xref: /linux/fs/nfs/inode.c (revision b88fe2b5dd018c2b856fd6c32b82f25033e908d4)
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 	flags |= nfsi->cache_validity;
209 	if (inode->i_mapping->nrpages == 0)
210 		flags &= ~NFS_INO_INVALID_DATA;
211 
212 	/* pairs with nfs_clear_invalid_mapping()'s smp_load_acquire() */
213 	smp_store_release(&nfsi->cache_validity, flags);
214 
215 	if (inode->i_mapping->nrpages == 0 ||
216 	    nfsi->cache_validity & NFS_INO_INVALID_DATA) {
217 		nfs_ooo_clear(nfsi);
218 	}
219 	trace_nfs_set_cache_invalid(inode, 0);
220 }
221 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
222 
223 /*
224  * Invalidate the local caches
225  */
226 static void nfs_zap_caches_locked(struct inode *inode)
227 {
228 	struct nfs_inode *nfsi = NFS_I(inode);
229 	int mode = inode->i_mode;
230 
231 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
232 
233 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
234 	nfsi->attrtimeo_timestamp = jiffies;
235 
236 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
237 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
238 						     NFS_INO_INVALID_DATA |
239 						     NFS_INO_INVALID_ACCESS |
240 						     NFS_INO_INVALID_ACL |
241 						     NFS_INO_INVALID_XATTR);
242 	else
243 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
244 						     NFS_INO_INVALID_ACCESS |
245 						     NFS_INO_INVALID_ACL |
246 						     NFS_INO_INVALID_XATTR);
247 	nfs_zap_label_cache_locked(nfsi);
248 }
249 
250 void nfs_zap_caches(struct inode *inode)
251 {
252 	spin_lock(&inode->i_lock);
253 	nfs_zap_caches_locked(inode);
254 	spin_unlock(&inode->i_lock);
255 }
256 
257 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
258 {
259 	if (mapping->nrpages != 0) {
260 		spin_lock(&inode->i_lock);
261 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
262 		spin_unlock(&inode->i_lock);
263 	}
264 }
265 
266 void nfs_zap_acl_cache(struct inode *inode)
267 {
268 	void (*clear_acl_cache)(struct inode *);
269 
270 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
271 	if (clear_acl_cache != NULL)
272 		clear_acl_cache(inode);
273 	spin_lock(&inode->i_lock);
274 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
275 	spin_unlock(&inode->i_lock);
276 }
277 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
278 
279 void nfs_invalidate_atime(struct inode *inode)
280 {
281 	if (nfs_have_delegated_atime(inode))
282 		return;
283 	spin_lock(&inode->i_lock);
284 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
285 	spin_unlock(&inode->i_lock);
286 }
287 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
288 
289 /*
290  * Invalidate, but do not unhash, the inode.
291  * NB: must be called with inode->i_lock held!
292  */
293 static void nfs_set_inode_stale_locked(struct inode *inode)
294 {
295 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
296 	nfs_zap_caches_locked(inode);
297 	trace_nfs_set_inode_stale(inode);
298 }
299 
300 void nfs_set_inode_stale(struct inode *inode)
301 {
302 	spin_lock(&inode->i_lock);
303 	nfs_set_inode_stale_locked(inode);
304 	spin_unlock(&inode->i_lock);
305 }
306 
307 struct nfs_find_desc {
308 	struct nfs_fh		*fh;
309 	struct nfs_fattr	*fattr;
310 };
311 
312 /*
313  * In NFSv3 we can have 64bit inode numbers. In order to support
314  * this, and re-exported directories (also seen in NFSv2)
315  * we are forced to allow 2 different inodes to have the same
316  * i_ino.
317  */
318 static int
319 nfs_find_actor(struct inode *inode, void *opaque)
320 {
321 	struct nfs_find_desc	*desc = opaque;
322 	struct nfs_fh		*fh = desc->fh;
323 	struct nfs_fattr	*fattr = desc->fattr;
324 
325 	if (NFS_FILEID(inode) != fattr->fileid)
326 		return 0;
327 	if (inode_wrong_type(inode, fattr->mode))
328 		return 0;
329 	if (nfs_compare_fh(NFS_FH(inode), fh))
330 		return 0;
331 	if (is_bad_inode(inode) || NFS_STALE(inode))
332 		return 0;
333 	return 1;
334 }
335 
336 static int
337 nfs_init_locked(struct inode *inode, void *opaque)
338 {
339 	struct nfs_find_desc	*desc = opaque;
340 	struct nfs_fattr	*fattr = desc->fattr;
341 
342 	set_nfs_fileid(inode, fattr->fileid);
343 	inode->i_mode = fattr->mode;
344 	nfs_copy_fh(NFS_FH(inode), desc->fh);
345 	return 0;
346 }
347 
348 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
349 static void nfs_clear_label_invalid(struct inode *inode)
350 {
351 	spin_lock(&inode->i_lock);
352 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
353 	spin_unlock(&inode->i_lock);
354 }
355 
356 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
357 {
358 	int error;
359 
360 	if (fattr->label == NULL)
361 		return;
362 
363 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
364 		error = security_inode_notifysecctx(inode, fattr->label->label,
365 				fattr->label->len);
366 		if (error)
367 			printk(KERN_ERR "%s() %s %d "
368 					"security_inode_notifysecctx() %d\n",
369 					__func__,
370 					(char *)fattr->label->label,
371 					fattr->label->len, error);
372 		nfs_clear_label_invalid(inode);
373 	}
374 }
375 
376 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
377 {
378 	struct nfs4_label *label;
379 
380 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
381 		return NULL;
382 
383 	label = kzalloc(sizeof(struct nfs4_label), flags);
384 	if (label == NULL)
385 		return ERR_PTR(-ENOMEM);
386 
387 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
388 	if (label->label == NULL) {
389 		kfree(label);
390 		return ERR_PTR(-ENOMEM);
391 	}
392 	label->len = NFS4_MAXLABELLEN;
393 
394 	return label;
395 }
396 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
397 #else
398 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
399 {
400 }
401 #endif
402 EXPORT_SYMBOL_GPL(nfs_setsecurity);
403 
404 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
405 struct inode *
406 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
407 {
408 	struct nfs_find_desc desc = {
409 		.fh	= fh,
410 		.fattr	= fattr,
411 	};
412 	struct inode *inode;
413 	unsigned long hash;
414 
415 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
416 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
417 		return NULL;
418 
419 	hash = nfs_fattr_to_ino_t(fattr);
420 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
421 
422 	dprintk("%s: returning %p\n", __func__, inode);
423 	return inode;
424 }
425 
426 static void nfs_inode_init_regular(struct nfs_inode *nfsi)
427 {
428 	atomic_long_set(&nfsi->nrequests, 0);
429 	atomic_long_set(&nfsi->redirtied_pages, 0);
430 	INIT_LIST_HEAD(&nfsi->commit_info.list);
431 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
432 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
433 	mutex_init(&nfsi->commit_mutex);
434 }
435 
436 static void nfs_inode_init_dir(struct nfs_inode *nfsi)
437 {
438 	nfsi->cache_change_attribute = 0;
439 	memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
440 	init_rwsem(&nfsi->rmdir_sem);
441 }
442 
443 /*
444  * This is our front-end to iget that looks up inodes by file handle
445  * instead of inode number.
446  */
447 struct inode *
448 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
449 {
450 	struct nfs_find_desc desc = {
451 		.fh	= fh,
452 		.fattr	= fattr
453 	};
454 	struct inode *inode = ERR_PTR(-ENOENT);
455 	u64 fattr_supported = NFS_SB(sb)->fattr_valid;
456 	unsigned long hash;
457 
458 	nfs_attr_check_mountpoint(sb, fattr);
459 
460 	if (nfs_attr_use_mounted_on_fileid(fattr))
461 		fattr->fileid = fattr->mounted_on_fileid;
462 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
463 		goto out_no_inode;
464 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
465 		goto out_no_inode;
466 
467 	hash = nfs_fattr_to_ino_t(fattr);
468 
469 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
470 	if (inode == NULL) {
471 		inode = ERR_PTR(-ENOMEM);
472 		goto out_no_inode;
473 	}
474 
475 	if (inode->i_state & I_NEW) {
476 		struct nfs_inode *nfsi = NFS_I(inode);
477 		unsigned long now = jiffies;
478 
479 		/* We set i_ino for the few things that still rely on it,
480 		 * such as stat(2) */
481 		inode->i_ino = hash;
482 
483 		/* We can't support update_atime(), since the server will reset it */
484 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
485 		inode->i_mode = fattr->mode;
486 		nfsi->cache_validity = 0;
487 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
488 				&& (fattr_supported & NFS_ATTR_FATTR_MODE))
489 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
490 		/* Why so? Because we want revalidate for devices/FIFOs, and
491 		 * that's precisely what we have in nfs_file_inode_operations.
492 		 */
493 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
494 		if (S_ISREG(inode->i_mode)) {
495 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
496 			inode->i_data.a_ops = &nfs_file_aops;
497 			nfs_inode_init_regular(nfsi);
498 			mapping_set_large_folios(inode->i_mapping);
499 		} else if (S_ISDIR(inode->i_mode)) {
500 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
501 			inode->i_fop = &nfs_dir_operations;
502 			inode->i_data.a_ops = &nfs_dir_aops;
503 			nfs_inode_init_dir(nfsi);
504 			/* Deal with crossing mountpoints */
505 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
506 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
507 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
508 					inode->i_op = &nfs_referral_inode_operations;
509 				else
510 					inode->i_op = &nfs_mountpoint_inode_operations;
511 				inode->i_fop = NULL;
512 				inode->i_flags |= S_AUTOMOUNT;
513 			}
514 		} else if (S_ISLNK(inode->i_mode)) {
515 			inode->i_op = &nfs_symlink_inode_operations;
516 			inode_nohighmem(inode);
517 		} else
518 			init_special_inode(inode, inode->i_mode, fattr->rdev);
519 
520 		inode_set_atime(inode, 0, 0);
521 		inode_set_mtime(inode, 0, 0);
522 		inode_set_ctime(inode, 0, 0);
523 		inode_set_iversion_raw(inode, 0);
524 		inode->i_size = 0;
525 		clear_nlink(inode);
526 		inode->i_uid = make_kuid(&init_user_ns, -2);
527 		inode->i_gid = make_kgid(&init_user_ns, -2);
528 		inode->i_blocks = 0;
529 		nfsi->write_io = 0;
530 		nfsi->read_io = 0;
531 
532 		nfsi->read_cache_jiffies = fattr->time_start;
533 		nfsi->attr_gencount = fattr->gencount;
534 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
535 			inode_set_atime_to_ts(inode, fattr->atime);
536 		else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
537 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
538 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
539 			inode_set_mtime_to_ts(inode, fattr->mtime);
540 		else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
541 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
542 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
543 			inode_set_ctime_to_ts(inode, fattr->ctime);
544 		else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
545 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
546 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
547 			inode_set_iversion_raw(inode, fattr->change_attr);
548 		else
549 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
550 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
551 			inode->i_size = nfs_size_to_loff_t(fattr->size);
552 		else
553 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
554 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
555 			set_nlink(inode, fattr->nlink);
556 		else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
557 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
558 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
559 			inode->i_uid = fattr->uid;
560 		else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
561 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
562 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
563 			inode->i_gid = fattr->gid;
564 		else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
565 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
566 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
567 			inode->i_blocks = fattr->du.nfs2.blocks;
568 		else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
569 			 fattr->size != 0)
570 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
571 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
572 			/*
573 			 * report the blocks in 512byte units
574 			 */
575 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
576 		} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
577 			   fattr->size != 0)
578 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
579 
580 		nfs_setsecurity(inode, fattr);
581 
582 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
583 		nfsi->attrtimeo_timestamp = now;
584 		nfsi->access_cache = RB_ROOT;
585 
586 		nfs_fscache_init_inode(inode);
587 
588 		unlock_new_inode(inode);
589 	} else {
590 		int err = nfs_refresh_inode(inode, fattr);
591 		if (err < 0) {
592 			iput(inode);
593 			inode = ERR_PTR(err);
594 			goto out_no_inode;
595 		}
596 	}
597 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
598 		inode->i_sb->s_id,
599 		(unsigned long long)NFS_FILEID(inode),
600 		nfs_display_fhandle_hash(fh),
601 		atomic_read(&inode->i_count));
602 
603 out:
604 	return inode;
605 
606 out_no_inode:
607 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
608 	goto out;
609 }
610 EXPORT_SYMBOL_GPL(nfs_fhget);
611 
612 static void
613 nfs_fattr_fixup_delegated(struct inode *inode, struct nfs_fattr *fattr)
614 {
615 	unsigned long cache_validity = NFS_I(inode)->cache_validity;
616 
617 	if (nfs_have_delegated_mtime(inode)) {
618 		if (!(cache_validity & NFS_INO_INVALID_CTIME))
619 			fattr->valid &= ~(NFS_ATTR_FATTR_PRECTIME |
620 					  NFS_ATTR_FATTR_CTIME);
621 
622 		if (!(cache_validity & NFS_INO_INVALID_MTIME))
623 			fattr->valid &= ~(NFS_ATTR_FATTR_PREMTIME |
624 					  NFS_ATTR_FATTR_MTIME);
625 
626 		if (!(cache_validity & NFS_INO_INVALID_ATIME))
627 			fattr->valid &= ~NFS_ATTR_FATTR_ATIME;
628 	} else if (nfs_have_delegated_atime(inode)) {
629 		if (!(cache_validity & NFS_INO_INVALID_ATIME))
630 			fattr->valid &= ~NFS_ATTR_FATTR_ATIME;
631 	}
632 }
633 
634 static void nfs_update_timestamps(struct inode *inode, unsigned int ia_valid)
635 {
636 	enum file_time_flags time_flags = 0;
637 	unsigned int cache_flags = 0;
638 
639 	if (ia_valid & ATTR_MTIME) {
640 		time_flags |= S_MTIME | S_CTIME;
641 		cache_flags |= NFS_INO_INVALID_CTIME | NFS_INO_INVALID_MTIME;
642 	}
643 	if (ia_valid & ATTR_ATIME) {
644 		time_flags |= S_ATIME;
645 		cache_flags |= NFS_INO_INVALID_ATIME;
646 	}
647 	inode_update_timestamps(inode, time_flags);
648 	NFS_I(inode)->cache_validity &= ~cache_flags;
649 }
650 
651 void nfs_update_delegated_atime(struct inode *inode)
652 {
653 	spin_lock(&inode->i_lock);
654 	if (nfs_have_delegated_atime(inode))
655 		nfs_update_timestamps(inode, ATTR_ATIME);
656 	spin_unlock(&inode->i_lock);
657 }
658 
659 void nfs_update_delegated_mtime_locked(struct inode *inode)
660 {
661 	if (nfs_have_delegated_mtime(inode))
662 		nfs_update_timestamps(inode, ATTR_MTIME);
663 }
664 
665 void nfs_update_delegated_mtime(struct inode *inode)
666 {
667 	spin_lock(&inode->i_lock);
668 	nfs_update_delegated_mtime_locked(inode);
669 	spin_unlock(&inode->i_lock);
670 }
671 EXPORT_SYMBOL_GPL(nfs_update_delegated_mtime);
672 
673 #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)
674 
675 int
676 nfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
677 	    struct iattr *attr)
678 {
679 	struct inode *inode = d_inode(dentry);
680 	struct nfs_fattr *fattr;
681 	int error = 0;
682 
683 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
684 
685 	/* skip mode change if it's just for clearing setuid/setgid */
686 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
687 		attr->ia_valid &= ~ATTR_MODE;
688 
689 	if (attr->ia_valid & ATTR_SIZE) {
690 		BUG_ON(!S_ISREG(inode->i_mode));
691 
692 		error = inode_newsize_ok(inode, attr->ia_size);
693 		if (error)
694 			return error;
695 
696 		if (attr->ia_size == i_size_read(inode))
697 			attr->ia_valid &= ~ATTR_SIZE;
698 	}
699 
700 	if (nfs_have_delegated_mtime(inode) && attr->ia_valid & ATTR_MTIME) {
701 		spin_lock(&inode->i_lock);
702 		nfs_update_timestamps(inode, attr->ia_valid);
703 		spin_unlock(&inode->i_lock);
704 		attr->ia_valid &= ~(ATTR_MTIME | ATTR_ATIME);
705 	} else if (nfs_have_delegated_atime(inode) &&
706 		   attr->ia_valid & ATTR_ATIME &&
707 		   !(attr->ia_valid & ATTR_MTIME)) {
708 		nfs_update_delegated_atime(inode);
709 		attr->ia_valid &= ~ATTR_ATIME;
710 	}
711 
712 	/* Optimization: if the end result is no change, don't RPC */
713 	if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
714 		return 0;
715 
716 	trace_nfs_setattr_enter(inode);
717 
718 	/* Write all dirty data */
719 	if (S_ISREG(inode->i_mode))
720 		nfs_sync_inode(inode);
721 
722 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
723 	if (fattr == NULL) {
724 		error = -ENOMEM;
725 		goto out;
726 	}
727 
728 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
729 	if (error == 0)
730 		error = nfs_refresh_inode(inode, fattr);
731 	nfs_free_fattr(fattr);
732 out:
733 	trace_nfs_setattr_exit(inode, error);
734 	return error;
735 }
736 EXPORT_SYMBOL_GPL(nfs_setattr);
737 
738 /**
739  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
740  * @inode: inode of the file used
741  * @offset: file offset to start truncating
742  *
743  * This is a copy of the common vmtruncate, but with the locking
744  * corrected to take into account the fact that NFS requires
745  * inode->i_size to be updated under the inode->i_lock.
746  * Note: must be called with inode->i_lock held!
747  */
748 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
749 {
750 	int err;
751 
752 	err = inode_newsize_ok(inode, offset);
753 	if (err)
754 		goto out;
755 
756 	trace_nfs_size_truncate(inode, offset);
757 	i_size_write(inode, offset);
758 	/* Optimisation */
759 	if (offset == 0) {
760 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
761 		nfs_ooo_clear(NFS_I(inode));
762 	}
763 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
764 
765 	spin_unlock(&inode->i_lock);
766 	truncate_pagecache(inode, offset);
767 	nfs_update_delegated_mtime_locked(inode);
768 	spin_lock(&inode->i_lock);
769 out:
770 	return err;
771 }
772 
773 /**
774  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
775  * @inode: pointer to struct inode
776  * @attr: pointer to struct iattr
777  * @fattr: pointer to struct nfs_fattr
778  *
779  * Note: we do this in the *proc.c in order to ensure that
780  *       it works for things like exclusive creates too.
781  */
782 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
783 		struct nfs_fattr *fattr)
784 {
785 	/* Barrier: bump the attribute generation count. */
786 	nfs_fattr_set_barrier(fattr);
787 
788 	spin_lock(&inode->i_lock);
789 	NFS_I(inode)->attr_gencount = fattr->gencount;
790 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
791 		if (!nfs_have_delegated_mtime(inode))
792 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
793 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
794 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
795 		nfs_vmtruncate(inode, attr->ia_size);
796 	}
797 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
798 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
799 		if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
800 		    inode->i_mode & S_ISUID)
801 			inode->i_mode &= ~S_ISUID;
802 		if (setattr_should_drop_sgid(&nop_mnt_idmap, inode))
803 			inode->i_mode &= ~S_ISGID;
804 		if ((attr->ia_valid & ATTR_MODE) != 0) {
805 			int mode = attr->ia_mode & S_IALLUGO;
806 			mode |= inode->i_mode & ~S_IALLUGO;
807 			inode->i_mode = mode;
808 		}
809 		if ((attr->ia_valid & ATTR_UID) != 0)
810 			inode->i_uid = attr->ia_uid;
811 		if ((attr->ia_valid & ATTR_GID) != 0)
812 			inode->i_gid = attr->ia_gid;
813 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
814 			inode_set_ctime_to_ts(inode, fattr->ctime);
815 		else
816 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
817 					| NFS_INO_INVALID_CTIME);
818 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
819 				| NFS_INO_INVALID_ACL);
820 	}
821 	if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
822 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
823 				| NFS_INO_INVALID_CTIME);
824 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
825 			inode_set_atime_to_ts(inode, fattr->atime);
826 		else if (attr->ia_valid & ATTR_ATIME_SET)
827 			inode_set_atime_to_ts(inode, attr->ia_atime);
828 		else
829 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
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 (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
838 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
839 				| NFS_INO_INVALID_CTIME);
840 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
841 			inode_set_mtime_to_ts(inode, fattr->mtime);
842 		else if (attr->ia_valid & ATTR_MTIME_SET)
843 			inode_set_mtime_to_ts(inode, attr->ia_mtime);
844 		else
845 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
846 
847 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
848 			inode_set_ctime_to_ts(inode, fattr->ctime);
849 		else
850 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
851 					| NFS_INO_INVALID_CTIME);
852 	}
853 	if (fattr->valid)
854 		nfs_update_inode(inode, fattr);
855 	spin_unlock(&inode->i_lock);
856 }
857 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
858 
859 /*
860  * Don't request help from readdirplus if the file is being written to,
861  * or if attribute caching is turned off
862  */
863 static bool nfs_getattr_readdirplus_enable(const struct inode *inode)
864 {
865 	return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) &&
866 	       !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ;
867 }
868 
869 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
870 {
871 	if (!IS_ROOT(dentry)) {
872 		struct dentry *parent = dget_parent(dentry);
873 		nfs_readdir_record_entry_cache_miss(d_inode(parent));
874 		dput(parent);
875 	}
876 }
877 
878 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
879 {
880 	if (!IS_ROOT(dentry)) {
881 		struct dentry *parent = dget_parent(dentry);
882 		nfs_readdir_record_entry_cache_hit(d_inode(parent));
883 		dput(parent);
884 	}
885 }
886 
887 static u32 nfs_get_valid_attrmask(struct inode *inode)
888 {
889 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
890 	u32 reply_mask = STATX_INO | STATX_TYPE;
891 
892 	if (!(cache_validity & NFS_INO_INVALID_ATIME))
893 		reply_mask |= STATX_ATIME;
894 	if (!(cache_validity & NFS_INO_INVALID_CTIME))
895 		reply_mask |= STATX_CTIME;
896 	if (!(cache_validity & NFS_INO_INVALID_MTIME))
897 		reply_mask |= STATX_MTIME;
898 	if (!(cache_validity & NFS_INO_INVALID_SIZE))
899 		reply_mask |= STATX_SIZE;
900 	if (!(cache_validity & NFS_INO_INVALID_NLINK))
901 		reply_mask |= STATX_NLINK;
902 	if (!(cache_validity & NFS_INO_INVALID_MODE))
903 		reply_mask |= STATX_MODE;
904 	if (!(cache_validity & NFS_INO_INVALID_OTHER))
905 		reply_mask |= STATX_UID | STATX_GID;
906 	if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
907 		reply_mask |= STATX_BLOCKS;
908 	if (!(cache_validity & NFS_INO_INVALID_CHANGE))
909 		reply_mask |= STATX_CHANGE_COOKIE;
910 	return reply_mask;
911 }
912 
913 int nfs_getattr(struct mnt_idmap *idmap, const struct path *path,
914 		struct kstat *stat, u32 request_mask, unsigned int query_flags)
915 {
916 	struct inode *inode = d_inode(path->dentry);
917 	struct nfs_server *server = NFS_SERVER(inode);
918 	unsigned long cache_validity;
919 	int err = 0;
920 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
921 	bool do_update = false;
922 	bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
923 
924 	trace_nfs_getattr_enter(inode);
925 
926 	request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
927 			STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
928 			STATX_INO | STATX_SIZE | STATX_BLOCKS |
929 			STATX_CHANGE_COOKIE;
930 
931 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
932 		if (readdirplus_enabled)
933 			nfs_readdirplus_parent_cache_hit(path->dentry);
934 		goto out_no_revalidate;
935 	}
936 
937 	/* Flush out writes to the server in order to update c/mtime/version.  */
938 	if ((request_mask & (STATX_CTIME | STATX_MTIME | STATX_CHANGE_COOKIE)) &&
939 	    S_ISREG(inode->i_mode)) {
940 		if (nfs_have_delegated_mtime(inode))
941 			filemap_fdatawrite(inode->i_mapping);
942 		else
943 			filemap_write_and_wait(inode->i_mapping);
944 	}
945 
946 	/*
947 	 * We may force a getattr if the user cares about atime.
948 	 *
949 	 * Note that we only have to check the vfsmount flags here:
950 	 *  - NFS always sets S_NOATIME by so checking it would give a
951 	 *    bogus result
952 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
953 	 *    no point in checking those.
954 	 */
955 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
956 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
957 		request_mask &= ~STATX_ATIME;
958 
959 	/* Is the user requesting attributes that might need revalidation? */
960 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
961 					STATX_MTIME|STATX_UID|STATX_GID|
962 					STATX_SIZE|STATX_BLOCKS|
963 					STATX_CHANGE_COOKIE)))
964 		goto out_no_revalidate;
965 
966 	/* Check whether the cached attributes are stale */
967 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
968 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
969 	do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
970 	if (request_mask & STATX_ATIME)
971 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
972 	if (request_mask & STATX_CTIME)
973 		do_update |= cache_validity & NFS_INO_INVALID_CTIME;
974 	if (request_mask & STATX_MTIME)
975 		do_update |= cache_validity & NFS_INO_INVALID_MTIME;
976 	if (request_mask & STATX_SIZE)
977 		do_update |= cache_validity & NFS_INO_INVALID_SIZE;
978 	if (request_mask & STATX_NLINK)
979 		do_update |= cache_validity & NFS_INO_INVALID_NLINK;
980 	if (request_mask & STATX_MODE)
981 		do_update |= cache_validity & NFS_INO_INVALID_MODE;
982 	if (request_mask & (STATX_UID | STATX_GID))
983 		do_update |= cache_validity & NFS_INO_INVALID_OTHER;
984 	if (request_mask & STATX_BLOCKS)
985 		do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
986 
987 	if (do_update) {
988 		if (readdirplus_enabled)
989 			nfs_readdirplus_parent_cache_miss(path->dentry);
990 		err = __nfs_revalidate_inode(server, inode);
991 		if (err)
992 			goto out;
993 	} else if (readdirplus_enabled)
994 		nfs_readdirplus_parent_cache_hit(path->dentry);
995 out_no_revalidate:
996 	/* Only return attributes that were revalidated. */
997 	stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
998 
999 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1000 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
1001 	stat->change_cookie = inode_peek_iversion_raw(inode);
1002 	stat->attributes_mask |= STATX_ATTR_CHANGE_MONOTONIC;
1003 	if (server->change_attr_type != NFS4_CHANGE_TYPE_IS_UNDEFINED)
1004 		stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
1005 	if (S_ISDIR(inode->i_mode))
1006 		stat->blksize = NFS_SERVER(inode)->dtsize;
1007 out:
1008 	trace_nfs_getattr_exit(inode, err);
1009 	return err;
1010 }
1011 EXPORT_SYMBOL_GPL(nfs_getattr);
1012 
1013 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
1014 {
1015 	refcount_set(&l_ctx->count, 1);
1016 	l_ctx->lockowner = current->files;
1017 	INIT_LIST_HEAD(&l_ctx->list);
1018 	atomic_set(&l_ctx->io_count, 0);
1019 }
1020 
1021 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
1022 {
1023 	struct nfs_lock_context *pos;
1024 
1025 	list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
1026 		if (pos->lockowner != current->files)
1027 			continue;
1028 		if (refcount_inc_not_zero(&pos->count))
1029 			return pos;
1030 	}
1031 	return NULL;
1032 }
1033 
1034 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
1035 {
1036 	struct nfs_lock_context *res, *new = NULL;
1037 	struct inode *inode = d_inode(ctx->dentry);
1038 
1039 	rcu_read_lock();
1040 	res = __nfs_find_lock_context(ctx);
1041 	rcu_read_unlock();
1042 	if (res == NULL) {
1043 		new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1044 		if (new == NULL)
1045 			return ERR_PTR(-ENOMEM);
1046 		nfs_init_lock_context(new);
1047 		spin_lock(&inode->i_lock);
1048 		res = __nfs_find_lock_context(ctx);
1049 		if (res == NULL) {
1050 			new->open_context = get_nfs_open_context(ctx);
1051 			if (new->open_context) {
1052 				list_add_tail_rcu(&new->list,
1053 						&ctx->lock_context.list);
1054 				res = new;
1055 				new = NULL;
1056 			} else
1057 				res = ERR_PTR(-EBADF);
1058 		}
1059 		spin_unlock(&inode->i_lock);
1060 		kfree(new);
1061 	}
1062 	return res;
1063 }
1064 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
1065 
1066 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
1067 {
1068 	struct nfs_open_context *ctx = l_ctx->open_context;
1069 	struct inode *inode = d_inode(ctx->dentry);
1070 
1071 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
1072 		return;
1073 	list_del_rcu(&l_ctx->list);
1074 	spin_unlock(&inode->i_lock);
1075 	put_nfs_open_context(ctx);
1076 	kfree_rcu(l_ctx, rcu_head);
1077 }
1078 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
1079 
1080 /**
1081  * nfs_close_context - Common close_context() routine NFSv2/v3
1082  * @ctx: pointer to context
1083  * @is_sync: is this a synchronous close
1084  *
1085  * Ensure that the attributes are up to date if we're mounted
1086  * with close-to-open semantics and we have cached data that will
1087  * need to be revalidated on open.
1088  */
1089 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1090 {
1091 	struct nfs_inode *nfsi;
1092 	struct inode *inode;
1093 
1094 	if (!(ctx->mode & FMODE_WRITE))
1095 		return;
1096 	if (!is_sync)
1097 		return;
1098 	inode = d_inode(ctx->dentry);
1099 	if (nfs_have_read_or_write_delegation(inode))
1100 		return;
1101 	nfsi = NFS_I(inode);
1102 	if (inode->i_mapping->nrpages == 0)
1103 		return;
1104 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1105 		return;
1106 	if (!list_empty(&nfsi->open_files))
1107 		return;
1108 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1109 		return;
1110 	nfs_revalidate_inode(inode,
1111 			     NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1112 }
1113 EXPORT_SYMBOL_GPL(nfs_close_context);
1114 
1115 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1116 						fmode_t f_mode,
1117 						struct file *filp)
1118 {
1119 	struct nfs_open_context *ctx;
1120 
1121 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1122 	if (!ctx)
1123 		return ERR_PTR(-ENOMEM);
1124 	nfs_sb_active(dentry->d_sb);
1125 	ctx->dentry = dget(dentry);
1126 	if (filp)
1127 		ctx->cred = get_cred(filp->f_cred);
1128 	else
1129 		ctx->cred = get_current_cred();
1130 	rcu_assign_pointer(ctx->ll_cred, NULL);
1131 	ctx->state = NULL;
1132 	ctx->mode = f_mode;
1133 	ctx->flags = 0;
1134 	ctx->error = 0;
1135 	ctx->flock_owner = (fl_owner_t)filp;
1136 	nfs_init_lock_context(&ctx->lock_context);
1137 	ctx->lock_context.open_context = ctx;
1138 	INIT_LIST_HEAD(&ctx->list);
1139 	ctx->mdsthreshold = NULL;
1140 	nfs_localio_file_init(&ctx->nfl);
1141 
1142 	return ctx;
1143 }
1144 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1145 
1146 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1147 {
1148 	if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1149 		return ctx;
1150 	return NULL;
1151 }
1152 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1153 
1154 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1155 {
1156 	struct inode *inode = d_inode(ctx->dentry);
1157 	struct super_block *sb = ctx->dentry->d_sb;
1158 
1159 	if (!refcount_dec_and_test(&ctx->lock_context.count))
1160 		return;
1161 	if (!list_empty(&ctx->list)) {
1162 		spin_lock(&inode->i_lock);
1163 		list_del_rcu(&ctx->list);
1164 		spin_unlock(&inode->i_lock);
1165 	}
1166 	if (inode != NULL)
1167 		NFS_PROTO(inode)->close_context(ctx, is_sync);
1168 	put_cred(ctx->cred);
1169 	dput(ctx->dentry);
1170 	nfs_sb_deactive(sb);
1171 	put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1172 	kfree(ctx->mdsthreshold);
1173 	nfs_close_local_fh(&ctx->nfl);
1174 	kfree_rcu(ctx, rcu_head);
1175 }
1176 
1177 void put_nfs_open_context(struct nfs_open_context *ctx)
1178 {
1179 	__put_nfs_open_context(ctx, 0);
1180 }
1181 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1182 
1183 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1184 {
1185 	__put_nfs_open_context(ctx, 1);
1186 }
1187 
1188 /*
1189  * Ensure that mmap has a recent RPC credential for use when writing out
1190  * shared pages
1191  */
1192 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1193 {
1194 	struct inode *inode = d_inode(ctx->dentry);
1195 	struct nfs_inode *nfsi = NFS_I(inode);
1196 
1197 	spin_lock(&inode->i_lock);
1198 	if (list_empty(&nfsi->open_files) &&
1199 	    nfs_ooo_test(nfsi))
1200 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1201 						     NFS_INO_REVAL_FORCED);
1202 	list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1203 	spin_unlock(&inode->i_lock);
1204 }
1205 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1206 
1207 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1208 {
1209 	filp->private_data = get_nfs_open_context(ctx);
1210 	set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1211 	if (list_empty(&ctx->list))
1212 		nfs_inode_attach_open_context(ctx);
1213 }
1214 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1215 
1216 /*
1217  * Given an inode, search for an open context with the desired characteristics
1218  */
1219 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1220 {
1221 	struct nfs_inode *nfsi = NFS_I(inode);
1222 	struct nfs_open_context *pos, *ctx = NULL;
1223 
1224 	rcu_read_lock();
1225 	list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1226 		if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1227 			continue;
1228 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1229 			continue;
1230 		if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1231 			continue;
1232 		ctx = get_nfs_open_context(pos);
1233 		if (ctx)
1234 			break;
1235 	}
1236 	rcu_read_unlock();
1237 	return ctx;
1238 }
1239 
1240 void nfs_file_clear_open_context(struct file *filp)
1241 {
1242 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1243 
1244 	if (ctx) {
1245 		struct inode *inode = d_inode(ctx->dentry);
1246 
1247 		clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1248 		/*
1249 		 * We fatal error on write before. Try to writeback
1250 		 * every page again.
1251 		 */
1252 		if (ctx->error < 0)
1253 			invalidate_inode_pages2(inode->i_mapping);
1254 		filp->private_data = NULL;
1255 		put_nfs_open_context_sync(ctx);
1256 	}
1257 }
1258 
1259 /*
1260  * These allocate and release file read/write context information.
1261  */
1262 int nfs_open(struct inode *inode, struct file *filp)
1263 {
1264 	struct nfs_open_context *ctx;
1265 
1266 	ctx = alloc_nfs_open_context(file_dentry(filp),
1267 				     flags_to_mode(filp->f_flags), filp);
1268 	if (IS_ERR(ctx))
1269 		return PTR_ERR(ctx);
1270 	nfs_file_set_open_context(filp, ctx);
1271 	put_nfs_open_context(ctx);
1272 	nfs_fscache_open_file(inode, filp);
1273 	return 0;
1274 }
1275 
1276 /*
1277  * This function is called whenever some part of NFS notices that
1278  * the cached attributes have to be refreshed.
1279  */
1280 int
1281 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1282 {
1283 	int		 status = -ESTALE;
1284 	struct nfs_fattr *fattr = NULL;
1285 	struct nfs_inode *nfsi = NFS_I(inode);
1286 
1287 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1288 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1289 
1290 	trace_nfs_revalidate_inode_enter(inode);
1291 
1292 	if (is_bad_inode(inode))
1293 		goto out;
1294 	if (NFS_STALE(inode))
1295 		goto out;
1296 
1297 	/* pNFS: Attributes aren't updated until we layoutcommit */
1298 	if (S_ISREG(inode->i_mode)) {
1299 		status = pnfs_sync_inode(inode, false);
1300 		if (status)
1301 			goto out;
1302 	}
1303 
1304 	status = -ENOMEM;
1305 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1306 	if (fattr == NULL)
1307 		goto out;
1308 
1309 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1310 
1311 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1312 	if (status != 0) {
1313 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1314 			 inode->i_sb->s_id,
1315 			 (unsigned long long)NFS_FILEID(inode), status);
1316 		switch (status) {
1317 		case -ETIMEDOUT:
1318 			/* A soft timeout occurred. Use cached information? */
1319 			if (server->flags & NFS_MOUNT_SOFTREVAL)
1320 				status = 0;
1321 			break;
1322 		case -ESTALE:
1323 			if (!S_ISDIR(inode->i_mode))
1324 				nfs_set_inode_stale(inode);
1325 			else
1326 				nfs_zap_caches(inode);
1327 		}
1328 		goto out;
1329 	}
1330 
1331 	status = nfs_refresh_inode(inode, fattr);
1332 	if (status) {
1333 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1334 			 inode->i_sb->s_id,
1335 			 (unsigned long long)NFS_FILEID(inode), status);
1336 		goto out;
1337 	}
1338 
1339 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1340 		nfs_zap_acl_cache(inode);
1341 
1342 	nfs_setsecurity(inode, fattr);
1343 
1344 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1345 		inode->i_sb->s_id,
1346 		(unsigned long long)NFS_FILEID(inode));
1347 
1348 out:
1349 	nfs_free_fattr(fattr);
1350 	trace_nfs_revalidate_inode_exit(inode, status);
1351 	return status;
1352 }
1353 
1354 int nfs_attribute_cache_expired(struct inode *inode)
1355 {
1356 	if (nfs_have_delegated_attributes(inode))
1357 		return 0;
1358 	return nfs_attribute_timeout(inode);
1359 }
1360 
1361 /**
1362  * nfs_revalidate_inode - Revalidate the inode attributes
1363  * @inode: pointer to inode struct
1364  * @flags: cache flags to check
1365  *
1366  * Updates inode attribute information by retrieving the data from the server.
1367  */
1368 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1369 {
1370 	if (!nfs_check_cache_invalid(inode, flags))
1371 		return NFS_STALE(inode) ? -ESTALE : 0;
1372 	return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1373 }
1374 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1375 
1376 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1377 {
1378 	int ret;
1379 
1380 	nfs_fscache_invalidate(inode, 0);
1381 	if (mapping->nrpages != 0) {
1382 		if (S_ISREG(inode->i_mode)) {
1383 			ret = nfs_sync_mapping(mapping);
1384 			if (ret < 0)
1385 				return ret;
1386 		}
1387 		ret = invalidate_inode_pages2(mapping);
1388 		if (ret < 0)
1389 			return ret;
1390 	}
1391 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1392 
1393 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1394 			inode->i_sb->s_id,
1395 			(unsigned long long)NFS_FILEID(inode));
1396 	return 0;
1397 }
1398 
1399 /**
1400  * nfs_clear_invalid_mapping - Conditionally clear a mapping
1401  * @mapping: pointer to mapping
1402  *
1403  * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1404  */
1405 int nfs_clear_invalid_mapping(struct address_space *mapping)
1406 {
1407 	struct inode *inode = mapping->host;
1408 	struct nfs_inode *nfsi = NFS_I(inode);
1409 	unsigned long *bitlock = &nfsi->flags;
1410 	int ret = 0;
1411 
1412 	/*
1413 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1414 	 * invalidations that come in while we're shooting down the mappings
1415 	 * are respected. But, that leaves a race window where one revalidator
1416 	 * can clear the flag, and then another checks it before the mapping
1417 	 * gets invalidated. Fix that by serializing access to this part of
1418 	 * the function.
1419 	 *
1420 	 * At the same time, we need to allow other tasks to see whether we
1421 	 * might be in the middle of invalidating the pages, so we only set
1422 	 * the bit lock here if it looks like we're going to be doing that.
1423 	 */
1424 	for (;;) {
1425 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1426 					 nfs_wait_bit_killable,
1427 					 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1428 		if (ret)
1429 			goto out;
1430 		smp_rmb(); /* pairs with smp_wmb() below */
1431 		if (test_bit(NFS_INO_INVALIDATING, bitlock))
1432 			continue;
1433 		/* pairs with nfs_set_cache_invalid()'s smp_store_release() */
1434 		if (!(smp_load_acquire(&nfsi->cache_validity) & NFS_INO_INVALID_DATA))
1435 			goto out;
1436 		/* Slow-path that double-checks with spinlock held */
1437 		spin_lock(&inode->i_lock);
1438 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1439 			spin_unlock(&inode->i_lock);
1440 			continue;
1441 		}
1442 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1443 			break;
1444 		spin_unlock(&inode->i_lock);
1445 		goto out;
1446 	}
1447 
1448 	set_bit(NFS_INO_INVALIDATING, bitlock);
1449 	smp_wmb();
1450 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1451 	nfs_ooo_clear(nfsi);
1452 	spin_unlock(&inode->i_lock);
1453 	trace_nfs_invalidate_mapping_enter(inode);
1454 	ret = nfs_invalidate_mapping(inode, mapping);
1455 	trace_nfs_invalidate_mapping_exit(inode, ret);
1456 
1457 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1458 	smp_mb__after_atomic();
1459 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1460 out:
1461 	return ret;
1462 }
1463 
1464 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1465 {
1466 	return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1467 		NFS_STALE(inode);
1468 }
1469 
1470 int nfs_revalidate_mapping_rcu(struct inode *inode)
1471 {
1472 	struct nfs_inode *nfsi = NFS_I(inode);
1473 	unsigned long *bitlock = &nfsi->flags;
1474 	int ret = 0;
1475 
1476 	if (IS_SWAPFILE(inode))
1477 		goto out;
1478 	if (nfs_mapping_need_revalidate_inode(inode)) {
1479 		ret = -ECHILD;
1480 		goto out;
1481 	}
1482 	spin_lock(&inode->i_lock);
1483 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1484 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1485 		ret = -ECHILD;
1486 	spin_unlock(&inode->i_lock);
1487 out:
1488 	return ret;
1489 }
1490 
1491 /**
1492  * nfs_revalidate_mapping - Revalidate the pagecache
1493  * @inode: pointer to host inode
1494  * @mapping: pointer to mapping
1495  */
1496 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1497 {
1498 	/* swapfiles are not supposed to be shared. */
1499 	if (IS_SWAPFILE(inode))
1500 		return 0;
1501 
1502 	if (nfs_mapping_need_revalidate_inode(inode)) {
1503 		int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1504 		if (ret < 0)
1505 			return ret;
1506 	}
1507 
1508 	return nfs_clear_invalid_mapping(mapping);
1509 }
1510 
1511 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1512 {
1513 	struct inode *inode = &nfsi->vfs_inode;
1514 
1515 	if (!S_ISREG(inode->i_mode))
1516 		return false;
1517 	if (list_empty(&nfsi->open_files))
1518 		return false;
1519 	return inode_is_open_for_write(inode);
1520 }
1521 
1522 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1523 {
1524 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1525 }
1526 
1527 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1528 {
1529 	struct timespec64 ts;
1530 
1531 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1532 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1533 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1534 		inode_set_iversion_raw(inode, fattr->change_attr);
1535 		if (S_ISDIR(inode->i_mode))
1536 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1537 		else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1538 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1539 	}
1540 	/* If we have atomic WCC data, we may update some attributes */
1541 	ts = inode_get_ctime(inode);
1542 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1543 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1544 			&& timespec64_equal(&ts, &fattr->pre_ctime)) {
1545 		inode_set_ctime_to_ts(inode, fattr->ctime);
1546 	}
1547 
1548 	ts = inode_get_mtime(inode);
1549 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1550 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1551 			&& timespec64_equal(&ts, &fattr->pre_mtime)) {
1552 		inode_set_mtime_to_ts(inode, fattr->mtime);
1553 	}
1554 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1555 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1556 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1557 			&& !nfs_have_writebacks(inode)) {
1558 		trace_nfs_size_wcc(inode, fattr->size);
1559 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1560 	}
1561 }
1562 
1563 /**
1564  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1565  * @inode: pointer to inode
1566  * @fattr: updated attributes
1567  *
1568  * Verifies the attribute cache. If we have just changed the attributes,
1569  * so that fattr carries weak cache consistency data, then it may
1570  * also update the ctime/mtime/change_attribute.
1571  */
1572 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1573 {
1574 	struct nfs_inode *nfsi = NFS_I(inode);
1575 	loff_t cur_size, new_isize;
1576 	unsigned long invalid = 0;
1577 	struct timespec64 ts;
1578 
1579 	if (nfs_have_delegated_attributes(inode))
1580 		return 0;
1581 
1582 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1583 		/* Only a mounted-on-fileid? Just exit */
1584 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1585 			return 0;
1586 	/* Has the inode gone and changed behind our back? */
1587 	} else if (nfsi->fileid != fattr->fileid) {
1588 		/* Is this perhaps the mounted-on fileid? */
1589 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1590 		    nfsi->fileid == fattr->mounted_on_fileid)
1591 			return 0;
1592 		return -ESTALE;
1593 	}
1594 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1595 		return -ESTALE;
1596 
1597 
1598 	if (!nfs_file_has_buffered_writers(nfsi)) {
1599 		/* Verify a few of the more important attributes */
1600 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1601 			invalid |= NFS_INO_INVALID_CHANGE;
1602 
1603 		ts = inode_get_mtime(inode);
1604 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1605 			invalid |= NFS_INO_INVALID_MTIME;
1606 
1607 		ts = inode_get_ctime(inode);
1608 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1609 			invalid |= NFS_INO_INVALID_CTIME;
1610 
1611 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1612 			cur_size = i_size_read(inode);
1613 			new_isize = nfs_size_to_loff_t(fattr->size);
1614 			if (cur_size != new_isize)
1615 				invalid |= NFS_INO_INVALID_SIZE;
1616 		}
1617 	}
1618 
1619 	/* Have any file permissions changed? */
1620 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1621 		invalid |= NFS_INO_INVALID_MODE;
1622 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1623 		invalid |= NFS_INO_INVALID_OTHER;
1624 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1625 		invalid |= NFS_INO_INVALID_OTHER;
1626 
1627 	/* Has the link count changed? */
1628 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1629 		invalid |= NFS_INO_INVALID_NLINK;
1630 
1631 	ts = inode_get_atime(inode);
1632 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1633 		invalid |= NFS_INO_INVALID_ATIME;
1634 
1635 	if (invalid != 0)
1636 		nfs_set_cache_invalid(inode, invalid);
1637 
1638 	nfsi->read_cache_jiffies = fattr->time_start;
1639 	return 0;
1640 }
1641 
1642 static atomic_long_t nfs_attr_generation_counter;
1643 
1644 static unsigned long nfs_read_attr_generation_counter(void)
1645 {
1646 	return atomic_long_read(&nfs_attr_generation_counter);
1647 }
1648 
1649 unsigned long nfs_inc_attr_generation_counter(void)
1650 {
1651 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1652 }
1653 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1654 
1655 void nfs_fattr_init(struct nfs_fattr *fattr)
1656 {
1657 	fattr->valid = 0;
1658 	fattr->time_start = jiffies;
1659 	fattr->gencount = nfs_inc_attr_generation_counter();
1660 	fattr->owner_name = NULL;
1661 	fattr->group_name = NULL;
1662 	fattr->mdsthreshold = NULL;
1663 }
1664 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1665 
1666 /**
1667  * nfs_fattr_set_barrier
1668  * @fattr: attributes
1669  *
1670  * Used to set a barrier after an attribute was updated. This
1671  * barrier ensures that older attributes from RPC calls that may
1672  * have raced with our update cannot clobber these new values.
1673  * Note that you are still responsible for ensuring that other
1674  * operations which change the attribute on the server do not
1675  * collide.
1676  */
1677 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1678 {
1679 	fattr->gencount = nfs_inc_attr_generation_counter();
1680 }
1681 
1682 struct nfs_fattr *nfs_alloc_fattr(void)
1683 {
1684 	struct nfs_fattr *fattr;
1685 
1686 	fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1687 	if (fattr != NULL) {
1688 		nfs_fattr_init(fattr);
1689 		fattr->label = NULL;
1690 	}
1691 	return fattr;
1692 }
1693 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1694 
1695 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1696 {
1697 	struct nfs_fattr *fattr = nfs_alloc_fattr();
1698 
1699 	if (!fattr)
1700 		return NULL;
1701 
1702 	fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1703 	if (IS_ERR(fattr->label)) {
1704 		kfree(fattr);
1705 		return NULL;
1706 	}
1707 
1708 	return fattr;
1709 }
1710 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1711 
1712 struct nfs_fh *nfs_alloc_fhandle(void)
1713 {
1714 	struct nfs_fh *fh;
1715 
1716 	fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1717 	if (fh != NULL)
1718 		fh->size = 0;
1719 	return fh;
1720 }
1721 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1722 
1723 #ifdef NFS_DEBUG
1724 /*
1725  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1726  *                             in the same way that wireshark does
1727  *
1728  * @fh: file handle
1729  *
1730  * For debugging only.
1731  */
1732 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1733 {
1734 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1735 	 * not on the result */
1736 	return nfs_fhandle_hash(fh);
1737 }
1738 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1739 
1740 /*
1741  * _nfs_display_fhandle - display an NFS file handle on the console
1742  *
1743  * @fh: file handle to display
1744  * @caption: display caption
1745  *
1746  * For debugging only.
1747  */
1748 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1749 {
1750 	unsigned short i;
1751 
1752 	if (fh == NULL || fh->size == 0) {
1753 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1754 		return;
1755 	}
1756 
1757 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1758 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1759 	for (i = 0; i < fh->size; i += 16) {
1760 		__be32 *pos = (__be32 *)&fh->data[i];
1761 
1762 		switch ((fh->size - i - 1) >> 2) {
1763 		case 0:
1764 			printk(KERN_DEFAULT " %08x\n",
1765 				be32_to_cpup(pos));
1766 			break;
1767 		case 1:
1768 			printk(KERN_DEFAULT " %08x %08x\n",
1769 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1770 			break;
1771 		case 2:
1772 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1773 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1774 				be32_to_cpup(pos + 2));
1775 			break;
1776 		default:
1777 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1778 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1779 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1780 		}
1781 	}
1782 }
1783 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1784 #endif
1785 
1786 /**
1787  * nfs_inode_attrs_cmp_generic - compare attributes
1788  * @fattr: attributes
1789  * @inode: pointer to inode
1790  *
1791  * Attempt to divine whether or not an RPC call reply carrying stale
1792  * attributes got scheduled after another call carrying updated ones.
1793  * Note also the check for wraparound of 'attr_gencount'
1794  *
1795  * The function returns '1' if it thinks the attributes in @fattr are
1796  * more recent than the ones cached in @inode. Otherwise it returns
1797  * the value '0'.
1798  */
1799 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1800 				       const struct inode *inode)
1801 {
1802 	unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1803 
1804 	return (long)(fattr->gencount - attr_gencount) > 0 ||
1805 	       (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1806 }
1807 
1808 /**
1809  * nfs_inode_attrs_cmp_monotonic - compare attributes
1810  * @fattr: attributes
1811  * @inode: pointer to inode
1812  *
1813  * Attempt to divine whether or not an RPC call reply carrying stale
1814  * attributes got scheduled after another call carrying updated ones.
1815  *
1816  * We assume that the server observes monotonic semantics for
1817  * the change attribute, so a larger value means that the attributes in
1818  * @fattr are more recent, in which case the function returns the
1819  * value '1'.
1820  * A return value of '0' indicates no measurable change
1821  * A return value of '-1' means that the attributes in @inode are
1822  * more recent.
1823  */
1824 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1825 					 const struct inode *inode)
1826 {
1827 	s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1828 	if (diff > 0)
1829 		return 1;
1830 	return diff == 0 ? 0 : -1;
1831 }
1832 
1833 /**
1834  * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1835  * @fattr: attributes
1836  * @inode: pointer to inode
1837  *
1838  * Attempt to divine whether or not an RPC call reply carrying stale
1839  * attributes got scheduled after another call carrying updated ones.
1840  *
1841  * We assume that the server observes strictly monotonic semantics for
1842  * the change attribute, so a larger value means that the attributes in
1843  * @fattr are more recent, in which case the function returns the
1844  * value '1'.
1845  * A return value of '-1' means that the attributes in @inode are
1846  * more recent or unchanged.
1847  */
1848 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1849 						const struct inode *inode)
1850 {
1851 	return  nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1852 }
1853 
1854 /**
1855  * nfs_inode_attrs_cmp - compare attributes
1856  * @fattr: attributes
1857  * @inode: pointer to inode
1858  *
1859  * This function returns '1' if it thinks the attributes in @fattr are
1860  * more recent than the ones cached in @inode. It returns '-1' if
1861  * the attributes in @inode are more recent than the ones in @fattr,
1862  * and it returns 0 if not sure.
1863  */
1864 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1865 			       const struct inode *inode)
1866 {
1867 	if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1868 		return 1;
1869 	switch (NFS_SERVER(inode)->change_attr_type) {
1870 	case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1871 		break;
1872 	case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1873 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1874 			break;
1875 		return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1876 	default:
1877 		if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1878 			break;
1879 		return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1880 	}
1881 	return 0;
1882 }
1883 
1884 /**
1885  * nfs_inode_finish_partial_attr_update - complete a previous inode update
1886  * @fattr: attributes
1887  * @inode: pointer to inode
1888  *
1889  * Returns '1' if the last attribute update left the inode cached
1890  * attributes in a partially unrevalidated state, and @fattr
1891  * matches the change attribute of that partial update.
1892  * Otherwise returns '0'.
1893  */
1894 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1895 						const struct inode *inode)
1896 {
1897 	const unsigned long check_valid =
1898 		NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1899 		NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1900 		NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1901 		NFS_INO_INVALID_NLINK;
1902 	unsigned long cache_validity = NFS_I(inode)->cache_validity;
1903 	enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1904 
1905 	if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1906 	    !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1907 	    (cache_validity & check_valid) != 0 &&
1908 	    (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1909 	    nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1910 		return 1;
1911 	return 0;
1912 }
1913 
1914 static void nfs_ooo_merge(struct nfs_inode *nfsi,
1915 			  u64 start, u64 end)
1916 {
1917 	int i, cnt;
1918 
1919 	if (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)
1920 		/* No point merging anything */
1921 		return;
1922 
1923 	if (!nfsi->ooo) {
1924 		nfsi->ooo = kmalloc(sizeof(*nfsi->ooo), GFP_ATOMIC);
1925 		if (!nfsi->ooo) {
1926 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1927 			return;
1928 		}
1929 		nfsi->ooo->cnt = 0;
1930 	}
1931 
1932 	/* add this range, merging if possible */
1933 	cnt = nfsi->ooo->cnt;
1934 	for (i = 0; i < cnt; i++) {
1935 		if (end == nfsi->ooo->gap[i].start)
1936 			end = nfsi->ooo->gap[i].end;
1937 		else if (start == nfsi->ooo->gap[i].end)
1938 			start = nfsi->ooo->gap[i].start;
1939 		else
1940 			continue;
1941 		/* Remove 'i' from table and loop to insert the new range */
1942 		cnt -= 1;
1943 		nfsi->ooo->gap[i] = nfsi->ooo->gap[cnt];
1944 		i = -1;
1945 	}
1946 	if (start != end) {
1947 		if (cnt >= ARRAY_SIZE(nfsi->ooo->gap)) {
1948 			nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1949 			kfree(nfsi->ooo);
1950 			nfsi->ooo = NULL;
1951 			return;
1952 		}
1953 		nfsi->ooo->gap[cnt].start = start;
1954 		nfsi->ooo->gap[cnt].end = end;
1955 		cnt += 1;
1956 	}
1957 	nfsi->ooo->cnt = cnt;
1958 }
1959 
1960 static void nfs_ooo_record(struct nfs_inode *nfsi,
1961 			   struct nfs_fattr *fattr)
1962 {
1963 	/* This reply was out-of-order, so record in the
1964 	 * pre/post change id, possibly cancelling
1965 	 * gaps created when iversion was jumpped forward.
1966 	 */
1967 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) &&
1968 	    (fattr->valid & NFS_ATTR_FATTR_PRECHANGE))
1969 		nfs_ooo_merge(nfsi,
1970 			      fattr->change_attr,
1971 			      fattr->pre_change_attr);
1972 }
1973 
1974 static int nfs_refresh_inode_locked(struct inode *inode,
1975 				    struct nfs_fattr *fattr)
1976 {
1977 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1978 	int ret = 0;
1979 
1980 	trace_nfs_refresh_inode_enter(inode);
1981 
1982 	if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1983 		ret = nfs_update_inode(inode, fattr);
1984 	else {
1985 		nfs_ooo_record(NFS_I(inode), fattr);
1986 
1987 		if (attr_cmp == 0)
1988 			ret = nfs_check_inode_attributes(inode, fattr);
1989 	}
1990 
1991 	trace_nfs_refresh_inode_exit(inode, ret);
1992 	return ret;
1993 }
1994 
1995 /**
1996  * nfs_refresh_inode - try to update the inode attribute cache
1997  * @inode: pointer to inode
1998  * @fattr: updated attributes
1999  *
2000  * Check that an RPC call that returned attributes has not overlapped with
2001  * other recent updates of the inode metadata, then decide whether it is
2002  * safe to do a full update of the inode attributes, or whether just to
2003  * call nfs_check_inode_attributes.
2004  */
2005 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
2006 {
2007 	int status;
2008 
2009 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
2010 		return 0;
2011 	spin_lock(&inode->i_lock);
2012 	status = nfs_refresh_inode_locked(inode, fattr);
2013 	spin_unlock(&inode->i_lock);
2014 
2015 	return status;
2016 }
2017 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
2018 
2019 static int nfs_post_op_update_inode_locked(struct inode *inode,
2020 		struct nfs_fattr *fattr, unsigned int invalid)
2021 {
2022 	if (S_ISDIR(inode->i_mode))
2023 		invalid |= NFS_INO_INVALID_DATA;
2024 	nfs_set_cache_invalid(inode, invalid);
2025 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
2026 		return 0;
2027 	return nfs_refresh_inode_locked(inode, fattr);
2028 }
2029 
2030 /**
2031  * nfs_post_op_update_inode - try to update the inode attribute cache
2032  * @inode: pointer to inode
2033  * @fattr: updated attributes
2034  *
2035  * After an operation that has changed the inode metadata, mark the
2036  * attribute cache as being invalid, then try to update it.
2037  *
2038  * NB: if the server didn't return any post op attributes, this
2039  * function will force the retrieval of attributes before the next
2040  * NFS request.  Thus it should be used only for operations that
2041  * are expected to change one or more attributes, to avoid
2042  * unnecessary NFS requests and trips through nfs_update_inode().
2043  */
2044 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2045 {
2046 	int status;
2047 
2048 	spin_lock(&inode->i_lock);
2049 	nfs_fattr_set_barrier(fattr);
2050 	status = nfs_post_op_update_inode_locked(inode, fattr,
2051 			NFS_INO_INVALID_CHANGE
2052 			| NFS_INO_INVALID_CTIME
2053 			| NFS_INO_REVAL_FORCED);
2054 	spin_unlock(&inode->i_lock);
2055 
2056 	return status;
2057 }
2058 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
2059 
2060 /**
2061  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
2062  * @inode: pointer to inode
2063  * @fattr: updated attributes
2064  *
2065  * After an operation that has changed the inode metadata, mark the
2066  * attribute cache as being invalid, then try to update it. Fake up
2067  * weak cache consistency data, if none exist.
2068  *
2069  * This function is mainly designed to be used by the ->write_done() functions.
2070  */
2071 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
2072 {
2073 	int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
2074 	int status;
2075 
2076 	/* Don't do a WCC update if these attributes are already stale */
2077 	if (attr_cmp < 0)
2078 		return 0;
2079 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
2080 		/* Record the pre/post change info before clearing PRECHANGE */
2081 		nfs_ooo_record(NFS_I(inode), fattr);
2082 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
2083 				| NFS_ATTR_FATTR_PRESIZE
2084 				| NFS_ATTR_FATTR_PREMTIME
2085 				| NFS_ATTR_FATTR_PRECTIME);
2086 		goto out_noforce;
2087 	}
2088 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
2089 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
2090 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
2091 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
2092 	}
2093 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
2094 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
2095 		fattr->pre_ctime = inode_get_ctime(inode);
2096 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
2097 	}
2098 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
2099 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
2100 		fattr->pre_mtime = inode_get_mtime(inode);
2101 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
2102 	}
2103 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
2104 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
2105 		fattr->pre_size = i_size_read(inode);
2106 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
2107 	}
2108 out_noforce:
2109 	status = nfs_post_op_update_inode_locked(inode, fattr,
2110 			NFS_INO_INVALID_CHANGE
2111 			| NFS_INO_INVALID_CTIME
2112 			| NFS_INO_INVALID_MTIME
2113 			| NFS_INO_INVALID_BLOCKS);
2114 	return status;
2115 }
2116 
2117 /**
2118  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
2119  * @inode: pointer to inode
2120  * @fattr: updated attributes
2121  *
2122  * After an operation that has changed the inode metadata, mark the
2123  * attribute cache as being invalid, then try to update it. Fake up
2124  * weak cache consistency data, if none exist.
2125  *
2126  * This function is mainly designed to be used by the ->write_done() functions.
2127  */
2128 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
2129 {
2130 	int status;
2131 
2132 	spin_lock(&inode->i_lock);
2133 	nfs_fattr_set_barrier(fattr);
2134 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
2135 	spin_unlock(&inode->i_lock);
2136 	return status;
2137 }
2138 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
2139 
2140 
2141 /*
2142  * Many nfs protocol calls return the new file attributes after
2143  * an operation.  Here we update the inode to reflect the state
2144  * of the server's inode.
2145  *
2146  * This is a bit tricky because we have to make sure all dirty pages
2147  * have been sent off to the server before calling invalidate_inode_pages.
2148  * To make sure no other process adds more write requests while we try
2149  * our best to flush them, we make them sleep during the attribute refresh.
2150  *
2151  * A very similar scenario holds for the dir cache.
2152  */
2153 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
2154 {
2155 	struct nfs_server *server = NFS_SERVER(inode);
2156 	struct nfs_inode *nfsi = NFS_I(inode);
2157 	loff_t cur_isize, new_isize;
2158 	u64 fattr_supported = server->fattr_valid;
2159 	unsigned long invalid = 0;
2160 	unsigned long now = jiffies;
2161 	unsigned long save_cache_validity;
2162 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
2163 	bool cache_revalidated = true;
2164 	bool attr_changed = false;
2165 	bool have_delegation;
2166 
2167 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
2168 			__func__, inode->i_sb->s_id, inode->i_ino,
2169 			nfs_display_fhandle_hash(NFS_FH(inode)),
2170 			atomic_read(&inode->i_count), fattr->valid);
2171 
2172 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2173 		/* Only a mounted-on-fileid? Just exit */
2174 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2175 			return 0;
2176 	/* Has the inode gone and changed behind our back? */
2177 	} else if (nfsi->fileid != fattr->fileid) {
2178 		/* Is this perhaps the mounted-on fileid? */
2179 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2180 		    nfsi->fileid == fattr->mounted_on_fileid)
2181 			return 0;
2182 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2183 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2184 			NFS_SERVER(inode)->nfs_client->cl_hostname,
2185 			inode->i_sb->s_id, (long long)nfsi->fileid,
2186 			(long long)fattr->fileid);
2187 		goto out_err;
2188 	}
2189 
2190 	/*
2191 	 * Make sure the inode's type hasn't changed.
2192 	 */
2193 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2194 		/*
2195 		* Big trouble! The inode has become a different object.
2196 		*/
2197 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2198 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
2199 		goto out_err;
2200 	}
2201 
2202 	/* Update the fsid? */
2203 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2204 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2205 			!IS_AUTOMOUNT(inode))
2206 		server->fsid = fattr->fsid;
2207 
2208 	/* Save the delegation state before clearing cache_validity */
2209 	have_delegation = nfs_have_delegated_attributes(inode);
2210 
2211 	/*
2212 	 * Update the read time so we don't revalidate too often.
2213 	 */
2214 	nfsi->read_cache_jiffies = fattr->time_start;
2215 
2216 	/* Fix up any delegated attributes in the struct nfs_fattr */
2217 	nfs_fattr_fixup_delegated(inode, fattr);
2218 
2219 	save_cache_validity = nfsi->cache_validity;
2220 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2221 			| NFS_INO_INVALID_ATIME
2222 			| NFS_INO_REVAL_FORCED
2223 			| NFS_INO_INVALID_BLOCKS);
2224 
2225 	/* Do atomic weak cache consistency updates */
2226 	nfs_wcc_update_inode(inode, fattr);
2227 
2228 	if (pnfs_layoutcommit_outstanding(inode)) {
2229 		nfsi->cache_validity |=
2230 			save_cache_validity &
2231 			(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2232 			 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2233 			 NFS_INO_INVALID_BLOCKS);
2234 		cache_revalidated = false;
2235 	}
2236 
2237 	/* More cache consistency checks */
2238 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2239 		if (!have_writers && nfsi->ooo && nfsi->ooo->cnt == 1 &&
2240 		    nfsi->ooo->gap[0].end == inode_peek_iversion_raw(inode)) {
2241 			/* There is one remaining gap that hasn't been
2242 			 * merged into iversion - do that now.
2243 			 */
2244 			inode_set_iversion_raw(inode, nfsi->ooo->gap[0].start);
2245 			kfree(nfsi->ooo);
2246 			nfsi->ooo = NULL;
2247 		}
2248 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2249 			/* Could it be a race with writeback? */
2250 			if (!(have_writers || have_delegation)) {
2251 				invalid |= NFS_INO_INVALID_DATA
2252 					| NFS_INO_INVALID_ACCESS
2253 					| NFS_INO_INVALID_ACL
2254 					| NFS_INO_INVALID_XATTR;
2255 				/* Force revalidate of all attributes */
2256 				save_cache_validity |= NFS_INO_INVALID_CTIME
2257 					| NFS_INO_INVALID_MTIME
2258 					| NFS_INO_INVALID_SIZE
2259 					| NFS_INO_INVALID_BLOCKS
2260 					| NFS_INO_INVALID_NLINK
2261 					| NFS_INO_INVALID_MODE
2262 					| NFS_INO_INVALID_OTHER;
2263 				if (S_ISDIR(inode->i_mode))
2264 					nfs_force_lookup_revalidate(inode);
2265 				attr_changed = true;
2266 				dprintk("NFS: change_attr change on server for file %s/%ld\n",
2267 						inode->i_sb->s_id,
2268 						inode->i_ino);
2269 			} else if (!have_delegation) {
2270 				nfs_ooo_record(nfsi, fattr);
2271 				nfs_ooo_merge(nfsi, inode_peek_iversion_raw(inode),
2272 					      fattr->change_attr);
2273 			}
2274 			inode_set_iversion_raw(inode, fattr->change_attr);
2275 		}
2276 	} else {
2277 		nfsi->cache_validity |=
2278 			save_cache_validity & NFS_INO_INVALID_CHANGE;
2279 		if (!have_delegation ||
2280 		    (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2281 			cache_revalidated = false;
2282 	}
2283 
2284 	if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2285 		inode_set_mtime_to_ts(inode, fattr->mtime);
2286 	else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2287 		nfsi->cache_validity |=
2288 			save_cache_validity & NFS_INO_INVALID_MTIME;
2289 
2290 	if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2291 		inode_set_ctime_to_ts(inode, fattr->ctime);
2292 	else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2293 		nfsi->cache_validity |=
2294 			save_cache_validity & NFS_INO_INVALID_CTIME;
2295 
2296 	/* Check if our cached file size is stale */
2297 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2298 		new_isize = nfs_size_to_loff_t(fattr->size);
2299 		cur_isize = i_size_read(inode);
2300 		if (new_isize != cur_isize && !have_delegation) {
2301 			/* Do we perhaps have any outstanding writes, or has
2302 			 * the file grown beyond our last write? */
2303 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2304 				trace_nfs_size_update(inode, new_isize);
2305 				i_size_write(inode, new_isize);
2306 				if (!have_writers)
2307 					invalid |= NFS_INO_INVALID_DATA;
2308 			}
2309 		}
2310 		if (new_isize == 0 &&
2311 		    !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2312 				      NFS_ATTR_FATTR_BLOCKS_USED))) {
2313 			fattr->du.nfs3.used = 0;
2314 			fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2315 		}
2316 	} else
2317 		nfsi->cache_validity |=
2318 			save_cache_validity & NFS_INO_INVALID_SIZE;
2319 
2320 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2321 		inode_set_atime_to_ts(inode, fattr->atime);
2322 	else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2323 		nfsi->cache_validity |=
2324 			save_cache_validity & NFS_INO_INVALID_ATIME;
2325 
2326 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2327 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2328 			umode_t newmode = inode->i_mode & S_IFMT;
2329 			newmode |= fattr->mode & S_IALLUGO;
2330 			inode->i_mode = newmode;
2331 			invalid |= NFS_INO_INVALID_ACCESS
2332 				| NFS_INO_INVALID_ACL;
2333 		}
2334 	} else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2335 		nfsi->cache_validity |=
2336 			save_cache_validity & NFS_INO_INVALID_MODE;
2337 
2338 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2339 		if (!uid_eq(inode->i_uid, fattr->uid)) {
2340 			invalid |= NFS_INO_INVALID_ACCESS
2341 				| NFS_INO_INVALID_ACL;
2342 			inode->i_uid = fattr->uid;
2343 		}
2344 	} else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2345 		nfsi->cache_validity |=
2346 			save_cache_validity & NFS_INO_INVALID_OTHER;
2347 
2348 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2349 		if (!gid_eq(inode->i_gid, fattr->gid)) {
2350 			invalid |= NFS_INO_INVALID_ACCESS
2351 				| NFS_INO_INVALID_ACL;
2352 			inode->i_gid = fattr->gid;
2353 		}
2354 	} else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2355 		nfsi->cache_validity |=
2356 			save_cache_validity & NFS_INO_INVALID_OTHER;
2357 
2358 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2359 		if (inode->i_nlink != fattr->nlink)
2360 			set_nlink(inode, fattr->nlink);
2361 	} else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2362 		nfsi->cache_validity |=
2363 			save_cache_validity & NFS_INO_INVALID_NLINK;
2364 
2365 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2366 		/*
2367 		 * report the blocks in 512byte units
2368 		 */
2369 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2370 	} else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2371 		nfsi->cache_validity |=
2372 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2373 
2374 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2375 		inode->i_blocks = fattr->du.nfs2.blocks;
2376 	else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2377 		nfsi->cache_validity |=
2378 			save_cache_validity & NFS_INO_INVALID_BLOCKS;
2379 
2380 	/* Update attrtimeo value if we're out of the unstable period */
2381 	if (attr_changed) {
2382 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2383 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2384 		nfsi->attrtimeo_timestamp = now;
2385 		/* Set barrier to be more recent than all outstanding updates */
2386 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2387 	} else {
2388 		if (cache_revalidated) {
2389 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2390 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2391 				nfsi->attrtimeo <<= 1;
2392 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2393 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2394 			}
2395 			nfsi->attrtimeo_timestamp = now;
2396 		}
2397 		/* Set the barrier to be more recent than this fattr */
2398 		if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2399 			nfsi->attr_gencount = fattr->gencount;
2400 	}
2401 
2402 	/* Don't invalidate the data if we were to blame */
2403 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2404 				|| S_ISLNK(inode->i_mode)))
2405 		invalid &= ~NFS_INO_INVALID_DATA;
2406 	nfs_set_cache_invalid(inode, invalid);
2407 
2408 	return 0;
2409  out_err:
2410 	/*
2411 	 * No need to worry about unhashing the dentry, as the
2412 	 * lookup validation will know that the inode is bad.
2413 	 * (But we fall through to invalidate the caches.)
2414 	 */
2415 	nfs_set_inode_stale_locked(inode);
2416 	return -ESTALE;
2417 }
2418 
2419 struct inode *nfs_alloc_inode(struct super_block *sb)
2420 {
2421 	struct nfs_inode *nfsi;
2422 	nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2423 	if (!nfsi)
2424 		return NULL;
2425 	nfsi->flags = 0UL;
2426 	nfsi->cache_validity = 0UL;
2427 	nfsi->ooo = NULL;
2428 #if IS_ENABLED(CONFIG_NFS_V4)
2429 	nfsi->nfs4_acl = NULL;
2430 #endif /* CONFIG_NFS_V4 */
2431 #ifdef CONFIG_NFS_V4_2
2432 	nfsi->xattr_cache = NULL;
2433 #endif
2434 	nfs_netfs_inode_init(nfsi);
2435 
2436 	return &nfsi->vfs_inode;
2437 }
2438 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2439 
2440 void nfs_free_inode(struct inode *inode)
2441 {
2442 	kfree(NFS_I(inode)->ooo);
2443 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2444 }
2445 EXPORT_SYMBOL_GPL(nfs_free_inode);
2446 
2447 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2448 {
2449 #if IS_ENABLED(CONFIG_NFS_V4)
2450 	INIT_LIST_HEAD(&nfsi->open_states);
2451 	nfsi->delegation = NULL;
2452 	init_rwsem(&nfsi->rwsem);
2453 	nfsi->layout = NULL;
2454 #endif
2455 }
2456 
2457 static void init_once(void *foo)
2458 {
2459 	struct nfs_inode *nfsi = foo;
2460 
2461 	inode_init_once(&nfsi->vfs_inode);
2462 	INIT_LIST_HEAD(&nfsi->open_files);
2463 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2464 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2465 	nfs4_init_once(nfsi);
2466 }
2467 
2468 static int __init nfs_init_inodecache(void)
2469 {
2470 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2471 					     sizeof(struct nfs_inode),
2472 					     0, (SLAB_RECLAIM_ACCOUNT|
2473 						SLAB_ACCOUNT),
2474 					     init_once);
2475 	if (nfs_inode_cachep == NULL)
2476 		return -ENOMEM;
2477 
2478 	return 0;
2479 }
2480 
2481 static void nfs_destroy_inodecache(void)
2482 {
2483 	/*
2484 	 * Make sure all delayed rcu free inodes are flushed before we
2485 	 * destroy cache.
2486 	 */
2487 	rcu_barrier();
2488 	kmem_cache_destroy(nfs_inode_cachep);
2489 }
2490 
2491 struct workqueue_struct *nfslocaliod_workqueue;
2492 struct workqueue_struct *nfsiod_workqueue;
2493 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2494 
2495 /*
2496  * Destroy the nfsiod workqueues
2497  */
2498 static void nfsiod_stop(void)
2499 {
2500 	struct workqueue_struct *wq;
2501 
2502 	wq = nfsiod_workqueue;
2503 	if (wq != NULL) {
2504 		nfsiod_workqueue = NULL;
2505 		destroy_workqueue(wq);
2506 	}
2507 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
2508 	wq = nfslocaliod_workqueue;
2509 	if (wq != NULL) {
2510 		nfslocaliod_workqueue = NULL;
2511 		destroy_workqueue(wq);
2512 	}
2513 #endif /* CONFIG_NFS_LOCALIO */
2514 }
2515 
2516 /*
2517  * Start the nfsiod workqueues
2518  */
2519 static int nfsiod_start(void)
2520 {
2521 	dprintk("RPC:       creating workqueue nfsiod\n");
2522 	nfsiod_workqueue = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2523 	if (nfsiod_workqueue == NULL)
2524 		return -ENOMEM;
2525 #if IS_ENABLED(CONFIG_NFS_LOCALIO)
2526 	/*
2527 	 * localio writes need to use a normal (non-memreclaim) workqueue.
2528 	 * When we start getting low on space, XFS goes and calls flush_work() on
2529 	 * a non-memreclaim work queue, which causes a priority inversion problem.
2530 	 */
2531 	dprintk("RPC:       creating workqueue nfslocaliod\n");
2532 	nfslocaliod_workqueue = alloc_workqueue("nfslocaliod", WQ_UNBOUND, 0);
2533 	if (unlikely(nfslocaliod_workqueue == NULL)) {
2534 		nfsiod_stop();
2535 		return -ENOMEM;
2536 	}
2537 #endif /* CONFIG_NFS_LOCALIO */
2538 	return 0;
2539 }
2540 
2541 unsigned int nfs_net_id;
2542 EXPORT_SYMBOL_GPL(nfs_net_id);
2543 
2544 static int nfs_net_init(struct net *net)
2545 {
2546 	struct nfs_net *nn = net_generic(net, nfs_net_id);
2547 
2548 	nfs_clients_init(net);
2549 
2550 	if (!rpc_proc_register(net, &nn->rpcstats)) {
2551 		nfs_clients_exit(net);
2552 		return -ENOMEM;
2553 	}
2554 
2555 	return nfs_fs_proc_net_init(net);
2556 }
2557 
2558 static void nfs_net_exit(struct net *net)
2559 {
2560 	rpc_proc_unregister(net, "nfs");
2561 	nfs_fs_proc_net_exit(net);
2562 	nfs_clients_exit(net);
2563 }
2564 
2565 static struct pernet_operations nfs_net_ops = {
2566 	.init = nfs_net_init,
2567 	.exit = nfs_net_exit,
2568 	.id   = &nfs_net_id,
2569 	.size = sizeof(struct nfs_net),
2570 };
2571 
2572 /*
2573  * Initialize NFS
2574  */
2575 static int __init init_nfs_fs(void)
2576 {
2577 	int err;
2578 
2579 	err = nfs_sysfs_init();
2580 	if (err < 0)
2581 		goto out10;
2582 
2583 	err = register_pernet_subsys(&nfs_net_ops);
2584 	if (err < 0)
2585 		goto out9;
2586 
2587 	err = nfsiod_start();
2588 	if (err)
2589 		goto out7;
2590 
2591 	err = nfs_fs_proc_init();
2592 	if (err)
2593 		goto out6;
2594 
2595 	err = nfs_init_nfspagecache();
2596 	if (err)
2597 		goto out5;
2598 
2599 	err = nfs_init_inodecache();
2600 	if (err)
2601 		goto out4;
2602 
2603 	err = nfs_init_readpagecache();
2604 	if (err)
2605 		goto out3;
2606 
2607 	err = nfs_init_writepagecache();
2608 	if (err)
2609 		goto out2;
2610 
2611 	err = nfs_init_directcache();
2612 	if (err)
2613 		goto out1;
2614 
2615 	err = register_nfs_fs();
2616 	if (err)
2617 		goto out0;
2618 
2619 	return 0;
2620 out0:
2621 	nfs_destroy_directcache();
2622 out1:
2623 	nfs_destroy_writepagecache();
2624 out2:
2625 	nfs_destroy_readpagecache();
2626 out3:
2627 	nfs_destroy_inodecache();
2628 out4:
2629 	nfs_destroy_nfspagecache();
2630 out5:
2631 	nfs_fs_proc_exit();
2632 out6:
2633 	nfsiod_stop();
2634 out7:
2635 	unregister_pernet_subsys(&nfs_net_ops);
2636 out9:
2637 	nfs_sysfs_exit();
2638 out10:
2639 	return err;
2640 }
2641 
2642 static void __exit exit_nfs_fs(void)
2643 {
2644 	nfs_destroy_directcache();
2645 	nfs_destroy_writepagecache();
2646 	nfs_destroy_readpagecache();
2647 	nfs_destroy_inodecache();
2648 	nfs_destroy_nfspagecache();
2649 	unregister_pernet_subsys(&nfs_net_ops);
2650 	unregister_nfs_fs();
2651 	nfs_fs_proc_exit();
2652 	nfsiod_stop();
2653 	nfs_sysfs_exit();
2654 }
2655 
2656 /* Not quite true; I just maintain it */
2657 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2658 MODULE_DESCRIPTION("NFS client support");
2659 MODULE_LICENSE("GPL");
2660 module_param(enable_ino64, bool, 0644);
2661 
2662 module_init(init_nfs_fs)
2663 module_exit(exit_nfs_fs)
2664