xref: /openbmc/linux/fs/nfs/inode.c (revision b240b419db5d624ce7a5a397d6f62a1a686009ec)
1 /*
2  *  linux/fs/nfs/inode.c
3  *
4  *  Copyright (C) 1992  Rick Sladkey
5  *
6  *  nfs inode and superblock handling functions
7  *
8  *  Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9  *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
10  *
11  *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12  *  J.S.Peatfield@damtp.cam.ac.uk
13  *
14  */
15 
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched/signal.h>
19 #include <linux/time.h>
20 #include <linux/kernel.h>
21 #include <linux/mm.h>
22 #include <linux/string.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/unistd.h>
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/metrics.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/vfs.h>
36 #include <linux/inet.h>
37 #include <linux/nfs_xdr.h>
38 #include <linux/slab.h>
39 #include <linux/compat.h>
40 #include <linux/freezer.h>
41 #include <linux/uaccess.h>
42 #include <linux/iversion.h>
43 
44 #include "nfs4_fs.h"
45 #include "callback.h"
46 #include "delegation.h"
47 #include "iostat.h"
48 #include "internal.h"
49 #include "fscache.h"
50 #include "pnfs.h"
51 #include "nfs.h"
52 #include "netns.h"
53 
54 #include "nfstrace.h"
55 
56 #define NFSDBG_FACILITY		NFSDBG_VFS
57 
58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED	1
59 
60 /* Default is to see 64-bit inode numbers */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
62 
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
65 
66 static struct kmem_cache * nfs_inode_cachep;
67 
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
70 {
71 	return nfs_fileid_to_ino_t(fattr->fileid);
72 }
73 
74 static int nfs_wait_killable(int mode)
75 {
76 	freezable_schedule_unsafe();
77 	if (signal_pending_state(mode, current))
78 		return -ERESTARTSYS;
79 	return 0;
80 }
81 
82 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
83 {
84 	return nfs_wait_killable(mode);
85 }
86 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
87 
88 /**
89  * nfs_compat_user_ino64 - returns the user-visible inode number
90  * @fileid: 64-bit fileid
91  *
92  * This function returns a 32-bit inode number if the boot parameter
93  * nfs.enable_ino64 is zero.
94  */
95 u64 nfs_compat_user_ino64(u64 fileid)
96 {
97 #ifdef CONFIG_COMPAT
98 	compat_ulong_t ino;
99 #else
100 	unsigned long ino;
101 #endif
102 
103 	if (enable_ino64)
104 		return fileid;
105 	ino = fileid;
106 	if (sizeof(ino) < sizeof(fileid))
107 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
108 	return ino;
109 }
110 
111 int nfs_drop_inode(struct inode *inode)
112 {
113 	return NFS_STALE(inode) || generic_drop_inode(inode);
114 }
115 EXPORT_SYMBOL_GPL(nfs_drop_inode);
116 
117 void nfs_clear_inode(struct inode *inode)
118 {
119 	/*
120 	 * The following should never happen...
121 	 */
122 	WARN_ON_ONCE(nfs_have_writebacks(inode));
123 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
124 	nfs_zap_acl_cache(inode);
125 	nfs_access_zap_cache(inode);
126 	nfs_fscache_clear_inode(inode);
127 }
128 EXPORT_SYMBOL_GPL(nfs_clear_inode);
129 
130 void nfs_evict_inode(struct inode *inode)
131 {
132 	truncate_inode_pages_final(&inode->i_data);
133 	clear_inode(inode);
134 	nfs_clear_inode(inode);
135 }
136 
137 int nfs_sync_inode(struct inode *inode)
138 {
139 	inode_dio_wait(inode);
140 	return nfs_wb_all(inode);
141 }
142 EXPORT_SYMBOL_GPL(nfs_sync_inode);
143 
144 /**
145  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
146  */
147 int nfs_sync_mapping(struct address_space *mapping)
148 {
149 	int ret = 0;
150 
151 	if (mapping->nrpages != 0) {
152 		unmap_mapping_range(mapping, 0, 0, 0);
153 		ret = nfs_wb_all(mapping->host);
154 	}
155 	return ret;
156 }
157 
158 static int nfs_attribute_timeout(struct inode *inode)
159 {
160 	struct nfs_inode *nfsi = NFS_I(inode);
161 
162 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
163 }
164 
165 static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
166 {
167 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
168 
169 	/* Special case for the pagecache or access cache */
170 	if (flags == NFS_INO_REVAL_PAGECACHE &&
171 	    !(cache_validity & NFS_INO_REVAL_FORCED))
172 		return false;
173 	return (cache_validity & flags) != 0;
174 }
175 
176 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
177 {
178 	unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
179 
180 	if ((cache_validity & flags) != 0)
181 		return true;
182 	if (nfs_attribute_timeout(inode))
183 		return true;
184 	return false;
185 }
186 
187 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
188 {
189 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
190 		return nfs_check_cache_invalid_delegated(inode, flags);
191 
192 	return nfs_check_cache_invalid_not_delegated(inode, flags);
193 }
194 
195 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
196 {
197 	struct nfs_inode *nfsi = NFS_I(inode);
198 
199 	if (inode->i_mapping->nrpages == 0)
200 		flags &= ~NFS_INO_INVALID_DATA;
201 	nfsi->cache_validity |= flags;
202 	if (flags & NFS_INO_INVALID_DATA)
203 		nfs_fscache_invalidate(inode);
204 }
205 
206 /*
207  * Invalidate the local caches
208  */
209 static void nfs_zap_caches_locked(struct inode *inode)
210 {
211 	struct nfs_inode *nfsi = NFS_I(inode);
212 	int mode = inode->i_mode;
213 
214 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
215 
216 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
217 	nfsi->attrtimeo_timestamp = jiffies;
218 
219 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
220 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
221 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
222 					| NFS_INO_INVALID_DATA
223 					| NFS_INO_INVALID_ACCESS
224 					| NFS_INO_INVALID_ACL
225 					| NFS_INO_REVAL_PAGECACHE);
226 	} else
227 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
228 					| NFS_INO_INVALID_ACCESS
229 					| NFS_INO_INVALID_ACL
230 					| NFS_INO_REVAL_PAGECACHE);
231 	nfs_zap_label_cache_locked(nfsi);
232 }
233 
234 void nfs_zap_caches(struct inode *inode)
235 {
236 	spin_lock(&inode->i_lock);
237 	nfs_zap_caches_locked(inode);
238 	spin_unlock(&inode->i_lock);
239 }
240 
241 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
242 {
243 	if (mapping->nrpages != 0) {
244 		spin_lock(&inode->i_lock);
245 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
246 		spin_unlock(&inode->i_lock);
247 	}
248 }
249 
250 void nfs_zap_acl_cache(struct inode *inode)
251 {
252 	void (*clear_acl_cache)(struct inode *);
253 
254 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
255 	if (clear_acl_cache != NULL)
256 		clear_acl_cache(inode);
257 	spin_lock(&inode->i_lock);
258 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
259 	spin_unlock(&inode->i_lock);
260 }
261 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
262 
263 void nfs_invalidate_atime(struct inode *inode)
264 {
265 	spin_lock(&inode->i_lock);
266 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
267 	spin_unlock(&inode->i_lock);
268 }
269 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
270 
271 /*
272  * Invalidate, but do not unhash, the inode.
273  * NB: must be called with inode->i_lock held!
274  */
275 static void nfs_invalidate_inode(struct inode *inode)
276 {
277 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
278 	nfs_zap_caches_locked(inode);
279 }
280 
281 struct nfs_find_desc {
282 	struct nfs_fh		*fh;
283 	struct nfs_fattr	*fattr;
284 };
285 
286 /*
287  * In NFSv3 we can have 64bit inode numbers. In order to support
288  * this, and re-exported directories (also seen in NFSv2)
289  * we are forced to allow 2 different inodes to have the same
290  * i_ino.
291  */
292 static int
293 nfs_find_actor(struct inode *inode, void *opaque)
294 {
295 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
296 	struct nfs_fh		*fh = desc->fh;
297 	struct nfs_fattr	*fattr = desc->fattr;
298 
299 	if (NFS_FILEID(inode) != fattr->fileid)
300 		return 0;
301 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
302 		return 0;
303 	if (nfs_compare_fh(NFS_FH(inode), fh))
304 		return 0;
305 	if (is_bad_inode(inode) || NFS_STALE(inode))
306 		return 0;
307 	return 1;
308 }
309 
310 static int
311 nfs_init_locked(struct inode *inode, void *opaque)
312 {
313 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
314 	struct nfs_fattr	*fattr = desc->fattr;
315 
316 	set_nfs_fileid(inode, fattr->fileid);
317 	inode->i_mode = fattr->mode;
318 	nfs_copy_fh(NFS_FH(inode), desc->fh);
319 	return 0;
320 }
321 
322 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
323 static void nfs_clear_label_invalid(struct inode *inode)
324 {
325 	spin_lock(&inode->i_lock);
326 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
327 	spin_unlock(&inode->i_lock);
328 }
329 
330 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
331 					struct nfs4_label *label)
332 {
333 	int error;
334 
335 	if (label == NULL)
336 		return;
337 
338 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
339 		error = security_inode_notifysecctx(inode, label->label,
340 				label->len);
341 		if (error)
342 			printk(KERN_ERR "%s() %s %d "
343 					"security_inode_notifysecctx() %d\n",
344 					__func__,
345 					(char *)label->label,
346 					label->len, error);
347 		nfs_clear_label_invalid(inode);
348 	}
349 }
350 
351 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
352 {
353 	struct nfs4_label *label = NULL;
354 	int minor_version = server->nfs_client->cl_minorversion;
355 
356 	if (minor_version < 2)
357 		return label;
358 
359 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
360 		return label;
361 
362 	label = kzalloc(sizeof(struct nfs4_label), flags);
363 	if (label == NULL)
364 		return ERR_PTR(-ENOMEM);
365 
366 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
367 	if (label->label == NULL) {
368 		kfree(label);
369 		return ERR_PTR(-ENOMEM);
370 	}
371 	label->len = NFS4_MAXLABELLEN;
372 
373 	return label;
374 }
375 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
376 #else
377 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
378 					struct nfs4_label *label)
379 {
380 }
381 #endif
382 EXPORT_SYMBOL_GPL(nfs_setsecurity);
383 
384 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
385 struct inode *
386 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
387 {
388 	struct nfs_find_desc desc = {
389 		.fh	= fh,
390 		.fattr	= fattr,
391 	};
392 	struct inode *inode;
393 	unsigned long hash;
394 
395 	if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
396 	    !(fattr->valid & NFS_ATTR_FATTR_TYPE))
397 		return NULL;
398 
399 	hash = nfs_fattr_to_ino_t(fattr);
400 	inode = ilookup5(sb, hash, nfs_find_actor, &desc);
401 
402 	dprintk("%s: returning %p\n", __func__, inode);
403 	return inode;
404 }
405 
406 /*
407  * This is our front-end to iget that looks up inodes by file handle
408  * instead of inode number.
409  */
410 struct inode *
411 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
412 {
413 	struct nfs_find_desc desc = {
414 		.fh	= fh,
415 		.fattr	= fattr
416 	};
417 	struct inode *inode = ERR_PTR(-ENOENT);
418 	unsigned long hash;
419 
420 	nfs_attr_check_mountpoint(sb, fattr);
421 
422 	if (nfs_attr_use_mounted_on_fileid(fattr))
423 		fattr->fileid = fattr->mounted_on_fileid;
424 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
425 		goto out_no_inode;
426 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
427 		goto out_no_inode;
428 
429 	hash = nfs_fattr_to_ino_t(fattr);
430 
431 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
432 	if (inode == NULL) {
433 		inode = ERR_PTR(-ENOMEM);
434 		goto out_no_inode;
435 	}
436 
437 	if (inode->i_state & I_NEW) {
438 		struct nfs_inode *nfsi = NFS_I(inode);
439 		unsigned long now = jiffies;
440 
441 		/* We set i_ino for the few things that still rely on it,
442 		 * such as stat(2) */
443 		inode->i_ino = hash;
444 
445 		/* We can't support update_atime(), since the server will reset it */
446 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
447 		inode->i_mode = fattr->mode;
448 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
449 				&& nfs_server_capable(inode, NFS_CAP_MODE))
450 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
451 		/* Why so? Because we want revalidate for devices/FIFOs, and
452 		 * that's precisely what we have in nfs_file_inode_operations.
453 		 */
454 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
455 		if (S_ISREG(inode->i_mode)) {
456 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
457 			inode->i_data.a_ops = &nfs_file_aops;
458 		} else if (S_ISDIR(inode->i_mode)) {
459 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
460 			inode->i_fop = &nfs_dir_operations;
461 			inode->i_data.a_ops = &nfs_dir_aops;
462 			/* Deal with crossing mountpoints */
463 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
464 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
465 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
466 					inode->i_op = &nfs_referral_inode_operations;
467 				else
468 					inode->i_op = &nfs_mountpoint_inode_operations;
469 				inode->i_fop = NULL;
470 				inode->i_flags |= S_AUTOMOUNT;
471 			}
472 		} else if (S_ISLNK(inode->i_mode)) {
473 			inode->i_op = &nfs_symlink_inode_operations;
474 			inode_nohighmem(inode);
475 		} else
476 			init_special_inode(inode, inode->i_mode, fattr->rdev);
477 
478 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
479 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
480 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
481 		inode_set_iversion_raw(inode, 0);
482 		inode->i_size = 0;
483 		clear_nlink(inode);
484 		inode->i_uid = make_kuid(&init_user_ns, -2);
485 		inode->i_gid = make_kgid(&init_user_ns, -2);
486 		inode->i_blocks = 0;
487 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
488 		nfsi->write_io = 0;
489 		nfsi->read_io = 0;
490 
491 		nfsi->read_cache_jiffies = fattr->time_start;
492 		nfsi->attr_gencount = fattr->gencount;
493 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
494 			inode->i_atime = fattr->atime;
495 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
496 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
497 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
498 			inode->i_mtime = fattr->mtime;
499 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
500 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
501 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
502 			inode->i_ctime = fattr->ctime;
503 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
504 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
505 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
506 			inode_set_iversion_raw(inode, fattr->change_attr);
507 		else
508 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
509 				| NFS_INO_REVAL_PAGECACHE);
510 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
511 			inode->i_size = nfs_size_to_loff_t(fattr->size);
512 		else
513 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
514 				| NFS_INO_REVAL_PAGECACHE);
515 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
516 			set_nlink(inode, fattr->nlink);
517 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
518 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
519 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
520 			inode->i_uid = fattr->uid;
521 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
522 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
523 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
524 			inode->i_gid = fattr->gid;
525 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
526 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
527 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
528 			inode->i_blocks = fattr->du.nfs2.blocks;
529 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
530 			/*
531 			 * report the blocks in 512byte units
532 			 */
533 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
534 		}
535 
536 		nfs_setsecurity(inode, fattr, label);
537 
538 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
539 		nfsi->attrtimeo_timestamp = now;
540 		nfsi->access_cache = RB_ROOT;
541 
542 		nfs_fscache_init_inode(inode);
543 
544 		unlock_new_inode(inode);
545 	} else {
546 		int err = nfs_refresh_inode(inode, fattr);
547 		if (err < 0) {
548 			iput(inode);
549 			inode = ERR_PTR(err);
550 			goto out_no_inode;
551 		}
552 	}
553 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
554 		inode->i_sb->s_id,
555 		(unsigned long long)NFS_FILEID(inode),
556 		nfs_display_fhandle_hash(fh),
557 		atomic_read(&inode->i_count));
558 
559 out:
560 	return inode;
561 
562 out_no_inode:
563 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
564 	goto out;
565 }
566 EXPORT_SYMBOL_GPL(nfs_fhget);
567 
568 #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)
569 
570 int
571 nfs_setattr(struct dentry *dentry, struct iattr *attr)
572 {
573 	struct inode *inode = d_inode(dentry);
574 	struct nfs_fattr *fattr;
575 	int error = 0;
576 
577 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
578 
579 	/* skip mode change if it's just for clearing setuid/setgid */
580 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
581 		attr->ia_valid &= ~ATTR_MODE;
582 
583 	if (attr->ia_valid & ATTR_SIZE) {
584 		BUG_ON(!S_ISREG(inode->i_mode));
585 
586 		error = inode_newsize_ok(inode, attr->ia_size);
587 		if (error)
588 			return error;
589 
590 		if (attr->ia_size == i_size_read(inode))
591 			attr->ia_valid &= ~ATTR_SIZE;
592 	}
593 
594 	/* Optimization: if the end result is no change, don't RPC */
595 	attr->ia_valid &= NFS_VALID_ATTRS;
596 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
597 		return 0;
598 
599 	trace_nfs_setattr_enter(inode);
600 
601 	/* Write all dirty data */
602 	if (S_ISREG(inode->i_mode))
603 		nfs_sync_inode(inode);
604 
605 	fattr = nfs_alloc_fattr();
606 	if (fattr == NULL) {
607 		error = -ENOMEM;
608 		goto out;
609 	}
610 
611 	/*
612 	 * Return any delegations if we're going to change ACLs
613 	 */
614 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
615 		NFS_PROTO(inode)->return_delegation(inode);
616 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
617 	if (error == 0)
618 		error = nfs_refresh_inode(inode, fattr);
619 	nfs_free_fattr(fattr);
620 out:
621 	trace_nfs_setattr_exit(inode, error);
622 	return error;
623 }
624 EXPORT_SYMBOL_GPL(nfs_setattr);
625 
626 /**
627  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
628  * @inode: inode of the file used
629  * @offset: file offset to start truncating
630  *
631  * This is a copy of the common vmtruncate, but with the locking
632  * corrected to take into account the fact that NFS requires
633  * inode->i_size to be updated under the inode->i_lock.
634  * Note: must be called with inode->i_lock held!
635  */
636 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
637 {
638 	int err;
639 
640 	err = inode_newsize_ok(inode, offset);
641 	if (err)
642 		goto out;
643 
644 	i_size_write(inode, offset);
645 	/* Optimisation */
646 	if (offset == 0)
647 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
648 
649 	spin_unlock(&inode->i_lock);
650 	truncate_pagecache(inode, offset);
651 	spin_lock(&inode->i_lock);
652 out:
653 	return err;
654 }
655 
656 /**
657  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
658  * @inode: pointer to struct inode
659  * @attr: pointer to struct iattr
660  *
661  * Note: we do this in the *proc.c in order to ensure that
662  *       it works for things like exclusive creates too.
663  */
664 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
665 		struct nfs_fattr *fattr)
666 {
667 	/* Barrier: bump the attribute generation count. */
668 	nfs_fattr_set_barrier(fattr);
669 
670 	spin_lock(&inode->i_lock);
671 	NFS_I(inode)->attr_gencount = fattr->gencount;
672 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
673 		if ((attr->ia_valid & ATTR_MODE) != 0) {
674 			int mode = attr->ia_mode & S_IALLUGO;
675 			mode |= inode->i_mode & ~S_IALLUGO;
676 			inode->i_mode = mode;
677 		}
678 		if ((attr->ia_valid & ATTR_UID) != 0)
679 			inode->i_uid = attr->ia_uid;
680 		if ((attr->ia_valid & ATTR_GID) != 0)
681 			inode->i_gid = attr->ia_gid;
682 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
683 				| NFS_INO_INVALID_ACL);
684 	}
685 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
686 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
687 		nfs_vmtruncate(inode, attr->ia_size);
688 	}
689 	if (fattr->valid)
690 		nfs_update_inode(inode, fattr);
691 	else
692 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
693 	spin_unlock(&inode->i_lock);
694 }
695 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
696 
697 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
698 {
699 	struct dentry *parent;
700 
701 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
702 		return;
703 	parent = dget_parent(dentry);
704 	nfs_force_use_readdirplus(d_inode(parent));
705 	dput(parent);
706 }
707 
708 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
709 {
710 	struct dentry *parent;
711 
712 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
713 		return;
714 	parent = dget_parent(dentry);
715 	nfs_advise_use_readdirplus(d_inode(parent));
716 	dput(parent);
717 }
718 
719 static bool nfs_need_revalidate_inode(struct inode *inode)
720 {
721 	if (NFS_I(inode)->cache_validity &
722 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
723 		return true;
724 	if (nfs_attribute_cache_expired(inode))
725 		return true;
726 	return false;
727 }
728 
729 int nfs_getattr(const struct path *path, struct kstat *stat,
730 		u32 request_mask, unsigned int query_flags)
731 {
732 	struct inode *inode = d_inode(path->dentry);
733 	struct nfs_server *server = NFS_SERVER(inode);
734 	unsigned long cache_validity;
735 	int err = 0;
736 	bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
737 	bool do_update = false;
738 
739 	trace_nfs_getattr_enter(inode);
740 
741 	if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync)
742 		goto out_no_update;
743 
744 	/* Flush out writes to the server in order to update c/mtime.  */
745 	if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
746 			S_ISREG(inode->i_mode)) {
747 		err = filemap_write_and_wait(inode->i_mapping);
748 		if (err)
749 			goto out;
750 	}
751 
752 	/*
753 	 * We may force a getattr if the user cares about atime.
754 	 *
755 	 * Note that we only have to check the vfsmount flags here:
756 	 *  - NFS always sets S_NOATIME by so checking it would give a
757 	 *    bogus result
758 	 *  - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
759 	 *    no point in checking those.
760 	 */
761 	if ((path->mnt->mnt_flags & MNT_NOATIME) ||
762 	    ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
763 		request_mask &= ~STATX_ATIME;
764 
765 	/* Is the user requesting attributes that might need revalidation? */
766 	if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
767 					STATX_MTIME|STATX_UID|STATX_GID|
768 					STATX_SIZE|STATX_BLOCKS)))
769 		goto out_no_revalidate;
770 
771 	/* Check whether the cached attributes are stale */
772 	do_update |= force_sync || nfs_attribute_cache_expired(inode);
773 	cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
774 	do_update |= cache_validity &
775 		(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
776 	if (request_mask & STATX_ATIME)
777 		do_update |= cache_validity & NFS_INO_INVALID_ATIME;
778 	if (request_mask & (STATX_CTIME|STATX_MTIME))
779 		do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
780 	if (do_update) {
781 		/* Update the attribute cache */
782 		if (!(server->flags & NFS_MOUNT_NOAC))
783 			nfs_readdirplus_parent_cache_miss(path->dentry);
784 		else
785 			nfs_readdirplus_parent_cache_hit(path->dentry);
786 		err = __nfs_revalidate_inode(server, inode);
787 		if (err)
788 			goto out;
789 	} else
790 		nfs_readdirplus_parent_cache_hit(path->dentry);
791 out_no_revalidate:
792 	/* Only return attributes that were revalidated. */
793 	stat->result_mask &= request_mask;
794 out_no_update:
795 	generic_fillattr(inode, stat);
796 	stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
797 	if (S_ISDIR(inode->i_mode))
798 		stat->blksize = NFS_SERVER(inode)->dtsize;
799 out:
800 	trace_nfs_getattr_exit(inode, err);
801 	return err;
802 }
803 EXPORT_SYMBOL_GPL(nfs_getattr);
804 
805 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
806 {
807 	refcount_set(&l_ctx->count, 1);
808 	l_ctx->lockowner = current->files;
809 	INIT_LIST_HEAD(&l_ctx->list);
810 	atomic_set(&l_ctx->io_count, 0);
811 }
812 
813 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
814 {
815 	struct nfs_lock_context *head = &ctx->lock_context;
816 	struct nfs_lock_context *pos = head;
817 
818 	do {
819 		if (pos->lockowner != current->files)
820 			continue;
821 		refcount_inc(&pos->count);
822 		return pos;
823 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
824 	return NULL;
825 }
826 
827 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
828 {
829 	struct nfs_lock_context *res, *new = NULL;
830 	struct inode *inode = d_inode(ctx->dentry);
831 
832 	spin_lock(&inode->i_lock);
833 	res = __nfs_find_lock_context(ctx);
834 	if (res == NULL) {
835 		spin_unlock(&inode->i_lock);
836 		new = kmalloc(sizeof(*new), GFP_KERNEL);
837 		if (new == NULL)
838 			return ERR_PTR(-ENOMEM);
839 		nfs_init_lock_context(new);
840 		spin_lock(&inode->i_lock);
841 		res = __nfs_find_lock_context(ctx);
842 		if (res == NULL) {
843 			list_add_tail(&new->list, &ctx->lock_context.list);
844 			new->open_context = ctx;
845 			res = new;
846 			new = NULL;
847 		}
848 	}
849 	spin_unlock(&inode->i_lock);
850 	kfree(new);
851 	return res;
852 }
853 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
854 
855 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
856 {
857 	struct nfs_open_context *ctx = l_ctx->open_context;
858 	struct inode *inode = d_inode(ctx->dentry);
859 
860 	if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
861 		return;
862 	list_del(&l_ctx->list);
863 	spin_unlock(&inode->i_lock);
864 	kfree(l_ctx);
865 }
866 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
867 
868 /**
869  * nfs_close_context - Common close_context() routine NFSv2/v3
870  * @ctx: pointer to context
871  * @is_sync: is this a synchronous close
872  *
873  * Ensure that the attributes are up to date if we're mounted
874  * with close-to-open semantics and we have cached data that will
875  * need to be revalidated on open.
876  */
877 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
878 {
879 	struct nfs_inode *nfsi;
880 	struct inode *inode;
881 	struct nfs_server *server;
882 
883 	if (!(ctx->mode & FMODE_WRITE))
884 		return;
885 	if (!is_sync)
886 		return;
887 	inode = d_inode(ctx->dentry);
888 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
889 		return;
890 	nfsi = NFS_I(inode);
891 	if (inode->i_mapping->nrpages == 0)
892 		return;
893 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
894 		return;
895 	if (!list_empty(&nfsi->open_files))
896 		return;
897 	server = NFS_SERVER(inode);
898 	if (server->flags & NFS_MOUNT_NOCTO)
899 		return;
900 	nfs_revalidate_inode(server, inode);
901 }
902 EXPORT_SYMBOL_GPL(nfs_close_context);
903 
904 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
905 						fmode_t f_mode,
906 						struct file *filp)
907 {
908 	struct nfs_open_context *ctx;
909 	struct rpc_cred *cred = rpc_lookup_cred();
910 	if (IS_ERR(cred))
911 		return ERR_CAST(cred);
912 
913 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
914 	if (!ctx) {
915 		put_rpccred(cred);
916 		return ERR_PTR(-ENOMEM);
917 	}
918 	nfs_sb_active(dentry->d_sb);
919 	ctx->dentry = dget(dentry);
920 	ctx->cred = cred;
921 	ctx->state = NULL;
922 	ctx->mode = f_mode;
923 	ctx->flags = 0;
924 	ctx->error = 0;
925 	ctx->flock_owner = (fl_owner_t)filp;
926 	nfs_init_lock_context(&ctx->lock_context);
927 	ctx->lock_context.open_context = ctx;
928 	INIT_LIST_HEAD(&ctx->list);
929 	ctx->mdsthreshold = NULL;
930 	return ctx;
931 }
932 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
933 
934 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
935 {
936 	if (ctx != NULL)
937 		refcount_inc(&ctx->lock_context.count);
938 	return ctx;
939 }
940 EXPORT_SYMBOL_GPL(get_nfs_open_context);
941 
942 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
943 {
944 	struct inode *inode = d_inode(ctx->dentry);
945 	struct super_block *sb = ctx->dentry->d_sb;
946 
947 	if (!list_empty(&ctx->list)) {
948 		if (!refcount_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
949 			return;
950 		list_del(&ctx->list);
951 		spin_unlock(&inode->i_lock);
952 	} else if (!refcount_dec_and_test(&ctx->lock_context.count))
953 		return;
954 	if (inode != NULL)
955 		NFS_PROTO(inode)->close_context(ctx, is_sync);
956 	if (ctx->cred != NULL)
957 		put_rpccred(ctx->cred);
958 	dput(ctx->dentry);
959 	nfs_sb_deactive(sb);
960 	kfree(ctx->mdsthreshold);
961 	kfree(ctx);
962 }
963 
964 void put_nfs_open_context(struct nfs_open_context *ctx)
965 {
966 	__put_nfs_open_context(ctx, 0);
967 }
968 EXPORT_SYMBOL_GPL(put_nfs_open_context);
969 
970 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
971 {
972 	__put_nfs_open_context(ctx, 1);
973 }
974 
975 /*
976  * Ensure that mmap has a recent RPC credential for use when writing out
977  * shared pages
978  */
979 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
980 {
981 	struct inode *inode = d_inode(ctx->dentry);
982 	struct nfs_inode *nfsi = NFS_I(inode);
983 
984 	spin_lock(&inode->i_lock);
985 	if (ctx->mode & FMODE_WRITE)
986 		list_add(&ctx->list, &nfsi->open_files);
987 	else
988 		list_add_tail(&ctx->list, &nfsi->open_files);
989 	spin_unlock(&inode->i_lock);
990 }
991 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
992 
993 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
994 {
995 	filp->private_data = get_nfs_open_context(ctx);
996 	if (list_empty(&ctx->list))
997 		nfs_inode_attach_open_context(ctx);
998 }
999 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1000 
1001 /*
1002  * Given an inode, search for an open context with the desired characteristics
1003  */
1004 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
1005 {
1006 	struct nfs_inode *nfsi = NFS_I(inode);
1007 	struct nfs_open_context *pos, *ctx = NULL;
1008 
1009 	spin_lock(&inode->i_lock);
1010 	list_for_each_entry(pos, &nfsi->open_files, list) {
1011 		if (cred != NULL && pos->cred != cred)
1012 			continue;
1013 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1014 			continue;
1015 		ctx = get_nfs_open_context(pos);
1016 		break;
1017 	}
1018 	spin_unlock(&inode->i_lock);
1019 	return ctx;
1020 }
1021 
1022 void nfs_file_clear_open_context(struct file *filp)
1023 {
1024 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1025 
1026 	if (ctx) {
1027 		struct inode *inode = d_inode(ctx->dentry);
1028 
1029 		/*
1030 		 * We fatal error on write before. Try to writeback
1031 		 * every page again.
1032 		 */
1033 		if (ctx->error < 0)
1034 			invalidate_inode_pages2(inode->i_mapping);
1035 		filp->private_data = NULL;
1036 		spin_lock(&inode->i_lock);
1037 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1038 		spin_unlock(&inode->i_lock);
1039 		put_nfs_open_context_sync(ctx);
1040 	}
1041 }
1042 
1043 /*
1044  * These allocate and release file read/write context information.
1045  */
1046 int nfs_open(struct inode *inode, struct file *filp)
1047 {
1048 	struct nfs_open_context *ctx;
1049 
1050 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1051 	if (IS_ERR(ctx))
1052 		return PTR_ERR(ctx);
1053 	nfs_file_set_open_context(filp, ctx);
1054 	put_nfs_open_context(ctx);
1055 	nfs_fscache_open_file(inode, filp);
1056 	return 0;
1057 }
1058 
1059 /*
1060  * This function is called whenever some part of NFS notices that
1061  * the cached attributes have to be refreshed.
1062  */
1063 int
1064 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1065 {
1066 	int		 status = -ESTALE;
1067 	struct nfs4_label *label = NULL;
1068 	struct nfs_fattr *fattr = NULL;
1069 	struct nfs_inode *nfsi = NFS_I(inode);
1070 
1071 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1072 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1073 
1074 	trace_nfs_revalidate_inode_enter(inode);
1075 
1076 	if (is_bad_inode(inode))
1077 		goto out;
1078 	if (NFS_STALE(inode))
1079 		goto out;
1080 
1081 	/* pNFS: Attributes aren't updated until we layoutcommit */
1082 	if (S_ISREG(inode->i_mode)) {
1083 		status = pnfs_sync_inode(inode, false);
1084 		if (status)
1085 			goto out;
1086 	}
1087 
1088 	status = -ENOMEM;
1089 	fattr = nfs_alloc_fattr();
1090 	if (fattr == NULL)
1091 		goto out;
1092 
1093 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1094 
1095 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1096 	if (IS_ERR(label)) {
1097 		status = PTR_ERR(label);
1098 		goto out;
1099 	}
1100 
1101 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
1102 	if (status != 0) {
1103 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1104 			 inode->i_sb->s_id,
1105 			 (unsigned long long)NFS_FILEID(inode), status);
1106 		if (status == -ESTALE) {
1107 			nfs_zap_caches(inode);
1108 			if (!S_ISDIR(inode->i_mode))
1109 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1110 		}
1111 		goto err_out;
1112 	}
1113 
1114 	status = nfs_refresh_inode(inode, fattr);
1115 	if (status) {
1116 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1117 			 inode->i_sb->s_id,
1118 			 (unsigned long long)NFS_FILEID(inode), status);
1119 		goto err_out;
1120 	}
1121 
1122 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1123 		nfs_zap_acl_cache(inode);
1124 
1125 	nfs_setsecurity(inode, fattr, label);
1126 
1127 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1128 		inode->i_sb->s_id,
1129 		(unsigned long long)NFS_FILEID(inode));
1130 
1131 err_out:
1132 	nfs4_label_free(label);
1133 out:
1134 	nfs_free_fattr(fattr);
1135 	trace_nfs_revalidate_inode_exit(inode, status);
1136 	return status;
1137 }
1138 
1139 int nfs_attribute_cache_expired(struct inode *inode)
1140 {
1141 	if (nfs_have_delegated_attributes(inode))
1142 		return 0;
1143 	return nfs_attribute_timeout(inode);
1144 }
1145 
1146 /**
1147  * nfs_revalidate_inode - Revalidate the inode attributes
1148  * @server - pointer to nfs_server struct
1149  * @inode - pointer to inode struct
1150  *
1151  * Updates inode attribute information by retrieving the data from the server.
1152  */
1153 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1154 {
1155 	if (!nfs_need_revalidate_inode(inode))
1156 		return NFS_STALE(inode) ? -ESTALE : 0;
1157 	return __nfs_revalidate_inode(server, inode);
1158 }
1159 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1160 
1161 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1162 {
1163 	struct nfs_inode *nfsi = NFS_I(inode);
1164 	int ret;
1165 
1166 	if (mapping->nrpages != 0) {
1167 		if (S_ISREG(inode->i_mode)) {
1168 			ret = nfs_sync_mapping(mapping);
1169 			if (ret < 0)
1170 				return ret;
1171 		}
1172 		ret = invalidate_inode_pages2(mapping);
1173 		if (ret < 0)
1174 			return ret;
1175 	}
1176 	if (S_ISDIR(inode->i_mode)) {
1177 		spin_lock(&inode->i_lock);
1178 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1179 		spin_unlock(&inode->i_lock);
1180 	}
1181 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1182 	nfs_fscache_wait_on_invalidate(inode);
1183 
1184 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1185 			inode->i_sb->s_id,
1186 			(unsigned long long)NFS_FILEID(inode));
1187 	return 0;
1188 }
1189 
1190 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1191 {
1192 	return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1193 		NFS_STALE(inode);
1194 }
1195 
1196 int nfs_revalidate_mapping_rcu(struct inode *inode)
1197 {
1198 	struct nfs_inode *nfsi = NFS_I(inode);
1199 	unsigned long *bitlock = &nfsi->flags;
1200 	int ret = 0;
1201 
1202 	if (IS_SWAPFILE(inode))
1203 		goto out;
1204 	if (nfs_mapping_need_revalidate_inode(inode)) {
1205 		ret = -ECHILD;
1206 		goto out;
1207 	}
1208 	spin_lock(&inode->i_lock);
1209 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1210 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1211 		ret = -ECHILD;
1212 	spin_unlock(&inode->i_lock);
1213 out:
1214 	return ret;
1215 }
1216 
1217 /**
1218  * nfs_revalidate_mapping - Revalidate the pagecache
1219  * @inode - pointer to host inode
1220  * @mapping - pointer to mapping
1221  */
1222 int nfs_revalidate_mapping(struct inode *inode,
1223 		struct address_space *mapping)
1224 {
1225 	struct nfs_inode *nfsi = NFS_I(inode);
1226 	unsigned long *bitlock = &nfsi->flags;
1227 	int ret = 0;
1228 
1229 	/* swapfiles are not supposed to be shared. */
1230 	if (IS_SWAPFILE(inode))
1231 		goto out;
1232 
1233 	if (nfs_mapping_need_revalidate_inode(inode)) {
1234 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1235 		if (ret < 0)
1236 			goto out;
1237 	}
1238 
1239 	/*
1240 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1241 	 * invalidations that come in while we're shooting down the mappings
1242 	 * are respected. But, that leaves a race window where one revalidator
1243 	 * can clear the flag, and then another checks it before the mapping
1244 	 * gets invalidated. Fix that by serializing access to this part of
1245 	 * the function.
1246 	 *
1247 	 * At the same time, we need to allow other tasks to see whether we
1248 	 * might be in the middle of invalidating the pages, so we only set
1249 	 * the bit lock here if it looks like we're going to be doing that.
1250 	 */
1251 	for (;;) {
1252 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1253 					 nfs_wait_bit_killable, TASK_KILLABLE);
1254 		if (ret)
1255 			goto out;
1256 		spin_lock(&inode->i_lock);
1257 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1258 			spin_unlock(&inode->i_lock);
1259 			continue;
1260 		}
1261 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1262 			break;
1263 		spin_unlock(&inode->i_lock);
1264 		goto out;
1265 	}
1266 
1267 	set_bit(NFS_INO_INVALIDATING, bitlock);
1268 	smp_wmb();
1269 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1270 	spin_unlock(&inode->i_lock);
1271 	trace_nfs_invalidate_mapping_enter(inode);
1272 	ret = nfs_invalidate_mapping(inode, mapping);
1273 	trace_nfs_invalidate_mapping_exit(inode, ret);
1274 
1275 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1276 	smp_mb__after_atomic();
1277 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1278 out:
1279 	return ret;
1280 }
1281 
1282 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1283 {
1284 	struct inode *inode = &nfsi->vfs_inode;
1285 
1286 	assert_spin_locked(&inode->i_lock);
1287 
1288 	if (!S_ISREG(inode->i_mode))
1289 		return false;
1290 	if (list_empty(&nfsi->open_files))
1291 		return false;
1292 	/* Note: This relies on nfsi->open_files being ordered with writers
1293 	 *       being placed at the head of the list.
1294 	 *       See nfs_inode_attach_open_context()
1295 	 */
1296 	return (list_first_entry(&nfsi->open_files,
1297 			struct nfs_open_context,
1298 			list)->mode & FMODE_WRITE) == FMODE_WRITE;
1299 }
1300 
1301 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1302 {
1303 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1304 }
1305 
1306 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1307 {
1308 	unsigned long ret = 0;
1309 
1310 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1311 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1312 			&& inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1313 		inode_set_iversion_raw(inode, fattr->change_attr);
1314 		if (S_ISDIR(inode->i_mode))
1315 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1316 		ret |= NFS_INO_INVALID_ATTR;
1317 	}
1318 	/* If we have atomic WCC data, we may update some attributes */
1319 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1320 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1321 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1322 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1323 		ret |= NFS_INO_INVALID_ATTR;
1324 	}
1325 
1326 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1327 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1328 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1329 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1330 		if (S_ISDIR(inode->i_mode))
1331 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1332 		ret |= NFS_INO_INVALID_ATTR;
1333 	}
1334 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1335 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1336 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1337 			&& !nfs_have_writebacks(inode)) {
1338 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1339 		ret |= NFS_INO_INVALID_ATTR;
1340 	}
1341 
1342 	return ret;
1343 }
1344 
1345 /**
1346  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1347  * @inode - pointer to inode
1348  * @fattr - updated attributes
1349  *
1350  * Verifies the attribute cache. If we have just changed the attributes,
1351  * so that fattr carries weak cache consistency data, then it may
1352  * also update the ctime/mtime/change_attribute.
1353  */
1354 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1355 {
1356 	struct nfs_inode *nfsi = NFS_I(inode);
1357 	loff_t cur_size, new_isize;
1358 	unsigned long invalid = 0;
1359 
1360 
1361 	if (nfs_have_delegated_attributes(inode))
1362 		return 0;
1363 	/* Has the inode gone and changed behind our back? */
1364 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1365 		return -ESTALE;
1366 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1367 		return -ESTALE;
1368 
1369 	if (!nfs_file_has_buffered_writers(nfsi)) {
1370 		/* Verify a few of the more important attributes */
1371 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1372 			invalid |= NFS_INO_INVALID_ATTR | NFS_INO_REVAL_PAGECACHE;
1373 
1374 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1375 			invalid |= NFS_INO_INVALID_ATTR;
1376 
1377 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&inode->i_ctime, &fattr->ctime))
1378 			invalid |= NFS_INO_INVALID_ATTR;
1379 
1380 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1381 			cur_size = i_size_read(inode);
1382 			new_isize = nfs_size_to_loff_t(fattr->size);
1383 			if (cur_size != new_isize)
1384 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1385 		}
1386 	}
1387 
1388 	/* Have any file permissions changed? */
1389 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1390 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1391 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1392 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1393 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1394 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1395 
1396 	/* Has the link count changed? */
1397 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1398 		invalid |= NFS_INO_INVALID_ATTR;
1399 
1400 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1401 		invalid |= NFS_INO_INVALID_ATIME;
1402 
1403 	if (invalid != 0)
1404 		nfs_set_cache_invalid(inode, invalid | NFS_INO_REVAL_FORCED);
1405 
1406 	nfsi->read_cache_jiffies = fattr->time_start;
1407 	return 0;
1408 }
1409 
1410 static atomic_long_t nfs_attr_generation_counter;
1411 
1412 static unsigned long nfs_read_attr_generation_counter(void)
1413 {
1414 	return atomic_long_read(&nfs_attr_generation_counter);
1415 }
1416 
1417 unsigned long nfs_inc_attr_generation_counter(void)
1418 {
1419 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1420 }
1421 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1422 
1423 void nfs_fattr_init(struct nfs_fattr *fattr)
1424 {
1425 	fattr->valid = 0;
1426 	fattr->time_start = jiffies;
1427 	fattr->gencount = nfs_inc_attr_generation_counter();
1428 	fattr->owner_name = NULL;
1429 	fattr->group_name = NULL;
1430 }
1431 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1432 
1433 /**
1434  * nfs_fattr_set_barrier
1435  * @fattr: attributes
1436  *
1437  * Used to set a barrier after an attribute was updated. This
1438  * barrier ensures that older attributes from RPC calls that may
1439  * have raced with our update cannot clobber these new values.
1440  * Note that you are still responsible for ensuring that other
1441  * operations which change the attribute on the server do not
1442  * collide.
1443  */
1444 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1445 {
1446 	fattr->gencount = nfs_inc_attr_generation_counter();
1447 }
1448 
1449 struct nfs_fattr *nfs_alloc_fattr(void)
1450 {
1451 	struct nfs_fattr *fattr;
1452 
1453 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1454 	if (fattr != NULL)
1455 		nfs_fattr_init(fattr);
1456 	return fattr;
1457 }
1458 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1459 
1460 struct nfs_fh *nfs_alloc_fhandle(void)
1461 {
1462 	struct nfs_fh *fh;
1463 
1464 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1465 	if (fh != NULL)
1466 		fh->size = 0;
1467 	return fh;
1468 }
1469 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1470 
1471 #ifdef NFS_DEBUG
1472 /*
1473  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1474  *                             in the same way that wireshark does
1475  *
1476  * @fh: file handle
1477  *
1478  * For debugging only.
1479  */
1480 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1481 {
1482 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1483 	 * not on the result */
1484 	return nfs_fhandle_hash(fh);
1485 }
1486 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1487 
1488 /*
1489  * _nfs_display_fhandle - display an NFS file handle on the console
1490  *
1491  * @fh: file handle to display
1492  * @caption: display caption
1493  *
1494  * For debugging only.
1495  */
1496 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1497 {
1498 	unsigned short i;
1499 
1500 	if (fh == NULL || fh->size == 0) {
1501 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1502 		return;
1503 	}
1504 
1505 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1506 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1507 	for (i = 0; i < fh->size; i += 16) {
1508 		__be32 *pos = (__be32 *)&fh->data[i];
1509 
1510 		switch ((fh->size - i - 1) >> 2) {
1511 		case 0:
1512 			printk(KERN_DEFAULT " %08x\n",
1513 				be32_to_cpup(pos));
1514 			break;
1515 		case 1:
1516 			printk(KERN_DEFAULT " %08x %08x\n",
1517 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1518 			break;
1519 		case 2:
1520 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1521 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1522 				be32_to_cpup(pos + 2));
1523 			break;
1524 		default:
1525 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1526 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1527 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1528 		}
1529 	}
1530 }
1531 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1532 #endif
1533 
1534 /**
1535  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1536  * @inode - pointer to inode
1537  * @fattr - attributes
1538  *
1539  * Attempt to divine whether or not an RPC call reply carrying stale
1540  * attributes got scheduled after another call carrying updated ones.
1541  *
1542  * To do so, the function first assumes that a more recent ctime means
1543  * that the attributes in fattr are newer, however it also attempt to
1544  * catch the case where ctime either didn't change, or went backwards
1545  * (if someone reset the clock on the server) by looking at whether
1546  * or not this RPC call was started after the inode was last updated.
1547  * Note also the check for wraparound of 'attr_gencount'
1548  *
1549  * The function returns 'true' if it thinks the attributes in 'fattr' are
1550  * more recent than the ones cached in the inode.
1551  *
1552  */
1553 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1554 {
1555 	const struct nfs_inode *nfsi = NFS_I(inode);
1556 
1557 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1558 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1559 }
1560 
1561 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1562 {
1563 	int ret;
1564 
1565 	trace_nfs_refresh_inode_enter(inode);
1566 
1567 	if (nfs_inode_attrs_need_update(inode, fattr))
1568 		ret = nfs_update_inode(inode, fattr);
1569 	else
1570 		ret = nfs_check_inode_attributes(inode, fattr);
1571 
1572 	trace_nfs_refresh_inode_exit(inode, ret);
1573 	return ret;
1574 }
1575 
1576 /**
1577  * nfs_refresh_inode - try to update the inode attribute cache
1578  * @inode - pointer to inode
1579  * @fattr - updated attributes
1580  *
1581  * Check that an RPC call that returned attributes has not overlapped with
1582  * other recent updates of the inode metadata, then decide whether it is
1583  * safe to do a full update of the inode attributes, or whether just to
1584  * call nfs_check_inode_attributes.
1585  */
1586 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1587 {
1588 	int status;
1589 
1590 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1591 		return 0;
1592 	spin_lock(&inode->i_lock);
1593 	status = nfs_refresh_inode_locked(inode, fattr);
1594 	spin_unlock(&inode->i_lock);
1595 
1596 	return status;
1597 }
1598 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1599 
1600 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1601 {
1602 	unsigned long invalid = NFS_INO_INVALID_ATTR;
1603 
1604 	if (S_ISDIR(inode->i_mode))
1605 		invalid |= NFS_INO_INVALID_DATA;
1606 	nfs_set_cache_invalid(inode, invalid);
1607 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1608 		return 0;
1609 	return nfs_refresh_inode_locked(inode, fattr);
1610 }
1611 
1612 /**
1613  * nfs_post_op_update_inode - try to update the inode attribute cache
1614  * @inode - pointer to inode
1615  * @fattr - updated attributes
1616  *
1617  * After an operation that has changed the inode metadata, mark the
1618  * attribute cache as being invalid, then try to update it.
1619  *
1620  * NB: if the server didn't return any post op attributes, this
1621  * function will force the retrieval of attributes before the next
1622  * NFS request.  Thus it should be used only for operations that
1623  * are expected to change one or more attributes, to avoid
1624  * unnecessary NFS requests and trips through nfs_update_inode().
1625  */
1626 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1627 {
1628 	int status;
1629 
1630 	spin_lock(&inode->i_lock);
1631 	nfs_fattr_set_barrier(fattr);
1632 	status = nfs_post_op_update_inode_locked(inode, fattr);
1633 	spin_unlock(&inode->i_lock);
1634 
1635 	return status;
1636 }
1637 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1638 
1639 /**
1640  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1641  * @inode - pointer to inode
1642  * @fattr - updated attributes
1643  *
1644  * After an operation that has changed the inode metadata, mark the
1645  * attribute cache as being invalid, then try to update it. Fake up
1646  * weak cache consistency data, if none exist.
1647  *
1648  * This function is mainly designed to be used by the ->write_done() functions.
1649  */
1650 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1651 {
1652 	int status;
1653 
1654 	/* Don't do a WCC update if these attributes are already stale */
1655 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1656 			!nfs_inode_attrs_need_update(inode, fattr)) {
1657 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1658 				| NFS_ATTR_FATTR_PRESIZE
1659 				| NFS_ATTR_FATTR_PREMTIME
1660 				| NFS_ATTR_FATTR_PRECTIME);
1661 		goto out_noforce;
1662 	}
1663 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1664 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1665 		fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1666 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1667 	}
1668 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1669 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1670 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1671 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1672 	}
1673 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1674 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1675 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1676 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1677 	}
1678 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1679 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1680 		fattr->pre_size = i_size_read(inode);
1681 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1682 	}
1683 out_noforce:
1684 	status = nfs_post_op_update_inode_locked(inode, fattr);
1685 	return status;
1686 }
1687 
1688 /**
1689  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1690  * @inode - pointer to inode
1691  * @fattr - updated attributes
1692  *
1693  * After an operation that has changed the inode metadata, mark the
1694  * attribute cache as being invalid, then try to update it. Fake up
1695  * weak cache consistency data, if none exist.
1696  *
1697  * This function is mainly designed to be used by the ->write_done() functions.
1698  */
1699 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1700 {
1701 	int status;
1702 
1703 	spin_lock(&inode->i_lock);
1704 	nfs_fattr_set_barrier(fattr);
1705 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1706 	spin_unlock(&inode->i_lock);
1707 	return status;
1708 }
1709 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1710 
1711 
1712 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1713 				    struct nfs_fattr *fattr)
1714 {
1715 	bool ret1 = true, ret2 = true;
1716 
1717 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1718 		ret1 = (nfsi->fileid == fattr->fileid);
1719 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1720 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1721 	return ret1 || ret2;
1722 }
1723 
1724 /*
1725  * Many nfs protocol calls return the new file attributes after
1726  * an operation.  Here we update the inode to reflect the state
1727  * of the server's inode.
1728  *
1729  * This is a bit tricky because we have to make sure all dirty pages
1730  * have been sent off to the server before calling invalidate_inode_pages.
1731  * To make sure no other process adds more write requests while we try
1732  * our best to flush them, we make them sleep during the attribute refresh.
1733  *
1734  * A very similar scenario holds for the dir cache.
1735  */
1736 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1737 {
1738 	struct nfs_server *server;
1739 	struct nfs_inode *nfsi = NFS_I(inode);
1740 	loff_t cur_isize, new_isize;
1741 	unsigned long invalid = 0;
1742 	unsigned long now = jiffies;
1743 	unsigned long save_cache_validity;
1744 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
1745 	bool cache_revalidated = true;
1746 
1747 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1748 			__func__, inode->i_sb->s_id, inode->i_ino,
1749 			nfs_display_fhandle_hash(NFS_FH(inode)),
1750 			atomic_read(&inode->i_count), fattr->valid);
1751 
1752 	if (!nfs_fileid_valid(nfsi, fattr)) {
1753 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1754 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1755 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1756 			inode->i_sb->s_id, (long long)nfsi->fileid,
1757 			(long long)fattr->fileid);
1758 		goto out_err;
1759 	}
1760 
1761 	/*
1762 	 * Make sure the inode's type hasn't changed.
1763 	 */
1764 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1765 		/*
1766 		* Big trouble! The inode has become a different object.
1767 		*/
1768 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1769 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1770 		goto out_err;
1771 	}
1772 
1773 	server = NFS_SERVER(inode);
1774 	/* Update the fsid? */
1775 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1776 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1777 			!IS_AUTOMOUNT(inode))
1778 		server->fsid = fattr->fsid;
1779 
1780 	/*
1781 	 * Update the read time so we don't revalidate too often.
1782 	 */
1783 	nfsi->read_cache_jiffies = fattr->time_start;
1784 
1785 	save_cache_validity = nfsi->cache_validity;
1786 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1787 			| NFS_INO_INVALID_ATIME
1788 			| NFS_INO_REVAL_FORCED
1789 			| NFS_INO_REVAL_PAGECACHE);
1790 
1791 	/* Do atomic weak cache consistency updates */
1792 	invalid |= nfs_wcc_update_inode(inode, fattr);
1793 
1794 	if (pnfs_layoutcommit_outstanding(inode)) {
1795 		nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1796 		cache_revalidated = false;
1797 	}
1798 
1799 	/* More cache consistency checks */
1800 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1801 		if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1802 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1803 					inode->i_sb->s_id, inode->i_ino);
1804 			/* Could it be a race with writeback? */
1805 			if (!have_writers) {
1806 				invalid |= NFS_INO_INVALID_ATTR
1807 					| NFS_INO_INVALID_DATA
1808 					| NFS_INO_INVALID_ACCESS
1809 					| NFS_INO_INVALID_ACL;
1810 				if (S_ISDIR(inode->i_mode))
1811 					nfs_force_lookup_revalidate(inode);
1812 			}
1813 			inode_set_iversion_raw(inode, fattr->change_attr);
1814 		}
1815 	} else {
1816 		nfsi->cache_validity |= save_cache_validity;
1817 		cache_revalidated = false;
1818 	}
1819 
1820 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1821 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1822 	} else if (server->caps & NFS_CAP_MTIME) {
1823 		nfsi->cache_validity |= save_cache_validity &
1824 				(NFS_INO_INVALID_ATTR
1825 				| NFS_INO_REVAL_FORCED);
1826 		cache_revalidated = false;
1827 	}
1828 
1829 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1830 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1831 	} else if (server->caps & NFS_CAP_CTIME) {
1832 		nfsi->cache_validity |= save_cache_validity &
1833 				(NFS_INO_INVALID_ATTR
1834 				| NFS_INO_REVAL_FORCED);
1835 		cache_revalidated = false;
1836 	}
1837 
1838 	/* Check if our cached file size is stale */
1839 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1840 		new_isize = nfs_size_to_loff_t(fattr->size);
1841 		cur_isize = i_size_read(inode);
1842 		if (new_isize != cur_isize) {
1843 			/* Do we perhaps have any outstanding writes, or has
1844 			 * the file grown beyond our last write? */
1845 			if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1846 				i_size_write(inode, new_isize);
1847 				if (!have_writers)
1848 					invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1849 			}
1850 			dprintk("NFS: isize change on server for file %s/%ld "
1851 					"(%Ld to %Ld)\n",
1852 					inode->i_sb->s_id,
1853 					inode->i_ino,
1854 					(long long)cur_isize,
1855 					(long long)new_isize);
1856 		}
1857 	} else {
1858 		nfsi->cache_validity |= save_cache_validity &
1859 				(NFS_INO_INVALID_ATTR
1860 				| NFS_INO_REVAL_PAGECACHE
1861 				| NFS_INO_REVAL_FORCED);
1862 		cache_revalidated = false;
1863 	}
1864 
1865 
1866 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1867 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1868 	else if (server->caps & NFS_CAP_ATIME) {
1869 		nfsi->cache_validity |= save_cache_validity &
1870 				(NFS_INO_INVALID_ATIME
1871 				| NFS_INO_REVAL_FORCED);
1872 		cache_revalidated = false;
1873 	}
1874 
1875 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1876 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1877 			umode_t newmode = inode->i_mode & S_IFMT;
1878 			newmode |= fattr->mode & S_IALLUGO;
1879 			inode->i_mode = newmode;
1880 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1881 		}
1882 	} else if (server->caps & NFS_CAP_MODE) {
1883 		nfsi->cache_validity |= save_cache_validity &
1884 				(NFS_INO_INVALID_ATTR
1885 				| NFS_INO_INVALID_ACCESS
1886 				| NFS_INO_INVALID_ACL
1887 				| NFS_INO_REVAL_FORCED);
1888 		cache_revalidated = false;
1889 	}
1890 
1891 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1892 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1893 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1894 			inode->i_uid = fattr->uid;
1895 		}
1896 	} else if (server->caps & NFS_CAP_OWNER) {
1897 		nfsi->cache_validity |= save_cache_validity &
1898 				(NFS_INO_INVALID_ATTR
1899 				| NFS_INO_INVALID_ACCESS
1900 				| NFS_INO_INVALID_ACL
1901 				| NFS_INO_REVAL_FORCED);
1902 		cache_revalidated = false;
1903 	}
1904 
1905 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1906 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1907 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1908 			inode->i_gid = fattr->gid;
1909 		}
1910 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
1911 		nfsi->cache_validity |= save_cache_validity &
1912 				(NFS_INO_INVALID_ATTR
1913 				| NFS_INO_INVALID_ACCESS
1914 				| NFS_INO_INVALID_ACL
1915 				| NFS_INO_REVAL_FORCED);
1916 		cache_revalidated = false;
1917 	}
1918 
1919 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1920 		if (inode->i_nlink != fattr->nlink) {
1921 			invalid |= NFS_INO_INVALID_ATTR;
1922 			if (S_ISDIR(inode->i_mode))
1923 				invalid |= NFS_INO_INVALID_DATA;
1924 			set_nlink(inode, fattr->nlink);
1925 		}
1926 	} else if (server->caps & NFS_CAP_NLINK) {
1927 		nfsi->cache_validity |= save_cache_validity &
1928 				(NFS_INO_INVALID_ATTR
1929 				| NFS_INO_REVAL_FORCED);
1930 		cache_revalidated = false;
1931 	}
1932 
1933 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1934 		/*
1935 		 * report the blocks in 512byte units
1936 		 */
1937 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1938 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1939 		inode->i_blocks = fattr->du.nfs2.blocks;
1940 	else
1941 		cache_revalidated = false;
1942 
1943 	/* Update attrtimeo value if we're out of the unstable period */
1944 	if (invalid & NFS_INO_INVALID_ATTR) {
1945 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1946 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1947 		nfsi->attrtimeo_timestamp = now;
1948 		/* Set barrier to be more recent than all outstanding updates */
1949 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1950 	} else {
1951 		if (cache_revalidated) {
1952 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1953 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1954 				nfsi->attrtimeo <<= 1;
1955 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1956 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1957 			}
1958 			nfsi->attrtimeo_timestamp = now;
1959 		}
1960 		/* Set the barrier to be more recent than this fattr */
1961 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1962 			nfsi->attr_gencount = fattr->gencount;
1963 	}
1964 
1965 	/* Don't declare attrcache up to date if there were no attrs! */
1966 	if (cache_revalidated)
1967 		invalid &= ~NFS_INO_INVALID_ATTR;
1968 
1969 	/* Don't invalidate the data if we were to blame */
1970 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1971 				|| S_ISLNK(inode->i_mode)))
1972 		invalid &= ~NFS_INO_INVALID_DATA;
1973 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1974 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1975 		nfs_set_cache_invalid(inode, invalid);
1976 
1977 	return 0;
1978  out_err:
1979 	/*
1980 	 * No need to worry about unhashing the dentry, as the
1981 	 * lookup validation will know that the inode is bad.
1982 	 * (But we fall through to invalidate the caches.)
1983 	 */
1984 	nfs_invalidate_inode(inode);
1985 	return -ESTALE;
1986 }
1987 
1988 struct inode *nfs_alloc_inode(struct super_block *sb)
1989 {
1990 	struct nfs_inode *nfsi;
1991 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1992 	if (!nfsi)
1993 		return NULL;
1994 	nfsi->flags = 0UL;
1995 	nfsi->cache_validity = 0UL;
1996 #if IS_ENABLED(CONFIG_NFS_V4)
1997 	nfsi->nfs4_acl = NULL;
1998 #endif /* CONFIG_NFS_V4 */
1999 	return &nfsi->vfs_inode;
2000 }
2001 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2002 
2003 static void nfs_i_callback(struct rcu_head *head)
2004 {
2005 	struct inode *inode = container_of(head, struct inode, i_rcu);
2006 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2007 }
2008 
2009 void nfs_destroy_inode(struct inode *inode)
2010 {
2011 	call_rcu(&inode->i_rcu, nfs_i_callback);
2012 }
2013 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
2014 
2015 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2016 {
2017 #if IS_ENABLED(CONFIG_NFS_V4)
2018 	INIT_LIST_HEAD(&nfsi->open_states);
2019 	nfsi->delegation = NULL;
2020 	init_rwsem(&nfsi->rwsem);
2021 	nfsi->layout = NULL;
2022 #endif
2023 }
2024 
2025 static void init_once(void *foo)
2026 {
2027 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2028 
2029 	inode_init_once(&nfsi->vfs_inode);
2030 	INIT_LIST_HEAD(&nfsi->open_files);
2031 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2032 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2033 	INIT_LIST_HEAD(&nfsi->commit_info.list);
2034 	atomic_long_set(&nfsi->nrequests, 0);
2035 	atomic_long_set(&nfsi->commit_info.ncommit, 0);
2036 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
2037 	init_rwsem(&nfsi->rmdir_sem);
2038 	mutex_init(&nfsi->commit_mutex);
2039 	nfs4_init_once(nfsi);
2040 }
2041 
2042 static int __init nfs_init_inodecache(void)
2043 {
2044 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2045 					     sizeof(struct nfs_inode),
2046 					     0, (SLAB_RECLAIM_ACCOUNT|
2047 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2048 					     init_once);
2049 	if (nfs_inode_cachep == NULL)
2050 		return -ENOMEM;
2051 
2052 	return 0;
2053 }
2054 
2055 static void nfs_destroy_inodecache(void)
2056 {
2057 	/*
2058 	 * Make sure all delayed rcu free inodes are flushed before we
2059 	 * destroy cache.
2060 	 */
2061 	rcu_barrier();
2062 	kmem_cache_destroy(nfs_inode_cachep);
2063 }
2064 
2065 struct workqueue_struct *nfsiod_workqueue;
2066 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2067 
2068 /*
2069  * start up the nfsiod workqueue
2070  */
2071 static int nfsiod_start(void)
2072 {
2073 	struct workqueue_struct *wq;
2074 	dprintk("RPC:       creating workqueue nfsiod\n");
2075 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2076 	if (wq == NULL)
2077 		return -ENOMEM;
2078 	nfsiod_workqueue = wq;
2079 	return 0;
2080 }
2081 
2082 /*
2083  * Destroy the nfsiod workqueue
2084  */
2085 static void nfsiod_stop(void)
2086 {
2087 	struct workqueue_struct *wq;
2088 
2089 	wq = nfsiod_workqueue;
2090 	if (wq == NULL)
2091 		return;
2092 	nfsiod_workqueue = NULL;
2093 	destroy_workqueue(wq);
2094 }
2095 
2096 unsigned int nfs_net_id;
2097 EXPORT_SYMBOL_GPL(nfs_net_id);
2098 
2099 static int nfs_net_init(struct net *net)
2100 {
2101 	nfs_clients_init(net);
2102 	return nfs_fs_proc_net_init(net);
2103 }
2104 
2105 static void nfs_net_exit(struct net *net)
2106 {
2107 	struct nfs_net *nn = net_generic(net, nfs_net_id);
2108 
2109 	nfs_fs_proc_net_exit(net);
2110 	nfs_cleanup_cb_ident_idr(net);
2111 	WARN_ON_ONCE(!list_empty(&nn->nfs_client_list));
2112 	WARN_ON_ONCE(!list_empty(&nn->nfs_volume_list));
2113 }
2114 
2115 static struct pernet_operations nfs_net_ops = {
2116 	.init = nfs_net_init,
2117 	.exit = nfs_net_exit,
2118 	.id   = &nfs_net_id,
2119 	.size = sizeof(struct nfs_net),
2120 };
2121 
2122 /*
2123  * Initialize NFS
2124  */
2125 static int __init init_nfs_fs(void)
2126 {
2127 	int err;
2128 
2129 	err = register_pernet_subsys(&nfs_net_ops);
2130 	if (err < 0)
2131 		goto out9;
2132 
2133 	err = nfs_fscache_register();
2134 	if (err < 0)
2135 		goto out8;
2136 
2137 	err = nfsiod_start();
2138 	if (err)
2139 		goto out7;
2140 
2141 	err = nfs_fs_proc_init();
2142 	if (err)
2143 		goto out6;
2144 
2145 	err = nfs_init_nfspagecache();
2146 	if (err)
2147 		goto out5;
2148 
2149 	err = nfs_init_inodecache();
2150 	if (err)
2151 		goto out4;
2152 
2153 	err = nfs_init_readpagecache();
2154 	if (err)
2155 		goto out3;
2156 
2157 	err = nfs_init_writepagecache();
2158 	if (err)
2159 		goto out2;
2160 
2161 	err = nfs_init_directcache();
2162 	if (err)
2163 		goto out1;
2164 
2165 	rpc_proc_register(&init_net, &nfs_rpcstat);
2166 
2167 	err = register_nfs_fs();
2168 	if (err)
2169 		goto out0;
2170 
2171 	return 0;
2172 out0:
2173 	rpc_proc_unregister(&init_net, "nfs");
2174 	nfs_destroy_directcache();
2175 out1:
2176 	nfs_destroy_writepagecache();
2177 out2:
2178 	nfs_destroy_readpagecache();
2179 out3:
2180 	nfs_destroy_inodecache();
2181 out4:
2182 	nfs_destroy_nfspagecache();
2183 out5:
2184 	nfs_fs_proc_exit();
2185 out6:
2186 	nfsiod_stop();
2187 out7:
2188 	nfs_fscache_unregister();
2189 out8:
2190 	unregister_pernet_subsys(&nfs_net_ops);
2191 out9:
2192 	return err;
2193 }
2194 
2195 static void __exit exit_nfs_fs(void)
2196 {
2197 	nfs_destroy_directcache();
2198 	nfs_destroy_writepagecache();
2199 	nfs_destroy_readpagecache();
2200 	nfs_destroy_inodecache();
2201 	nfs_destroy_nfspagecache();
2202 	nfs_fscache_unregister();
2203 	unregister_pernet_subsys(&nfs_net_ops);
2204 	rpc_proc_unregister(&init_net, "nfs");
2205 	unregister_nfs_fs();
2206 	nfs_fs_proc_exit();
2207 	nfsiod_stop();
2208 }
2209 
2210 /* Not quite true; I just maintain it */
2211 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2212 MODULE_LICENSE("GPL");
2213 module_param(enable_ino64, bool, 0644);
2214 
2215 module_init(init_nfs_fs)
2216 module_exit(exit_nfs_fs)
2217