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