Merge master.kernel.org:/home/rmk/linux-2.6-arm
[linux-2.6] / fs / nfs / inode.c
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         nfs_wb_all(inode);
150         /*
151          * The following should never happen...
152          */
153         if (nfs_have_writebacks(inode)) {
154                 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
155         }
156
157         clear_inode(inode);
158 }
159
160 static void
161 nfs_clear_inode(struct inode *inode)
162 {
163         struct nfs_inode *nfsi = NFS_I(inode);
164         struct rpc_cred *cred;
165
166         nfs_wb_all(inode);
167         BUG_ON (!list_empty(&nfsi->open_files));
168         nfs_zap_acl_cache(inode);
169         cred = nfsi->cache_access.cred;
170         if (cred)
171                 put_rpccred(cred);
172         BUG_ON(atomic_read(&nfsi->data_updates) != 0);
173 }
174
175 void
176 nfs_umount_begin(struct super_block *sb)
177 {
178         struct rpc_clnt *rpc = NFS_SB(sb)->client;
179
180         /* -EIO all pending I/O */
181         if (!IS_ERR(rpc))
182                 rpc_killall_tasks(rpc);
183         rpc = NFS_SB(sb)->client_acl;
184         if (!IS_ERR(rpc))
185                 rpc_killall_tasks(rpc);
186 }
187
188
189 static inline unsigned long
190 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
191 {
192         /* make sure blocksize is a power of two */
193         if ((bsize & (bsize - 1)) || nrbitsp) {
194                 unsigned char   nrbits;
195
196                 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
197                         ;
198                 bsize = 1 << nrbits;
199                 if (nrbitsp)
200                         *nrbitsp = nrbits;
201         }
202
203         return bsize;
204 }
205
206 /*
207  * Calculate the number of 512byte blocks used.
208  */
209 static inline unsigned long
210 nfs_calc_block_size(u64 tsize)
211 {
212         loff_t used = (tsize + 511) >> 9;
213         return (used > ULONG_MAX) ? ULONG_MAX : used;
214 }
215
216 /*
217  * Compute and set NFS server blocksize
218  */
219 static inline unsigned long
220 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
221 {
222         if (bsize < 1024)
223                 bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
224         else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
225                 bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
226
227         return nfs_block_bits(bsize, nrbitsp);
228 }
229
230 /*
231  * Obtain the root inode of the file system.
232  */
233 static struct inode *
234 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
235 {
236         struct nfs_server       *server = NFS_SB(sb);
237         struct inode *rooti;
238         int                     error;
239
240         error = server->rpc_ops->getroot(server, rootfh, fsinfo);
241         if (error < 0) {
242                 dprintk("nfs_get_root: getattr error = %d\n", -error);
243                 return ERR_PTR(error);
244         }
245
246         rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
247         if (!rooti)
248                 return ERR_PTR(-ENOMEM);
249         return rooti;
250 }
251
252 /*
253  * Do NFS version-independent mount processing, and sanity checking
254  */
255 static int
256 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
257 {
258         struct nfs_server       *server;
259         struct inode            *root_inode;
260         struct nfs_fattr        fattr;
261         struct nfs_fsinfo       fsinfo = {
262                                         .fattr = &fattr,
263                                 };
264         struct nfs_pathconf pathinfo = {
265                         .fattr = &fattr,
266         };
267         int no_root_error = 0;
268         unsigned long max_rpc_payload;
269
270         /* We probably want something more informative here */
271         snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
272
273         server = NFS_SB(sb);
274
275         sb->s_magic      = NFS_SUPER_MAGIC;
276
277         root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
278         /* Did getting the root inode fail? */
279         if (IS_ERR(root_inode)) {
280                 no_root_error = PTR_ERR(root_inode);
281                 goto out_no_root;
282         }
283         sb->s_root = d_alloc_root(root_inode);
284         if (!sb->s_root) {
285                 no_root_error = -ENOMEM;
286                 goto out_no_root;
287         }
288         sb->s_root->d_op = server->rpc_ops->dentry_ops;
289
290         /* Get some general file system info */
291         if (server->namelen == 0 &&
292             server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
293                 server->namelen = pathinfo.max_namelen;
294         /* Work out a lot of parameters */
295         if (server->rsize == 0)
296                 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
297         if (server->wsize == 0)
298                 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
299
300         if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
301                 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
302         if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
303                 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
304
305         max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
306         if (server->rsize > max_rpc_payload)
307                 server->rsize = max_rpc_payload;
308         if (server->wsize > max_rpc_payload)
309                 server->wsize = max_rpc_payload;
310
311         server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
312         if (server->rpages > NFS_READ_MAXIOV) {
313                 server->rpages = NFS_READ_MAXIOV;
314                 server->rsize = server->rpages << PAGE_CACHE_SHIFT;
315         }
316
317         server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
318         if (server->wpages > NFS_WRITE_MAXIOV) {
319                 server->wpages = NFS_WRITE_MAXIOV;
320                 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
321         }
322
323         if (sb->s_blocksize == 0)
324                 sb->s_blocksize = nfs_block_bits(server->wsize,
325                                                          &sb->s_blocksize_bits);
326         server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
327
328         server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
329         if (server->dtsize > PAGE_CACHE_SIZE)
330                 server->dtsize = PAGE_CACHE_SIZE;
331         if (server->dtsize > server->rsize)
332                 server->dtsize = server->rsize;
333
334         if (server->flags & NFS_MOUNT_NOAC) {
335                 server->acregmin = server->acregmax = 0;
336                 server->acdirmin = server->acdirmax = 0;
337                 sb->s_flags |= MS_SYNCHRONOUS;
338         }
339         server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
340
341         sb->s_maxbytes = fsinfo.maxfilesize;
342         if (sb->s_maxbytes > MAX_LFS_FILESIZE) 
343                 sb->s_maxbytes = MAX_LFS_FILESIZE; 
344
345         server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
346         server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
347
348         /* We're airborne Set socket buffersize */
349         rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
350         return 0;
351         /* Yargs. It didn't work out. */
352 out_no_root:
353         dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
354         if (!IS_ERR(root_inode))
355                 iput(root_inode);
356         return no_root_error;
357 }
358
359 /*
360  * Create an RPC client handle.
361  */
362 static struct rpc_clnt *
363 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
364 {
365         struct rpc_timeout      timeparms;
366         struct rpc_xprt         *xprt = NULL;
367         struct rpc_clnt         *clnt = NULL;
368         int                     tcp   = (data->flags & NFS_MOUNT_TCP);
369
370         /* Initialize timeout values */
371         timeparms.to_initval = data->timeo * HZ / 10;
372         timeparms.to_retries = data->retrans;
373         timeparms.to_maxval  = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
374         timeparms.to_exponential = 1;
375
376         if (!timeparms.to_initval)
377                 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
378         if (!timeparms.to_retries)
379                 timeparms.to_retries = 5;
380
381         /* create transport and client */
382         xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
383                                  &server->addr, &timeparms);
384         if (IS_ERR(xprt)) {
385                 dprintk("%s: cannot create RPC transport. Error = %ld\n",
386                                 __FUNCTION__, PTR_ERR(xprt));
387                 return (struct rpc_clnt *)xprt;
388         }
389         clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
390                                  server->rpc_ops->version, data->pseudoflavor);
391         if (IS_ERR(clnt)) {
392                 dprintk("%s: cannot create RPC client. Error = %ld\n",
393                                 __FUNCTION__, PTR_ERR(xprt));
394                 goto out_fail;
395         }
396
397         clnt->cl_intr     = 1;
398         clnt->cl_softrtry = 1;
399         clnt->cl_chatty   = 1;
400
401         return clnt;
402
403 out_fail:
404         return clnt;
405 }
406
407 /*
408  * The way this works is that the mount process passes a structure
409  * in the data argument which contains the server's IP address
410  * and the root file handle obtained from the server's mount
411  * daemon. We stash these away in the private superblock fields.
412  */
413 static int
414 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
415 {
416         struct nfs_server       *server;
417         rpc_authflavor_t        authflavor;
418
419         server           = NFS_SB(sb);
420         sb->s_blocksize_bits = 0;
421         sb->s_blocksize = 0;
422         if (data->bsize)
423                 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
424         if (data->rsize)
425                 server->rsize = nfs_block_size(data->rsize, NULL);
426         if (data->wsize)
427                 server->wsize = nfs_block_size(data->wsize, NULL);
428         server->flags    = data->flags & NFS_MOUNT_FLAGMASK;
429
430         server->acregmin = data->acregmin*HZ;
431         server->acregmax = data->acregmax*HZ;
432         server->acdirmin = data->acdirmin*HZ;
433         server->acdirmax = data->acdirmax*HZ;
434
435         /* Start lockd here, before we might error out */
436         if (!(server->flags & NFS_MOUNT_NONLM))
437                 lockd_up();
438
439         server->namelen  = data->namlen;
440         server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
441         if (!server->hostname)
442                 return -ENOMEM;
443         strcpy(server->hostname, data->hostname);
444
445         /* Check NFS protocol revision and initialize RPC op vector
446          * and file handle pool. */
447 #ifdef CONFIG_NFS_V3
448         if (server->flags & NFS_MOUNT_VER3) {
449                 server->rpc_ops = &nfs_v3_clientops;
450                 server->caps |= NFS_CAP_READDIRPLUS;
451         } else {
452                 server->rpc_ops = &nfs_v2_clientops;
453         }
454 #else
455         server->rpc_ops = &nfs_v2_clientops;
456 #endif
457
458         /* Fill in pseudoflavor for mount version < 5 */
459         if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
460                 data->pseudoflavor = RPC_AUTH_UNIX;
461         authflavor = data->pseudoflavor;        /* save for sb_init() */
462         /* XXX maybe we want to add a server->pseudoflavor field */
463
464         /* Create RPC client handles */
465         server->client = nfs_create_client(server, data);
466         if (IS_ERR(server->client))
467                 return PTR_ERR(server->client);
468         /* RFC 2623, sec 2.3.2 */
469         if (authflavor != RPC_AUTH_UNIX) {
470                 struct rpc_auth *auth;
471
472                 server->client_sys = rpc_clone_client(server->client);
473                 if (IS_ERR(server->client_sys))
474                         return PTR_ERR(server->client_sys);
475                 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
476                 if (IS_ERR(auth))
477                         return PTR_ERR(auth);
478         } else {
479                 atomic_inc(&server->client->cl_count);
480                 server->client_sys = server->client;
481         }
482         if (server->flags & NFS_MOUNT_VER3) {
483 #ifdef CONFIG_NFS_V3_ACL
484                 if (!(server->flags & NFS_MOUNT_NOACL)) {
485                         server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
486                         /* No errors! Assume that Sun nfsacls are supported */
487                         if (!IS_ERR(server->client_acl))
488                                 server->caps |= NFS_CAP_ACLS;
489                 }
490 #else
491                 server->flags &= ~NFS_MOUNT_NOACL;
492 #endif /* CONFIG_NFS_V3_ACL */
493                 /*
494                  * The VFS shouldn't apply the umask to mode bits. We will
495                  * do so ourselves when necessary.
496                  */
497                 sb->s_flags |= MS_POSIXACL;
498                 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
499                         server->namelen = NFS3_MAXNAMLEN;
500                 sb->s_time_gran = 1;
501         } else {
502                 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
503                         server->namelen = NFS2_MAXNAMLEN;
504         }
505
506         sb->s_op = &nfs_sops;
507         return nfs_sb_init(sb, authflavor);
508 }
509
510 static int
511 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
512 {
513         struct nfs_server *server = NFS_SB(sb);
514         unsigned char blockbits;
515         unsigned long blockres;
516         struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
517         struct nfs_fattr fattr;
518         struct nfs_fsstat res = {
519                         .fattr = &fattr,
520         };
521         int error;
522
523         lock_kernel();
524
525         error = server->rpc_ops->statfs(server, rootfh, &res);
526         buf->f_type = NFS_SUPER_MAGIC;
527         if (error < 0)
528                 goto out_err;
529
530         /*
531          * Current versions of glibc do not correctly handle the
532          * case where f_frsize != f_bsize.  Eventually we want to
533          * report the value of wtmult in this field.
534          */
535         buf->f_frsize = sb->s_blocksize;
536
537         /*
538          * On most *nix systems, f_blocks, f_bfree, and f_bavail
539          * are reported in units of f_frsize.  Linux hasn't had
540          * an f_frsize field in its statfs struct until recently,
541          * thus historically Linux's sys_statfs reports these
542          * fields in units of f_bsize.
543          */
544         buf->f_bsize = sb->s_blocksize;
545         blockbits = sb->s_blocksize_bits;
546         blockres = (1 << blockbits) - 1;
547         buf->f_blocks = (res.tbytes + blockres) >> blockbits;
548         buf->f_bfree = (res.fbytes + blockres) >> blockbits;
549         buf->f_bavail = (res.abytes + blockres) >> blockbits;
550
551         buf->f_files = res.tfiles;
552         buf->f_ffree = res.afiles;
553
554         buf->f_namelen = server->namelen;
555  out:
556         unlock_kernel();
557
558         return 0;
559
560  out_err:
561         printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
562         buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
563         goto out;
564
565 }
566
567 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
568 {
569         static struct proc_nfs_info {
570                 int flag;
571                 char *str;
572                 char *nostr;
573         } nfs_info[] = {
574                 { NFS_MOUNT_SOFT, ",soft", ",hard" },
575                 { NFS_MOUNT_INTR, ",intr", "" },
576                 { NFS_MOUNT_POSIX, ",posix", "" },
577                 { NFS_MOUNT_TCP, ",tcp", ",udp" },
578                 { NFS_MOUNT_NOCTO, ",nocto", "" },
579                 { NFS_MOUNT_NOAC, ",noac", "" },
580                 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
581                 { NFS_MOUNT_NOACL, ",noacl", "" },
582                 { 0, NULL, NULL }
583         };
584         struct proc_nfs_info *nfs_infop;
585         struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
586
587         seq_printf(m, ",v%d", nfss->rpc_ops->version);
588         seq_printf(m, ",rsize=%d", nfss->rsize);
589         seq_printf(m, ",wsize=%d", nfss->wsize);
590         if (nfss->acregmin != 3*HZ)
591                 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
592         if (nfss->acregmax != 60*HZ)
593                 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
594         if (nfss->acdirmin != 30*HZ)
595                 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
596         if (nfss->acdirmax != 60*HZ)
597                 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
598         for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
599                 if (nfss->flags & nfs_infop->flag)
600                         seq_puts(m, nfs_infop->str);
601                 else
602                         seq_puts(m, nfs_infop->nostr);
603         }
604         seq_puts(m, ",addr=");
605         seq_escape(m, nfss->hostname, " \t\n\\");
606         return 0;
607 }
608
609 /*
610  * Invalidate the local caches
611  */
612 void
613 nfs_zap_caches(struct inode *inode)
614 {
615         struct nfs_inode *nfsi = NFS_I(inode);
616         int mode = inode->i_mode;
617
618         NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
619         NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
620
621         memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
622         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
623                 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
624         else
625                 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
626 }
627
628 static void nfs_zap_acl_cache(struct inode *inode)
629 {
630         void (*clear_acl_cache)(struct inode *);
631
632         clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
633         if (clear_acl_cache != NULL)
634                 clear_acl_cache(inode);
635         NFS_I(inode)->flags &= ~NFS_INO_INVALID_ACL;
636 }
637
638 /*
639  * Invalidate, but do not unhash, the inode
640  */
641 static void
642 nfs_invalidate_inode(struct inode *inode)
643 {
644         umode_t save_mode = inode->i_mode;
645
646         make_bad_inode(inode);
647         inode->i_mode = save_mode;
648         nfs_zap_caches(inode);
649 }
650
651 struct nfs_find_desc {
652         struct nfs_fh           *fh;
653         struct nfs_fattr        *fattr;
654 };
655
656 /*
657  * In NFSv3 we can have 64bit inode numbers. In order to support
658  * this, and re-exported directories (also seen in NFSv2)
659  * we are forced to allow 2 different inodes to have the same
660  * i_ino.
661  */
662 static int
663 nfs_find_actor(struct inode *inode, void *opaque)
664 {
665         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
666         struct nfs_fh           *fh = desc->fh;
667         struct nfs_fattr        *fattr = desc->fattr;
668
669         if (NFS_FILEID(inode) != fattr->fileid)
670                 return 0;
671         if (nfs_compare_fh(NFS_FH(inode), fh))
672                 return 0;
673         if (is_bad_inode(inode) || NFS_STALE(inode))
674                 return 0;
675         return 1;
676 }
677
678 static int
679 nfs_init_locked(struct inode *inode, void *opaque)
680 {
681         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
682         struct nfs_fattr        *fattr = desc->fattr;
683
684         NFS_FILEID(inode) = fattr->fileid;
685         nfs_copy_fh(NFS_FH(inode), desc->fh);
686         return 0;
687 }
688
689 /* Don't use READDIRPLUS on directories that we believe are too large */
690 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
691
692 /*
693  * This is our front-end to iget that looks up inodes by file handle
694  * instead of inode number.
695  */
696 struct inode *
697 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
698 {
699         struct nfs_find_desc desc = {
700                 .fh     = fh,
701                 .fattr  = fattr
702         };
703         struct inode *inode = NULL;
704         unsigned long hash;
705
706         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
707                 goto out_no_inode;
708
709         if (!fattr->nlink) {
710                 printk("NFS: Buggy server - nlink == 0!\n");
711                 goto out_no_inode;
712         }
713
714         hash = nfs_fattr_to_ino_t(fattr);
715
716         if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
717                 goto out_no_inode;
718
719         if (inode->i_state & I_NEW) {
720                 struct nfs_inode *nfsi = NFS_I(inode);
721
722                 /* We set i_ino for the few things that still rely on it,
723                  * such as stat(2) */
724                 inode->i_ino = hash;
725
726                 /* We can't support update_atime(), since the server will reset it */
727                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
728                 inode->i_mode = fattr->mode;
729                 /* Why so? Because we want revalidate for devices/FIFOs, and
730                  * that's precisely what we have in nfs_file_inode_operations.
731                  */
732                 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
733                 if (S_ISREG(inode->i_mode)) {
734                         inode->i_fop = &nfs_file_operations;
735                         inode->i_data.a_ops = &nfs_file_aops;
736                         inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
737                 } else if (S_ISDIR(inode->i_mode)) {
738                         inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
739                         inode->i_fop = &nfs_dir_operations;
740                         if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
741                             && fattr->size <= NFS_LIMIT_READDIRPLUS)
742                                 NFS_FLAGS(inode) |= NFS_INO_ADVISE_RDPLUS;
743                 } else if (S_ISLNK(inode->i_mode))
744                         inode->i_op = &nfs_symlink_inode_operations;
745                 else
746                         init_special_inode(inode, inode->i_mode, fattr->rdev);
747
748                 nfsi->read_cache_jiffies = fattr->timestamp;
749                 inode->i_atime = fattr->atime;
750                 inode->i_mtime = fattr->mtime;
751                 inode->i_ctime = fattr->ctime;
752                 if (fattr->valid & NFS_ATTR_FATTR_V4)
753                         nfsi->change_attr = fattr->change_attr;
754                 inode->i_size = nfs_size_to_loff_t(fattr->size);
755                 inode->i_nlink = fattr->nlink;
756                 inode->i_uid = fattr->uid;
757                 inode->i_gid = fattr->gid;
758                 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
759                         /*
760                          * report the blocks in 512byte units
761                          */
762                         inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
763                         inode->i_blksize = inode->i_sb->s_blocksize;
764                 } else {
765                         inode->i_blocks = fattr->du.nfs2.blocks;
766                         inode->i_blksize = fattr->du.nfs2.blocksize;
767                 }
768                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
769                 nfsi->attrtimeo_timestamp = jiffies;
770                 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
771                 nfsi->cache_access.cred = NULL;
772
773                 unlock_new_inode(inode);
774         } else
775                 nfs_refresh_inode(inode, fattr);
776         dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
777                 inode->i_sb->s_id,
778                 (long long)NFS_FILEID(inode),
779                 atomic_read(&inode->i_count));
780
781 out:
782         return inode;
783
784 out_no_inode:
785         printk("nfs_fhget: iget failed\n");
786         goto out;
787 }
788
789 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
790
791 int
792 nfs_setattr(struct dentry *dentry, struct iattr *attr)
793 {
794         struct inode *inode = dentry->d_inode;
795         struct nfs_fattr fattr;
796         int error;
797
798         if (attr->ia_valid & ATTR_SIZE) {
799                 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
800                         attr->ia_valid &= ~ATTR_SIZE;
801         }
802
803         /* Optimization: if the end result is no change, don't RPC */
804         attr->ia_valid &= NFS_VALID_ATTRS;
805         if (attr->ia_valid == 0)
806                 return 0;
807
808         lock_kernel();
809         nfs_begin_data_update(inode);
810         /* Write all dirty data if we're changing file permissions or size */
811         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
812                 if (filemap_fdatawrite(inode->i_mapping) == 0)
813                         filemap_fdatawait(inode->i_mapping);
814                 nfs_wb_all(inode);
815         }
816         error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
817         if (error == 0) {
818                 nfs_refresh_inode(inode, &fattr);
819                 if ((attr->ia_valid & ATTR_MODE) != 0) {
820                         int mode;
821                         mode = inode->i_mode & ~S_IALLUGO;
822                         mode |= attr->ia_mode & S_IALLUGO;
823                         inode->i_mode = mode;
824                 }
825                 if ((attr->ia_valid & ATTR_UID) != 0)
826                         inode->i_uid = attr->ia_uid;
827                 if ((attr->ia_valid & ATTR_GID) != 0)
828                         inode->i_gid = attr->ia_gid;
829                 if ((attr->ia_valid & ATTR_SIZE) != 0) {
830                         inode->i_size = attr->ia_size;
831                         vmtruncate(inode, attr->ia_size);
832                 }
833         }
834         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
835                 NFS_FLAGS(inode) |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
836         nfs_end_data_update(inode);
837         unlock_kernel();
838         return error;
839 }
840
841 /*
842  * Wait for the inode to get unlocked.
843  * (Used for NFS_INO_LOCKED and NFS_INO_REVALIDATING).
844  */
845 static int
846 nfs_wait_on_inode(struct inode *inode, int flag)
847 {
848         struct rpc_clnt *clnt = NFS_CLIENT(inode);
849         struct nfs_inode *nfsi = NFS_I(inode);
850
851         int error;
852         if (!(NFS_FLAGS(inode) & flag))
853                 return 0;
854         atomic_inc(&inode->i_count);
855         error = nfs_wait_event(clnt, nfsi->nfs_i_wait,
856                                 !(NFS_FLAGS(inode) & flag));
857         iput(inode);
858         return error;
859 }
860
861 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
862 {
863         struct inode *inode = dentry->d_inode;
864         struct nfs_inode *nfsi = NFS_I(inode);
865         int need_atime = nfsi->flags & NFS_INO_INVALID_ATIME;
866         int err;
867
868         if (__IS_FLG(inode, MS_NOATIME))
869                 need_atime = 0;
870         else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
871                 need_atime = 0;
872         /* We may force a getattr if the user cares about atime */
873         if (need_atime)
874                 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
875         else
876                 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
877         if (!err)
878                 generic_fillattr(inode, stat);
879         return err;
880 }
881
882 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
883 {
884         struct nfs_open_context *ctx;
885
886         ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
887         if (ctx != NULL) {
888                 atomic_set(&ctx->count, 1);
889                 ctx->dentry = dget(dentry);
890                 ctx->cred = get_rpccred(cred);
891                 ctx->state = NULL;
892                 ctx->lockowner = current->files;
893                 ctx->error = 0;
894                 ctx->dir_cookie = 0;
895         }
896         return ctx;
897 }
898
899 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
900 {
901         if (ctx != NULL)
902                 atomic_inc(&ctx->count);
903         return ctx;
904 }
905
906 void put_nfs_open_context(struct nfs_open_context *ctx)
907 {
908         if (atomic_dec_and_test(&ctx->count)) {
909                 if (!list_empty(&ctx->list)) {
910                         struct inode *inode = ctx->dentry->d_inode;
911                         spin_lock(&inode->i_lock);
912                         list_del(&ctx->list);
913                         spin_unlock(&inode->i_lock);
914                 }
915                 if (ctx->state != NULL)
916                         nfs4_close_state(ctx->state, ctx->mode);
917                 if (ctx->cred != NULL)
918                         put_rpccred(ctx->cred);
919                 dput(ctx->dentry);
920                 kfree(ctx);
921         }
922 }
923
924 /*
925  * Ensure that mmap has a recent RPC credential for use when writing out
926  * shared pages
927  */
928 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
929 {
930         struct inode *inode = filp->f_dentry->d_inode;
931         struct nfs_inode *nfsi = NFS_I(inode);
932
933         filp->private_data = get_nfs_open_context(ctx);
934         spin_lock(&inode->i_lock);
935         list_add(&ctx->list, &nfsi->open_files);
936         spin_unlock(&inode->i_lock);
937 }
938
939 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
940 {
941         struct nfs_inode *nfsi = NFS_I(inode);
942         struct nfs_open_context *pos, *ctx = NULL;
943
944         spin_lock(&inode->i_lock);
945         list_for_each_entry(pos, &nfsi->open_files, list) {
946                 if ((pos->mode & mode) == mode) {
947                         ctx = get_nfs_open_context(pos);
948                         break;
949                 }
950         }
951         spin_unlock(&inode->i_lock);
952         return ctx;
953 }
954
955 void nfs_file_clear_open_context(struct file *filp)
956 {
957         struct inode *inode = filp->f_dentry->d_inode;
958         struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
959
960         if (ctx) {
961                 filp->private_data = NULL;
962                 spin_lock(&inode->i_lock);
963                 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
964                 spin_unlock(&inode->i_lock);
965                 put_nfs_open_context(ctx);
966         }
967 }
968
969 /*
970  * These allocate and release file read/write context information.
971  */
972 int nfs_open(struct inode *inode, struct file *filp)
973 {
974         struct nfs_open_context *ctx;
975         struct rpc_cred *cred;
976
977         cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
978         if (IS_ERR(cred))
979                 return PTR_ERR(cred);
980         ctx = alloc_nfs_open_context(filp->f_dentry, cred);
981         put_rpccred(cred);
982         if (ctx == NULL)
983                 return -ENOMEM;
984         ctx->mode = filp->f_mode;
985         nfs_file_set_open_context(filp, ctx);
986         put_nfs_open_context(ctx);
987         if ((filp->f_mode & FMODE_WRITE) != 0)
988                 nfs_begin_data_update(inode);
989         return 0;
990 }
991
992 int nfs_release(struct inode *inode, struct file *filp)
993 {
994         if ((filp->f_mode & FMODE_WRITE) != 0)
995                 nfs_end_data_update(inode);
996         nfs_file_clear_open_context(filp);
997         return 0;
998 }
999
1000 /*
1001  * This function is called whenever some part of NFS notices that
1002  * the cached attributes have to be refreshed.
1003  */
1004 int
1005 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1006 {
1007         int              status = -ESTALE;
1008         struct nfs_fattr fattr;
1009         struct nfs_inode *nfsi = NFS_I(inode);
1010         unsigned long verifier;
1011         unsigned int flags;
1012
1013         dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1014                 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1015
1016         lock_kernel();
1017         if (!inode || is_bad_inode(inode))
1018                 goto out_nowait;
1019         if (NFS_STALE(inode))
1020                 goto out_nowait;
1021
1022         while (NFS_REVALIDATING(inode)) {
1023                 status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
1024                 if (status < 0)
1025                         goto out_nowait;
1026                 if (NFS_ATTRTIMEO(inode) == 0)
1027                         continue;
1028                 if (NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME))
1029                         continue;
1030                 status = NFS_STALE(inode) ? -ESTALE : 0;
1031                 goto out_nowait;
1032         }
1033         NFS_FLAGS(inode) |= NFS_INO_REVALIDATING;
1034
1035         /* Protect against RPC races by saving the change attribute */
1036         verifier = nfs_save_change_attribute(inode);
1037         status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1038         if (status != 0) {
1039                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1040                          inode->i_sb->s_id,
1041                          (long long)NFS_FILEID(inode), status);
1042                 if (status == -ESTALE) {
1043                         nfs_zap_caches(inode);
1044                         if (!S_ISDIR(inode->i_mode))
1045                                 NFS_FLAGS(inode) |= NFS_INO_STALE;
1046                 }
1047                 goto out;
1048         }
1049
1050         status = nfs_update_inode(inode, &fattr, verifier);
1051         if (status) {
1052                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1053                          inode->i_sb->s_id,
1054                          (long long)NFS_FILEID(inode), status);
1055                 goto out;
1056         }
1057         flags = nfsi->flags;
1058         nfsi->flags &= ~NFS_INO_REVAL_PAGECACHE;
1059         /*
1060          * We may need to keep the attributes marked as invalid if
1061          * we raced with nfs_end_attr_update().
1062          */
1063         if (verifier == nfsi->cache_change_attribute)
1064                 nfsi->flags &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1065         /* Do the page cache invalidation */
1066         nfs_revalidate_mapping(inode, inode->i_mapping);
1067         if (flags & NFS_INO_INVALID_ACL)
1068                 nfs_zap_acl_cache(inode);
1069         dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1070                 inode->i_sb->s_id,
1071                 (long long)NFS_FILEID(inode));
1072
1073 out:
1074         NFS_FLAGS(inode) &= ~NFS_INO_REVALIDATING;
1075         wake_up(&nfsi->nfs_i_wait);
1076  out_nowait:
1077         unlock_kernel();
1078         return status;
1079 }
1080
1081 int nfs_attribute_timeout(struct inode *inode)
1082 {
1083         struct nfs_inode *nfsi = NFS_I(inode);
1084
1085         if (nfs_have_delegation(inode, FMODE_READ))
1086                 return 0;
1087         return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1088 }
1089
1090 /**
1091  * nfs_revalidate_inode - Revalidate the inode attributes
1092  * @server - pointer to nfs_server struct
1093  * @inode - pointer to inode struct
1094  *
1095  * Updates inode attribute information by retrieving the data from the server.
1096  */
1097 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1098 {
1099         if (!(NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1100                         && !nfs_attribute_timeout(inode))
1101                 return NFS_STALE(inode) ? -ESTALE : 0;
1102         return __nfs_revalidate_inode(server, inode);
1103 }
1104
1105 /**
1106  * nfs_revalidate_mapping - Revalidate the pagecache
1107  * @inode - pointer to host inode
1108  * @mapping - pointer to mapping
1109  */
1110 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1111 {
1112         struct nfs_inode *nfsi = NFS_I(inode);
1113
1114         if (nfsi->flags & NFS_INO_INVALID_DATA) {
1115                 if (S_ISREG(inode->i_mode)) {
1116                         if (filemap_fdatawrite(mapping) == 0)
1117                                 filemap_fdatawait(mapping);
1118                         nfs_wb_all(inode);
1119                 }
1120                 invalidate_inode_pages2(mapping);
1121                 nfsi->flags &= ~NFS_INO_INVALID_DATA;
1122                 if (S_ISDIR(inode->i_mode)) {
1123                         memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1124                         /* This ensures we revalidate child dentries */
1125                         nfsi->cache_change_attribute++;
1126                 }
1127                 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1128                                 inode->i_sb->s_id,
1129                                 (long long)NFS_FILEID(inode));
1130         }
1131 }
1132
1133 /**
1134  * nfs_begin_data_update
1135  * @inode - pointer to inode
1136  * Declare that a set of operations will update file data on the server
1137  */
1138 void nfs_begin_data_update(struct inode *inode)
1139 {
1140         atomic_inc(&NFS_I(inode)->data_updates);
1141 }
1142
1143 /**
1144  * nfs_end_data_update
1145  * @inode - pointer to inode
1146  * Declare end of the operations that will update file data
1147  * This will mark the inode as immediately needing revalidation
1148  * of its attribute cache.
1149  */
1150 void nfs_end_data_update(struct inode *inode)
1151 {
1152         struct nfs_inode *nfsi = NFS_I(inode);
1153
1154         if (!nfs_have_delegation(inode, FMODE_READ)) {
1155                 /* Mark the attribute cache for revalidation */
1156                 nfsi->flags |= NFS_INO_INVALID_ATTR;
1157                 /* Directories and symlinks: invalidate page cache too */
1158                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1159                         nfsi->flags |= NFS_INO_INVALID_DATA;
1160         }
1161         nfsi->cache_change_attribute ++;
1162         atomic_dec(&nfsi->data_updates);
1163 }
1164
1165 /**
1166  * nfs_refresh_inode - verify consistency of the inode attribute cache
1167  * @inode - pointer to inode
1168  * @fattr - updated attributes
1169  *
1170  * Verifies the attribute cache. If we have just changed the attributes,
1171  * so that fattr carries weak cache consistency data, then it may
1172  * also update the ctime/mtime/change_attribute.
1173  */
1174 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1175 {
1176         struct nfs_inode *nfsi = NFS_I(inode);
1177         loff_t cur_size, new_isize;
1178         int data_unstable;
1179
1180         /* Do we hold a delegation? */
1181         if (nfs_have_delegation(inode, FMODE_READ))
1182                 return 0;
1183
1184         /* Are we in the process of updating data on the server? */
1185         data_unstable = nfs_caches_unstable(inode);
1186
1187         if (fattr->valid & NFS_ATTR_FATTR_V4) {
1188                 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1189                                 && nfsi->change_attr == fattr->pre_change_attr)
1190                         nfsi->change_attr = fattr->change_attr;
1191                 if (nfsi->change_attr != fattr->change_attr) {
1192                         nfsi->flags |= NFS_INO_INVALID_ATTR;
1193                         if (!data_unstable)
1194                                 nfsi->flags |= NFS_INO_REVAL_PAGECACHE;
1195                 }
1196         }
1197
1198         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1199                 return 0;
1200
1201         /* Has the inode gone and changed behind our back? */
1202         if (nfsi->fileid != fattr->fileid
1203                         || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1204                 return -EIO;
1205
1206         cur_size = i_size_read(inode);
1207         new_isize = nfs_size_to_loff_t(fattr->size);
1208
1209         /* If we have atomic WCC data, we may update some attributes */
1210         if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1211                 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1212                         memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1213                 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1214                         memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1215         }
1216
1217         /* Verify a few of the more important attributes */
1218         if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1219                 nfsi->flags |= NFS_INO_INVALID_ATTR;
1220                 if (!data_unstable)
1221                         nfsi->flags |= NFS_INO_REVAL_PAGECACHE;
1222         }
1223         if (cur_size != new_isize) {
1224                 nfsi->flags |= NFS_INO_INVALID_ATTR;
1225                 if (nfsi->npages == 0)
1226                         nfsi->flags |= NFS_INO_REVAL_PAGECACHE;
1227         }
1228
1229         /* Have any file permissions changed? */
1230         if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1231                         || inode->i_uid != fattr->uid
1232                         || inode->i_gid != fattr->gid)
1233                 nfsi->flags |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1234
1235         /* Has the link count changed? */
1236         if (inode->i_nlink != fattr->nlink)
1237                 nfsi->flags |= NFS_INO_INVALID_ATTR;
1238
1239         if (!timespec_equal(&inode->i_atime, &fattr->atime))
1240                 nfsi->flags |= NFS_INO_INVALID_ATIME;
1241
1242         nfsi->read_cache_jiffies = fattr->timestamp;
1243         return 0;
1244 }
1245
1246 /*
1247  * Many nfs protocol calls return the new file attributes after
1248  * an operation.  Here we update the inode to reflect the state
1249  * of the server's inode.
1250  *
1251  * This is a bit tricky because we have to make sure all dirty pages
1252  * have been sent off to the server before calling invalidate_inode_pages.
1253  * To make sure no other process adds more write requests while we try
1254  * our best to flush them, we make them sleep during the attribute refresh.
1255  *
1256  * A very similar scenario holds for the dir cache.
1257  */
1258 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1259 {
1260         struct nfs_inode *nfsi = NFS_I(inode);
1261         loff_t cur_isize, new_isize;
1262         unsigned int    invalid = 0;
1263         int data_unstable;
1264
1265         dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1266                         __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1267                         atomic_read(&inode->i_count), fattr->valid);
1268
1269         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1270                 return 0;
1271
1272         if (nfsi->fileid != fattr->fileid) {
1273                 printk(KERN_ERR "%s: inode number mismatch\n"
1274                        "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1275                        __FUNCTION__,
1276                        inode->i_sb->s_id, (long long)nfsi->fileid,
1277                        inode->i_sb->s_id, (long long)fattr->fileid);
1278                 goto out_err;
1279         }
1280
1281         /*
1282          * Make sure the inode's type hasn't changed.
1283          */
1284         if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1285                 goto out_changed;
1286
1287         /*
1288          * Update the read time so we don't revalidate too often.
1289          */
1290         nfsi->read_cache_jiffies = fattr->timestamp;
1291
1292         /* Are we racing with known updates of the metadata on the server? */
1293         data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1294
1295         /* Check if our cached file size is stale */
1296         new_isize = nfs_size_to_loff_t(fattr->size);
1297         cur_isize = i_size_read(inode);
1298         if (new_isize != cur_isize) {
1299                 /* Do we perhaps have any outstanding writes? */
1300                 if (nfsi->npages == 0) {
1301                         /* No, but did we race with nfs_end_data_update()? */
1302                         if (verifier  ==  nfsi->cache_change_attribute) {
1303                                 inode->i_size = new_isize;
1304                                 invalid |= NFS_INO_INVALID_DATA;
1305                         }
1306                         invalid |= NFS_INO_INVALID_ATTR;
1307                 } else if (new_isize > cur_isize) {
1308                         inode->i_size = new_isize;
1309                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1310                 }
1311                 dprintk("NFS: isize change on server for file %s/%ld\n",
1312                                 inode->i_sb->s_id, inode->i_ino);
1313         }
1314
1315         /* Check if the mtime agrees */
1316         if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1317                 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1318                 dprintk("NFS: mtime change on server for file %s/%ld\n",
1319                                 inode->i_sb->s_id, inode->i_ino);
1320                 if (!data_unstable)
1321                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1322         }
1323
1324         if ((fattr->valid & NFS_ATTR_FATTR_V4)
1325             && nfsi->change_attr != fattr->change_attr) {
1326                 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1327                        inode->i_sb->s_id, inode->i_ino);
1328                 nfsi->change_attr = fattr->change_attr;
1329                 if (!data_unstable)
1330                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1331         }
1332
1333         /* If ctime has changed we should definitely clear access+acl caches */
1334         if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1335                 if (!data_unstable)
1336                         invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1337                 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1338         }
1339         memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1340
1341         if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1342             inode->i_uid != fattr->uid ||
1343             inode->i_gid != fattr->gid)
1344                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1345
1346         inode->i_mode = fattr->mode;
1347         inode->i_nlink = fattr->nlink;
1348         inode->i_uid = fattr->uid;
1349         inode->i_gid = fattr->gid;
1350
1351         if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1352                 /*
1353                  * report the blocks in 512byte units
1354                  */
1355                 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1356                 inode->i_blksize = inode->i_sb->s_blocksize;
1357         } else {
1358                 inode->i_blocks = fattr->du.nfs2.blocks;
1359                 inode->i_blksize = fattr->du.nfs2.blocksize;
1360         }
1361
1362         /* Update attrtimeo value if we're out of the unstable period */
1363         if (invalid & NFS_INO_INVALID_ATTR) {
1364                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1365                 nfsi->attrtimeo_timestamp = jiffies;
1366         } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1367                 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1368                         nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1369                 nfsi->attrtimeo_timestamp = jiffies;
1370         }
1371         /* Don't invalidate the data if we were to blame */
1372         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1373                                 || S_ISLNK(inode->i_mode)))
1374                 invalid &= ~NFS_INO_INVALID_DATA;
1375         if (!nfs_have_delegation(inode, FMODE_READ))
1376                 nfsi->flags |= invalid;
1377
1378         return 0;
1379  out_changed:
1380         /*
1381          * Big trouble! The inode has become a different object.
1382          */
1383 #ifdef NFS_PARANOIA
1384         printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1385                         __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1386 #endif
1387         /*
1388          * No need to worry about unhashing the dentry, as the
1389          * lookup validation will know that the inode is bad.
1390          * (But we fall through to invalidate the caches.)
1391          */
1392         nfs_invalidate_inode(inode);
1393  out_err:
1394         NFS_FLAGS(inode) |= NFS_INO_STALE;
1395         return -ESTALE;
1396 }
1397
1398 /*
1399  * File system information
1400  */
1401
1402 static int nfs_set_super(struct super_block *s, void *data)
1403 {
1404         s->s_fs_info = data;
1405         return set_anon_super(s, data);
1406 }
1407  
1408 static int nfs_compare_super(struct super_block *sb, void *data)
1409 {
1410         struct nfs_server *server = data;
1411         struct nfs_server *old = NFS_SB(sb);
1412
1413         if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1414                 return 0;
1415         if (old->addr.sin_port != server->addr.sin_port)
1416                 return 0;
1417         return !nfs_compare_fh(&old->fh, &server->fh);
1418 }
1419
1420 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1421         int flags, const char *dev_name, void *raw_data)
1422 {
1423         int error;
1424         struct nfs_server *server = NULL;
1425         struct super_block *s;
1426         struct nfs_fh *root;
1427         struct nfs_mount_data *data = raw_data;
1428
1429         s = ERR_PTR(-EINVAL);
1430         if (data == NULL) {
1431                 dprintk("%s: missing data argument\n", __FUNCTION__);
1432                 goto out_err;
1433         }
1434         if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1435                 dprintk("%s: bad mount version\n", __FUNCTION__);
1436                 goto out_err;
1437         }
1438         switch (data->version) {
1439                 case 1:
1440                         data->namlen = 0;
1441                 case 2:
1442                         data->bsize  = 0;
1443                 case 3:
1444                         if (data->flags & NFS_MOUNT_VER3) {
1445                                 dprintk("%s: mount structure version %d does not support NFSv3\n",
1446                                                 __FUNCTION__,
1447                                                 data->version);
1448                                 goto out_err;
1449                         }
1450                         data->root.size = NFS2_FHSIZE;
1451                         memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1452                 case 4:
1453                         if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1454                                 dprintk("%s: mount structure version %d does not support strong security\n",
1455                                                 __FUNCTION__,
1456                                                 data->version);
1457                                 goto out_err;
1458                         }
1459                 case 5:
1460                         memset(data->context, 0, sizeof(data->context));
1461         }
1462 #ifndef CONFIG_NFS_V3
1463         /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1464         s = ERR_PTR(-EPROTONOSUPPORT);
1465         if (data->flags & NFS_MOUNT_VER3) {
1466                 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1467                 goto out_err;
1468         }
1469 #endif /* CONFIG_NFS_V3 */
1470
1471         s = ERR_PTR(-ENOMEM);
1472         server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1473         if (!server)
1474                 goto out_err;
1475         memset(server, 0, sizeof(struct nfs_server));
1476         /* Zero out the NFS state stuff */
1477         init_nfsv4_state(server);
1478         server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1479
1480         root = &server->fh;
1481         if (data->flags & NFS_MOUNT_VER3)
1482                 root->size = data->root.size;
1483         else
1484                 root->size = NFS2_FHSIZE;
1485         s = ERR_PTR(-EINVAL);
1486         if (root->size > sizeof(root->data)) {
1487                 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1488                 goto out_err;
1489         }
1490         memcpy(root->data, data->root.data, root->size);
1491
1492         /* We now require that the mount process passes the remote address */
1493         memcpy(&server->addr, &data->addr, sizeof(server->addr));
1494         if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1495                 dprintk("%s: mount program didn't pass remote address!\n",
1496                                 __FUNCTION__);
1497                 goto out_err;
1498         }
1499
1500         /* Fire up rpciod if not yet running */
1501         s = ERR_PTR(rpciod_up());
1502         if (IS_ERR(s)) {
1503                 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1504                                 __FUNCTION__, PTR_ERR(s));
1505                 goto out_err;
1506         }
1507
1508         s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1509         if (IS_ERR(s) || s->s_root)
1510                 goto out_rpciod_down;
1511
1512         s->s_flags = flags;
1513
1514         error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1515         if (error) {
1516                 up_write(&s->s_umount);
1517                 deactivate_super(s);
1518                 return ERR_PTR(error);
1519         }
1520         s->s_flags |= MS_ACTIVE;
1521         return s;
1522 out_rpciod_down:
1523         rpciod_down();
1524 out_err:
1525         kfree(server);
1526         return s;
1527 }
1528
1529 static void nfs_kill_super(struct super_block *s)
1530 {
1531         struct nfs_server *server = NFS_SB(s);
1532
1533         kill_anon_super(s);
1534
1535         if (!IS_ERR(server->client))
1536                 rpc_shutdown_client(server->client);
1537         if (!IS_ERR(server->client_sys))
1538                 rpc_shutdown_client(server->client_sys);
1539         if (!IS_ERR(server->client_acl))
1540                 rpc_shutdown_client(server->client_acl);
1541
1542         if (!(server->flags & NFS_MOUNT_NONLM))
1543                 lockd_down();   /* release rpc.lockd */
1544
1545         rpciod_down();          /* release rpciod */
1546
1547         if (server->hostname != NULL)
1548                 kfree(server->hostname);
1549         kfree(server);
1550 }
1551
1552 static struct file_system_type nfs_fs_type = {
1553         .owner          = THIS_MODULE,
1554         .name           = "nfs",
1555         .get_sb         = nfs_get_sb,
1556         .kill_sb        = nfs_kill_super,
1557         .fs_flags       = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1558 };
1559
1560 #ifdef CONFIG_NFS_V4
1561
1562 static void nfs4_clear_inode(struct inode *);
1563
1564
1565 static struct super_operations nfs4_sops = { 
1566         .alloc_inode    = nfs_alloc_inode,
1567         .destroy_inode  = nfs_destroy_inode,
1568         .write_inode    = nfs_write_inode,
1569         .delete_inode   = nfs_delete_inode,
1570         .statfs         = nfs_statfs,
1571         .clear_inode    = nfs4_clear_inode,
1572         .umount_begin   = nfs_umount_begin,
1573         .show_options   = nfs_show_options,
1574 };
1575
1576 /*
1577  * Clean out any remaining NFSv4 state that might be left over due
1578  * to open() calls that passed nfs_atomic_lookup, but failed to call
1579  * nfs_open().
1580  */
1581 static void nfs4_clear_inode(struct inode *inode)
1582 {
1583         struct nfs_inode *nfsi = NFS_I(inode);
1584
1585         /* If we are holding a delegation, return it! */
1586         if (nfsi->delegation != NULL)
1587                 nfs_inode_return_delegation(inode);
1588         /* First call standard NFS clear_inode() code */
1589         nfs_clear_inode(inode);
1590         /* Now clear out any remaining state */
1591         while (!list_empty(&nfsi->open_states)) {
1592                 struct nfs4_state *state;
1593                 
1594                 state = list_entry(nfsi->open_states.next,
1595                                 struct nfs4_state,
1596                                 inode_states);
1597                 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1598                                 __FUNCTION__,
1599                                 inode->i_sb->s_id,
1600                                 (long long)NFS_FILEID(inode),
1601                                 state);
1602                 BUG_ON(atomic_read(&state->count) != 1);
1603                 nfs4_close_state(state, state->state);
1604         }
1605 }
1606
1607
1608 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1609 {
1610         struct nfs_server *server;
1611         struct nfs4_client *clp = NULL;
1612         struct rpc_xprt *xprt = NULL;
1613         struct rpc_clnt *clnt = NULL;
1614         struct rpc_timeout timeparms;
1615         rpc_authflavor_t authflavour;
1616         int proto, err = -EIO;
1617
1618         sb->s_blocksize_bits = 0;
1619         sb->s_blocksize = 0;
1620         server = NFS_SB(sb);
1621         if (data->rsize != 0)
1622                 server->rsize = nfs_block_size(data->rsize, NULL);
1623         if (data->wsize != 0)
1624                 server->wsize = nfs_block_size(data->wsize, NULL);
1625         server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1626         server->caps = NFS_CAP_ATOMIC_OPEN;
1627
1628         server->acregmin = data->acregmin*HZ;
1629         server->acregmax = data->acregmax*HZ;
1630         server->acdirmin = data->acdirmin*HZ;
1631         server->acdirmax = data->acdirmax*HZ;
1632
1633         server->rpc_ops = &nfs_v4_clientops;
1634         /* Initialize timeout values */
1635
1636         timeparms.to_initval = data->timeo * HZ / 10;
1637         timeparms.to_retries = data->retrans;
1638         timeparms.to_exponential = 1;
1639         if (!timeparms.to_retries)
1640                 timeparms.to_retries = 5;
1641
1642         proto = data->proto;
1643         /* Which IP protocol do we use? */
1644         switch (proto) {
1645         case IPPROTO_TCP:
1646                 timeparms.to_maxval  = RPC_MAX_TCP_TIMEOUT;
1647                 if (!timeparms.to_initval)
1648                         timeparms.to_initval = 600 * HZ / 10;
1649                 break;
1650         case IPPROTO_UDP:
1651                 timeparms.to_maxval  = RPC_MAX_UDP_TIMEOUT;
1652                 if (!timeparms.to_initval)
1653                         timeparms.to_initval = 11 * HZ / 10;
1654                 break;
1655         default:
1656                 return -EINVAL;
1657         }
1658
1659         clp = nfs4_get_client(&server->addr.sin_addr);
1660         if (!clp) {
1661                 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1662                 return -EIO;
1663         }
1664
1665         /* Now create transport and client */
1666         authflavour = RPC_AUTH_UNIX;
1667         if (data->auth_flavourlen != 0) {
1668                 if (data->auth_flavourlen != 1) {
1669                         dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1670                                         __FUNCTION__, data->auth_flavourlen);
1671                         err = -EINVAL;
1672                         goto out_fail;
1673                 }
1674                 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1675                         err = -EFAULT;
1676                         goto out_fail;
1677                 }
1678         }
1679
1680         down_write(&clp->cl_sem);
1681         if (IS_ERR(clp->cl_rpcclient)) {
1682                 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1683                 if (IS_ERR(xprt)) {
1684                         up_write(&clp->cl_sem);
1685                         err = PTR_ERR(xprt);
1686                         dprintk("%s: cannot create RPC transport. Error = %d\n",
1687                                         __FUNCTION__, err);
1688                         goto out_fail;
1689                 }
1690                 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1691                                 server->rpc_ops->version, authflavour);
1692                 if (IS_ERR(clnt)) {
1693                         up_write(&clp->cl_sem);
1694                         err = PTR_ERR(clnt);
1695                         dprintk("%s: cannot create RPC client. Error = %d\n",
1696                                         __FUNCTION__, err);
1697                         goto out_fail;
1698                 }
1699                 clnt->cl_intr     = 1;
1700                 clnt->cl_softrtry = 1;
1701                 clnt->cl_chatty   = 1;
1702                 clp->cl_rpcclient = clnt;
1703                 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1704                 if (IS_ERR(clp->cl_cred)) {
1705                         up_write(&clp->cl_sem);
1706                         err = PTR_ERR(clp->cl_cred);
1707                         clp->cl_cred = NULL;
1708                         goto out_fail;
1709                 }
1710                 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1711                 nfs_idmap_new(clp);
1712         }
1713         if (list_empty(&clp->cl_superblocks)) {
1714                 err = nfs4_init_client(clp);
1715                 if (err != 0) {
1716                         up_write(&clp->cl_sem);
1717                         goto out_fail;
1718                 }
1719         }
1720         list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1721         clnt = rpc_clone_client(clp->cl_rpcclient);
1722         if (!IS_ERR(clnt))
1723                         server->nfs4_state = clp;
1724         up_write(&clp->cl_sem);
1725         clp = NULL;
1726
1727         if (IS_ERR(clnt)) {
1728                 err = PTR_ERR(clnt);
1729                 dprintk("%s: cannot create RPC client. Error = %d\n",
1730                                 __FUNCTION__, err);
1731                 return err;
1732         }
1733
1734         server->client    = clnt;
1735
1736         if (server->nfs4_state->cl_idmap == NULL) {
1737                 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1738                 return -ENOMEM;
1739         }
1740
1741         if (clnt->cl_auth->au_flavor != authflavour) {
1742                 struct rpc_auth *auth;
1743
1744                 auth = rpcauth_create(authflavour, clnt);
1745                 if (IS_ERR(auth)) {
1746                         dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1747                         return PTR_ERR(auth);
1748                 }
1749         }
1750
1751         sb->s_time_gran = 1;
1752
1753         sb->s_op = &nfs4_sops;
1754         err = nfs_sb_init(sb, authflavour);
1755         if (err == 0)
1756                 return 0;
1757 out_fail:
1758         if (clp)
1759                 nfs4_put_client(clp);
1760         return err;
1761 }
1762
1763 static int nfs4_compare_super(struct super_block *sb, void *data)
1764 {
1765         struct nfs_server *server = data;
1766         struct nfs_server *old = NFS_SB(sb);
1767
1768         if (strcmp(server->hostname, old->hostname) != 0)
1769                 return 0;
1770         if (strcmp(server->mnt_path, old->mnt_path) != 0)
1771                 return 0;
1772         return 1;
1773 }
1774
1775 static void *
1776 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1777 {
1778         void *p = NULL;
1779
1780         if (!src->len)
1781                 return ERR_PTR(-EINVAL);
1782         if (src->len < maxlen)
1783                 maxlen = src->len;
1784         if (dst == NULL) {
1785                 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1786                 if (p == NULL)
1787                         return ERR_PTR(-ENOMEM);
1788         }
1789         if (copy_from_user(dst, src->data, maxlen)) {
1790                 if (p != NULL)
1791                         kfree(p);
1792                 return ERR_PTR(-EFAULT);
1793         }
1794         dst[maxlen] = '\0';
1795         return dst;
1796 }
1797
1798 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1799         int flags, const char *dev_name, void *raw_data)
1800 {
1801         int error;
1802         struct nfs_server *server;
1803         struct super_block *s;
1804         struct nfs4_mount_data *data = raw_data;
1805         void *p;
1806
1807         if (data == NULL) {
1808                 dprintk("%s: missing data argument\n", __FUNCTION__);
1809                 return ERR_PTR(-EINVAL);
1810         }
1811         if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1812                 dprintk("%s: bad mount version\n", __FUNCTION__);
1813                 return ERR_PTR(-EINVAL);
1814         }
1815
1816         server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1817         if (!server)
1818                 return ERR_PTR(-ENOMEM);
1819         memset(server, 0, sizeof(struct nfs_server));
1820         /* Zero out the NFS state stuff */
1821         init_nfsv4_state(server);
1822         server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1823
1824         p = nfs_copy_user_string(NULL, &data->hostname, 256);
1825         if (IS_ERR(p))
1826                 goto out_err;
1827         server->hostname = p;
1828
1829         p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1830         if (IS_ERR(p))
1831                 goto out_err;
1832         server->mnt_path = p;
1833
1834         p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1835                         sizeof(server->ip_addr) - 1);
1836         if (IS_ERR(p))
1837                 goto out_err;
1838
1839         /* We now require that the mount process passes the remote address */
1840         if (data->host_addrlen != sizeof(server->addr)) {
1841                 s = ERR_PTR(-EINVAL);
1842                 goto out_free;
1843         }
1844         if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1845                 s = ERR_PTR(-EFAULT);
1846                 goto out_free;
1847         }
1848         if (server->addr.sin_family != AF_INET ||
1849             server->addr.sin_addr.s_addr == INADDR_ANY) {
1850                 dprintk("%s: mount program didn't pass remote IP address!\n",
1851                                 __FUNCTION__);
1852                 s = ERR_PTR(-EINVAL);
1853                 goto out_free;
1854         }
1855
1856         /* Fire up rpciod if not yet running */
1857         s = ERR_PTR(rpciod_up());
1858         if (IS_ERR(s)) {
1859                 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1860                                 __FUNCTION__, PTR_ERR(s));
1861                 goto out_free;
1862         }
1863
1864         s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1865
1866         if (IS_ERR(s) || s->s_root)
1867                 goto out_free;
1868
1869         s->s_flags = flags;
1870
1871         error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1872         if (error) {
1873                 up_write(&s->s_umount);
1874                 deactivate_super(s);
1875                 return ERR_PTR(error);
1876         }
1877         s->s_flags |= MS_ACTIVE;
1878         return s;
1879 out_err:
1880         s = (struct super_block *)p;
1881 out_free:
1882         if (server->mnt_path)
1883                 kfree(server->mnt_path);
1884         if (server->hostname)
1885                 kfree(server->hostname);
1886         kfree(server);
1887         return s;
1888 }
1889
1890 static void nfs4_kill_super(struct super_block *sb)
1891 {
1892         struct nfs_server *server = NFS_SB(sb);
1893
1894         nfs_return_all_delegations(sb);
1895         kill_anon_super(sb);
1896
1897         nfs4_renewd_prepare_shutdown(server);
1898
1899         if (server->client != NULL && !IS_ERR(server->client))
1900                 rpc_shutdown_client(server->client);
1901         rpciod_down();          /* release rpciod */
1902
1903         destroy_nfsv4_state(server);
1904
1905         if (server->hostname != NULL)
1906                 kfree(server->hostname);
1907         kfree(server);
1908 }
1909
1910 static struct file_system_type nfs4_fs_type = {
1911         .owner          = THIS_MODULE,
1912         .name           = "nfs4",
1913         .get_sb         = nfs4_get_sb,
1914         .kill_sb        = nfs4_kill_super,
1915         .fs_flags       = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1916 };
1917
1918 #define nfs4_init_once(nfsi) \
1919         do { \
1920                 INIT_LIST_HEAD(&(nfsi)->open_states); \
1921                 nfsi->delegation = NULL; \
1922                 nfsi->delegation_state = 0; \
1923                 init_rwsem(&nfsi->rwsem); \
1924         } while(0)
1925 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1926 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1927 #else
1928 #define nfs4_init_once(nfsi) \
1929         do { } while (0)
1930 #define register_nfs4fs() (0)
1931 #define unregister_nfs4fs()
1932 #endif
1933
1934 extern int nfs_init_nfspagecache(void);
1935 extern void nfs_destroy_nfspagecache(void);
1936 extern int nfs_init_readpagecache(void);
1937 extern void nfs_destroy_readpagecache(void);
1938 extern int nfs_init_writepagecache(void);
1939 extern void nfs_destroy_writepagecache(void);
1940 #ifdef CONFIG_NFS_DIRECTIO
1941 extern int nfs_init_directcache(void);
1942 extern void nfs_destroy_directcache(void);
1943 #endif
1944
1945 static kmem_cache_t * nfs_inode_cachep;
1946
1947 static struct inode *nfs_alloc_inode(struct super_block *sb)
1948 {
1949         struct nfs_inode *nfsi;
1950         nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
1951         if (!nfsi)
1952                 return NULL;
1953         nfsi->flags = 0;
1954 #ifdef CONFIG_NFS_V3_ACL
1955         nfsi->acl_access = ERR_PTR(-EAGAIN);
1956         nfsi->acl_default = ERR_PTR(-EAGAIN);
1957 #endif
1958 #ifdef CONFIG_NFS_V4
1959         nfsi->nfs4_acl = NULL;
1960 #endif /* CONFIG_NFS_V4 */
1961         return &nfsi->vfs_inode;
1962 }
1963
1964 static void nfs_destroy_inode(struct inode *inode)
1965 {
1966         kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1967 }
1968
1969 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
1970 {
1971         struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1972
1973         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1974             SLAB_CTOR_CONSTRUCTOR) {
1975                 inode_init_once(&nfsi->vfs_inode);
1976                 spin_lock_init(&nfsi->req_lock);
1977                 INIT_LIST_HEAD(&nfsi->dirty);
1978                 INIT_LIST_HEAD(&nfsi->commit);
1979                 INIT_LIST_HEAD(&nfsi->open_files);
1980                 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1981                 atomic_set(&nfsi->data_updates, 0);
1982                 nfsi->ndirty = 0;
1983                 nfsi->ncommit = 0;
1984                 nfsi->npages = 0;
1985                 init_waitqueue_head(&nfsi->nfs_i_wait);
1986                 nfs4_init_once(nfsi);
1987         }
1988 }
1989  
1990 static int nfs_init_inodecache(void)
1991 {
1992         nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1993                                              sizeof(struct nfs_inode),
1994                                              0, SLAB_RECLAIM_ACCOUNT,
1995                                              init_once, NULL);
1996         if (nfs_inode_cachep == NULL)
1997                 return -ENOMEM;
1998
1999         return 0;
2000 }
2001
2002 static void nfs_destroy_inodecache(void)
2003 {
2004         if (kmem_cache_destroy(nfs_inode_cachep))
2005                 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2006 }
2007
2008 /*
2009  * Initialize NFS
2010  */
2011 static int __init init_nfs_fs(void)
2012 {
2013         int err;
2014
2015         err = nfs_init_nfspagecache();
2016         if (err)
2017                 goto out4;
2018
2019         err = nfs_init_inodecache();
2020         if (err)
2021                 goto out3;
2022
2023         err = nfs_init_readpagecache();
2024         if (err)
2025                 goto out2;
2026
2027         err = nfs_init_writepagecache();
2028         if (err)
2029                 goto out1;
2030
2031 #ifdef CONFIG_NFS_DIRECTIO
2032         err = nfs_init_directcache();
2033         if (err)
2034                 goto out0;
2035 #endif
2036
2037 #ifdef CONFIG_PROC_FS
2038         rpc_proc_register(&nfs_rpcstat);
2039 #endif
2040         err = register_filesystem(&nfs_fs_type);
2041         if (err)
2042                 goto out;
2043         if ((err = register_nfs4fs()) != 0)
2044                 goto out;
2045         return 0;
2046 out:
2047 #ifdef CONFIG_PROC_FS
2048         rpc_proc_unregister("nfs");
2049 #endif
2050         nfs_destroy_writepagecache();
2051 #ifdef CONFIG_NFS_DIRECTIO
2052 out0:
2053         nfs_destroy_directcache();
2054 #endif
2055 out1:
2056         nfs_destroy_readpagecache();
2057 out2:
2058         nfs_destroy_inodecache();
2059 out3:
2060         nfs_destroy_nfspagecache();
2061 out4:
2062         return err;
2063 }
2064
2065 static void __exit exit_nfs_fs(void)
2066 {
2067 #ifdef CONFIG_NFS_DIRECTIO
2068         nfs_destroy_directcache();
2069 #endif
2070         nfs_destroy_writepagecache();
2071         nfs_destroy_readpagecache();
2072         nfs_destroy_inodecache();
2073         nfs_destroy_nfspagecache();
2074 #ifdef CONFIG_PROC_FS
2075         rpc_proc_unregister("nfs");
2076 #endif
2077         unregister_filesystem(&nfs_fs_type);
2078         unregister_nfs4fs();
2079 }
2080
2081 /* Not quite true; I just maintain it */
2082 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2083 MODULE_LICENSE("GPL");
2084
2085 module_init(init_nfs_fs)
2086 module_exit(exit_nfs_fs)