xref: /openbmc/linux/fs/nfs/inode.c (revision 1da177e4)
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 "delegation.h"
43 
44 #define NFSDBG_FACILITY		NFSDBG_VFS
45 #define NFS_PARANOIA 1
46 
47 /* Maximum number of readahead requests
48  * FIXME: this should really be a sysctl so that users may tune it to suit
49  *        their needs. People that do NFS over a slow network, might for
50  *        instance want to reduce it to something closer to 1 for improved
51  *        interactive response.
52  */
53 #define NFS_MAX_READAHEAD	(RPC_DEF_SLOT_TABLE - 1)
54 
55 static void nfs_invalidate_inode(struct inode *);
56 static int nfs_update_inode(struct inode *, struct nfs_fattr *, unsigned long);
57 
58 static struct inode *nfs_alloc_inode(struct super_block *sb);
59 static void nfs_destroy_inode(struct inode *);
60 static int nfs_write_inode(struct inode *,int);
61 static void nfs_delete_inode(struct inode *);
62 static void nfs_clear_inode(struct inode *);
63 static void nfs_umount_begin(struct super_block *);
64 static int  nfs_statfs(struct super_block *, struct kstatfs *);
65 static int  nfs_show_options(struct seq_file *, struct vfsmount *);
66 
67 static struct rpc_program	nfs_program;
68 
69 static struct super_operations nfs_sops = {
70 	.alloc_inode	= nfs_alloc_inode,
71 	.destroy_inode	= nfs_destroy_inode,
72 	.write_inode	= nfs_write_inode,
73 	.delete_inode	= nfs_delete_inode,
74 	.statfs		= nfs_statfs,
75 	.clear_inode	= nfs_clear_inode,
76 	.umount_begin	= nfs_umount_begin,
77 	.show_options	= nfs_show_options,
78 };
79 
80 /*
81  * RPC cruft for NFS
82  */
83 static struct rpc_stat		nfs_rpcstat = {
84 	.program		= &nfs_program
85 };
86 static struct rpc_version *	nfs_version[] = {
87 	NULL,
88 	NULL,
89 	&nfs_version2,
90 #if defined(CONFIG_NFS_V3)
91 	&nfs_version3,
92 #elif defined(CONFIG_NFS_V4)
93 	NULL,
94 #endif
95 #if defined(CONFIG_NFS_V4)
96 	&nfs_version4,
97 #endif
98 };
99 
100 static struct rpc_program	nfs_program = {
101 	.name			= "nfs",
102 	.number			= NFS_PROGRAM,
103 	.nrvers			= sizeof(nfs_version) / sizeof(nfs_version[0]),
104 	.version		= nfs_version,
105 	.stats			= &nfs_rpcstat,
106 	.pipe_dir_name		= "/nfs",
107 };
108 
109 static inline unsigned long
110 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
111 {
112 	return nfs_fileid_to_ino_t(fattr->fileid);
113 }
114 
115 static int
116 nfs_write_inode(struct inode *inode, int sync)
117 {
118 	int flags = sync ? FLUSH_WAIT : 0;
119 	int ret;
120 
121 	ret = nfs_commit_inode(inode, 0, 0, flags);
122 	if (ret < 0)
123 		return ret;
124 	return 0;
125 }
126 
127 static void
128 nfs_delete_inode(struct inode * inode)
129 {
130 	dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
131 
132 	nfs_wb_all(inode);
133 	/*
134 	 * The following should never happen...
135 	 */
136 	if (nfs_have_writebacks(inode)) {
137 		printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
138 	}
139 
140 	clear_inode(inode);
141 }
142 
143 /*
144  * For the moment, the only task for the NFS clear_inode method is to
145  * release the mmap credential
146  */
147 static void
148 nfs_clear_inode(struct inode *inode)
149 {
150 	struct nfs_inode *nfsi = NFS_I(inode);
151 	struct rpc_cred *cred;
152 
153 	nfs_wb_all(inode);
154 	BUG_ON (!list_empty(&nfsi->open_files));
155 	cred = nfsi->cache_access.cred;
156 	if (cred)
157 		put_rpccred(cred);
158 	BUG_ON(atomic_read(&nfsi->data_updates) != 0);
159 }
160 
161 void
162 nfs_umount_begin(struct super_block *sb)
163 {
164 	struct nfs_server *server = NFS_SB(sb);
165 	struct rpc_clnt	*rpc;
166 
167 	/* -EIO all pending I/O */
168 	if ((rpc = server->client) != NULL)
169 		rpc_killall_tasks(rpc);
170 }
171 
172 
173 static inline unsigned long
174 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
175 {
176 	/* make sure blocksize is a power of two */
177 	if ((bsize & (bsize - 1)) || nrbitsp) {
178 		unsigned char	nrbits;
179 
180 		for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
181 			;
182 		bsize = 1 << nrbits;
183 		if (nrbitsp)
184 			*nrbitsp = nrbits;
185 	}
186 
187 	return bsize;
188 }
189 
190 /*
191  * Calculate the number of 512byte blocks used.
192  */
193 static inline unsigned long
194 nfs_calc_block_size(u64 tsize)
195 {
196 	loff_t used = (tsize + 511) >> 9;
197 	return (used > ULONG_MAX) ? ULONG_MAX : used;
198 }
199 
200 /*
201  * Compute and set NFS server blocksize
202  */
203 static inline unsigned long
204 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
205 {
206 	if (bsize < 1024)
207 		bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
208 	else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
209 		bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
210 
211 	return nfs_block_bits(bsize, nrbitsp);
212 }
213 
214 /*
215  * Obtain the root inode of the file system.
216  */
217 static struct inode *
218 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
219 {
220 	struct nfs_server	*server = NFS_SB(sb);
221 	struct inode *rooti;
222 	int			error;
223 
224 	error = server->rpc_ops->getroot(server, rootfh, fsinfo);
225 	if (error < 0) {
226 		dprintk("nfs_get_root: getattr error = %d\n", -error);
227 		return ERR_PTR(error);
228 	}
229 
230 	rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
231 	if (!rooti)
232 		return ERR_PTR(-ENOMEM);
233 	return rooti;
234 }
235 
236 /*
237  * Do NFS version-independent mount processing, and sanity checking
238  */
239 static int
240 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
241 {
242 	struct nfs_server	*server;
243 	struct inode		*root_inode;
244 	struct nfs_fattr	fattr;
245 	struct nfs_fsinfo	fsinfo = {
246 					.fattr = &fattr,
247 				};
248 	struct nfs_pathconf pathinfo = {
249 			.fattr = &fattr,
250 	};
251 	int no_root_error = 0;
252 	unsigned long max_rpc_payload;
253 
254 	/* We probably want something more informative here */
255 	snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
256 
257 	server = NFS_SB(sb);
258 
259 	sb->s_magic      = NFS_SUPER_MAGIC;
260 
261 	root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
262 	/* Did getting the root inode fail? */
263 	if (IS_ERR(root_inode)) {
264 		no_root_error = PTR_ERR(root_inode);
265 		goto out_no_root;
266 	}
267 	sb->s_root = d_alloc_root(root_inode);
268 	if (!sb->s_root) {
269 		no_root_error = -ENOMEM;
270 		goto out_no_root;
271 	}
272 	sb->s_root->d_op = server->rpc_ops->dentry_ops;
273 
274 	/* Get some general file system info */
275 	if (server->namelen == 0 &&
276 	    server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
277 		server->namelen = pathinfo.max_namelen;
278 	/* Work out a lot of parameters */
279 	if (server->rsize == 0)
280 		server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
281 	if (server->wsize == 0)
282 		server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
283 
284 	if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
285 		server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
286 	if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
287 		server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
288 
289 	max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
290 	if (server->rsize > max_rpc_payload)
291 		server->rsize = max_rpc_payload;
292 	if (server->wsize > max_rpc_payload)
293 		server->wsize = max_rpc_payload;
294 
295 	server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
296 	if (server->rpages > NFS_READ_MAXIOV) {
297 		server->rpages = NFS_READ_MAXIOV;
298 		server->rsize = server->rpages << PAGE_CACHE_SHIFT;
299 	}
300 
301 	server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
302         if (server->wpages > NFS_WRITE_MAXIOV) {
303 		server->wpages = NFS_WRITE_MAXIOV;
304                 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
305 	}
306 
307 	if (sb->s_blocksize == 0)
308 		sb->s_blocksize = nfs_block_bits(server->wsize,
309 							 &sb->s_blocksize_bits);
310 	server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
311 
312 	server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
313 	if (server->dtsize > PAGE_CACHE_SIZE)
314 		server->dtsize = PAGE_CACHE_SIZE;
315 	if (server->dtsize > server->rsize)
316 		server->dtsize = server->rsize;
317 
318 	if (server->flags & NFS_MOUNT_NOAC) {
319 		server->acregmin = server->acregmax = 0;
320 		server->acdirmin = server->acdirmax = 0;
321 		sb->s_flags |= MS_SYNCHRONOUS;
322 	}
323 	server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
324 
325 	sb->s_maxbytes = fsinfo.maxfilesize;
326 	if (sb->s_maxbytes > MAX_LFS_FILESIZE)
327 		sb->s_maxbytes = MAX_LFS_FILESIZE;
328 
329 	server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
330 	server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
331 
332 	/* We're airborne Set socket buffersize */
333 	rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
334 	return 0;
335 	/* Yargs. It didn't work out. */
336 out_no_root:
337 	dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
338 	if (!IS_ERR(root_inode))
339 		iput(root_inode);
340 	return no_root_error;
341 }
342 
343 /*
344  * Create an RPC client handle.
345  */
346 static struct rpc_clnt *
347 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
348 {
349 	struct rpc_timeout	timeparms;
350 	struct rpc_xprt		*xprt = NULL;
351 	struct rpc_clnt		*clnt = NULL;
352 	int			tcp   = (data->flags & NFS_MOUNT_TCP);
353 
354 	/* Initialize timeout values */
355 	timeparms.to_initval = data->timeo * HZ / 10;
356 	timeparms.to_retries = data->retrans;
357 	timeparms.to_maxval  = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
358 	timeparms.to_exponential = 1;
359 
360 	if (!timeparms.to_initval)
361 		timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
362 	if (!timeparms.to_retries)
363 		timeparms.to_retries = 5;
364 
365 	/* create transport and client */
366 	xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
367 				 &server->addr, &timeparms);
368 	if (IS_ERR(xprt)) {
369 		printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
370 		return (struct rpc_clnt *)xprt;
371 	}
372 	clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
373 				 server->rpc_ops->version, data->pseudoflavor);
374 	if (IS_ERR(clnt)) {
375 		printk(KERN_WARNING "NFS: cannot create RPC client.\n");
376 		goto out_fail;
377 	}
378 
379 	clnt->cl_intr     = 1;
380 	clnt->cl_softrtry = 1;
381 	clnt->cl_chatty   = 1;
382 
383 	return clnt;
384 
385 out_fail:
386 	xprt_destroy(xprt);
387 	return clnt;
388 }
389 
390 /*
391  * The way this works is that the mount process passes a structure
392  * in the data argument which contains the server's IP address
393  * and the root file handle obtained from the server's mount
394  * daemon. We stash these away in the private superblock fields.
395  */
396 static int
397 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
398 {
399 	struct nfs_server	*server;
400 	rpc_authflavor_t	authflavor;
401 
402 	server           = NFS_SB(sb);
403 	sb->s_blocksize_bits = 0;
404 	sb->s_blocksize = 0;
405 	if (data->bsize)
406 		sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
407 	if (data->rsize)
408 		server->rsize = nfs_block_size(data->rsize, NULL);
409 	if (data->wsize)
410 		server->wsize = nfs_block_size(data->wsize, NULL);
411 	server->flags    = data->flags & NFS_MOUNT_FLAGMASK;
412 
413 	server->acregmin = data->acregmin*HZ;
414 	server->acregmax = data->acregmax*HZ;
415 	server->acdirmin = data->acdirmin*HZ;
416 	server->acdirmax = data->acdirmax*HZ;
417 
418 	/* Start lockd here, before we might error out */
419 	if (!(server->flags & NFS_MOUNT_NONLM))
420 		lockd_up();
421 
422 	server->namelen  = data->namlen;
423 	server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
424 	if (!server->hostname)
425 		return -ENOMEM;
426 	strcpy(server->hostname, data->hostname);
427 
428 	/* Check NFS protocol revision and initialize RPC op vector
429 	 * and file handle pool. */
430 	if (server->flags & NFS_MOUNT_VER3) {
431 #ifdef CONFIG_NFS_V3
432 		server->rpc_ops = &nfs_v3_clientops;
433 		server->caps |= NFS_CAP_READDIRPLUS;
434 		if (data->version < 4) {
435 			printk(KERN_NOTICE "NFS: NFSv3 not supported by mount program.\n");
436 			return -EIO;
437 		}
438 #else
439 		printk(KERN_NOTICE "NFS: NFSv3 not supported.\n");
440 		return -EIO;
441 #endif
442 	} else {
443 		server->rpc_ops = &nfs_v2_clientops;
444 	}
445 
446 	/* Fill in pseudoflavor for mount version < 5 */
447 	if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
448 		data->pseudoflavor = RPC_AUTH_UNIX;
449 	authflavor = data->pseudoflavor;	/* save for sb_init() */
450 	/* XXX maybe we want to add a server->pseudoflavor field */
451 
452 	/* Create RPC client handles */
453 	server->client = nfs_create_client(server, data);
454 	if (IS_ERR(server->client))
455 		return PTR_ERR(server->client);
456 	/* RFC 2623, sec 2.3.2 */
457 	if (authflavor != RPC_AUTH_UNIX) {
458 		server->client_sys = rpc_clone_client(server->client);
459 		if (IS_ERR(server->client_sys))
460 			return PTR_ERR(server->client_sys);
461 		if (!rpcauth_create(RPC_AUTH_UNIX, server->client_sys))
462 			return -ENOMEM;
463 	} else {
464 		atomic_inc(&server->client->cl_count);
465 		server->client_sys = server->client;
466 	}
467 
468 	if (server->flags & NFS_MOUNT_VER3) {
469 		if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
470 			server->namelen = NFS3_MAXNAMLEN;
471 		sb->s_time_gran = 1;
472 	} else {
473 		if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
474 			server->namelen = NFS2_MAXNAMLEN;
475 	}
476 
477 	sb->s_op = &nfs_sops;
478 	return nfs_sb_init(sb, authflavor);
479 }
480 
481 static int
482 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
483 {
484 	struct nfs_server *server = NFS_SB(sb);
485 	unsigned char blockbits;
486 	unsigned long blockres;
487 	struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
488 	struct nfs_fattr fattr;
489 	struct nfs_fsstat res = {
490 			.fattr = &fattr,
491 	};
492 	int error;
493 
494 	lock_kernel();
495 
496 	error = server->rpc_ops->statfs(server, rootfh, &res);
497 	buf->f_type = NFS_SUPER_MAGIC;
498 	if (error < 0)
499 		goto out_err;
500 
501 	/*
502 	 * Current versions of glibc do not correctly handle the
503 	 * case where f_frsize != f_bsize.  Eventually we want to
504 	 * report the value of wtmult in this field.
505 	 */
506 	buf->f_frsize = sb->s_blocksize;
507 
508 	/*
509 	 * On most *nix systems, f_blocks, f_bfree, and f_bavail
510 	 * are reported in units of f_frsize.  Linux hasn't had
511 	 * an f_frsize field in its statfs struct until recently,
512 	 * thus historically Linux's sys_statfs reports these
513 	 * fields in units of f_bsize.
514 	 */
515 	buf->f_bsize = sb->s_blocksize;
516 	blockbits = sb->s_blocksize_bits;
517 	blockres = (1 << blockbits) - 1;
518 	buf->f_blocks = (res.tbytes + blockres) >> blockbits;
519 	buf->f_bfree = (res.fbytes + blockres) >> blockbits;
520 	buf->f_bavail = (res.abytes + blockres) >> blockbits;
521 
522 	buf->f_files = res.tfiles;
523 	buf->f_ffree = res.afiles;
524 
525 	buf->f_namelen = server->namelen;
526  out:
527 	unlock_kernel();
528 
529 	return 0;
530 
531  out_err:
532 	printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
533 	buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
534 	goto out;
535 
536 }
537 
538 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
539 {
540 	static struct proc_nfs_info {
541 		int flag;
542 		char *str;
543 		char *nostr;
544 	} nfs_info[] = {
545 		{ NFS_MOUNT_SOFT, ",soft", ",hard" },
546 		{ NFS_MOUNT_INTR, ",intr", "" },
547 		{ NFS_MOUNT_POSIX, ",posix", "" },
548 		{ NFS_MOUNT_TCP, ",tcp", ",udp" },
549 		{ NFS_MOUNT_NOCTO, ",nocto", "" },
550 		{ NFS_MOUNT_NOAC, ",noac", "" },
551 		{ NFS_MOUNT_NONLM, ",nolock", ",lock" },
552 		{ 0, NULL, NULL }
553 	};
554 	struct proc_nfs_info *nfs_infop;
555 	struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
556 
557 	seq_printf(m, ",v%d", nfss->rpc_ops->version);
558 	seq_printf(m, ",rsize=%d", nfss->rsize);
559 	seq_printf(m, ",wsize=%d", nfss->wsize);
560 	if (nfss->acregmin != 3*HZ)
561 		seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
562 	if (nfss->acregmax != 60*HZ)
563 		seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
564 	if (nfss->acdirmin != 30*HZ)
565 		seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
566 	if (nfss->acdirmax != 60*HZ)
567 		seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
568 	for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
569 		if (nfss->flags & nfs_infop->flag)
570 			seq_puts(m, nfs_infop->str);
571 		else
572 			seq_puts(m, nfs_infop->nostr);
573 	}
574 	seq_puts(m, ",addr=");
575 	seq_escape(m, nfss->hostname, " \t\n\\");
576 	return 0;
577 }
578 
579 /*
580  * Invalidate the local caches
581  */
582 void
583 nfs_zap_caches(struct inode *inode)
584 {
585 	struct nfs_inode *nfsi = NFS_I(inode);
586 	int mode = inode->i_mode;
587 
588 	NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
589 	NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
590 
591 	memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
592 	if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
593 		nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS;
594 	else
595 		nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS;
596 }
597 
598 /*
599  * Invalidate, but do not unhash, the inode
600  */
601 static void
602 nfs_invalidate_inode(struct inode *inode)
603 {
604 	umode_t save_mode = inode->i_mode;
605 
606 	make_bad_inode(inode);
607 	inode->i_mode = save_mode;
608 	nfs_zap_caches(inode);
609 }
610 
611 struct nfs_find_desc {
612 	struct nfs_fh		*fh;
613 	struct nfs_fattr	*fattr;
614 };
615 
616 /*
617  * In NFSv3 we can have 64bit inode numbers. In order to support
618  * this, and re-exported directories (also seen in NFSv2)
619  * we are forced to allow 2 different inodes to have the same
620  * i_ino.
621  */
622 static int
623 nfs_find_actor(struct inode *inode, void *opaque)
624 {
625 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
626 	struct nfs_fh		*fh = desc->fh;
627 	struct nfs_fattr	*fattr = desc->fattr;
628 
629 	if (NFS_FILEID(inode) != fattr->fileid)
630 		return 0;
631 	if (nfs_compare_fh(NFS_FH(inode), fh))
632 		return 0;
633 	if (is_bad_inode(inode) || NFS_STALE(inode))
634 		return 0;
635 	return 1;
636 }
637 
638 static int
639 nfs_init_locked(struct inode *inode, void *opaque)
640 {
641 	struct nfs_find_desc	*desc = (struct nfs_find_desc *)opaque;
642 	struct nfs_fattr	*fattr = desc->fattr;
643 
644 	NFS_FILEID(inode) = fattr->fileid;
645 	nfs_copy_fh(NFS_FH(inode), desc->fh);
646 	return 0;
647 }
648 
649 /* Don't use READDIRPLUS on directories that we believe are too large */
650 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
651 
652 /*
653  * This is our front-end to iget that looks up inodes by file handle
654  * instead of inode number.
655  */
656 struct inode *
657 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
658 {
659 	struct nfs_find_desc desc = {
660 		.fh	= fh,
661 		.fattr	= fattr
662 	};
663 	struct inode *inode = NULL;
664 	unsigned long hash;
665 
666 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
667 		goto out_no_inode;
668 
669 	if (!fattr->nlink) {
670 		printk("NFS: Buggy server - nlink == 0!\n");
671 		goto out_no_inode;
672 	}
673 
674 	hash = nfs_fattr_to_ino_t(fattr);
675 
676 	if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
677 		goto out_no_inode;
678 
679 	if (inode->i_state & I_NEW) {
680 		struct nfs_inode *nfsi = NFS_I(inode);
681 
682 		/* We set i_ino for the few things that still rely on it,
683 		 * such as stat(2) */
684 		inode->i_ino = hash;
685 
686 		/* We can't support update_atime(), since the server will reset it */
687 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
688 		inode->i_mode = fattr->mode;
689 		/* Why so? Because we want revalidate for devices/FIFOs, and
690 		 * that's precisely what we have in nfs_file_inode_operations.
691 		 */
692 		inode->i_op = &nfs_file_inode_operations;
693 		if (S_ISREG(inode->i_mode)) {
694 			inode->i_fop = &nfs_file_operations;
695 			inode->i_data.a_ops = &nfs_file_aops;
696 			inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
697 		} else if (S_ISDIR(inode->i_mode)) {
698 			inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
699 			inode->i_fop = &nfs_dir_operations;
700 			if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
701 			    && fattr->size <= NFS_LIMIT_READDIRPLUS)
702 				NFS_FLAGS(inode) |= NFS_INO_ADVISE_RDPLUS;
703 		} else if (S_ISLNK(inode->i_mode))
704 			inode->i_op = &nfs_symlink_inode_operations;
705 		else
706 			init_special_inode(inode, inode->i_mode, fattr->rdev);
707 
708 		nfsi->read_cache_jiffies = fattr->timestamp;
709 		inode->i_atime = fattr->atime;
710 		inode->i_mtime = fattr->mtime;
711 		inode->i_ctime = fattr->ctime;
712 		if (fattr->valid & NFS_ATTR_FATTR_V4)
713 			nfsi->change_attr = fattr->change_attr;
714 		inode->i_size = nfs_size_to_loff_t(fattr->size);
715 		inode->i_nlink = fattr->nlink;
716 		inode->i_uid = fattr->uid;
717 		inode->i_gid = fattr->gid;
718 		if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
719 			/*
720 			 * report the blocks in 512byte units
721 			 */
722 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
723 			inode->i_blksize = inode->i_sb->s_blocksize;
724 		} else {
725 			inode->i_blocks = fattr->du.nfs2.blocks;
726 			inode->i_blksize = fattr->du.nfs2.blocksize;
727 		}
728 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
729 		nfsi->attrtimeo_timestamp = jiffies;
730 		memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
731 		nfsi->cache_access.cred = NULL;
732 
733 		unlock_new_inode(inode);
734 	} else
735 		nfs_refresh_inode(inode, fattr);
736 	dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
737 		inode->i_sb->s_id,
738 		(long long)NFS_FILEID(inode),
739 		atomic_read(&inode->i_count));
740 
741 out:
742 	return inode;
743 
744 out_no_inode:
745 	printk("nfs_fhget: iget failed\n");
746 	goto out;
747 }
748 
749 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
750 
751 int
752 nfs_setattr(struct dentry *dentry, struct iattr *attr)
753 {
754 	struct inode *inode = dentry->d_inode;
755 	struct nfs_fattr fattr;
756 	int error;
757 
758 	if (attr->ia_valid & ATTR_SIZE) {
759 		if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
760 			attr->ia_valid &= ~ATTR_SIZE;
761 	}
762 
763 	/* Optimization: if the end result is no change, don't RPC */
764 	attr->ia_valid &= NFS_VALID_ATTRS;
765 	if (attr->ia_valid == 0)
766 		return 0;
767 
768 	lock_kernel();
769 	nfs_begin_data_update(inode);
770 	/* Write all dirty data if we're changing file permissions or size */
771 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
772 		if (filemap_fdatawrite(inode->i_mapping) == 0)
773 			filemap_fdatawait(inode->i_mapping);
774 		nfs_wb_all(inode);
775 	}
776 	error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
777 	if (error == 0) {
778 		nfs_refresh_inode(inode, &fattr);
779 		if ((attr->ia_valid & ATTR_MODE) != 0) {
780 			int mode;
781 			mode = inode->i_mode & ~S_IALLUGO;
782 			mode |= attr->ia_mode & S_IALLUGO;
783 			inode->i_mode = mode;
784 		}
785 		if ((attr->ia_valid & ATTR_UID) != 0)
786 			inode->i_uid = attr->ia_uid;
787 		if ((attr->ia_valid & ATTR_GID) != 0)
788 			inode->i_gid = attr->ia_gid;
789 		if ((attr->ia_valid & ATTR_SIZE) != 0) {
790 			inode->i_size = attr->ia_size;
791 			vmtruncate(inode, attr->ia_size);
792 		}
793 	}
794 	if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
795 		NFS_FLAGS(inode) |= NFS_INO_INVALID_ACCESS;
796 	nfs_end_data_update(inode);
797 	unlock_kernel();
798 	return error;
799 }
800 
801 /*
802  * Wait for the inode to get unlocked.
803  * (Used for NFS_INO_LOCKED and NFS_INO_REVALIDATING).
804  */
805 static int
806 nfs_wait_on_inode(struct inode *inode, int flag)
807 {
808 	struct rpc_clnt	*clnt = NFS_CLIENT(inode);
809 	struct nfs_inode *nfsi = NFS_I(inode);
810 
811 	int error;
812 	if (!(NFS_FLAGS(inode) & flag))
813 		return 0;
814 	atomic_inc(&inode->i_count);
815 	error = nfs_wait_event(clnt, nfsi->nfs_i_wait,
816 				!(NFS_FLAGS(inode) & flag));
817 	iput(inode);
818 	return error;
819 }
820 
821 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
822 {
823 	struct inode *inode = dentry->d_inode;
824 	struct nfs_inode *nfsi = NFS_I(inode);
825 	int need_atime = nfsi->flags & NFS_INO_INVALID_ATIME;
826 	int err;
827 
828 	if (__IS_FLG(inode, MS_NOATIME))
829 		need_atime = 0;
830 	else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
831 		need_atime = 0;
832 	/* We may force a getattr if the user cares about atime */
833 	if (need_atime)
834 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
835 	else
836 		err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
837 	if (!err)
838 		generic_fillattr(inode, stat);
839 	return err;
840 }
841 
842 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
843 {
844 	struct nfs_open_context *ctx;
845 
846 	ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
847 	if (ctx != NULL) {
848 		atomic_set(&ctx->count, 1);
849 		ctx->dentry = dget(dentry);
850 		ctx->cred = get_rpccred(cred);
851 		ctx->state = NULL;
852 		ctx->lockowner = current->files;
853 		ctx->error = 0;
854 		init_waitqueue_head(&ctx->waitq);
855 	}
856 	return ctx;
857 }
858 
859 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
860 {
861 	if (ctx != NULL)
862 		atomic_inc(&ctx->count);
863 	return ctx;
864 }
865 
866 void put_nfs_open_context(struct nfs_open_context *ctx)
867 {
868 	if (atomic_dec_and_test(&ctx->count)) {
869 		if (!list_empty(&ctx->list)) {
870 			struct inode *inode = ctx->dentry->d_inode;
871 			spin_lock(&inode->i_lock);
872 			list_del(&ctx->list);
873 			spin_unlock(&inode->i_lock);
874 		}
875 		if (ctx->state != NULL)
876 			nfs4_close_state(ctx->state, ctx->mode);
877 		if (ctx->cred != NULL)
878 			put_rpccred(ctx->cred);
879 		dput(ctx->dentry);
880 		kfree(ctx);
881 	}
882 }
883 
884 /*
885  * Ensure that mmap has a recent RPC credential for use when writing out
886  * shared pages
887  */
888 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
889 {
890 	struct inode *inode = filp->f_dentry->d_inode;
891 	struct nfs_inode *nfsi = NFS_I(inode);
892 
893 	filp->private_data = get_nfs_open_context(ctx);
894 	spin_lock(&inode->i_lock);
895 	list_add(&ctx->list, &nfsi->open_files);
896 	spin_unlock(&inode->i_lock);
897 }
898 
899 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
900 {
901 	struct nfs_inode *nfsi = NFS_I(inode);
902 	struct nfs_open_context *pos, *ctx = NULL;
903 
904 	spin_lock(&inode->i_lock);
905 	list_for_each_entry(pos, &nfsi->open_files, list) {
906 		if ((pos->mode & mode) == mode) {
907 			ctx = get_nfs_open_context(pos);
908 			break;
909 		}
910 	}
911 	spin_unlock(&inode->i_lock);
912 	return ctx;
913 }
914 
915 void nfs_file_clear_open_context(struct file *filp)
916 {
917 	struct inode *inode = filp->f_dentry->d_inode;
918 	struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
919 
920 	if (ctx) {
921 		filp->private_data = NULL;
922 		spin_lock(&inode->i_lock);
923 		list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
924 		spin_unlock(&inode->i_lock);
925 		put_nfs_open_context(ctx);
926 	}
927 }
928 
929 /*
930  * These allocate and release file read/write context information.
931  */
932 int nfs_open(struct inode *inode, struct file *filp)
933 {
934 	struct nfs_open_context *ctx;
935 	struct rpc_cred *cred;
936 
937 	cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
938 	if (IS_ERR(cred))
939 		return PTR_ERR(cred);
940 	ctx = alloc_nfs_open_context(filp->f_dentry, cred);
941 	put_rpccred(cred);
942 	if (ctx == NULL)
943 		return -ENOMEM;
944 	ctx->mode = filp->f_mode;
945 	nfs_file_set_open_context(filp, ctx);
946 	put_nfs_open_context(ctx);
947 	if ((filp->f_mode & FMODE_WRITE) != 0)
948 		nfs_begin_data_update(inode);
949 	return 0;
950 }
951 
952 int nfs_release(struct inode *inode, struct file *filp)
953 {
954 	if ((filp->f_mode & FMODE_WRITE) != 0)
955 		nfs_end_data_update(inode);
956 	nfs_file_clear_open_context(filp);
957 	return 0;
958 }
959 
960 /*
961  * This function is called whenever some part of NFS notices that
962  * the cached attributes have to be refreshed.
963  */
964 int
965 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
966 {
967 	int		 status = -ESTALE;
968 	struct nfs_fattr fattr;
969 	struct nfs_inode *nfsi = NFS_I(inode);
970 	unsigned long verifier;
971 	unsigned int flags;
972 
973 	dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
974 		inode->i_sb->s_id, (long long)NFS_FILEID(inode));
975 
976 	lock_kernel();
977 	if (!inode || is_bad_inode(inode))
978  		goto out_nowait;
979 	if (NFS_STALE(inode))
980  		goto out_nowait;
981 
982 	while (NFS_REVALIDATING(inode)) {
983 		status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
984 		if (status < 0)
985 			goto out_nowait;
986 		if (NFS_ATTRTIMEO(inode) == 0)
987 			continue;
988 		if (NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME))
989 			continue;
990 		status = NFS_STALE(inode) ? -ESTALE : 0;
991 		goto out_nowait;
992 	}
993 	NFS_FLAGS(inode) |= NFS_INO_REVALIDATING;
994 
995 	/* Protect against RPC races by saving the change attribute */
996 	verifier = nfs_save_change_attribute(inode);
997 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
998 	if (status != 0) {
999 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1000 			 inode->i_sb->s_id,
1001 			 (long long)NFS_FILEID(inode), status);
1002 		if (status == -ESTALE) {
1003 			nfs_zap_caches(inode);
1004 			if (!S_ISDIR(inode->i_mode))
1005 				NFS_FLAGS(inode) |= NFS_INO_STALE;
1006 		}
1007 		goto out;
1008 	}
1009 
1010 	status = nfs_update_inode(inode, &fattr, verifier);
1011 	if (status) {
1012 		dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1013 			 inode->i_sb->s_id,
1014 			 (long long)NFS_FILEID(inode), status);
1015 		goto out;
1016 	}
1017 	flags = nfsi->flags;
1018 	/*
1019 	 * We may need to keep the attributes marked as invalid if
1020 	 * we raced with nfs_end_attr_update().
1021 	 */
1022 	if (verifier == nfsi->cache_change_attribute)
1023 		nfsi->flags &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1024 	/* Do the page cache invalidation */
1025 	if (flags & NFS_INO_INVALID_DATA) {
1026 		if (S_ISREG(inode->i_mode)) {
1027 			if (filemap_fdatawrite(inode->i_mapping) == 0)
1028 				filemap_fdatawait(inode->i_mapping);
1029 			nfs_wb_all(inode);
1030 		}
1031 		nfsi->flags &= ~NFS_INO_INVALID_DATA;
1032 		invalidate_inode_pages2(inode->i_mapping);
1033 		memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
1034 		dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1035 				inode->i_sb->s_id,
1036 				(long long)NFS_FILEID(inode));
1037 		/* This ensures we revalidate dentries */
1038 		nfsi->cache_change_attribute++;
1039 	}
1040 	dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1041 		inode->i_sb->s_id,
1042 		(long long)NFS_FILEID(inode));
1043 
1044 out:
1045 	NFS_FLAGS(inode) &= ~NFS_INO_REVALIDATING;
1046 	wake_up(&nfsi->nfs_i_wait);
1047  out_nowait:
1048 	unlock_kernel();
1049 	return status;
1050 }
1051 
1052 int nfs_attribute_timeout(struct inode *inode)
1053 {
1054 	struct nfs_inode *nfsi = NFS_I(inode);
1055 
1056 	if (nfs_have_delegation(inode, FMODE_READ))
1057 		return 0;
1058 	return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1059 }
1060 
1061 /**
1062  * nfs_revalidate_inode - Revalidate the inode attributes
1063  * @server - pointer to nfs_server struct
1064  * @inode - pointer to inode struct
1065  *
1066  * Updates inode attribute information by retrieving the data from the server.
1067  */
1068 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1069 {
1070 	if (!(NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1071 			&& !nfs_attribute_timeout(inode))
1072 		return NFS_STALE(inode) ? -ESTALE : 0;
1073 	return __nfs_revalidate_inode(server, inode);
1074 }
1075 
1076 /**
1077  * nfs_begin_data_update
1078  * @inode - pointer to inode
1079  * Declare that a set of operations will update file data on the server
1080  */
1081 void nfs_begin_data_update(struct inode *inode)
1082 {
1083 	atomic_inc(&NFS_I(inode)->data_updates);
1084 }
1085 
1086 /**
1087  * nfs_end_data_update
1088  * @inode - pointer to inode
1089  * Declare end of the operations that will update file data
1090  * This will mark the inode as immediately needing revalidation
1091  * of its attribute cache.
1092  */
1093 void nfs_end_data_update(struct inode *inode)
1094 {
1095 	struct nfs_inode *nfsi = NFS_I(inode);
1096 
1097 	if (!nfs_have_delegation(inode, FMODE_READ)) {
1098 		/* Mark the attribute cache for revalidation */
1099 		nfsi->flags |= NFS_INO_INVALID_ATTR;
1100 		/* Directories and symlinks: invalidate page cache too */
1101 		if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1102 			nfsi->flags |= NFS_INO_INVALID_DATA;
1103 	}
1104 	nfsi->cache_change_attribute ++;
1105 	atomic_dec(&nfsi->data_updates);
1106 }
1107 
1108 /**
1109  * nfs_end_data_update_defer
1110  * @inode - pointer to inode
1111  * Declare end of the operations that will update file data
1112  * This will defer marking the inode as needing revalidation
1113  * unless there are no other pending updates.
1114  */
1115 void nfs_end_data_update_defer(struct inode *inode)
1116 {
1117 	struct nfs_inode *nfsi = NFS_I(inode);
1118 
1119 	if (atomic_dec_and_test(&nfsi->data_updates)) {
1120 		/* Mark the attribute cache for revalidation */
1121 		nfsi->flags |= NFS_INO_INVALID_ATTR;
1122 		/* Directories and symlinks: invalidate page cache too */
1123 		if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1124 			nfsi->flags |= NFS_INO_INVALID_DATA;
1125 		nfsi->cache_change_attribute ++;
1126 	}
1127 }
1128 
1129 /**
1130  * nfs_refresh_inode - verify consistency of the inode attribute cache
1131  * @inode - pointer to inode
1132  * @fattr - updated attributes
1133  *
1134  * Verifies the attribute cache. If we have just changed the attributes,
1135  * so that fattr carries weak cache consistency data, then it may
1136  * also update the ctime/mtime/change_attribute.
1137  */
1138 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1139 {
1140 	struct nfs_inode *nfsi = NFS_I(inode);
1141 	loff_t cur_size, new_isize;
1142 	int data_unstable;
1143 
1144 	/* Do we hold a delegation? */
1145 	if (nfs_have_delegation(inode, FMODE_READ))
1146 		return 0;
1147 
1148 	/* Are we in the process of updating data on the server? */
1149 	data_unstable = nfs_caches_unstable(inode);
1150 
1151 	if (fattr->valid & NFS_ATTR_FATTR_V4) {
1152 		if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1153 				&& nfsi->change_attr == fattr->pre_change_attr)
1154 			nfsi->change_attr = fattr->change_attr;
1155 		if (!data_unstable && nfsi->change_attr != fattr->change_attr)
1156 			nfsi->flags |= NFS_INO_INVALID_ATTR;
1157 	}
1158 
1159 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1160 		return 0;
1161 
1162 	/* Has the inode gone and changed behind our back? */
1163 	if (nfsi->fileid != fattr->fileid
1164 			|| (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1165 		return -EIO;
1166 
1167 	cur_size = i_size_read(inode);
1168  	new_isize = nfs_size_to_loff_t(fattr->size);
1169 
1170 	/* If we have atomic WCC data, we may update some attributes */
1171 	if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1172 		if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1173 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1174 		if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1175 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1176 	}
1177 
1178 	/* Verify a few of the more important attributes */
1179 	if (!data_unstable) {
1180 		if (!timespec_equal(&inode->i_mtime, &fattr->mtime)
1181 				|| cur_size != new_isize)
1182 			nfsi->flags |= NFS_INO_INVALID_ATTR;
1183 	} else if (S_ISREG(inode->i_mode) && new_isize > cur_size)
1184 			nfsi->flags |= NFS_INO_INVALID_ATTR;
1185 
1186 	/* Have any file permissions changed? */
1187 	if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1188 			|| inode->i_uid != fattr->uid
1189 			|| inode->i_gid != fattr->gid)
1190 		nfsi->flags |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1191 
1192 	/* Has the link count changed? */
1193 	if (inode->i_nlink != fattr->nlink)
1194 		nfsi->flags |= NFS_INO_INVALID_ATTR;
1195 
1196 	if (!timespec_equal(&inode->i_atime, &fattr->atime))
1197 		nfsi->flags |= NFS_INO_INVALID_ATIME;
1198 
1199 	nfsi->read_cache_jiffies = fattr->timestamp;
1200 	return 0;
1201 }
1202 
1203 /*
1204  * Many nfs protocol calls return the new file attributes after
1205  * an operation.  Here we update the inode to reflect the state
1206  * of the server's inode.
1207  *
1208  * This is a bit tricky because we have to make sure all dirty pages
1209  * have been sent off to the server before calling invalidate_inode_pages.
1210  * To make sure no other process adds more write requests while we try
1211  * our best to flush them, we make them sleep during the attribute refresh.
1212  *
1213  * A very similar scenario holds for the dir cache.
1214  */
1215 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1216 {
1217 	struct nfs_inode *nfsi = NFS_I(inode);
1218 	__u64		new_size;
1219 	loff_t		new_isize;
1220 	unsigned int	invalid = 0;
1221 	loff_t		cur_isize;
1222 	int data_unstable;
1223 
1224 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1225 			__FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1226 			atomic_read(&inode->i_count), fattr->valid);
1227 
1228 	if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1229 		return 0;
1230 
1231 	if (nfsi->fileid != fattr->fileid) {
1232 		printk(KERN_ERR "%s: inode number mismatch\n"
1233 		       "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1234 		       __FUNCTION__,
1235 		       inode->i_sb->s_id, (long long)nfsi->fileid,
1236 		       inode->i_sb->s_id, (long long)fattr->fileid);
1237 		goto out_err;
1238 	}
1239 
1240 	/*
1241 	 * Make sure the inode's type hasn't changed.
1242 	 */
1243 	if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1244 		goto out_changed;
1245 
1246 	/*
1247 	 * Update the read time so we don't revalidate too often.
1248 	 */
1249 	nfsi->read_cache_jiffies = fattr->timestamp;
1250 
1251 	/* Are we racing with known updates of the metadata on the server? */
1252 	data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1253 
1254 	/* Check if the file size agrees */
1255 	new_size = fattr->size;
1256  	new_isize = nfs_size_to_loff_t(fattr->size);
1257 	cur_isize = i_size_read(inode);
1258 	if (cur_isize != new_size) {
1259 #ifdef NFS_DEBUG_VERBOSE
1260 		printk(KERN_DEBUG "NFS: isize change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1261 #endif
1262 		/*
1263 		 * If we have pending writebacks, things can get
1264 		 * messy.
1265 		 */
1266 		if (S_ISREG(inode->i_mode) && data_unstable) {
1267 			if (new_isize > cur_isize) {
1268 				inode->i_size = new_isize;
1269 				invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1270 			}
1271 		} else {
1272 			inode->i_size = new_isize;
1273 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1274 		}
1275 	}
1276 
1277 	/*
1278 	 * Note: we don't check inode->i_mtime since pipes etc.
1279 	 *       can change this value in VFS without requiring a
1280 	 *	 cache revalidation.
1281 	 */
1282 	if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1283 		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1284 #ifdef NFS_DEBUG_VERBOSE
1285 		printk(KERN_DEBUG "NFS: mtime change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1286 #endif
1287 		if (!data_unstable)
1288 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1289 	}
1290 
1291 	if ((fattr->valid & NFS_ATTR_FATTR_V4)
1292 	    && nfsi->change_attr != fattr->change_attr) {
1293 #ifdef NFS_DEBUG_VERBOSE
1294 		printk(KERN_DEBUG "NFS: change_attr change on %s/%ld\n",
1295 		       inode->i_sb->s_id, inode->i_ino);
1296 #endif
1297 		nfsi->change_attr = fattr->change_attr;
1298 		if (!data_unstable)
1299 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS;
1300 	}
1301 
1302 	memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1303 	memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1304 
1305 	if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1306 	    inode->i_uid != fattr->uid ||
1307 	    inode->i_gid != fattr->gid)
1308 		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS;
1309 
1310 	inode->i_mode = fattr->mode;
1311 	inode->i_nlink = fattr->nlink;
1312 	inode->i_uid = fattr->uid;
1313 	inode->i_gid = fattr->gid;
1314 
1315 	if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1316 		/*
1317 		 * report the blocks in 512byte units
1318 		 */
1319 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1320 		inode->i_blksize = inode->i_sb->s_blocksize;
1321  	} else {
1322  		inode->i_blocks = fattr->du.nfs2.blocks;
1323  		inode->i_blksize = fattr->du.nfs2.blocksize;
1324  	}
1325 
1326 	/* Update attrtimeo value if we're out of the unstable period */
1327 	if (invalid & NFS_INO_INVALID_ATTR) {
1328 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1329 		nfsi->attrtimeo_timestamp = jiffies;
1330 	} else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1331 		if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1332 			nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1333 		nfsi->attrtimeo_timestamp = jiffies;
1334 	}
1335 	/* Don't invalidate the data if we were to blame */
1336 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1337 				|| S_ISLNK(inode->i_mode)))
1338 		invalid &= ~NFS_INO_INVALID_DATA;
1339 	if (!nfs_have_delegation(inode, FMODE_READ))
1340 		nfsi->flags |= invalid;
1341 
1342 	return 0;
1343  out_changed:
1344 	/*
1345 	 * Big trouble! The inode has become a different object.
1346 	 */
1347 #ifdef NFS_PARANOIA
1348 	printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1349 			__FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1350 #endif
1351 	/*
1352 	 * No need to worry about unhashing the dentry, as the
1353 	 * lookup validation will know that the inode is bad.
1354 	 * (But we fall through to invalidate the caches.)
1355 	 */
1356 	nfs_invalidate_inode(inode);
1357  out_err:
1358 	NFS_FLAGS(inode) |= NFS_INO_STALE;
1359 	return -ESTALE;
1360 }
1361 
1362 /*
1363  * File system information
1364  */
1365 
1366 static int nfs_set_super(struct super_block *s, void *data)
1367 {
1368 	s->s_fs_info = data;
1369 	return set_anon_super(s, data);
1370 }
1371 
1372 static int nfs_compare_super(struct super_block *sb, void *data)
1373 {
1374 	struct nfs_server *server = data;
1375 	struct nfs_server *old = NFS_SB(sb);
1376 
1377 	if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1378 		return 0;
1379 	if (old->addr.sin_port != server->addr.sin_port)
1380 		return 0;
1381 	return !nfs_compare_fh(&old->fh, &server->fh);
1382 }
1383 
1384 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1385 	int flags, const char *dev_name, void *raw_data)
1386 {
1387 	int error;
1388 	struct nfs_server *server;
1389 	struct super_block *s;
1390 	struct nfs_fh *root;
1391 	struct nfs_mount_data *data = raw_data;
1392 
1393 	if (!data) {
1394 		printk("nfs_read_super: missing data argument\n");
1395 		return ERR_PTR(-EINVAL);
1396 	}
1397 
1398 	server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1399 	if (!server)
1400 		return ERR_PTR(-ENOMEM);
1401 	memset(server, 0, sizeof(struct nfs_server));
1402 	/* Zero out the NFS state stuff */
1403 	init_nfsv4_state(server);
1404 
1405 	if (data->version != NFS_MOUNT_VERSION) {
1406 		printk("nfs warning: mount version %s than kernel\n",
1407 			data->version < NFS_MOUNT_VERSION ? "older" : "newer");
1408 		if (data->version < 2)
1409 			data->namlen = 0;
1410 		if (data->version < 3)
1411 			data->bsize  = 0;
1412 		if (data->version < 4) {
1413 			data->flags &= ~NFS_MOUNT_VER3;
1414 			data->root.size = NFS2_FHSIZE;
1415 			memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1416 		}
1417 		if (data->version < 5)
1418 			data->flags &= ~NFS_MOUNT_SECFLAVOUR;
1419 	}
1420 
1421 	root = &server->fh;
1422 	if (data->flags & NFS_MOUNT_VER3)
1423 		root->size = data->root.size;
1424 	else
1425 		root->size = NFS2_FHSIZE;
1426 	if (root->size > sizeof(root->data)) {
1427 		printk("nfs_get_sb: invalid root filehandle\n");
1428 		kfree(server);
1429 		return ERR_PTR(-EINVAL);
1430 	}
1431 	memcpy(root->data, data->root.data, root->size);
1432 
1433 	/* We now require that the mount process passes the remote address */
1434 	memcpy(&server->addr, &data->addr, sizeof(server->addr));
1435 	if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1436 		printk("NFS: mount program didn't pass remote address!\n");
1437 		kfree(server);
1438 		return ERR_PTR(-EINVAL);
1439 	}
1440 
1441 	s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1442 
1443 	if (IS_ERR(s) || s->s_root) {
1444 		kfree(server);
1445 		return s;
1446 	}
1447 
1448 	s->s_flags = flags;
1449 
1450 	/* Fire up rpciod if not yet running */
1451 	if (rpciod_up() != 0) {
1452 		printk(KERN_WARNING "NFS: couldn't start rpciod!\n");
1453 		kfree(server);
1454 		return ERR_PTR(-EIO);
1455 	}
1456 
1457 	error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1458 	if (error) {
1459 		up_write(&s->s_umount);
1460 		deactivate_super(s);
1461 		return ERR_PTR(error);
1462 	}
1463 	s->s_flags |= MS_ACTIVE;
1464 	return s;
1465 }
1466 
1467 static void nfs_kill_super(struct super_block *s)
1468 {
1469 	struct nfs_server *server = NFS_SB(s);
1470 
1471 	kill_anon_super(s);
1472 
1473 	if (server->client != NULL && !IS_ERR(server->client))
1474 		rpc_shutdown_client(server->client);
1475 	if (server->client_sys != NULL && !IS_ERR(server->client_sys))
1476 		rpc_shutdown_client(server->client_sys);
1477 
1478 	if (!(server->flags & NFS_MOUNT_NONLM))
1479 		lockd_down();	/* release rpc.lockd */
1480 
1481 	rpciod_down();		/* release rpciod */
1482 
1483 	if (server->hostname != NULL)
1484 		kfree(server->hostname);
1485 	kfree(server);
1486 }
1487 
1488 static struct file_system_type nfs_fs_type = {
1489 	.owner		= THIS_MODULE,
1490 	.name		= "nfs",
1491 	.get_sb		= nfs_get_sb,
1492 	.kill_sb	= nfs_kill_super,
1493 	.fs_flags	= FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1494 };
1495 
1496 #ifdef CONFIG_NFS_V4
1497 
1498 static void nfs4_clear_inode(struct inode *);
1499 
1500 
1501 static struct super_operations nfs4_sops = {
1502 	.alloc_inode	= nfs_alloc_inode,
1503 	.destroy_inode	= nfs_destroy_inode,
1504 	.write_inode	= nfs_write_inode,
1505 	.delete_inode	= nfs_delete_inode,
1506 	.statfs		= nfs_statfs,
1507 	.clear_inode	= nfs4_clear_inode,
1508 	.umount_begin	= nfs_umount_begin,
1509 	.show_options	= nfs_show_options,
1510 };
1511 
1512 /*
1513  * Clean out any remaining NFSv4 state that might be left over due
1514  * to open() calls that passed nfs_atomic_lookup, but failed to call
1515  * nfs_open().
1516  */
1517 static void nfs4_clear_inode(struct inode *inode)
1518 {
1519 	struct nfs_inode *nfsi = NFS_I(inode);
1520 
1521 	/* If we are holding a delegation, return it! */
1522 	if (nfsi->delegation != NULL)
1523 		nfs_inode_return_delegation(inode);
1524 	/* First call standard NFS clear_inode() code */
1525 	nfs_clear_inode(inode);
1526 	/* Now clear out any remaining state */
1527 	while (!list_empty(&nfsi->open_states)) {
1528 		struct nfs4_state *state;
1529 
1530 		state = list_entry(nfsi->open_states.next,
1531 				struct nfs4_state,
1532 				inode_states);
1533 		dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1534 				__FUNCTION__,
1535 				inode->i_sb->s_id,
1536 				(long long)NFS_FILEID(inode),
1537 				state);
1538 		BUG_ON(atomic_read(&state->count) != 1);
1539 		nfs4_close_state(state, state->state);
1540 	}
1541 }
1542 
1543 
1544 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1545 {
1546 	struct nfs_server *server;
1547 	struct nfs4_client *clp = NULL;
1548 	struct rpc_xprt *xprt = NULL;
1549 	struct rpc_clnt *clnt = NULL;
1550 	struct rpc_timeout timeparms;
1551 	rpc_authflavor_t authflavour;
1552 	int proto, err = -EIO;
1553 
1554 	sb->s_blocksize_bits = 0;
1555 	sb->s_blocksize = 0;
1556 	server = NFS_SB(sb);
1557 	if (data->rsize != 0)
1558 		server->rsize = nfs_block_size(data->rsize, NULL);
1559 	if (data->wsize != 0)
1560 		server->wsize = nfs_block_size(data->wsize, NULL);
1561 	server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1562 	server->caps = NFS_CAP_ATOMIC_OPEN;
1563 
1564 	server->acregmin = data->acregmin*HZ;
1565 	server->acregmax = data->acregmax*HZ;
1566 	server->acdirmin = data->acdirmin*HZ;
1567 	server->acdirmax = data->acdirmax*HZ;
1568 
1569 	server->rpc_ops = &nfs_v4_clientops;
1570 	/* Initialize timeout values */
1571 
1572 	timeparms.to_initval = data->timeo * HZ / 10;
1573 	timeparms.to_retries = data->retrans;
1574 	timeparms.to_exponential = 1;
1575 	if (!timeparms.to_retries)
1576 		timeparms.to_retries = 5;
1577 
1578 	proto = data->proto;
1579 	/* Which IP protocol do we use? */
1580 	switch (proto) {
1581 	case IPPROTO_TCP:
1582 		timeparms.to_maxval  = RPC_MAX_TCP_TIMEOUT;
1583 		if (!timeparms.to_initval)
1584 			timeparms.to_initval = 600 * HZ / 10;
1585 		break;
1586 	case IPPROTO_UDP:
1587 		timeparms.to_maxval  = RPC_MAX_UDP_TIMEOUT;
1588 		if (!timeparms.to_initval)
1589 			timeparms.to_initval = 11 * HZ / 10;
1590 		break;
1591 	default:
1592 		return -EINVAL;
1593 	}
1594 
1595 	clp = nfs4_get_client(&server->addr.sin_addr);
1596 	if (!clp) {
1597 		printk(KERN_WARNING "NFS: failed to create NFS4 client.\n");
1598 		return -EIO;
1599 	}
1600 
1601 	/* Now create transport and client */
1602 	authflavour = RPC_AUTH_UNIX;
1603 	if (data->auth_flavourlen != 0) {
1604 		if (data->auth_flavourlen > 1)
1605 			printk(KERN_INFO "NFS: cannot yet deal with multiple auth flavours.\n");
1606 		if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1607 			err = -EFAULT;
1608 			goto out_fail;
1609 		}
1610 	}
1611 
1612 	down_write(&clp->cl_sem);
1613 	if (clp->cl_rpcclient == NULL) {
1614 		xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1615 		if (IS_ERR(xprt)) {
1616 			up_write(&clp->cl_sem);
1617 			printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
1618 			err = PTR_ERR(xprt);
1619 			goto out_fail;
1620 		}
1621 		clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1622 				server->rpc_ops->version, authflavour);
1623 		if (IS_ERR(clnt)) {
1624 			up_write(&clp->cl_sem);
1625 			printk(KERN_WARNING "NFS: cannot create RPC client.\n");
1626 			xprt_destroy(xprt);
1627 			err = PTR_ERR(clnt);
1628 			goto out_fail;
1629 		}
1630 		clnt->cl_intr     = 1;
1631 		clnt->cl_softrtry = 1;
1632 		clnt->cl_chatty   = 1;
1633 		clp->cl_rpcclient = clnt;
1634 		clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1635 		if (IS_ERR(clp->cl_cred)) {
1636 			up_write(&clp->cl_sem);
1637 			err = PTR_ERR(clp->cl_cred);
1638 			clp->cl_cred = NULL;
1639 			goto out_fail;
1640 		}
1641 		memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1642 		nfs_idmap_new(clp);
1643 	}
1644 	if (list_empty(&clp->cl_superblocks)) {
1645 		err = nfs4_init_client(clp);
1646 		if (err != 0) {
1647 			up_write(&clp->cl_sem);
1648 			goto out_fail;
1649 		}
1650 	}
1651 	list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1652 	clnt = rpc_clone_client(clp->cl_rpcclient);
1653 	if (!IS_ERR(clnt))
1654 			server->nfs4_state = clp;
1655 	up_write(&clp->cl_sem);
1656 	clp = NULL;
1657 
1658 	if (IS_ERR(clnt)) {
1659 		printk(KERN_WARNING "NFS: cannot create RPC client.\n");
1660 		return PTR_ERR(clnt);
1661 	}
1662 
1663 	server->client    = clnt;
1664 
1665 	if (server->nfs4_state->cl_idmap == NULL) {
1666 		printk(KERN_WARNING "NFS: failed to create idmapper.\n");
1667 		return -ENOMEM;
1668 	}
1669 
1670 	if (clnt->cl_auth->au_flavor != authflavour) {
1671 		if (rpcauth_create(authflavour, clnt) == NULL) {
1672 			printk(KERN_WARNING "NFS: couldn't create credcache!\n");
1673 			return -ENOMEM;
1674 		}
1675 	}
1676 
1677 	sb->s_time_gran = 1;
1678 
1679 	sb->s_op = &nfs4_sops;
1680 	err = nfs_sb_init(sb, authflavour);
1681 	if (err == 0)
1682 		return 0;
1683 out_fail:
1684 	if (clp)
1685 		nfs4_put_client(clp);
1686 	return err;
1687 }
1688 
1689 static int nfs4_compare_super(struct super_block *sb, void *data)
1690 {
1691 	struct nfs_server *server = data;
1692 	struct nfs_server *old = NFS_SB(sb);
1693 
1694 	if (strcmp(server->hostname, old->hostname) != 0)
1695 		return 0;
1696 	if (strcmp(server->mnt_path, old->mnt_path) != 0)
1697 		return 0;
1698 	return 1;
1699 }
1700 
1701 static void *
1702 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1703 {
1704 	void *p = NULL;
1705 
1706 	if (!src->len)
1707 		return ERR_PTR(-EINVAL);
1708 	if (src->len < maxlen)
1709 		maxlen = src->len;
1710 	if (dst == NULL) {
1711 		p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1712 		if (p == NULL)
1713 			return ERR_PTR(-ENOMEM);
1714 	}
1715 	if (copy_from_user(dst, src->data, maxlen)) {
1716 		if (p != NULL)
1717 			kfree(p);
1718 		return ERR_PTR(-EFAULT);
1719 	}
1720 	dst[maxlen] = '\0';
1721 	return dst;
1722 }
1723 
1724 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1725 	int flags, const char *dev_name, void *raw_data)
1726 {
1727 	int error;
1728 	struct nfs_server *server;
1729 	struct super_block *s;
1730 	struct nfs4_mount_data *data = raw_data;
1731 	void *p;
1732 
1733 	if (!data) {
1734 		printk("nfs_read_super: missing data argument\n");
1735 		return ERR_PTR(-EINVAL);
1736 	}
1737 
1738 	server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1739 	if (!server)
1740 		return ERR_PTR(-ENOMEM);
1741 	memset(server, 0, sizeof(struct nfs_server));
1742 	/* Zero out the NFS state stuff */
1743 	init_nfsv4_state(server);
1744 
1745 	if (data->version != NFS4_MOUNT_VERSION) {
1746 		printk("nfs warning: mount version %s than kernel\n",
1747 			data->version < NFS4_MOUNT_VERSION ? "older" : "newer");
1748 	}
1749 
1750 	p = nfs_copy_user_string(NULL, &data->hostname, 256);
1751 	if (IS_ERR(p))
1752 		goto out_err;
1753 	server->hostname = p;
1754 
1755 	p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1756 	if (IS_ERR(p))
1757 		goto out_err;
1758 	server->mnt_path = p;
1759 
1760 	p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1761 			sizeof(server->ip_addr) - 1);
1762 	if (IS_ERR(p))
1763 		goto out_err;
1764 
1765 	/* We now require that the mount process passes the remote address */
1766 	if (data->host_addrlen != sizeof(server->addr)) {
1767 		s = ERR_PTR(-EINVAL);
1768 		goto out_free;
1769 	}
1770 	if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1771 		s = ERR_PTR(-EFAULT);
1772 		goto out_free;
1773 	}
1774 	if (server->addr.sin_family != AF_INET ||
1775 	    server->addr.sin_addr.s_addr == INADDR_ANY) {
1776 		printk("NFS: mount program didn't pass remote IP address!\n");
1777 		s = ERR_PTR(-EINVAL);
1778 		goto out_free;
1779 	}
1780 
1781 	s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1782 
1783 	if (IS_ERR(s) || s->s_root)
1784 		goto out_free;
1785 
1786 	s->s_flags = flags;
1787 
1788 	/* Fire up rpciod if not yet running */
1789 	if (rpciod_up() != 0) {
1790 		printk(KERN_WARNING "NFS: couldn't start rpciod!\n");
1791 		s = ERR_PTR(-EIO);
1792 		goto out_free;
1793 	}
1794 
1795 	error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1796 	if (error) {
1797 		up_write(&s->s_umount);
1798 		deactivate_super(s);
1799 		return ERR_PTR(error);
1800 	}
1801 	s->s_flags |= MS_ACTIVE;
1802 	return s;
1803 out_err:
1804 	s = (struct super_block *)p;
1805 out_free:
1806 	if (server->mnt_path)
1807 		kfree(server->mnt_path);
1808 	if (server->hostname)
1809 		kfree(server->hostname);
1810 	kfree(server);
1811 	return s;
1812 }
1813 
1814 static void nfs4_kill_super(struct super_block *sb)
1815 {
1816 	struct nfs_server *server = NFS_SB(sb);
1817 
1818 	nfs_return_all_delegations(sb);
1819 	kill_anon_super(sb);
1820 
1821 	nfs4_renewd_prepare_shutdown(server);
1822 
1823 	if (server->client != NULL && !IS_ERR(server->client))
1824 		rpc_shutdown_client(server->client);
1825 	rpciod_down();		/* release rpciod */
1826 
1827 	destroy_nfsv4_state(server);
1828 
1829 	if (server->hostname != NULL)
1830 		kfree(server->hostname);
1831 	kfree(server);
1832 }
1833 
1834 static struct file_system_type nfs4_fs_type = {
1835 	.owner		= THIS_MODULE,
1836 	.name		= "nfs4",
1837 	.get_sb		= nfs4_get_sb,
1838 	.kill_sb	= nfs4_kill_super,
1839 	.fs_flags	= FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1840 };
1841 
1842 #define nfs4_init_once(nfsi) \
1843 	do { \
1844 		INIT_LIST_HEAD(&(nfsi)->open_states); \
1845 		nfsi->delegation = NULL; \
1846 		nfsi->delegation_state = 0; \
1847 		init_rwsem(&nfsi->rwsem); \
1848 	} while(0)
1849 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1850 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1851 #else
1852 #define nfs4_init_once(nfsi) \
1853 	do { } while (0)
1854 #define register_nfs4fs() (0)
1855 #define unregister_nfs4fs()
1856 #endif
1857 
1858 extern int nfs_init_nfspagecache(void);
1859 extern void nfs_destroy_nfspagecache(void);
1860 extern int nfs_init_readpagecache(void);
1861 extern void nfs_destroy_readpagecache(void);
1862 extern int nfs_init_writepagecache(void);
1863 extern void nfs_destroy_writepagecache(void);
1864 #ifdef CONFIG_NFS_DIRECTIO
1865 extern int nfs_init_directcache(void);
1866 extern void nfs_destroy_directcache(void);
1867 #endif
1868 
1869 static kmem_cache_t * nfs_inode_cachep;
1870 
1871 static struct inode *nfs_alloc_inode(struct super_block *sb)
1872 {
1873 	struct nfs_inode *nfsi;
1874 	nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
1875 	if (!nfsi)
1876 		return NULL;
1877 	nfsi->flags = 0;
1878 	return &nfsi->vfs_inode;
1879 }
1880 
1881 static void nfs_destroy_inode(struct inode *inode)
1882 {
1883 	kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1884 }
1885 
1886 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
1887 {
1888 	struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1889 
1890 	if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1891 	    SLAB_CTOR_CONSTRUCTOR) {
1892 		inode_init_once(&nfsi->vfs_inode);
1893 		spin_lock_init(&nfsi->req_lock);
1894 		INIT_LIST_HEAD(&nfsi->dirty);
1895 		INIT_LIST_HEAD(&nfsi->commit);
1896 		INIT_LIST_HEAD(&nfsi->open_files);
1897 		INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1898 		atomic_set(&nfsi->data_updates, 0);
1899 		nfsi->ndirty = 0;
1900 		nfsi->ncommit = 0;
1901 		nfsi->npages = 0;
1902 		init_waitqueue_head(&nfsi->nfs_i_wait);
1903 		nfs4_init_once(nfsi);
1904 	}
1905 }
1906 
1907 int nfs_init_inodecache(void)
1908 {
1909 	nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1910 					     sizeof(struct nfs_inode),
1911 					     0, SLAB_RECLAIM_ACCOUNT,
1912 					     init_once, NULL);
1913 	if (nfs_inode_cachep == NULL)
1914 		return -ENOMEM;
1915 
1916 	return 0;
1917 }
1918 
1919 void nfs_destroy_inodecache(void)
1920 {
1921 	if (kmem_cache_destroy(nfs_inode_cachep))
1922 		printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
1923 }
1924 
1925 /*
1926  * Initialize NFS
1927  */
1928 static int __init init_nfs_fs(void)
1929 {
1930 	int err;
1931 
1932 	err = nfs_init_nfspagecache();
1933 	if (err)
1934 		goto out4;
1935 
1936 	err = nfs_init_inodecache();
1937 	if (err)
1938 		goto out3;
1939 
1940 	err = nfs_init_readpagecache();
1941 	if (err)
1942 		goto out2;
1943 
1944 	err = nfs_init_writepagecache();
1945 	if (err)
1946 		goto out1;
1947 
1948 #ifdef CONFIG_NFS_DIRECTIO
1949 	err = nfs_init_directcache();
1950 	if (err)
1951 		goto out0;
1952 #endif
1953 
1954 #ifdef CONFIG_PROC_FS
1955 	rpc_proc_register(&nfs_rpcstat);
1956 #endif
1957         err = register_filesystem(&nfs_fs_type);
1958 	if (err)
1959 		goto out;
1960 	if ((err = register_nfs4fs()) != 0)
1961 		goto out;
1962 	return 0;
1963 out:
1964 #ifdef CONFIG_PROC_FS
1965 	rpc_proc_unregister("nfs");
1966 #endif
1967 	nfs_destroy_writepagecache();
1968 #ifdef CONFIG_NFS_DIRECTIO
1969 out0:
1970 	nfs_destroy_directcache();
1971 #endif
1972 out1:
1973 	nfs_destroy_readpagecache();
1974 out2:
1975 	nfs_destroy_inodecache();
1976 out3:
1977 	nfs_destroy_nfspagecache();
1978 out4:
1979 	return err;
1980 }
1981 
1982 static void __exit exit_nfs_fs(void)
1983 {
1984 #ifdef CONFIG_NFS_DIRECTIO
1985 	nfs_destroy_directcache();
1986 #endif
1987 	nfs_destroy_writepagecache();
1988 	nfs_destroy_readpagecache();
1989 	nfs_destroy_inodecache();
1990 	nfs_destroy_nfspagecache();
1991 #ifdef CONFIG_PROC_FS
1992 	rpc_proc_unregister("nfs");
1993 #endif
1994 	unregister_filesystem(&nfs_fs_type);
1995 	unregister_nfs4fs();
1996 }
1997 
1998 /* Not quite true; I just maintain it */
1999 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2000 MODULE_LICENSE("GPL");
2001 
2002 module_init(init_nfs_fs)
2003 module_exit(exit_nfs_fs)
2004