xref: /openbmc/linux/fs/nfs/inode.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
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.Cox@linux.org>, 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/config.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
19 
20 #include <linux/time.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
38 
39 #include <asm/system.h>
40 #include <asm/uaccess.h>
41 
42 #include "nfs4_fs.h"
43 #include "delegation.h"
44 
45 #define NFSDBG_FACILITY		NFSDBG_VFS
46 #define NFS_PARANOIA 1
47 
48 /* Maximum number of readahead requests
49  * FIXME: this should really be a sysctl so that users may tune it to suit
50  *        their needs. People that do NFS over a slow network, might for
51  *        instance want to reduce it to something closer to 1 for improved
52  *        interactive response.
53  */
54 #define NFS_MAX_READAHEAD	(RPC_DEF_SLOT_TABLE - 1)
55 
56 static void nfs_invalidate_inode(struct inode *);
57 static int nfs_update_inode(struct inode *, struct nfs_fattr *, unsigned long);
58 
59 static struct inode *nfs_alloc_inode(struct super_block *sb);
60 static void nfs_destroy_inode(struct inode *);
61 static int nfs_write_inode(struct inode *,int);
62 static void nfs_delete_inode(struct inode *);
63 static void nfs_clear_inode(struct inode *);
64 static void nfs_umount_begin(struct super_block *);
65 static int  nfs_statfs(struct super_block *, struct kstatfs *);
66 static int  nfs_show_options(struct seq_file *, struct vfsmount *);
67 static void nfs_zap_acl_cache(struct inode *);
68 
69 static struct rpc_program	nfs_program;
70 
71 static struct super_operations nfs_sops = {
72 	.alloc_inode	= nfs_alloc_inode,
73 	.destroy_inode	= nfs_destroy_inode,
74 	.write_inode	= nfs_write_inode,
75 	.delete_inode	= nfs_delete_inode,
76 	.statfs		= nfs_statfs,
77 	.clear_inode	= nfs_clear_inode,
78 	.umount_begin	= nfs_umount_begin,
79 	.show_options	= nfs_show_options,
80 };
81 
82 /*
83  * RPC cruft for NFS
84  */
85 static struct rpc_stat		nfs_rpcstat = {
86 	.program		= &nfs_program
87 };
88 static struct rpc_version *	nfs_version[] = {
89 	NULL,
90 	NULL,
91 	&nfs_version2,
92 #if defined(CONFIG_NFS_V3)
93 	&nfs_version3,
94 #elif defined(CONFIG_NFS_V4)
95 	NULL,
96 #endif
97 #if defined(CONFIG_NFS_V4)
98 	&nfs_version4,
99 #endif
100 };
101 
102 static struct rpc_program	nfs_program = {
103 	.name			= "nfs",
104 	.number			= NFS_PROGRAM,
105 	.nrvers			= sizeof(nfs_version) / sizeof(nfs_version[0]),
106 	.version		= nfs_version,
107 	.stats			= &nfs_rpcstat,
108 	.pipe_dir_name		= "/nfs",
109 };
110 
111 #ifdef CONFIG_NFS_V3_ACL
112 static struct rpc_stat		nfsacl_rpcstat = { &nfsacl_program };
113 static struct rpc_version *	nfsacl_version[] = {
114 	[3]			= &nfsacl_version3,
115 };
116 
117 struct rpc_program		nfsacl_program = {
118 	.name =			"nfsacl",
119 	.number =		NFS_ACL_PROGRAM,
120 	.nrvers =		sizeof(nfsacl_version) / sizeof(nfsacl_version[0]),
121 	.version =		nfsacl_version,
122 	.stats =		&nfsacl_rpcstat,
123 };
124 #endif  /* CONFIG_NFS_V3_ACL */
125 
126 static inline unsigned long
127 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
128 {
129 	return nfs_fileid_to_ino_t(fattr->fileid);
130 }
131 
132 static int
133 nfs_write_inode(struct inode *inode, int sync)
134 {
135 	int flags = sync ? FLUSH_WAIT : 0;
136 	int ret;
137 
138 	ret = nfs_commit_inode(inode, flags);
139 	if (ret < 0)
140 		return ret;
141 	return 0;
142 }
143 
144 static void
145 nfs_delete_inode(struct inode * inode)
146 {
147 	dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
148 
149 	truncate_inode_pages(&inode->i_data, 0);
150 
151 	nfs_wb_all(inode);
152 	/*
153 	 * The following should never happen...
154 	 */
155 	if (nfs_have_writebacks(inode)) {
156 		printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
157 	}
158 
159 	clear_inode(inode);
160 }
161 
162 static void
163 nfs_clear_inode(struct inode *inode)
164 {
165 	struct nfs_inode *nfsi = NFS_I(inode);
166 	struct rpc_cred *cred;
167 
168 	nfs_wb_all(inode);
169 	BUG_ON (!list_empty(&nfsi->open_files));
170 	nfs_zap_acl_cache(inode);
171 	cred = nfsi->cache_access.cred;
172 	if (cred)
173 		put_rpccred(cred);
174 	BUG_ON(atomic_read(&nfsi->data_updates) != 0);
175 }
176 
177 void
178 nfs_umount_begin(struct super_block *sb)
179 {
180 	struct rpc_clnt	*rpc = NFS_SB(sb)->client;
181 
182 	/* -EIO all pending I/O */
183 	if (!IS_ERR(rpc))
184 		rpc_killall_tasks(rpc);
185 	rpc = NFS_SB(sb)->client_acl;
186 	if (!IS_ERR(rpc))
187 		rpc_killall_tasks(rpc);
188 }
189 
190 
191 static inline unsigned long
192 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
193 {
194 	/* make sure blocksize is a power of two */
195 	if ((bsize & (bsize - 1)) || nrbitsp) {
196 		unsigned char	nrbits;
197 
198 		for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
199 			;
200 		bsize = 1 << nrbits;
201 		if (nrbitsp)
202 			*nrbitsp = nrbits;
203 	}
204 
205 	return bsize;
206 }
207 
208 /*
209  * Calculate the number of 512byte blocks used.
210  */
211 static inline unsigned long
212 nfs_calc_block_size(u64 tsize)
213 {
214 	loff_t used = (tsize + 511) >> 9;
215 	return (used > ULONG_MAX) ? ULONG_MAX : used;
216 }
217 
218 /*
219  * Compute and set NFS server blocksize
220  */
221 static inline unsigned long
222 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
223 {
224 	if (bsize < 1024)
225 		bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
226 	else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
227 		bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
228 
229 	return nfs_block_bits(bsize, nrbitsp);
230 }
231 
232 /*
233  * Obtain the root inode of the file system.
234  */
235 static struct inode *
236 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
237 {
238 	struct nfs_server	*server = NFS_SB(sb);
239 	struct inode *rooti;
240 	int			error;
241 
242 	error = server->rpc_ops->getroot(server, rootfh, fsinfo);
243 	if (error < 0) {
244 		dprintk("nfs_get_root: getattr error = %d\n", -error);
245 		return ERR_PTR(error);
246 	}
247 
248 	rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
249 	if (!rooti)
250 		return ERR_PTR(-ENOMEM);
251 	return rooti;
252 }
253 
254 /*
255  * Do NFS version-independent mount processing, and sanity checking
256  */
257 static int
258 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
259 {
260 	struct nfs_server	*server;
261 	struct inode		*root_inode;
262 	struct nfs_fattr	fattr;
263 	struct nfs_fsinfo	fsinfo = {
264 					.fattr = &fattr,
265 				};
266 	struct nfs_pathconf pathinfo = {
267 			.fattr = &fattr,
268 	};
269 	int no_root_error = 0;
270 	unsigned long max_rpc_payload;
271 
272 	/* We probably want something more informative here */
273 	snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
274 
275 	server = NFS_SB(sb);
276 
277 	sb->s_magic      = NFS_SUPER_MAGIC;
278 
279 	root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
280 	/* Did getting the root inode fail? */
281 	if (IS_ERR(root_inode)) {
282 		no_root_error = PTR_ERR(root_inode);
283 		goto out_no_root;
284 	}
285 	sb->s_root = d_alloc_root(root_inode);
286 	if (!sb->s_root) {
287 		no_root_error = -ENOMEM;
288 		goto out_no_root;
289 	}
290 	sb->s_root->d_op = server->rpc_ops->dentry_ops;
291 
292 	/* Get some general file system info */
293 	if (server->namelen == 0 &&
294 	    server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
295 		server->namelen = pathinfo.max_namelen;
296 	/* Work out a lot of parameters */
297 	if (server->rsize == 0)
298 		server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
299 	if (server->wsize == 0)
300 		server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
301 
302 	if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
303 		server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
304 	if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
305 		server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
306 
307 	max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
308 	if (server->rsize > max_rpc_payload)
309 		server->rsize = max_rpc_payload;
310 	if (server->wsize > max_rpc_payload)
311 		server->wsize = max_rpc_payload;
312 
313 	server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
314 	if (server->rpages > NFS_READ_MAXIOV) {
315 		server->rpages = NFS_READ_MAXIOV;
316 		server->rsize = server->rpages << PAGE_CACHE_SHIFT;
317 	}
318 
319 	server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
320         if (server->wpages > NFS_WRITE_MAXIOV) {
321 		server->wpages = NFS_WRITE_MAXIOV;
322                 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
323 	}
324 
325 	if (sb->s_blocksize == 0)
326 		sb->s_blocksize = nfs_block_bits(server->wsize,
327 							 &sb->s_blocksize_bits);
328 	server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
329 
330 	server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
331 	if (server->dtsize > PAGE_CACHE_SIZE)
332 		server->dtsize = PAGE_CACHE_SIZE;
333 	if (server->dtsize > server->rsize)
334 		server->dtsize = server->rsize;
335 
336 	if (server->flags & NFS_MOUNT_NOAC) {
337 		server->acregmin = server->acregmax = 0;
338 		server->acdirmin = server->acdirmax = 0;
339 		sb->s_flags |= MS_SYNCHRONOUS;
340 	}
341 	server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
342 
343 	sb->s_maxbytes = fsinfo.maxfilesize;
344 	if (sb->s_maxbytes > MAX_LFS_FILESIZE)
345 		sb->s_maxbytes = MAX_LFS_FILESIZE;
346 
347 	server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
348 	server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
349 
350 	/* We're airborne Set socket buffersize */
351 	rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
352 	return 0;
353 	/* Yargs. It didn't work out. */
354 out_no_root:
355 	dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
356 	if (!IS_ERR(root_inode))
357 		iput(root_inode);
358 	return no_root_error;
359 }
360 
361 static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
362 {
363 	to->to_initval = timeo * HZ / 10;
364 	to->to_retries = retrans;
365 	if (!to->to_retries)
366 		to->to_retries = 2;
367 
368 	switch (proto) {
369 	case IPPROTO_TCP:
370 		if (!to->to_initval)
371 			to->to_initval = 60 * HZ;
372 		if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
373 			to->to_initval = NFS_MAX_TCP_TIMEOUT;
374 		to->to_increment = to->to_initval;
375 		to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
376 		to->to_exponential = 0;
377 		break;
378 	case IPPROTO_UDP:
379 	default:
380 		if (!to->to_initval)
381 			to->to_initval = 11 * HZ / 10;
382 		if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
383 			to->to_initval = NFS_MAX_UDP_TIMEOUT;
384 		to->to_maxval = NFS_MAX_UDP_TIMEOUT;
385 		to->to_exponential = 1;
386 		break;
387 	}
388 }
389 
390 /*
391  * Create an RPC client handle.
392  */
393 static struct rpc_clnt *
394 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
395 {
396 	struct rpc_timeout	timeparms;
397 	struct rpc_xprt		*xprt = NULL;
398 	struct rpc_clnt		*clnt = NULL;
399 	int			proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
400 
401 	nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
402 
403 	/* create transport and client */
404 	xprt = xprt_create_proto(proto, &server->addr, &timeparms);
405 	if (IS_ERR(xprt)) {
406 		dprintk("%s: cannot create RPC transport. Error = %ld\n",
407 				__FUNCTION__, PTR_ERR(xprt));
408 		return (struct rpc_clnt *)xprt;
409 	}
410 	clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
411 				 server->rpc_ops->version, data->pseudoflavor);
412 	if (IS_ERR(clnt)) {
413 		dprintk("%s: cannot create RPC client. Error = %ld\n",
414 				__FUNCTION__, PTR_ERR(xprt));
415 		goto out_fail;
416 	}
417 
418 	clnt->cl_intr     = 1;
419 	clnt->cl_softrtry = 1;
420 	clnt->cl_chatty   = 1;
421 
422 	return clnt;
423 
424 out_fail:
425 	return clnt;
426 }
427 
428 /*
429  * The way this works is that the mount process passes a structure
430  * in the data argument which contains the server's IP address
431  * and the root file handle obtained from the server's mount
432  * daemon. We stash these away in the private superblock fields.
433  */
434 static int
435 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
436 {
437 	struct nfs_server	*server;
438 	rpc_authflavor_t	authflavor;
439 
440 	server           = NFS_SB(sb);
441 	sb->s_blocksize_bits = 0;
442 	sb->s_blocksize = 0;
443 	if (data->bsize)
444 		sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
445 	if (data->rsize)
446 		server->rsize = nfs_block_size(data->rsize, NULL);
447 	if (data->wsize)
448 		server->wsize = nfs_block_size(data->wsize, NULL);
449 	server->flags    = data->flags & NFS_MOUNT_FLAGMASK;
450 
451 	server->acregmin = data->acregmin*HZ;
452 	server->acregmax = data->acregmax*HZ;
453 	server->acdirmin = data->acdirmin*HZ;
454 	server->acdirmax = data->acdirmax*HZ;
455 
456 	/* Start lockd here, before we might error out */
457 	if (!(server->flags & NFS_MOUNT_NONLM))
458 		lockd_up();
459 
460 	server->namelen  = data->namlen;
461 	server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
462 	if (!server->hostname)
463 		return -ENOMEM;
464 	strcpy(server->hostname, data->hostname);
465 
466 	/* Check NFS protocol revision and initialize RPC op vector
467 	 * and file handle pool. */
468 #ifdef CONFIG_NFS_V3
469 	if (server->flags & NFS_MOUNT_VER3) {
470 		server->rpc_ops = &nfs_v3_clientops;
471 		server->caps |= NFS_CAP_READDIRPLUS;
472 	} else {
473 		server->rpc_ops = &nfs_v2_clientops;
474 	}
475 #else
476 	server->rpc_ops = &nfs_v2_clientops;
477 #endif
478 
479 	/* Fill in pseudoflavor for mount version < 5 */
480 	if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
481 		data->pseudoflavor = RPC_AUTH_UNIX;
482 	authflavor = data->pseudoflavor;	/* save for sb_init() */
483 	/* XXX maybe we want to add a server->pseudoflavor field */
484 
485 	/* Create RPC client handles */
486 	server->client = nfs_create_client(server, data);
487 	if (IS_ERR(server->client))
488 		return PTR_ERR(server->client);
489 	/* RFC 2623, sec 2.3.2 */
490 	if (authflavor != RPC_AUTH_UNIX) {
491 		struct rpc_auth *auth;
492 
493 		server->client_sys = rpc_clone_client(server->client);
494 		if (IS_ERR(server->client_sys))
495 			return PTR_ERR(server->client_sys);
496 		auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
497 		if (IS_ERR(auth))
498 			return PTR_ERR(auth);
499 	} else {
500 		atomic_inc(&server->client->cl_count);
501 		server->client_sys = server->client;
502 	}
503 	if (server->flags & NFS_MOUNT_VER3) {
504 #ifdef CONFIG_NFS_V3_ACL
505 		if (!(server->flags & NFS_MOUNT_NOACL)) {
506 			server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
507 			/* No errors! Assume that Sun nfsacls are supported */
508 			if (!IS_ERR(server->client_acl))
509 				server->caps |= NFS_CAP_ACLS;
510 		}
511 #else
512 		server->flags &= ~NFS_MOUNT_NOACL;
513 #endif /* CONFIG_NFS_V3_ACL */
514 		/*
515 		 * The VFS shouldn't apply the umask to mode bits. We will
516 		 * do so ourselves when necessary.
517 		 */
518 		sb->s_flags |= MS_POSIXACL;
519 		if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
520 			server->namelen = NFS3_MAXNAMLEN;
521 		sb->s_time_gran = 1;
522 	} else {
523 		if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
524 			server->namelen = NFS2_MAXNAMLEN;
525 	}
526 
527 	sb->s_op = &nfs_sops;
528 	return nfs_sb_init(sb, authflavor);
529 }
530 
531 static int
532 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
533 {
534 	struct nfs_server *server = NFS_SB(sb);
535 	unsigned char blockbits;
536 	unsigned long blockres;
537 	struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
538 	struct nfs_fattr fattr;
539 	struct nfs_fsstat res = {
540 			.fattr = &fattr,
541 	};
542 	int error;
543 
544 	lock_kernel();
545 
546 	error = server->rpc_ops->statfs(server, rootfh, &res);
547 	buf->f_type = NFS_SUPER_MAGIC;
548 	if (error < 0)
549 		goto out_err;
550 
551 	/*
552 	 * Current versions of glibc do not correctly handle the
553 	 * case where f_frsize != f_bsize.  Eventually we want to
554 	 * report the value of wtmult in this field.
555 	 */
556 	buf->f_frsize = sb->s_blocksize;
557 
558 	/*
559 	 * On most *nix systems, f_blocks, f_bfree, and f_bavail
560 	 * are reported in units of f_frsize.  Linux hasn't had
561 	 * an f_frsize field in its statfs struct until recently,
562 	 * thus historically Linux's sys_statfs reports these
563 	 * fields in units of f_bsize.
564 	 */
565 	buf->f_bsize = sb->s_blocksize;
566 	blockbits = sb->s_blocksize_bits;
567 	blockres = (1 << blockbits) - 1;
568 	buf->f_blocks = (res.tbytes + blockres) >> blockbits;
569 	buf->f_bfree = (res.fbytes + blockres) >> blockbits;
570 	buf->f_bavail = (res.abytes + blockres) >> blockbits;
571 
572 	buf->f_files = res.tfiles;
573 	buf->f_ffree = res.afiles;
574 
575 	buf->f_namelen = server->namelen;
576  out:
577 	unlock_kernel();
578 
579 	return 0;
580 
581  out_err:
582 	printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
583 	buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
584 	goto out;
585 
586 }
587 
588 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
589 {
590 	static struct proc_nfs_info {
591 		int flag;
592 		char *str;
593 		char *nostr;
594 	} nfs_info[] = {
595 		{ NFS_MOUNT_SOFT, ",soft", ",hard" },
596 		{ NFS_MOUNT_INTR, ",intr", "" },
597 		{ NFS_MOUNT_POSIX, ",posix", "" },
598 		{ NFS_MOUNT_NOCTO, ",nocto", "" },
599 		{ NFS_MOUNT_NOAC, ",noac", "" },
600 		{ NFS_MOUNT_NONLM, ",nolock", ",lock" },
601 		{ NFS_MOUNT_NOACL, ",noacl", "" },
602 		{ 0, NULL, NULL }
603 	};
604 	struct proc_nfs_info *nfs_infop;
605 	struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
606 	char buf[12];
607 	char *proto;
608 
609 	seq_printf(m, ",v%d", nfss->rpc_ops->version);
610 	seq_printf(m, ",rsize=%d", nfss->rsize);
611 	seq_printf(m, ",wsize=%d", nfss->wsize);
612 	if (nfss->acregmin != 3*HZ)
613 		seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
614 	if (nfss->acregmax != 60*HZ)
615 		seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
616 	if (nfss->acdirmin != 30*HZ)
617 		seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
618 	if (nfss->acdirmax != 60*HZ)
619 		seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
620 	for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
621 		if (nfss->flags & nfs_infop->flag)
622 			seq_puts(m, nfs_infop->str);
623 		else
624 			seq_puts(m, nfs_infop->nostr);
625 	}
626 	switch (nfss->client->cl_xprt->prot) {
627 		case IPPROTO_TCP:
628 			proto = "tcp";
629 			break;
630 		case IPPROTO_UDP:
631 			proto = "udp";
632 			break;
633 		default:
634 			snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
635 			proto = buf;
636 	}
637 	seq_printf(m, ",proto=%s", proto);
638 	seq_puts(m, ",addr=");
639 	seq_escape(m, nfss->hostname, " \t\n\\");
640 	return 0;
641 }
642 
643 /*
644  * Invalidate the local caches
645  */
646 void
647 nfs_zap_caches(struct inode *inode)
648 {
649 	struct nfs_inode *nfsi = NFS_I(inode);
650 	int mode = inode->i_mode;
651 
652 	spin_lock(&inode->i_lock);
653 
654 	NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
655 	NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
656 
657 	memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
658 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
659 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
660 	else
661 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
662 
663 	spin_unlock(&inode->i_lock);
664 }
665 
666 static void nfs_zap_acl_cache(struct inode *inode)
667 {
668 	void (*clear_acl_cache)(struct inode *);
669 
670 	clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
671 	if (clear_acl_cache != NULL)
672 		clear_acl_cache(inode);
673 	spin_lock(&inode->i_lock);
674 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
675 	spin_unlock(&inode->i_lock);
676 }
677 
678 /*
679  * Invalidate, but do not unhash, the inode
680  */
681 static void
682 nfs_invalidate_inode(struct inode *inode)
683 {
684 	umode_t save_mode = inode->i_mode;
685 
686 	make_bad_inode(inode);
687 	inode->i_mode = save_mode;
688 	nfs_zap_caches(inode);
689 }
690 
691 struct nfs_find_desc {
692 	struct nfs_fh		*fh;
693 	struct nfs_fattr	*fattr;
694 };
695 
696 /*
697  * In NFSv3 we can have 64bit inode numbers. In order to support
698  * this, and re-exported directories (also seen in NFSv2)
699  * we are forced to allow 2 different inodes to have the same
700  * i_ino.
701  */
702 static int
703 nfs_find_actor(struct inode *inode, void *opaque)
704 {
705 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
706 	struct nfs_fh		*fh = desc->fh;
707 	struct nfs_fattr	*fattr = desc->fattr;
708 
709 	if (NFS_FILEID(inode) != fattr->fileid)
710 		return 0;
711 	if (nfs_compare_fh(NFS_FH(inode), fh))
712 		return 0;
713 	if (is_bad_inode(inode) || NFS_STALE(inode))
714 		return 0;
715 	return 1;
716 }
717 
718 static int
719 nfs_init_locked(struct inode *inode, void *opaque)
720 {
721 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
722 	struct nfs_fattr	*fattr = desc->fattr;
723 
724 	NFS_FILEID(inode) = fattr->fileid;
725 	nfs_copy_fh(NFS_FH(inode), desc->fh);
726 	return 0;
727 }
728 
729 /* Don't use READDIRPLUS on directories that we believe are too large */
730 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
731 
732 /*
733  * This is our front-end to iget that looks up inodes by file handle
734  * instead of inode number.
735  */
736 struct inode *
737 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
738 {
739 	struct nfs_find_desc desc = {
740 		.fh	= fh,
741 		.fattr	= fattr
742 	};
743 	struct inode *inode = NULL;
744 	unsigned long hash;
745 
746 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
747 		goto out_no_inode;
748 
749 	if (!fattr->nlink) {
750 		printk("NFS: Buggy server - nlink == 0!\n");
751 		goto out_no_inode;
752 	}
753 
754 	hash = nfs_fattr_to_ino_t(fattr);
755 
756 	if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
757 		goto out_no_inode;
758 
759 	if (inode->i_state & I_NEW) {
760 		struct nfs_inode *nfsi = NFS_I(inode);
761 
762 		/* We set i_ino for the few things that still rely on it,
763 		 * such as stat(2) */
764 		inode->i_ino = hash;
765 
766 		/* We can't support update_atime(), since the server will reset it */
767 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
768 		inode->i_mode = fattr->mode;
769 		/* Why so? Because we want revalidate for devices/FIFOs, and
770 		 * that's precisely what we have in nfs_file_inode_operations.
771 		 */
772 		inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
773 		if (S_ISREG(inode->i_mode)) {
774 			inode->i_fop = &nfs_file_operations;
775 			inode->i_data.a_ops = &nfs_file_aops;
776 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
777 		} else if (S_ISDIR(inode->i_mode)) {
778 			inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
779 			inode->i_fop = &nfs_dir_operations;
780 			if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
781 			    && fattr->size <= NFS_LIMIT_READDIRPLUS)
782 				set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
783 		} else if (S_ISLNK(inode->i_mode))
784 			inode->i_op = &nfs_symlink_inode_operations;
785 		else
786 			init_special_inode(inode, inode->i_mode, fattr->rdev);
787 
788 		nfsi->read_cache_jiffies = fattr->time_start;
789 		nfsi->last_updated = jiffies;
790 		inode->i_atime = fattr->atime;
791 		inode->i_mtime = fattr->mtime;
792 		inode->i_ctime = fattr->ctime;
793 		if (fattr->valid & NFS_ATTR_FATTR_V4)
794 			nfsi->change_attr = fattr->change_attr;
795 		inode->i_size = nfs_size_to_loff_t(fattr->size);
796 		inode->i_nlink = fattr->nlink;
797 		inode->i_uid = fattr->uid;
798 		inode->i_gid = fattr->gid;
799 		if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
800 			/*
801 			 * report the blocks in 512byte units
802 			 */
803 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
804 			inode->i_blksize = inode->i_sb->s_blocksize;
805 		} else {
806 			inode->i_blocks = fattr->du.nfs2.blocks;
807 			inode->i_blksize = fattr->du.nfs2.blocksize;
808 		}
809 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
810 		nfsi->attrtimeo_timestamp = jiffies;
811 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
812 		nfsi->cache_access.cred = NULL;
813 
814 		unlock_new_inode(inode);
815 	} else
816 		nfs_refresh_inode(inode, fattr);
817 	dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
818 		inode->i_sb->s_id,
819 		(long long)NFS_FILEID(inode),
820 		atomic_read(&inode->i_count));
821 
822 out:
823 	return inode;
824 
825 out_no_inode:
826 	printk("nfs_fhget: iget failed\n");
827 	goto out;
828 }
829 
830 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
831 
832 int
833 nfs_setattr(struct dentry *dentry, struct iattr *attr)
834 {
835 	struct inode *inode = dentry->d_inode;
836 	struct nfs_fattr fattr;
837 	int error;
838 
839 	if (attr->ia_valid & ATTR_SIZE) {
840 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
841 			attr->ia_valid &= ~ATTR_SIZE;
842 	}
843 
844 	/* Optimization: if the end result is no change, don't RPC */
845 	attr->ia_valid &= NFS_VALID_ATTRS;
846 	if (attr->ia_valid == 0)
847 		return 0;
848 
849 	lock_kernel();
850 	nfs_begin_data_update(inode);
851 	/* Write all dirty data if we're changing file permissions or size */
852 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
853 		if (filemap_fdatawrite(inode->i_mapping) == 0)
854 			filemap_fdatawait(inode->i_mapping);
855 		nfs_wb_all(inode);
856 	}
857 	/*
858 	 * Return any delegations if we're going to change ACLs
859 	 */
860 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
861 		nfs_inode_return_delegation(inode);
862 	error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
863 	if (error == 0)
864 		nfs_refresh_inode(inode, &fattr);
865 	nfs_end_data_update(inode);
866 	unlock_kernel();
867 	return error;
868 }
869 
870 /**
871  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
872  * @inode: pointer to struct inode
873  * @attr: pointer to struct iattr
874  *
875  * Note: we do this in the *proc.c in order to ensure that
876  *       it works for things like exclusive creates too.
877  */
878 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
879 {
880 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
881 		if ((attr->ia_valid & ATTR_MODE) != 0) {
882 			int mode = attr->ia_mode & S_IALLUGO;
883 			mode |= inode->i_mode & ~S_IALLUGO;
884 			inode->i_mode = mode;
885 		}
886 		if ((attr->ia_valid & ATTR_UID) != 0)
887 			inode->i_uid = attr->ia_uid;
888 		if ((attr->ia_valid & ATTR_GID) != 0)
889 			inode->i_gid = attr->ia_gid;
890 		spin_lock(&inode->i_lock);
891 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
892 		spin_unlock(&inode->i_lock);
893 	}
894 	if ((attr->ia_valid & ATTR_SIZE) != 0) {
895 		inode->i_size = attr->ia_size;
896 		vmtruncate(inode, attr->ia_size);
897 	}
898 }
899 
900 static int nfs_wait_schedule(void *word)
901 {
902 	if (signal_pending(current))
903 		return -ERESTARTSYS;
904 	schedule();
905 	return 0;
906 }
907 
908 /*
909  * Wait for the inode to get unlocked.
910  */
911 static int nfs_wait_on_inode(struct inode *inode)
912 {
913 	struct rpc_clnt	*clnt = NFS_CLIENT(inode);
914 	struct nfs_inode *nfsi = NFS_I(inode);
915 	sigset_t oldmask;
916 	int error;
917 
918 	rpc_clnt_sigmask(clnt, &oldmask);
919 	error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
920 					nfs_wait_schedule, TASK_INTERRUPTIBLE);
921 	rpc_clnt_sigunmask(clnt, &oldmask);
922 
923 	return error;
924 }
925 
926 static void nfs_wake_up_inode(struct inode *inode)
927 {
928 	struct nfs_inode *nfsi = NFS_I(inode);
929 
930 	clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
931 	smp_mb__after_clear_bit();
932 	wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
933 }
934 
935 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
936 {
937 	struct inode *inode = dentry->d_inode;
938 	int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
939 	int err;
940 
941 	if (__IS_FLG(inode, MS_NOATIME))
942 		need_atime = 0;
943 	else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
944 		need_atime = 0;
945 	/* We may force a getattr if the user cares about atime */
946 	if (need_atime)
947 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
948 	else
949 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
950 	if (!err)
951 		generic_fillattr(inode, stat);
952 	return err;
953 }
954 
955 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
956 {
957 	struct nfs_open_context *ctx;
958 
959 	ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
960 	if (ctx != NULL) {
961 		atomic_set(&ctx->count, 1);
962 		ctx->dentry = dget(dentry);
963 		ctx->cred = get_rpccred(cred);
964 		ctx->state = NULL;
965 		ctx->lockowner = current->files;
966 		ctx->error = 0;
967 		ctx->dir_cookie = 0;
968 	}
969 	return ctx;
970 }
971 
972 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
973 {
974 	if (ctx != NULL)
975 		atomic_inc(&ctx->count);
976 	return ctx;
977 }
978 
979 void put_nfs_open_context(struct nfs_open_context *ctx)
980 {
981 	if (atomic_dec_and_test(&ctx->count)) {
982 		if (!list_empty(&ctx->list)) {
983 			struct inode *inode = ctx->dentry->d_inode;
984 			spin_lock(&inode->i_lock);
985 			list_del(&ctx->list);
986 			spin_unlock(&inode->i_lock);
987 		}
988 		if (ctx->state != NULL)
989 			nfs4_close_state(ctx->state, ctx->mode);
990 		if (ctx->cred != NULL)
991 			put_rpccred(ctx->cred);
992 		dput(ctx->dentry);
993 		kfree(ctx);
994 	}
995 }
996 
997 /*
998  * Ensure that mmap has a recent RPC credential for use when writing out
999  * shared pages
1000  */
1001 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1002 {
1003 	struct inode *inode = filp->f_dentry->d_inode;
1004 	struct nfs_inode *nfsi = NFS_I(inode);
1005 
1006 	filp->private_data = get_nfs_open_context(ctx);
1007 	spin_lock(&inode->i_lock);
1008 	list_add(&ctx->list, &nfsi->open_files);
1009 	spin_unlock(&inode->i_lock);
1010 }
1011 
1012 /*
1013  * Given an inode, search for an open context with the desired characteristics
1014  */
1015 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
1016 {
1017 	struct nfs_inode *nfsi = NFS_I(inode);
1018 	struct nfs_open_context *pos, *ctx = NULL;
1019 
1020 	spin_lock(&inode->i_lock);
1021 	list_for_each_entry(pos, &nfsi->open_files, list) {
1022 		if (cred != NULL && pos->cred != cred)
1023 			continue;
1024 		if ((pos->mode & mode) == mode) {
1025 			ctx = get_nfs_open_context(pos);
1026 			break;
1027 		}
1028 	}
1029 	spin_unlock(&inode->i_lock);
1030 	return ctx;
1031 }
1032 
1033 void nfs_file_clear_open_context(struct file *filp)
1034 {
1035 	struct inode *inode = filp->f_dentry->d_inode;
1036 	struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
1037 
1038 	if (ctx) {
1039 		filp->private_data = NULL;
1040 		spin_lock(&inode->i_lock);
1041 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1042 		spin_unlock(&inode->i_lock);
1043 		put_nfs_open_context(ctx);
1044 	}
1045 }
1046 
1047 /*
1048  * These allocate and release file read/write context information.
1049  */
1050 int nfs_open(struct inode *inode, struct file *filp)
1051 {
1052 	struct nfs_open_context *ctx;
1053 	struct rpc_cred *cred;
1054 
1055 	cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1056 	if (IS_ERR(cred))
1057 		return PTR_ERR(cred);
1058 	ctx = alloc_nfs_open_context(filp->f_dentry, cred);
1059 	put_rpccred(cred);
1060 	if (ctx == NULL)
1061 		return -ENOMEM;
1062 	ctx->mode = filp->f_mode;
1063 	nfs_file_set_open_context(filp, ctx);
1064 	put_nfs_open_context(ctx);
1065 	return 0;
1066 }
1067 
1068 int nfs_release(struct inode *inode, struct file *filp)
1069 {
1070 	nfs_file_clear_open_context(filp);
1071 	return 0;
1072 }
1073 
1074 /*
1075  * This function is called whenever some part of NFS notices that
1076  * the cached attributes have to be refreshed.
1077  */
1078 int
1079 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1080 {
1081 	int		 status = -ESTALE;
1082 	struct nfs_fattr fattr;
1083 	struct nfs_inode *nfsi = NFS_I(inode);
1084 	unsigned long verifier;
1085 	unsigned long cache_validity;
1086 
1087 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1088 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1089 
1090 	lock_kernel();
1091 	if (!inode || is_bad_inode(inode))
1092  		goto out_nowait;
1093 	if (NFS_STALE(inode))
1094  		goto out_nowait;
1095 
1096 	status = nfs_wait_on_inode(inode);
1097 	if (status < 0)
1098 		goto out;
1099 	if (NFS_STALE(inode)) {
1100 		status = -ESTALE;
1101 		/* Do we trust the cached ESTALE? */
1102 		if (NFS_ATTRTIMEO(inode) != 0) {
1103 			if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1104 				/* no */
1105 			} else
1106 				goto out;
1107 		}
1108 	}
1109 
1110 	/* Protect against RPC races by saving the change attribute */
1111 	verifier = nfs_save_change_attribute(inode);
1112 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1113 	if (status != 0) {
1114 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1115 			 inode->i_sb->s_id,
1116 			 (long long)NFS_FILEID(inode), status);
1117 		if (status == -ESTALE) {
1118 			nfs_zap_caches(inode);
1119 			if (!S_ISDIR(inode->i_mode))
1120 				set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1121 		}
1122 		goto out;
1123 	}
1124 
1125 	spin_lock(&inode->i_lock);
1126 	status = nfs_update_inode(inode, &fattr, verifier);
1127 	if (status) {
1128 		spin_unlock(&inode->i_lock);
1129 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1130 			 inode->i_sb->s_id,
1131 			 (long long)NFS_FILEID(inode), status);
1132 		goto out;
1133 	}
1134 	cache_validity = nfsi->cache_validity;
1135 	nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1136 
1137 	/*
1138 	 * We may need to keep the attributes marked as invalid if
1139 	 * we raced with nfs_end_attr_update().
1140 	 */
1141 	if (time_after_eq(verifier, nfsi->cache_change_attribute))
1142 		nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1143 	spin_unlock(&inode->i_lock);
1144 
1145 	nfs_revalidate_mapping(inode, inode->i_mapping);
1146 
1147 	if (cache_validity & NFS_INO_INVALID_ACL)
1148 		nfs_zap_acl_cache(inode);
1149 
1150 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1151 		inode->i_sb->s_id,
1152 		(long long)NFS_FILEID(inode));
1153 
1154  out:
1155 	nfs_wake_up_inode(inode);
1156 
1157  out_nowait:
1158 	unlock_kernel();
1159 	return status;
1160 }
1161 
1162 int nfs_attribute_timeout(struct inode *inode)
1163 {
1164 	struct nfs_inode *nfsi = NFS_I(inode);
1165 
1166 	if (nfs_have_delegation(inode, FMODE_READ))
1167 		return 0;
1168 	return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1169 }
1170 
1171 /**
1172  * nfs_revalidate_inode - Revalidate the inode attributes
1173  * @server - pointer to nfs_server struct
1174  * @inode - pointer to inode struct
1175  *
1176  * Updates inode attribute information by retrieving the data from the server.
1177  */
1178 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1179 {
1180 	if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1181 			&& !nfs_attribute_timeout(inode))
1182 		return NFS_STALE(inode) ? -ESTALE : 0;
1183 	return __nfs_revalidate_inode(server, inode);
1184 }
1185 
1186 /**
1187  * nfs_revalidate_mapping - Revalidate the pagecache
1188  * @inode - pointer to host inode
1189  * @mapping - pointer to mapping
1190  */
1191 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1192 {
1193 	struct nfs_inode *nfsi = NFS_I(inode);
1194 
1195 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1196 		if (S_ISREG(inode->i_mode)) {
1197 			if (filemap_fdatawrite(mapping) == 0)
1198 				filemap_fdatawait(mapping);
1199 			nfs_wb_all(inode);
1200 		}
1201 		invalidate_inode_pages2(mapping);
1202 
1203 		spin_lock(&inode->i_lock);
1204 		nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1205 		if (S_ISDIR(inode->i_mode)) {
1206 			memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1207 			/* This ensures we revalidate child dentries */
1208 			nfsi->cache_change_attribute = jiffies;
1209 		}
1210 		spin_unlock(&inode->i_lock);
1211 
1212 		dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1213 				inode->i_sb->s_id,
1214 				(long long)NFS_FILEID(inode));
1215 	}
1216 }
1217 
1218 /**
1219  * nfs_begin_data_update
1220  * @inode - pointer to inode
1221  * Declare that a set of operations will update file data on the server
1222  */
1223 void nfs_begin_data_update(struct inode *inode)
1224 {
1225 	atomic_inc(&NFS_I(inode)->data_updates);
1226 }
1227 
1228 /**
1229  * nfs_end_data_update
1230  * @inode - pointer to inode
1231  * Declare end of the operations that will update file data
1232  * This will mark the inode as immediately needing revalidation
1233  * of its attribute cache.
1234  */
1235 void nfs_end_data_update(struct inode *inode)
1236 {
1237 	struct nfs_inode *nfsi = NFS_I(inode);
1238 
1239 	if (!nfs_have_delegation(inode, FMODE_READ)) {
1240 		/* Directories and symlinks: invalidate page cache */
1241 		if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
1242 			spin_lock(&inode->i_lock);
1243 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1244 			spin_unlock(&inode->i_lock);
1245 		}
1246 	}
1247 	nfsi->cache_change_attribute = jiffies;
1248 	atomic_dec(&nfsi->data_updates);
1249 }
1250 
1251 /**
1252  * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1253  * @inode - pointer to inode
1254  * @fattr - updated attributes
1255  *
1256  * Verifies the attribute cache. If we have just changed the attributes,
1257  * so that fattr carries weak cache consistency data, then it may
1258  * also update the ctime/mtime/change_attribute.
1259  */
1260 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1261 {
1262 	struct nfs_inode *nfsi = NFS_I(inode);
1263 	loff_t cur_size, new_isize;
1264 	int data_unstable;
1265 
1266 
1267 	/* Are we in the process of updating data on the server? */
1268 	data_unstable = nfs_caches_unstable(inode);
1269 
1270 	if (fattr->valid & NFS_ATTR_FATTR_V4) {
1271 		if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1272 				&& nfsi->change_attr == fattr->pre_change_attr)
1273 			nfsi->change_attr = fattr->change_attr;
1274 		if (nfsi->change_attr != fattr->change_attr) {
1275 			nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1276 			if (!data_unstable)
1277 				nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1278 		}
1279 	}
1280 
1281 	if ((fattr->valid & NFS_ATTR_FATTR) == 0) {
1282 		return 0;
1283 	}
1284 
1285 	/* Has the inode gone and changed behind our back? */
1286 	if (nfsi->fileid != fattr->fileid
1287 			|| (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1288 		return -EIO;
1289 	}
1290 
1291 	cur_size = i_size_read(inode);
1292  	new_isize = nfs_size_to_loff_t(fattr->size);
1293 
1294 	/* If we have atomic WCC data, we may update some attributes */
1295 	if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1296 		if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1297 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1298 		if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1299 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1300 	}
1301 
1302 	/* Verify a few of the more important attributes */
1303 	if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1304 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1305 		if (!data_unstable)
1306 			nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1307 	}
1308 	if (cur_size != new_isize) {
1309 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1310 		if (nfsi->npages == 0)
1311 			nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1312 	}
1313 
1314 	/* Have any file permissions changed? */
1315 	if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1316 			|| inode->i_uid != fattr->uid
1317 			|| inode->i_gid != fattr->gid)
1318 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1319 
1320 	/* Has the link count changed? */
1321 	if (inode->i_nlink != fattr->nlink)
1322 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1323 
1324 	if (!timespec_equal(&inode->i_atime, &fattr->atime))
1325 		nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1326 
1327 	nfsi->read_cache_jiffies = fattr->time_start;
1328 	return 0;
1329 }
1330 
1331 /**
1332  * nfs_refresh_inode - try to update the inode attribute cache
1333  * @inode - pointer to inode
1334  * @fattr - updated attributes
1335  *
1336  * Check that an RPC call that returned attributes has not overlapped with
1337  * other recent updates of the inode metadata, then decide whether it is
1338  * safe to do a full update of the inode attributes, or whether just to
1339  * call nfs_check_inode_attributes.
1340  */
1341 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1342 {
1343 	struct nfs_inode *nfsi = NFS_I(inode);
1344 	int status;
1345 
1346 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1347 		return 0;
1348 	spin_lock(&inode->i_lock);
1349 	nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1350 	if (nfs_verify_change_attribute(inode, fattr->time_start))
1351 		nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1352 	if (time_after(fattr->time_start, nfsi->last_updated))
1353 		status = nfs_update_inode(inode, fattr, fattr->time_start);
1354 	else
1355 		status = nfs_check_inode_attributes(inode, fattr);
1356 
1357 	spin_unlock(&inode->i_lock);
1358 	return status;
1359 }
1360 
1361 /**
1362  * nfs_post_op_update_inode - try to update the inode attribute cache
1363  * @inode - pointer to inode
1364  * @fattr - updated attributes
1365  *
1366  * After an operation that has changed the inode metadata, mark the
1367  * attribute cache as being invalid, then try to update it.
1368  */
1369 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1370 {
1371 	struct nfs_inode *nfsi = NFS_I(inode);
1372 	int status = 0;
1373 
1374 	spin_lock(&inode->i_lock);
1375 	if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
1376 		nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1377 		goto out;
1378 	}
1379 	status = nfs_update_inode(inode, fattr, fattr->time_start);
1380 	if (time_after_eq(fattr->time_start, nfsi->cache_change_attribute))
1381 		nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1382 	nfsi->cache_change_attribute = jiffies;
1383 out:
1384 	spin_unlock(&inode->i_lock);
1385 	return status;
1386 }
1387 
1388 /*
1389  * Many nfs protocol calls return the new file attributes after
1390  * an operation.  Here we update the inode to reflect the state
1391  * of the server's inode.
1392  *
1393  * This is a bit tricky because we have to make sure all dirty pages
1394  * have been sent off to the server before calling invalidate_inode_pages.
1395  * To make sure no other process adds more write requests while we try
1396  * our best to flush them, we make them sleep during the attribute refresh.
1397  *
1398  * A very similar scenario holds for the dir cache.
1399  */
1400 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1401 {
1402 	struct nfs_inode *nfsi = NFS_I(inode);
1403 	loff_t cur_isize, new_isize;
1404 	unsigned int	invalid = 0;
1405 	int data_unstable;
1406 
1407 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1408 			__FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1409 			atomic_read(&inode->i_count), fattr->valid);
1410 
1411 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1412 		return 0;
1413 
1414 	if (nfsi->fileid != fattr->fileid) {
1415 		printk(KERN_ERR "%s: inode number mismatch\n"
1416 		       "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1417 		       __FUNCTION__,
1418 		       inode->i_sb->s_id, (long long)nfsi->fileid,
1419 		       inode->i_sb->s_id, (long long)fattr->fileid);
1420 		goto out_err;
1421 	}
1422 
1423 	/*
1424 	 * Make sure the inode's type hasn't changed.
1425 	 */
1426 	if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1427 		goto out_changed;
1428 
1429 	/*
1430 	 * Update the read time so we don't revalidate too often.
1431 	 */
1432 	nfsi->read_cache_jiffies = fattr->time_start;
1433 	nfsi->last_updated = jiffies;
1434 
1435 	/* Are we racing with known updates of the metadata on the server? */
1436 	data_unstable = ! (nfs_verify_change_attribute(inode, verifier) ||
1437 		(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE));
1438 
1439 	/* Check if our cached file size is stale */
1440  	new_isize = nfs_size_to_loff_t(fattr->size);
1441 	cur_isize = i_size_read(inode);
1442 	if (new_isize != cur_isize) {
1443 		/* Do we perhaps have any outstanding writes? */
1444 		if (nfsi->npages == 0) {
1445 			/* No, but did we race with nfs_end_data_update()? */
1446 			if (time_after_eq(verifier,  nfsi->cache_change_attribute)) {
1447 				inode->i_size = new_isize;
1448 				invalid |= NFS_INO_INVALID_DATA;
1449 			}
1450 			invalid |= NFS_INO_INVALID_ATTR;
1451 		} else if (new_isize > cur_isize) {
1452 			inode->i_size = new_isize;
1453 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1454 		}
1455 		dprintk("NFS: isize change on server for file %s/%ld\n",
1456 				inode->i_sb->s_id, inode->i_ino);
1457 	}
1458 
1459 	/* Check if the mtime agrees */
1460 	if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1461 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1462 		dprintk("NFS: mtime change on server for file %s/%ld\n",
1463 				inode->i_sb->s_id, inode->i_ino);
1464 		if (!data_unstable)
1465 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1466 	}
1467 
1468 	if ((fattr->valid & NFS_ATTR_FATTR_V4)
1469 	    && nfsi->change_attr != fattr->change_attr) {
1470 		dprintk("NFS: change_attr change on server for file %s/%ld\n",
1471 		       inode->i_sb->s_id, inode->i_ino);
1472 		nfsi->change_attr = fattr->change_attr;
1473 		if (!data_unstable)
1474 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1475 	}
1476 
1477 	/* If ctime has changed we should definitely clear access+acl caches */
1478 	if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1479 		if (!data_unstable)
1480 			invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1481 		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1482 	}
1483 	memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1484 
1485 	if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1486 	    inode->i_uid != fattr->uid ||
1487 	    inode->i_gid != fattr->gid)
1488 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1489 
1490 	inode->i_mode = fattr->mode;
1491 	inode->i_nlink = fattr->nlink;
1492 	inode->i_uid = fattr->uid;
1493 	inode->i_gid = fattr->gid;
1494 
1495 	if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1496 		/*
1497 		 * report the blocks in 512byte units
1498 		 */
1499 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1500 		inode->i_blksize = inode->i_sb->s_blocksize;
1501  	} else {
1502  		inode->i_blocks = fattr->du.nfs2.blocks;
1503  		inode->i_blksize = fattr->du.nfs2.blocksize;
1504  	}
1505 
1506 	/* Update attrtimeo value if we're out of the unstable period */
1507 	if (invalid & NFS_INO_INVALID_ATTR) {
1508 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1509 		nfsi->attrtimeo_timestamp = jiffies;
1510 	} else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1511 		if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1512 			nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1513 		nfsi->attrtimeo_timestamp = jiffies;
1514 	}
1515 	/* Don't invalidate the data if we were to blame */
1516 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1517 				|| S_ISLNK(inode->i_mode)))
1518 		invalid &= ~NFS_INO_INVALID_DATA;
1519 	if (!nfs_have_delegation(inode, FMODE_READ))
1520 		nfsi->cache_validity |= invalid;
1521 
1522 	return 0;
1523  out_changed:
1524 	/*
1525 	 * Big trouble! The inode has become a different object.
1526 	 */
1527 #ifdef NFS_PARANOIA
1528 	printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1529 			__FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1530 #endif
1531 	/*
1532 	 * No need to worry about unhashing the dentry, as the
1533 	 * lookup validation will know that the inode is bad.
1534 	 * (But we fall through to invalidate the caches.)
1535 	 */
1536 	nfs_invalidate_inode(inode);
1537  out_err:
1538 	set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1539 	return -ESTALE;
1540 }
1541 
1542 /*
1543  * File system information
1544  */
1545 
1546 static int nfs_set_super(struct super_block *s, void *data)
1547 {
1548 	s->s_fs_info = data;
1549 	return set_anon_super(s, data);
1550 }
1551 
1552 static int nfs_compare_super(struct super_block *sb, void *data)
1553 {
1554 	struct nfs_server *server = data;
1555 	struct nfs_server *old = NFS_SB(sb);
1556 
1557 	if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1558 		return 0;
1559 	if (old->addr.sin_port != server->addr.sin_port)
1560 		return 0;
1561 	return !nfs_compare_fh(&old->fh, &server->fh);
1562 }
1563 
1564 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1565 	int flags, const char *dev_name, void *raw_data)
1566 {
1567 	int error;
1568 	struct nfs_server *server = NULL;
1569 	struct super_block *s;
1570 	struct nfs_fh *root;
1571 	struct nfs_mount_data *data = raw_data;
1572 
1573 	s = ERR_PTR(-EINVAL);
1574 	if (data == NULL) {
1575 		dprintk("%s: missing data argument\n", __FUNCTION__);
1576 		goto out_err;
1577 	}
1578 	if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1579 		dprintk("%s: bad mount version\n", __FUNCTION__);
1580 		goto out_err;
1581 	}
1582 	switch (data->version) {
1583 		case 1:
1584 			data->namlen = 0;
1585 		case 2:
1586 			data->bsize  = 0;
1587 		case 3:
1588 			if (data->flags & NFS_MOUNT_VER3) {
1589 				dprintk("%s: mount structure version %d does not support NFSv3\n",
1590 						__FUNCTION__,
1591 						data->version);
1592 				goto out_err;
1593 			}
1594 			data->root.size = NFS2_FHSIZE;
1595 			memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1596 		case 4:
1597 			if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1598 				dprintk("%s: mount structure version %d does not support strong security\n",
1599 						__FUNCTION__,
1600 						data->version);
1601 				goto out_err;
1602 			}
1603 		case 5:
1604 			memset(data->context, 0, sizeof(data->context));
1605 	}
1606 #ifndef CONFIG_NFS_V3
1607 	/* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1608 	s = ERR_PTR(-EPROTONOSUPPORT);
1609 	if (data->flags & NFS_MOUNT_VER3) {
1610 		dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1611 		goto out_err;
1612 	}
1613 #endif /* CONFIG_NFS_V3 */
1614 
1615 	s = ERR_PTR(-ENOMEM);
1616 	server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1617 	if (!server)
1618 		goto out_err;
1619 	memset(server, 0, sizeof(struct nfs_server));
1620 	/* Zero out the NFS state stuff */
1621 	init_nfsv4_state(server);
1622 	server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1623 
1624 	root = &server->fh;
1625 	if (data->flags & NFS_MOUNT_VER3)
1626 		root->size = data->root.size;
1627 	else
1628 		root->size = NFS2_FHSIZE;
1629 	s = ERR_PTR(-EINVAL);
1630 	if (root->size > sizeof(root->data)) {
1631 		dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1632 		goto out_err;
1633 	}
1634 	memcpy(root->data, data->root.data, root->size);
1635 
1636 	/* We now require that the mount process passes the remote address */
1637 	memcpy(&server->addr, &data->addr, sizeof(server->addr));
1638 	if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1639 		dprintk("%s: mount program didn't pass remote address!\n",
1640 				__FUNCTION__);
1641 		goto out_err;
1642 	}
1643 
1644 	/* Fire up rpciod if not yet running */
1645 	s = ERR_PTR(rpciod_up());
1646 	if (IS_ERR(s)) {
1647 		dprintk("%s: couldn't start rpciod! Error = %ld\n",
1648 				__FUNCTION__, PTR_ERR(s));
1649 		goto out_err;
1650 	}
1651 
1652 	s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1653 	if (IS_ERR(s) || s->s_root)
1654 		goto out_rpciod_down;
1655 
1656 	s->s_flags = flags;
1657 
1658 	error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1659 	if (error) {
1660 		up_write(&s->s_umount);
1661 		deactivate_super(s);
1662 		return ERR_PTR(error);
1663 	}
1664 	s->s_flags |= MS_ACTIVE;
1665 	return s;
1666 out_rpciod_down:
1667 	rpciod_down();
1668 out_err:
1669 	kfree(server);
1670 	return s;
1671 }
1672 
1673 static void nfs_kill_super(struct super_block *s)
1674 {
1675 	struct nfs_server *server = NFS_SB(s);
1676 
1677 	kill_anon_super(s);
1678 
1679 	if (!IS_ERR(server->client))
1680 		rpc_shutdown_client(server->client);
1681 	if (!IS_ERR(server->client_sys))
1682 		rpc_shutdown_client(server->client_sys);
1683 	if (!IS_ERR(server->client_acl))
1684 		rpc_shutdown_client(server->client_acl);
1685 
1686 	if (!(server->flags & NFS_MOUNT_NONLM))
1687 		lockd_down();	/* release rpc.lockd */
1688 
1689 	rpciod_down();		/* release rpciod */
1690 
1691 	if (server->hostname != NULL)
1692 		kfree(server->hostname);
1693 	kfree(server);
1694 }
1695 
1696 static struct file_system_type nfs_fs_type = {
1697 	.owner		= THIS_MODULE,
1698 	.name		= "nfs",
1699 	.get_sb		= nfs_get_sb,
1700 	.kill_sb	= nfs_kill_super,
1701 	.fs_flags	= FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1702 };
1703 
1704 #ifdef CONFIG_NFS_V4
1705 
1706 static void nfs4_clear_inode(struct inode *);
1707 
1708 
1709 static struct super_operations nfs4_sops = {
1710 	.alloc_inode	= nfs_alloc_inode,
1711 	.destroy_inode	= nfs_destroy_inode,
1712 	.write_inode	= nfs_write_inode,
1713 	.delete_inode	= nfs_delete_inode,
1714 	.statfs		= nfs_statfs,
1715 	.clear_inode	= nfs4_clear_inode,
1716 	.umount_begin	= nfs_umount_begin,
1717 	.show_options	= nfs_show_options,
1718 };
1719 
1720 /*
1721  * Clean out any remaining NFSv4 state that might be left over due
1722  * to open() calls that passed nfs_atomic_lookup, but failed to call
1723  * nfs_open().
1724  */
1725 static void nfs4_clear_inode(struct inode *inode)
1726 {
1727 	struct nfs_inode *nfsi = NFS_I(inode);
1728 
1729 	/* If we are holding a delegation, return it! */
1730 	nfs_inode_return_delegation(inode);
1731 	/* First call standard NFS clear_inode() code */
1732 	nfs_clear_inode(inode);
1733 	/* Now clear out any remaining state */
1734 	while (!list_empty(&nfsi->open_states)) {
1735 		struct nfs4_state *state;
1736 
1737 		state = list_entry(nfsi->open_states.next,
1738 				struct nfs4_state,
1739 				inode_states);
1740 		dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1741 				__FUNCTION__,
1742 				inode->i_sb->s_id,
1743 				(long long)NFS_FILEID(inode),
1744 				state);
1745 		BUG_ON(atomic_read(&state->count) != 1);
1746 		nfs4_close_state(state, state->state);
1747 	}
1748 }
1749 
1750 
1751 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1752 {
1753 	struct nfs_server *server;
1754 	struct nfs4_client *clp = NULL;
1755 	struct rpc_xprt *xprt = NULL;
1756 	struct rpc_clnt *clnt = NULL;
1757 	struct rpc_timeout timeparms;
1758 	rpc_authflavor_t authflavour;
1759 	int err = -EIO;
1760 
1761 	sb->s_blocksize_bits = 0;
1762 	sb->s_blocksize = 0;
1763 	server = NFS_SB(sb);
1764 	if (data->rsize != 0)
1765 		server->rsize = nfs_block_size(data->rsize, NULL);
1766 	if (data->wsize != 0)
1767 		server->wsize = nfs_block_size(data->wsize, NULL);
1768 	server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1769 	server->caps = NFS_CAP_ATOMIC_OPEN;
1770 
1771 	server->acregmin = data->acregmin*HZ;
1772 	server->acregmax = data->acregmax*HZ;
1773 	server->acdirmin = data->acdirmin*HZ;
1774 	server->acdirmax = data->acdirmax*HZ;
1775 
1776 	server->rpc_ops = &nfs_v4_clientops;
1777 
1778 	nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
1779 
1780 	clp = nfs4_get_client(&server->addr.sin_addr);
1781 	if (!clp) {
1782 		dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1783 		return -EIO;
1784 	}
1785 
1786 	/* Now create transport and client */
1787 	authflavour = RPC_AUTH_UNIX;
1788 	if (data->auth_flavourlen != 0) {
1789 		if (data->auth_flavourlen != 1) {
1790 			dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1791 					__FUNCTION__, data->auth_flavourlen);
1792 			err = -EINVAL;
1793 			goto out_fail;
1794 		}
1795 		if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1796 			err = -EFAULT;
1797 			goto out_fail;
1798 		}
1799 	}
1800 
1801 	down_write(&clp->cl_sem);
1802 	if (IS_ERR(clp->cl_rpcclient)) {
1803 		xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
1804 		if (IS_ERR(xprt)) {
1805 			up_write(&clp->cl_sem);
1806 			err = PTR_ERR(xprt);
1807 			dprintk("%s: cannot create RPC transport. Error = %d\n",
1808 					__FUNCTION__, err);
1809 			goto out_fail;
1810 		}
1811 		clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1812 				server->rpc_ops->version, authflavour);
1813 		if (IS_ERR(clnt)) {
1814 			up_write(&clp->cl_sem);
1815 			err = PTR_ERR(clnt);
1816 			dprintk("%s: cannot create RPC client. Error = %d\n",
1817 					__FUNCTION__, err);
1818 			goto out_fail;
1819 		}
1820 		clnt->cl_intr     = 1;
1821 		clnt->cl_softrtry = 1;
1822 		clnt->cl_chatty   = 1;
1823 		clp->cl_rpcclient = clnt;
1824 		clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1825 		if (IS_ERR(clp->cl_cred)) {
1826 			up_write(&clp->cl_sem);
1827 			err = PTR_ERR(clp->cl_cred);
1828 			clp->cl_cred = NULL;
1829 			goto out_fail;
1830 		}
1831 		memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1832 		nfs_idmap_new(clp);
1833 	}
1834 	if (list_empty(&clp->cl_superblocks)) {
1835 		err = nfs4_init_client(clp);
1836 		if (err != 0) {
1837 			up_write(&clp->cl_sem);
1838 			goto out_fail;
1839 		}
1840 	}
1841 	list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1842 	clnt = rpc_clone_client(clp->cl_rpcclient);
1843 	if (!IS_ERR(clnt))
1844 			server->nfs4_state = clp;
1845 	up_write(&clp->cl_sem);
1846 	clp = NULL;
1847 
1848 	if (IS_ERR(clnt)) {
1849 		err = PTR_ERR(clnt);
1850 		dprintk("%s: cannot create RPC client. Error = %d\n",
1851 				__FUNCTION__, err);
1852 		return err;
1853 	}
1854 
1855 	server->client    = clnt;
1856 
1857 	if (server->nfs4_state->cl_idmap == NULL) {
1858 		dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1859 		return -ENOMEM;
1860 	}
1861 
1862 	if (clnt->cl_auth->au_flavor != authflavour) {
1863 		struct rpc_auth *auth;
1864 
1865 		auth = rpcauth_create(authflavour, clnt);
1866 		if (IS_ERR(auth)) {
1867 			dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1868 			return PTR_ERR(auth);
1869 		}
1870 	}
1871 
1872 	sb->s_time_gran = 1;
1873 
1874 	sb->s_op = &nfs4_sops;
1875 	err = nfs_sb_init(sb, authflavour);
1876 	if (err == 0)
1877 		return 0;
1878 out_fail:
1879 	if (clp)
1880 		nfs4_put_client(clp);
1881 	return err;
1882 }
1883 
1884 static int nfs4_compare_super(struct super_block *sb, void *data)
1885 {
1886 	struct nfs_server *server = data;
1887 	struct nfs_server *old = NFS_SB(sb);
1888 
1889 	if (strcmp(server->hostname, old->hostname) != 0)
1890 		return 0;
1891 	if (strcmp(server->mnt_path, old->mnt_path) != 0)
1892 		return 0;
1893 	return 1;
1894 }
1895 
1896 static void *
1897 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1898 {
1899 	void *p = NULL;
1900 
1901 	if (!src->len)
1902 		return ERR_PTR(-EINVAL);
1903 	if (src->len < maxlen)
1904 		maxlen = src->len;
1905 	if (dst == NULL) {
1906 		p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1907 		if (p == NULL)
1908 			return ERR_PTR(-ENOMEM);
1909 	}
1910 	if (copy_from_user(dst, src->data, maxlen)) {
1911 		if (p != NULL)
1912 			kfree(p);
1913 		return ERR_PTR(-EFAULT);
1914 	}
1915 	dst[maxlen] = '\0';
1916 	return dst;
1917 }
1918 
1919 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1920 	int flags, const char *dev_name, void *raw_data)
1921 {
1922 	int error;
1923 	struct nfs_server *server;
1924 	struct super_block *s;
1925 	struct nfs4_mount_data *data = raw_data;
1926 	void *p;
1927 
1928 	if (data == NULL) {
1929 		dprintk("%s: missing data argument\n", __FUNCTION__);
1930 		return ERR_PTR(-EINVAL);
1931 	}
1932 	if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1933 		dprintk("%s: bad mount version\n", __FUNCTION__);
1934 		return ERR_PTR(-EINVAL);
1935 	}
1936 
1937 	server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1938 	if (!server)
1939 		return ERR_PTR(-ENOMEM);
1940 	memset(server, 0, sizeof(struct nfs_server));
1941 	/* Zero out the NFS state stuff */
1942 	init_nfsv4_state(server);
1943 	server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1944 
1945 	p = nfs_copy_user_string(NULL, &data->hostname, 256);
1946 	if (IS_ERR(p))
1947 		goto out_err;
1948 	server->hostname = p;
1949 
1950 	p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1951 	if (IS_ERR(p))
1952 		goto out_err;
1953 	server->mnt_path = p;
1954 
1955 	p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1956 			sizeof(server->ip_addr) - 1);
1957 	if (IS_ERR(p))
1958 		goto out_err;
1959 
1960 	/* We now require that the mount process passes the remote address */
1961 	if (data->host_addrlen != sizeof(server->addr)) {
1962 		s = ERR_PTR(-EINVAL);
1963 		goto out_free;
1964 	}
1965 	if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1966 		s = ERR_PTR(-EFAULT);
1967 		goto out_free;
1968 	}
1969 	if (server->addr.sin_family != AF_INET ||
1970 	    server->addr.sin_addr.s_addr == INADDR_ANY) {
1971 		dprintk("%s: mount program didn't pass remote IP address!\n",
1972 				__FUNCTION__);
1973 		s = ERR_PTR(-EINVAL);
1974 		goto out_free;
1975 	}
1976 
1977 	/* Fire up rpciod if not yet running */
1978 	s = ERR_PTR(rpciod_up());
1979 	if (IS_ERR(s)) {
1980 		dprintk("%s: couldn't start rpciod! Error = %ld\n",
1981 				__FUNCTION__, PTR_ERR(s));
1982 		goto out_free;
1983 	}
1984 
1985 	s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1986 
1987 	if (IS_ERR(s) || s->s_root)
1988 		goto out_free;
1989 
1990 	s->s_flags = flags;
1991 
1992 	error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1993 	if (error) {
1994 		up_write(&s->s_umount);
1995 		deactivate_super(s);
1996 		return ERR_PTR(error);
1997 	}
1998 	s->s_flags |= MS_ACTIVE;
1999 	return s;
2000 out_err:
2001 	s = (struct super_block *)p;
2002 out_free:
2003 	if (server->mnt_path)
2004 		kfree(server->mnt_path);
2005 	if (server->hostname)
2006 		kfree(server->hostname);
2007 	kfree(server);
2008 	return s;
2009 }
2010 
2011 static void nfs4_kill_super(struct super_block *sb)
2012 {
2013 	struct nfs_server *server = NFS_SB(sb);
2014 
2015 	nfs_return_all_delegations(sb);
2016 	kill_anon_super(sb);
2017 
2018 	nfs4_renewd_prepare_shutdown(server);
2019 
2020 	if (server->client != NULL && !IS_ERR(server->client))
2021 		rpc_shutdown_client(server->client);
2022 	rpciod_down();		/* release rpciod */
2023 
2024 	destroy_nfsv4_state(server);
2025 
2026 	if (server->hostname != NULL)
2027 		kfree(server->hostname);
2028 	kfree(server);
2029 }
2030 
2031 static struct file_system_type nfs4_fs_type = {
2032 	.owner		= THIS_MODULE,
2033 	.name		= "nfs4",
2034 	.get_sb		= nfs4_get_sb,
2035 	.kill_sb	= nfs4_kill_super,
2036 	.fs_flags	= FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2037 };
2038 
2039 #define nfs4_init_once(nfsi) \
2040 	do { \
2041 		INIT_LIST_HEAD(&(nfsi)->open_states); \
2042 		nfsi->delegation = NULL; \
2043 		nfsi->delegation_state = 0; \
2044 		init_rwsem(&nfsi->rwsem); \
2045 	} while(0)
2046 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
2047 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
2048 #else
2049 #define nfs4_init_once(nfsi) \
2050 	do { } while (0)
2051 #define register_nfs4fs() (0)
2052 #define unregister_nfs4fs()
2053 #endif
2054 
2055 extern int nfs_init_nfspagecache(void);
2056 extern void nfs_destroy_nfspagecache(void);
2057 extern int nfs_init_readpagecache(void);
2058 extern void nfs_destroy_readpagecache(void);
2059 extern int nfs_init_writepagecache(void);
2060 extern void nfs_destroy_writepagecache(void);
2061 #ifdef CONFIG_NFS_DIRECTIO
2062 extern int nfs_init_directcache(void);
2063 extern void nfs_destroy_directcache(void);
2064 #endif
2065 
2066 static kmem_cache_t * nfs_inode_cachep;
2067 
2068 static struct inode *nfs_alloc_inode(struct super_block *sb)
2069 {
2070 	struct nfs_inode *nfsi;
2071 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2072 	if (!nfsi)
2073 		return NULL;
2074 	nfsi->flags = 0UL;
2075 	nfsi->cache_validity = 0UL;
2076 #ifdef CONFIG_NFS_V3_ACL
2077 	nfsi->acl_access = ERR_PTR(-EAGAIN);
2078 	nfsi->acl_default = ERR_PTR(-EAGAIN);
2079 #endif
2080 #ifdef CONFIG_NFS_V4
2081 	nfsi->nfs4_acl = NULL;
2082 #endif /* CONFIG_NFS_V4 */
2083 	return &nfsi->vfs_inode;
2084 }
2085 
2086 static void nfs_destroy_inode(struct inode *inode)
2087 {
2088 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2089 }
2090 
2091 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2092 {
2093 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2094 
2095 	if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2096 	    SLAB_CTOR_CONSTRUCTOR) {
2097 		inode_init_once(&nfsi->vfs_inode);
2098 		spin_lock_init(&nfsi->req_lock);
2099 		INIT_LIST_HEAD(&nfsi->dirty);
2100 		INIT_LIST_HEAD(&nfsi->commit);
2101 		INIT_LIST_HEAD(&nfsi->open_files);
2102 		INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2103 		atomic_set(&nfsi->data_updates, 0);
2104 		nfsi->ndirty = 0;
2105 		nfsi->ncommit = 0;
2106 		nfsi->npages = 0;
2107 		nfs4_init_once(nfsi);
2108 	}
2109 }
2110 
2111 static int nfs_init_inodecache(void)
2112 {
2113 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2114 					     sizeof(struct nfs_inode),
2115 					     0, SLAB_RECLAIM_ACCOUNT,
2116 					     init_once, NULL);
2117 	if (nfs_inode_cachep == NULL)
2118 		return -ENOMEM;
2119 
2120 	return 0;
2121 }
2122 
2123 static void nfs_destroy_inodecache(void)
2124 {
2125 	if (kmem_cache_destroy(nfs_inode_cachep))
2126 		printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2127 }
2128 
2129 /*
2130  * Initialize NFS
2131  */
2132 static int __init init_nfs_fs(void)
2133 {
2134 	int err;
2135 
2136 	err = nfs_init_nfspagecache();
2137 	if (err)
2138 		goto out4;
2139 
2140 	err = nfs_init_inodecache();
2141 	if (err)
2142 		goto out3;
2143 
2144 	err = nfs_init_readpagecache();
2145 	if (err)
2146 		goto out2;
2147 
2148 	err = nfs_init_writepagecache();
2149 	if (err)
2150 		goto out1;
2151 
2152 #ifdef CONFIG_NFS_DIRECTIO
2153 	err = nfs_init_directcache();
2154 	if (err)
2155 		goto out0;
2156 #endif
2157 
2158 #ifdef CONFIG_PROC_FS
2159 	rpc_proc_register(&nfs_rpcstat);
2160 #endif
2161         err = register_filesystem(&nfs_fs_type);
2162 	if (err)
2163 		goto out;
2164 	if ((err = register_nfs4fs()) != 0)
2165 		goto out;
2166 	return 0;
2167 out:
2168 #ifdef CONFIG_PROC_FS
2169 	rpc_proc_unregister("nfs");
2170 #endif
2171 	nfs_destroy_writepagecache();
2172 #ifdef CONFIG_NFS_DIRECTIO
2173 out0:
2174 	nfs_destroy_directcache();
2175 #endif
2176 out1:
2177 	nfs_destroy_readpagecache();
2178 out2:
2179 	nfs_destroy_inodecache();
2180 out3:
2181 	nfs_destroy_nfspagecache();
2182 out4:
2183 	return err;
2184 }
2185 
2186 static void __exit exit_nfs_fs(void)
2187 {
2188 #ifdef CONFIG_NFS_DIRECTIO
2189 	nfs_destroy_directcache();
2190 #endif
2191 	nfs_destroy_writepagecache();
2192 	nfs_destroy_readpagecache();
2193 	nfs_destroy_inodecache();
2194 	nfs_destroy_nfspagecache();
2195 #ifdef CONFIG_PROC_FS
2196 	rpc_proc_unregister("nfs");
2197 #endif
2198 	unregister_filesystem(&nfs_fs_type);
2199 	unregister_nfs4fs();
2200 }
2201 
2202 /* Not quite true; I just maintain it */
2203 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2204 MODULE_LICENSE("GPL");
2205 
2206 module_init(init_nfs_fs)
2207 module_exit(exit_nfs_fs)
2208