xref: /openbmc/linux/fs/nfs/inode.c (revision 3745488e)
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 void nfs_invalidate_inode(struct inode *);
66 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
67 
68 static struct kmem_cache * nfs_inode_cachep;
69 
70 static inline unsigned long
71 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
72 {
73 	return nfs_fileid_to_ino_t(fattr->fileid);
74 }
75 
76 static int nfs_wait_killable(int mode)
77 {
78 	freezable_schedule_unsafe();
79 	if (signal_pending_state(mode, current))
80 		return -ERESTARTSYS;
81 	return 0;
82 }
83 
84 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
85 {
86 	return nfs_wait_killable(mode);
87 }
88 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
89 
90 /**
91  * nfs_compat_user_ino64 - returns the user-visible inode number
92  * @fileid: 64-bit fileid
93  *
94  * This function returns a 32-bit inode number if the boot parameter
95  * nfs.enable_ino64 is zero.
96  */
97 u64 nfs_compat_user_ino64(u64 fileid)
98 {
99 #ifdef CONFIG_COMPAT
100 	compat_ulong_t ino;
101 #else
102 	unsigned long ino;
103 #endif
104 
105 	if (enable_ino64)
106 		return fileid;
107 	ino = fileid;
108 	if (sizeof(ino) < sizeof(fileid))
109 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
110 	return ino;
111 }
112 
113 int nfs_drop_inode(struct inode *inode)
114 {
115 	return NFS_STALE(inode) || generic_drop_inode(inode);
116 }
117 EXPORT_SYMBOL_GPL(nfs_drop_inode);
118 
119 void nfs_clear_inode(struct inode *inode)
120 {
121 	/*
122 	 * The following should never happen...
123 	 */
124 	WARN_ON_ONCE(nfs_have_writebacks(inode));
125 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
126 	nfs_zap_acl_cache(inode);
127 	nfs_access_zap_cache(inode);
128 	nfs_fscache_clear_inode(inode);
129 }
130 EXPORT_SYMBOL_GPL(nfs_clear_inode);
131 
132 void nfs_evict_inode(struct inode *inode)
133 {
134 	truncate_inode_pages_final(&inode->i_data);
135 	clear_inode(inode);
136 	nfs_clear_inode(inode);
137 }
138 
139 int nfs_sync_inode(struct inode *inode)
140 {
141 	inode_dio_wait(inode);
142 	return nfs_wb_all(inode);
143 }
144 EXPORT_SYMBOL_GPL(nfs_sync_inode);
145 
146 /**
147  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
148  * @mapping: pointer to struct address_space
149  */
150 int nfs_sync_mapping(struct address_space *mapping)
151 {
152 	int ret = 0;
153 
154 	if (mapping->nrpages != 0) {
155 		unmap_mapping_range(mapping, 0, 0, 0);
156 		ret = nfs_wb_all(mapping->host);
157 	}
158 	return ret;
159 }
160 
161 static int nfs_attribute_timeout(struct inode *inode)
162 {
163 	struct nfs_inode *nfsi = NFS_I(inode);
164 
165 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
166 }
167 
168 static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
169 {
170 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
171 
172 	/* Special case for the pagecache or access cache */
173 	if (flags == NFS_INO_REVAL_PAGECACHE &&
174 	    !(cache_validity & NFS_INO_REVAL_FORCED))
175 		return false;
176 	return (cache_validity & flags) != 0;
177 }
178 
179 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
180 {
181 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
182 
183 	if ((cache_validity & flags) != 0)
184 		return true;
185 	if (nfs_attribute_timeout(inode))
186 		return true;
187 	return false;
188 }
189 
190 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
191 {
192 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
193 		return nfs_check_cache_invalid_delegated(inode, flags);
194 
195 	return nfs_check_cache_invalid_not_delegated(inode, flags);
196 }
197 
198 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
199 {
200 	struct nfs_inode *nfsi = NFS_I(inode);
201 	bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
202 
203 	if (have_delegation) {
204 		if (!(flags & NFS_INO_REVAL_FORCED))
205 			flags &= ~NFS_INO_INVALID_OTHER;
206 		flags &= ~(NFS_INO_INVALID_CHANGE
207 				| NFS_INO_INVALID_SIZE
208 				| NFS_INO_REVAL_PAGECACHE);
209 	}
210 
211 	if (inode->i_mapping->nrpages == 0)
212 		flags &= ~(NFS_INO_INVALID_DATA|NFS_INO_DATA_INVAL_DEFER);
213 	nfsi->cache_validity |= flags;
214 	if (flags & NFS_INO_INVALID_DATA)
215 		nfs_fscache_invalidate(inode);
216 }
217 
218 /*
219  * Invalidate the local caches
220  */
221 static void nfs_zap_caches_locked(struct inode *inode)
222 {
223 	struct nfs_inode *nfsi = NFS_I(inode);
224 	int mode = inode->i_mode;
225 
226 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
227 
228 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
229 	nfsi->attrtimeo_timestamp = jiffies;
230 
231 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
232 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
233 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
234 					| NFS_INO_INVALID_DATA
235 					| NFS_INO_INVALID_ACCESS
236 					| NFS_INO_INVALID_ACL
237 					| NFS_INO_REVAL_PAGECACHE);
238 	} else
239 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
240 					| NFS_INO_INVALID_ACCESS
241 					| NFS_INO_INVALID_ACL
242 					| NFS_INO_REVAL_PAGECACHE);
243 	nfs_zap_label_cache_locked(nfsi);
244 }
245 
246 void nfs_zap_caches(struct inode *inode)
247 {
248 	spin_lock(&inode->i_lock);
249 	nfs_zap_caches_locked(inode);
250 	spin_unlock(&inode->i_lock);
251 }
252 
253 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
254 {
255 	if (mapping->nrpages != 0) {
256 		spin_lock(&inode->i_lock);
257 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
258 		spin_unlock(&inode->i_lock);
259 	}
260 }
261 
262 void nfs_zap_acl_cache(struct inode *inode)
263 {
264 	void (*clear_acl_cache)(struct inode *);
265 
266 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
267 	if (clear_acl_cache != NULL)
268 		clear_acl_cache(inode);
269 	spin_lock(&inode->i_lock);
270 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
271 	spin_unlock(&inode->i_lock);
272 }
273 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
274 
275 void nfs_invalidate_atime(struct inode *inode)
276 {
277 	spin_lock(&inode->i_lock);
278 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
279 	spin_unlock(&inode->i_lock);
280 }
281 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
282 
283 /*
284  * Invalidate, but do not unhash, the inode.
285  * NB: must be called with inode->i_lock held!
286  */
287 static void nfs_invalidate_inode(struct inode *inode)
288 {
289 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
290 	nfs_zap_caches_locked(inode);
291 }
292 
293 struct nfs_find_desc {
294 	struct nfs_fh		*fh;
295 	struct nfs_fattr	*fattr;
296 };
297 
298 /*
299  * In NFSv3 we can have 64bit inode numbers. In order to support
300  * this, and re-exported directories (also seen in NFSv2)
301  * we are forced to allow 2 different inodes to have the same
302  * i_ino.
303  */
304 static int
305 nfs_find_actor(struct inode *inode, void *opaque)
306 {
307 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
308 	struct nfs_fh		*fh = desc->fh;
309 	struct nfs_fattr	*fattr = desc->fattr;
310 
311 	if (NFS_FILEID(inode) != fattr->fileid)
312 		return 0;
313 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
314 		return 0;
315 	if (nfs_compare_fh(NFS_FH(inode), fh))
316 		return 0;
317 	if (is_bad_inode(inode) || NFS_STALE(inode))
318 		return 0;
319 	return 1;
320 }
321 
322 static int
323 nfs_init_locked(struct inode *inode, void *opaque)
324 {
325 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
326 	struct nfs_fattr	*fattr = desc->fattr;
327 
328 	set_nfs_fileid(inode, fattr->fileid);
329 	inode->i_mode = fattr->mode;
330 	nfs_copy_fh(NFS_FH(inode), desc->fh);
331 	return 0;
332 }
333 
334 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
335 static void nfs_clear_label_invalid(struct inode *inode)
336 {
337 	spin_lock(&inode->i_lock);
338 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
339 	spin_unlock(&inode->i_lock);
340 }
341 
342 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
343 					struct nfs4_label *label)
344 {
345 	int error;
346 
347 	if (label == NULL)
348 		return;
349 
350 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
351 		error = security_inode_notifysecctx(inode, label->label,
352 				label->len);
353 		if (error)
354 			printk(KERN_ERR "%s() %s %d "
355 					"security_inode_notifysecctx() %d\n",
356 					__func__,
357 					(char *)label->label,
358 					label->len, error);
359 		nfs_clear_label_invalid(inode);
360 	}
361 }
362 
363 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
364 {
365 	struct nfs4_label *label = NULL;
366 	int minor_version = server->nfs_client->cl_minorversion;
367 
368 	if (minor_version < 2)
369 		return label;
370 
371 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
372 		return label;
373 
374 	label = kzalloc(sizeof(struct nfs4_label), flags);
375 	if (label == NULL)
376 		return ERR_PTR(-ENOMEM);
377 
378 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
379 	if (label->label == NULL) {
380 		kfree(label);
381 		return ERR_PTR(-ENOMEM);
382 	}
383 	label->len = NFS4_MAXLABELLEN;
384 
385 	return label;
386 }
387 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
388 #else
389 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
390 					struct nfs4_label *label)
391 {
392 }
393 #endif
394 EXPORT_SYMBOL_GPL(nfs_setsecurity);
395 
396 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
397 struct inode *
398 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
399 {
400 	struct nfs_find_desc desc = {
401 		.fh	= fh,
402 		.fattr	= fattr,
403 	};
404 	struct inode *inode;
405 	unsigned long hash;
406 
407 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
408 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
409 		return NULL;
410 
411 	hash = nfs_fattr_to_ino_t(fattr);
412 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
413 
414 	dprintk("%s: returning %p\n", __func__, inode);
415 	return inode;
416 }
417 
418 /*
419  * This is our front-end to iget that looks up inodes by file handle
420  * instead of inode number.
421  */
422 struct inode *
423 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
424 {
425 	struct nfs_find_desc desc = {
426 		.fh	= fh,
427 		.fattr	= fattr
428 	};
429 	struct inode *inode = ERR_PTR(-ENOENT);
430 	unsigned long hash;
431 
432 	nfs_attr_check_mountpoint(sb, fattr);
433 
434 	if (nfs_attr_use_mounted_on_fileid(fattr))
435 		fattr->fileid = fattr->mounted_on_fileid;
436 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
437 		goto out_no_inode;
438 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
439 		goto out_no_inode;
440 
441 	hash = nfs_fattr_to_ino_t(fattr);
442 
443 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
444 	if (inode == NULL) {
445 		inode = ERR_PTR(-ENOMEM);
446 		goto out_no_inode;
447 	}
448 
449 	if (inode->i_state & I_NEW) {
450 		struct nfs_inode *nfsi = NFS_I(inode);
451 		unsigned long now = jiffies;
452 
453 		/* We set i_ino for the few things that still rely on it,
454 		 * such as stat(2) */
455 		inode->i_ino = hash;
456 
457 		/* We can't support update_atime(), since the server will reset it */
458 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
459 		inode->i_mode = fattr->mode;
460 		nfsi->cache_validity = 0;
461 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
462 				&& nfs_server_capable(inode, NFS_CAP_MODE))
463 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
464 		/* Why so? Because we want revalidate for devices/FIFOs, and
465 		 * that's precisely what we have in nfs_file_inode_operations.
466 		 */
467 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
468 		if (S_ISREG(inode->i_mode)) {
469 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
470 			inode->i_data.a_ops = &nfs_file_aops;
471 		} else if (S_ISDIR(inode->i_mode)) {
472 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
473 			inode->i_fop = &nfs_dir_operations;
474 			inode->i_data.a_ops = &nfs_dir_aops;
475 			/* Deal with crossing mountpoints */
476 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
477 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
478 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
479 					inode->i_op = &nfs_referral_inode_operations;
480 				else
481 					inode->i_op = &nfs_mountpoint_inode_operations;
482 				inode->i_fop = NULL;
483 				inode->i_flags |= S_AUTOMOUNT;
484 			}
485 		} else if (S_ISLNK(inode->i_mode)) {
486 			inode->i_op = &nfs_symlink_inode_operations;
487 			inode_nohighmem(inode);
488 		} else
489 			init_special_inode(inode, inode->i_mode, fattr->rdev);
490 
491 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
492 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
493 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
494 		inode_set_iversion_raw(inode, 0);
495 		inode->i_size = 0;
496 		clear_nlink(inode);
497 		inode->i_uid = make_kuid(&init_user_ns, -2);
498 		inode->i_gid = make_kgid(&init_user_ns, -2);
499 		inode->i_blocks = 0;
500 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
501 		nfsi->write_io = 0;
502 		nfsi->read_io = 0;
503 
504 		nfsi->read_cache_jiffies = fattr->time_start;
505 		nfsi->attr_gencount = fattr->gencount;
506 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
507 			inode->i_atime = fattr->atime;
508 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
509 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
510 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
511 			inode->i_mtime = fattr->mtime;
512 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
513 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
514 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
515 			inode->i_ctime = fattr->ctime;
516 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
517 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
518 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
519 			inode_set_iversion_raw(inode, fattr->change_attr);
520 		else
521 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
522 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
523 			inode->i_size = nfs_size_to_loff_t(fattr->size);
524 		else
525 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
526 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
527 			set_nlink(inode, fattr->nlink);
528 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
529 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
530 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
531 			inode->i_uid = fattr->uid;
532 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
533 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
534 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
535 			inode->i_gid = fattr->gid;
536 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
537 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
538 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
539 			inode->i_blocks = fattr->du.nfs2.blocks;
540 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
541 			/*
542 			 * report the blocks in 512byte units
543 			 */
544 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
545 		}
546 
547 		if (nfsi->cache_validity != 0)
548 			nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
549 
550 		nfs_setsecurity(inode, fattr, label);
551 
552 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
553 		nfsi->attrtimeo_timestamp = now;
554 		nfsi->access_cache = RB_ROOT;
555 
556 		nfs_fscache_init_inode(inode);
557 
558 		unlock_new_inode(inode);
559 	} else {
560 		int err = nfs_refresh_inode(inode, fattr);
561 		if (err < 0) {
562 			iput(inode);
563 			inode = ERR_PTR(err);
564 			goto out_no_inode;
565 		}
566 	}
567 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
568 		inode->i_sb->s_id,
569 		(unsigned long long)NFS_FILEID(inode),
570 		nfs_display_fhandle_hash(fh),
571 		atomic_read(&inode->i_count));
572 
573 out:
574 	return inode;
575 
576 out_no_inode:
577 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
578 	goto out;
579 }
580 EXPORT_SYMBOL_GPL(nfs_fhget);
581 
582 #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)
583 
584 int
585 nfs_setattr(struct dentry *dentry, struct iattr *attr)
586 {
587 	struct inode *inode = d_inode(dentry);
588 	struct nfs_fattr *fattr;
589 	int error = 0;
590 
591 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
592 
593 	/* skip mode change if it's just for clearing setuid/setgid */
594 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
595 		attr->ia_valid &= ~ATTR_MODE;
596 
597 	if (attr->ia_valid & ATTR_SIZE) {
598 		BUG_ON(!S_ISREG(inode->i_mode));
599 
600 		error = inode_newsize_ok(inode, attr->ia_size);
601 		if (error)
602 			return error;
603 
604 		if (attr->ia_size == i_size_read(inode))
605 			attr->ia_valid &= ~ATTR_SIZE;
606 	}
607 
608 	/* Optimization: if the end result is no change, don't RPC */
609 	attr->ia_valid &= NFS_VALID_ATTRS;
610 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
611 		return 0;
612 
613 	trace_nfs_setattr_enter(inode);
614 
615 	/* Write all dirty data */
616 	if (S_ISREG(inode->i_mode))
617 		nfs_sync_inode(inode);
618 
619 	fattr = nfs_alloc_fattr();
620 	if (fattr == NULL) {
621 		error = -ENOMEM;
622 		goto out;
623 	}
624 
625 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
626 	if (error == 0)
627 		error = nfs_refresh_inode(inode, fattr);
628 	nfs_free_fattr(fattr);
629 out:
630 	trace_nfs_setattr_exit(inode, error);
631 	return error;
632 }
633 EXPORT_SYMBOL_GPL(nfs_setattr);
634 
635 /**
636  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
637  * @inode: inode of the file used
638  * @offset: file offset to start truncating
639  *
640  * This is a copy of the common vmtruncate, but with the locking
641  * corrected to take into account the fact that NFS requires
642  * inode->i_size to be updated under the inode->i_lock.
643  * Note: must be called with inode->i_lock held!
644  */
645 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
646 {
647 	int err;
648 
649 	err = inode_newsize_ok(inode, offset);
650 	if (err)
651 		goto out;
652 
653 	i_size_write(inode, offset);
654 	/* Optimisation */
655 	if (offset == 0)
656 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
657 				NFS_INO_DATA_INVAL_DEFER);
658 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
659 
660 	spin_unlock(&inode->i_lock);
661 	truncate_pagecache(inode, offset);
662 	spin_lock(&inode->i_lock);
663 out:
664 	return err;
665 }
666 
667 /**
668  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
669  * @inode: pointer to struct inode
670  * @attr: pointer to struct iattr
671  * @fattr: pointer to struct nfs_fattr
672  *
673  * Note: we do this in the *proc.c in order to ensure that
674  *       it works for things like exclusive creates too.
675  */
676 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
677 		struct nfs_fattr *fattr)
678 {
679 	/* Barrier: bump the attribute generation count. */
680 	nfs_fattr_set_barrier(fattr);
681 
682 	spin_lock(&inode->i_lock);
683 	NFS_I(inode)->attr_gencount = fattr->gencount;
684 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
685 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
686 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
687 		nfs_vmtruncate(inode, attr->ia_size);
688 	}
689 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
690 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
691 		if ((attr->ia_valid & ATTR_MODE) != 0) {
692 			int mode = attr->ia_mode & S_IALLUGO;
693 			mode |= inode->i_mode & ~S_IALLUGO;
694 			inode->i_mode = mode;
695 		}
696 		if ((attr->ia_valid & ATTR_UID) != 0)
697 			inode->i_uid = attr->ia_uid;
698 		if ((attr->ia_valid & ATTR_GID) != 0)
699 			inode->i_gid = attr->ia_gid;
700 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
701 			inode->i_ctime = fattr->ctime;
702 		else
703 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
704 					| NFS_INO_INVALID_CTIME);
705 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
706 				| NFS_INO_INVALID_ACL);
707 	}
708 	if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
709 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
710 				| NFS_INO_INVALID_CTIME);
711 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
712 			inode->i_atime = fattr->atime;
713 		else if (attr->ia_valid & ATTR_ATIME_SET)
714 			inode->i_atime = attr->ia_atime;
715 		else
716 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
717 
718 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
719 			inode->i_ctime = fattr->ctime;
720 		else
721 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
722 					| NFS_INO_INVALID_CTIME);
723 	}
724 	if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
725 		NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
726 				| NFS_INO_INVALID_CTIME);
727 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
728 			inode->i_mtime = fattr->mtime;
729 		else if (attr->ia_valid & ATTR_MTIME_SET)
730 			inode->i_mtime = attr->ia_mtime;
731 		else
732 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
733 
734 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
735 			inode->i_ctime = fattr->ctime;
736 		else
737 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
738 					| NFS_INO_INVALID_CTIME);
739 	}
740 	if (fattr->valid)
741 		nfs_update_inode(inode, fattr);
742 	spin_unlock(&inode->i_lock);
743 }
744 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
745 
746 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
747 {
748 	struct dentry *parent;
749 
750 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
751 		return;
752 	parent = dget_parent(dentry);
753 	nfs_force_use_readdirplus(d_inode(parent));
754 	dput(parent);
755 }
756 
757 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
758 {
759 	struct dentry *parent;
760 
761 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
762 		return;
763 	parent = dget_parent(dentry);
764 	nfs_advise_use_readdirplus(d_inode(parent));
765 	dput(parent);
766 }
767 
768 static bool nfs_need_revalidate_inode(struct inode *inode)
769 {
770 	if (NFS_I(inode)->cache_validity &
771 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
772 		return true;
773 	if (nfs_attribute_cache_expired(inode))
774 		return true;
775 	return false;
776 }
777 
778 int nfs_getattr(const struct path *path, struct kstat *stat,
779 		u32 request_mask, unsigned int query_flags)
780 {
781 	struct inode *inode = d_inode(path->dentry);
782 	struct nfs_server *server = NFS_SERVER(inode);
783 	unsigned long cache_validity;
784 	int err = 0;
785 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
786 	bool do_update = false;
787 
788 	trace_nfs_getattr_enter(inode);
789 
790 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync)
791 		goto out_no_update;
792 
793 	/* Flush out writes to the server in order to update c/mtime.  */
794 	if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
795 			S_ISREG(inode->i_mode)) {
796 		err = filemap_write_and_wait(inode->i_mapping);
797 		if (err)
798 			goto out;
799 	}
800 
801 	/*
802 	 * We may force a getattr if the user cares about atime.
803 	 *
804 	 * Note that we only have to check the vfsmount flags here:
805 	 *  - NFS always sets S_NOATIME by so checking it would give a
806 	 *    bogus result
807 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
808 	 *    no point in checking those.
809 	 */
810 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
811 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
812 		request_mask &= ~STATX_ATIME;
813 
814 	/* Is the user requesting attributes that might need revalidation? */
815 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
816 					STATX_MTIME|STATX_UID|STATX_GID|
817 					STATX_SIZE|STATX_BLOCKS)))
818 		goto out_no_revalidate;
819 
820 	/* Check whether the cached attributes are stale */
821 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
822 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
823 	do_update |= cache_validity &
824 		(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
825 	if (request_mask & STATX_ATIME)
826 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
827 	if (request_mask & (STATX_CTIME|STATX_MTIME))
828 		do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
829 	if (do_update) {
830 		/* Update the attribute cache */
831 		if (!(server->flags & NFS_MOUNT_NOAC))
832 			nfs_readdirplus_parent_cache_miss(path->dentry);
833 		else
834 			nfs_readdirplus_parent_cache_hit(path->dentry);
835 		err = __nfs_revalidate_inode(server, inode);
836 		if (err)
837 			goto out;
838 	} else
839 		nfs_readdirplus_parent_cache_hit(path->dentry);
840 out_no_revalidate:
841 	/* Only return attributes that were revalidated. */
842 	stat->result_mask &= request_mask;
843 out_no_update:
844 	generic_fillattr(inode, stat);
845 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
846 	if (S_ISDIR(inode->i_mode))
847 		stat->blksize = NFS_SERVER(inode)->dtsize;
848 out:
849 	trace_nfs_getattr_exit(inode, err);
850 	return err;
851 }
852 EXPORT_SYMBOL_GPL(nfs_getattr);
853 
854 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
855 {
856 	refcount_set(&l_ctx->count, 1);
857 	l_ctx->lockowner = current->files;
858 	INIT_LIST_HEAD(&l_ctx->list);
859 	atomic_set(&l_ctx->io_count, 0);
860 }
861 
862 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
863 {
864 	struct nfs_lock_context *pos;
865 
866 	list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
867 		if (pos->lockowner != current->files)
868 			continue;
869 		if (refcount_inc_not_zero(&pos->count))
870 			return pos;
871 	}
872 	return NULL;
873 }
874 
875 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
876 {
877 	struct nfs_lock_context *res, *new = NULL;
878 	struct inode *inode = d_inode(ctx->dentry);
879 
880 	rcu_read_lock();
881 	res = __nfs_find_lock_context(ctx);
882 	rcu_read_unlock();
883 	if (res == NULL) {
884 		new = kmalloc(sizeof(*new), GFP_KERNEL);
885 		if (new == NULL)
886 			return ERR_PTR(-ENOMEM);
887 		nfs_init_lock_context(new);
888 		spin_lock(&inode->i_lock);
889 		res = __nfs_find_lock_context(ctx);
890 		if (res == NULL) {
891 			new->open_context = get_nfs_open_context(ctx);
892 			if (new->open_context) {
893 				list_add_tail_rcu(&new->list,
894 						&ctx->lock_context.list);
895 				res = new;
896 				new = NULL;
897 			} else
898 				res = ERR_PTR(-EBADF);
899 		}
900 		spin_unlock(&inode->i_lock);
901 		kfree(new);
902 	}
903 	return res;
904 }
905 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
906 
907 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
908 {
909 	struct nfs_open_context *ctx = l_ctx->open_context;
910 	struct inode *inode = d_inode(ctx->dentry);
911 
912 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
913 		return;
914 	list_del_rcu(&l_ctx->list);
915 	spin_unlock(&inode->i_lock);
916 	put_nfs_open_context(ctx);
917 	kfree_rcu(l_ctx, rcu_head);
918 }
919 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
920 
921 /**
922  * nfs_close_context - Common close_context() routine NFSv2/v3
923  * @ctx: pointer to context
924  * @is_sync: is this a synchronous close
925  *
926  * Ensure that the attributes are up to date if we're mounted
927  * with close-to-open semantics and we have cached data that will
928  * need to be revalidated on open.
929  */
930 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
931 {
932 	struct nfs_inode *nfsi;
933 	struct inode *inode;
934 	struct nfs_server *server;
935 
936 	if (!(ctx->mode & FMODE_WRITE))
937 		return;
938 	if (!is_sync)
939 		return;
940 	inode = d_inode(ctx->dentry);
941 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
942 		return;
943 	nfsi = NFS_I(inode);
944 	if (inode->i_mapping->nrpages == 0)
945 		return;
946 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
947 		return;
948 	if (!list_empty(&nfsi->open_files))
949 		return;
950 	server = NFS_SERVER(inode);
951 	if (server->flags & NFS_MOUNT_NOCTO)
952 		return;
953 	nfs_revalidate_inode(server, inode);
954 }
955 EXPORT_SYMBOL_GPL(nfs_close_context);
956 
957 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
958 						fmode_t f_mode,
959 						struct file *filp)
960 {
961 	struct nfs_open_context *ctx;
962 	const struct cred *cred = get_current_cred();
963 
964 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
965 	if (!ctx) {
966 		put_cred(cred);
967 		return ERR_PTR(-ENOMEM);
968 	}
969 	nfs_sb_active(dentry->d_sb);
970 	ctx->dentry = dget(dentry);
971 	ctx->cred = cred;
972 	ctx->ll_cred = NULL;
973 	ctx->state = NULL;
974 	ctx->mode = f_mode;
975 	ctx->flags = 0;
976 	ctx->error = 0;
977 	ctx->flock_owner = (fl_owner_t)filp;
978 	nfs_init_lock_context(&ctx->lock_context);
979 	ctx->lock_context.open_context = ctx;
980 	INIT_LIST_HEAD(&ctx->list);
981 	ctx->mdsthreshold = NULL;
982 	return ctx;
983 }
984 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
985 
986 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
987 {
988 	if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
989 		return ctx;
990 	return NULL;
991 }
992 EXPORT_SYMBOL_GPL(get_nfs_open_context);
993 
994 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
995 {
996 	struct inode *inode = d_inode(ctx->dentry);
997 	struct super_block *sb = ctx->dentry->d_sb;
998 
999 	if (!refcount_dec_and_test(&ctx->lock_context.count))
1000 		return;
1001 	if (!list_empty(&ctx->list)) {
1002 		spin_lock(&inode->i_lock);
1003 		list_del_rcu(&ctx->list);
1004 		spin_unlock(&inode->i_lock);
1005 	}
1006 	if (inode != NULL)
1007 		NFS_PROTO(inode)->close_context(ctx, is_sync);
1008 	put_cred(ctx->cred);
1009 	dput(ctx->dentry);
1010 	nfs_sb_deactive(sb);
1011 	put_rpccred(ctx->ll_cred);
1012 	kfree(ctx->mdsthreshold);
1013 	kfree_rcu(ctx, rcu_head);
1014 }
1015 
1016 void put_nfs_open_context(struct nfs_open_context *ctx)
1017 {
1018 	__put_nfs_open_context(ctx, 0);
1019 }
1020 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1021 
1022 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1023 {
1024 	__put_nfs_open_context(ctx, 1);
1025 }
1026 
1027 /*
1028  * Ensure that mmap has a recent RPC credential for use when writing out
1029  * shared pages
1030  */
1031 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1032 {
1033 	struct inode *inode = d_inode(ctx->dentry);
1034 	struct nfs_inode *nfsi = NFS_I(inode);
1035 
1036 	spin_lock(&inode->i_lock);
1037 	if (list_empty(&nfsi->open_files) &&
1038 	    (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1039 		nfsi->cache_validity |= NFS_INO_INVALID_DATA |
1040 			NFS_INO_REVAL_FORCED;
1041 	list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1042 	spin_unlock(&inode->i_lock);
1043 }
1044 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1045 
1046 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1047 {
1048 	filp->private_data = get_nfs_open_context(ctx);
1049 	if (list_empty(&ctx->list))
1050 		nfs_inode_attach_open_context(ctx);
1051 }
1052 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1053 
1054 /*
1055  * Given an inode, search for an open context with the desired characteristics
1056  */
1057 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1058 {
1059 	struct nfs_inode *nfsi = NFS_I(inode);
1060 	struct nfs_open_context *pos, *ctx = NULL;
1061 
1062 	rcu_read_lock();
1063 	list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1064 		if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1065 			continue;
1066 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1067 			continue;
1068 		ctx = get_nfs_open_context(pos);
1069 		if (ctx)
1070 			break;
1071 	}
1072 	rcu_read_unlock();
1073 	return ctx;
1074 }
1075 
1076 void nfs_file_clear_open_context(struct file *filp)
1077 {
1078 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1079 
1080 	if (ctx) {
1081 		struct inode *inode = d_inode(ctx->dentry);
1082 
1083 		/*
1084 		 * We fatal error on write before. Try to writeback
1085 		 * every page again.
1086 		 */
1087 		if (ctx->error < 0)
1088 			invalidate_inode_pages2(inode->i_mapping);
1089 		filp->private_data = NULL;
1090 		put_nfs_open_context_sync(ctx);
1091 	}
1092 }
1093 
1094 /*
1095  * These allocate and release file read/write context information.
1096  */
1097 int nfs_open(struct inode *inode, struct file *filp)
1098 {
1099 	struct nfs_open_context *ctx;
1100 
1101 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1102 	if (IS_ERR(ctx))
1103 		return PTR_ERR(ctx);
1104 	nfs_file_set_open_context(filp, ctx);
1105 	put_nfs_open_context(ctx);
1106 	nfs_fscache_open_file(inode, filp);
1107 	return 0;
1108 }
1109 EXPORT_SYMBOL_GPL(nfs_open);
1110 
1111 /*
1112  * This function is called whenever some part of NFS notices that
1113  * the cached attributes have to be refreshed.
1114  */
1115 int
1116 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1117 {
1118 	int		 status = -ESTALE;
1119 	struct nfs4_label *label = NULL;
1120 	struct nfs_fattr *fattr = NULL;
1121 	struct nfs_inode *nfsi = NFS_I(inode);
1122 
1123 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1124 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1125 
1126 	trace_nfs_revalidate_inode_enter(inode);
1127 
1128 	if (is_bad_inode(inode))
1129 		goto out;
1130 	if (NFS_STALE(inode))
1131 		goto out;
1132 
1133 	/* pNFS: Attributes aren't updated until we layoutcommit */
1134 	if (S_ISREG(inode->i_mode)) {
1135 		status = pnfs_sync_inode(inode, false);
1136 		if (status)
1137 			goto out;
1138 	}
1139 
1140 	status = -ENOMEM;
1141 	fattr = nfs_alloc_fattr();
1142 	if (fattr == NULL)
1143 		goto out;
1144 
1145 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1146 
1147 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1148 	if (IS_ERR(label)) {
1149 		status = PTR_ERR(label);
1150 		goto out;
1151 	}
1152 
1153 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1154 			label, inode);
1155 	if (status != 0) {
1156 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1157 			 inode->i_sb->s_id,
1158 			 (unsigned long long)NFS_FILEID(inode), status);
1159 		switch (status) {
1160 		case -ETIMEDOUT:
1161 			/* A soft timeout occurred. Use cached information? */
1162 			if (server->flags & NFS_MOUNT_SOFTREVAL)
1163 				status = 0;
1164 			break;
1165 		case -ESTALE:
1166 			nfs_zap_caches(inode);
1167 			if (!S_ISDIR(inode->i_mode))
1168 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1169 		}
1170 		goto err_out;
1171 	}
1172 
1173 	status = nfs_refresh_inode(inode, fattr);
1174 	if (status) {
1175 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1176 			 inode->i_sb->s_id,
1177 			 (unsigned long long)NFS_FILEID(inode), status);
1178 		goto err_out;
1179 	}
1180 
1181 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1182 		nfs_zap_acl_cache(inode);
1183 
1184 	nfs_setsecurity(inode, fattr, label);
1185 
1186 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1187 		inode->i_sb->s_id,
1188 		(unsigned long long)NFS_FILEID(inode));
1189 
1190 err_out:
1191 	nfs4_label_free(label);
1192 out:
1193 	nfs_free_fattr(fattr);
1194 	trace_nfs_revalidate_inode_exit(inode, status);
1195 	return status;
1196 }
1197 
1198 int nfs_attribute_cache_expired(struct inode *inode)
1199 {
1200 	if (nfs_have_delegated_attributes(inode))
1201 		return 0;
1202 	return nfs_attribute_timeout(inode);
1203 }
1204 
1205 /**
1206  * nfs_revalidate_inode - Revalidate the inode attributes
1207  * @server: pointer to nfs_server struct
1208  * @inode: pointer to inode struct
1209  *
1210  * Updates inode attribute information by retrieving the data from the server.
1211  */
1212 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1213 {
1214 	if (!nfs_need_revalidate_inode(inode))
1215 		return NFS_STALE(inode) ? -ESTALE : 0;
1216 	return __nfs_revalidate_inode(server, inode);
1217 }
1218 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1219 
1220 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1221 {
1222 	struct nfs_inode *nfsi = NFS_I(inode);
1223 	int ret;
1224 
1225 	if (mapping->nrpages != 0) {
1226 		if (S_ISREG(inode->i_mode)) {
1227 			ret = nfs_sync_mapping(mapping);
1228 			if (ret < 0)
1229 				return ret;
1230 		}
1231 		ret = invalidate_inode_pages2(mapping);
1232 		if (ret < 0)
1233 			return ret;
1234 	}
1235 	if (S_ISDIR(inode->i_mode)) {
1236 		spin_lock(&inode->i_lock);
1237 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1238 		spin_unlock(&inode->i_lock);
1239 	}
1240 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1241 	nfs_fscache_wait_on_invalidate(inode);
1242 
1243 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1244 			inode->i_sb->s_id,
1245 			(unsigned long long)NFS_FILEID(inode));
1246 	return 0;
1247 }
1248 
1249 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1250 {
1251 	return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1252 		NFS_STALE(inode);
1253 }
1254 
1255 int nfs_revalidate_mapping_rcu(struct inode *inode)
1256 {
1257 	struct nfs_inode *nfsi = NFS_I(inode);
1258 	unsigned long *bitlock = &nfsi->flags;
1259 	int ret = 0;
1260 
1261 	if (IS_SWAPFILE(inode))
1262 		goto out;
1263 	if (nfs_mapping_need_revalidate_inode(inode)) {
1264 		ret = -ECHILD;
1265 		goto out;
1266 	}
1267 	spin_lock(&inode->i_lock);
1268 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1269 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1270 		ret = -ECHILD;
1271 	spin_unlock(&inode->i_lock);
1272 out:
1273 	return ret;
1274 }
1275 
1276 /**
1277  * nfs_revalidate_mapping - Revalidate the pagecache
1278  * @inode: pointer to host inode
1279  * @mapping: pointer to mapping
1280  */
1281 int nfs_revalidate_mapping(struct inode *inode,
1282 		struct address_space *mapping)
1283 {
1284 	struct nfs_inode *nfsi = NFS_I(inode);
1285 	unsigned long *bitlock = &nfsi->flags;
1286 	int ret = 0;
1287 
1288 	/* swapfiles are not supposed to be shared. */
1289 	if (IS_SWAPFILE(inode))
1290 		goto out;
1291 
1292 	if (nfs_mapping_need_revalidate_inode(inode)) {
1293 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1294 		if (ret < 0)
1295 			goto out;
1296 	}
1297 
1298 	/*
1299 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1300 	 * invalidations that come in while we're shooting down the mappings
1301 	 * are respected. But, that leaves a race window where one revalidator
1302 	 * can clear the flag, and then another checks it before the mapping
1303 	 * gets invalidated. Fix that by serializing access to this part of
1304 	 * the function.
1305 	 *
1306 	 * At the same time, we need to allow other tasks to see whether we
1307 	 * might be in the middle of invalidating the pages, so we only set
1308 	 * the bit lock here if it looks like we're going to be doing that.
1309 	 */
1310 	for (;;) {
1311 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1312 					 nfs_wait_bit_killable, TASK_KILLABLE);
1313 		if (ret)
1314 			goto out;
1315 		spin_lock(&inode->i_lock);
1316 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1317 			spin_unlock(&inode->i_lock);
1318 			continue;
1319 		}
1320 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1321 			break;
1322 		spin_unlock(&inode->i_lock);
1323 		goto out;
1324 	}
1325 
1326 	set_bit(NFS_INO_INVALIDATING, bitlock);
1327 	smp_wmb();
1328 	nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA|
1329 			NFS_INO_DATA_INVAL_DEFER);
1330 	spin_unlock(&inode->i_lock);
1331 	trace_nfs_invalidate_mapping_enter(inode);
1332 	ret = nfs_invalidate_mapping(inode, mapping);
1333 	trace_nfs_invalidate_mapping_exit(inode, ret);
1334 
1335 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1336 	smp_mb__after_atomic();
1337 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1338 out:
1339 	return ret;
1340 }
1341 
1342 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1343 {
1344 	struct inode *inode = &nfsi->vfs_inode;
1345 
1346 	if (!S_ISREG(inode->i_mode))
1347 		return false;
1348 	if (list_empty(&nfsi->open_files))
1349 		return false;
1350 	return inode_is_open_for_write(inode);
1351 }
1352 
1353 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1354 {
1355 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1356 }
1357 
1358 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1359 {
1360 	struct timespec64 ts;
1361 
1362 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1363 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1364 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1365 		inode_set_iversion_raw(inode, fattr->change_attr);
1366 		if (S_ISDIR(inode->i_mode))
1367 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1368 	}
1369 	/* If we have atomic WCC data, we may update some attributes */
1370 	ts = inode->i_ctime;
1371 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1372 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1373 			&& timespec64_equal(&ts, &fattr->pre_ctime)) {
1374 		inode->i_ctime = fattr->ctime;
1375 	}
1376 
1377 	ts = inode->i_mtime;
1378 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1379 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1380 			&& timespec64_equal(&ts, &fattr->pre_mtime)) {
1381 		inode->i_mtime = fattr->mtime;
1382 		if (S_ISDIR(inode->i_mode))
1383 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1384 	}
1385 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1386 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1387 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1388 			&& !nfs_have_writebacks(inode)) {
1389 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1390 	}
1391 }
1392 
1393 /**
1394  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1395  * @inode: pointer to inode
1396  * @fattr: updated attributes
1397  *
1398  * Verifies the attribute cache. If we have just changed the attributes,
1399  * so that fattr carries weak cache consistency data, then it may
1400  * also update the ctime/mtime/change_attribute.
1401  */
1402 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1403 {
1404 	struct nfs_inode *nfsi = NFS_I(inode);
1405 	loff_t cur_size, new_isize;
1406 	unsigned long invalid = 0;
1407 	struct timespec64 ts;
1408 
1409 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1410 		return 0;
1411 
1412 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1413 		/* Only a mounted-on-fileid? Just exit */
1414 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1415 			return 0;
1416 	/* Has the inode gone and changed behind our back? */
1417 	} else if (nfsi->fileid != fattr->fileid) {
1418 		/* Is this perhaps the mounted-on fileid? */
1419 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1420 		    nfsi->fileid == fattr->mounted_on_fileid)
1421 			return 0;
1422 		return -ESTALE;
1423 	}
1424 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1425 		return -ESTALE;
1426 
1427 
1428 	if (!nfs_file_has_buffered_writers(nfsi)) {
1429 		/* Verify a few of the more important attributes */
1430 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1431 			invalid |= NFS_INO_INVALID_CHANGE
1432 				| NFS_INO_REVAL_PAGECACHE;
1433 
1434 		ts = inode->i_mtime;
1435 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1436 			invalid |= NFS_INO_INVALID_MTIME;
1437 
1438 		ts = inode->i_ctime;
1439 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1440 			invalid |= NFS_INO_INVALID_CTIME;
1441 
1442 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1443 			cur_size = i_size_read(inode);
1444 			new_isize = nfs_size_to_loff_t(fattr->size);
1445 			if (cur_size != new_isize)
1446 				invalid |= NFS_INO_INVALID_SIZE
1447 					| NFS_INO_REVAL_PAGECACHE;
1448 		}
1449 	}
1450 
1451 	/* Have any file permissions changed? */
1452 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1453 		invalid |= NFS_INO_INVALID_ACCESS
1454 			| NFS_INO_INVALID_ACL
1455 			| NFS_INO_INVALID_OTHER;
1456 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1457 		invalid |= NFS_INO_INVALID_ACCESS
1458 			| NFS_INO_INVALID_ACL
1459 			| NFS_INO_INVALID_OTHER;
1460 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1461 		invalid |= NFS_INO_INVALID_ACCESS
1462 			| NFS_INO_INVALID_ACL
1463 			| NFS_INO_INVALID_OTHER;
1464 
1465 	/* Has the link count changed? */
1466 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1467 		invalid |= NFS_INO_INVALID_OTHER;
1468 
1469 	ts = inode->i_atime;
1470 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1471 		invalid |= NFS_INO_INVALID_ATIME;
1472 
1473 	if (invalid != 0)
1474 		nfs_set_cache_invalid(inode, invalid);
1475 
1476 	nfsi->read_cache_jiffies = fattr->time_start;
1477 	return 0;
1478 }
1479 
1480 static atomic_long_t nfs_attr_generation_counter;
1481 
1482 static unsigned long nfs_read_attr_generation_counter(void)
1483 {
1484 	return atomic_long_read(&nfs_attr_generation_counter);
1485 }
1486 
1487 unsigned long nfs_inc_attr_generation_counter(void)
1488 {
1489 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1490 }
1491 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1492 
1493 void nfs_fattr_init(struct nfs_fattr *fattr)
1494 {
1495 	fattr->valid = 0;
1496 	fattr->time_start = jiffies;
1497 	fattr->gencount = nfs_inc_attr_generation_counter();
1498 	fattr->owner_name = NULL;
1499 	fattr->group_name = NULL;
1500 }
1501 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1502 
1503 /**
1504  * nfs_fattr_set_barrier
1505  * @fattr: attributes
1506  *
1507  * Used to set a barrier after an attribute was updated. This
1508  * barrier ensures that older attributes from RPC calls that may
1509  * have raced with our update cannot clobber these new values.
1510  * Note that you are still responsible for ensuring that other
1511  * operations which change the attribute on the server do not
1512  * collide.
1513  */
1514 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1515 {
1516 	fattr->gencount = nfs_inc_attr_generation_counter();
1517 }
1518 
1519 struct nfs_fattr *nfs_alloc_fattr(void)
1520 {
1521 	struct nfs_fattr *fattr;
1522 
1523 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1524 	if (fattr != NULL)
1525 		nfs_fattr_init(fattr);
1526 	return fattr;
1527 }
1528 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1529 
1530 struct nfs_fh *nfs_alloc_fhandle(void)
1531 {
1532 	struct nfs_fh *fh;
1533 
1534 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1535 	if (fh != NULL)
1536 		fh->size = 0;
1537 	return fh;
1538 }
1539 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1540 
1541 #ifdef NFS_DEBUG
1542 /*
1543  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1544  *                             in the same way that wireshark does
1545  *
1546  * @fh: file handle
1547  *
1548  * For debugging only.
1549  */
1550 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1551 {
1552 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1553 	 * not on the result */
1554 	return nfs_fhandle_hash(fh);
1555 }
1556 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1557 
1558 /*
1559  * _nfs_display_fhandle - display an NFS file handle on the console
1560  *
1561  * @fh: file handle to display
1562  * @caption: display caption
1563  *
1564  * For debugging only.
1565  */
1566 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1567 {
1568 	unsigned short i;
1569 
1570 	if (fh == NULL || fh->size == 0) {
1571 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1572 		return;
1573 	}
1574 
1575 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1576 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1577 	for (i = 0; i < fh->size; i += 16) {
1578 		__be32 *pos = (__be32 *)&fh->data[i];
1579 
1580 		switch ((fh->size - i - 1) >> 2) {
1581 		case 0:
1582 			printk(KERN_DEFAULT " %08x\n",
1583 				be32_to_cpup(pos));
1584 			break;
1585 		case 1:
1586 			printk(KERN_DEFAULT " %08x %08x\n",
1587 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1588 			break;
1589 		case 2:
1590 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1591 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1592 				be32_to_cpup(pos + 2));
1593 			break;
1594 		default:
1595 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1596 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1597 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1598 		}
1599 	}
1600 }
1601 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1602 #endif
1603 
1604 /**
1605  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1606  * @inode: pointer to inode
1607  * @fattr: attributes
1608  *
1609  * Attempt to divine whether or not an RPC call reply carrying stale
1610  * attributes got scheduled after another call carrying updated ones.
1611  *
1612  * To do so, the function first assumes that a more recent ctime means
1613  * that the attributes in fattr are newer, however it also attempt to
1614  * catch the case where ctime either didn't change, or went backwards
1615  * (if someone reset the clock on the server) by looking at whether
1616  * or not this RPC call was started after the inode was last updated.
1617  * Note also the check for wraparound of 'attr_gencount'
1618  *
1619  * The function returns 'true' if it thinks the attributes in 'fattr' are
1620  * more recent than the ones cached in the inode.
1621  *
1622  */
1623 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1624 {
1625 	const struct nfs_inode *nfsi = NFS_I(inode);
1626 
1627 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1628 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1629 }
1630 
1631 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1632 {
1633 	int ret;
1634 
1635 	trace_nfs_refresh_inode_enter(inode);
1636 
1637 	if (nfs_inode_attrs_need_update(inode, fattr))
1638 		ret = nfs_update_inode(inode, fattr);
1639 	else
1640 		ret = nfs_check_inode_attributes(inode, fattr);
1641 
1642 	trace_nfs_refresh_inode_exit(inode, ret);
1643 	return ret;
1644 }
1645 
1646 /**
1647  * nfs_refresh_inode - try to update the inode attribute cache
1648  * @inode: pointer to inode
1649  * @fattr: updated attributes
1650  *
1651  * Check that an RPC call that returned attributes has not overlapped with
1652  * other recent updates of the inode metadata, then decide whether it is
1653  * safe to do a full update of the inode attributes, or whether just to
1654  * call nfs_check_inode_attributes.
1655  */
1656 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1657 {
1658 	int status;
1659 
1660 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1661 		return 0;
1662 	spin_lock(&inode->i_lock);
1663 	status = nfs_refresh_inode_locked(inode, fattr);
1664 	spin_unlock(&inode->i_lock);
1665 
1666 	return status;
1667 }
1668 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1669 
1670 static int nfs_post_op_update_inode_locked(struct inode *inode,
1671 		struct nfs_fattr *fattr, unsigned int invalid)
1672 {
1673 	if (S_ISDIR(inode->i_mode))
1674 		invalid |= NFS_INO_INVALID_DATA;
1675 	nfs_set_cache_invalid(inode, invalid);
1676 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1677 		return 0;
1678 	return nfs_refresh_inode_locked(inode, fattr);
1679 }
1680 
1681 /**
1682  * nfs_post_op_update_inode - try to update the inode attribute cache
1683  * @inode: pointer to inode
1684  * @fattr: updated attributes
1685  *
1686  * After an operation that has changed the inode metadata, mark the
1687  * attribute cache as being invalid, then try to update it.
1688  *
1689  * NB: if the server didn't return any post op attributes, this
1690  * function will force the retrieval of attributes before the next
1691  * NFS request.  Thus it should be used only for operations that
1692  * are expected to change one or more attributes, to avoid
1693  * unnecessary NFS requests and trips through nfs_update_inode().
1694  */
1695 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1696 {
1697 	int status;
1698 
1699 	spin_lock(&inode->i_lock);
1700 	nfs_fattr_set_barrier(fattr);
1701 	status = nfs_post_op_update_inode_locked(inode, fattr,
1702 			NFS_INO_INVALID_CHANGE
1703 			| NFS_INO_INVALID_CTIME
1704 			| NFS_INO_REVAL_FORCED);
1705 	spin_unlock(&inode->i_lock);
1706 
1707 	return status;
1708 }
1709 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1710 
1711 /**
1712  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1713  * @inode: pointer to inode
1714  * @fattr: updated attributes
1715  *
1716  * After an operation that has changed the inode metadata, mark the
1717  * attribute cache as being invalid, then try to update it. Fake up
1718  * weak cache consistency data, if none exist.
1719  *
1720  * This function is mainly designed to be used by the ->write_done() functions.
1721  */
1722 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1723 {
1724 	int status;
1725 
1726 	/* Don't do a WCC update if these attributes are already stale */
1727 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1728 			!nfs_inode_attrs_need_update(inode, fattr)) {
1729 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1730 				| NFS_ATTR_FATTR_PRESIZE
1731 				| NFS_ATTR_FATTR_PREMTIME
1732 				| NFS_ATTR_FATTR_PRECTIME);
1733 		goto out_noforce;
1734 	}
1735 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1736 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1737 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1738 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1739 	}
1740 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1741 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1742 		fattr->pre_ctime = inode->i_ctime;
1743 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1744 	}
1745 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1746 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1747 		fattr->pre_mtime = inode->i_mtime;
1748 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1749 	}
1750 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1751 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1752 		fattr->pre_size = i_size_read(inode);
1753 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1754 	}
1755 out_noforce:
1756 	status = nfs_post_op_update_inode_locked(inode, fattr,
1757 			NFS_INO_INVALID_CHANGE
1758 			| NFS_INO_INVALID_CTIME
1759 			| NFS_INO_INVALID_MTIME);
1760 	return status;
1761 }
1762 
1763 /**
1764  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1765  * @inode: pointer to inode
1766  * @fattr: updated attributes
1767  *
1768  * After an operation that has changed the inode metadata, mark the
1769  * attribute cache as being invalid, then try to update it. Fake up
1770  * weak cache consistency data, if none exist.
1771  *
1772  * This function is mainly designed to be used by the ->write_done() functions.
1773  */
1774 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1775 {
1776 	int status;
1777 
1778 	spin_lock(&inode->i_lock);
1779 	nfs_fattr_set_barrier(fattr);
1780 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1781 	spin_unlock(&inode->i_lock);
1782 	return status;
1783 }
1784 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1785 
1786 
1787 /*
1788  * Many nfs protocol calls return the new file attributes after
1789  * an operation.  Here we update the inode to reflect the state
1790  * of the server's inode.
1791  *
1792  * This is a bit tricky because we have to make sure all dirty pages
1793  * have been sent off to the server before calling invalidate_inode_pages.
1794  * To make sure no other process adds more write requests while we try
1795  * our best to flush them, we make them sleep during the attribute refresh.
1796  *
1797  * A very similar scenario holds for the dir cache.
1798  */
1799 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1800 {
1801 	struct nfs_server *server;
1802 	struct nfs_inode *nfsi = NFS_I(inode);
1803 	loff_t cur_isize, new_isize;
1804 	unsigned long invalid = 0;
1805 	unsigned long now = jiffies;
1806 	unsigned long save_cache_validity;
1807 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
1808 	bool cache_revalidated = true;
1809 	bool attr_changed = false;
1810 	bool have_delegation;
1811 
1812 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1813 			__func__, inode->i_sb->s_id, inode->i_ino,
1814 			nfs_display_fhandle_hash(NFS_FH(inode)),
1815 			atomic_read(&inode->i_count), fattr->valid);
1816 
1817 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1818 		/* Only a mounted-on-fileid? Just exit */
1819 		if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1820 			return 0;
1821 	/* Has the inode gone and changed behind our back? */
1822 	} else if (nfsi->fileid != fattr->fileid) {
1823 		/* Is this perhaps the mounted-on fileid? */
1824 		if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1825 		    nfsi->fileid == fattr->mounted_on_fileid)
1826 			return 0;
1827 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1828 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1829 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1830 			inode->i_sb->s_id, (long long)nfsi->fileid,
1831 			(long long)fattr->fileid);
1832 		goto out_err;
1833 	}
1834 
1835 	/*
1836 	 * Make sure the inode's type hasn't changed.
1837 	 */
1838 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1839 		/*
1840 		* Big trouble! The inode has become a different object.
1841 		*/
1842 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1843 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1844 		goto out_err;
1845 	}
1846 
1847 	server = NFS_SERVER(inode);
1848 	/* Update the fsid? */
1849 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1850 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1851 			!IS_AUTOMOUNT(inode))
1852 		server->fsid = fattr->fsid;
1853 
1854 	/* Save the delegation state before clearing cache_validity */
1855 	have_delegation = nfs_have_delegated_attributes(inode);
1856 
1857 	/*
1858 	 * Update the read time so we don't revalidate too often.
1859 	 */
1860 	nfsi->read_cache_jiffies = fattr->time_start;
1861 
1862 	save_cache_validity = nfsi->cache_validity;
1863 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1864 			| NFS_INO_INVALID_ATIME
1865 			| NFS_INO_REVAL_FORCED
1866 			| NFS_INO_REVAL_PAGECACHE);
1867 
1868 	/* Do atomic weak cache consistency updates */
1869 	nfs_wcc_update_inode(inode, fattr);
1870 
1871 	if (pnfs_layoutcommit_outstanding(inode)) {
1872 		nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1873 		cache_revalidated = false;
1874 	}
1875 
1876 	/* More cache consistency checks */
1877 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1878 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1879 			/* Could it be a race with writeback? */
1880 			if (!(have_writers || have_delegation)) {
1881 				invalid |= NFS_INO_INVALID_DATA
1882 					| NFS_INO_INVALID_ACCESS
1883 					| NFS_INO_INVALID_ACL;
1884 				/* Force revalidate of all attributes */
1885 				save_cache_validity |= NFS_INO_INVALID_CTIME
1886 					| NFS_INO_INVALID_MTIME
1887 					| NFS_INO_INVALID_SIZE
1888 					| NFS_INO_INVALID_OTHER;
1889 				if (S_ISDIR(inode->i_mode))
1890 					nfs_force_lookup_revalidate(inode);
1891 				dprintk("NFS: change_attr change on server for file %s/%ld\n",
1892 						inode->i_sb->s_id,
1893 						inode->i_ino);
1894 			} else if (!have_delegation)
1895 				nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
1896 			inode_set_iversion_raw(inode, fattr->change_attr);
1897 			attr_changed = true;
1898 		}
1899 	} else {
1900 		nfsi->cache_validity |= save_cache_validity &
1901 				(NFS_INO_INVALID_CHANGE
1902 				| NFS_INO_REVAL_PAGECACHE
1903 				| NFS_INO_REVAL_FORCED);
1904 		cache_revalidated = false;
1905 	}
1906 
1907 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1908 		inode->i_mtime = fattr->mtime;
1909 	} else if (server->caps & NFS_CAP_MTIME) {
1910 		nfsi->cache_validity |= save_cache_validity &
1911 				(NFS_INO_INVALID_MTIME
1912 				| NFS_INO_REVAL_FORCED);
1913 		cache_revalidated = false;
1914 	}
1915 
1916 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1917 		inode->i_ctime = fattr->ctime;
1918 	} else if (server->caps & NFS_CAP_CTIME) {
1919 		nfsi->cache_validity |= save_cache_validity &
1920 				(NFS_INO_INVALID_CTIME
1921 				| NFS_INO_REVAL_FORCED);
1922 		cache_revalidated = false;
1923 	}
1924 
1925 	/* Check if our cached file size is stale */
1926 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1927 		new_isize = nfs_size_to_loff_t(fattr->size);
1928 		cur_isize = i_size_read(inode);
1929 		if (new_isize != cur_isize && !have_delegation) {
1930 			/* Do we perhaps have any outstanding writes, or has
1931 			 * the file grown beyond our last write? */
1932 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1933 				i_size_write(inode, new_isize);
1934 				if (!have_writers)
1935 					invalid |= NFS_INO_INVALID_DATA;
1936 				attr_changed = true;
1937 			}
1938 			dprintk("NFS: isize change on server for file %s/%ld "
1939 					"(%Ld to %Ld)\n",
1940 					inode->i_sb->s_id,
1941 					inode->i_ino,
1942 					(long long)cur_isize,
1943 					(long long)new_isize);
1944 		}
1945 	} else {
1946 		nfsi->cache_validity |= save_cache_validity &
1947 				(NFS_INO_INVALID_SIZE
1948 				| NFS_INO_REVAL_PAGECACHE
1949 				| NFS_INO_REVAL_FORCED);
1950 		cache_revalidated = false;
1951 	}
1952 
1953 
1954 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1955 		inode->i_atime = fattr->atime;
1956 	else if (server->caps & NFS_CAP_ATIME) {
1957 		nfsi->cache_validity |= save_cache_validity &
1958 				(NFS_INO_INVALID_ATIME
1959 				| NFS_INO_REVAL_FORCED);
1960 		cache_revalidated = false;
1961 	}
1962 
1963 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1964 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1965 			umode_t newmode = inode->i_mode & S_IFMT;
1966 			newmode |= fattr->mode & S_IALLUGO;
1967 			inode->i_mode = newmode;
1968 			invalid |= NFS_INO_INVALID_ACCESS
1969 				| NFS_INO_INVALID_ACL;
1970 			attr_changed = true;
1971 		}
1972 	} else if (server->caps & NFS_CAP_MODE) {
1973 		nfsi->cache_validity |= save_cache_validity &
1974 				(NFS_INO_INVALID_OTHER
1975 				| NFS_INO_REVAL_FORCED);
1976 		cache_revalidated = false;
1977 	}
1978 
1979 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1980 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1981 			invalid |= NFS_INO_INVALID_ACCESS
1982 				| NFS_INO_INVALID_ACL;
1983 			inode->i_uid = fattr->uid;
1984 			attr_changed = true;
1985 		}
1986 	} else if (server->caps & NFS_CAP_OWNER) {
1987 		nfsi->cache_validity |= save_cache_validity &
1988 				(NFS_INO_INVALID_OTHER
1989 				| NFS_INO_REVAL_FORCED);
1990 		cache_revalidated = false;
1991 	}
1992 
1993 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1994 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1995 			invalid |= NFS_INO_INVALID_ACCESS
1996 				| NFS_INO_INVALID_ACL;
1997 			inode->i_gid = fattr->gid;
1998 			attr_changed = true;
1999 		}
2000 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
2001 		nfsi->cache_validity |= save_cache_validity &
2002 				(NFS_INO_INVALID_OTHER
2003 				| NFS_INO_REVAL_FORCED);
2004 		cache_revalidated = false;
2005 	}
2006 
2007 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2008 		if (inode->i_nlink != fattr->nlink) {
2009 			if (S_ISDIR(inode->i_mode))
2010 				invalid |= NFS_INO_INVALID_DATA;
2011 			set_nlink(inode, fattr->nlink);
2012 			attr_changed = true;
2013 		}
2014 	} else if (server->caps & NFS_CAP_NLINK) {
2015 		nfsi->cache_validity |= save_cache_validity &
2016 				(NFS_INO_INVALID_OTHER
2017 				| NFS_INO_REVAL_FORCED);
2018 		cache_revalidated = false;
2019 	}
2020 
2021 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2022 		/*
2023 		 * report the blocks in 512byte units
2024 		 */
2025 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2026 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2027 		inode->i_blocks = fattr->du.nfs2.blocks;
2028 	else
2029 		cache_revalidated = false;
2030 
2031 	/* Update attrtimeo value if we're out of the unstable period */
2032 	if (attr_changed) {
2033 		invalid &= ~NFS_INO_INVALID_ATTR;
2034 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2035 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2036 		nfsi->attrtimeo_timestamp = now;
2037 		/* Set barrier to be more recent than all outstanding updates */
2038 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2039 	} else {
2040 		if (cache_revalidated) {
2041 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2042 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2043 				nfsi->attrtimeo <<= 1;
2044 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2045 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2046 			}
2047 			nfsi->attrtimeo_timestamp = now;
2048 		}
2049 		/* Set the barrier to be more recent than this fattr */
2050 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
2051 			nfsi->attr_gencount = fattr->gencount;
2052 	}
2053 
2054 	/* Don't invalidate the data if we were to blame */
2055 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2056 				|| S_ISLNK(inode->i_mode)))
2057 		invalid &= ~NFS_INO_INVALID_DATA;
2058 	nfs_set_cache_invalid(inode, invalid);
2059 
2060 	return 0;
2061  out_err:
2062 	/*
2063 	 * No need to worry about unhashing the dentry, as the
2064 	 * lookup validation will know that the inode is bad.
2065 	 * (But we fall through to invalidate the caches.)
2066 	 */
2067 	nfs_invalidate_inode(inode);
2068 	return -ESTALE;
2069 }
2070 
2071 struct inode *nfs_alloc_inode(struct super_block *sb)
2072 {
2073 	struct nfs_inode *nfsi;
2074 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2075 	if (!nfsi)
2076 		return NULL;
2077 	nfsi->flags = 0UL;
2078 	nfsi->cache_validity = 0UL;
2079 #if IS_ENABLED(CONFIG_NFS_V4)
2080 	nfsi->nfs4_acl = NULL;
2081 #endif /* CONFIG_NFS_V4 */
2082 	return &nfsi->vfs_inode;
2083 }
2084 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2085 
2086 void nfs_free_inode(struct inode *inode)
2087 {
2088 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2089 }
2090 EXPORT_SYMBOL_GPL(nfs_free_inode);
2091 
2092 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2093 {
2094 #if IS_ENABLED(CONFIG_NFS_V4)
2095 	INIT_LIST_HEAD(&nfsi->open_states);
2096 	nfsi->delegation = NULL;
2097 	init_rwsem(&nfsi->rwsem);
2098 	nfsi->layout = NULL;
2099 #endif
2100 }
2101 
2102 static void init_once(void *foo)
2103 {
2104 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2105 
2106 	inode_init_once(&nfsi->vfs_inode);
2107 	INIT_LIST_HEAD(&nfsi->open_files);
2108 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2109 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2110 	INIT_LIST_HEAD(&nfsi->commit_info.list);
2111 	atomic_long_set(&nfsi->nrequests, 0);
2112 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
2113 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
2114 	init_rwsem(&nfsi->rmdir_sem);
2115 	mutex_init(&nfsi->commit_mutex);
2116 	nfs4_init_once(nfsi);
2117 	nfsi->cache_change_attribute = 0;
2118 }
2119 
2120 static int __init nfs_init_inodecache(void)
2121 {
2122 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2123 					     sizeof(struct nfs_inode),
2124 					     0, (SLAB_RECLAIM_ACCOUNT|
2125 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2126 					     init_once);
2127 	if (nfs_inode_cachep == NULL)
2128 		return -ENOMEM;
2129 
2130 	return 0;
2131 }
2132 
2133 static void nfs_destroy_inodecache(void)
2134 {
2135 	/*
2136 	 * Make sure all delayed rcu free inodes are flushed before we
2137 	 * destroy cache.
2138 	 */
2139 	rcu_barrier();
2140 	kmem_cache_destroy(nfs_inode_cachep);
2141 }
2142 
2143 struct workqueue_struct *nfsiod_workqueue;
2144 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2145 
2146 /*
2147  * start up the nfsiod workqueue
2148  */
2149 static int nfsiod_start(void)
2150 {
2151 	struct workqueue_struct *wq;
2152 	dprintk("RPC:       creating workqueue nfsiod\n");
2153 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2154 	if (wq == NULL)
2155 		return -ENOMEM;
2156 	nfsiod_workqueue = wq;
2157 	return 0;
2158 }
2159 
2160 /*
2161  * Destroy the nfsiod workqueue
2162  */
2163 static void nfsiod_stop(void)
2164 {
2165 	struct workqueue_struct *wq;
2166 
2167 	wq = nfsiod_workqueue;
2168 	if (wq == NULL)
2169 		return;
2170 	nfsiod_workqueue = NULL;
2171 	destroy_workqueue(wq);
2172 }
2173 
2174 unsigned int nfs_net_id;
2175 EXPORT_SYMBOL_GPL(nfs_net_id);
2176 
2177 static int nfs_net_init(struct net *net)
2178 {
2179 	nfs_clients_init(net);
2180 	return nfs_fs_proc_net_init(net);
2181 }
2182 
2183 static void nfs_net_exit(struct net *net)
2184 {
2185 	nfs_fs_proc_net_exit(net);
2186 	nfs_clients_exit(net);
2187 }
2188 
2189 static struct pernet_operations nfs_net_ops = {
2190 	.init = nfs_net_init,
2191 	.exit = nfs_net_exit,
2192 	.id   = &nfs_net_id,
2193 	.size = sizeof(struct nfs_net),
2194 };
2195 
2196 /*
2197  * Initialize NFS
2198  */
2199 static int __init init_nfs_fs(void)
2200 {
2201 	int err;
2202 
2203 	err = nfs_sysfs_init();
2204 	if (err < 0)
2205 		goto out10;
2206 
2207 	err = register_pernet_subsys(&nfs_net_ops);
2208 	if (err < 0)
2209 		goto out9;
2210 
2211 	err = nfs_fscache_register();
2212 	if (err < 0)
2213 		goto out8;
2214 
2215 	err = nfsiod_start();
2216 	if (err)
2217 		goto out7;
2218 
2219 	err = nfs_fs_proc_init();
2220 	if (err)
2221 		goto out6;
2222 
2223 	err = nfs_init_nfspagecache();
2224 	if (err)
2225 		goto out5;
2226 
2227 	err = nfs_init_inodecache();
2228 	if (err)
2229 		goto out4;
2230 
2231 	err = nfs_init_readpagecache();
2232 	if (err)
2233 		goto out3;
2234 
2235 	err = nfs_init_writepagecache();
2236 	if (err)
2237 		goto out2;
2238 
2239 	err = nfs_init_directcache();
2240 	if (err)
2241 		goto out1;
2242 
2243 	rpc_proc_register(&init_net, &nfs_rpcstat);
2244 
2245 	err = register_nfs_fs();
2246 	if (err)
2247 		goto out0;
2248 
2249 	return 0;
2250 out0:
2251 	rpc_proc_unregister(&init_net, "nfs");
2252 	nfs_destroy_directcache();
2253 out1:
2254 	nfs_destroy_writepagecache();
2255 out2:
2256 	nfs_destroy_readpagecache();
2257 out3:
2258 	nfs_destroy_inodecache();
2259 out4:
2260 	nfs_destroy_nfspagecache();
2261 out5:
2262 	nfs_fs_proc_exit();
2263 out6:
2264 	nfsiod_stop();
2265 out7:
2266 	nfs_fscache_unregister();
2267 out8:
2268 	unregister_pernet_subsys(&nfs_net_ops);
2269 out9:
2270 	nfs_sysfs_exit();
2271 out10:
2272 	return err;
2273 }
2274 
2275 static void __exit exit_nfs_fs(void)
2276 {
2277 	nfs_destroy_directcache();
2278 	nfs_destroy_writepagecache();
2279 	nfs_destroy_readpagecache();
2280 	nfs_destroy_inodecache();
2281 	nfs_destroy_nfspagecache();
2282 	nfs_fscache_unregister();
2283 	unregister_pernet_subsys(&nfs_net_ops);
2284 	rpc_proc_unregister(&init_net, "nfs");
2285 	unregister_nfs_fs();
2286 	nfs_fs_proc_exit();
2287 	nfsiod_stop();
2288 	nfs_sysfs_exit();
2289 }
2290 
2291 /* Not quite true; I just maintain it */
2292 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2293 MODULE_LICENSE("GPL");
2294 module_param(enable_ino64, bool, 0644);
2295 
2296 module_init(init_nfs_fs)
2297 module_exit(exit_nfs_fs)
2298