xref: /openbmc/linux/fs/nfs/inode.c (revision e23feb16)
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 /**
75  * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76  * @word: long word containing the bit lock
77  */
78 int nfs_wait_bit_killable(void *word)
79 {
80 	if (fatal_signal_pending(current))
81 		return -ERESTARTSYS;
82 	freezable_schedule_unsafe();
83 	return 0;
84 }
85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
86 
87 /**
88  * nfs_compat_user_ino64 - returns the user-visible inode number
89  * @fileid: 64-bit fileid
90  *
91  * This function returns a 32-bit inode number if the boot parameter
92  * nfs.enable_ino64 is zero.
93  */
94 u64 nfs_compat_user_ino64(u64 fileid)
95 {
96 #ifdef CONFIG_COMPAT
97 	compat_ulong_t ino;
98 #else
99 	unsigned long ino;
100 #endif
101 
102 	if (enable_ino64)
103 		return fileid;
104 	ino = fileid;
105 	if (sizeof(ino) < sizeof(fileid))
106 		ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
107 	return ino;
108 }
109 
110 int nfs_drop_inode(struct inode *inode)
111 {
112 	return NFS_STALE(inode) || generic_drop_inode(inode);
113 }
114 EXPORT_SYMBOL_GPL(nfs_drop_inode);
115 
116 void nfs_clear_inode(struct inode *inode)
117 {
118 	/*
119 	 * The following should never happen...
120 	 */
121 	WARN_ON_ONCE(nfs_have_writebacks(inode));
122 	WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
123 	nfs_zap_acl_cache(inode);
124 	nfs_access_zap_cache(inode);
125 	nfs_fscache_release_inode_cookie(inode);
126 }
127 EXPORT_SYMBOL_GPL(nfs_clear_inode);
128 
129 void nfs_evict_inode(struct inode *inode)
130 {
131 	truncate_inode_pages(&inode->i_data, 0);
132 	clear_inode(inode);
133 	nfs_clear_inode(inode);
134 }
135 
136 /**
137  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
138  */
139 int nfs_sync_mapping(struct address_space *mapping)
140 {
141 	int ret = 0;
142 
143 	if (mapping->nrpages != 0) {
144 		unmap_mapping_range(mapping, 0, 0, 0);
145 		ret = nfs_wb_all(mapping->host);
146 	}
147 	return ret;
148 }
149 
150 /*
151  * Invalidate the local caches
152  */
153 static void nfs_zap_caches_locked(struct inode *inode)
154 {
155 	struct nfs_inode *nfsi = NFS_I(inode);
156 	int mode = inode->i_mode;
157 
158 	nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
159 
160 	nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
161 	nfsi->attrtimeo_timestamp = jiffies;
162 
163 	memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
164 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
165 		nfs_fscache_invalidate(inode);
166 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR
167 					| NFS_INO_INVALID_LABEL
168 					| NFS_INO_INVALID_DATA
169 					| NFS_INO_INVALID_ACCESS
170 					| NFS_INO_INVALID_ACL
171 					| NFS_INO_REVAL_PAGECACHE;
172 	} else
173 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR
174 					| NFS_INO_INVALID_LABEL
175 					| NFS_INO_INVALID_ACCESS
176 					| NFS_INO_INVALID_ACL
177 					| NFS_INO_REVAL_PAGECACHE;
178 }
179 
180 void nfs_zap_caches(struct inode *inode)
181 {
182 	spin_lock(&inode->i_lock);
183 	nfs_zap_caches_locked(inode);
184 	spin_unlock(&inode->i_lock);
185 }
186 
187 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
188 {
189 	if (mapping->nrpages != 0) {
190 		spin_lock(&inode->i_lock);
191 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
192 		nfs_fscache_invalidate(inode);
193 		spin_unlock(&inode->i_lock);
194 	}
195 }
196 
197 void nfs_zap_acl_cache(struct inode *inode)
198 {
199 	void (*clear_acl_cache)(struct inode *);
200 
201 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
202 	if (clear_acl_cache != NULL)
203 		clear_acl_cache(inode);
204 	spin_lock(&inode->i_lock);
205 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
206 	spin_unlock(&inode->i_lock);
207 }
208 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
209 
210 void nfs_invalidate_atime(struct inode *inode)
211 {
212 	spin_lock(&inode->i_lock);
213 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
214 	spin_unlock(&inode->i_lock);
215 }
216 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
217 
218 /*
219  * Invalidate, but do not unhash, the inode.
220  * NB: must be called with inode->i_lock held!
221  */
222 static void nfs_invalidate_inode(struct inode *inode)
223 {
224 	set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
225 	nfs_zap_caches_locked(inode);
226 }
227 
228 struct nfs_find_desc {
229 	struct nfs_fh		*fh;
230 	struct nfs_fattr	*fattr;
231 };
232 
233 /*
234  * In NFSv3 we can have 64bit inode numbers. In order to support
235  * this, and re-exported directories (also seen in NFSv2)
236  * we are forced to allow 2 different inodes to have the same
237  * i_ino.
238  */
239 static int
240 nfs_find_actor(struct inode *inode, void *opaque)
241 {
242 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
243 	struct nfs_fh		*fh = desc->fh;
244 	struct nfs_fattr	*fattr = desc->fattr;
245 
246 	if (NFS_FILEID(inode) != fattr->fileid)
247 		return 0;
248 	if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
249 		return 0;
250 	if (nfs_compare_fh(NFS_FH(inode), fh))
251 		return 0;
252 	if (is_bad_inode(inode) || NFS_STALE(inode))
253 		return 0;
254 	return 1;
255 }
256 
257 static int
258 nfs_init_locked(struct inode *inode, void *opaque)
259 {
260 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
261 	struct nfs_fattr	*fattr = desc->fattr;
262 
263 	set_nfs_fileid(inode, fattr->fileid);
264 	nfs_copy_fh(NFS_FH(inode), desc->fh);
265 	return 0;
266 }
267 
268 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
269 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
270 					struct nfs4_label *label)
271 {
272 	int error;
273 
274 	if (label == NULL)
275 		return;
276 
277 	if (nfs_server_capable(inode, NFS_CAP_SECURITY_LABEL) == 0)
278 		return;
279 
280 	if (NFS_SERVER(inode)->nfs_client->cl_minorversion < 2)
281 		return;
282 
283 	if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
284 		error = security_inode_notifysecctx(inode, label->label,
285 				label->len);
286 		if (error)
287 			printk(KERN_ERR "%s() %s %d "
288 					"security_inode_notifysecctx() %d\n",
289 					__func__,
290 					(char *)label->label,
291 					label->len, error);
292 	}
293 }
294 
295 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
296 {
297 	struct nfs4_label *label = NULL;
298 	int minor_version = server->nfs_client->cl_minorversion;
299 
300 	if (minor_version < 2)
301 		return label;
302 
303 	if (!(server->caps & NFS_CAP_SECURITY_LABEL))
304 		return label;
305 
306 	label = kzalloc(sizeof(struct nfs4_label), flags);
307 	if (label == NULL)
308 		return ERR_PTR(-ENOMEM);
309 
310 	label->label = kzalloc(NFS4_MAXLABELLEN, flags);
311 	if (label->label == NULL) {
312 		kfree(label);
313 		return ERR_PTR(-ENOMEM);
314 	}
315 	label->len = NFS4_MAXLABELLEN;
316 
317 	return label;
318 }
319 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
320 #else
321 void inline nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
322 					struct nfs4_label *label)
323 {
324 }
325 #endif
326 EXPORT_SYMBOL_GPL(nfs_setsecurity);
327 
328 /*
329  * This is our front-end to iget that looks up inodes by file handle
330  * instead of inode number.
331  */
332 struct inode *
333 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
334 {
335 	struct nfs_find_desc desc = {
336 		.fh	= fh,
337 		.fattr	= fattr
338 	};
339 	struct inode *inode = ERR_PTR(-ENOENT);
340 	unsigned long hash;
341 
342 	nfs_attr_check_mountpoint(sb, fattr);
343 
344 	if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
345 	    !nfs_attr_use_mounted_on_fileid(fattr))
346 		goto out_no_inode;
347 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
348 		goto out_no_inode;
349 
350 	hash = nfs_fattr_to_ino_t(fattr);
351 
352 	inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
353 	if (inode == NULL) {
354 		inode = ERR_PTR(-ENOMEM);
355 		goto out_no_inode;
356 	}
357 
358 	if (inode->i_state & I_NEW) {
359 		struct nfs_inode *nfsi = NFS_I(inode);
360 		unsigned long now = jiffies;
361 
362 		/* We set i_ino for the few things that still rely on it,
363 		 * such as stat(2) */
364 		inode->i_ino = hash;
365 
366 		/* We can't support update_atime(), since the server will reset it */
367 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
368 		inode->i_mode = fattr->mode;
369 		if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
370 				&& nfs_server_capable(inode, NFS_CAP_MODE))
371 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
372 		/* Why so? Because we want revalidate for devices/FIFOs, and
373 		 * that's precisely what we have in nfs_file_inode_operations.
374 		 */
375 		inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
376 		if (S_ISREG(inode->i_mode)) {
377 			inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
378 			inode->i_data.a_ops = &nfs_file_aops;
379 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
380 		} else if (S_ISDIR(inode->i_mode)) {
381 			inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
382 			inode->i_fop = &nfs_dir_operations;
383 			inode->i_data.a_ops = &nfs_dir_aops;
384 			/* Deal with crossing mountpoints */
385 			if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
386 					fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
387 				if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
388 					inode->i_op = &nfs_referral_inode_operations;
389 				else
390 					inode->i_op = &nfs_mountpoint_inode_operations;
391 				inode->i_fop = NULL;
392 				inode->i_flags |= S_AUTOMOUNT;
393 			}
394 		} else if (S_ISLNK(inode->i_mode))
395 			inode->i_op = &nfs_symlink_inode_operations;
396 		else
397 			init_special_inode(inode, inode->i_mode, fattr->rdev);
398 
399 		memset(&inode->i_atime, 0, sizeof(inode->i_atime));
400 		memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
401 		memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
402 		inode->i_version = 0;
403 		inode->i_size = 0;
404 		clear_nlink(inode);
405 		inode->i_uid = make_kuid(&init_user_ns, -2);
406 		inode->i_gid = make_kgid(&init_user_ns, -2);
407 		inode->i_blocks = 0;
408 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
409 		nfsi->write_io = 0;
410 		nfsi->read_io = 0;
411 
412 		nfsi->read_cache_jiffies = fattr->time_start;
413 		nfsi->attr_gencount = fattr->gencount;
414 		if (fattr->valid & NFS_ATTR_FATTR_ATIME)
415 			inode->i_atime = fattr->atime;
416 		else if (nfs_server_capable(inode, NFS_CAP_ATIME))
417 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
418 		if (fattr->valid & NFS_ATTR_FATTR_MTIME)
419 			inode->i_mtime = fattr->mtime;
420 		else if (nfs_server_capable(inode, NFS_CAP_MTIME))
421 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
422 		if (fattr->valid & NFS_ATTR_FATTR_CTIME)
423 			inode->i_ctime = fattr->ctime;
424 		else if (nfs_server_capable(inode, NFS_CAP_CTIME))
425 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
426 		if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
427 			inode->i_version = fattr->change_attr;
428 		else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
429 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
430 		if (fattr->valid & NFS_ATTR_FATTR_SIZE)
431 			inode->i_size = nfs_size_to_loff_t(fattr->size);
432 		else
433 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR
434 				| NFS_INO_REVAL_PAGECACHE;
435 		if (fattr->valid & NFS_ATTR_FATTR_NLINK)
436 			set_nlink(inode, fattr->nlink);
437 		else if (nfs_server_capable(inode, NFS_CAP_NLINK))
438 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
439 		if (fattr->valid & NFS_ATTR_FATTR_OWNER)
440 			inode->i_uid = fattr->uid;
441 		else if (nfs_server_capable(inode, NFS_CAP_OWNER))
442 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
443 		if (fattr->valid & NFS_ATTR_FATTR_GROUP)
444 			inode->i_gid = fattr->gid;
445 		else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
446 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
447 		if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
448 			inode->i_blocks = fattr->du.nfs2.blocks;
449 		if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
450 			/*
451 			 * report the blocks in 512byte units
452 			 */
453 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
454 		}
455 
456 		nfs_setsecurity(inode, fattr, label);
457 
458 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
459 		nfsi->attrtimeo_timestamp = now;
460 		nfsi->access_cache = RB_ROOT;
461 
462 		nfs_fscache_init_inode_cookie(inode);
463 
464 		unlock_new_inode(inode);
465 	} else
466 		nfs_refresh_inode(inode, fattr);
467 	dprintk("NFS: nfs_fhget(%s/%Ld fh_crc=0x%08x ct=%d)\n",
468 		inode->i_sb->s_id,
469 		(long long)NFS_FILEID(inode),
470 		nfs_display_fhandle_hash(fh),
471 		atomic_read(&inode->i_count));
472 
473 out:
474 	return inode;
475 
476 out_no_inode:
477 	dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
478 	goto out;
479 }
480 EXPORT_SYMBOL_GPL(nfs_fhget);
481 
482 #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)
483 
484 int
485 nfs_setattr(struct dentry *dentry, struct iattr *attr)
486 {
487 	struct inode *inode = dentry->d_inode;
488 	struct nfs_fattr *fattr;
489 	int error = -ENOMEM;
490 
491 	nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
492 
493 	/* skip mode change if it's just for clearing setuid/setgid */
494 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
495 		attr->ia_valid &= ~ATTR_MODE;
496 
497 	if (attr->ia_valid & ATTR_SIZE) {
498 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
499 			attr->ia_valid &= ~ATTR_SIZE;
500 	}
501 
502 	/* Optimization: if the end result is no change, don't RPC */
503 	attr->ia_valid &= NFS_VALID_ATTRS;
504 	if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
505 		return 0;
506 
507 	trace_nfs_setattr_enter(inode);
508 
509 	/* Write all dirty data */
510 	if (S_ISREG(inode->i_mode)) {
511 		nfs_inode_dio_wait(inode);
512 		nfs_wb_all(inode);
513 	}
514 
515 	fattr = nfs_alloc_fattr();
516 	if (fattr == NULL)
517 		goto out;
518 	/*
519 	 * Return any delegations if we're going to change ACLs
520 	 */
521 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
522 		NFS_PROTO(inode)->return_delegation(inode);
523 	error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
524 	if (error == 0)
525 		error = nfs_refresh_inode(inode, fattr);
526 	nfs_free_fattr(fattr);
527 out:
528 	trace_nfs_setattr_exit(inode, error);
529 	return error;
530 }
531 EXPORT_SYMBOL_GPL(nfs_setattr);
532 
533 /**
534  * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
535  * @inode: inode of the file used
536  * @offset: file offset to start truncating
537  *
538  * This is a copy of the common vmtruncate, but with the locking
539  * corrected to take into account the fact that NFS requires
540  * inode->i_size to be updated under the inode->i_lock.
541  */
542 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
543 {
544 	int err;
545 
546 	err = inode_newsize_ok(inode, offset);
547 	if (err)
548 		goto out;
549 
550 	spin_lock(&inode->i_lock);
551 	i_size_write(inode, offset);
552 	spin_unlock(&inode->i_lock);
553 
554 	truncate_pagecache(inode, offset);
555 out:
556 	return err;
557 }
558 
559 /**
560  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
561  * @inode: pointer to struct inode
562  * @attr: pointer to struct iattr
563  *
564  * Note: we do this in the *proc.c in order to ensure that
565  *       it works for things like exclusive creates too.
566  */
567 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
568 {
569 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
570 		spin_lock(&inode->i_lock);
571 		if ((attr->ia_valid & ATTR_MODE) != 0) {
572 			int mode = attr->ia_mode & S_IALLUGO;
573 			mode |= inode->i_mode & ~S_IALLUGO;
574 			inode->i_mode = mode;
575 		}
576 		if ((attr->ia_valid & ATTR_UID) != 0)
577 			inode->i_uid = attr->ia_uid;
578 		if ((attr->ia_valid & ATTR_GID) != 0)
579 			inode->i_gid = attr->ia_gid;
580 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
581 		spin_unlock(&inode->i_lock);
582 	}
583 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
584 		nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
585 		nfs_vmtruncate(inode, attr->ia_size);
586 	}
587 }
588 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
589 
590 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
591 {
592 	struct inode *inode = dentry->d_inode;
593 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
594 	int err;
595 
596 	trace_nfs_getattr_enter(inode);
597 	/* Flush out writes to the server in order to update c/mtime.  */
598 	if (S_ISREG(inode->i_mode)) {
599 		nfs_inode_dio_wait(inode);
600 		err = filemap_write_and_wait(inode->i_mapping);
601 		if (err)
602 			goto out;
603 	}
604 
605 	/*
606 	 * We may force a getattr if the user cares about atime.
607 	 *
608 	 * Note that we only have to check the vfsmount flags here:
609 	 *  - NFS always sets S_NOATIME by so checking it would give a
610 	 *    bogus result
611 	 *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
612 	 *    no point in checking those.
613 	 */
614  	if ((mnt->mnt_flags & MNT_NOATIME) ||
615  	    ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
616 		need_atime = 0;
617 
618 	if (need_atime)
619 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
620 	else
621 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
622 	if (!err) {
623 		generic_fillattr(inode, stat);
624 		stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
625 	}
626 out:
627 	trace_nfs_getattr_exit(inode, err);
628 	return err;
629 }
630 EXPORT_SYMBOL_GPL(nfs_getattr);
631 
632 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
633 {
634 	atomic_set(&l_ctx->count, 1);
635 	l_ctx->lockowner.l_owner = current->files;
636 	l_ctx->lockowner.l_pid = current->tgid;
637 	INIT_LIST_HEAD(&l_ctx->list);
638 	nfs_iocounter_init(&l_ctx->io_count);
639 }
640 
641 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
642 {
643 	struct nfs_lock_context *head = &ctx->lock_context;
644 	struct nfs_lock_context *pos = head;
645 
646 	do {
647 		if (pos->lockowner.l_owner != current->files)
648 			continue;
649 		if (pos->lockowner.l_pid != current->tgid)
650 			continue;
651 		atomic_inc(&pos->count);
652 		return pos;
653 	} while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
654 	return NULL;
655 }
656 
657 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
658 {
659 	struct nfs_lock_context *res, *new = NULL;
660 	struct inode *inode = ctx->dentry->d_inode;
661 
662 	spin_lock(&inode->i_lock);
663 	res = __nfs_find_lock_context(ctx);
664 	if (res == NULL) {
665 		spin_unlock(&inode->i_lock);
666 		new = kmalloc(sizeof(*new), GFP_KERNEL);
667 		if (new == NULL)
668 			return ERR_PTR(-ENOMEM);
669 		nfs_init_lock_context(new);
670 		spin_lock(&inode->i_lock);
671 		res = __nfs_find_lock_context(ctx);
672 		if (res == NULL) {
673 			list_add_tail(&new->list, &ctx->lock_context.list);
674 			new->open_context = ctx;
675 			res = new;
676 			new = NULL;
677 		}
678 	}
679 	spin_unlock(&inode->i_lock);
680 	kfree(new);
681 	return res;
682 }
683 
684 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
685 {
686 	struct nfs_open_context *ctx = l_ctx->open_context;
687 	struct inode *inode = ctx->dentry->d_inode;
688 
689 	if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
690 		return;
691 	list_del(&l_ctx->list);
692 	spin_unlock(&inode->i_lock);
693 	kfree(l_ctx);
694 }
695 
696 /**
697  * nfs_close_context - Common close_context() routine NFSv2/v3
698  * @ctx: pointer to context
699  * @is_sync: is this a synchronous close
700  *
701  * always ensure that the attributes are up to date if we're mounted
702  * with close-to-open semantics
703  */
704 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
705 {
706 	struct inode *inode;
707 	struct nfs_server *server;
708 
709 	if (!(ctx->mode & FMODE_WRITE))
710 		return;
711 	if (!is_sync)
712 		return;
713 	inode = ctx->dentry->d_inode;
714 	if (!list_empty(&NFS_I(inode)->open_files))
715 		return;
716 	server = NFS_SERVER(inode);
717 	if (server->flags & NFS_MOUNT_NOCTO)
718 		return;
719 	nfs_revalidate_inode(server, inode);
720 }
721 EXPORT_SYMBOL_GPL(nfs_close_context);
722 
723 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
724 {
725 	struct nfs_open_context *ctx;
726 	struct rpc_cred *cred = rpc_lookup_cred();
727 	if (IS_ERR(cred))
728 		return ERR_CAST(cred);
729 
730 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
731 	if (!ctx) {
732 		put_rpccred(cred);
733 		return ERR_PTR(-ENOMEM);
734 	}
735 	nfs_sb_active(dentry->d_sb);
736 	ctx->dentry = dget(dentry);
737 	ctx->cred = cred;
738 	ctx->state = NULL;
739 	ctx->mode = f_mode;
740 	ctx->flags = 0;
741 	ctx->error = 0;
742 	nfs_init_lock_context(&ctx->lock_context);
743 	ctx->lock_context.open_context = ctx;
744 	INIT_LIST_HEAD(&ctx->list);
745 	ctx->mdsthreshold = NULL;
746 	return ctx;
747 }
748 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
749 
750 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
751 {
752 	if (ctx != NULL)
753 		atomic_inc(&ctx->lock_context.count);
754 	return ctx;
755 }
756 EXPORT_SYMBOL_GPL(get_nfs_open_context);
757 
758 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
759 {
760 	struct inode *inode = ctx->dentry->d_inode;
761 	struct super_block *sb = ctx->dentry->d_sb;
762 
763 	if (!list_empty(&ctx->list)) {
764 		if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
765 			return;
766 		list_del(&ctx->list);
767 		spin_unlock(&inode->i_lock);
768 	} else if (!atomic_dec_and_test(&ctx->lock_context.count))
769 		return;
770 	if (inode != NULL)
771 		NFS_PROTO(inode)->close_context(ctx, is_sync);
772 	if (ctx->cred != NULL)
773 		put_rpccred(ctx->cred);
774 	dput(ctx->dentry);
775 	nfs_sb_deactive(sb);
776 	kfree(ctx->mdsthreshold);
777 	kfree(ctx);
778 }
779 
780 void put_nfs_open_context(struct nfs_open_context *ctx)
781 {
782 	__put_nfs_open_context(ctx, 0);
783 }
784 EXPORT_SYMBOL_GPL(put_nfs_open_context);
785 
786 /*
787  * Ensure that mmap has a recent RPC credential for use when writing out
788  * shared pages
789  */
790 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
791 {
792 	struct inode *inode = ctx->dentry->d_inode;
793 	struct nfs_inode *nfsi = NFS_I(inode);
794 
795 	spin_lock(&inode->i_lock);
796 	list_add(&ctx->list, &nfsi->open_files);
797 	spin_unlock(&inode->i_lock);
798 }
799 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
800 
801 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
802 {
803 	filp->private_data = get_nfs_open_context(ctx);
804 	if (list_empty(&ctx->list))
805 		nfs_inode_attach_open_context(ctx);
806 }
807 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
808 
809 /*
810  * Given an inode, search for an open context with the desired characteristics
811  */
812 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
813 {
814 	struct nfs_inode *nfsi = NFS_I(inode);
815 	struct nfs_open_context *pos, *ctx = NULL;
816 
817 	spin_lock(&inode->i_lock);
818 	list_for_each_entry(pos, &nfsi->open_files, list) {
819 		if (cred != NULL && pos->cred != cred)
820 			continue;
821 		if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
822 			continue;
823 		ctx = get_nfs_open_context(pos);
824 		break;
825 	}
826 	spin_unlock(&inode->i_lock);
827 	return ctx;
828 }
829 
830 static void nfs_file_clear_open_context(struct file *filp)
831 {
832 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
833 
834 	if (ctx) {
835 		struct inode *inode = ctx->dentry->d_inode;
836 
837 		filp->private_data = NULL;
838 		spin_lock(&inode->i_lock);
839 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
840 		spin_unlock(&inode->i_lock);
841 		__put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
842 	}
843 }
844 
845 /*
846  * These allocate and release file read/write context information.
847  */
848 int nfs_open(struct inode *inode, struct file *filp)
849 {
850 	struct nfs_open_context *ctx;
851 
852 	ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
853 	if (IS_ERR(ctx))
854 		return PTR_ERR(ctx);
855 	nfs_file_set_open_context(filp, ctx);
856 	put_nfs_open_context(ctx);
857 	nfs_fscache_set_inode_cookie(inode, filp);
858 	return 0;
859 }
860 
861 int nfs_release(struct inode *inode, struct file *filp)
862 {
863 	nfs_file_clear_open_context(filp);
864 	return 0;
865 }
866 
867 /*
868  * This function is called whenever some part of NFS notices that
869  * the cached attributes have to be refreshed.
870  */
871 int
872 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
873 {
874 	int		 status = -ESTALE;
875 	struct nfs4_label *label = NULL;
876 	struct nfs_fattr *fattr = NULL;
877 	struct nfs_inode *nfsi = NFS_I(inode);
878 
879 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
880 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
881 
882 	trace_nfs_revalidate_inode_enter(inode);
883 
884 	if (is_bad_inode(inode))
885 		goto out;
886 	if (NFS_STALE(inode))
887 		goto out;
888 
889 	status = -ENOMEM;
890 	fattr = nfs_alloc_fattr();
891 	if (fattr == NULL)
892 		goto out;
893 
894 	nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
895 
896 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
897 	if (IS_ERR(label)) {
898 		status = PTR_ERR(label);
899 		goto out;
900 	}
901 
902 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
903 	if (status != 0) {
904 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
905 			 inode->i_sb->s_id,
906 			 (long long)NFS_FILEID(inode), status);
907 		if (status == -ESTALE) {
908 			nfs_zap_caches(inode);
909 			if (!S_ISDIR(inode->i_mode))
910 				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
911 		}
912 		goto err_out;
913 	}
914 
915 	status = nfs_refresh_inode(inode, fattr);
916 	if (status) {
917 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
918 			 inode->i_sb->s_id,
919 			 (long long)NFS_FILEID(inode), status);
920 		goto err_out;
921 	}
922 
923 	if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
924 		nfs_zap_acl_cache(inode);
925 
926 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
927 		inode->i_sb->s_id,
928 		(long long)NFS_FILEID(inode));
929 
930 err_out:
931 	nfs4_label_free(label);
932 out:
933 	nfs_free_fattr(fattr);
934 	trace_nfs_revalidate_inode_exit(inode, status);
935 	return status;
936 }
937 
938 int nfs_attribute_timeout(struct inode *inode)
939 {
940 	struct nfs_inode *nfsi = NFS_I(inode);
941 
942 	return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
943 }
944 
945 int nfs_attribute_cache_expired(struct inode *inode)
946 {
947 	if (nfs_have_delegated_attributes(inode))
948 		return 0;
949 	return nfs_attribute_timeout(inode);
950 }
951 
952 /**
953  * nfs_revalidate_inode - Revalidate the inode attributes
954  * @server - pointer to nfs_server struct
955  * @inode - pointer to inode struct
956  *
957  * Updates inode attribute information by retrieving the data from the server.
958  */
959 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
960 {
961 	if (!(NFS_I(inode)->cache_validity &
962 			(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
963 			&& !nfs_attribute_cache_expired(inode))
964 		return NFS_STALE(inode) ? -ESTALE : 0;
965 	return __nfs_revalidate_inode(server, inode);
966 }
967 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
968 
969 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
970 {
971 	struct nfs_inode *nfsi = NFS_I(inode);
972 	int ret;
973 
974 	if (mapping->nrpages != 0) {
975 		if (S_ISREG(inode->i_mode)) {
976 			ret = nfs_sync_mapping(mapping);
977 			if (ret < 0)
978 				return ret;
979 		}
980 		ret = invalidate_inode_pages2(mapping);
981 		if (ret < 0)
982 			return ret;
983 	}
984 	spin_lock(&inode->i_lock);
985 	nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
986 	if (S_ISDIR(inode->i_mode))
987 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
988 	spin_unlock(&inode->i_lock);
989 	nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
990 	nfs_fscache_wait_on_invalidate(inode);
991 
992 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
993 			inode->i_sb->s_id, (long long)NFS_FILEID(inode));
994 	return 0;
995 }
996 
997 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
998 {
999 	if (nfs_have_delegated_attributes(inode))
1000 		return false;
1001 	return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1002 		|| nfs_attribute_timeout(inode)
1003 		|| NFS_STALE(inode);
1004 }
1005 
1006 /**
1007  * nfs_revalidate_mapping - Revalidate the pagecache
1008  * @inode - pointer to host inode
1009  * @mapping - pointer to mapping
1010  */
1011 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1012 {
1013 	struct nfs_inode *nfsi = NFS_I(inode);
1014 	int ret = 0;
1015 
1016 	/* swapfiles are not supposed to be shared. */
1017 	if (IS_SWAPFILE(inode))
1018 		goto out;
1019 
1020 	if (nfs_mapping_need_revalidate_inode(inode)) {
1021 		ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1022 		if (ret < 0)
1023 			goto out;
1024 	}
1025 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1026 		trace_nfs_invalidate_mapping_enter(inode);
1027 		ret = nfs_invalidate_mapping(inode, mapping);
1028 		trace_nfs_invalidate_mapping_exit(inode, ret);
1029 	}
1030 
1031 out:
1032 	return ret;
1033 }
1034 
1035 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1036 {
1037 	struct nfs_inode *nfsi = NFS_I(inode);
1038 	unsigned long ret = 0;
1039 
1040 	if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1041 			&& (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1042 			&& inode->i_version == fattr->pre_change_attr) {
1043 		inode->i_version = fattr->change_attr;
1044 		if (S_ISDIR(inode->i_mode))
1045 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1046 		ret |= NFS_INO_INVALID_ATTR;
1047 	}
1048 	/* If we have atomic WCC data, we may update some attributes */
1049 	if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1050 			&& (fattr->valid & NFS_ATTR_FATTR_CTIME)
1051 			&& timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1052 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1053 		ret |= NFS_INO_INVALID_ATTR;
1054 	}
1055 
1056 	if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1057 			&& (fattr->valid & NFS_ATTR_FATTR_MTIME)
1058 			&& timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1059 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1060 		if (S_ISDIR(inode->i_mode))
1061 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1062 		ret |= NFS_INO_INVALID_ATTR;
1063 	}
1064 	if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1065 			&& (fattr->valid & NFS_ATTR_FATTR_SIZE)
1066 			&& i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1067 			&& nfsi->npages == 0) {
1068 		i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1069 		ret |= NFS_INO_INVALID_ATTR;
1070 	}
1071 
1072 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1073 		nfs_fscache_invalidate(inode);
1074 
1075 	return ret;
1076 }
1077 
1078 /**
1079  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1080  * @inode - pointer to inode
1081  * @fattr - updated attributes
1082  *
1083  * Verifies the attribute cache. If we have just changed the attributes,
1084  * so that fattr carries weak cache consistency data, then it may
1085  * also update the ctime/mtime/change_attribute.
1086  */
1087 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1088 {
1089 	struct nfs_inode *nfsi = NFS_I(inode);
1090 	loff_t cur_size, new_isize;
1091 	unsigned long invalid = 0;
1092 
1093 
1094 	if (nfs_have_delegated_attributes(inode))
1095 		return 0;
1096 	/* Has the inode gone and changed behind our back? */
1097 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1098 		return -EIO;
1099 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1100 		return -EIO;
1101 
1102 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1103 			inode->i_version != fattr->change_attr)
1104 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1105 
1106 	/* Verify a few of the more important attributes */
1107 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1108 		invalid |= NFS_INO_INVALID_ATTR;
1109 
1110 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1111 		cur_size = i_size_read(inode);
1112 		new_isize = nfs_size_to_loff_t(fattr->size);
1113 		if (cur_size != new_isize && nfsi->npages == 0)
1114 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1115 	}
1116 
1117 	/* Have any file permissions changed? */
1118 	if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1119 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1120 	if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1121 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1122 	if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1123 		invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1124 
1125 	/* Has the link count changed? */
1126 	if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1127 		invalid |= NFS_INO_INVALID_ATTR;
1128 
1129 	if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1130 		invalid |= NFS_INO_INVALID_ATIME;
1131 
1132 	if (invalid != 0)
1133 		nfsi->cache_validity |= invalid;
1134 
1135 	nfsi->read_cache_jiffies = fattr->time_start;
1136 	return 0;
1137 }
1138 
1139 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1140 {
1141 	if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1142 		return 0;
1143 	return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1144 }
1145 
1146 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1147 {
1148 	if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1149 		return 0;
1150 	return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1151 }
1152 
1153 static atomic_long_t nfs_attr_generation_counter;
1154 
1155 static unsigned long nfs_read_attr_generation_counter(void)
1156 {
1157 	return atomic_long_read(&nfs_attr_generation_counter);
1158 }
1159 
1160 unsigned long nfs_inc_attr_generation_counter(void)
1161 {
1162 	return atomic_long_inc_return(&nfs_attr_generation_counter);
1163 }
1164 
1165 void nfs_fattr_init(struct nfs_fattr *fattr)
1166 {
1167 	fattr->valid = 0;
1168 	fattr->time_start = jiffies;
1169 	fattr->gencount = nfs_inc_attr_generation_counter();
1170 	fattr->owner_name = NULL;
1171 	fattr->group_name = NULL;
1172 }
1173 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1174 
1175 struct nfs_fattr *nfs_alloc_fattr(void)
1176 {
1177 	struct nfs_fattr *fattr;
1178 
1179 	fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1180 	if (fattr != NULL)
1181 		nfs_fattr_init(fattr);
1182 	return fattr;
1183 }
1184 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1185 
1186 struct nfs_fh *nfs_alloc_fhandle(void)
1187 {
1188 	struct nfs_fh *fh;
1189 
1190 	fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1191 	if (fh != NULL)
1192 		fh->size = 0;
1193 	return fh;
1194 }
1195 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1196 
1197 #ifdef NFS_DEBUG
1198 /*
1199  * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1200  *                             in the same way that wireshark does
1201  *
1202  * @fh: file handle
1203  *
1204  * For debugging only.
1205  */
1206 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1207 {
1208 	/* wireshark uses 32-bit AUTODIN crc and does a bitwise
1209 	 * not on the result */
1210 	return nfs_fhandle_hash(fh);
1211 }
1212 
1213 /*
1214  * _nfs_display_fhandle - display an NFS file handle on the console
1215  *
1216  * @fh: file handle to display
1217  * @caption: display caption
1218  *
1219  * For debugging only.
1220  */
1221 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1222 {
1223 	unsigned short i;
1224 
1225 	if (fh == NULL || fh->size == 0) {
1226 		printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1227 		return;
1228 	}
1229 
1230 	printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1231 	       caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1232 	for (i = 0; i < fh->size; i += 16) {
1233 		__be32 *pos = (__be32 *)&fh->data[i];
1234 
1235 		switch ((fh->size - i - 1) >> 2) {
1236 		case 0:
1237 			printk(KERN_DEFAULT " %08x\n",
1238 				be32_to_cpup(pos));
1239 			break;
1240 		case 1:
1241 			printk(KERN_DEFAULT " %08x %08x\n",
1242 				be32_to_cpup(pos), be32_to_cpup(pos + 1));
1243 			break;
1244 		case 2:
1245 			printk(KERN_DEFAULT " %08x %08x %08x\n",
1246 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1247 				be32_to_cpup(pos + 2));
1248 			break;
1249 		default:
1250 			printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1251 				be32_to_cpup(pos), be32_to_cpup(pos + 1),
1252 				be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1253 		}
1254 	}
1255 }
1256 #endif
1257 
1258 /**
1259  * nfs_inode_attrs_need_update - check if the inode attributes need updating
1260  * @inode - pointer to inode
1261  * @fattr - attributes
1262  *
1263  * Attempt to divine whether or not an RPC call reply carrying stale
1264  * attributes got scheduled after another call carrying updated ones.
1265  *
1266  * To do so, the function first assumes that a more recent ctime means
1267  * that the attributes in fattr are newer, however it also attempt to
1268  * catch the case where ctime either didn't change, or went backwards
1269  * (if someone reset the clock on the server) by looking at whether
1270  * or not this RPC call was started after the inode was last updated.
1271  * Note also the check for wraparound of 'attr_gencount'
1272  *
1273  * The function returns 'true' if it thinks the attributes in 'fattr' are
1274  * more recent than the ones cached in the inode.
1275  *
1276  */
1277 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1278 {
1279 	const struct nfs_inode *nfsi = NFS_I(inode);
1280 
1281 	return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1282 		nfs_ctime_need_update(inode, fattr) ||
1283 		nfs_size_need_update(inode, fattr) ||
1284 		((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1285 }
1286 
1287 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1288 {
1289 	int ret;
1290 
1291 	trace_nfs_refresh_inode_enter(inode);
1292 
1293 	if (nfs_inode_attrs_need_update(inode, fattr))
1294 		ret = nfs_update_inode(inode, fattr);
1295 	else
1296 		ret = nfs_check_inode_attributes(inode, fattr);
1297 
1298 	trace_nfs_refresh_inode_exit(inode, ret);
1299 	return ret;
1300 }
1301 
1302 /**
1303  * nfs_refresh_inode - try to update the inode attribute cache
1304  * @inode - pointer to inode
1305  * @fattr - updated attributes
1306  *
1307  * Check that an RPC call that returned attributes has not overlapped with
1308  * other recent updates of the inode metadata, then decide whether it is
1309  * safe to do a full update of the inode attributes, or whether just to
1310  * call nfs_check_inode_attributes.
1311  */
1312 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1313 {
1314 	int status;
1315 
1316 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1317 		return 0;
1318 	spin_lock(&inode->i_lock);
1319 	status = nfs_refresh_inode_locked(inode, fattr);
1320 	spin_unlock(&inode->i_lock);
1321 
1322 	return status;
1323 }
1324 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1325 
1326 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1327 {
1328 	struct nfs_inode *nfsi = NFS_I(inode);
1329 
1330 	nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1331 	if (S_ISDIR(inode->i_mode)) {
1332 		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1333 		nfs_fscache_invalidate(inode);
1334 	}
1335 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1336 		return 0;
1337 	return nfs_refresh_inode_locked(inode, fattr);
1338 }
1339 
1340 /**
1341  * nfs_post_op_update_inode - try to update the inode attribute cache
1342  * @inode - pointer to inode
1343  * @fattr - updated attributes
1344  *
1345  * After an operation that has changed the inode metadata, mark the
1346  * attribute cache as being invalid, then try to update it.
1347  *
1348  * NB: if the server didn't return any post op attributes, this
1349  * function will force the retrieval of attributes before the next
1350  * NFS request.  Thus it should be used only for operations that
1351  * are expected to change one or more attributes, to avoid
1352  * unnecessary NFS requests and trips through nfs_update_inode().
1353  */
1354 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1355 {
1356 	int status;
1357 
1358 	spin_lock(&inode->i_lock);
1359 	status = nfs_post_op_update_inode_locked(inode, fattr);
1360 	spin_unlock(&inode->i_lock);
1361 
1362 	return status;
1363 }
1364 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1365 
1366 /**
1367  * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1368  * @inode - pointer to inode
1369  * @fattr - updated attributes
1370  *
1371  * After an operation that has changed the inode metadata, mark the
1372  * attribute cache as being invalid, then try to update it. Fake up
1373  * weak cache consistency data, if none exist.
1374  *
1375  * This function is mainly designed to be used by the ->write_done() functions.
1376  */
1377 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1378 {
1379 	int status;
1380 
1381 	spin_lock(&inode->i_lock);
1382 	/* Don't do a WCC update if these attributes are already stale */
1383 	if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1384 			!nfs_inode_attrs_need_update(inode, fattr)) {
1385 		fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1386 				| NFS_ATTR_FATTR_PRESIZE
1387 				| NFS_ATTR_FATTR_PREMTIME
1388 				| NFS_ATTR_FATTR_PRECTIME);
1389 		goto out_noforce;
1390 	}
1391 	if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1392 			(fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1393 		fattr->pre_change_attr = inode->i_version;
1394 		fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1395 	}
1396 	if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1397 			(fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1398 		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1399 		fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1400 	}
1401 	if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1402 			(fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1403 		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1404 		fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1405 	}
1406 	if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1407 			(fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1408 		fattr->pre_size = i_size_read(inode);
1409 		fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1410 	}
1411 out_noforce:
1412 	status = nfs_post_op_update_inode_locked(inode, fattr);
1413 	spin_unlock(&inode->i_lock);
1414 	return status;
1415 }
1416 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1417 
1418 /*
1419  * Many nfs protocol calls return the new file attributes after
1420  * an operation.  Here we update the inode to reflect the state
1421  * of the server's inode.
1422  *
1423  * This is a bit tricky because we have to make sure all dirty pages
1424  * have been sent off to the server before calling invalidate_inode_pages.
1425  * To make sure no other process adds more write requests while we try
1426  * our best to flush them, we make them sleep during the attribute refresh.
1427  *
1428  * A very similar scenario holds for the dir cache.
1429  */
1430 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1431 {
1432 	struct nfs_server *server;
1433 	struct nfs_inode *nfsi = NFS_I(inode);
1434 	loff_t cur_isize, new_isize;
1435 	unsigned long invalid = 0;
1436 	unsigned long now = jiffies;
1437 	unsigned long save_cache_validity;
1438 
1439 	dfprintk(VFS, "NFS: %s(%s/%ld fh_crc=0x%08x ct=%d info=0x%x)\n",
1440 			__func__, inode->i_sb->s_id, inode->i_ino,
1441 			nfs_display_fhandle_hash(NFS_FH(inode)),
1442 			atomic_read(&inode->i_count), fattr->valid);
1443 
1444 	if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1445 		printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1446 			"fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1447 			NFS_SERVER(inode)->nfs_client->cl_hostname,
1448 			inode->i_sb->s_id, (long long)nfsi->fileid,
1449 			(long long)fattr->fileid);
1450 		goto out_err;
1451 	}
1452 
1453 	/*
1454 	 * Make sure the inode's type hasn't changed.
1455 	 */
1456 	if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1457 		/*
1458 		* Big trouble! The inode has become a different object.
1459 		*/
1460 		printk(KERN_DEBUG "NFS: %s: inode %ld mode changed, %07o to %07o\n",
1461 				__func__, inode->i_ino, inode->i_mode, fattr->mode);
1462 		goto out_err;
1463 	}
1464 
1465 	server = NFS_SERVER(inode);
1466 	/* Update the fsid? */
1467 	if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1468 			!nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1469 			!IS_AUTOMOUNT(inode))
1470 		server->fsid = fattr->fsid;
1471 
1472 	/*
1473 	 * Update the read time so we don't revalidate too often.
1474 	 */
1475 	nfsi->read_cache_jiffies = fattr->time_start;
1476 
1477 	save_cache_validity = nfsi->cache_validity;
1478 	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1479 			| NFS_INO_INVALID_ATIME
1480 			| NFS_INO_REVAL_FORCED
1481 			| NFS_INO_REVAL_PAGECACHE);
1482 
1483 	/* Do atomic weak cache consistency updates */
1484 	invalid |= nfs_wcc_update_inode(inode, fattr);
1485 
1486 	/* More cache consistency checks */
1487 	if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1488 		if (inode->i_version != fattr->change_attr) {
1489 			dprintk("NFS: change_attr change on server for file %s/%ld\n",
1490 					inode->i_sb->s_id, inode->i_ino);
1491 			invalid |= NFS_INO_INVALID_ATTR
1492 				| NFS_INO_INVALID_DATA
1493 				| NFS_INO_INVALID_ACCESS
1494 				| NFS_INO_INVALID_ACL
1495 				| NFS_INO_REVAL_PAGECACHE;
1496 			if (S_ISDIR(inode->i_mode))
1497 				nfs_force_lookup_revalidate(inode);
1498 			inode->i_version = fattr->change_attr;
1499 		}
1500 	} else if (server->caps & NFS_CAP_CHANGE_ATTR)
1501 		invalid |= save_cache_validity;
1502 
1503 	if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1504 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1505 	} else if (server->caps & NFS_CAP_MTIME)
1506 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1507 				| NFS_INO_REVAL_FORCED);
1508 
1509 	if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1510 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1511 	} else if (server->caps & NFS_CAP_CTIME)
1512 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1513 				| NFS_INO_REVAL_FORCED);
1514 
1515 	/* Check if our cached file size is stale */
1516 	if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1517 		new_isize = nfs_size_to_loff_t(fattr->size);
1518 		cur_isize = i_size_read(inode);
1519 		if (new_isize != cur_isize) {
1520 			/* Do we perhaps have any outstanding writes, or has
1521 			 * the file grown beyond our last write? */
1522 			if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) ||
1523 			     new_isize > cur_isize) {
1524 				i_size_write(inode, new_isize);
1525 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1526 			}
1527 			dprintk("NFS: isize change on server for file %s/%ld "
1528 					"(%Ld to %Ld)\n",
1529 					inode->i_sb->s_id,
1530 					inode->i_ino,
1531 					(long long)cur_isize,
1532 					(long long)new_isize);
1533 		}
1534 	} else
1535 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1536 				| NFS_INO_REVAL_PAGECACHE
1537 				| NFS_INO_REVAL_FORCED);
1538 
1539 
1540 	if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1541 		memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1542 	else if (server->caps & NFS_CAP_ATIME)
1543 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1544 				| NFS_INO_REVAL_FORCED);
1545 
1546 	if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1547 		if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1548 			umode_t newmode = inode->i_mode & S_IFMT;
1549 			newmode |= fattr->mode & S_IALLUGO;
1550 			inode->i_mode = newmode;
1551 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1552 		}
1553 	} else if (server->caps & NFS_CAP_MODE)
1554 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1555 				| NFS_INO_INVALID_ACCESS
1556 				| NFS_INO_INVALID_ACL
1557 				| NFS_INO_REVAL_FORCED);
1558 
1559 	if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1560 		if (!uid_eq(inode->i_uid, fattr->uid)) {
1561 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1562 			inode->i_uid = fattr->uid;
1563 		}
1564 	} else if (server->caps & NFS_CAP_OWNER)
1565 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1566 				| NFS_INO_INVALID_ACCESS
1567 				| NFS_INO_INVALID_ACL
1568 				| NFS_INO_REVAL_FORCED);
1569 
1570 	if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1571 		if (!gid_eq(inode->i_gid, fattr->gid)) {
1572 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1573 			inode->i_gid = fattr->gid;
1574 		}
1575 	} else if (server->caps & NFS_CAP_OWNER_GROUP)
1576 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1577 				| NFS_INO_INVALID_ACCESS
1578 				| NFS_INO_INVALID_ACL
1579 				| NFS_INO_REVAL_FORCED);
1580 
1581 	if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1582 		if (inode->i_nlink != fattr->nlink) {
1583 			invalid |= NFS_INO_INVALID_ATTR;
1584 			if (S_ISDIR(inode->i_mode))
1585 				invalid |= NFS_INO_INVALID_DATA;
1586 			set_nlink(inode, fattr->nlink);
1587 		}
1588 	} else if (server->caps & NFS_CAP_NLINK)
1589 		invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1590 				| NFS_INO_REVAL_FORCED);
1591 
1592 	if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1593 		/*
1594 		 * report the blocks in 512byte units
1595 		 */
1596 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1597  	}
1598 	if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1599 		inode->i_blocks = fattr->du.nfs2.blocks;
1600 
1601 	/* Update attrtimeo value if we're out of the unstable period */
1602 	if (invalid & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL)) {
1603 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1604 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1605 		nfsi->attrtimeo_timestamp = now;
1606 		nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1607 	} else {
1608 		if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1609 			if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1610 				nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1611 			nfsi->attrtimeo_timestamp = now;
1612 		}
1613 	}
1614 	invalid &= ~NFS_INO_INVALID_ATTR;
1615 	invalid &= ~NFS_INO_INVALID_LABEL;
1616 	/* Don't invalidate the data if we were to blame */
1617 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1618 				|| S_ISLNK(inode->i_mode)))
1619 		invalid &= ~NFS_INO_INVALID_DATA;
1620 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1621 			(save_cache_validity & NFS_INO_REVAL_FORCED))
1622 		nfsi->cache_validity |= invalid;
1623 
1624 	if (invalid & NFS_INO_INVALID_DATA)
1625 		nfs_fscache_invalidate(inode);
1626 
1627 	return 0;
1628  out_err:
1629 	/*
1630 	 * No need to worry about unhashing the dentry, as the
1631 	 * lookup validation will know that the inode is bad.
1632 	 * (But we fall through to invalidate the caches.)
1633 	 */
1634 	nfs_invalidate_inode(inode);
1635 	return -ESTALE;
1636 }
1637 
1638 struct inode *nfs_alloc_inode(struct super_block *sb)
1639 {
1640 	struct nfs_inode *nfsi;
1641 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1642 	if (!nfsi)
1643 		return NULL;
1644 	nfsi->flags = 0UL;
1645 	nfsi->cache_validity = 0UL;
1646 #ifdef CONFIG_NFS_V3_ACL
1647 	nfsi->acl_access = ERR_PTR(-EAGAIN);
1648 	nfsi->acl_default = ERR_PTR(-EAGAIN);
1649 #endif
1650 #if IS_ENABLED(CONFIG_NFS_V4)
1651 	nfsi->nfs4_acl = NULL;
1652 #endif /* CONFIG_NFS_V4 */
1653 	return &nfsi->vfs_inode;
1654 }
1655 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1656 
1657 static void nfs_i_callback(struct rcu_head *head)
1658 {
1659 	struct inode *inode = container_of(head, struct inode, i_rcu);
1660 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1661 }
1662 
1663 void nfs_destroy_inode(struct inode *inode)
1664 {
1665 	call_rcu(&inode->i_rcu, nfs_i_callback);
1666 }
1667 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1668 
1669 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1670 {
1671 #if IS_ENABLED(CONFIG_NFS_V4)
1672 	INIT_LIST_HEAD(&nfsi->open_states);
1673 	nfsi->delegation = NULL;
1674 	nfsi->delegation_state = 0;
1675 	init_rwsem(&nfsi->rwsem);
1676 	nfsi->layout = NULL;
1677 #endif
1678 }
1679 
1680 static void init_once(void *foo)
1681 {
1682 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1683 
1684 	inode_init_once(&nfsi->vfs_inode);
1685 	INIT_LIST_HEAD(&nfsi->open_files);
1686 	INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1687 	INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1688 	INIT_LIST_HEAD(&nfsi->commit_info.list);
1689 	nfsi->npages = 0;
1690 	nfsi->commit_info.ncommit = 0;
1691 	atomic_set(&nfsi->commit_info.rpcs_out, 0);
1692 	atomic_set(&nfsi->silly_count, 1);
1693 	INIT_HLIST_HEAD(&nfsi->silly_list);
1694 	init_waitqueue_head(&nfsi->waitqueue);
1695 	nfs4_init_once(nfsi);
1696 }
1697 
1698 static int __init nfs_init_inodecache(void)
1699 {
1700 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1701 					     sizeof(struct nfs_inode),
1702 					     0, (SLAB_RECLAIM_ACCOUNT|
1703 						SLAB_MEM_SPREAD),
1704 					     init_once);
1705 	if (nfs_inode_cachep == NULL)
1706 		return -ENOMEM;
1707 
1708 	return 0;
1709 }
1710 
1711 static void nfs_destroy_inodecache(void)
1712 {
1713 	/*
1714 	 * Make sure all delayed rcu free inodes are flushed before we
1715 	 * destroy cache.
1716 	 */
1717 	rcu_barrier();
1718 	kmem_cache_destroy(nfs_inode_cachep);
1719 }
1720 
1721 struct workqueue_struct *nfsiod_workqueue;
1722 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1723 
1724 /*
1725  * start up the nfsiod workqueue
1726  */
1727 static int nfsiod_start(void)
1728 {
1729 	struct workqueue_struct *wq;
1730 	dprintk("RPC:       creating workqueue nfsiod\n");
1731 	wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1732 	if (wq == NULL)
1733 		return -ENOMEM;
1734 	nfsiod_workqueue = wq;
1735 	return 0;
1736 }
1737 
1738 /*
1739  * Destroy the nfsiod workqueue
1740  */
1741 static void nfsiod_stop(void)
1742 {
1743 	struct workqueue_struct *wq;
1744 
1745 	wq = nfsiod_workqueue;
1746 	if (wq == NULL)
1747 		return;
1748 	nfsiod_workqueue = NULL;
1749 	destroy_workqueue(wq);
1750 }
1751 
1752 int nfs_net_id;
1753 EXPORT_SYMBOL_GPL(nfs_net_id);
1754 
1755 static int nfs_net_init(struct net *net)
1756 {
1757 	nfs_clients_init(net);
1758 	return 0;
1759 }
1760 
1761 static void nfs_net_exit(struct net *net)
1762 {
1763 	nfs_cleanup_cb_ident_idr(net);
1764 }
1765 
1766 static struct pernet_operations nfs_net_ops = {
1767 	.init = nfs_net_init,
1768 	.exit = nfs_net_exit,
1769 	.id   = &nfs_net_id,
1770 	.size = sizeof(struct nfs_net),
1771 };
1772 
1773 /*
1774  * Initialize NFS
1775  */
1776 static int __init init_nfs_fs(void)
1777 {
1778 	int err;
1779 
1780 	err = register_pernet_subsys(&nfs_net_ops);
1781 	if (err < 0)
1782 		goto out9;
1783 
1784 	err = nfs_fscache_register();
1785 	if (err < 0)
1786 		goto out8;
1787 
1788 	err = nfsiod_start();
1789 	if (err)
1790 		goto out7;
1791 
1792 	err = nfs_fs_proc_init();
1793 	if (err)
1794 		goto out6;
1795 
1796 	err = nfs_init_nfspagecache();
1797 	if (err)
1798 		goto out5;
1799 
1800 	err = nfs_init_inodecache();
1801 	if (err)
1802 		goto out4;
1803 
1804 	err = nfs_init_readpagecache();
1805 	if (err)
1806 		goto out3;
1807 
1808 	err = nfs_init_writepagecache();
1809 	if (err)
1810 		goto out2;
1811 
1812 	err = nfs_init_directcache();
1813 	if (err)
1814 		goto out1;
1815 
1816 #ifdef CONFIG_PROC_FS
1817 	rpc_proc_register(&init_net, &nfs_rpcstat);
1818 #endif
1819 	if ((err = register_nfs_fs()) != 0)
1820 		goto out0;
1821 
1822 	return 0;
1823 out0:
1824 #ifdef CONFIG_PROC_FS
1825 	rpc_proc_unregister(&init_net, "nfs");
1826 #endif
1827 	nfs_destroy_directcache();
1828 out1:
1829 	nfs_destroy_writepagecache();
1830 out2:
1831 	nfs_destroy_readpagecache();
1832 out3:
1833 	nfs_destroy_inodecache();
1834 out4:
1835 	nfs_destroy_nfspagecache();
1836 out5:
1837 	nfs_fs_proc_exit();
1838 out6:
1839 	nfsiod_stop();
1840 out7:
1841 	nfs_fscache_unregister();
1842 out8:
1843 	unregister_pernet_subsys(&nfs_net_ops);
1844 out9:
1845 	return err;
1846 }
1847 
1848 static void __exit exit_nfs_fs(void)
1849 {
1850 	nfs_destroy_directcache();
1851 	nfs_destroy_writepagecache();
1852 	nfs_destroy_readpagecache();
1853 	nfs_destroy_inodecache();
1854 	nfs_destroy_nfspagecache();
1855 	nfs_fscache_unregister();
1856 	unregister_pernet_subsys(&nfs_net_ops);
1857 #ifdef CONFIG_PROC_FS
1858 	rpc_proc_unregister(&init_net, "nfs");
1859 #endif
1860 	unregister_nfs_fs();
1861 	nfs_fs_proc_exit();
1862 	nfsiod_stop();
1863 }
1864 
1865 /* Not quite true; I just maintain it */
1866 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1867 MODULE_LICENSE("GPL");
1868 module_param(enable_ino64, bool, 0644);
1869 
1870 module_init(init_nfs_fs)
1871 module_exit(exit_nfs_fs)
1872