Merge /spare/repo/linux-2.6/
[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         truncate_inode_pages(&inode->i_data, 0);
150
151         nfs_wb_all(inode);
152         /*
153          * The following should never happen...
154          */
155         if (nfs_have_writebacks(inode)) {
156                 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
157         }
158
159         clear_inode(inode);
160 }
161
162 static void
163 nfs_clear_inode(struct inode *inode)
164 {
165         struct nfs_inode *nfsi = NFS_I(inode);
166         struct rpc_cred *cred;
167
168         nfs_wb_all(inode);
169         BUG_ON (!list_empty(&nfsi->open_files));
170         nfs_zap_acl_cache(inode);
171         cred = nfsi->cache_access.cred;
172         if (cred)
173                 put_rpccred(cred);
174         BUG_ON(atomic_read(&nfsi->data_updates) != 0);
175 }
176
177 void
178 nfs_umount_begin(struct super_block *sb)
179 {
180         struct rpc_clnt *rpc = NFS_SB(sb)->client;
181
182         /* -EIO all pending I/O */
183         if (!IS_ERR(rpc))
184                 rpc_killall_tasks(rpc);
185         rpc = NFS_SB(sb)->client_acl;
186         if (!IS_ERR(rpc))
187                 rpc_killall_tasks(rpc);
188 }
189
190
191 static inline unsigned long
192 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
193 {
194         /* make sure blocksize is a power of two */
195         if ((bsize & (bsize - 1)) || nrbitsp) {
196                 unsigned char   nrbits;
197
198                 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
199                         ;
200                 bsize = 1 << nrbits;
201                 if (nrbitsp)
202                         *nrbitsp = nrbits;
203         }
204
205         return bsize;
206 }
207
208 /*
209  * Calculate the number of 512byte blocks used.
210  */
211 static inline unsigned long
212 nfs_calc_block_size(u64 tsize)
213 {
214         loff_t used = (tsize + 511) >> 9;
215         return (used > ULONG_MAX) ? ULONG_MAX : used;
216 }
217
218 /*
219  * Compute and set NFS server blocksize
220  */
221 static inline unsigned long
222 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
223 {
224         if (bsize < 1024)
225                 bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
226         else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
227                 bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
228
229         return nfs_block_bits(bsize, nrbitsp);
230 }
231
232 /*
233  * Obtain the root inode of the file system.
234  */
235 static struct inode *
236 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
237 {
238         struct nfs_server       *server = NFS_SB(sb);
239         struct inode *rooti;
240         int                     error;
241
242         error = server->rpc_ops->getroot(server, rootfh, fsinfo);
243         if (error < 0) {
244                 dprintk("nfs_get_root: getattr error = %d\n", -error);
245                 return ERR_PTR(error);
246         }
247
248         rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
249         if (!rooti)
250                 return ERR_PTR(-ENOMEM);
251         return rooti;
252 }
253
254 /*
255  * Do NFS version-independent mount processing, and sanity checking
256  */
257 static int
258 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
259 {
260         struct nfs_server       *server;
261         struct inode            *root_inode;
262         struct nfs_fattr        fattr;
263         struct nfs_fsinfo       fsinfo = {
264                                         .fattr = &fattr,
265                                 };
266         struct nfs_pathconf pathinfo = {
267                         .fattr = &fattr,
268         };
269         int no_root_error = 0;
270         unsigned long max_rpc_payload;
271
272         /* We probably want something more informative here */
273         snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
274
275         server = NFS_SB(sb);
276
277         sb->s_magic      = NFS_SUPER_MAGIC;
278
279         root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
280         /* Did getting the root inode fail? */
281         if (IS_ERR(root_inode)) {
282                 no_root_error = PTR_ERR(root_inode);
283                 goto out_no_root;
284         }
285         sb->s_root = d_alloc_root(root_inode);
286         if (!sb->s_root) {
287                 no_root_error = -ENOMEM;
288                 goto out_no_root;
289         }
290         sb->s_root->d_op = server->rpc_ops->dentry_ops;
291
292         /* Get some general file system info */
293         if (server->namelen == 0 &&
294             server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
295                 server->namelen = pathinfo.max_namelen;
296         /* Work out a lot of parameters */
297         if (server->rsize == 0)
298                 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
299         if (server->wsize == 0)
300                 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
301
302         if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
303                 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
304         if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
305                 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
306
307         max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
308         if (server->rsize > max_rpc_payload)
309                 server->rsize = max_rpc_payload;
310         if (server->wsize > max_rpc_payload)
311                 server->wsize = max_rpc_payload;
312
313         server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
314         if (server->rpages > NFS_READ_MAXIOV) {
315                 server->rpages = NFS_READ_MAXIOV;
316                 server->rsize = server->rpages << PAGE_CACHE_SHIFT;
317         }
318
319         server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
320         if (server->wpages > NFS_WRITE_MAXIOV) {
321                 server->wpages = NFS_WRITE_MAXIOV;
322                 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
323         }
324
325         if (sb->s_blocksize == 0)
326                 sb->s_blocksize = nfs_block_bits(server->wsize,
327                                                          &sb->s_blocksize_bits);
328         server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
329
330         server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
331         if (server->dtsize > PAGE_CACHE_SIZE)
332                 server->dtsize = PAGE_CACHE_SIZE;
333         if (server->dtsize > server->rsize)
334                 server->dtsize = server->rsize;
335
336         if (server->flags & NFS_MOUNT_NOAC) {
337                 server->acregmin = server->acregmax = 0;
338                 server->acdirmin = server->acdirmax = 0;
339                 sb->s_flags |= MS_SYNCHRONOUS;
340         }
341         server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
342
343         sb->s_maxbytes = fsinfo.maxfilesize;
344         if (sb->s_maxbytes > MAX_LFS_FILESIZE) 
345                 sb->s_maxbytes = MAX_LFS_FILESIZE; 
346
347         server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
348         server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
349
350         /* We're airborne Set socket buffersize */
351         rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
352         return 0;
353         /* Yargs. It didn't work out. */
354 out_no_root:
355         dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
356         if (!IS_ERR(root_inode))
357                 iput(root_inode);
358         return no_root_error;
359 }
360
361 /*
362  * Create an RPC client handle.
363  */
364 static struct rpc_clnt *
365 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
366 {
367         struct rpc_timeout      timeparms;
368         struct rpc_xprt         *xprt = NULL;
369         struct rpc_clnt         *clnt = NULL;
370         int                     tcp   = (data->flags & NFS_MOUNT_TCP);
371
372         /* Initialize timeout values */
373         timeparms.to_initval = data->timeo * HZ / 10;
374         timeparms.to_retries = data->retrans;
375         timeparms.to_maxval  = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
376         timeparms.to_exponential = 1;
377
378         if (!timeparms.to_initval)
379                 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
380         if (!timeparms.to_retries)
381                 timeparms.to_retries = 5;
382
383         /* create transport and client */
384         xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
385                                  &server->addr, &timeparms);
386         if (IS_ERR(xprt)) {
387                 dprintk("%s: cannot create RPC transport. Error = %ld\n",
388                                 __FUNCTION__, PTR_ERR(xprt));
389                 return (struct rpc_clnt *)xprt;
390         }
391         clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
392                                  server->rpc_ops->version, data->pseudoflavor);
393         if (IS_ERR(clnt)) {
394                 dprintk("%s: cannot create RPC client. Error = %ld\n",
395                                 __FUNCTION__, PTR_ERR(xprt));
396                 goto out_fail;
397         }
398
399         clnt->cl_intr     = 1;
400         clnt->cl_softrtry = 1;
401         clnt->cl_chatty   = 1;
402
403         return clnt;
404
405 out_fail:
406         return clnt;
407 }
408
409 /*
410  * The way this works is that the mount process passes a structure
411  * in the data argument which contains the server's IP address
412  * and the root file handle obtained from the server's mount
413  * daemon. We stash these away in the private superblock fields.
414  */
415 static int
416 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
417 {
418         struct nfs_server       *server;
419         rpc_authflavor_t        authflavor;
420
421         server           = NFS_SB(sb);
422         sb->s_blocksize_bits = 0;
423         sb->s_blocksize = 0;
424         if (data->bsize)
425                 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
426         if (data->rsize)
427                 server->rsize = nfs_block_size(data->rsize, NULL);
428         if (data->wsize)
429                 server->wsize = nfs_block_size(data->wsize, NULL);
430         server->flags    = data->flags & NFS_MOUNT_FLAGMASK;
431
432         server->acregmin = data->acregmin*HZ;
433         server->acregmax = data->acregmax*HZ;
434         server->acdirmin = data->acdirmin*HZ;
435         server->acdirmax = data->acdirmax*HZ;
436
437         /* Start lockd here, before we might error out */
438         if (!(server->flags & NFS_MOUNT_NONLM))
439                 lockd_up();
440
441         server->namelen  = data->namlen;
442         server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
443         if (!server->hostname)
444                 return -ENOMEM;
445         strcpy(server->hostname, data->hostname);
446
447         /* Check NFS protocol revision and initialize RPC op vector
448          * and file handle pool. */
449 #ifdef CONFIG_NFS_V3
450         if (server->flags & NFS_MOUNT_VER3) {
451                 server->rpc_ops = &nfs_v3_clientops;
452                 server->caps |= NFS_CAP_READDIRPLUS;
453         } else {
454                 server->rpc_ops = &nfs_v2_clientops;
455         }
456 #else
457         server->rpc_ops = &nfs_v2_clientops;
458 #endif
459
460         /* Fill in pseudoflavor for mount version < 5 */
461         if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
462                 data->pseudoflavor = RPC_AUTH_UNIX;
463         authflavor = data->pseudoflavor;        /* save for sb_init() */
464         /* XXX maybe we want to add a server->pseudoflavor field */
465
466         /* Create RPC client handles */
467         server->client = nfs_create_client(server, data);
468         if (IS_ERR(server->client))
469                 return PTR_ERR(server->client);
470         /* RFC 2623, sec 2.3.2 */
471         if (authflavor != RPC_AUTH_UNIX) {
472                 struct rpc_auth *auth;
473
474                 server->client_sys = rpc_clone_client(server->client);
475                 if (IS_ERR(server->client_sys))
476                         return PTR_ERR(server->client_sys);
477                 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
478                 if (IS_ERR(auth))
479                         return PTR_ERR(auth);
480         } else {
481                 atomic_inc(&server->client->cl_count);
482                 server->client_sys = server->client;
483         }
484         if (server->flags & NFS_MOUNT_VER3) {
485 #ifdef CONFIG_NFS_V3_ACL
486                 if (!(server->flags & NFS_MOUNT_NOACL)) {
487                         server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
488                         /* No errors! Assume that Sun nfsacls are supported */
489                         if (!IS_ERR(server->client_acl))
490                                 server->caps |= NFS_CAP_ACLS;
491                 }
492 #else
493                 server->flags &= ~NFS_MOUNT_NOACL;
494 #endif /* CONFIG_NFS_V3_ACL */
495                 /*
496                  * The VFS shouldn't apply the umask to mode bits. We will
497                  * do so ourselves when necessary.
498                  */
499                 sb->s_flags |= MS_POSIXACL;
500                 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
501                         server->namelen = NFS3_MAXNAMLEN;
502                 sb->s_time_gran = 1;
503         } else {
504                 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
505                         server->namelen = NFS2_MAXNAMLEN;
506         }
507
508         sb->s_op = &nfs_sops;
509         return nfs_sb_init(sb, authflavor);
510 }
511
512 static int
513 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
514 {
515         struct nfs_server *server = NFS_SB(sb);
516         unsigned char blockbits;
517         unsigned long blockres;
518         struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
519         struct nfs_fattr fattr;
520         struct nfs_fsstat res = {
521                         .fattr = &fattr,
522         };
523         int error;
524
525         lock_kernel();
526
527         error = server->rpc_ops->statfs(server, rootfh, &res);
528         buf->f_type = NFS_SUPER_MAGIC;
529         if (error < 0)
530                 goto out_err;
531
532         /*
533          * Current versions of glibc do not correctly handle the
534          * case where f_frsize != f_bsize.  Eventually we want to
535          * report the value of wtmult in this field.
536          */
537         buf->f_frsize = sb->s_blocksize;
538
539         /*
540          * On most *nix systems, f_blocks, f_bfree, and f_bavail
541          * are reported in units of f_frsize.  Linux hasn't had
542          * an f_frsize field in its statfs struct until recently,
543          * thus historically Linux's sys_statfs reports these
544          * fields in units of f_bsize.
545          */
546         buf->f_bsize = sb->s_blocksize;
547         blockbits = sb->s_blocksize_bits;
548         blockres = (1 << blockbits) - 1;
549         buf->f_blocks = (res.tbytes + blockres) >> blockbits;
550         buf->f_bfree = (res.fbytes + blockres) >> blockbits;
551         buf->f_bavail = (res.abytes + blockres) >> blockbits;
552
553         buf->f_files = res.tfiles;
554         buf->f_ffree = res.afiles;
555
556         buf->f_namelen = server->namelen;
557  out:
558         unlock_kernel();
559
560         return 0;
561
562  out_err:
563         printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
564         buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
565         goto out;
566
567 }
568
569 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
570 {
571         static struct proc_nfs_info {
572                 int flag;
573                 char *str;
574                 char *nostr;
575         } nfs_info[] = {
576                 { NFS_MOUNT_SOFT, ",soft", ",hard" },
577                 { NFS_MOUNT_INTR, ",intr", "" },
578                 { NFS_MOUNT_POSIX, ",posix", "" },
579                 { NFS_MOUNT_TCP, ",tcp", ",udp" },
580                 { NFS_MOUNT_NOCTO, ",nocto", "" },
581                 { NFS_MOUNT_NOAC, ",noac", "" },
582                 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
583                 { NFS_MOUNT_NOACL, ",noacl", "" },
584                 { 0, NULL, NULL }
585         };
586         struct proc_nfs_info *nfs_infop;
587         struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
588
589         seq_printf(m, ",v%d", nfss->rpc_ops->version);
590         seq_printf(m, ",rsize=%d", nfss->rsize);
591         seq_printf(m, ",wsize=%d", nfss->wsize);
592         if (nfss->acregmin != 3*HZ)
593                 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
594         if (nfss->acregmax != 60*HZ)
595                 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
596         if (nfss->acdirmin != 30*HZ)
597                 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
598         if (nfss->acdirmax != 60*HZ)
599                 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
600         for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
601                 if (nfss->flags & nfs_infop->flag)
602                         seq_puts(m, nfs_infop->str);
603                 else
604                         seq_puts(m, nfs_infop->nostr);
605         }
606         seq_puts(m, ",addr=");
607         seq_escape(m, nfss->hostname, " \t\n\\");
608         return 0;
609 }
610
611 /*
612  * Invalidate the local caches
613  */
614 void
615 nfs_zap_caches(struct inode *inode)
616 {
617         struct nfs_inode *nfsi = NFS_I(inode);
618         int mode = inode->i_mode;
619
620         spin_lock(&inode->i_lock);
621
622         NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
623         NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
624
625         memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
626         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
627                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
628         else
629                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
630
631         spin_unlock(&inode->i_lock);
632 }
633
634 static void nfs_zap_acl_cache(struct inode *inode)
635 {
636         void (*clear_acl_cache)(struct inode *);
637
638         clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
639         if (clear_acl_cache != NULL)
640                 clear_acl_cache(inode);
641         spin_lock(&inode->i_lock);
642         NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
643         spin_unlock(&inode->i_lock);
644 }
645
646 /*
647  * Invalidate, but do not unhash, the inode
648  */
649 static void
650 nfs_invalidate_inode(struct inode *inode)
651 {
652         umode_t save_mode = inode->i_mode;
653
654         make_bad_inode(inode);
655         inode->i_mode = save_mode;
656         nfs_zap_caches(inode);
657 }
658
659 struct nfs_find_desc {
660         struct nfs_fh           *fh;
661         struct nfs_fattr        *fattr;
662 };
663
664 /*
665  * In NFSv3 we can have 64bit inode numbers. In order to support
666  * this, and re-exported directories (also seen in NFSv2)
667  * we are forced to allow 2 different inodes to have the same
668  * i_ino.
669  */
670 static int
671 nfs_find_actor(struct inode *inode, void *opaque)
672 {
673         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
674         struct nfs_fh           *fh = desc->fh;
675         struct nfs_fattr        *fattr = desc->fattr;
676
677         if (NFS_FILEID(inode) != fattr->fileid)
678                 return 0;
679         if (nfs_compare_fh(NFS_FH(inode), fh))
680                 return 0;
681         if (is_bad_inode(inode) || NFS_STALE(inode))
682                 return 0;
683         return 1;
684 }
685
686 static int
687 nfs_init_locked(struct inode *inode, void *opaque)
688 {
689         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
690         struct nfs_fattr        *fattr = desc->fattr;
691
692         NFS_FILEID(inode) = fattr->fileid;
693         nfs_copy_fh(NFS_FH(inode), desc->fh);
694         return 0;
695 }
696
697 /* Don't use READDIRPLUS on directories that we believe are too large */
698 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
699
700 /*
701  * This is our front-end to iget that looks up inodes by file handle
702  * instead of inode number.
703  */
704 struct inode *
705 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
706 {
707         struct nfs_find_desc desc = {
708                 .fh     = fh,
709                 .fattr  = fattr
710         };
711         struct inode *inode = NULL;
712         unsigned long hash;
713
714         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
715                 goto out_no_inode;
716
717         if (!fattr->nlink) {
718                 printk("NFS: Buggy server - nlink == 0!\n");
719                 goto out_no_inode;
720         }
721
722         hash = nfs_fattr_to_ino_t(fattr);
723
724         if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
725                 goto out_no_inode;
726
727         if (inode->i_state & I_NEW) {
728                 struct nfs_inode *nfsi = NFS_I(inode);
729
730                 /* We set i_ino for the few things that still rely on it,
731                  * such as stat(2) */
732                 inode->i_ino = hash;
733
734                 /* We can't support update_atime(), since the server will reset it */
735                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
736                 inode->i_mode = fattr->mode;
737                 /* Why so? Because we want revalidate for devices/FIFOs, and
738                  * that's precisely what we have in nfs_file_inode_operations.
739                  */
740                 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
741                 if (S_ISREG(inode->i_mode)) {
742                         inode->i_fop = &nfs_file_operations;
743                         inode->i_data.a_ops = &nfs_file_aops;
744                         inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
745                 } else if (S_ISDIR(inode->i_mode)) {
746                         inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
747                         inode->i_fop = &nfs_dir_operations;
748                         if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
749                             && fattr->size <= NFS_LIMIT_READDIRPLUS)
750                                 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
751                 } else if (S_ISLNK(inode->i_mode))
752                         inode->i_op = &nfs_symlink_inode_operations;
753                 else
754                         init_special_inode(inode, inode->i_mode, fattr->rdev);
755
756                 nfsi->read_cache_jiffies = fattr->timestamp;
757                 inode->i_atime = fattr->atime;
758                 inode->i_mtime = fattr->mtime;
759                 inode->i_ctime = fattr->ctime;
760                 if (fattr->valid & NFS_ATTR_FATTR_V4)
761                         nfsi->change_attr = fattr->change_attr;
762                 inode->i_size = nfs_size_to_loff_t(fattr->size);
763                 inode->i_nlink = fattr->nlink;
764                 inode->i_uid = fattr->uid;
765                 inode->i_gid = fattr->gid;
766                 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
767                         /*
768                          * report the blocks in 512byte units
769                          */
770                         inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
771                         inode->i_blksize = inode->i_sb->s_blocksize;
772                 } else {
773                         inode->i_blocks = fattr->du.nfs2.blocks;
774                         inode->i_blksize = fattr->du.nfs2.blocksize;
775                 }
776                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
777                 nfsi->attrtimeo_timestamp = jiffies;
778                 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
779                 nfsi->cache_access.cred = NULL;
780
781                 unlock_new_inode(inode);
782         } else
783                 nfs_refresh_inode(inode, fattr);
784         dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
785                 inode->i_sb->s_id,
786                 (long long)NFS_FILEID(inode),
787                 atomic_read(&inode->i_count));
788
789 out:
790         return inode;
791
792 out_no_inode:
793         printk("nfs_fhget: iget failed\n");
794         goto out;
795 }
796
797 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
798
799 int
800 nfs_setattr(struct dentry *dentry, struct iattr *attr)
801 {
802         struct inode *inode = dentry->d_inode;
803         struct nfs_fattr fattr;
804         int error;
805
806         if (attr->ia_valid & ATTR_SIZE) {
807                 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
808                         attr->ia_valid &= ~ATTR_SIZE;
809         }
810
811         /* Optimization: if the end result is no change, don't RPC */
812         attr->ia_valid &= NFS_VALID_ATTRS;
813         if (attr->ia_valid == 0)
814                 return 0;
815
816         lock_kernel();
817         nfs_begin_data_update(inode);
818         /* Write all dirty data if we're changing file permissions or size */
819         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
820                 if (filemap_fdatawrite(inode->i_mapping) == 0)
821                         filemap_fdatawait(inode->i_mapping);
822                 nfs_wb_all(inode);
823         }
824         error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
825         if (error == 0)
826                 nfs_refresh_inode(inode, &fattr);
827         nfs_end_data_update(inode);
828         unlock_kernel();
829         return error;
830 }
831
832 /**
833  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
834  * @inode: pointer to struct inode
835  * @attr: pointer to struct iattr
836  *
837  * Note: we do this in the *proc.c in order to ensure that
838  *       it works for things like exclusive creates too.
839  */
840 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
841 {
842         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
843                 if ((attr->ia_valid & ATTR_MODE) != 0) {
844                         int mode = attr->ia_mode & S_IALLUGO;
845                         mode |= inode->i_mode & ~S_IALLUGO;
846                         inode->i_mode = mode;
847                 }
848                 if ((attr->ia_valid & ATTR_UID) != 0)
849                         inode->i_uid = attr->ia_uid;
850                 if ((attr->ia_valid & ATTR_GID) != 0)
851                         inode->i_gid = attr->ia_gid;
852                 spin_lock(&inode->i_lock);
853                 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
854                 spin_unlock(&inode->i_lock);
855         }
856         if ((attr->ia_valid & ATTR_SIZE) != 0) {
857                 inode->i_size = attr->ia_size;
858                 vmtruncate(inode, attr->ia_size);
859         }
860 }
861
862 static int nfs_wait_schedule(void *word)
863 {
864         if (signal_pending(current))
865                 return -ERESTARTSYS;
866         schedule();
867         return 0;
868 }
869
870 /*
871  * Wait for the inode to get unlocked.
872  */
873 static int nfs_wait_on_inode(struct inode *inode)
874 {
875         struct rpc_clnt *clnt = NFS_CLIENT(inode);
876         struct nfs_inode *nfsi = NFS_I(inode);
877         sigset_t oldmask;
878         int error;
879
880         atomic_inc(&inode->i_count);
881         rpc_clnt_sigmask(clnt, &oldmask);
882         error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
883                                         nfs_wait_schedule, TASK_INTERRUPTIBLE);
884         rpc_clnt_sigunmask(clnt, &oldmask);
885         iput(inode);
886
887         return error;
888 }
889
890 static void nfs_wake_up_inode(struct inode *inode)
891 {
892         struct nfs_inode *nfsi = NFS_I(inode);
893
894         clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
895         smp_mb__after_clear_bit();
896         wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
897 }
898
899 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
900 {
901         struct inode *inode = dentry->d_inode;
902         int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
903         int err;
904
905         if (__IS_FLG(inode, MS_NOATIME))
906                 need_atime = 0;
907         else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
908                 need_atime = 0;
909         /* We may force a getattr if the user cares about atime */
910         if (need_atime)
911                 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
912         else
913                 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
914         if (!err)
915                 generic_fillattr(inode, stat);
916         return err;
917 }
918
919 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
920 {
921         struct nfs_open_context *ctx;
922
923         ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
924         if (ctx != NULL) {
925                 atomic_set(&ctx->count, 1);
926                 ctx->dentry = dget(dentry);
927                 ctx->cred = get_rpccred(cred);
928                 ctx->state = NULL;
929                 ctx->lockowner = current->files;
930                 ctx->error = 0;
931                 ctx->dir_cookie = 0;
932         }
933         return ctx;
934 }
935
936 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
937 {
938         if (ctx != NULL)
939                 atomic_inc(&ctx->count);
940         return ctx;
941 }
942
943 void put_nfs_open_context(struct nfs_open_context *ctx)
944 {
945         if (atomic_dec_and_test(&ctx->count)) {
946                 if (!list_empty(&ctx->list)) {
947                         struct inode *inode = ctx->dentry->d_inode;
948                         spin_lock(&inode->i_lock);
949                         list_del(&ctx->list);
950                         spin_unlock(&inode->i_lock);
951                 }
952                 if (ctx->state != NULL)
953                         nfs4_close_state(ctx->state, ctx->mode);
954                 if (ctx->cred != NULL)
955                         put_rpccred(ctx->cred);
956                 dput(ctx->dentry);
957                 kfree(ctx);
958         }
959 }
960
961 /*
962  * Ensure that mmap has a recent RPC credential for use when writing out
963  * shared pages
964  */
965 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
966 {
967         struct inode *inode = filp->f_dentry->d_inode;
968         struct nfs_inode *nfsi = NFS_I(inode);
969
970         filp->private_data = get_nfs_open_context(ctx);
971         spin_lock(&inode->i_lock);
972         list_add(&ctx->list, &nfsi->open_files);
973         spin_unlock(&inode->i_lock);
974 }
975
976 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
977 {
978         struct nfs_inode *nfsi = NFS_I(inode);
979         struct nfs_open_context *pos, *ctx = NULL;
980
981         spin_lock(&inode->i_lock);
982         list_for_each_entry(pos, &nfsi->open_files, list) {
983                 if ((pos->mode & mode) == mode) {
984                         ctx = get_nfs_open_context(pos);
985                         break;
986                 }
987         }
988         spin_unlock(&inode->i_lock);
989         return ctx;
990 }
991
992 void nfs_file_clear_open_context(struct file *filp)
993 {
994         struct inode *inode = filp->f_dentry->d_inode;
995         struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
996
997         if (ctx) {
998                 filp->private_data = NULL;
999                 spin_lock(&inode->i_lock);
1000                 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1001                 spin_unlock(&inode->i_lock);
1002                 put_nfs_open_context(ctx);
1003         }
1004 }
1005
1006 /*
1007  * These allocate and release file read/write context information.
1008  */
1009 int nfs_open(struct inode *inode, struct file *filp)
1010 {
1011         struct nfs_open_context *ctx;
1012         struct rpc_cred *cred;
1013
1014         cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1015         if (IS_ERR(cred))
1016                 return PTR_ERR(cred);
1017         ctx = alloc_nfs_open_context(filp->f_dentry, cred);
1018         put_rpccred(cred);
1019         if (ctx == NULL)
1020                 return -ENOMEM;
1021         ctx->mode = filp->f_mode;
1022         nfs_file_set_open_context(filp, ctx);
1023         put_nfs_open_context(ctx);
1024         if ((filp->f_mode & FMODE_WRITE) != 0)
1025                 nfs_begin_data_update(inode);
1026         return 0;
1027 }
1028
1029 int nfs_release(struct inode *inode, struct file *filp)
1030 {
1031         if ((filp->f_mode & FMODE_WRITE) != 0)
1032                 nfs_end_data_update(inode);
1033         nfs_file_clear_open_context(filp);
1034         return 0;
1035 }
1036
1037 /*
1038  * This function is called whenever some part of NFS notices that
1039  * the cached attributes have to be refreshed.
1040  */
1041 int
1042 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1043 {
1044         int              status = -ESTALE;
1045         struct nfs_fattr fattr;
1046         struct nfs_inode *nfsi = NFS_I(inode);
1047         unsigned long verifier;
1048         unsigned long cache_validity;
1049
1050         dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1051                 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1052
1053         lock_kernel();
1054         if (!inode || is_bad_inode(inode))
1055                 goto out_nowait;
1056         if (NFS_STALE(inode))
1057                 goto out_nowait;
1058
1059         status = nfs_wait_on_inode(inode);
1060         if (status < 0)
1061                 goto out;
1062         if (NFS_STALE(inode)) {
1063                 status = -ESTALE;
1064                 /* Do we trust the cached ESTALE? */
1065                 if (NFS_ATTRTIMEO(inode) != 0) {
1066                         if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1067                                 /* no */
1068                         } else
1069                                 goto out;
1070                 }
1071         }
1072
1073         /* Protect against RPC races by saving the change attribute */
1074         verifier = nfs_save_change_attribute(inode);
1075         status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1076         if (status != 0) {
1077                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1078                          inode->i_sb->s_id,
1079                          (long long)NFS_FILEID(inode), status);
1080                 if (status == -ESTALE) {
1081                         nfs_zap_caches(inode);
1082                         if (!S_ISDIR(inode->i_mode))
1083                                 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1084                 }
1085                 goto out;
1086         }
1087
1088         status = nfs_update_inode(inode, &fattr, verifier);
1089         if (status) {
1090                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1091                          inode->i_sb->s_id,
1092                          (long long)NFS_FILEID(inode), status);
1093                 goto out;
1094         }
1095         spin_lock(&inode->i_lock);
1096         cache_validity = nfsi->cache_validity;
1097         nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1098
1099         /*
1100          * We may need to keep the attributes marked as invalid if
1101          * we raced with nfs_end_attr_update().
1102          */
1103         if (verifier == nfsi->cache_change_attribute)
1104                 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1105         spin_unlock(&inode->i_lock);
1106
1107         nfs_revalidate_mapping(inode, inode->i_mapping);
1108
1109         if (cache_validity & NFS_INO_INVALID_ACL)
1110                 nfs_zap_acl_cache(inode);
1111
1112         dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1113                 inode->i_sb->s_id,
1114                 (long long)NFS_FILEID(inode));
1115
1116  out:
1117         nfs_wake_up_inode(inode);
1118
1119  out_nowait:
1120         unlock_kernel();
1121         return status;
1122 }
1123
1124 int nfs_attribute_timeout(struct inode *inode)
1125 {
1126         struct nfs_inode *nfsi = NFS_I(inode);
1127
1128         if (nfs_have_delegation(inode, FMODE_READ))
1129                 return 0;
1130         return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1131 }
1132
1133 /**
1134  * nfs_revalidate_inode - Revalidate the inode attributes
1135  * @server - pointer to nfs_server struct
1136  * @inode - pointer to inode struct
1137  *
1138  * Updates inode attribute information by retrieving the data from the server.
1139  */
1140 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1141 {
1142         if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1143                         && !nfs_attribute_timeout(inode))
1144                 return NFS_STALE(inode) ? -ESTALE : 0;
1145         return __nfs_revalidate_inode(server, inode);
1146 }
1147
1148 /**
1149  * nfs_revalidate_mapping - Revalidate the pagecache
1150  * @inode - pointer to host inode
1151  * @mapping - pointer to mapping
1152  */
1153 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1154 {
1155         struct nfs_inode *nfsi = NFS_I(inode);
1156
1157         if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1158                 if (S_ISREG(inode->i_mode)) {
1159                         if (filemap_fdatawrite(mapping) == 0)
1160                                 filemap_fdatawait(mapping);
1161                         nfs_wb_all(inode);
1162                 }
1163                 invalidate_inode_pages2(mapping);
1164
1165                 spin_lock(&inode->i_lock);
1166                 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1167                 if (S_ISDIR(inode->i_mode)) {
1168                         memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1169                         /* This ensures we revalidate child dentries */
1170                         nfsi->cache_change_attribute++;
1171                 }
1172                 spin_unlock(&inode->i_lock);
1173
1174                 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1175                                 inode->i_sb->s_id,
1176                                 (long long)NFS_FILEID(inode));
1177         }
1178 }
1179
1180 /**
1181  * nfs_begin_data_update
1182  * @inode - pointer to inode
1183  * Declare that a set of operations will update file data on the server
1184  */
1185 void nfs_begin_data_update(struct inode *inode)
1186 {
1187         atomic_inc(&NFS_I(inode)->data_updates);
1188 }
1189
1190 /**
1191  * nfs_end_data_update
1192  * @inode - pointer to inode
1193  * Declare end of the operations that will update file data
1194  * This will mark the inode as immediately needing revalidation
1195  * of its attribute cache.
1196  */
1197 void nfs_end_data_update(struct inode *inode)
1198 {
1199         struct nfs_inode *nfsi = NFS_I(inode);
1200
1201         if (!nfs_have_delegation(inode, FMODE_READ)) {
1202                 /* Mark the attribute cache for revalidation */
1203                 spin_lock(&inode->i_lock);
1204                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1205                 /* Directories and symlinks: invalidate page cache too */
1206                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1207                         nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1208                 spin_unlock(&inode->i_lock);
1209         }
1210         nfsi->cache_change_attribute ++;
1211         atomic_dec(&nfsi->data_updates);
1212 }
1213
1214 /**
1215  * nfs_refresh_inode - verify consistency of the inode attribute cache
1216  * @inode - pointer to inode
1217  * @fattr - updated attributes
1218  *
1219  * Verifies the attribute cache. If we have just changed the attributes,
1220  * so that fattr carries weak cache consistency data, then it may
1221  * also update the ctime/mtime/change_attribute.
1222  */
1223 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1224 {
1225         struct nfs_inode *nfsi = NFS_I(inode);
1226         loff_t cur_size, new_isize;
1227         int data_unstable;
1228
1229         /* Do we hold a delegation? */
1230         if (nfs_have_delegation(inode, FMODE_READ))
1231                 return 0;
1232
1233         spin_lock(&inode->i_lock);
1234
1235         /* Are we in the process of updating data on the server? */
1236         data_unstable = nfs_caches_unstable(inode);
1237
1238         if (fattr->valid & NFS_ATTR_FATTR_V4) {
1239                 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1240                                 && nfsi->change_attr == fattr->pre_change_attr)
1241                         nfsi->change_attr = fattr->change_attr;
1242                 if (nfsi->change_attr != fattr->change_attr) {
1243                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1244                         if (!data_unstable)
1245                                 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1246                 }
1247         }
1248
1249         if ((fattr->valid & NFS_ATTR_FATTR) == 0) {
1250                 spin_unlock(&inode->i_lock);
1251                 return 0;
1252         }
1253
1254         /* Has the inode gone and changed behind our back? */
1255         if (nfsi->fileid != fattr->fileid
1256                         || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1257                 spin_unlock(&inode->i_lock);
1258                 return -EIO;
1259         }
1260
1261         cur_size = i_size_read(inode);
1262         new_isize = nfs_size_to_loff_t(fattr->size);
1263
1264         /* If we have atomic WCC data, we may update some attributes */
1265         if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1266                 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1267                         memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1268                 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1269                         memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1270         }
1271
1272         /* Verify a few of the more important attributes */
1273         if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1274                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1275                 if (!data_unstable)
1276                         nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1277         }
1278         if (cur_size != new_isize) {
1279                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1280                 if (nfsi->npages == 0)
1281                         nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1282         }
1283
1284         /* Have any file permissions changed? */
1285         if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1286                         || inode->i_uid != fattr->uid
1287                         || inode->i_gid != fattr->gid)
1288                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1289
1290         /* Has the link count changed? */
1291         if (inode->i_nlink != fattr->nlink)
1292                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1293
1294         if (!timespec_equal(&inode->i_atime, &fattr->atime))
1295                 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1296
1297         nfsi->read_cache_jiffies = fattr->timestamp;
1298         spin_unlock(&inode->i_lock);
1299         return 0;
1300 }
1301
1302 /*
1303  * Many nfs protocol calls return the new file attributes after
1304  * an operation.  Here we update the inode to reflect the state
1305  * of the server's inode.
1306  *
1307  * This is a bit tricky because we have to make sure all dirty pages
1308  * have been sent off to the server before calling invalidate_inode_pages.
1309  * To make sure no other process adds more write requests while we try
1310  * our best to flush them, we make them sleep during the attribute refresh.
1311  *
1312  * A very similar scenario holds for the dir cache.
1313  */
1314 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1315 {
1316         struct nfs_inode *nfsi = NFS_I(inode);
1317         loff_t cur_isize, new_isize;
1318         unsigned int    invalid = 0;
1319         int data_unstable;
1320
1321         dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1322                         __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1323                         atomic_read(&inode->i_count), fattr->valid);
1324
1325         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1326                 return 0;
1327
1328         if (nfsi->fileid != fattr->fileid) {
1329                 printk(KERN_ERR "%s: inode number mismatch\n"
1330                        "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1331                        __FUNCTION__,
1332                        inode->i_sb->s_id, (long long)nfsi->fileid,
1333                        inode->i_sb->s_id, (long long)fattr->fileid);
1334                 goto out_err;
1335         }
1336
1337         spin_lock(&inode->i_lock);
1338
1339         /*
1340          * Make sure the inode's type hasn't changed.
1341          */
1342         if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1343                 spin_unlock(&inode->i_lock);
1344                 goto out_changed;
1345         }
1346
1347         /*
1348          * Update the read time so we don't revalidate too often.
1349          */
1350         nfsi->read_cache_jiffies = fattr->timestamp;
1351
1352         /* Are we racing with known updates of the metadata on the server? */
1353         data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1354
1355         /* Check if our cached file size is stale */
1356         new_isize = nfs_size_to_loff_t(fattr->size);
1357         cur_isize = i_size_read(inode);
1358         if (new_isize != cur_isize) {
1359                 /* Do we perhaps have any outstanding writes? */
1360                 if (nfsi->npages == 0) {
1361                         /* No, but did we race with nfs_end_data_update()? */
1362                         if (verifier  ==  nfsi->cache_change_attribute) {
1363                                 inode->i_size = new_isize;
1364                                 invalid |= NFS_INO_INVALID_DATA;
1365                         }
1366                         invalid |= NFS_INO_INVALID_ATTR;
1367                 } else if (new_isize > cur_isize) {
1368                         inode->i_size = new_isize;
1369                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1370                 }
1371                 dprintk("NFS: isize change on server for file %s/%ld\n",
1372                                 inode->i_sb->s_id, inode->i_ino);
1373         }
1374
1375         /* Check if the mtime agrees */
1376         if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1377                 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1378                 dprintk("NFS: mtime change on server for file %s/%ld\n",
1379                                 inode->i_sb->s_id, inode->i_ino);
1380                 if (!data_unstable)
1381                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1382         }
1383
1384         if ((fattr->valid & NFS_ATTR_FATTR_V4)
1385             && nfsi->change_attr != fattr->change_attr) {
1386                 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1387                        inode->i_sb->s_id, inode->i_ino);
1388                 nfsi->change_attr = fattr->change_attr;
1389                 if (!data_unstable)
1390                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1391         }
1392
1393         /* If ctime has changed we should definitely clear access+acl caches */
1394         if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1395                 if (!data_unstable)
1396                         invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1397                 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1398         }
1399         memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1400
1401         if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1402             inode->i_uid != fattr->uid ||
1403             inode->i_gid != fattr->gid)
1404                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1405
1406         inode->i_mode = fattr->mode;
1407         inode->i_nlink = fattr->nlink;
1408         inode->i_uid = fattr->uid;
1409         inode->i_gid = fattr->gid;
1410
1411         if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1412                 /*
1413                  * report the blocks in 512byte units
1414                  */
1415                 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1416                 inode->i_blksize = inode->i_sb->s_blocksize;
1417         } else {
1418                 inode->i_blocks = fattr->du.nfs2.blocks;
1419                 inode->i_blksize = fattr->du.nfs2.blocksize;
1420         }
1421
1422         /* Update attrtimeo value if we're out of the unstable period */
1423         if (invalid & NFS_INO_INVALID_ATTR) {
1424                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1425                 nfsi->attrtimeo_timestamp = jiffies;
1426         } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1427                 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1428                         nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1429                 nfsi->attrtimeo_timestamp = jiffies;
1430         }
1431         /* Don't invalidate the data if we were to blame */
1432         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1433                                 || S_ISLNK(inode->i_mode)))
1434                 invalid &= ~NFS_INO_INVALID_DATA;
1435         if (!nfs_have_delegation(inode, FMODE_READ))
1436                 nfsi->cache_validity |= invalid;
1437
1438         spin_unlock(&inode->i_lock);
1439         return 0;
1440  out_changed:
1441         /*
1442          * Big trouble! The inode has become a different object.
1443          */
1444 #ifdef NFS_PARANOIA
1445         printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1446                         __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1447 #endif
1448         /*
1449          * No need to worry about unhashing the dentry, as the
1450          * lookup validation will know that the inode is bad.
1451          * (But we fall through to invalidate the caches.)
1452          */
1453         nfs_invalidate_inode(inode);
1454  out_err:
1455         set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1456         return -ESTALE;
1457 }
1458
1459 /*
1460  * File system information
1461  */
1462
1463 static int nfs_set_super(struct super_block *s, void *data)
1464 {
1465         s->s_fs_info = data;
1466         return set_anon_super(s, data);
1467 }
1468  
1469 static int nfs_compare_super(struct super_block *sb, void *data)
1470 {
1471         struct nfs_server *server = data;
1472         struct nfs_server *old = NFS_SB(sb);
1473
1474         if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1475                 return 0;
1476         if (old->addr.sin_port != server->addr.sin_port)
1477                 return 0;
1478         return !nfs_compare_fh(&old->fh, &server->fh);
1479 }
1480
1481 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1482         int flags, const char *dev_name, void *raw_data)
1483 {
1484         int error;
1485         struct nfs_server *server = NULL;
1486         struct super_block *s;
1487         struct nfs_fh *root;
1488         struct nfs_mount_data *data = raw_data;
1489
1490         s = ERR_PTR(-EINVAL);
1491         if (data == NULL) {
1492                 dprintk("%s: missing data argument\n", __FUNCTION__);
1493                 goto out_err;
1494         }
1495         if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1496                 dprintk("%s: bad mount version\n", __FUNCTION__);
1497                 goto out_err;
1498         }
1499         switch (data->version) {
1500                 case 1:
1501                         data->namlen = 0;
1502                 case 2:
1503                         data->bsize  = 0;
1504                 case 3:
1505                         if (data->flags & NFS_MOUNT_VER3) {
1506                                 dprintk("%s: mount structure version %d does not support NFSv3\n",
1507                                                 __FUNCTION__,
1508                                                 data->version);
1509                                 goto out_err;
1510                         }
1511                         data->root.size = NFS2_FHSIZE;
1512                         memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1513                 case 4:
1514                         if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1515                                 dprintk("%s: mount structure version %d does not support strong security\n",
1516                                                 __FUNCTION__,
1517                                                 data->version);
1518                                 goto out_err;
1519                         }
1520                 case 5:
1521                         memset(data->context, 0, sizeof(data->context));
1522         }
1523 #ifndef CONFIG_NFS_V3
1524         /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1525         s = ERR_PTR(-EPROTONOSUPPORT);
1526         if (data->flags & NFS_MOUNT_VER3) {
1527                 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1528                 goto out_err;
1529         }
1530 #endif /* CONFIG_NFS_V3 */
1531
1532         s = ERR_PTR(-ENOMEM);
1533         server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1534         if (!server)
1535                 goto out_err;
1536         memset(server, 0, sizeof(struct nfs_server));
1537         /* Zero out the NFS state stuff */
1538         init_nfsv4_state(server);
1539         server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1540
1541         root = &server->fh;
1542         if (data->flags & NFS_MOUNT_VER3)
1543                 root->size = data->root.size;
1544         else
1545                 root->size = NFS2_FHSIZE;
1546         s = ERR_PTR(-EINVAL);
1547         if (root->size > sizeof(root->data)) {
1548                 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1549                 goto out_err;
1550         }
1551         memcpy(root->data, data->root.data, root->size);
1552
1553         /* We now require that the mount process passes the remote address */
1554         memcpy(&server->addr, &data->addr, sizeof(server->addr));
1555         if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1556                 dprintk("%s: mount program didn't pass remote address!\n",
1557                                 __FUNCTION__);
1558                 goto out_err;
1559         }
1560
1561         /* Fire up rpciod if not yet running */
1562         s = ERR_PTR(rpciod_up());
1563         if (IS_ERR(s)) {
1564                 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1565                                 __FUNCTION__, PTR_ERR(s));
1566                 goto out_err;
1567         }
1568
1569         s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1570         if (IS_ERR(s) || s->s_root)
1571                 goto out_rpciod_down;
1572
1573         s->s_flags = flags;
1574
1575         error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1576         if (error) {
1577                 up_write(&s->s_umount);
1578                 deactivate_super(s);
1579                 return ERR_PTR(error);
1580         }
1581         s->s_flags |= MS_ACTIVE;
1582         return s;
1583 out_rpciod_down:
1584         rpciod_down();
1585 out_err:
1586         kfree(server);
1587         return s;
1588 }
1589
1590 static void nfs_kill_super(struct super_block *s)
1591 {
1592         struct nfs_server *server = NFS_SB(s);
1593
1594         kill_anon_super(s);
1595
1596         if (!IS_ERR(server->client))
1597                 rpc_shutdown_client(server->client);
1598         if (!IS_ERR(server->client_sys))
1599                 rpc_shutdown_client(server->client_sys);
1600         if (!IS_ERR(server->client_acl))
1601                 rpc_shutdown_client(server->client_acl);
1602
1603         if (!(server->flags & NFS_MOUNT_NONLM))
1604                 lockd_down();   /* release rpc.lockd */
1605
1606         rpciod_down();          /* release rpciod */
1607
1608         if (server->hostname != NULL)
1609                 kfree(server->hostname);
1610         kfree(server);
1611 }
1612
1613 static struct file_system_type nfs_fs_type = {
1614         .owner          = THIS_MODULE,
1615         .name           = "nfs",
1616         .get_sb         = nfs_get_sb,
1617         .kill_sb        = nfs_kill_super,
1618         .fs_flags       = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1619 };
1620
1621 #ifdef CONFIG_NFS_V4
1622
1623 static void nfs4_clear_inode(struct inode *);
1624
1625
1626 static struct super_operations nfs4_sops = { 
1627         .alloc_inode    = nfs_alloc_inode,
1628         .destroy_inode  = nfs_destroy_inode,
1629         .write_inode    = nfs_write_inode,
1630         .delete_inode   = nfs_delete_inode,
1631         .statfs         = nfs_statfs,
1632         .clear_inode    = nfs4_clear_inode,
1633         .umount_begin   = nfs_umount_begin,
1634         .show_options   = nfs_show_options,
1635 };
1636
1637 /*
1638  * Clean out any remaining NFSv4 state that might be left over due
1639  * to open() calls that passed nfs_atomic_lookup, but failed to call
1640  * nfs_open().
1641  */
1642 static void nfs4_clear_inode(struct inode *inode)
1643 {
1644         struct nfs_inode *nfsi = NFS_I(inode);
1645
1646         /* If we are holding a delegation, return it! */
1647         if (nfsi->delegation != NULL)
1648                 nfs_inode_return_delegation(inode);
1649         /* First call standard NFS clear_inode() code */
1650         nfs_clear_inode(inode);
1651         /* Now clear out any remaining state */
1652         while (!list_empty(&nfsi->open_states)) {
1653                 struct nfs4_state *state;
1654                 
1655                 state = list_entry(nfsi->open_states.next,
1656                                 struct nfs4_state,
1657                                 inode_states);
1658                 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1659                                 __FUNCTION__,
1660                                 inode->i_sb->s_id,
1661                                 (long long)NFS_FILEID(inode),
1662                                 state);
1663                 BUG_ON(atomic_read(&state->count) != 1);
1664                 nfs4_close_state(state, state->state);
1665         }
1666 }
1667
1668
1669 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1670 {
1671         struct nfs_server *server;
1672         struct nfs4_client *clp = NULL;
1673         struct rpc_xprt *xprt = NULL;
1674         struct rpc_clnt *clnt = NULL;
1675         struct rpc_timeout timeparms;
1676         rpc_authflavor_t authflavour;
1677         int proto, err = -EIO;
1678
1679         sb->s_blocksize_bits = 0;
1680         sb->s_blocksize = 0;
1681         server = NFS_SB(sb);
1682         if (data->rsize != 0)
1683                 server->rsize = nfs_block_size(data->rsize, NULL);
1684         if (data->wsize != 0)
1685                 server->wsize = nfs_block_size(data->wsize, NULL);
1686         server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1687         server->caps = NFS_CAP_ATOMIC_OPEN;
1688
1689         server->acregmin = data->acregmin*HZ;
1690         server->acregmax = data->acregmax*HZ;
1691         server->acdirmin = data->acdirmin*HZ;
1692         server->acdirmax = data->acdirmax*HZ;
1693
1694         server->rpc_ops = &nfs_v4_clientops;
1695         /* Initialize timeout values */
1696
1697         timeparms.to_initval = data->timeo * HZ / 10;
1698         timeparms.to_retries = data->retrans;
1699         timeparms.to_exponential = 1;
1700         if (!timeparms.to_retries)
1701                 timeparms.to_retries = 5;
1702
1703         proto = data->proto;
1704         /* Which IP protocol do we use? */
1705         switch (proto) {
1706         case IPPROTO_TCP:
1707                 timeparms.to_maxval  = RPC_MAX_TCP_TIMEOUT;
1708                 if (!timeparms.to_initval)
1709                         timeparms.to_initval = 600 * HZ / 10;
1710                 break;
1711         case IPPROTO_UDP:
1712                 timeparms.to_maxval  = RPC_MAX_UDP_TIMEOUT;
1713                 if (!timeparms.to_initval)
1714                         timeparms.to_initval = 11 * HZ / 10;
1715                 break;
1716         default:
1717                 return -EINVAL;
1718         }
1719
1720         clp = nfs4_get_client(&server->addr.sin_addr);
1721         if (!clp) {
1722                 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1723                 return -EIO;
1724         }
1725
1726         /* Now create transport and client */
1727         authflavour = RPC_AUTH_UNIX;
1728         if (data->auth_flavourlen != 0) {
1729                 if (data->auth_flavourlen != 1) {
1730                         dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1731                                         __FUNCTION__, data->auth_flavourlen);
1732                         err = -EINVAL;
1733                         goto out_fail;
1734                 }
1735                 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1736                         err = -EFAULT;
1737                         goto out_fail;
1738                 }
1739         }
1740
1741         down_write(&clp->cl_sem);
1742         if (IS_ERR(clp->cl_rpcclient)) {
1743                 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1744                 if (IS_ERR(xprt)) {
1745                         up_write(&clp->cl_sem);
1746                         err = PTR_ERR(xprt);
1747                         dprintk("%s: cannot create RPC transport. Error = %d\n",
1748                                         __FUNCTION__, err);
1749                         goto out_fail;
1750                 }
1751                 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1752                                 server->rpc_ops->version, authflavour);
1753                 if (IS_ERR(clnt)) {
1754                         up_write(&clp->cl_sem);
1755                         err = PTR_ERR(clnt);
1756                         dprintk("%s: cannot create RPC client. Error = %d\n",
1757                                         __FUNCTION__, err);
1758                         goto out_fail;
1759                 }
1760                 clnt->cl_intr     = 1;
1761                 clnt->cl_softrtry = 1;
1762                 clnt->cl_chatty   = 1;
1763                 clp->cl_rpcclient = clnt;
1764                 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1765                 if (IS_ERR(clp->cl_cred)) {
1766                         up_write(&clp->cl_sem);
1767                         err = PTR_ERR(clp->cl_cred);
1768                         clp->cl_cred = NULL;
1769                         goto out_fail;
1770                 }
1771                 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1772                 nfs_idmap_new(clp);
1773         }
1774         if (list_empty(&clp->cl_superblocks)) {
1775                 err = nfs4_init_client(clp);
1776                 if (err != 0) {
1777                         up_write(&clp->cl_sem);
1778                         goto out_fail;
1779                 }
1780         }
1781         list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1782         clnt = rpc_clone_client(clp->cl_rpcclient);
1783         if (!IS_ERR(clnt))
1784                         server->nfs4_state = clp;
1785         up_write(&clp->cl_sem);
1786         clp = NULL;
1787
1788         if (IS_ERR(clnt)) {
1789                 err = PTR_ERR(clnt);
1790                 dprintk("%s: cannot create RPC client. Error = %d\n",
1791                                 __FUNCTION__, err);
1792                 return err;
1793         }
1794
1795         server->client    = clnt;
1796
1797         if (server->nfs4_state->cl_idmap == NULL) {
1798                 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1799                 return -ENOMEM;
1800         }
1801
1802         if (clnt->cl_auth->au_flavor != authflavour) {
1803                 struct rpc_auth *auth;
1804
1805                 auth = rpcauth_create(authflavour, clnt);
1806                 if (IS_ERR(auth)) {
1807                         dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1808                         return PTR_ERR(auth);
1809                 }
1810         }
1811
1812         sb->s_time_gran = 1;
1813
1814         sb->s_op = &nfs4_sops;
1815         err = nfs_sb_init(sb, authflavour);
1816         if (err == 0)
1817                 return 0;
1818 out_fail:
1819         if (clp)
1820                 nfs4_put_client(clp);
1821         return err;
1822 }
1823
1824 static int nfs4_compare_super(struct super_block *sb, void *data)
1825 {
1826         struct nfs_server *server = data;
1827         struct nfs_server *old = NFS_SB(sb);
1828
1829         if (strcmp(server->hostname, old->hostname) != 0)
1830                 return 0;
1831         if (strcmp(server->mnt_path, old->mnt_path) != 0)
1832                 return 0;
1833         return 1;
1834 }
1835
1836 static void *
1837 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1838 {
1839         void *p = NULL;
1840
1841         if (!src->len)
1842                 return ERR_PTR(-EINVAL);
1843         if (src->len < maxlen)
1844                 maxlen = src->len;
1845         if (dst == NULL) {
1846                 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1847                 if (p == NULL)
1848                         return ERR_PTR(-ENOMEM);
1849         }
1850         if (copy_from_user(dst, src->data, maxlen)) {
1851                 if (p != NULL)
1852                         kfree(p);
1853                 return ERR_PTR(-EFAULT);
1854         }
1855         dst[maxlen] = '\0';
1856         return dst;
1857 }
1858
1859 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1860         int flags, const char *dev_name, void *raw_data)
1861 {
1862         int error;
1863         struct nfs_server *server;
1864         struct super_block *s;
1865         struct nfs4_mount_data *data = raw_data;
1866         void *p;
1867
1868         if (data == NULL) {
1869                 dprintk("%s: missing data argument\n", __FUNCTION__);
1870                 return ERR_PTR(-EINVAL);
1871         }
1872         if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1873                 dprintk("%s: bad mount version\n", __FUNCTION__);
1874                 return ERR_PTR(-EINVAL);
1875         }
1876
1877         server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1878         if (!server)
1879                 return ERR_PTR(-ENOMEM);
1880         memset(server, 0, sizeof(struct nfs_server));
1881         /* Zero out the NFS state stuff */
1882         init_nfsv4_state(server);
1883         server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1884
1885         p = nfs_copy_user_string(NULL, &data->hostname, 256);
1886         if (IS_ERR(p))
1887                 goto out_err;
1888         server->hostname = p;
1889
1890         p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1891         if (IS_ERR(p))
1892                 goto out_err;
1893         server->mnt_path = p;
1894
1895         p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1896                         sizeof(server->ip_addr) - 1);
1897         if (IS_ERR(p))
1898                 goto out_err;
1899
1900         /* We now require that the mount process passes the remote address */
1901         if (data->host_addrlen != sizeof(server->addr)) {
1902                 s = ERR_PTR(-EINVAL);
1903                 goto out_free;
1904         }
1905         if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1906                 s = ERR_PTR(-EFAULT);
1907                 goto out_free;
1908         }
1909         if (server->addr.sin_family != AF_INET ||
1910             server->addr.sin_addr.s_addr == INADDR_ANY) {
1911                 dprintk("%s: mount program didn't pass remote IP address!\n",
1912                                 __FUNCTION__);
1913                 s = ERR_PTR(-EINVAL);
1914                 goto out_free;
1915         }
1916
1917         /* Fire up rpciod if not yet running */
1918         s = ERR_PTR(rpciod_up());
1919         if (IS_ERR(s)) {
1920                 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1921                                 __FUNCTION__, PTR_ERR(s));
1922                 goto out_free;
1923         }
1924
1925         s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1926
1927         if (IS_ERR(s) || s->s_root)
1928                 goto out_free;
1929
1930         s->s_flags = flags;
1931
1932         error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1933         if (error) {
1934                 up_write(&s->s_umount);
1935                 deactivate_super(s);
1936                 return ERR_PTR(error);
1937         }
1938         s->s_flags |= MS_ACTIVE;
1939         return s;
1940 out_err:
1941         s = (struct super_block *)p;
1942 out_free:
1943         if (server->mnt_path)
1944                 kfree(server->mnt_path);
1945         if (server->hostname)
1946                 kfree(server->hostname);
1947         kfree(server);
1948         return s;
1949 }
1950
1951 static void nfs4_kill_super(struct super_block *sb)
1952 {
1953         struct nfs_server *server = NFS_SB(sb);
1954
1955         nfs_return_all_delegations(sb);
1956         kill_anon_super(sb);
1957
1958         nfs4_renewd_prepare_shutdown(server);
1959
1960         if (server->client != NULL && !IS_ERR(server->client))
1961                 rpc_shutdown_client(server->client);
1962         rpciod_down();          /* release rpciod */
1963
1964         destroy_nfsv4_state(server);
1965
1966         if (server->hostname != NULL)
1967                 kfree(server->hostname);
1968         kfree(server);
1969 }
1970
1971 static struct file_system_type nfs4_fs_type = {
1972         .owner          = THIS_MODULE,
1973         .name           = "nfs4",
1974         .get_sb         = nfs4_get_sb,
1975         .kill_sb        = nfs4_kill_super,
1976         .fs_flags       = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1977 };
1978
1979 #define nfs4_init_once(nfsi) \
1980         do { \
1981                 INIT_LIST_HEAD(&(nfsi)->open_states); \
1982                 nfsi->delegation = NULL; \
1983                 nfsi->delegation_state = 0; \
1984                 init_rwsem(&nfsi->rwsem); \
1985         } while(0)
1986 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1987 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1988 #else
1989 #define nfs4_init_once(nfsi) \
1990         do { } while (0)
1991 #define register_nfs4fs() (0)
1992 #define unregister_nfs4fs()
1993 #endif
1994
1995 extern int nfs_init_nfspagecache(void);
1996 extern void nfs_destroy_nfspagecache(void);
1997 extern int nfs_init_readpagecache(void);
1998 extern void nfs_destroy_readpagecache(void);
1999 extern int nfs_init_writepagecache(void);
2000 extern void nfs_destroy_writepagecache(void);
2001 #ifdef CONFIG_NFS_DIRECTIO
2002 extern int nfs_init_directcache(void);
2003 extern void nfs_destroy_directcache(void);
2004 #endif
2005
2006 static kmem_cache_t * nfs_inode_cachep;
2007
2008 static struct inode *nfs_alloc_inode(struct super_block *sb)
2009 {
2010         struct nfs_inode *nfsi;
2011         nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2012         if (!nfsi)
2013                 return NULL;
2014         nfsi->flags = 0UL;
2015         nfsi->cache_validity = 0UL;
2016 #ifdef CONFIG_NFS_V3_ACL
2017         nfsi->acl_access = ERR_PTR(-EAGAIN);
2018         nfsi->acl_default = ERR_PTR(-EAGAIN);
2019 #endif
2020 #ifdef CONFIG_NFS_V4
2021         nfsi->nfs4_acl = NULL;
2022 #endif /* CONFIG_NFS_V4 */
2023         return &nfsi->vfs_inode;
2024 }
2025
2026 static void nfs_destroy_inode(struct inode *inode)
2027 {
2028         kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2029 }
2030
2031 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2032 {
2033         struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2034
2035         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2036             SLAB_CTOR_CONSTRUCTOR) {
2037                 inode_init_once(&nfsi->vfs_inode);
2038                 spin_lock_init(&nfsi->req_lock);
2039                 INIT_LIST_HEAD(&nfsi->dirty);
2040                 INIT_LIST_HEAD(&nfsi->commit);
2041                 INIT_LIST_HEAD(&nfsi->open_files);
2042                 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2043                 atomic_set(&nfsi->data_updates, 0);
2044                 nfsi->ndirty = 0;
2045                 nfsi->ncommit = 0;
2046                 nfsi->npages = 0;
2047                 nfs4_init_once(nfsi);
2048         }
2049 }
2050  
2051 static int nfs_init_inodecache(void)
2052 {
2053         nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2054                                              sizeof(struct nfs_inode),
2055                                              0, SLAB_RECLAIM_ACCOUNT,
2056                                              init_once, NULL);
2057         if (nfs_inode_cachep == NULL)
2058                 return -ENOMEM;
2059
2060         return 0;
2061 }
2062
2063 static void nfs_destroy_inodecache(void)
2064 {
2065         if (kmem_cache_destroy(nfs_inode_cachep))
2066                 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2067 }
2068
2069 /*
2070  * Initialize NFS
2071  */
2072 static int __init init_nfs_fs(void)
2073 {
2074         int err;
2075
2076         err = nfs_init_nfspagecache();
2077         if (err)
2078                 goto out4;
2079
2080         err = nfs_init_inodecache();
2081         if (err)
2082                 goto out3;
2083
2084         err = nfs_init_readpagecache();
2085         if (err)
2086                 goto out2;
2087
2088         err = nfs_init_writepagecache();
2089         if (err)
2090                 goto out1;
2091
2092 #ifdef CONFIG_NFS_DIRECTIO
2093         err = nfs_init_directcache();
2094         if (err)
2095                 goto out0;
2096 #endif
2097
2098 #ifdef CONFIG_PROC_FS
2099         rpc_proc_register(&nfs_rpcstat);
2100 #endif
2101         err = register_filesystem(&nfs_fs_type);
2102         if (err)
2103                 goto out;
2104         if ((err = register_nfs4fs()) != 0)
2105                 goto out;
2106         return 0;
2107 out:
2108 #ifdef CONFIG_PROC_FS
2109         rpc_proc_unregister("nfs");
2110 #endif
2111         nfs_destroy_writepagecache();
2112 #ifdef CONFIG_NFS_DIRECTIO
2113 out0:
2114         nfs_destroy_directcache();
2115 #endif
2116 out1:
2117         nfs_destroy_readpagecache();
2118 out2:
2119         nfs_destroy_inodecache();
2120 out3:
2121         nfs_destroy_nfspagecache();
2122 out4:
2123         return err;
2124 }
2125
2126 static void __exit exit_nfs_fs(void)
2127 {
2128 #ifdef CONFIG_NFS_DIRECTIO
2129         nfs_destroy_directcache();
2130 #endif
2131         nfs_destroy_writepagecache();
2132         nfs_destroy_readpagecache();
2133         nfs_destroy_inodecache();
2134         nfs_destroy_nfspagecache();
2135 #ifdef CONFIG_PROC_FS
2136         rpc_proc_unregister("nfs");
2137 #endif
2138         unregister_filesystem(&nfs_fs_type);
2139         unregister_nfs4fs();
2140 }
2141
2142 /* Not quite true; I just maintain it */
2143 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2144 MODULE_LICENSE("GPL");
2145
2146 module_init(init_nfs_fs)
2147 module_exit(exit_nfs_fs)