xref: /openbmc/linux/fs/nfs/inode.c (revision bc5aa3a0)
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_request_parent_use_readdirplus(struct dentry *dentry)
638 {
639 	struct dentry *parent;
640 
641 	parent = dget_parent(dentry);
642 	nfs_force_use_readdirplus(d_inode(parent));
643 	dput(parent);
644 }
645 
646 static bool nfs_need_revalidate_inode(struct inode *inode)
647 {
648 	if (NFS_I(inode)->cache_validity &
649 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
650 		return true;
651 	if (nfs_attribute_cache_expired(inode))
652 		return true;
653 	return false;
654 }
655 
656 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
657 {
658 	struct inode *inode = d_inode(dentry);
659 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
660 	int err = 0;
661 
662 	trace_nfs_getattr_enter(inode);
663 	/* Flush out writes to the server in order to update c/mtime.  */
664 	if (S_ISREG(inode->i_mode)) {
665 		err = filemap_write_and_wait(inode->i_mapping);
666 		if (err)
667 			goto out;
668 	}
669 
670 	/*
671 	 * We may force a getattr if the user cares about atime.
672 	 *
673 	 * Note that we only have to check the vfsmount flags here:
674 	 *  - NFS always sets S_NOATIME by so checking it would give a
675 	 *    bogus result
676 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
677 	 *    no point in checking those.
678 	 */
679  	if ((mnt->mnt_flags & MNT_NOATIME) ||
680  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
681 		need_atime = 0;
682 
683 	if (need_atime || nfs_need_revalidate_inode(inode)) {
684 		struct nfs_server *server = NFS_SERVER(inode);
685 
686 		if (server->caps & NFS_CAP_READDIRPLUS)
687 			nfs_request_parent_use_readdirplus(dentry);
688 		err = __nfs_revalidate_inode(server, inode);
689 	}
690 	if (!err) {
691 		generic_fillattr(inode, stat);
692 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
693 		if (S_ISDIR(inode->i_mode))
694 			stat->blksize = NFS_SERVER(inode)->dtsize;
695 	}
696 out:
697 	trace_nfs_getattr_exit(inode, err);
698 	return err;
699 }
700 EXPORT_SYMBOL_GPL(nfs_getattr);
701 
702 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
703 {
704 	atomic_set(&l_ctx->count, 1);
705 	l_ctx->lockowner.l_owner = current->files;
706 	l_ctx->lockowner.l_pid = current->tgid;
707 	INIT_LIST_HEAD(&l_ctx->list);
708 	atomic_set(&l_ctx->io_count, 0);
709 }
710 
711 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
712 {
713 	struct nfs_lock_context *head = &ctx->lock_context;
714 	struct nfs_lock_context *pos = head;
715 
716 	do {
717 		if (pos->lockowner.l_owner != current->files)
718 			continue;
719 		if (pos->lockowner.l_pid != current->tgid)
720 			continue;
721 		atomic_inc(&pos->count);
722 		return pos;
723 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
724 	return NULL;
725 }
726 
727 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
728 {
729 	struct nfs_lock_context *res, *new = NULL;
730 	struct inode *inode = d_inode(ctx->dentry);
731 
732 	spin_lock(&inode->i_lock);
733 	res = __nfs_find_lock_context(ctx);
734 	if (res == NULL) {
735 		spin_unlock(&inode->i_lock);
736 		new = kmalloc(sizeof(*new), GFP_KERNEL);
737 		if (new == NULL)
738 			return ERR_PTR(-ENOMEM);
739 		nfs_init_lock_context(new);
740 		spin_lock(&inode->i_lock);
741 		res = __nfs_find_lock_context(ctx);
742 		if (res == NULL) {
743 			list_add_tail(&new->list, &ctx->lock_context.list);
744 			new->open_context = ctx;
745 			res = new;
746 			new = NULL;
747 		}
748 	}
749 	spin_unlock(&inode->i_lock);
750 	kfree(new);
751 	return res;
752 }
753 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
754 
755 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
756 {
757 	struct nfs_open_context *ctx = l_ctx->open_context;
758 	struct inode *inode = d_inode(ctx->dentry);
759 
760 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
761 		return;
762 	list_del(&l_ctx->list);
763 	spin_unlock(&inode->i_lock);
764 	kfree(l_ctx);
765 }
766 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
767 
768 /**
769  * nfs_close_context - Common close_context() routine NFSv2/v3
770  * @ctx: pointer to context
771  * @is_sync: is this a synchronous close
772  *
773  * Ensure that the attributes are up to date if we're mounted
774  * with close-to-open semantics and we have cached data that will
775  * need to be revalidated on open.
776  */
777 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
778 {
779 	struct nfs_inode *nfsi;
780 	struct inode *inode;
781 	struct nfs_server *server;
782 
783 	if (!(ctx->mode & FMODE_WRITE))
784 		return;
785 	if (!is_sync)
786 		return;
787 	inode = d_inode(ctx->dentry);
788 	nfsi = NFS_I(inode);
789 	if (inode->i_mapping->nrpages == 0)
790 		return;
791 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
792 		return;
793 	if (!list_empty(&nfsi->open_files))
794 		return;
795 	server = NFS_SERVER(inode);
796 	if (server->flags & NFS_MOUNT_NOCTO)
797 		return;
798 	nfs_revalidate_inode(server, inode);
799 }
800 EXPORT_SYMBOL_GPL(nfs_close_context);
801 
802 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
803 {
804 	struct nfs_open_context *ctx;
805 	struct rpc_cred *cred = rpc_lookup_cred();
806 	if (IS_ERR(cred))
807 		return ERR_CAST(cred);
808 
809 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
810 	if (!ctx) {
811 		put_rpccred(cred);
812 		return ERR_PTR(-ENOMEM);
813 	}
814 	nfs_sb_active(dentry->d_sb);
815 	ctx->dentry = dget(dentry);
816 	ctx->cred = cred;
817 	ctx->state = NULL;
818 	ctx->mode = f_mode;
819 	ctx->flags = 0;
820 	ctx->error = 0;
821 	nfs_init_lock_context(&ctx->lock_context);
822 	ctx->lock_context.open_context = ctx;
823 	INIT_LIST_HEAD(&ctx->list);
824 	ctx->mdsthreshold = NULL;
825 	return ctx;
826 }
827 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
828 
829 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
830 {
831 	if (ctx != NULL)
832 		atomic_inc(&ctx->lock_context.count);
833 	return ctx;
834 }
835 EXPORT_SYMBOL_GPL(get_nfs_open_context);
836 
837 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
838 {
839 	struct inode *inode = d_inode(ctx->dentry);
840 	struct super_block *sb = ctx->dentry->d_sb;
841 
842 	if (!list_empty(&ctx->list)) {
843 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
844 			return;
845 		list_del(&ctx->list);
846 		spin_unlock(&inode->i_lock);
847 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
848 		return;
849 	if (inode != NULL)
850 		NFS_PROTO(inode)->close_context(ctx, is_sync);
851 	if (ctx->cred != NULL)
852 		put_rpccred(ctx->cred);
853 	dput(ctx->dentry);
854 	nfs_sb_deactive(sb);
855 	kfree(ctx->mdsthreshold);
856 	kfree(ctx);
857 }
858 
859 void put_nfs_open_context(struct nfs_open_context *ctx)
860 {
861 	__put_nfs_open_context(ctx, 0);
862 }
863 EXPORT_SYMBOL_GPL(put_nfs_open_context);
864 
865 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
866 {
867 	__put_nfs_open_context(ctx, 1);
868 }
869 
870 /*
871  * Ensure that mmap has a recent RPC credential for use when writing out
872  * shared pages
873  */
874 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
875 {
876 	struct inode *inode = d_inode(ctx->dentry);
877 	struct nfs_inode *nfsi = NFS_I(inode);
878 
879 	spin_lock(&inode->i_lock);
880 	if (ctx->mode & FMODE_WRITE)
881 		list_add(&ctx->list, &nfsi->open_files);
882 	else
883 		list_add_tail(&ctx->list, &nfsi->open_files);
884 	spin_unlock(&inode->i_lock);
885 }
886 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
887 
888 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
889 {
890 	filp->private_data = get_nfs_open_context(ctx);
891 	if (list_empty(&ctx->list))
892 		nfs_inode_attach_open_context(ctx);
893 }
894 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
895 
896 /*
897  * Given an inode, search for an open context with the desired characteristics
898  */
899 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
900 {
901 	struct nfs_inode *nfsi = NFS_I(inode);
902 	struct nfs_open_context *pos, *ctx = NULL;
903 
904 	spin_lock(&inode->i_lock);
905 	list_for_each_entry(pos, &nfsi->open_files, list) {
906 		if (cred != NULL && pos->cred != cred)
907 			continue;
908 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
909 			continue;
910 		ctx = get_nfs_open_context(pos);
911 		break;
912 	}
913 	spin_unlock(&inode->i_lock);
914 	return ctx;
915 }
916 
917 void nfs_file_clear_open_context(struct file *filp)
918 {
919 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
920 
921 	if (ctx) {
922 		struct inode *inode = d_inode(ctx->dentry);
923 
924 		/*
925 		 * We fatal error on write before. Try to writeback
926 		 * every page again.
927 		 */
928 		if (ctx->error < 0)
929 			invalidate_inode_pages2(inode->i_mapping);
930 		filp->private_data = NULL;
931 		spin_lock(&inode->i_lock);
932 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
933 		spin_unlock(&inode->i_lock);
934 		put_nfs_open_context_sync(ctx);
935 	}
936 }
937 
938 /*
939  * These allocate and release file read/write context information.
940  */
941 int nfs_open(struct inode *inode, struct file *filp)
942 {
943 	struct nfs_open_context *ctx;
944 
945 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode);
946 	if (IS_ERR(ctx))
947 		return PTR_ERR(ctx);
948 	nfs_file_set_open_context(filp, ctx);
949 	put_nfs_open_context(ctx);
950 	nfs_fscache_open_file(inode, filp);
951 	return 0;
952 }
953 
954 /*
955  * This function is called whenever some part of NFS notices that
956  * the cached attributes have to be refreshed.
957  */
958 int
959 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
960 {
961 	int		 status = -ESTALE;
962 	struct nfs4_label *label = NULL;
963 	struct nfs_fattr *fattr = NULL;
964 	struct nfs_inode *nfsi = NFS_I(inode);
965 
966 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
967 		inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
968 
969 	trace_nfs_revalidate_inode_enter(inode);
970 
971 	if (is_bad_inode(inode))
972 		goto out;
973 	if (NFS_STALE(inode))
974 		goto out;
975 
976 	/* pNFS: Attributes aren't updated until we layoutcommit */
977 	if (S_ISREG(inode->i_mode)) {
978 		status = pnfs_sync_inode(inode, false);
979 		if (status)
980 			goto out;
981 	}
982 
983 	status = -ENOMEM;
984 	fattr = nfs_alloc_fattr();
985 	if (fattr == NULL)
986 		goto out;
987 
988 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
989 
990 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
991 	if (IS_ERR(label)) {
992 		status = PTR_ERR(label);
993 		goto out;
994 	}
995 
996 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
997 	if (status != 0) {
998 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
999 			 inode->i_sb->s_id,
1000 			 (unsigned long long)NFS_FILEID(inode), status);
1001 		if (status == -ESTALE) {
1002 			nfs_zap_caches(inode);
1003 			if (!S_ISDIR(inode->i_mode))
1004 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1005 		}
1006 		goto err_out;
1007 	}
1008 
1009 	status = nfs_refresh_inode(inode, fattr);
1010 	if (status) {
1011 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1012 			 inode->i_sb->s_id,
1013 			 (unsigned long long)NFS_FILEID(inode), status);
1014 		goto err_out;
1015 	}
1016 
1017 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1018 		nfs_zap_acl_cache(inode);
1019 
1020 	nfs_setsecurity(inode, fattr, label);
1021 
1022 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1023 		inode->i_sb->s_id,
1024 		(unsigned long long)NFS_FILEID(inode));
1025 
1026 err_out:
1027 	nfs4_label_free(label);
1028 out:
1029 	nfs_free_fattr(fattr);
1030 	trace_nfs_revalidate_inode_exit(inode, status);
1031 	return status;
1032 }
1033 
1034 int nfs_attribute_timeout(struct inode *inode)
1035 {
1036 	struct nfs_inode *nfsi = NFS_I(inode);
1037 
1038 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1039 }
1040 
1041 int nfs_attribute_cache_expired(struct inode *inode)
1042 {
1043 	if (nfs_have_delegated_attributes(inode))
1044 		return 0;
1045 	return nfs_attribute_timeout(inode);
1046 }
1047 
1048 /**
1049  * nfs_revalidate_inode - Revalidate the inode attributes
1050  * @server - pointer to nfs_server struct
1051  * @inode - pointer to inode struct
1052  *
1053  * Updates inode attribute information by retrieving the data from the server.
1054  */
1055 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1056 {
1057 	if (!nfs_need_revalidate_inode(inode))
1058 		return NFS_STALE(inode) ? -ESTALE : 0;
1059 	return __nfs_revalidate_inode(server, inode);
1060 }
1061 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1062 
1063 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1064 {
1065 	if (!(NFS_I(inode)->cache_validity &
1066 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1067 			&& !nfs_attribute_cache_expired(inode))
1068 		return NFS_STALE(inode) ? -ESTALE : 0;
1069 	return -ECHILD;
1070 }
1071 
1072 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1073 {
1074 	struct nfs_inode *nfsi = NFS_I(inode);
1075 	int ret;
1076 
1077 	if (mapping->nrpages != 0) {
1078 		if (S_ISREG(inode->i_mode)) {
1079 			unmap_mapping_range(mapping, 0, 0, 0);
1080 			ret = nfs_sync_mapping(mapping);
1081 			if (ret < 0)
1082 				return ret;
1083 		}
1084 		ret = invalidate_inode_pages2(mapping);
1085 		if (ret < 0)
1086 			return ret;
1087 	}
1088 	if (S_ISDIR(inode->i_mode)) {
1089 		spin_lock(&inode->i_lock);
1090 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1091 		spin_unlock(&inode->i_lock);
1092 	}
1093 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1094 	nfs_fscache_wait_on_invalidate(inode);
1095 
1096 	dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1097 			inode->i_sb->s_id,
1098 			(unsigned long long)NFS_FILEID(inode));
1099 	return 0;
1100 }
1101 
1102 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1103 {
1104 	if (nfs_have_delegated_attributes(inode))
1105 		return false;
1106 	return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1107 		|| nfs_attribute_timeout(inode)
1108 		|| NFS_STALE(inode);
1109 }
1110 
1111 int nfs_revalidate_mapping_rcu(struct inode *inode)
1112 {
1113 	struct nfs_inode *nfsi = NFS_I(inode);
1114 	unsigned long *bitlock = &nfsi->flags;
1115 	int ret = 0;
1116 
1117 	if (IS_SWAPFILE(inode))
1118 		goto out;
1119 	if (nfs_mapping_need_revalidate_inode(inode)) {
1120 		ret = -ECHILD;
1121 		goto out;
1122 	}
1123 	spin_lock(&inode->i_lock);
1124 	if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1125 	    (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1126 		ret = -ECHILD;
1127 	spin_unlock(&inode->i_lock);
1128 out:
1129 	return ret;
1130 }
1131 
1132 /**
1133  * nfs_revalidate_mapping - Revalidate the pagecache
1134  * @inode - pointer to host inode
1135  * @mapping - pointer to mapping
1136  */
1137 int nfs_revalidate_mapping(struct inode *inode,
1138 		struct address_space *mapping)
1139 {
1140 	struct nfs_inode *nfsi = NFS_I(inode);
1141 	unsigned long *bitlock = &nfsi->flags;
1142 	int ret = 0;
1143 
1144 	/* swapfiles are not supposed to be shared. */
1145 	if (IS_SWAPFILE(inode))
1146 		goto out;
1147 
1148 	if (nfs_mapping_need_revalidate_inode(inode)) {
1149 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1150 		if (ret < 0)
1151 			goto out;
1152 	}
1153 
1154 	/*
1155 	 * We must clear NFS_INO_INVALID_DATA first to ensure that
1156 	 * invalidations that come in while we're shooting down the mappings
1157 	 * are respected. But, that leaves a race window where one revalidator
1158 	 * can clear the flag, and then another checks it before the mapping
1159 	 * gets invalidated. Fix that by serializing access to this part of
1160 	 * the function.
1161 	 *
1162 	 * At the same time, we need to allow other tasks to see whether we
1163 	 * might be in the middle of invalidating the pages, so we only set
1164 	 * the bit lock here if it looks like we're going to be doing that.
1165 	 */
1166 	for (;;) {
1167 		ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1168 					 nfs_wait_bit_killable, TASK_KILLABLE);
1169 		if (ret)
1170 			goto out;
1171 		spin_lock(&inode->i_lock);
1172 		if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1173 			spin_unlock(&inode->i_lock);
1174 			continue;
1175 		}
1176 		if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1177 			break;
1178 		spin_unlock(&inode->i_lock);
1179 		goto out;
1180 	}
1181 
1182 	set_bit(NFS_INO_INVALIDATING, bitlock);
1183 	smp_wmb();
1184 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1185 	spin_unlock(&inode->i_lock);
1186 	trace_nfs_invalidate_mapping_enter(inode);
1187 	ret = nfs_invalidate_mapping(inode, mapping);
1188 	trace_nfs_invalidate_mapping_exit(inode, ret);
1189 
1190 	clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1191 	smp_mb__after_atomic();
1192 	wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1193 out:
1194 	return ret;
1195 }
1196 
1197 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1198 {
1199 	struct inode *inode = &nfsi->vfs_inode;
1200 
1201 	assert_spin_locked(&inode->i_lock);
1202 
1203 	if (!S_ISREG(inode->i_mode))
1204 		return false;
1205 	if (list_empty(&nfsi->open_files))
1206 		return false;
1207 	/* Note: This relies on nfsi->open_files being ordered with writers
1208 	 *       being placed at the head of the list.
1209 	 *       See nfs_inode_attach_open_context()
1210 	 */
1211 	return (list_first_entry(&nfsi->open_files,
1212 			struct nfs_open_context,
1213 			list)->mode & FMODE_WRITE) == FMODE_WRITE;
1214 }
1215 
1216 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1217 {
1218 	return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1219 }
1220 
1221 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1222 {
1223 	struct nfs_inode *nfsi = NFS_I(inode);
1224 	unsigned long ret = 0;
1225 
1226 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1227 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1228 			&& inode->i_version == fattr->pre_change_attr) {
1229 		inode->i_version = fattr->change_attr;
1230 		if (S_ISDIR(inode->i_mode))
1231 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1232 		ret |= NFS_INO_INVALID_ATTR;
1233 	}
1234 	/* If we have atomic WCC data, we may update some attributes */
1235 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1236 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1237 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1238 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1239 		ret |= NFS_INO_INVALID_ATTR;
1240 	}
1241 
1242 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1243 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1244 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1245 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1246 		if (S_ISDIR(inode->i_mode))
1247 			nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1248 		ret |= NFS_INO_INVALID_ATTR;
1249 	}
1250 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1251 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1252 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1253 			&& nfsi->nrequests == 0) {
1254 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1255 		ret |= NFS_INO_INVALID_ATTR;
1256 	}
1257 
1258 	return ret;
1259 }
1260 
1261 /**
1262  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1263  * @inode - pointer to inode
1264  * @fattr - updated attributes
1265  *
1266  * Verifies the attribute cache. If we have just changed the attributes,
1267  * so that fattr carries weak cache consistency data, then it may
1268  * also update the ctime/mtime/change_attribute.
1269  */
1270 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1271 {
1272 	struct nfs_inode *nfsi = NFS_I(inode);
1273 	loff_t cur_size, new_isize;
1274 	unsigned long invalid = 0;
1275 
1276 
1277 	if (nfs_have_delegated_attributes(inode))
1278 		return 0;
1279 	/* Has the inode gone and changed behind our back? */
1280 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1281 		return -EIO;
1282 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1283 		return -EIO;
1284 
1285 	if (!nfs_file_has_buffered_writers(nfsi)) {
1286 		/* Verify a few of the more important attributes */
1287 		if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && inode->i_version != fattr->change_attr)
1288 			invalid |= NFS_INO_INVALID_ATTR | NFS_INO_REVAL_PAGECACHE;
1289 
1290 		if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1291 			invalid |= NFS_INO_INVALID_ATTR;
1292 
1293 		if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&inode->i_ctime, &fattr->ctime))
1294 			invalid |= NFS_INO_INVALID_ATTR;
1295 
1296 		if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1297 			cur_size = i_size_read(inode);
1298 			new_isize = nfs_size_to_loff_t(fattr->size);
1299 			if (cur_size != new_isize)
1300 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1301 		}
1302 	}
1303 
1304 	/* Have any file permissions changed? */
1305 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1306 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1307 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1308 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1309 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1310 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1311 
1312 	/* Has the link count changed? */
1313 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1314 		invalid |= NFS_INO_INVALID_ATTR;
1315 
1316 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1317 		invalid |= NFS_INO_INVALID_ATIME;
1318 
1319 	if (invalid != 0)
1320 		nfs_set_cache_invalid(inode, invalid);
1321 
1322 	nfsi->read_cache_jiffies = fattr->time_start;
1323 	return 0;
1324 }
1325 
1326 static atomic_long_t nfs_attr_generation_counter;
1327 
1328 static unsigned long nfs_read_attr_generation_counter(void)
1329 {
1330 	return atomic_long_read(&nfs_attr_generation_counter);
1331 }
1332 
1333 unsigned long nfs_inc_attr_generation_counter(void)
1334 {
1335 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1336 }
1337 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1338 
1339 void nfs_fattr_init(struct nfs_fattr *fattr)
1340 {
1341 	fattr->valid = 0;
1342 	fattr->time_start = jiffies;
1343 	fattr->gencount = nfs_inc_attr_generation_counter();
1344 	fattr->owner_name = NULL;
1345 	fattr->group_name = NULL;
1346 }
1347 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1348 
1349 /**
1350  * nfs_fattr_set_barrier
1351  * @fattr: attributes
1352  *
1353  * Used to set a barrier after an attribute was updated. This
1354  * barrier ensures that older attributes from RPC calls that may
1355  * have raced with our update cannot clobber these new values.
1356  * Note that you are still responsible for ensuring that other
1357  * operations which change the attribute on the server do not
1358  * collide.
1359  */
1360 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1361 {
1362 	fattr->gencount = nfs_inc_attr_generation_counter();
1363 }
1364 
1365 struct nfs_fattr *nfs_alloc_fattr(void)
1366 {
1367 	struct nfs_fattr *fattr;
1368 
1369 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1370 	if (fattr != NULL)
1371 		nfs_fattr_init(fattr);
1372 	return fattr;
1373 }
1374 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1375 
1376 struct nfs_fh *nfs_alloc_fhandle(void)
1377 {
1378 	struct nfs_fh *fh;
1379 
1380 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1381 	if (fh != NULL)
1382 		fh->size = 0;
1383 	return fh;
1384 }
1385 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1386 
1387 #ifdef NFS_DEBUG
1388 /*
1389  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1390  *                             in the same way that wireshark does
1391  *
1392  * @fh: file handle
1393  *
1394  * For debugging only.
1395  */
1396 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1397 {
1398 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1399 	 * not on the result */
1400 	return nfs_fhandle_hash(fh);
1401 }
1402 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1403 
1404 /*
1405  * _nfs_display_fhandle - display an NFS file handle on the console
1406  *
1407  * @fh: file handle to display
1408  * @caption: display caption
1409  *
1410  * For debugging only.
1411  */
1412 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1413 {
1414 	unsigned short i;
1415 
1416 	if (fh == NULL || fh->size == 0) {
1417 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1418 		return;
1419 	}
1420 
1421 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1422 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1423 	for (i = 0; i < fh->size; i += 16) {
1424 		__be32 *pos = (__be32 *)&fh->data[i];
1425 
1426 		switch ((fh->size - i - 1) >> 2) {
1427 		case 0:
1428 			printk(KERN_DEFAULT " %08x\n",
1429 				be32_to_cpup(pos));
1430 			break;
1431 		case 1:
1432 			printk(KERN_DEFAULT " %08x %08x\n",
1433 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1434 			break;
1435 		case 2:
1436 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1437 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1438 				be32_to_cpup(pos + 2));
1439 			break;
1440 		default:
1441 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1442 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1443 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1444 		}
1445 	}
1446 }
1447 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1448 #endif
1449 
1450 /**
1451  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1452  * @inode - pointer to inode
1453  * @fattr - attributes
1454  *
1455  * Attempt to divine whether or not an RPC call reply carrying stale
1456  * attributes got scheduled after another call carrying updated ones.
1457  *
1458  * To do so, the function first assumes that a more recent ctime means
1459  * that the attributes in fattr are newer, however it also attempt to
1460  * catch the case where ctime either didn't change, or went backwards
1461  * (if someone reset the clock on the server) by looking at whether
1462  * or not this RPC call was started after the inode was last updated.
1463  * Note also the check for wraparound of 'attr_gencount'
1464  *
1465  * The function returns 'true' if it thinks the attributes in 'fattr' are
1466  * more recent than the ones cached in the inode.
1467  *
1468  */
1469 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1470 {
1471 	const struct nfs_inode *nfsi = NFS_I(inode);
1472 
1473 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1474 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1475 }
1476 
1477 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1478 {
1479 	int ret;
1480 
1481 	trace_nfs_refresh_inode_enter(inode);
1482 
1483 	if (nfs_inode_attrs_need_update(inode, fattr))
1484 		ret = nfs_update_inode(inode, fattr);
1485 	else
1486 		ret = nfs_check_inode_attributes(inode, fattr);
1487 
1488 	trace_nfs_refresh_inode_exit(inode, ret);
1489 	return ret;
1490 }
1491 
1492 /**
1493  * nfs_refresh_inode - try to update the inode attribute cache
1494  * @inode - pointer to inode
1495  * @fattr - updated attributes
1496  *
1497  * Check that an RPC call that returned attributes has not overlapped with
1498  * other recent updates of the inode metadata, then decide whether it is
1499  * safe to do a full update of the inode attributes, or whether just to
1500  * call nfs_check_inode_attributes.
1501  */
1502 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1503 {
1504 	int status;
1505 
1506 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1507 		return 0;
1508 	spin_lock(&inode->i_lock);
1509 	status = nfs_refresh_inode_locked(inode, fattr);
1510 	spin_unlock(&inode->i_lock);
1511 
1512 	return status;
1513 }
1514 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1515 
1516 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1517 {
1518 	unsigned long invalid = NFS_INO_INVALID_ATTR;
1519 
1520 	/*
1521 	 * Don't revalidate the pagecache if we hold a delegation, but do
1522 	 * force an attribute update
1523 	 */
1524 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1525 		invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_FORCED;
1526 
1527 	if (S_ISDIR(inode->i_mode))
1528 		invalid |= NFS_INO_INVALID_DATA;
1529 	nfs_set_cache_invalid(inode, invalid);
1530 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1531 		return 0;
1532 	return nfs_refresh_inode_locked(inode, fattr);
1533 }
1534 
1535 /**
1536  * nfs_post_op_update_inode - try to update the inode attribute cache
1537  * @inode - pointer to inode
1538  * @fattr - updated attributes
1539  *
1540  * After an operation that has changed the inode metadata, mark the
1541  * attribute cache as being invalid, then try to update it.
1542  *
1543  * NB: if the server didn't return any post op attributes, this
1544  * function will force the retrieval of attributes before the next
1545  * NFS request.  Thus it should be used only for operations that
1546  * are expected to change one or more attributes, to avoid
1547  * unnecessary NFS requests and trips through nfs_update_inode().
1548  */
1549 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1550 {
1551 	int status;
1552 
1553 	spin_lock(&inode->i_lock);
1554 	nfs_fattr_set_barrier(fattr);
1555 	status = nfs_post_op_update_inode_locked(inode, fattr);
1556 	spin_unlock(&inode->i_lock);
1557 
1558 	return status;
1559 }
1560 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1561 
1562 /**
1563  * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1564  * @inode - pointer to inode
1565  * @fattr - updated attributes
1566  *
1567  * After an operation that has changed the inode metadata, mark the
1568  * attribute cache as being invalid, then try to update it. Fake up
1569  * weak cache consistency data, if none exist.
1570  *
1571  * This function is mainly designed to be used by the ->write_done() functions.
1572  */
1573 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1574 {
1575 	int status;
1576 
1577 	/* Don't do a WCC update if these attributes are already stale */
1578 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1579 			!nfs_inode_attrs_need_update(inode, fattr)) {
1580 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1581 				| NFS_ATTR_FATTR_PRESIZE
1582 				| NFS_ATTR_FATTR_PREMTIME
1583 				| NFS_ATTR_FATTR_PRECTIME);
1584 		goto out_noforce;
1585 	}
1586 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1587 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1588 		fattr->pre_change_attr = inode->i_version;
1589 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1590 	}
1591 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1592 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1593 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1594 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1595 	}
1596 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1597 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1598 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1599 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1600 	}
1601 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1602 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1603 		fattr->pre_size = i_size_read(inode);
1604 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1605 	}
1606 out_noforce:
1607 	status = nfs_post_op_update_inode_locked(inode, fattr);
1608 	return status;
1609 }
1610 
1611 /**
1612  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1613  * @inode - pointer to inode
1614  * @fattr - updated attributes
1615  *
1616  * After an operation that has changed the inode metadata, mark the
1617  * attribute cache as being invalid, then try to update it. Fake up
1618  * weak cache consistency data, if none exist.
1619  *
1620  * This function is mainly designed to be used by the ->write_done() functions.
1621  */
1622 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1623 {
1624 	int status;
1625 
1626 	spin_lock(&inode->i_lock);
1627 	nfs_fattr_set_barrier(fattr);
1628 	status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1629 	spin_unlock(&inode->i_lock);
1630 	return status;
1631 }
1632 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1633 
1634 
1635 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1636 				    struct nfs_fattr *fattr)
1637 {
1638 	bool ret1 = true, ret2 = true;
1639 
1640 	if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1641 		ret1 = (nfsi->fileid == fattr->fileid);
1642 	if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1643 		ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1644 	return ret1 || ret2;
1645 }
1646 
1647 /*
1648  * Many nfs protocol calls return the new file attributes after
1649  * an operation.  Here we update the inode to reflect the state
1650  * of the server's inode.
1651  *
1652  * This is a bit tricky because we have to make sure all dirty pages
1653  * have been sent off to the server before calling invalidate_inode_pages.
1654  * To make sure no other process adds more write requests while we try
1655  * our best to flush them, we make them sleep during the attribute refresh.
1656  *
1657  * A very similar scenario holds for the dir cache.
1658  */
1659 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1660 {
1661 	struct nfs_server *server;
1662 	struct nfs_inode *nfsi = NFS_I(inode);
1663 	loff_t cur_isize, new_isize;
1664 	unsigned long invalid = 0;
1665 	unsigned long now = jiffies;
1666 	unsigned long save_cache_validity;
1667 	bool have_writers = nfs_file_has_buffered_writers(nfsi);
1668 	bool cache_revalidated = true;
1669 
1670 	dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1671 			__func__, inode->i_sb->s_id, inode->i_ino,
1672 			nfs_display_fhandle_hash(NFS_FH(inode)),
1673 			atomic_read(&inode->i_count), fattr->valid);
1674 
1675 	if (!nfs_fileid_valid(nfsi, fattr)) {
1676 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1677 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1678 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1679 			inode->i_sb->s_id, (long long)nfsi->fileid,
1680 			(long long)fattr->fileid);
1681 		goto out_err;
1682 	}
1683 
1684 	/*
1685 	 * Make sure the inode's type hasn't changed.
1686 	 */
1687 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1688 		/*
1689 		* Big trouble! The inode has become a different object.
1690 		*/
1691 		printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1692 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1693 		goto out_err;
1694 	}
1695 
1696 	server = NFS_SERVER(inode);
1697 	/* Update the fsid? */
1698 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1699 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1700 			!IS_AUTOMOUNT(inode))
1701 		server->fsid = fattr->fsid;
1702 
1703 	/*
1704 	 * Update the read time so we don't revalidate too often.
1705 	 */
1706 	nfsi->read_cache_jiffies = fattr->time_start;
1707 
1708 	save_cache_validity = nfsi->cache_validity;
1709 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1710 			| NFS_INO_INVALID_ATIME
1711 			| NFS_INO_REVAL_FORCED
1712 			| NFS_INO_REVAL_PAGECACHE);
1713 
1714 	/* Do atomic weak cache consistency updates */
1715 	invalid |= nfs_wcc_update_inode(inode, fattr);
1716 
1717 	if (pnfs_layoutcommit_outstanding(inode)) {
1718 		nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1719 		cache_revalidated = false;
1720 	}
1721 
1722 	/* More cache consistency checks */
1723 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1724 		if (inode->i_version != fattr->change_attr) {
1725 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1726 					inode->i_sb->s_id, inode->i_ino);
1727 			/* Could it be a race with writeback? */
1728 			if (!have_writers) {
1729 				invalid |= NFS_INO_INVALID_ATTR
1730 					| NFS_INO_INVALID_DATA
1731 					| NFS_INO_INVALID_ACCESS
1732 					| NFS_INO_INVALID_ACL;
1733 				if (S_ISDIR(inode->i_mode))
1734 					nfs_force_lookup_revalidate(inode);
1735 			}
1736 			inode->i_version = fattr->change_attr;
1737 		}
1738 	} else {
1739 		nfsi->cache_validity |= save_cache_validity;
1740 		cache_revalidated = false;
1741 	}
1742 
1743 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1744 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1745 	} else if (server->caps & NFS_CAP_MTIME) {
1746 		nfsi->cache_validity |= save_cache_validity &
1747 				(NFS_INO_INVALID_ATTR
1748 				| NFS_INO_REVAL_FORCED);
1749 		cache_revalidated = false;
1750 	}
1751 
1752 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1753 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1754 	} else if (server->caps & NFS_CAP_CTIME) {
1755 		nfsi->cache_validity |= save_cache_validity &
1756 				(NFS_INO_INVALID_ATTR
1757 				| NFS_INO_REVAL_FORCED);
1758 		cache_revalidated = false;
1759 	}
1760 
1761 	/* Check if our cached file size is stale */
1762 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1763 		new_isize = nfs_size_to_loff_t(fattr->size);
1764 		cur_isize = i_size_read(inode);
1765 		if (new_isize != cur_isize) {
1766 			/* Do we perhaps have any outstanding writes, or has
1767 			 * the file grown beyond our last write? */
1768 			if (nfsi->nrequests == 0 || new_isize > cur_isize) {
1769 				i_size_write(inode, new_isize);
1770 				if (!have_writers)
1771 					invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1772 			}
1773 			dprintk("NFS: isize change on server for file %s/%ld "
1774 					"(%Ld to %Ld)\n",
1775 					inode->i_sb->s_id,
1776 					inode->i_ino,
1777 					(long long)cur_isize,
1778 					(long long)new_isize);
1779 		}
1780 	} else {
1781 		nfsi->cache_validity |= save_cache_validity &
1782 				(NFS_INO_INVALID_ATTR
1783 				| NFS_INO_REVAL_PAGECACHE
1784 				| NFS_INO_REVAL_FORCED);
1785 		cache_revalidated = false;
1786 	}
1787 
1788 
1789 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1790 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1791 	else if (server->caps & NFS_CAP_ATIME) {
1792 		nfsi->cache_validity |= save_cache_validity &
1793 				(NFS_INO_INVALID_ATIME
1794 				| NFS_INO_REVAL_FORCED);
1795 		cache_revalidated = false;
1796 	}
1797 
1798 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1799 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1800 			umode_t newmode = inode->i_mode & S_IFMT;
1801 			newmode |= fattr->mode & S_IALLUGO;
1802 			inode->i_mode = newmode;
1803 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1804 		}
1805 	} else if (server->caps & NFS_CAP_MODE) {
1806 		nfsi->cache_validity |= save_cache_validity &
1807 				(NFS_INO_INVALID_ATTR
1808 				| NFS_INO_INVALID_ACCESS
1809 				| NFS_INO_INVALID_ACL
1810 				| NFS_INO_REVAL_FORCED);
1811 		cache_revalidated = false;
1812 	}
1813 
1814 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1815 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1816 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1817 			inode->i_uid = fattr->uid;
1818 		}
1819 	} else if (server->caps & NFS_CAP_OWNER) {
1820 		nfsi->cache_validity |= save_cache_validity &
1821 				(NFS_INO_INVALID_ATTR
1822 				| NFS_INO_INVALID_ACCESS
1823 				| NFS_INO_INVALID_ACL
1824 				| NFS_INO_REVAL_FORCED);
1825 		cache_revalidated = false;
1826 	}
1827 
1828 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1829 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1830 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1831 			inode->i_gid = fattr->gid;
1832 		}
1833 	} else if (server->caps & NFS_CAP_OWNER_GROUP) {
1834 		nfsi->cache_validity |= save_cache_validity &
1835 				(NFS_INO_INVALID_ATTR
1836 				| NFS_INO_INVALID_ACCESS
1837 				| NFS_INO_INVALID_ACL
1838 				| NFS_INO_REVAL_FORCED);
1839 		cache_revalidated = false;
1840 	}
1841 
1842 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1843 		if (inode->i_nlink != fattr->nlink) {
1844 			invalid |= NFS_INO_INVALID_ATTR;
1845 			if (S_ISDIR(inode->i_mode))
1846 				invalid |= NFS_INO_INVALID_DATA;
1847 			set_nlink(inode, fattr->nlink);
1848 		}
1849 	} else if (server->caps & NFS_CAP_NLINK) {
1850 		nfsi->cache_validity |= save_cache_validity &
1851 				(NFS_INO_INVALID_ATTR
1852 				| NFS_INO_REVAL_FORCED);
1853 		cache_revalidated = false;
1854 	}
1855 
1856 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1857 		/*
1858 		 * report the blocks in 512byte units
1859 		 */
1860 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1861 	} else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1862 		inode->i_blocks = fattr->du.nfs2.blocks;
1863 	else
1864 		cache_revalidated = false;
1865 
1866 	/* Update attrtimeo value if we're out of the unstable period */
1867 	if (invalid & NFS_INO_INVALID_ATTR) {
1868 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1869 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1870 		nfsi->attrtimeo_timestamp = now;
1871 		/* Set barrier to be more recent than all outstanding updates */
1872 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1873 	} else {
1874 		if (cache_revalidated) {
1875 			if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
1876 				nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1877 				nfsi->attrtimeo <<= 1;
1878 				if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
1879 					nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1880 			}
1881 			nfsi->attrtimeo_timestamp = now;
1882 		}
1883 		/* Set the barrier to be more recent than this fattr */
1884 		if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
1885 			nfsi->attr_gencount = fattr->gencount;
1886 	}
1887 
1888 	/* Don't declare attrcache up to date if there were no attrs! */
1889 	if (cache_revalidated)
1890 		invalid &= ~NFS_INO_INVALID_ATTR;
1891 
1892 	/* Don't invalidate the data if we were to blame */
1893 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1894 				|| S_ISLNK(inode->i_mode)))
1895 		invalid &= ~NFS_INO_INVALID_DATA;
1896 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1897 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1898 		nfs_set_cache_invalid(inode, invalid);
1899 
1900 	return 0;
1901  out_err:
1902 	/*
1903 	 * No need to worry about unhashing the dentry, as the
1904 	 * lookup validation will know that the inode is bad.
1905 	 * (But we fall through to invalidate the caches.)
1906 	 */
1907 	nfs_invalidate_inode(inode);
1908 	return -ESTALE;
1909 }
1910 
1911 struct inode *nfs_alloc_inode(struct super_block *sb)
1912 {
1913 	struct nfs_inode *nfsi;
1914 	nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1915 	if (!nfsi)
1916 		return NULL;
1917 	nfsi->flags = 0UL;
1918 	nfsi->cache_validity = 0UL;
1919 #if IS_ENABLED(CONFIG_NFS_V4)
1920 	nfsi->nfs4_acl = NULL;
1921 #endif /* CONFIG_NFS_V4 */
1922 	return &nfsi->vfs_inode;
1923 }
1924 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1925 
1926 static void nfs_i_callback(struct rcu_head *head)
1927 {
1928 	struct inode *inode = container_of(head, struct inode, i_rcu);
1929 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1930 }
1931 
1932 void nfs_destroy_inode(struct inode *inode)
1933 {
1934 	call_rcu(&inode->i_rcu, nfs_i_callback);
1935 }
1936 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1937 
1938 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1939 {
1940 #if IS_ENABLED(CONFIG_NFS_V4)
1941 	INIT_LIST_HEAD(&nfsi->open_states);
1942 	nfsi->delegation = NULL;
1943 	init_rwsem(&nfsi->rwsem);
1944 	nfsi->layout = NULL;
1945 #endif
1946 }
1947 
1948 static void init_once(void *foo)
1949 {
1950 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1951 
1952 	inode_init_once(&nfsi->vfs_inode);
1953 	INIT_LIST_HEAD(&nfsi->open_files);
1954 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1955 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1956 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1957 	nfsi->nrequests = 0;
1958 	nfsi->commit_info.ncommit = 0;
1959 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1960 	init_rwsem(&nfsi->rmdir_sem);
1961 	nfs4_init_once(nfsi);
1962 }
1963 
1964 static int __init nfs_init_inodecache(void)
1965 {
1966 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1967 					     sizeof(struct nfs_inode),
1968 					     0, (SLAB_RECLAIM_ACCOUNT|
1969 						SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1970 					     init_once);
1971 	if (nfs_inode_cachep == NULL)
1972 		return -ENOMEM;
1973 
1974 	return 0;
1975 }
1976 
1977 static void nfs_destroy_inodecache(void)
1978 {
1979 	/*
1980 	 * Make sure all delayed rcu free inodes are flushed before we
1981 	 * destroy cache.
1982 	 */
1983 	rcu_barrier();
1984 	kmem_cache_destroy(nfs_inode_cachep);
1985 }
1986 
1987 struct workqueue_struct *nfsiod_workqueue;
1988 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1989 
1990 /*
1991  * start up the nfsiod workqueue
1992  */
1993 static int nfsiod_start(void)
1994 {
1995 	struct workqueue_struct *wq;
1996 	dprintk("RPC:       creating workqueue nfsiod\n");
1997 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1998 	if (wq == NULL)
1999 		return -ENOMEM;
2000 	nfsiod_workqueue = wq;
2001 	return 0;
2002 }
2003 
2004 /*
2005  * Destroy the nfsiod workqueue
2006  */
2007 static void nfsiod_stop(void)
2008 {
2009 	struct workqueue_struct *wq;
2010 
2011 	wq = nfsiod_workqueue;
2012 	if (wq == NULL)
2013 		return;
2014 	nfsiod_workqueue = NULL;
2015 	destroy_workqueue(wq);
2016 }
2017 
2018 int nfs_net_id;
2019 EXPORT_SYMBOL_GPL(nfs_net_id);
2020 
2021 static int nfs_net_init(struct net *net)
2022 {
2023 	nfs_clients_init(net);
2024 	return nfs_fs_proc_net_init(net);
2025 }
2026 
2027 static void nfs_net_exit(struct net *net)
2028 {
2029 	nfs_fs_proc_net_exit(net);
2030 	nfs_cleanup_cb_ident_idr(net);
2031 }
2032 
2033 static struct pernet_operations nfs_net_ops = {
2034 	.init = nfs_net_init,
2035 	.exit = nfs_net_exit,
2036 	.id   = &nfs_net_id,
2037 	.size = sizeof(struct nfs_net),
2038 };
2039 
2040 /*
2041  * Initialize NFS
2042  */
2043 static int __init init_nfs_fs(void)
2044 {
2045 	int err;
2046 
2047 	err = register_pernet_subsys(&nfs_net_ops);
2048 	if (err < 0)
2049 		goto out9;
2050 
2051 	err = nfs_fscache_register();
2052 	if (err < 0)
2053 		goto out8;
2054 
2055 	err = nfsiod_start();
2056 	if (err)
2057 		goto out7;
2058 
2059 	err = nfs_fs_proc_init();
2060 	if (err)
2061 		goto out6;
2062 
2063 	err = nfs_init_nfspagecache();
2064 	if (err)
2065 		goto out5;
2066 
2067 	err = nfs_init_inodecache();
2068 	if (err)
2069 		goto out4;
2070 
2071 	err = nfs_init_readpagecache();
2072 	if (err)
2073 		goto out3;
2074 
2075 	err = nfs_init_writepagecache();
2076 	if (err)
2077 		goto out2;
2078 
2079 	err = nfs_init_directcache();
2080 	if (err)
2081 		goto out1;
2082 
2083 	rpc_proc_register(&init_net, &nfs_rpcstat);
2084 
2085 	err = register_nfs_fs();
2086 	if (err)
2087 		goto out0;
2088 
2089 	return 0;
2090 out0:
2091 	rpc_proc_unregister(&init_net, "nfs");
2092 	nfs_destroy_directcache();
2093 out1:
2094 	nfs_destroy_writepagecache();
2095 out2:
2096 	nfs_destroy_readpagecache();
2097 out3:
2098 	nfs_destroy_inodecache();
2099 out4:
2100 	nfs_destroy_nfspagecache();
2101 out5:
2102 	nfs_fs_proc_exit();
2103 out6:
2104 	nfsiod_stop();
2105 out7:
2106 	nfs_fscache_unregister();
2107 out8:
2108 	unregister_pernet_subsys(&nfs_net_ops);
2109 out9:
2110 	return err;
2111 }
2112 
2113 static void __exit exit_nfs_fs(void)
2114 {
2115 	nfs_destroy_directcache();
2116 	nfs_destroy_writepagecache();
2117 	nfs_destroy_readpagecache();
2118 	nfs_destroy_inodecache();
2119 	nfs_destroy_nfspagecache();
2120 	nfs_fscache_unregister();
2121 	unregister_pernet_subsys(&nfs_net_ops);
2122 	rpc_proc_unregister(&init_net, "nfs");
2123 	unregister_nfs_fs();
2124 	nfs_fs_proc_exit();
2125 	nfsiod_stop();
2126 }
2127 
2128 /* Not quite true; I just maintain it */
2129 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2130 MODULE_LICENSE("GPL");
2131 module_param(enable_ino64, bool, 0644);
2132 
2133 module_init(init_nfs_fs)
2134 module_exit(exit_nfs_fs)
2135