xref: /linux/fs/nfs/inode.c (revision 32786fdc9506aeba98278c1844d4bfb766863832)
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.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 
42 #include <asm/uaccess.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 int nfs_wait_atomic_killable(atomic_t *p)
89 {
90 	return nfs_wait_killable(TASK_KILLABLE);
91 }
92 
93 /**
94  * nfs_compat_user_ino64 - returns the user-visible inode number
95  * @fileid: 64-bit fileid
96  *
97  * This function returns a 32-bit inode number if the boot parameter
98  * nfs.enable_ino64 is zero.
99  */
100 u64 nfs_compat_user_ino64(u64 fileid)
101 {
102 #ifdef CONFIG_COMPAT
103 	compat_ulong_t ino;
104 #else
105 	unsigned long ino;
106 #endif
107 
108 	if (enable_ino64)
109 		return fileid;
110 	ino = fileid;
111 	if (sizeof(ino) < sizeof(fileid))
112 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
113 	return ino;
114 }
115 
116 int nfs_drop_inode(struct inode *inode)
117 {
118 	return NFS_STALE(inode) || generic_drop_inode(inode);
119 }
120 EXPORT_SYMBOL_GPL(nfs_drop_inode);
121 
122 void nfs_clear_inode(struct inode *inode)
123 {
124 	/*
125 	 * The following should never happen...
126 	 */
127 	WARN_ON_ONCE(nfs_have_writebacks(inode));
128 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
129 	nfs_zap_acl_cache(inode);
130 	nfs_access_zap_cache(inode);
131 	nfs_fscache_clear_inode(inode);
132 }
133 EXPORT_SYMBOL_GPL(nfs_clear_inode);
134 
135 void nfs_evict_inode(struct inode *inode)
136 {
137 	truncate_inode_pages_final(&inode->i_data);
138 	clear_inode(inode);
139 	nfs_clear_inode(inode);
140 }
141 
142 int nfs_sync_inode(struct inode *inode)
143 {
144 	inode_dio_wait(inode);
145 	return nfs_wb_all(inode);
146 }
147 EXPORT_SYMBOL_GPL(nfs_sync_inode);
148 
149 /**
150  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
151  */
152 int nfs_sync_mapping(struct address_space *mapping)
153 {
154 	int ret = 0;
155 
156 	if (mapping->nrpages != 0) {
157 		unmap_mapping_range(mapping, 0, 0, 0);
158 		ret = nfs_wb_all(mapping->host);
159 	}
160 	return ret;
161 }
162 
163 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
164 {
165 	struct nfs_inode *nfsi = NFS_I(inode);
166 
167 	if (inode->i_mapping->nrpages == 0)
168 		flags &= ~NFS_INO_INVALID_DATA;
169 	nfsi->cache_validity |= flags;
170 	if (flags & NFS_INO_INVALID_DATA)
171 		nfs_fscache_invalidate(inode);
172 }
173 
174 /*
175  * Invalidate the local caches
176  */
177 static void nfs_zap_caches_locked(struct inode *inode)
178 {
179 	struct nfs_inode *nfsi = NFS_I(inode);
180 	int mode = inode->i_mode;
181 
182 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
183 
184 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
185 	nfsi->attrtimeo_timestamp = jiffies;
186 
187 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
188 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
189 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
190 					| NFS_INO_INVALID_DATA
191 					| NFS_INO_INVALID_ACCESS
192 					| NFS_INO_INVALID_ACL
193 					| NFS_INO_REVAL_PAGECACHE);
194 	} else
195 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
196 					| NFS_INO_INVALID_ACCESS
197 					| NFS_INO_INVALID_ACL
198 					| NFS_INO_REVAL_PAGECACHE);
199 	nfs_zap_label_cache_locked(nfsi);
200 }
201 
202 void nfs_zap_caches(struct inode *inode)
203 {
204 	spin_lock(&inode->i_lock);
205 	nfs_zap_caches_locked(inode);
206 	spin_unlock(&inode->i_lock);
207 }
208 
209 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
210 {
211 	if (mapping->nrpages != 0) {
212 		spin_lock(&inode->i_lock);
213 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
214 		spin_unlock(&inode->i_lock);
215 	}
216 }
217 
218 void nfs_zap_acl_cache(struct inode *inode)
219 {
220 	void (*clear_acl_cache)(struct inode *);
221 
222 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
223 	if (clear_acl_cache != NULL)
224 		clear_acl_cache(inode);
225 	spin_lock(&inode->i_lock);
226 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
227 	spin_unlock(&inode->i_lock);
228 }
229 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
230 
231 void nfs_invalidate_atime(struct inode *inode)
232 {
233 	spin_lock(&inode->i_lock);
234 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
235 	spin_unlock(&inode->i_lock);
236 }
237 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
238 
239 /*
240  * Invalidate, but do not unhash, the inode.
241  * NB: must be called with inode->i_lock held!
242  */
243 static void nfs_invalidate_inode(struct inode *inode)
244 {
245 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
246 	nfs_zap_caches_locked(inode);
247 }
248 
249 struct nfs_find_desc {
250 	struct nfs_fh		*fh;
251 	struct nfs_fattr	*fattr;
252 };
253 
254 /*
255  * In NFSv3 we can have 64bit inode numbers. In order to support
256  * this, and re-exported directories (also seen in NFSv2)
257  * we are forced to allow 2 different inodes to have the same
258  * i_ino.
259  */
260 static int
261 nfs_find_actor(struct inode *inode, void *opaque)
262 {
263 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
264 	struct nfs_fh		*fh = desc->fh;
265 	struct nfs_fattr	*fattr = desc->fattr;
266 
267 	if (NFS_FILEID(inode) != fattr->fileid)
268 		return 0;
269 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
270 		return 0;
271 	if (nfs_compare_fh(NFS_FH(inode), fh))
272 		return 0;
273 	if (is_bad_inode(inode) || NFS_STALE(inode))
274 		return 0;
275 	return 1;
276 }
277 
278 static int
279 nfs_init_locked(struct inode *inode, void *opaque)
280 {
281 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
282 	struct nfs_fattr	*fattr = desc->fattr;
283 
284 	set_nfs_fileid(inode, fattr->fileid);
285 	inode->i_mode = fattr->mode;
286 	nfs_copy_fh(NFS_FH(inode), desc->fh);
287 	return 0;
288 }
289 
290 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
291 static void nfs_clear_label_invalid(struct inode *inode)
292 {
293 	spin_lock(&inode->i_lock);
294 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
295 	spin_unlock(&inode->i_lock);
296 }
297 
298 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
299 					struct nfs4_label *label)
300 {
301 	int error;
302 
303 	if (label == NULL)
304 		return;
305 
306 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
307 		error = security_inode_notifysecctx(inode, label->label,
308 				label->len);
309 		if (error)
310 			printk(KERN_ERR "%s() %s %d "
311 					"security_inode_notifysecctx() %d\n",
312 					__func__,
313 					(char *)label->label,
314 					label->len, error);
315 		nfs_clear_label_invalid(inode);
316 	}
317 }
318 
319 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
320 {
321 	struct nfs4_label *label = NULL;
322 	int minor_version = server->nfs_client->cl_minorversion;
323 
324 	if (minor_version < 2)
325 		return label;
326 
327 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
328 		return label;
329 
330 	label = kzalloc(sizeof(struct nfs4_label), flags);
331 	if (label == NULL)
332 		return ERR_PTR(-ENOMEM);
333 
334 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
335 	if (label->label == NULL) {
336 		kfree(label);
337 		return ERR_PTR(-ENOMEM);
338 	}
339 	label->len = NFS4_MAXLABELLEN;
340 
341 	return label;
342 }
343 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
344 #else
345 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
346 					struct nfs4_label *label)
347 {
348 }
349 #endif
350 EXPORT_SYMBOL_GPL(nfs_setsecurity);
351 
352 /*
353  * This is our front-end to iget that looks up inodes by file handle
354  * instead of inode number.
355  */
356 struct inode *
357 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
358 {
359 	struct nfs_find_desc desc = {
360 		.fh	= fh,
361 		.fattr	= fattr
362 	};
363 	struct inode *inode = ERR_PTR(-ENOENT);
364 	unsigned long hash;
365 
366 	nfs_attr_check_mountpoint(sb, fattr);
367 
368 	if (nfs_attr_use_mounted_on_fileid(fattr))
369 		fattr->fileid = fattr->mounted_on_fileid;
370 	else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
371 		goto out_no_inode;
372 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
373 		goto out_no_inode;
374 
375 	hash = nfs_fattr_to_ino_t(fattr);
376 
377 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
378 	if (inode == NULL) {
379 		inode = ERR_PTR(-ENOMEM);
380 		goto out_no_inode;
381 	}
382 
383 	if (inode->i_state & I_NEW) {
384 		struct nfs_inode *nfsi = NFS_I(inode);
385 		unsigned long now = jiffies;
386 
387 		/* We set i_ino for the few things that still rely on it,
388 		 * such as stat(2) */
389 		inode->i_ino = hash;
390 
391 		/* We can't support update_atime(), since the server will reset it */
392 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
393 		inode->i_mode = fattr->mode;
394 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
395 				&& nfs_server_capable(inode, NFS_CAP_MODE))
396 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
397 		/* Why so? Because we want revalidate for devices/FIFOs, and
398 		 * that's precisely what we have in nfs_file_inode_operations.
399 		 */
400 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
401 		if (S_ISREG(inode->i_mode)) {
402 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
403 			inode->i_data.a_ops = &nfs_file_aops;
404 		} else if (S_ISDIR(inode->i_mode)) {
405 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
406 			inode->i_fop = &nfs_dir_operations;
407 			inode->i_data.a_ops = &nfs_dir_aops;
408 			/* Deal with crossing mountpoints */
409 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
410 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
411 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
412 					inode->i_op = &nfs_referral_inode_operations;
413 				else
414 					inode->i_op = &nfs_mountpoint_inode_operations;
415 				inode->i_fop = NULL;
416 				inode->i_flags |= S_AUTOMOUNT;
417 			}
418 		} else if (S_ISLNK(inode->i_mode)) {
419 			inode->i_op = &nfs_symlink_inode_operations;
420 			inode_nohighmem(inode);
421 		} else
422 			init_special_inode(inode, inode->i_mode, fattr->rdev);
423 
424 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
425 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
426 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
427 		inode->i_version = 0;
428 		inode->i_size = 0;
429 		clear_nlink(inode);
430 		inode->i_uid = make_kuid(&init_user_ns, -2);
431 		inode->i_gid = make_kgid(&init_user_ns, -2);
432 		inode->i_blocks = 0;
433 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
434 		nfsi->write_io = 0;
435 		nfsi->read_io = 0;
436 
437 		nfsi->read_cache_jiffies = fattr->time_start;
438 		nfsi->attr_gencount = fattr->gencount;
439 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
440 			inode->i_atime = fattr->atime;
441 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
442 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
443 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
444 			inode->i_mtime = fattr->mtime;
445 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
446 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
447 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
448 			inode->i_ctime = fattr->ctime;
449 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
450 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
451 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
452 			inode->i_version = fattr->change_attr;
453 		else
454 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
455 				| NFS_INO_REVAL_PAGECACHE);
456 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
457 			inode->i_size = nfs_size_to_loff_t(fattr->size);
458 		else
459 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
460 				| NFS_INO_REVAL_PAGECACHE);
461 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
462 			set_nlink(inode, fattr->nlink);
463 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
464 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
465 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
466 			inode->i_uid = fattr->uid;
467 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
468 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
469 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
470 			inode->i_gid = fattr->gid;
471 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
472 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
473 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
474 			inode->i_blocks = fattr->du.nfs2.blocks;
475 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
476 			/*
477 			 * report the blocks in 512byte units
478 			 */
479 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
480 		}
481 
482 		nfs_setsecurity(inode, fattr, label);
483 
484 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
485 		nfsi->attrtimeo_timestamp = now;
486 		nfsi->access_cache = RB_ROOT;
487 
488 		nfs_fscache_init_inode(inode);
489 
490 		unlock_new_inode(inode);
491 	} else
492 		nfs_refresh_inode(inode, fattr);
493 	dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
494 		inode->i_sb->s_id,
495 		(unsigned long long)NFS_FILEID(inode),
496 		nfs_display_fhandle_hash(fh),
497 		atomic_read(&inode->i_count));
498 
499 out:
500 	return inode;
501 
502 out_no_inode:
503 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
504 	goto out;
505 }
506 EXPORT_SYMBOL_GPL(nfs_fhget);
507 
508 #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)
509 
510 int
511 nfs_setattr(struct dentry *dentry, struct iattr *attr)
512 {
513 	struct inode *inode = d_inode(dentry);
514 	struct nfs_fattr *fattr;
515 	int error = 0;
516 
517 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
518 
519 	/* skip mode change if it's just for clearing setuid/setgid */
520 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
521 		attr->ia_valid &= ~ATTR_MODE;
522 
523 	if (attr->ia_valid & ATTR_SIZE) {
524 		BUG_ON(!S_ISREG(inode->i_mode));
525 
526 		error = inode_newsize_ok(inode, attr->ia_size);
527 		if (error)
528 			return error;
529 
530 		if (attr->ia_size == i_size_read(inode))
531 			attr->ia_valid &= ~ATTR_SIZE;
532 	}
533 
534 	/* Optimization: if the end result is no change, don't RPC */
535 	attr->ia_valid &= NFS_VALID_ATTRS;
536 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
537 		return 0;
538 
539 	trace_nfs_setattr_enter(inode);
540 
541 	/* Write all dirty data */
542 	if (S_ISREG(inode->i_mode))
543 		nfs_sync_inode(inode);
544 
545 	fattr = nfs_alloc_fattr();
546 	if (fattr == NULL) {
547 		error = -ENOMEM;
548 		goto out;
549 	}
550 
551 	/*
552 	 * Return any delegations if we're going to change ACLs
553 	 */
554 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
555 		NFS_PROTO(inode)->return_delegation(inode);
556 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
557 	if (error == 0)
558 		error = nfs_refresh_inode(inode, fattr);
559 	nfs_free_fattr(fattr);
560 out:
561 	trace_nfs_setattr_exit(inode, error);
562 	return error;
563 }
564 EXPORT_SYMBOL_GPL(nfs_setattr);
565 
566 /**
567  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
568  * @inode: inode of the file used
569  * @offset: file offset to start truncating
570  *
571  * This is a copy of the common vmtruncate, but with the locking
572  * corrected to take into account the fact that NFS requires
573  * inode->i_size to be updated under the inode->i_lock.
574  * Note: must be called with inode->i_lock held!
575  */
576 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
577 {
578 	int err;
579 
580 	err = inode_newsize_ok(inode, offset);
581 	if (err)
582 		goto out;
583 
584 	i_size_write(inode, offset);
585 	/* Optimisation */
586 	if (offset == 0)
587 		NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
588 
589 	spin_unlock(&inode->i_lock);
590 	truncate_pagecache(inode, offset);
591 	spin_lock(&inode->i_lock);
592 out:
593 	return err;
594 }
595 
596 /**
597  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
598  * @inode: pointer to struct inode
599  * @attr: pointer to struct iattr
600  *
601  * Note: we do this in the *proc.c in order to ensure that
602  *       it works for things like exclusive creates too.
603  */
604 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
605 		struct nfs_fattr *fattr)
606 {
607 	/* Barrier: bump the attribute generation count. */
608 	nfs_fattr_set_barrier(fattr);
609 
610 	spin_lock(&inode->i_lock);
611 	NFS_I(inode)->attr_gencount = fattr->gencount;
612 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
613 		if ((attr->ia_valid & ATTR_MODE) != 0) {
614 			int mode = attr->ia_mode & S_IALLUGO;
615 			mode |= inode->i_mode & ~S_IALLUGO;
616 			inode->i_mode = mode;
617 		}
618 		if ((attr->ia_valid & ATTR_UID) != 0)
619 			inode->i_uid = attr->ia_uid;
620 		if ((attr->ia_valid & ATTR_GID) != 0)
621 			inode->i_gid = attr->ia_gid;
622 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
623 				| NFS_INO_INVALID_ACL);
624 	}
625 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
626 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
627 		nfs_vmtruncate(inode, attr->ia_size);
628 	}
629 	if (fattr->valid)
630 		nfs_update_inode(inode, fattr);
631 	else
632 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
633 	spin_unlock(&inode->i_lock);
634 }
635 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
636 
637 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
638 {
639 	struct dentry *parent;
640 
641 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
642 		return;
643 	parent = dget_parent(dentry);
644 	nfs_force_use_readdirplus(d_inode(parent));
645 	dput(parent);
646 }
647 
648 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
649 {
650 	struct dentry *parent;
651 
652 	if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
653 		return;
654 	parent = dget_parent(dentry);
655 	nfs_advise_use_readdirplus(d_inode(parent));
656 	dput(parent);
657 }
658 
659 static bool nfs_need_revalidate_inode(struct inode *inode)
660 {
661 	if (NFS_I(inode)->cache_validity &
662 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
663 		return true;
664 	if (nfs_attribute_cache_expired(inode))
665 		return true;
666 	return false;
667 }
668 
669 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
670 {
671 	struct inode *inode = d_inode(dentry);
672 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
673 	int err = 0;
674 
675 	trace_nfs_getattr_enter(inode);
676 	/* Flush out writes to the server in order to update c/mtime.  */
677 	if (S_ISREG(inode->i_mode)) {
678 		err = filemap_write_and_wait(inode->i_mapping);
679 		if (err)
680 			goto out;
681 	}
682 
683 	/*
684 	 * We may force a getattr if the user cares about atime.
685 	 *
686 	 * Note that we only have to check the vfsmount flags here:
687 	 *  - NFS always sets S_NOATIME by so checking it would give a
688 	 *    bogus result
689 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
690 	 *    no point in checking those.
691 	 */
692  	if ((mnt->mnt_flags & MNT_NOATIME) ||
693  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
694 		need_atime = 0;
695 
696 	if (need_atime || nfs_need_revalidate_inode(inode)) {
697 		struct nfs_server *server = NFS_SERVER(inode);
698 
699 		nfs_readdirplus_parent_cache_miss(dentry);
700 		err = __nfs_revalidate_inode(server, inode);
701 	} else
702 		nfs_readdirplus_parent_cache_hit(dentry);
703 	if (!err) {
704 		generic_fillattr(inode, stat);
705 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
706 		if (S_ISDIR(inode->i_mode))
707 			stat->blksize = NFS_SERVER(inode)->dtsize;
708 	}
709 out:
710 	trace_nfs_getattr_exit(inode, err);
711 	return err;
712 }
713 EXPORT_SYMBOL_GPL(nfs_getattr);
714 
715 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
716 {
717 	atomic_set(&l_ctx->count, 1);
718 	l_ctx->lockowner = current->files;
719 	INIT_LIST_HEAD(&l_ctx->list);
720 	atomic_set(&l_ctx->io_count, 0);
721 }
722 
723 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
724 {
725 	struct nfs_lock_context *head = &ctx->lock_context;
726 	struct nfs_lock_context *pos = head;
727 
728 	do {
729 		if (pos->lockowner != current->files)
730 			continue;
731 		atomic_inc(&pos->count);
732 		return pos;
733 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
734 	return NULL;
735 }
736 
737 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
738 {
739 	struct nfs_lock_context *res, *new = NULL;
740 	struct inode *inode = d_inode(ctx->dentry);
741 
742 	spin_lock(&inode->i_lock);
743 	res = __nfs_find_lock_context(ctx);
744 	if (res == NULL) {
745 		spin_unlock(&inode->i_lock);
746 		new = kmalloc(sizeof(*new), GFP_KERNEL);
747 		if (new == NULL)
748 			return ERR_PTR(-ENOMEM);
749 		nfs_init_lock_context(new);
750 		spin_lock(&inode->i_lock);
751 		res = __nfs_find_lock_context(ctx);
752 		if (res == NULL) {
753 			list_add_tail(&new->list, &ctx->lock_context.list);
754 			new->open_context = ctx;
755 			res = new;
756 			new = NULL;
757 		}
758 	}
759 	spin_unlock(&inode->i_lock);
760 	kfree(new);
761 	return res;
762 }
763 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
764 
765 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
766 {
767 	struct nfs_open_context *ctx = l_ctx->open_context;
768 	struct inode *inode = d_inode(ctx->dentry);
769 
770 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
771 		return;
772 	list_del(&l_ctx->list);
773 	spin_unlock(&inode->i_lock);
774 	kfree(l_ctx);
775 }
776 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
777 
778 /**
779  * nfs_close_context - Common close_context() routine NFSv2/v3
780  * @ctx: pointer to context
781  * @is_sync: is this a synchronous close
782  *
783  * Ensure that the attributes are up to date if we're mounted
784  * with close-to-open semantics and we have cached data that will
785  * need to be revalidated on open.
786  */
787 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
788 {
789 	struct nfs_inode *nfsi;
790 	struct inode *inode;
791 	struct nfs_server *server;
792 
793 	if (!(ctx->mode & FMODE_WRITE))
794 		return;
795 	if (!is_sync)
796 		return;
797 	inode = d_inode(ctx->dentry);
798 	nfsi = NFS_I(inode);
799 	if (inode->i_mapping->nrpages == 0)
800 		return;
801 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
802 		return;
803 	if (!list_empty(&nfsi->open_files))
804 		return;
805 	server = NFS_SERVER(inode);
806 	if (server->flags & NFS_MOUNT_NOCTO)
807 		return;
808 	nfs_revalidate_inode(server, inode);
809 }
810 EXPORT_SYMBOL_GPL(nfs_close_context);
811 
812 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
813 						fmode_t f_mode,
814 						struct file *filp)
815 {
816 	struct nfs_open_context *ctx;
817 	struct rpc_cred *cred = rpc_lookup_cred();
818 	if (IS_ERR(cred))
819 		return ERR_CAST(cred);
820 
821 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
822 	if (!ctx) {
823 		put_rpccred(cred);
824 		return ERR_PTR(-ENOMEM);
825 	}
826 	nfs_sb_active(dentry->d_sb);
827 	ctx->dentry = dget(dentry);
828 	ctx->cred = cred;
829 	ctx->state = NULL;
830 	ctx->mode = f_mode;
831 	ctx->flags = 0;
832 	ctx->error = 0;
833 	ctx->flock_owner = (fl_owner_t)filp;
834 	nfs_init_lock_context(&ctx->lock_context);
835 	ctx->lock_context.open_context = ctx;
836 	INIT_LIST_HEAD(&ctx->list);
837 	ctx->mdsthreshold = NULL;
838 	return ctx;
839 }
840 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
841 
842 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
843 {
844 	if (ctx != NULL)
845 		atomic_inc(&ctx->lock_context.count);
846 	return ctx;
847 }
848 EXPORT_SYMBOL_GPL(get_nfs_open_context);
849 
850 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
851 {
852 	struct inode *inode = d_inode(ctx->dentry);
853 	struct super_block *sb = ctx->dentry->d_sb;
854 
855 	if (!list_empty(&ctx->list)) {
856 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
857 			return;
858 		list_del(&ctx->list);
859 		spin_unlock(&inode->i_lock);
860 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
861 		return;
862 	if (inode != NULL)
863 		NFS_PROTO(inode)->close_context(ctx, is_sync);
864 	if (ctx->cred != NULL)
865 		put_rpccred(ctx->cred);
866 	dput(ctx->dentry);
867 	nfs_sb_deactive(sb);
868 	kfree(ctx->mdsthreshold);
869 	kfree(ctx);
870 }
871 
872 void put_nfs_open_context(struct nfs_open_context *ctx)
873 {
874 	__put_nfs_open_context(ctx, 0);
875 }
876 EXPORT_SYMBOL_GPL(put_nfs_open_context);
877 
878 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
879 {
880 	__put_nfs_open_context(ctx, 1);
881 }
882 
883 /*
884  * Ensure that mmap has a recent RPC credential for use when writing out
885  * shared pages
886  */
887 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
888 {
889 	struct inode *inode = d_inode(ctx->dentry);
890 	struct nfs_inode *nfsi = NFS_I(inode);
891 
892 	spin_lock(&inode->i_lock);
893 	if (ctx->mode & FMODE_WRITE)
894 		list_add(&ctx->list, &nfsi->open_files);
895 	else
896 		list_add_tail(&ctx->list, &nfsi->open_files);
897 	spin_unlock(&inode->i_lock);
898 }
899 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
900 
901 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
902 {
903 	filp->private_data = get_nfs_open_context(ctx);
904 	if (list_empty(&ctx->list))
905 		nfs_inode_attach_open_context(ctx);
906 }
907 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
908 
909 /*
910  * Given an inode, search for an open context with the desired characteristics
911  */
912 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
913 {
914 	struct nfs_inode *nfsi = NFS_I(inode);
915 	struct nfs_open_context *pos, *ctx = NULL;
916 
917 	spin_lock(&inode->i_lock);
918 	list_for_each_entry(pos, &nfsi->open_files, list) {
919 		if (cred != NULL && pos->cred != cred)
920 			continue;
921 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
922 			continue;
923 		ctx = get_nfs_open_context(pos);
924 		break;
925 	}
926 	spin_unlock(&inode->i_lock);
927 	return ctx;
928 }
929 
930 void nfs_file_clear_open_context(struct file *filp)
931 {
932 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
933 
934 	if (ctx) {
935 		struct inode *inode = d_inode(ctx->dentry);
936 
937 		/*
938 		 * We fatal error on write before. Try to writeback
939 		 * every page again.
940 		 */
941 		if (ctx->error < 0)
942 			invalidate_inode_pages2(inode->i_mapping);
943 		filp->private_data = NULL;
944 		spin_lock(&inode->i_lock);
945 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
946 		spin_unlock(&inode->i_lock);
947 		put_nfs_open_context_sync(ctx);
948 	}
949 }
950 
951 /*
952  * These allocate and release file read/write context information.
953  */
954 int nfs_open(struct inode *inode, struct file *filp)
955 {
956 	struct nfs_open_context *ctx;
957 
958 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
959 	if (IS_ERR(ctx))
960 		return PTR_ERR(ctx);
961 	nfs_file_set_open_context(filp, ctx);
962 	put_nfs_open_context(ctx);
963 	nfs_fscache_open_file(inode, filp);
964 	return 0;
965 }
966 
967 /*
968  * This function is called whenever some part of NFS notices that
969  * the cached attributes have to be refreshed.
970  */
971 int
972 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
973 {
974 	int		 status = -ESTALE;
975 	struct nfs4_label *label = NULL;
976 	struct nfs_fattr *fattr = NULL;
977 	struct nfs_inode *nfsi = NFS_I(inode);
978 
979 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
980 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
981 
982 	trace_nfs_revalidate_inode_enter(inode);
983 
984 	if (is_bad_inode(inode))
985 		goto out;
986 	if (NFS_STALE(inode))
987 		goto out;
988 
989 	/* pNFS: Attributes aren't updated until we layoutcommit */
990 	if (S_ISREG(inode->i_mode)) {
991 		status = pnfs_sync_inode(inode, false);
992 		if (status)
993 			goto out;
994 	}
995 
996 	status = -ENOMEM;
997 	fattr = nfs_alloc_fattr();
998 	if (fattr == NULL)
999 		goto out;
1000 
1001 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1002 
1003 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1004 	if (IS_ERR(label)) {
1005 		status = PTR_ERR(label);
1006 		goto out;
1007 	}
1008 
1009 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
1010 	if (status != 0) {
1011 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1012 			 inode->i_sb->s_id,
1013 			 (unsigned long long)NFS_FILEID(inode), status);
1014 		if (status == -ESTALE) {
1015 			nfs_zap_caches(inode);
1016 			if (!S_ISDIR(inode->i_mode))
1017 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1018 		}
1019 		goto err_out;
1020 	}
1021 
1022 	status = nfs_refresh_inode(inode, fattr);
1023 	if (status) {
1024 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1025 			 inode->i_sb->s_id,
1026 			 (unsigned long long)NFS_FILEID(inode), status);
1027 		goto err_out;
1028 	}
1029 
1030 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1031 		nfs_zap_acl_cache(inode);
1032 
1033 	nfs_setsecurity(inode, fattr, label);
1034 
1035 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1036 		inode->i_sb->s_id,
1037 		(unsigned long long)NFS_FILEID(inode));
1038 
1039 err_out:
1040 	nfs4_label_free(label);
1041 out:
1042 	nfs_free_fattr(fattr);
1043 	trace_nfs_revalidate_inode_exit(inode, status);
1044 	return status;
1045 }
1046 
1047 int nfs_attribute_timeout(struct inode *inode)
1048 {
1049 	struct nfs_inode *nfsi = NFS_I(inode);
1050 
1051 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1052 }
1053 
1054 int nfs_attribute_cache_expired(struct inode *inode)
1055 {
1056 	if (nfs_have_delegated_attributes(inode))
1057 		return 0;
1058 	return nfs_attribute_timeout(inode);
1059 }
1060 
1061 /**
1062  * nfs_revalidate_inode - Revalidate the inode attributes
1063  * @server - pointer to nfs_server struct
1064  * @inode - pointer to inode struct
1065  *
1066  * Updates inode attribute information by retrieving the data from the server.
1067  */
1068 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1069 {
1070 	if (!nfs_need_revalidate_inode(inode))
1071 		return NFS_STALE(inode) ? -ESTALE : 0;
1072 	return __nfs_revalidate_inode(server, inode);
1073 }
1074 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1075 
1076 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1077 {
1078 	if (!(NFS_I(inode)->cache_validity &
1079 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1080 			&& !nfs_attribute_cache_expired(inode))
1081 		return NFS_STALE(inode) ? -ESTALE : 0;
1082 	return -ECHILD;
1083 }
1084 
1085 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1086 {
1087 	struct nfs_inode *nfsi = NFS_I(inode);
1088 	int ret;
1089 
1090 	if (mapping->nrpages != 0) {
1091 		if (S_ISREG(inode->i_mode)) {
1092 			unmap_mapping_range(mapping, 0, 0, 0);
1093 			ret = nfs_sync_mapping(mapping);
1094 			if (ret < 0)
1095 				return ret;
1096 		}
1097 		ret = invalidate_inode_pages2(mapping);
1098 		if (ret < 0)
1099 			return ret;
1100 	}
1101 	if (S_ISDIR(inode->i_mode)) {
1102 		spin_lock(&inode->i_lock);
1103 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1104 		spin_unlock(&inode->i_lock);
1105 	}
1106 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1107 	nfs_fscache_wait_on_invalidate(inode);
1108 
1109 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1110 			inode->i_sb->s_id,
1111 			(unsigned long long)NFS_FILEID(inode));
1112 	return 0;
1113 }
1114 
1115 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1116 {
1117 	unsigned long cache_validity = NFS_I(inode)->cache_validity;
1118 
1119 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
1120 		const unsigned long force_reval =
1121 			NFS_INO_REVAL_PAGECACHE|NFS_INO_REVAL_FORCED;
1122 		return (cache_validity & force_reval) == force_reval;
1123 	}
1124 
1125 	return (cache_validity & NFS_INO_REVAL_PAGECACHE)
1126 		|| nfs_attribute_timeout(inode)
1127 		|| NFS_STALE(inode);
1128 }
1129 
1130 int nfs_revalidate_mapping_rcu(struct inode *inode)
1131 {
1132 	struct nfs_inode *nfsi = NFS_I(inode);
1133 	unsigned long *bitlock = &nfsi->flags;
1134 	int ret = 0;
1135 
1136 	if (IS_SWAPFILE(inode))
1137 		goto out;
1138 	if (nfs_mapping_need_revalidate_inode(inode)) {
1139 		ret = -ECHILD;
1140 		goto out;
1141 	}
1142 	spin_lock(&inode->i_lock);
1143 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1144 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1145 		ret = -ECHILD;
1146 	spin_unlock(&inode->i_lock);
1147 out:
1148 	return ret;
1149 }
1150 
1151 /**
1152  * nfs_revalidate_mapping - Revalidate the pagecache
1153  * @inode - pointer to host inode
1154  * @mapping - pointer to mapping
1155  */
1156 int nfs_revalidate_mapping(struct inode *inode,
1157 		struct address_space *mapping)
1158 {
1159 	struct nfs_inode *nfsi = NFS_I(inode);
1160 	unsigned long *bitlock = &nfsi->flags;
1161 	int ret = 0;
1162 
1163 	/* swapfiles are not supposed to be shared. */
1164 	if (IS_SWAPFILE(inode))
1165 		goto out;
1166 
1167 	if (nfs_mapping_need_revalidate_inode(inode)) {
1168 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1169 		if (ret < 0)
1170 			goto out;
1171 	}
1172 
1173 	/*
1174 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1175 	 * invalidations that come in while we're shooting down the mappings
1176 	 * are respected. But, that leaves a race window where one revalidator
1177 	 * can clear the flag, and then another checks it before the mapping
1178 	 * gets invalidated. Fix that by serializing access to this part of
1179 	 * the function.
1180 	 *
1181 	 * At the same time, we need to allow other tasks to see whether we
1182 	 * might be in the middle of invalidating the pages, so we only set
1183 	 * the bit lock here if it looks like we're going to be doing that.
1184 	 */
1185 	for (;;) {
1186 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1187 					 nfs_wait_bit_killable, TASK_KILLABLE);
1188 		if (ret)
1189 			goto out;
1190 		spin_lock(&inode->i_lock);
1191 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1192 			spin_unlock(&inode->i_lock);
1193 			continue;
1194 		}
1195 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1196 			break;
1197 		spin_unlock(&inode->i_lock);
1198 		goto out;
1199 	}
1200 
1201 	set_bit(NFS_INO_INVALIDATING, bitlock);
1202 	smp_wmb();
1203 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1204 	spin_unlock(&inode->i_lock);
1205 	trace_nfs_invalidate_mapping_enter(inode);
1206 	ret = nfs_invalidate_mapping(inode, mapping);
1207 	trace_nfs_invalidate_mapping_exit(inode, ret);
1208 
1209 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1210 	smp_mb__after_atomic();
1211 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1212 out:
1213 	return ret;
1214 }
1215 
1216 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1217 {
1218 	struct inode *inode = &nfsi->vfs_inode;
1219 
1220 	assert_spin_locked(&inode->i_lock);
1221 
1222 	if (!S_ISREG(inode->i_mode))
1223 		return false;
1224 	if (list_empty(&nfsi->open_files))
1225 		return false;
1226 	/* Note: This relies on nfsi->open_files being ordered with writers
1227 	 *       being placed at the head of the list.
1228 	 *       See nfs_inode_attach_open_context()
1229 	 */
1230 	return (list_first_entry(&nfsi->open_files,
1231 			struct nfs_open_context,
1232 			list)->mode & FMODE_WRITE) == FMODE_WRITE;
1233 }
1234 
1235 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1236 {
1237 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1238 }
1239 
1240 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1241 {
1242 	struct nfs_inode *nfsi = NFS_I(inode);
1243 	unsigned long ret = 0;
1244 
1245 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1246 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1247 			&& inode->i_version == fattr->pre_change_attr) {
1248 		inode->i_version = fattr->change_attr;
1249 		if (S_ISDIR(inode->i_mode))
1250 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1251 		ret |= NFS_INO_INVALID_ATTR;
1252 	}
1253 	/* If we have atomic WCC data, we may update some attributes */
1254 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1255 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1256 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1257 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1258 		ret |= NFS_INO_INVALID_ATTR;
1259 	}
1260 
1261 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1262 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1263 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1264 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1265 		if (S_ISDIR(inode->i_mode))
1266 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1267 		ret |= NFS_INO_INVALID_ATTR;
1268 	}
1269 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1270 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1271 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1272 			&& nfsi->nrequests == 0) {
1273 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1274 		ret |= NFS_INO_INVALID_ATTR;
1275 	}
1276 
1277 	return ret;
1278 }
1279 
1280 /**
1281  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1282  * @inode - pointer to inode
1283  * @fattr - updated attributes
1284  *
1285  * Verifies the attribute cache. If we have just changed the attributes,
1286  * so that fattr carries weak cache consistency data, then it may
1287  * also update the ctime/mtime/change_attribute.
1288  */
1289 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1290 {
1291 	struct nfs_inode *nfsi = NFS_I(inode);
1292 	loff_t cur_size, new_isize;
1293 	unsigned long invalid = 0;
1294 
1295 
1296 	if (nfs_have_delegated_attributes(inode))
1297 		return 0;
1298 	/* Has the inode gone and changed behind our back? */
1299 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1300 		return -EIO;
1301 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1302 		return -EIO;
1303 
1304 	if (!nfs_file_has_buffered_writers(nfsi)) {
1305 		/* Verify a few of the more important attributes */
1306 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && inode->i_version != fattr->change_attr)
1307 			invalid |= NFS_INO_INVALID_ATTR | NFS_INO_REVAL_PAGECACHE;
1308 
1309 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1310 			invalid |= NFS_INO_INVALID_ATTR;
1311 
1312 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&inode->i_ctime, &fattr->ctime))
1313 			invalid |= NFS_INO_INVALID_ATTR;
1314 
1315 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1316 			cur_size = i_size_read(inode);
1317 			new_isize = nfs_size_to_loff_t(fattr->size);
1318 			if (cur_size != new_isize)
1319 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1320 		}
1321 	}
1322 
1323 	/* Have any file permissions changed? */
1324 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1325 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1326 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1327 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1328 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1329 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1330 
1331 	/* Has the link count changed? */
1332 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1333 		invalid |= NFS_INO_INVALID_ATTR;
1334 
1335 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1336 		invalid |= NFS_INO_INVALID_ATIME;
1337 
1338 	if (invalid != 0)
1339 		nfs_set_cache_invalid(inode, invalid | NFS_INO_REVAL_FORCED);
1340 
1341 	nfsi->read_cache_jiffies = fattr->time_start;
1342 	return 0;
1343 }
1344 
1345 static atomic_long_t nfs_attr_generation_counter;
1346 
1347 static unsigned long nfs_read_attr_generation_counter(void)
1348 {
1349 	return atomic_long_read(&nfs_attr_generation_counter);
1350 }
1351 
1352 unsigned long nfs_inc_attr_generation_counter(void)
1353 {
1354 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1355 }
1356 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1357 
1358 void nfs_fattr_init(struct nfs_fattr *fattr)
1359 {
1360 	fattr->valid = 0;
1361 	fattr->time_start = jiffies;
1362 	fattr->gencount = nfs_inc_attr_generation_counter();
1363 	fattr->owner_name = NULL;
1364 	fattr->group_name = NULL;
1365 }
1366 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1367 
1368 /**
1369  * nfs_fattr_set_barrier
1370  * @fattr: attributes
1371  *
1372  * Used to set a barrier after an attribute was updated. This
1373  * barrier ensures that older attributes from RPC calls that may
1374  * have raced with our update cannot clobber these new values.
1375  * Note that you are still responsible for ensuring that other
1376  * operations which change the attribute on the server do not
1377  * collide.
1378  */
1379 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1380 {
1381 	fattr->gencount = nfs_inc_attr_generation_counter();
1382 }
1383 
1384 struct nfs_fattr *nfs_alloc_fattr(void)
1385 {
1386 	struct nfs_fattr *fattr;
1387 
1388 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1389 	if (fattr != NULL)
1390 		nfs_fattr_init(fattr);
1391 	return fattr;
1392 }
1393 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1394 
1395 struct nfs_fh *nfs_alloc_fhandle(void)
1396 {
1397 	struct nfs_fh *fh;
1398 
1399 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1400 	if (fh != NULL)
1401 		fh->size = 0;
1402 	return fh;
1403 }
1404 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1405 
1406 #ifdef NFS_DEBUG
1407 /*
1408  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1409  *                             in the same way that wireshark does
1410  *
1411  * @fh: file handle
1412  *
1413  * For debugging only.
1414  */
1415 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1416 {
1417 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1418 	 * not on the result */
1419 	return nfs_fhandle_hash(fh);
1420 }
1421 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1422 
1423 /*
1424  * _nfs_display_fhandle - display an NFS file handle on the console
1425  *
1426  * @fh: file handle to display
1427  * @caption: display caption
1428  *
1429  * For debugging only.
1430  */
1431 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1432 {
1433 	unsigned short i;
1434 
1435 	if (fh == NULL || fh->size == 0) {
1436 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1437 		return;
1438 	}
1439 
1440 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1441 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1442 	for (i = 0; i < fh->size; i += 16) {
1443 		__be32 *pos = (__be32 *)&fh->data[i];
1444 
1445 		switch ((fh->size - i - 1) >> 2) {
1446 		case 0:
1447 			printk(KERN_DEFAULT " %08x\n",
1448 				be32_to_cpup(pos));
1449 			break;
1450 		case 1:
1451 			printk(KERN_DEFAULT " %08x %08x\n",
1452 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1453 			break;
1454 		case 2:
1455 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1456 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1457 				be32_to_cpup(pos + 2));
1458 			break;
1459 		default:
1460 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1461 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1462 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1463 		}
1464 	}
1465 }
1466 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1467 #endif
1468 
1469 /**
1470  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1471  * @inode - pointer to inode
1472  * @fattr - attributes
1473  *
1474  * Attempt to divine whether or not an RPC call reply carrying stale
1475  * attributes got scheduled after another call carrying updated ones.
1476  *
1477  * To do so, the function first assumes that a more recent ctime means
1478  * that the attributes in fattr are newer, however it also attempt to
1479  * catch the case where ctime either didn't change, or went backwards
1480  * (if someone reset the clock on the server) by looking at whether
1481  * or not this RPC call was started after the inode was last updated.
1482  * Note also the check for wraparound of 'attr_gencount'
1483  *
1484  * The function returns 'true' if it thinks the attributes in 'fattr' are
1485  * more recent than the ones cached in the inode.
1486  *
1487  */
1488 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1489 {
1490 	const struct nfs_inode *nfsi = NFS_I(inode);
1491 
1492 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1493 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1494 }
1495 
1496 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1497 {
1498 	int ret;
1499 
1500 	trace_nfs_refresh_inode_enter(inode);
1501 
1502 	if (nfs_inode_attrs_need_update(inode, fattr))
1503 		ret = nfs_update_inode(inode, fattr);
1504 	else
1505 		ret = nfs_check_inode_attributes(inode, fattr);
1506 
1507 	trace_nfs_refresh_inode_exit(inode, ret);
1508 	return ret;
1509 }
1510 
1511 /**
1512  * nfs_refresh_inode - try to update the inode attribute cache
1513  * @inode - pointer to inode
1514  * @fattr - updated attributes
1515  *
1516  * Check that an RPC call that returned attributes has not overlapped with
1517  * other recent updates of the inode metadata, then decide whether it is
1518  * safe to do a full update of the inode attributes, or whether just to
1519  * call nfs_check_inode_attributes.
1520  */
1521 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1522 {
1523 	int status;
1524 
1525 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1526 		return 0;
1527 	spin_lock(&inode->i_lock);
1528 	status = nfs_refresh_inode_locked(inode, fattr);
1529 	spin_unlock(&inode->i_lock);
1530 
1531 	return status;
1532 }
1533 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1534 
1535 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1536 {
1537 	unsigned long invalid = NFS_INO_INVALID_ATTR;
1538 
1539 	/*
1540 	 * Don't revalidate the pagecache if we hold a delegation, but do
1541 	 * force an attribute update
1542 	 */
1543 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1544 		invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_FORCED;
1545 
1546 	if (S_ISDIR(inode->i_mode))
1547 		invalid |= NFS_INO_INVALID_DATA;
1548 	nfs_set_cache_invalid(inode, invalid);
1549 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1550 		return 0;
1551 	return nfs_refresh_inode_locked(inode, fattr);
1552 }
1553 
1554 /**
1555  * nfs_post_op_update_inode - try to update the inode attribute cache
1556  * @inode - pointer to inode
1557  * @fattr - updated attributes
1558  *
1559  * After an operation that has changed the inode metadata, mark the
1560  * attribute cache as being invalid, then try to update it.
1561  *
1562  * NB: if the server didn't return any post op attributes, this
1563  * function will force the retrieval of attributes before the next
1564  * NFS request.  Thus it should be used only for operations that
1565  * are expected to change one or more attributes, to avoid
1566  * unnecessary NFS requests and trips through nfs_update_inode().
1567  */
1568 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1569 {
1570 	int status;
1571 
1572 	spin_lock(&inode->i_lock);
1573 	nfs_fattr_set_barrier(fattr);
1574 	status = nfs_post_op_update_inode_locked(inode, fattr);
1575 	spin_unlock(&inode->i_lock);
1576 
1577 	return status;
1578 }
1579 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1580 
1581 /**
1582  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1583  * @inode - pointer to inode
1584  * @fattr - updated attributes
1585  *
1586  * After an operation that has changed the inode metadata, mark the
1587  * attribute cache as being invalid, then try to update it. Fake up
1588  * weak cache consistency data, if none exist.
1589  *
1590  * This function is mainly designed to be used by the ->write_done() functions.
1591  */
1592 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1593 {
1594 	int status;
1595 
1596 	/* Don't do a WCC update if these attributes are already stale */
1597 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1598 			!nfs_inode_attrs_need_update(inode, fattr)) {
1599 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1600 				| NFS_ATTR_FATTR_PRESIZE
1601 				| NFS_ATTR_FATTR_PREMTIME
1602 				| NFS_ATTR_FATTR_PRECTIME);
1603 		goto out_noforce;
1604 	}
1605 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1606 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1607 		fattr->pre_change_attr = inode->i_version;
1608 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1609 	}
1610 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1611 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1612 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1613 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1614 	}
1615 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1616 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1617 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1618 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1619 	}
1620 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1621 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1622 		fattr->pre_size = i_size_read(inode);
1623 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1624 	}
1625 out_noforce:
1626 	status = nfs_post_op_update_inode_locked(inode, fattr);
1627 	return status;
1628 }
1629 
1630 /**
1631  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1632  * @inode - pointer to inode
1633  * @fattr - updated attributes
1634  *
1635  * After an operation that has changed the inode metadata, mark the
1636  * attribute cache as being invalid, then try to update it. Fake up
1637  * weak cache consistency data, if none exist.
1638  *
1639  * This function is mainly designed to be used by the ->write_done() functions.
1640  */
1641 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1642 {
1643 	int status;
1644 
1645 	spin_lock(&inode->i_lock);
1646 	nfs_fattr_set_barrier(fattr);
1647 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1648 	spin_unlock(&inode->i_lock);
1649 	return status;
1650 }
1651 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1652 
1653 
1654 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1655 				    struct nfs_fattr *fattr)
1656 {
1657 	bool ret1 = true, ret2 = true;
1658 
1659 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1660 		ret1 = (nfsi->fileid == fattr->fileid);
1661 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1662 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1663 	return ret1 || ret2;
1664 }
1665 
1666 /*
1667  * Many nfs protocol calls return the new file attributes after
1668  * an operation.  Here we update the inode to reflect the state
1669  * of the server's inode.
1670  *
1671  * This is a bit tricky because we have to make sure all dirty pages
1672  * have been sent off to the server before calling invalidate_inode_pages.
1673  * To make sure no other process adds more write requests while we try
1674  * our best to flush them, we make them sleep during the attribute refresh.
1675  *
1676  * A very similar scenario holds for the dir cache.
1677  */
1678 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1679 {
1680 	struct nfs_server *server;
1681 	struct nfs_inode *nfsi = NFS_I(inode);
1682 	loff_t cur_isize, new_isize;
1683 	unsigned long invalid = 0;
1684 	unsigned long now = jiffies;
1685 	unsigned long save_cache_validity;
1686 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
1687 	bool cache_revalidated = true;
1688 
1689 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1690 			__func__, inode->i_sb->s_id, inode->i_ino,
1691 			nfs_display_fhandle_hash(NFS_FH(inode)),
1692 			atomic_read(&inode->i_count), fattr->valid);
1693 
1694 	if (!nfs_fileid_valid(nfsi, fattr)) {
1695 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1696 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1697 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1698 			inode->i_sb->s_id, (long long)nfsi->fileid,
1699 			(long long)fattr->fileid);
1700 		goto out_err;
1701 	}
1702 
1703 	/*
1704 	 * Make sure the inode's type hasn't changed.
1705 	 */
1706 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1707 		/*
1708 		* Big trouble! The inode has become a different object.
1709 		*/
1710 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1711 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1712 		goto out_err;
1713 	}
1714 
1715 	server = NFS_SERVER(inode);
1716 	/* Update the fsid? */
1717 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1718 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1719 			!IS_AUTOMOUNT(inode))
1720 		server->fsid = fattr->fsid;
1721 
1722 	/*
1723 	 * Update the read time so we don't revalidate too often.
1724 	 */
1725 	nfsi->read_cache_jiffies = fattr->time_start;
1726 
1727 	save_cache_validity = nfsi->cache_validity;
1728 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1729 			| NFS_INO_INVALID_ATIME
1730 			| NFS_INO_REVAL_FORCED
1731 			| NFS_INO_REVAL_PAGECACHE);
1732 
1733 	/* Do atomic weak cache consistency updates */
1734 	invalid |= nfs_wcc_update_inode(inode, fattr);
1735 
1736 	if (pnfs_layoutcommit_outstanding(inode)) {
1737 		nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1738 		cache_revalidated = false;
1739 	}
1740 
1741 	/* More cache consistency checks */
1742 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1743 		if (inode->i_version != fattr->change_attr) {
1744 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1745 					inode->i_sb->s_id, inode->i_ino);
1746 			/* Could it be a race with writeback? */
1747 			if (!have_writers) {
1748 				invalid |= NFS_INO_INVALID_ATTR
1749 					| NFS_INO_INVALID_DATA
1750 					| NFS_INO_INVALID_ACCESS
1751 					| NFS_INO_INVALID_ACL;
1752 				if (S_ISDIR(inode->i_mode))
1753 					nfs_force_lookup_revalidate(inode);
1754 			}
1755 			inode->i_version = fattr->change_attr;
1756 		}
1757 	} else {
1758 		nfsi->cache_validity |= save_cache_validity;
1759 		cache_revalidated = false;
1760 	}
1761 
1762 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1763 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1764 	} else if (server->caps & NFS_CAP_MTIME) {
1765 		nfsi->cache_validity |= save_cache_validity &
1766 				(NFS_INO_INVALID_ATTR
1767 				| NFS_INO_REVAL_FORCED);
1768 		cache_revalidated = false;
1769 	}
1770 
1771 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1772 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1773 	} else if (server->caps & NFS_CAP_CTIME) {
1774 		nfsi->cache_validity |= save_cache_validity &
1775 				(NFS_INO_INVALID_ATTR
1776 				| NFS_INO_REVAL_FORCED);
1777 		cache_revalidated = false;
1778 	}
1779 
1780 	/* Check if our cached file size is stale */
1781 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1782 		new_isize = nfs_size_to_loff_t(fattr->size);
1783 		cur_isize = i_size_read(inode);
1784 		if (new_isize != cur_isize) {
1785 			/* Do we perhaps have any outstanding writes, or has
1786 			 * the file grown beyond our last write? */
1787 			if (nfsi->nrequests == 0 || new_isize > cur_isize) {
1788 				i_size_write(inode, new_isize);
1789 				if (!have_writers)
1790 					invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1791 			}
1792 			dprintk("NFS: isize change on server for file %s/%ld "
1793 					"(%Ld to %Ld)\n",
1794 					inode->i_sb->s_id,
1795 					inode->i_ino,
1796 					(long long)cur_isize,
1797 					(long long)new_isize);
1798 		}
1799 	} else {
1800 		nfsi->cache_validity |= save_cache_validity &
1801 				(NFS_INO_INVALID_ATTR
1802 				| NFS_INO_REVAL_PAGECACHE
1803 				| NFS_INO_REVAL_FORCED);
1804 		cache_revalidated = false;
1805 	}
1806 
1807 
1808 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1809 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1810 	else if (server->caps & NFS_CAP_ATIME) {
1811 		nfsi->cache_validity |= save_cache_validity &
1812 				(NFS_INO_INVALID_ATIME
1813 				| NFS_INO_REVAL_FORCED);
1814 		cache_revalidated = false;
1815 	}
1816 
1817 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1818 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1819 			umode_t newmode = inode->i_mode & S_IFMT;
1820 			newmode |= fattr->mode & S_IALLUGO;
1821 			inode->i_mode = newmode;
1822 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1823 		}
1824 	} else if (server->caps & NFS_CAP_MODE) {
1825 		nfsi->cache_validity |= save_cache_validity &
1826 				(NFS_INO_INVALID_ATTR
1827 				| NFS_INO_INVALID_ACCESS
1828 				| NFS_INO_INVALID_ACL
1829 				| NFS_INO_REVAL_FORCED);
1830 		cache_revalidated = false;
1831 	}
1832 
1833 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1834 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1835 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1836 			inode->i_uid = fattr->uid;
1837 		}
1838 	} else if (server->caps & NFS_CAP_OWNER) {
1839 		nfsi->cache_validity |= save_cache_validity &
1840 				(NFS_INO_INVALID_ATTR
1841 				| NFS_INO_INVALID_ACCESS
1842 				| NFS_INO_INVALID_ACL
1843 				| NFS_INO_REVAL_FORCED);
1844 		cache_revalidated = false;
1845 	}
1846 
1847 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1848 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1849 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1850 			inode->i_gid = fattr->gid;
1851 		}
1852 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
1853 		nfsi->cache_validity |= save_cache_validity &
1854 				(NFS_INO_INVALID_ATTR
1855 				| NFS_INO_INVALID_ACCESS
1856 				| NFS_INO_INVALID_ACL
1857 				| NFS_INO_REVAL_FORCED);
1858 		cache_revalidated = false;
1859 	}
1860 
1861 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1862 		if (inode->i_nlink != fattr->nlink) {
1863 			invalid |= NFS_INO_INVALID_ATTR;
1864 			if (S_ISDIR(inode->i_mode))
1865 				invalid |= NFS_INO_INVALID_DATA;
1866 			set_nlink(inode, fattr->nlink);
1867 		}
1868 	} else if (server->caps & NFS_CAP_NLINK) {
1869 		nfsi->cache_validity |= save_cache_validity &
1870 				(NFS_INO_INVALID_ATTR
1871 				| NFS_INO_REVAL_FORCED);
1872 		cache_revalidated = false;
1873 	}
1874 
1875 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1876 		/*
1877 		 * report the blocks in 512byte units
1878 		 */
1879 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1880 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1881 		inode->i_blocks = fattr->du.nfs2.blocks;
1882 	else
1883 		cache_revalidated = false;
1884 
1885 	/* Update attrtimeo value if we're out of the unstable period */
1886 	if (invalid & NFS_INO_INVALID_ATTR) {
1887 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1888 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1889 		nfsi->attrtimeo_timestamp = now;
1890 		/* Set barrier to be more recent than all outstanding updates */
1891 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1892 	} else {
1893 		if (cache_revalidated) {
1894 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1895 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1896 				nfsi->attrtimeo <<= 1;
1897 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1898 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1899 			}
1900 			nfsi->attrtimeo_timestamp = now;
1901 		}
1902 		/* Set the barrier to be more recent than this fattr */
1903 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1904 			nfsi->attr_gencount = fattr->gencount;
1905 	}
1906 
1907 	/* Don't declare attrcache up to date if there were no attrs! */
1908 	if (cache_revalidated)
1909 		invalid &= ~NFS_INO_INVALID_ATTR;
1910 
1911 	/* Don't invalidate the data if we were to blame */
1912 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1913 				|| S_ISLNK(inode->i_mode)))
1914 		invalid &= ~NFS_INO_INVALID_DATA;
1915 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1916 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1917 		nfs_set_cache_invalid(inode, invalid);
1918 
1919 	return 0;
1920  out_err:
1921 	/*
1922 	 * No need to worry about unhashing the dentry, as the
1923 	 * lookup validation will know that the inode is bad.
1924 	 * (But we fall through to invalidate the caches.)
1925 	 */
1926 	nfs_invalidate_inode(inode);
1927 	return -ESTALE;
1928 }
1929 
1930 struct inode *nfs_alloc_inode(struct super_block *sb)
1931 {
1932 	struct nfs_inode *nfsi;
1933 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1934 	if (!nfsi)
1935 		return NULL;
1936 	nfsi->flags = 0UL;
1937 	nfsi->cache_validity = 0UL;
1938 #if IS_ENABLED(CONFIG_NFS_V4)
1939 	nfsi->nfs4_acl = NULL;
1940 #endif /* CONFIG_NFS_V4 */
1941 	return &nfsi->vfs_inode;
1942 }
1943 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1944 
1945 static void nfs_i_callback(struct rcu_head *head)
1946 {
1947 	struct inode *inode = container_of(head, struct inode, i_rcu);
1948 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1949 }
1950 
1951 void nfs_destroy_inode(struct inode *inode)
1952 {
1953 	call_rcu(&inode->i_rcu, nfs_i_callback);
1954 }
1955 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1956 
1957 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1958 {
1959 #if IS_ENABLED(CONFIG_NFS_V4)
1960 	INIT_LIST_HEAD(&nfsi->open_states);
1961 	nfsi->delegation = NULL;
1962 	init_rwsem(&nfsi->rwsem);
1963 	nfsi->layout = NULL;
1964 #endif
1965 }
1966 
1967 static void init_once(void *foo)
1968 {
1969 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1970 
1971 	inode_init_once(&nfsi->vfs_inode);
1972 	INIT_LIST_HEAD(&nfsi->open_files);
1973 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1974 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1975 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1976 	nfsi->nrequests = 0;
1977 	nfsi->commit_info.ncommit = 0;
1978 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1979 	init_rwsem(&nfsi->rmdir_sem);
1980 	nfs4_init_once(nfsi);
1981 }
1982 
1983 static int __init nfs_init_inodecache(void)
1984 {
1985 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1986 					     sizeof(struct nfs_inode),
1987 					     0, (SLAB_RECLAIM_ACCOUNT|
1988 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1989 					     init_once);
1990 	if (nfs_inode_cachep == NULL)
1991 		return -ENOMEM;
1992 
1993 	return 0;
1994 }
1995 
1996 static void nfs_destroy_inodecache(void)
1997 {
1998 	/*
1999 	 * Make sure all delayed rcu free inodes are flushed before we
2000 	 * destroy cache.
2001 	 */
2002 	rcu_barrier();
2003 	kmem_cache_destroy(nfs_inode_cachep);
2004 }
2005 
2006 struct workqueue_struct *nfsiod_workqueue;
2007 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2008 
2009 /*
2010  * start up the nfsiod workqueue
2011  */
2012 static int nfsiod_start(void)
2013 {
2014 	struct workqueue_struct *wq;
2015 	dprintk("RPC:       creating workqueue nfsiod\n");
2016 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2017 	if (wq == NULL)
2018 		return -ENOMEM;
2019 	nfsiod_workqueue = wq;
2020 	return 0;
2021 }
2022 
2023 /*
2024  * Destroy the nfsiod workqueue
2025  */
2026 static void nfsiod_stop(void)
2027 {
2028 	struct workqueue_struct *wq;
2029 
2030 	wq = nfsiod_workqueue;
2031 	if (wq == NULL)
2032 		return;
2033 	nfsiod_workqueue = NULL;
2034 	destroy_workqueue(wq);
2035 }
2036 
2037 unsigned int nfs_net_id;
2038 EXPORT_SYMBOL_GPL(nfs_net_id);
2039 
2040 static int nfs_net_init(struct net *net)
2041 {
2042 	nfs_clients_init(net);
2043 	return nfs_fs_proc_net_init(net);
2044 }
2045 
2046 static void nfs_net_exit(struct net *net)
2047 {
2048 	nfs_fs_proc_net_exit(net);
2049 	nfs_cleanup_cb_ident_idr(net);
2050 }
2051 
2052 static struct pernet_operations nfs_net_ops = {
2053 	.init = nfs_net_init,
2054 	.exit = nfs_net_exit,
2055 	.id   = &nfs_net_id,
2056 	.size = sizeof(struct nfs_net),
2057 };
2058 
2059 /*
2060  * Initialize NFS
2061  */
2062 static int __init init_nfs_fs(void)
2063 {
2064 	int err;
2065 
2066 	err = register_pernet_subsys(&nfs_net_ops);
2067 	if (err < 0)
2068 		goto out9;
2069 
2070 	err = nfs_fscache_register();
2071 	if (err < 0)
2072 		goto out8;
2073 
2074 	err = nfsiod_start();
2075 	if (err)
2076 		goto out7;
2077 
2078 	err = nfs_fs_proc_init();
2079 	if (err)
2080 		goto out6;
2081 
2082 	err = nfs_init_nfspagecache();
2083 	if (err)
2084 		goto out5;
2085 
2086 	err = nfs_init_inodecache();
2087 	if (err)
2088 		goto out4;
2089 
2090 	err = nfs_init_readpagecache();
2091 	if (err)
2092 		goto out3;
2093 
2094 	err = nfs_init_writepagecache();
2095 	if (err)
2096 		goto out2;
2097 
2098 	err = nfs_init_directcache();
2099 	if (err)
2100 		goto out1;
2101 
2102 	rpc_proc_register(&init_net, &nfs_rpcstat);
2103 
2104 	err = register_nfs_fs();
2105 	if (err)
2106 		goto out0;
2107 
2108 	return 0;
2109 out0:
2110 	rpc_proc_unregister(&init_net, "nfs");
2111 	nfs_destroy_directcache();
2112 out1:
2113 	nfs_destroy_writepagecache();
2114 out2:
2115 	nfs_destroy_readpagecache();
2116 out3:
2117 	nfs_destroy_inodecache();
2118 out4:
2119 	nfs_destroy_nfspagecache();
2120 out5:
2121 	nfs_fs_proc_exit();
2122 out6:
2123 	nfsiod_stop();
2124 out7:
2125 	nfs_fscache_unregister();
2126 out8:
2127 	unregister_pernet_subsys(&nfs_net_ops);
2128 out9:
2129 	return err;
2130 }
2131 
2132 static void __exit exit_nfs_fs(void)
2133 {
2134 	nfs_destroy_directcache();
2135 	nfs_destroy_writepagecache();
2136 	nfs_destroy_readpagecache();
2137 	nfs_destroy_inodecache();
2138 	nfs_destroy_nfspagecache();
2139 	nfs_fscache_unregister();
2140 	unregister_pernet_subsys(&nfs_net_ops);
2141 	rpc_proc_unregister(&init_net, "nfs");
2142 	unregister_nfs_fs();
2143 	nfs_fs_proc_exit();
2144 	nfsiod_stop();
2145 }
2146 
2147 /* Not quite true; I just maintain it */
2148 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2149 MODULE_LICENSE("GPL");
2150 module_param(enable_ino64, bool, 0644);
2151 
2152 module_init(init_nfs_fs)
2153 module_exit(exit_nfs_fs)
2154