 	if (!nfs_check_cache_invalid(inode, flags))
1352 		return NFS_STALE(inode) ? -ESTALE : 0;
1353 	return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1354 }
1355 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1356 
1357 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1358 {
1359 	int ret;
1360 
1361 	nfs_fscache_invalidate(inode, 0);
1362 	if (mapping->nrpages != 0) {
1363 		if (S_ISREG(inode->i_mode)) {
1364 			ret = nfs_sync_mapping(mapping);
1365 			if (ret < 0)
1366 				return ret;
1367 		}
1368 		ret = invalidate_inode_pages2(mapping);
1369 		if (ret < 0)
1370 			return ret;
1371 	}
1372 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1373 
1374 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1375 			inode->i_sb->s_id,
1376 			(unsigned long long)NFS_FILEID(inode));
1377 	return 0;
1378 }
1379 
1380 /**
1381  * nfs_clear_invalid_mapping - Conditionally clear a mapping
1382  * @mapping: pointer to mapping
1383  *
1384  * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1385  */
1386 int nfs_clear_invalid_mapping(struct address_space *mapping)
1387 {
1388 	struct inode *inode = mapping->host;
1389 	struct nfs_inode *nfsi = NFS_I(inode);
1390 	unsigned long *bitlock = &nfsi->flags;
1391 	int ret = 0;
1392 
1393 	/*
1394 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1395 	 * invalidations that come in while we're shooting down the mappings
1396 	 * are respected. But, that leaves a race window where one revalidator
1397 	 * can clear the flag, and then another checks it before the mapping
1398 	 * gets invalidated. Fix that by serializing access to this part of
1399 	 * the function.
1400 	 *
1401 	 * At the same time, we need to allow other tasks to see whether we
1402 	 * might be in the middle of invalidating the pages, so we only set
1403 	 * the bit lock here if it looks like we're going to be doing that.
1404 	 */
1405 	for (;;) {
1406 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1407 					 nfs_wait_bit_killable,
1408 					 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1409 		if (ret)
1410 			goto out;
1411 		spin_lock(&inode->i_lock);
1412 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1413 			spin_unlock(&inode->i_lock);
1414 			continue;
1415 		}
1416 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1417 			break;
1418 		spin_unlock(&inode->i_lock);
1419 		goto out;
1420 	}
1421 
1422 	set_bit(NFS_INO_INVALIDATING, bitlock);
1423 	smp_wmb();
1424 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1425 	nfs_ooo_clear(nfsi);
1426 	spin_unlock(&inode->i_lock);
1427 	trace_nfs_invalidate_mapping_enter(inode);
1428 	ret = nfs_invalidate_mapping(inode, mapping);
1429 	trace_nfs_invalidate_mapping_exit(inode, ret);
1430 
1431 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1432 	smp_mb__after_atomic();
1433 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1434 out:
1435 	return ret;
1436 }
1437 
1438 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1439 {
1440 	return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1441 		NFS_STALE(inode);
1442 }
1443 
1444 int nfs_revalidate_mapping_rcu(struct inode *inode)
1445 {
1446 	struct nfs_inode *nfsi = NFS_I(inode);
1447 	unsigned long *bitlock = &nfsi->flags;
1448 	int ret = 0;
1449 
1450 	if (IS_SWAPFILE(inode))
1451 		goto out;
1452 	if (nfs_mapping_need_revalidate_inode(inode)) {
1453 		ret = -ECHILD;
1454 		goto out;
1455 	}
1456 	spin_lock(&inode->i_lock);
1457 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1458 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1459 		ret = -ECHILD;
1460 	spin_unlock(&inode->i_lock);
1461 out:
1462 	return ret;
1463 }
1464 
1465 /**
1466  * nfs_revalidate_mapping - Revalidate the pagecache
1467  * @inode: pointer to host inode
1468  * @mapping: pointer to mapping
1469  */
1470 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1471 {
1472 	/* swapfiles are not supposed to be shared. */
1473 	if (IS_SWAPFILE(inode))
1474 		return 0;
1475 
1476 	if (nfs_mapping_need_revalidate_inode(inode)) {
1477 		int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1478 		if (ret < 0)
1479 			return ret;
1480 	}
1481 
1482 	return nfs_clear_invalid_mapping(mapping);
1483 }
1484 
1485 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1486 {
1487 	struct inode *inode = &nfsi->vfs_inode;
1488 
1489 	if (!S_ISREG(inode->i_mode))
1490 		return false;
1491 	if (list_empty(&nfsi->open_files))
1492 		return false;
1493 	return inode_is_open_for_write(inode);
1494 }
1495 
1496 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1497 {
1498 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1499 }
1500 
1501 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1502 {
1503 	struct timespec64 ts;
1504 
1505 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1506 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1507 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1508 		inode_set_iversion_raw(inode, fattr->change_attr);
1509 		if (S_ISDIR(inode->i_mode))
1510 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1511 		else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1512 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1513 	}
1514 	/* If we have atomic WCC data, we may update some attributes */
1515 	ts = inode_get_ctime(inode);
1516 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1517 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1518 			&& timespec64_equal(&ts, &fattr->pre_ctime)) {
1519 		inode_set_ctime_to_ts(inode, fattr->ctime);
1520 	}
1521 
1522 	ts = inode_get_mtime(inode);
1523 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1524 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1525 			&& timespec64_equal(&ts, &fattr->pre_mtime)) {
1526 		inode_set_mtime_to_ts(inode, fattr->mtime);
1527 	}
1528 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1529 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1530 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1531 			&& !nfs_have_writebacks(inode)) {
1532 		trace_nfs_size_wcc(inode, fattr->size);
1533 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1534 	}
1535 }
1536 
1537 /**
1538  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1539  * @inode: pointer to inode
1540  * @fattr: updated attributes
1541  *
1542  * Verifies the attribute cache. If we have just changed the attributes,
1543  * so that fattr carries weak cache consistency data, then it may
1544  * also update the ctime/mtime/change_attribute.
1545  */
1546 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1547 {
1548 	struct nfs_inode *nfsi = NFS_I(inode);
1549 	loff_t cur_size, new_isize;
1550 	unsigned long invalid = 0;
1551 	struct timespec64 ts;
1552 
1553 	if (nfs_have_delegated_attributes(inode))
1554 		return 0;
1555 
1556 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1557 		/* Only a mounted-on-fileid? Just exit */
1558 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1559 			return 0;
1560 	/* Has the inode gone and changed behind our back? */
1561 	} else if (nfsi->fileid != fattr->fileid) {
1562 		/* Is this perhaps the mounted-on fileid? */
1563 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1564 		    nfsi->fileid == fattr->mounted_on_fileid)
1565 			return 0;
1566 		return -ESTALE;
1567 	}
1568 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1569 		return -ESTALE;
1570 
1571 
1572 	if (!nfs_file_has_buffered_writers(nfsi)) {
1573 		/* Verify a few of the more important attributes */
1574 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1575 			invalid |= NFS_INO_INVALID_CHANGE;
1576 
1577 		ts = inode_get_mtime(inode);
1578 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1579 			invalid |= NFS_INO_INVALID_MTIME;
1580 
1581 		ts = inode_get_ctime(inode);
1582 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1583 			invalid |= NFS_INO_INVALID_CTIME;
1584 
1585 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1586 			cur_size = i_size_read(inode);
1587 			new_isize = nfs_size_to_loff_t(fattr->size);
1588 			if (cur_size != new_isize)
1589 				invalid |= NFS_INO_INVALID_SIZE;
1590 		}
1591 	}
1592 
1593 	/* Have any file permissions changed? */
1594 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1595 		invalid |= NFS_INO_INVALID_MODE;
1596 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1597 		invalid |= NFS_INO_INVALID_OTHER;
1598 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1599 		invalid |= NFS_INO_INVALID_OTHER;
1600 
1601 	/* Has the link count changed? */
1602 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1603 		invalid |= NFS_INO_INVALID_NLINK;
1604 
1605 	ts = inode_get_atime(inode);
1606 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1607 		invalid |= NFS_INO_INVALID_ATIME;
1608 
1609 	if (invalid != 0)
1610 		nfs_set_cache_invalid(inode, invalid);
1611 
1612 	nfsi->read_cache_jiffies = fattr->time_start;
1613 	return 0;
1614 }
1615 
1616 static atomic_long_t nfs_attr_generation_counter;
1617 
1618 static unsigned long nfs_read_attr_generation_counter(void)
1619 {
1620 	return atomic_long_read(&nfs_attr_generation_counter);
1621 }
1622 
1623 unsigned long nfs_inc_attr_generation_counter(void)
1624 {
1625 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1626 }
1627 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1628 
1629 void nfs_fattr_init(struct nfs_fattr *fattr)
1630 {
1631 	fattr->valid = 0;
1632 	fattr->time_start = jiffies;
1633 	fattr->gencount = nfs_inc_attr_generation_counter();
1634 	fattr->owner_name = NULL;
1635 	fattr->group_name = NULL;
1636 }
1637 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1638 
1639 /**
1640  * nfs_fattr_set_barrier
1641  * @fattr: attributes
1642  *
1643  * Used to set a barrier after an attribute was updated. This
1644  * barrier ensures that older attributes from RPC calls that may
1645  * have raced with our update cannot clobber these new values.
1646  * Note that you are still responsible for ensuring that other
1647  * operations which change the attribute on the server do not
1648  * collide.
1649  */
1650 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1651 {
1652 	fattr->gencount = nfs_inc_attr_generation_counter();
1653 }
1654 
1655 struct nfs_fattr *nfs_alloc_fattr(void)
1656 {
1657 	struct nfs_fattr *fattr;
1658 
1659 	fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1660 	if (fattr != NULL) {
1661 		nfs_fattr_init(fattr);
1662 		fattr->label = NULL;
1663 	}
1664 	return fattr;
1665 }
1666 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1667 
1668 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1669 {
1670 	struct nfs_fattr *fattr = nfs_alloc_fattr();
1671 
1672 	if (!fattr)
1673 		return NULL;
1674 
1675 	fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1676 	if (IS_ERR(fattr->label)) {
1677 		kfree(fattr);
1678 		return NULL;
1679 	}
1680 
1681 	return fattr;
1682 }
1683 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1684 
1685 struct nfs_fh *nfs_alloc_fhandle(void)
1686 {
1687 	struct nfs_fh *fh;
1688 
1689 	fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1690 	if (fh != NULL)
1691 		fh->size = 0;
1692 	return fh;
1693 }
1694 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1695 
1696 #ifdef NFS_DEBUG
1697 /*
1698  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1699  *                             in the same way that wireshark does
1700  *
1701  * @fh: file handle
1702  *
1703  * For debugging only.
1704  */
1705 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1706 {
1707 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1708 	 * not on the result */
1709 	return nfs_fhandle_hash(fh);
1710 }
1711 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1712 
1713 /*
1714  * _nfs_display_fhandle - display an NFS file handle on the console
1715  *
1716  * @fh: file handle to display
1717  * @caption: display caption
1718  *
1719  * For debugging only.
1720  */
1721 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1722 {
1723 	unsigned short i;
1724 
1725 	if (fh == NULL || fh->size == 0) {
1726 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1727 		return;
1728 	}
1729 
1730 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1731 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1732 	for (i = 0; i < fh->size; i += 16) {
1733 		__be32 *pos = (__be32 *)&fh->data[i];
1734 
1735 		switch ((fh->size - i - 1) >> 2) {
1736 		case 0:
1737 			printk(KERN_DEFAULT " %08x\n",
1738 				be32_to_cpup(pos));
1739 			break;
1740 		case 1:
1741 			printk(KERN_DEFAULT " %08x %08x\n",
1742 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1743 			break;
1744 		case 2:
1745 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1746 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1747 				be32_to_cpup(pos + 2));
1748 			break;
1749 		default:
1750 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1751 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1752 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1753 		}
1754 	}
1755 }
1756 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1757 #endif
1758 
1759 /**
1760  * nfs_inode_attrs_cmp_generic - compare attributes
1761  * @fattr: attributes
1762  * @inode: pointer to inode
1763  *
1764  * Attempt to divine whether or not an RPC call reply carrying stale
1765  * attributes got scheduled after another call carrying updated ones.
1766  * Note also the check for wraparound of 'attr_gencount'
1767  *
1768  * The function returns '1' if it thinks the attributes in @fattr are
1769  * more recent than the ones cached in @inode. Otherwise it returns
1770  * the value '0'.
1771  */
1772 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1773 				       const struct inode *inode)
1774 {
1775 	unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1776 
1777 	return (long)(fattr->gencount - attr_gencount) > 0 ||
1778 	       (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1779 }
1780 
1781 /**
1782  * nfs_inode_attrs_cmp_monotonic - compare attributes
1783  * @fattr: attributes
1784  * @inode: pointer to inode
1785  *
1786  * Attempt to divine whether or not an RPC call reply carrying stale
1787  * attributes got scheduled after another call carrying updated ones.
1788  *
1789  * We assume that the server observes monotonic semantics for
1790  * the change attribute, so a larger value means that the attributes in
1791  * @fattr are more recent, in which case the function returns the
1792  * value '1'.
1793  * A return value of '0' indicates no measurable change
1794  * A return value of '-1' means that the attributes in @inode are
1795  * more recent.
1796  */
1797 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1798 					 const struct inode *inode)
1799 {
1800 	s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1801 	if (diff > 0)
1802 		return 1;
1803 	return diff == 0 ? 0 : -1;
1804 }
1805 
1806 /**
1807  * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1808  * @fattr: attributes
1809  * @inode: pointer to inode
1810  *
1811  * Attempt to divine whether or not an RPC call reply carrying stale
1812  * attributes got scheduled after another call carrying updated ones.
1813  *
1814  * We assume that the server observes strictly monotonic semantics for
1815  * the change attribute, so a larger value means that the attributes in
1816  * @fattr are more recent, in which case the function returns the
1817  * value '1'.
1818  * A return value of '-1' means that the attributes in @inode are
1819  * more recent or unchanged.
1820  */
1821 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1822 						const struct inode *inode)
1823 {
1824 	return  nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1825 }
1826 
1827 /**
1828  * nfs_inode_attrs_cmp - compare attributes
1829  * @fattr: attributes
1830  * @inode: pointer to inode
1831  *
1832  * This function returns '1' if it thinks the attributes in @fattr are
1833  * more recent than the ones cached in @inode. It returns '-1' if
1834  * the attributes in @inode are more recent than the ones in @fattr,
1835  * and it returns 0 if not sure.
1836  */
1837 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1838 			       const struct inode *inode)
1839 {
1840 	if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1841 		return 1;
1842 	switch (NFS_SERVER(inode)->change_attr_type) {
1843 	case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1844 		break;
1845 	case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1846 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1847 			break;
1848 		return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1849 	default:
1850 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1851 			break;
1852 		return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1853 	}
1854 	return 0;
1855 }
1856 
1857 /**
1858  * nfs_inode_finish_partial_attr_update - complete a previous inode update
1859  * @fattr: attributes
1860  * @inode: pointer to inode
1861  *
1862  * Returns '1' if the last attribute update left the inode cached
1863  * attributes in a partially unrevalidated state, and @fattr
1864  * matches the change attribute of that partial update.
1865  * Otherwise returns '0'.
1866  */
1867 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1868 						const struct inode *inode)
1869 {
1870 	const unsigned long check_valid =
1871 		NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1872 		NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1873 		NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1874 		NFS_INO_INVALID_NLINK;
1875 	unsigned long cache_validity = NFS_I(inode)->cache_validity;
1876 	enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1877 
1878 	if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1879 	    !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1880 	    (cache_validity & check_valid) != 0 &&
1881 	    (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1882 	    nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1883 		return 1;
1884 	return 0;
1885 }
1886 
1887 static void nfs_ooo_merge(struct nfs_inode *nfsi,
1888 			  u64 start, u64 end)
1889 {
1890 	int i, cnt;
1891 
1892 	if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)
1893 		/* No point merging anything */
1894 		return;
1895 
1896 	if (!nfsi->ooo) {
1897 		nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC);
1898 		if (!nfsi->ooo) {
1899 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1900 			return;
1901 		}
1902 		nfsi->ooo->cnt = 0;
1903 	}
1904 
1905 	/* add this range, merging if possible */
1906 	cnt = nfsi->ooo->cnt;
1907 	for (i = 0; i < cnt; i++) {
1908 		if (end == nfsi->ooo->gap[i].start)
1909 			end = nfsi->ooo->gap[i].end;
1910 		else if (start == nfsi->ooo->gap[i].end)
1911 			start = nfsi->ooo->gap[i].start;
1912 		else
1913 			continue;
1914 		/* Remove 'i' from table and loop to insert the new range */
1915 		cnt -= 1;
1916 		nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt];
1917 		i = -1;
1918 	}
1919 	if (start != end) {
1920 		if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) {
1921 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1922 			kfree(nfsi->ooo);
1923 			nfsi->ooo = NULL;
1924 			return;
1925 		}
1926 		nfsi->ooo->gap[cnt].start = start;
1927 		nfsi->ooo->gap[cnt].end = end;
1928 		cnt += 1;
1929 	}
1930 	nfsi->ooo->cnt = cnt;
1931 }
1932 
1933 static void nfs_ooo_record(struct nfs_inode *nfsi,
1934 			   struct nfs_fattr *fattr)
1935 {
1936 	/* This reply was out-of-order, so record in the
1937 	 * pre/post change id, possibly cancelling
1938 	 * gaps created when iversion was jumpped forward.
1939 	 */
1940 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) &&
1941 	    (fattr->valid & NFS_ATTR_FATTR_PRECHANGE))
1942 		nfs_ooo_merge(nfsi,
1943 			      fattr->change_attr,
1944 			      fattr->pre_change_attr);
1945 }
1946 
1947 static int nfs_refresh_inode_locked(struct inode *inode,
1948 				    struct nfs_fattr *fattr)
1949 {
1950 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1951 	int ret = 0;
1952 
1953 	trace_nfs_refresh_inode_enter(inode);
1954 
1955 	if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1956 		ret = nfs_update_inode(inode, fattr);
1957 	else {
1958 		nfs_ooo_record(NFS_I(inode), fattr);
1959 
1960 		if (attr_cmp == 0)
1961 			ret = nfs_check_inode_attributes(inode, fattr);
1962 	}
1963 
1964 	trace_nfs_refresh_inode_exit(inode, ret);
1965 	return ret;
1966 }
1967 
1968 /**
1969  * nfs_refresh_inode - try to update the inode attribute cache
1970  * @inode: pointer to inode
1971  * @fattr: updated attributes
1972  *
1973  * Check that an RPC call that returned attributes has not overlapped with
1974  * other recent updates of the inode metadata, then decide whether it is
1975  * safe to do a full update of the inode attributes, or whether just to
1976  * call nfs_check_inode_attributes.
1977  */
1978 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1979 {
1980 	int status;
1981 
1982 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1983 		return 0;
1984 	spin_lock(&inode->i_lock);
1985 	status = nfs_refresh_inode_locked(inode, fattr);
1986 	spin_unlock(&inode->i_lock);
1987 
1988 	return status;
1989 }
1990 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1991 
1992 static int nfs_post_op_update_inode_locked(struct inode *inode,
1993 		struct nfs_fattr *fattr, unsigned int invalid)
1994 {
1995 	if (S_ISDIR(inode->i_mode))
1996 		invalid |= NFS_INO_INVALID_DATA;
1997 	nfs_set_cache_invalid(inode, invalid);
1998 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1999 		return 0;
2000 	return nfs_refresh_inode_locked(inode, fattr);
2001 }
2002 
2003 /**
2004  * nfs_post_op_update_inode - try to update the inode attribute cache
2005  * @inode: pointer to inode
2006  * @fattr: updated attributes
2007  *
2008  * After an operation that has changed the inode metadata, mark the
2009  * attribute cache as being invalid, then try to update it.
2010  *
2011  * NB: if the server didn't return any post op attributes, this
2012  * function will force the retrieval of attributes before the next
2013  * NFS request.  Thus it should be used only for operations that
2014  * are expected to change one or more attributes, to avoid
2015  * unnecessary NFS requests and trips through nfs_update_inode().
2016  */
2017 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2018 {
2019 	int status;
2020 
2021 	spin_lock(&inode->i_lock);
2022 	nfs_fattr_set_barrier(fattr);
2023 	status = nfs_post_op_update_inode_locked(inode, fattr,
2024 			NFS_INO_INVALID_CHANGE
2025 			| NFS_INO_INVALID_CTIME
2026 			| NFS_INO_REVAL_FORCED);
2027 	spin_unlock(&inode->i_lock);
2028 
2029 	return status;
2030 }
2031 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
2032 
2033 /**
2034  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
2035  * @inode: pointer to inode
2036  * @fattr: updated attributes
2037  *
2038  * After an operation that has changed the inode metadata, mark the
2039  * attribute cache as being invalid, then try to update it. Fake up
2040  * weak cache consistency data, if none exist.
2041  *
2042  * This function is mainly designed to be used by the ->write_done() functions.
2043  */
2044 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
2045 {
2046 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
2047 	int status;
2048 
2049 	/* Don't do a WCC update if these attributes are already stale */
2050 	if (attr_cmp < 0)
2051 		return 0;
2052 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
2053 		/* Record the pre/post change info before clearing PRECHANGE */
2054 		nfs_ooo_record(NFS_I(inode), fattr);
2055 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
2056 				| NFS_ATTR_FATTR_PRESIZE
2057 				| NFS_ATTR_FATTR_PREMTIME
2058 				| NFS_ATTR_FATTR_PRECTIME);
2059 		goto out_noforce;
2060 	}
2061 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
2062 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
2063 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
2064 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
2065 	}
2066 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
2067 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
2068 		fattr->pre_ctime = inode_get_ctime(inode);
2069 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
2070 	}
2071 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
2072 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
2073 		fattr->pre_mtime = inode_get_mtime(inode);
2074 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
2075 	}
2076 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
2077 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
2078 		fattr->pre_size = i_size_read(inode);
2079 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
2080 	}
2081 out_noforce:
2082 	status = nfs_post_op_update_inode_locked(inode, fattr,
2083 			NFS_INO_INVALID_CHANGE
2084 			| NFS_INO_INVALID_CTIME
2085 			| NFS_INO_INVALID_MTIME
2086 			| NFS_INO_INVALID_BLOCKS);
2087 	return status;
2088 }
2089 
2090 /**
2091  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
2092  * @inode: pointer to inode
2093  * @fattr: updated attributes
2094  *
2095  * After an operation that has changed the inode metadata, mark the
2096  * attribute cache as being invalid, then try to update it. Fake up
2097  * weak cache consistency data, if none exist.
2098  *
2099  * This function is mainly designed to be used by the ->write_done() functions.
2100  */
2101 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
2102 {
2103 	int status;
2104 
2105 	spin_lock(&inode->i_lock);
2106 	nfs_fattr_set_barrier(fattr);
2107 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
2108 	spin_unlock(&inode->i_lock);
2109 	return status;
2110 }
2111 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
2112 
2113 
2114 /*
2115  * Many nfs protocol calls return the new file attributes after
2116  * an operation.  Here we update the inode to reflect the state
2117  * of the server's inode.
2118  *
2119  * This is a bit tricky because we have to make sure all dirty pages
2120  * have been sent off to the server before calling invalidate_inode_pages.
2121  * To make sure no other process adds more write requests while we try
2122  * our best to flush them, we make them sleep during the attribute refresh.
2123  *
2124  * A very similar scenario holds for the dir cache.
2125  */
2126 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2127 {
2128 	struct nfs_server *server = NFS_SERVER(inode);
2129 	struct nfs_inode *nfsi = NFS_I(inode);
2130 	loff_t cur_isize, new_isize;
2131 	u64 fattr_supported = server->fattr_valid;
2132 	unsigned long invalid = 0;
2133 	unsigned long now = jiffies;
2134 	unsigned long save_cache_validity;
2135 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
2136 	bool cache_revalidated = true;
2137 	bool attr_changed = false;
2138 	bool have_delegation;
2139 
2140 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
2141 			__func__, inode->i_sb->s_id, inode->i_ino,
2142 			nfs_display_fhandle_hash(NFS_FH(inode)),
2143 			atomic_read(&inode->i_count), fattr->valid);
2144 
2145 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2146 		/* Only a mounted-on-fileid? Just exit */
2147 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2148 			return 0;
2149 	/* Has the inode gone and changed behind our back? */
2150 	} else if (nfsi->fileid != fattr->fileid) {
2151 		/* Is this perhaps the mounted-on fileid? */
2152 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2153 		    nfsi->fileid == fattr->mounted_on_fileid)
2154 			return 0;
2155 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2156 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2157 			NFS_SERVER(inode)->nfs_client->cl_hostname,
2158 			inode->i_sb->s_id, (long long)nfsi->fileid,
2159 			(long long)fattr->fileid);
2160 		goto out_err;
2161 	}
2162 
2163 	/*
2164 	 * Make sure the inode's type hasn't changed.
2165 	 */
2166 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2167 		/*
2168 		* Big trouble! The inode has become a different object.
2169 		*/
2170 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2171 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
2172 		goto out_err;
2173 	}
2174 
2175 	/* Update the fsid? */
2176 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2177 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2178 			!IS_AUTOMOUNT(inode))
2179 		server->fsid = fattr->fsid;
2180 
2181 	/* Save the delegation state before clearing cache_validity */
2182 	have_delegation = nfs_have_delegated_attributes(inode);
2183 
2184 	/*
2185 	 * Update the read time so we don't revalidate too often.
2186 	 */
2187 	nfsi->read_cache_jiffies = fattr->time_start;
2188 
2189 	/* Fix up any delegated attributes in the struct nfs_fattr */
2190 	nfs_fattr_fixup_delegated(inode, fattr);
2191 
2192 	save_cache_validity = nfsi->cache_validity;
2193 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2194 			| NFS_INO_INVALID_ATIME
2195 			| NFS_INO_REVAL_FORCED
2196 			| NFS_INO_INVALID_BLOCKS);
2197 
2198 	/* Do atomic weak cache consistency updates */
2199 	nfs_wcc_update_inode(inode, fattr);
2200 
2201 	if (pnfs_layoutcommit_outstanding(inode)) {
2202 		nfsi->cache_validity |=
2203 			save_cache_validity &
2204 			(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2205 			 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2206 			 NFS_INO_INVALID_BLOCKS);
2207 		cache_revalidated = false;
2208 	}
2209 
2210 	/* More cache consistency checks */
2211 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2212 		if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 &&
2213 		    nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) {
2214 			/* There is one remaining gap that hasn't been
2215 			 * merged into iversion - do that now.
2216 			 */
2217 			inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start);
2218 			kfree(nfsi->ooo);
2219 			nfsi->ooo = NULL;
2220 		}
2221 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2222 			/* Could it be a race with writeback? */
2223 			if (!(have_writers || have_delegation)) {
2224 				invalid |= NFS_INO_INVALID_DATA
2225 					| NFS_INO_INVALID_ACCESS
2226 					| NFS_INO_INVALID_ACL
2227 					| NFS_INO_INVALID_XATTR;
2228 				/* Force revalidate of all attributes */
2229 				save_cache_validity |= NFS_INO_INVALID_CTIME
2230 					| NFS_INO_INVALID_MTIME
2231 					| NFS_INO_INVALID_SIZE
2232 					| NFS_INO_INVALID_BLOCKS
2233 					| NFS_INO_INVALID_NLINK
2234 					| NFS_INO_INVALID_MODE
2235 					| NFS_INO_INVALID_OTHER;
2236 				if (S_ISDIR(inode->i_mode))
2237 					nfs_force_lookup_revalidate(inode);
2238 				attr_changed = true;
2239 				dprintk("NFS: change_attr change on server for file %s/%ld\n",
2240 						inode->i_sb->s_id,
2241 						inode->i_ino);
2242 			} else if (!have_delegation) {
2243 				nfs_ooo_record(nfsi, fattr);
2244 				nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode),
2245 					      fattr->change_attr);
2246 			}
2247 			inode_set_iversion_raw(inode, fattr->change_attr);
2248 		}
2249 	} else {
2250 		nfsi->cache_validity |=
2251 			save_cache_validity & NFS_INO_INVALID_CHANGE;
2252 		if (!have_delegation ||
2253 		    (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2254 			cache_revalidated = false;
2255 	}
2256 
2257 	if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2258 		inode_set_mtime_to_ts(inode, fattr->mtime);
2259 	else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2260 		nfsi->cache_validity |=
2261 			save_cache_validity & NFS_INO_INVALID_MTIME;
2262 
2263 	if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2264 		inode_set_ctime_to_ts(inode, fattr->ctime);
2265 	else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2266 		nfsi->cache_validity |=
2267 			save_cache_validity & NFS_INO_INVALID_CTIME;
2268 
2269 	/* Check if our cached file size is stale */
2270 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2271 		new_isize = nfs_size_to_loff_t(fattr->size);
2272 		cur_isize = i_size_read(inode);
2273 		if (new_isize != cur_isize && !have_delegation) {
2274 			/* Do we perhaps have any outstanding writes, or has
2275 			 * the file grown beyond our last write? */
2276 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2277 				trace_nfs_size_update(inode, new_isize);
2278 				i_size_write(inode, new_isize);
2279 				if (!have_writers)
2280 					invalid |= NFS_INO_INVALID_DATA;
2281 			}
2282 		}
2283 		if (new_isize == 0 &&
2284 		    !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2285 				      NFS_ATTR_FATTR_BLOCKS_USED))) {
2286 			fattr->du.nfs3.used = 0;
2287 			fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2288 		}
2289 	} else
2290 		nfsi->cache_validity |=
2291 			save_cache_validity & NFS_INO_INVALID_SIZE;
2292 
2293 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2294 		inode_set_atime_to_ts(inode, fattr->atime);
2295 	else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2296 		nfsi->cache_validity |=
2297 			save_cache_validity & NFS_INO_INVALID_ATIME;
2298 
2299 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2300 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2301 			umode_t newmode = inode->i_mode & S_IFMT;
2302 			newmode |= fattr->mode & S_IALLUGO;
2303 			inode->i_mode = newmode;
2304 			invalid |= NFS_INO_INVALID_ACCESS
2305 				| NFS_INO_INVALID_ACL;
2306 		}
2307 	} else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2308 		nfsi->cache_validity |=
2309 			save_cache_validity & NFS_INO_INVALID_MODE;
2310 
2311 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2312 		if (!uid_eq(inode->i_uid, fattr->uid)) {
2313 			invalid |= NFS_INO_INVALID_ACCESS
2314 				| NFS_INO_INVALID_ACL;
2315 			inode->i_uid = fattr->uid;
2316 		}
2317 	} else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2318 		nfsi->cache_validity |=
2319 			save_cache_validity & NFS_INO_INVALID_OTHER;
2320 
2321 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2322 		if (!gid_eq(inode->i_gid, fattr->gid)) {
2323 			invalid |= NFS_INO_INVALID_ACCESS
2324 				| NFS_INO_INVALID_ACL;
2325 			inode->i_gid = fattr->gid;
2326 		}
2327 	} else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2328 		nfsi->cache_validity |=
2329 			save_cache_validity & NFS_INO_INVALID_OTHER;
2330 
2331 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2332 		if (inode->i_nlink != fattr->nlink)
2333 			set_nlink(inode, fattr->nlink);
2334 	} else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2335 		nfsi->cache_validity |=
2336 			save_cache_validity & NFS_INO_INVALID_NLINK;
2337 
2338 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2339 		/*
2340 		 * report the blocks in 512byte units
2341 		 */
2342 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2343 	} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2344 		nfsi->cache_validity |=
2345 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2346 
2347 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2348 		inode->i_blocks = fattr->du.nfs2.blocks;
2349 	else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2350 		nfsi->cache_validity |=
2351 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2352 
2353 	/* Update attrtimeo value if we're out of the unstable period */
2354 	if (attr_changed) {
2355 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2356 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2357 		nfsi->attrtimeo_timestamp = now;
2358 		/* Set barrier to be more recent than all outstanding updates */
2359 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2360 	} else {
2361 		if (cache_revalidated) {
2362 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2363 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2364 				nfsi->attrtimeo <<= 1;
2365 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2366 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2367 			}
2368 			nfsi->attrtimeo_timestamp = now;
2369 		}
2370 		/* Set the barrier to be more recent than this fattr */
2371 		if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2372 			nfsi->attr_gencount = fattr->gencount;
2373 	}
2374 
2375 	/* Don't invalidate the data if we were to blame */
2376 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2377 				|| S_ISLNK(inode->i_mode)))
2378 		invalid &= ~NFS_INO_INVALID_DATA;
2379 	nfs_set_cache_invalid(inode, invalid);
2380 
2381 	return 0;
2382  out_err:
2383 	/*
2384 	 * No need to worry about unhashing the dentry, as the
2385 	 * lookup validation will know that the inode is bad.
2386 	 * (But we fall through to invalidate the caches.)
2387 	 */
2388 	nfs_set_inode_stale_locked(inode);
2389 	return -ESTALE;
2390 }
2391 
2392 struct inode *nfs_alloc_inode(struct super_block *sb)
2393 {
2394 	struct nfs_inode *nfsi;
2395 	nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2396 	if (!nfsi)
2397 		return NULL;
2398 	nfsi->flags = 0UL;
2399 	nfsi->cache_validity = 0UL;
2400 	nfsi->ooo = NULL;
2401 #if IS_ENABLED(CONFIG_NFS_V4)
2402 	nfsi->nfs4_acl = NULL;
2403 #endif /* CONFIG_NFS_V4 */
2404 #ifdef CONFIG_NFS_V4_2
2405 	nfsi->xattr_cache = NULL;
2406 #endif
2407 	nfs_netfs_inode_init(nfsi);
2408 
2409 	return &nfsi->vfs_inode;
2410 }
2411 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2412 
2413 void nfs_free_inode(struct inode *inode)
2414 {
2415 	kfree(NFS_I(inode)->ooo);
2416 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2417 }
2418 EXPORT_SYMBOL_GPL(nfs_free_inode);
2419 
2420 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2421 {
2422 #if IS_ENABLED(CONFIG_NFS_V4)
2423 	INIT_LIST_HEAD(&nfsi->open_states);
2424 	nfsi->delegation = NULL;
2425 	init_rwsem(&nfsi->rwsem);
2426 	nfsi->layout = NULL;
2427 #endif
2428 }
2429 
2430 static void init_once(void *foo)
2431 {
2432 	struct nfs_inode *nfsi = foo;
2433 
2434 	inode_init_once(&nfsi->vfs_inode);
2435 	INIT_LIST_HEAD(&nfsi->open_files);
2436 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2437 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2438 	nfs4_init_once(nfsi);
2439 }
2440 
2441 static int __init nfs_init_inodecache(void)
2442 {
2443 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2444 					     sizeof(struct nfs_inode),
2445 					     0, (SLAB_RECLAIM_ACCOUNT|
2446 						SLAB_ACCOUNT),
2447 					     init_once);
2448 	if (nfs_inode_cachep == NULL)
2449 		return -ENOMEM;
2450 
2451 	return 0;
2452 }
2453 
2454 static void nfs_destroy_inodecache(void)
2455 {
2456 	/*
2457 	 * Make sure all delayed rcu free inodes are flushed before we
2458 	 * destroy cache.
2459 	 */
2460 	rcu_barrier();
2461 	kmem_cache_destroy(nfs_inode_cachep);
2462 }
2463 
2464 struct workqueue_struct *nfslocaliod_workqueue;
2465 struct workqueue_struct *nfsiod_workqueue;
2466 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2467 
2468 /*
2469  * Destroy the nfsiod workqueues
2470  */
2471 static void nfsiod_stop(void)
2472 {
2473 	struct workqueue_struct *wq;
2474 
2475 	wq = nfsiod_workqueue;
2476 	if (wq != NULL) {
2477 		nfsiod_workqueue = NULL;
2478 		destroy_workqueue(wq);
2479 	}
2480 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
2481 	wq = nfslocaliod_workqueue;
2482 	if (wq != NULL) {
2483 		nfslocaliod_workqueue = NULL;
2484 		destroy_workqueue(wq);
2485 	}
2486 #endif /* CONFIG_NFS_LOCALIO */
2487 }
2488 
2489 /*
2490  * Start the nfsiod workqueues
2491  */
2492 static int nfsiod_start(void)
2493 {
2494 	dprintk("RPC:       creating workqueue nfsiod\n");
2495 	nfsiod_workqueue = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2496 	if (nfsiod_workqueue == NULL)
2497 		return -ENOMEM;
2498 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
2499 	/*
2500 	 * localio writes need to use a normal (non-memreclaim) workqueue.
2501 	 * When we start getting low on space, XFS goes and calls flush_work() on
2502 	 * a non-memreclaim work queue, which causes a priority inversion problem.
2503 	 */
2504 	dprintk("RPC:       creating workqueue nfslocaliod\n");
2505 	nfslocaliod_workqueue = alloc_workqueue("nfslocaliod", WQ_UNBOUND, 0);
2506 	if (unlikely(nfslocaliod_workqueue == NULL)) {
2507 		nfsiod_stop();
2508 		return -ENOMEM;
2509 	}
2510 #endif /* CONFIG_NFS_LOCALIO */
2511 	return 0;
2512 }
2513 
2514 unsigned int nfs_net_id;
2515 EXPORT_SYMBOL_GPL(nfs_net_id);
2516 
2517 static int nfs_net_init(struct net *net)
2518 {
2519 	struct nfs_net *nn = net_generic(net, nfs_net_id);
2520 
2521 	nfs_clients_init(net);
2522 
2523 	if (!rpc_proc_register(net, &nn->rpcstats)) {
2524 		nfs_clients_exit(net);
2525 		return -ENOMEM;
2526 	}
2527 
2528 	return nfs_fs_proc_net_init(net);
2529 }
2530 
2531 static void nfs_net_exit(struct net *net)
2532 {
2533 	rpc_proc_unregister(net, "nfs");
2534 	nfs_fs_proc_net_exit(net);
2535 	nfs_clients_exit(net);
2536 }
2537 
2538 static struct pernet_operations nfs_net_ops = {
2539 	.init = nfs_net_init,
2540 	.exit = nfs_net_exit,
2541 	.id   = &nfs_net_id,
2542 	.size = sizeof(struct nfs_net),
2543 };
2544 
2545 /*
2546  * Initialize NFS
2547  */
2548 static int __init init_nfs_fs(void)
2549 {
2550 	int err;
2551 
2552 	err = nfs_sysfs_init();
2553 	if (err < 0)
2554 		goto out10;
2555 
2556 	err = register_pernet_subsys(&nfs_net_ops);
2557 	if (err < 0)
2558 		goto out9;
2559 
2560 	err = nfsiod_start();
2561 	if (err)
2562 		goto out7;
2563 
2564 	err = nfs_fs_proc_init();
2565 	if (err)
2566 		goto out6;
2567 
2568 	err = nfs_init_nfspagecache();
2569 	if (err)
2570 		goto out5;
2571 
2572 	err = nfs_init_inodecache();
2573 	if (err)
2574 		goto out4;
2575 
2576 	err = nfs_init_readpagecache();
2577 	if (err)
2578 		goto out3;
2579 
2580 	err = nfs_init_writepagecache();
2581 	if (err)
2582 		goto out2;
2583 
2584 	err = nfs_init_directcache();
2585 	if (err)
2586 		goto out1;
2587 
2588 	err = register_nfs_fs();
2589 	if (err)
2590 		goto out0;
2591 
2592 	return 0;
2593 out0:
2594 	nfs_destroy_directcache();
2595 out1:
2596 	nfs_destroy_writepagecache();
2597 out2:
2598 	nfs_destroy_readpagecache();
2599 out3:
2600 	nfs_destroy_inodecache();
2601 out4:
2602 	nfs_destroy_nfspagecache();
2603 out5:
2604 	nfs_fs_proc_exit();
2605 out6:
2606 	nfsiod_stop();
2607 out7:
2608 	unregister_pernet_subsys(&nfs_net_ops);
2609 out9:
2610 	nfs_sysfs_exit();
2611 out10:
2612 	return err;
2613 }
2614 
2615 static void __exit exit_nfs_fs(void)
2616 {
2617 	nfs_destroy_directcache();
2618 	nfs_destroy_writepagecache();
2619 	nfs_destroy_readpagecache();
2620 	nfs_destroy_inodecache();
2621 	nfs_destroy_nfspagecache();
2622 	unregister_pernet_subsys(&nfs_net_ops);
2623 	unregister_nfs_fs();
2624 	nfs_fs_proc_exit();
2625 	nfsiod_stop();
2626 	nfs_sysfs_exit();
2627 }
2628 
2629 /* Not quite true; I just maintain it */
2630 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2631 MODULE_DESCRIPTION("NFS client support");
2632 MODULE_LICENSE("GPL");
2633 module_param(enable_ino64, bool, 0644);
2634 
2635 module_init(init_nfs_fs)
2636 module_exit(exit_nfs_fs)
2637