[PATCH] NFS: use atomic bitops to manipulate flags in nfsi->flags
[pandora-kernel.git] / fs / nfs / inode.c
1 /*
2  *  linux/fs/nfs/inode.c
3  *
4  *  Copyright (C) 1992  Rick Sladkey
5  *
6  *  nfs inode and superblock handling functions
7  *
8  *  Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some
9  *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
10  *
11  *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12  *  J.S.Peatfield@damtp.cam.ac.uk
13  *
14  */
15
16 #include <linux/config.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
19
20 #include <linux/time.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
38
39 #include <asm/system.h>
40 #include <asm/uaccess.h>
41
42 #include "nfs4_fs.h"
43 #include "delegation.h"
44
45 #define NFSDBG_FACILITY         NFSDBG_VFS
46 #define NFS_PARANOIA 1
47
48 /* Maximum number of readahead requests
49  * FIXME: this should really be a sysctl so that users may tune it to suit
50  *        their needs. People that do NFS over a slow network, might for
51  *        instance want to reduce it to something closer to 1 for improved
52  *        interactive response.
53  */
54 #define NFS_MAX_READAHEAD       (RPC_DEF_SLOT_TABLE - 1)
55
56 static void nfs_invalidate_inode(struct inode *);
57 static int nfs_update_inode(struct inode *, struct nfs_fattr *, unsigned long);
58
59 static struct inode *nfs_alloc_inode(struct super_block *sb);
60 static void nfs_destroy_inode(struct inode *);
61 static int nfs_write_inode(struct inode *,int);
62 static void nfs_delete_inode(struct inode *);
63 static void nfs_clear_inode(struct inode *);
64 static void nfs_umount_begin(struct super_block *);
65 static int  nfs_statfs(struct super_block *, struct kstatfs *);
66 static int  nfs_show_options(struct seq_file *, struct vfsmount *);
67 static void nfs_zap_acl_cache(struct inode *);
68
69 static struct rpc_program       nfs_program;
70
71 static struct super_operations nfs_sops = { 
72         .alloc_inode    = nfs_alloc_inode,
73         .destroy_inode  = nfs_destroy_inode,
74         .write_inode    = nfs_write_inode,
75         .delete_inode   = nfs_delete_inode,
76         .statfs         = nfs_statfs,
77         .clear_inode    = nfs_clear_inode,
78         .umount_begin   = nfs_umount_begin,
79         .show_options   = nfs_show_options,
80 };
81
82 /*
83  * RPC cruft for NFS
84  */
85 static struct rpc_stat          nfs_rpcstat = {
86         .program                = &nfs_program
87 };
88 static struct rpc_version *     nfs_version[] = {
89         NULL,
90         NULL,
91         &nfs_version2,
92 #if defined(CONFIG_NFS_V3)
93         &nfs_version3,
94 #elif defined(CONFIG_NFS_V4)
95         NULL,
96 #endif
97 #if defined(CONFIG_NFS_V4)
98         &nfs_version4,
99 #endif
100 };
101
102 static struct rpc_program       nfs_program = {
103         .name                   = "nfs",
104         .number                 = NFS_PROGRAM,
105         .nrvers                 = sizeof(nfs_version) / sizeof(nfs_version[0]),
106         .version                = nfs_version,
107         .stats                  = &nfs_rpcstat,
108         .pipe_dir_name          = "/nfs",
109 };
110
111 #ifdef CONFIG_NFS_V3_ACL
112 static struct rpc_stat          nfsacl_rpcstat = { &nfsacl_program };
113 static struct rpc_version *     nfsacl_version[] = {
114         [3]                     = &nfsacl_version3,
115 };
116
117 struct rpc_program              nfsacl_program = {
118         .name =                 "nfsacl",
119         .number =               NFS_ACL_PROGRAM,
120         .nrvers =               sizeof(nfsacl_version) / sizeof(nfsacl_version[0]),
121         .version =              nfsacl_version,
122         .stats =                &nfsacl_rpcstat,
123 };
124 #endif  /* CONFIG_NFS_V3_ACL */
125
126 static inline unsigned long
127 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
128 {
129         return nfs_fileid_to_ino_t(fattr->fileid);
130 }
131
132 static int
133 nfs_write_inode(struct inode *inode, int sync)
134 {
135         int flags = sync ? FLUSH_WAIT : 0;
136         int ret;
137
138         ret = nfs_commit_inode(inode, flags);
139         if (ret < 0)
140                 return ret;
141         return 0;
142 }
143
144 static void
145 nfs_delete_inode(struct inode * inode)
146 {
147         dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
148
149         nfs_wb_all(inode);
150         /*
151          * The following should never happen...
152          */
153         if (nfs_have_writebacks(inode)) {
154                 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
155         }
156
157         clear_inode(inode);
158 }
159
160 static void
161 nfs_clear_inode(struct inode *inode)
162 {
163         struct nfs_inode *nfsi = NFS_I(inode);
164         struct rpc_cred *cred;
165
166         nfs_wb_all(inode);
167         BUG_ON (!list_empty(&nfsi->open_files));
168         nfs_zap_acl_cache(inode);
169         cred = nfsi->cache_access.cred;
170         if (cred)
171                 put_rpccred(cred);
172         BUG_ON(atomic_read(&nfsi->data_updates) != 0);
173 }
174
175 void
176 nfs_umount_begin(struct super_block *sb)
177 {
178         struct rpc_clnt *rpc = NFS_SB(sb)->client;
179
180         /* -EIO all pending I/O */
181         if (!IS_ERR(rpc))
182                 rpc_killall_tasks(rpc);
183         rpc = NFS_SB(sb)->client_acl;
184         if (!IS_ERR(rpc))
185                 rpc_killall_tasks(rpc);
186 }
187
188
189 static inline unsigned long
190 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
191 {
192         /* make sure blocksize is a power of two */
193         if ((bsize & (bsize - 1)) || nrbitsp) {
194                 unsigned char   nrbits;
195
196                 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
197                         ;
198                 bsize = 1 << nrbits;
199                 if (nrbitsp)
200                         *nrbitsp = nrbits;
201         }
202
203         return bsize;
204 }
205
206 /*
207  * Calculate the number of 512byte blocks used.
208  */
209 static inline unsigned long
210 nfs_calc_block_size(u64 tsize)
211 {
212         loff_t used = (tsize + 511) >> 9;
213         return (used > ULONG_MAX) ? ULONG_MAX : used;
214 }
215
216 /*
217  * Compute and set NFS server blocksize
218  */
219 static inline unsigned long
220 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
221 {
222         if (bsize < 1024)
223                 bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
224         else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
225                 bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
226
227         return nfs_block_bits(bsize, nrbitsp);
228 }
229
230 /*
231  * Obtain the root inode of the file system.
232  */
233 static struct inode *
234 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
235 {
236         struct nfs_server       *server = NFS_SB(sb);
237         struct inode *rooti;
238         int                     error;
239
240         error = server->rpc_ops->getroot(server, rootfh, fsinfo);
241         if (error < 0) {
242                 dprintk("nfs_get_root: getattr error = %d\n", -error);
243                 return ERR_PTR(error);
244         }
245
246         rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
247         if (!rooti)
248                 return ERR_PTR(-ENOMEM);
249         return rooti;
250 }
251
252 /*
253  * Do NFS version-independent mount processing, and sanity checking
254  */
255 static int
256 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
257 {
258         struct nfs_server       *server;
259         struct inode            *root_inode;
260         struct nfs_fattr        fattr;
261         struct nfs_fsinfo       fsinfo = {
262                                         .fattr = &fattr,
263                                 };
264         struct nfs_pathconf pathinfo = {
265                         .fattr = &fattr,
266         };
267         int no_root_error = 0;
268         unsigned long max_rpc_payload;
269
270         /* We probably want something more informative here */
271         snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
272
273         server = NFS_SB(sb);
274
275         sb->s_magic      = NFS_SUPER_MAGIC;
276
277         root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
278         /* Did getting the root inode fail? */
279         if (IS_ERR(root_inode)) {
280                 no_root_error = PTR_ERR(root_inode);
281                 goto out_no_root;
282         }
283         sb->s_root = d_alloc_root(root_inode);
284         if (!sb->s_root) {
285                 no_root_error = -ENOMEM;
286                 goto out_no_root;
287         }
288         sb->s_root->d_op = server->rpc_ops->dentry_ops;
289
290         /* Get some general file system info */
291         if (server->namelen == 0 &&
292             server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
293                 server->namelen = pathinfo.max_namelen;
294         /* Work out a lot of parameters */
295         if (server->rsize == 0)
296                 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
297         if (server->wsize == 0)
298                 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
299
300         if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
301                 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
302         if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
303                 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
304
305         max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
306         if (server->rsize > max_rpc_payload)
307                 server->rsize = max_rpc_payload;
308         if (server->wsize > max_rpc_payload)
309                 server->wsize = max_rpc_payload;
310
311         server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
312         if (server->rpages > NFS_READ_MAXIOV) {
313                 server->rpages = NFS_READ_MAXIOV;
314                 server->rsize = server->rpages << PAGE_CACHE_SHIFT;
315         }
316
317         server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
318         if (server->wpages > NFS_WRITE_MAXIOV) {
319                 server->wpages = NFS_WRITE_MAXIOV;
320                 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
321         }
322
323         if (sb->s_blocksize == 0)
324                 sb->s_blocksize = nfs_block_bits(server->wsize,
325                                                          &sb->s_blocksize_bits);
326         server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
327
328         server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
329         if (server->dtsize > PAGE_CACHE_SIZE)
330                 server->dtsize = PAGE_CACHE_SIZE;
331         if (server->dtsize > server->rsize)
332                 server->dtsize = server->rsize;
333
334         if (server->flags & NFS_MOUNT_NOAC) {
335                 server->acregmin = server->acregmax = 0;
336                 server->acdirmin = server->acdirmax = 0;
337                 sb->s_flags |= MS_SYNCHRONOUS;
338         }
339         server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
340
341         sb->s_maxbytes = fsinfo.maxfilesize;
342         if (sb->s_maxbytes > MAX_LFS_FILESIZE) 
343                 sb->s_maxbytes = MAX_LFS_FILESIZE; 
344
345         server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
346         server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
347
348         /* We're airborne Set socket buffersize */
349         rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
350         return 0;
351         /* Yargs. It didn't work out. */
352 out_no_root:
353         dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
354         if (!IS_ERR(root_inode))
355                 iput(root_inode);
356         return no_root_error;
357 }
358
359 /*
360  * Create an RPC client handle.
361  */
362 static struct rpc_clnt *
363 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
364 {
365         struct rpc_timeout      timeparms;
366         struct rpc_xprt         *xprt = NULL;
367         struct rpc_clnt         *clnt = NULL;
368         int                     tcp   = (data->flags & NFS_MOUNT_TCP);
369
370         /* Initialize timeout values */
371         timeparms.to_initval = data->timeo * HZ / 10;
372         timeparms.to_retries = data->retrans;
373         timeparms.to_maxval  = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
374         timeparms.to_exponential = 1;
375
376         if (!timeparms.to_initval)
377                 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
378         if (!timeparms.to_retries)
379                 timeparms.to_retries = 5;
380
381         /* create transport and client */
382         xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
383                                  &server->addr, &timeparms);
384         if (IS_ERR(xprt)) {
385                 dprintk("%s: cannot create RPC transport. Error = %ld\n",
386                                 __FUNCTION__, PTR_ERR(xprt));
387                 return (struct rpc_clnt *)xprt;
388         }
389         clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
390                                  server->rpc_ops->version, data->pseudoflavor);
391         if (IS_ERR(clnt)) {
392                 dprintk("%s: cannot create RPC client. Error = %ld\n",
393                                 __FUNCTION__, PTR_ERR(xprt));
394                 goto out_fail;
395         }
396
397         clnt->cl_intr     = 1;
398         clnt->cl_softrtry = 1;
399         clnt->cl_chatty   = 1;
400
401         return clnt;
402
403 out_fail:
404         return clnt;
405 }
406
407 /*
408  * The way this works is that the mount process passes a structure
409  * in the data argument which contains the server's IP address
410  * and the root file handle obtained from the server's mount
411  * daemon. We stash these away in the private superblock fields.
412  */
413 static int
414 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
415 {
416         struct nfs_server       *server;
417         rpc_authflavor_t        authflavor;
418
419         server           = NFS_SB(sb);
420         sb->s_blocksize_bits = 0;
421         sb->s_blocksize = 0;
422         if (data->bsize)
423                 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
424         if (data->rsize)
425                 server->rsize = nfs_block_size(data->rsize, NULL);
426         if (data->wsize)
427                 server->wsize = nfs_block_size(data->wsize, NULL);
428         server->flags    = data->flags & NFS_MOUNT_FLAGMASK;
429
430         server->acregmin = data->acregmin*HZ;
431         server->acregmax = data->acregmax*HZ;
432         server->acdirmin = data->acdirmin*HZ;
433         server->acdirmax = data->acdirmax*HZ;
434
435         /* Start lockd here, before we might error out */
436         if (!(server->flags & NFS_MOUNT_NONLM))
437                 lockd_up();
438
439         server->namelen  = data->namlen;
440         server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
441         if (!server->hostname)
442                 return -ENOMEM;
443         strcpy(server->hostname, data->hostname);
444
445         /* Check NFS protocol revision and initialize RPC op vector
446          * and file handle pool. */
447 #ifdef CONFIG_NFS_V3
448         if (server->flags & NFS_MOUNT_VER3) {
449                 server->rpc_ops = &nfs_v3_clientops;
450                 server->caps |= NFS_CAP_READDIRPLUS;
451         } else {
452                 server->rpc_ops = &nfs_v2_clientops;
453         }
454 #else
455         server->rpc_ops = &nfs_v2_clientops;
456 #endif
457
458         /* Fill in pseudoflavor for mount version < 5 */
459         if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
460                 data->pseudoflavor = RPC_AUTH_UNIX;
461         authflavor = data->pseudoflavor;        /* save for sb_init() */
462         /* XXX maybe we want to add a server->pseudoflavor field */
463
464         /* Create RPC client handles */
465         server->client = nfs_create_client(server, data);
466         if (IS_ERR(server->client))
467                 return PTR_ERR(server->client);
468         /* RFC 2623, sec 2.3.2 */
469         if (authflavor != RPC_AUTH_UNIX) {
470                 struct rpc_auth *auth;
471
472                 server->client_sys = rpc_clone_client(server->client);
473                 if (IS_ERR(server->client_sys))
474                         return PTR_ERR(server->client_sys);
475                 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
476                 if (IS_ERR(auth))
477                         return PTR_ERR(auth);
478         } else {
479                 atomic_inc(&server->client->cl_count);
480                 server->client_sys = server->client;
481         }
482         if (server->flags & NFS_MOUNT_VER3) {
483 #ifdef CONFIG_NFS_V3_ACL
484                 if (!(server->flags & NFS_MOUNT_NOACL)) {
485                         server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
486                         /* No errors! Assume that Sun nfsacls are supported */
487                         if (!IS_ERR(server->client_acl))
488                                 server->caps |= NFS_CAP_ACLS;
489                 }
490 #else
491                 server->flags &= ~NFS_MOUNT_NOACL;
492 #endif /* CONFIG_NFS_V3_ACL */
493                 /*
494                  * The VFS shouldn't apply the umask to mode bits. We will
495                  * do so ourselves when necessary.
496                  */
497                 sb->s_flags |= MS_POSIXACL;
498                 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
499                         server->namelen = NFS3_MAXNAMLEN;
500                 sb->s_time_gran = 1;
501         } else {
502                 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
503                         server->namelen = NFS2_MAXNAMLEN;
504         }
505
506         sb->s_op = &nfs_sops;
507         return nfs_sb_init(sb, authflavor);
508 }
509
510 static int
511 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
512 {
513         struct nfs_server *server = NFS_SB(sb);
514         unsigned char blockbits;
515         unsigned long blockres;
516         struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
517         struct nfs_fattr fattr;
518         struct nfs_fsstat res = {
519                         .fattr = &fattr,
520         };
521         int error;
522
523         lock_kernel();
524
525         error = server->rpc_ops->statfs(server, rootfh, &res);
526         buf->f_type = NFS_SUPER_MAGIC;
527         if (error < 0)
528                 goto out_err;
529
530         /*
531          * Current versions of glibc do not correctly handle the
532          * case where f_frsize != f_bsize.  Eventually we want to
533          * report the value of wtmult in this field.
534          */
535         buf->f_frsize = sb->s_blocksize;
536
537         /*
538          * On most *nix systems, f_blocks, f_bfree, and f_bavail
539          * are reported in units of f_frsize.  Linux hasn't had
540          * an f_frsize field in its statfs struct until recently,
541          * thus historically Linux's sys_statfs reports these
542          * fields in units of f_bsize.
543          */
544         buf->f_bsize = sb->s_blocksize;
545         blockbits = sb->s_blocksize_bits;
546         blockres = (1 << blockbits) - 1;
547         buf->f_blocks = (res.tbytes + blockres) >> blockbits;
548         buf->f_bfree = (res.fbytes + blockres) >> blockbits;
549         buf->f_bavail = (res.abytes + blockres) >> blockbits;
550
551         buf->f_files = res.tfiles;
552         buf->f_ffree = res.afiles;
553
554         buf->f_namelen = server->namelen;
555  out:
556         unlock_kernel();
557
558         return 0;
559
560  out_err:
561         printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
562         buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
563         goto out;
564
565 }
566
567 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
568 {
569         static struct proc_nfs_info {
570                 int flag;
571                 char *str;
572                 char *nostr;
573         } nfs_info[] = {
574                 { NFS_MOUNT_SOFT, ",soft", ",hard" },
575                 { NFS_MOUNT_INTR, ",intr", "" },
576                 { NFS_MOUNT_POSIX, ",posix", "" },
577                 { NFS_MOUNT_TCP, ",tcp", ",udp" },
578                 { NFS_MOUNT_NOCTO, ",nocto", "" },
579                 { NFS_MOUNT_NOAC, ",noac", "" },
580                 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
581                 { NFS_MOUNT_NOACL, ",noacl", "" },
582                 { 0, NULL, NULL }
583         };
584         struct proc_nfs_info *nfs_infop;
585         struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
586
587         seq_printf(m, ",v%d", nfss->rpc_ops->version);
588         seq_printf(m, ",rsize=%d", nfss->rsize);
589         seq_printf(m, ",wsize=%d", nfss->wsize);
590         if (nfss->acregmin != 3*HZ)
591                 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
592         if (nfss->acregmax != 60*HZ)
593                 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
594         if (nfss->acdirmin != 30*HZ)
595                 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
596         if (nfss->acdirmax != 60*HZ)
597                 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
598         for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
599                 if (nfss->flags & nfs_infop->flag)
600                         seq_puts(m, nfs_infop->str);
601                 else
602                         seq_puts(m, nfs_infop->nostr);
603         }
604         seq_puts(m, ",addr=");
605         seq_escape(m, nfss->hostname, " \t\n\\");
606         return 0;
607 }
608
609 /*
610  * Invalidate the local caches
611  */
612 void
613 nfs_zap_caches(struct inode *inode)
614 {
615         struct nfs_inode *nfsi = NFS_I(inode);
616         int mode = inode->i_mode;
617
618         NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
619         NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
620
621         memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
622         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
623                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
624         else
625                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
626 }
627
628 static void nfs_zap_acl_cache(struct inode *inode)
629 {
630         void (*clear_acl_cache)(struct inode *);
631
632         clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
633         if (clear_acl_cache != NULL)
634                 clear_acl_cache(inode);
635         NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
636 }
637
638 /*
639  * Invalidate, but do not unhash, the inode
640  */
641 static void
642 nfs_invalidate_inode(struct inode *inode)
643 {
644         umode_t save_mode = inode->i_mode;
645
646         make_bad_inode(inode);
647         inode->i_mode = save_mode;
648         nfs_zap_caches(inode);
649 }
650
651 struct nfs_find_desc {
652         struct nfs_fh           *fh;
653         struct nfs_fattr        *fattr;
654 };
655
656 /*
657  * In NFSv3 we can have 64bit inode numbers. In order to support
658  * this, and re-exported directories (also seen in NFSv2)
659  * we are forced to allow 2 different inodes to have the same
660  * i_ino.
661  */
662 static int
663 nfs_find_actor(struct inode *inode, void *opaque)
664 {
665         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
666         struct nfs_fh           *fh = desc->fh;
667         struct nfs_fattr        *fattr = desc->fattr;
668
669         if (NFS_FILEID(inode) != fattr->fileid)
670                 return 0;
671         if (nfs_compare_fh(NFS_FH(inode), fh))
672                 return 0;
673         if (is_bad_inode(inode) || NFS_STALE(inode))
674                 return 0;
675         return 1;
676 }
677
678 static int
679 nfs_init_locked(struct inode *inode, void *opaque)
680 {
681         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
682         struct nfs_fattr        *fattr = desc->fattr;
683
684         NFS_FILEID(inode) = fattr->fileid;
685         nfs_copy_fh(NFS_FH(inode), desc->fh);
686         return 0;
687 }
688
689 /* Don't use READDIRPLUS on directories that we believe are too large */
690 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
691
692 /*
693  * This is our front-end to iget that looks up inodes by file handle
694  * instead of inode number.
695  */
696 struct inode *
697 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
698 {
699         struct nfs_find_desc desc = {
700                 .fh     = fh,
701                 .fattr  = fattr
702         };
703         struct inode *inode = NULL;
704         unsigned long hash;
705
706         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
707                 goto out_no_inode;
708
709         if (!fattr->nlink) {
710                 printk("NFS: Buggy server - nlink == 0!\n");
711                 goto out_no_inode;
712         }
713
714         hash = nfs_fattr_to_ino_t(fattr);
715
716         if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
717                 goto out_no_inode;
718
719         if (inode->i_state & I_NEW) {
720                 struct nfs_inode *nfsi = NFS_I(inode);
721
722                 /* We set i_ino for the few things that still rely on it,
723                  * such as stat(2) */
724                 inode->i_ino = hash;
725
726                 /* We can't support update_atime(), since the server will reset it */
727                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
728                 inode->i_mode = fattr->mode;
729                 /* Why so? Because we want revalidate for devices/FIFOs, and
730                  * that's precisely what we have in nfs_file_inode_operations.
731                  */
732                 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
733                 if (S_ISREG(inode->i_mode)) {
734                         inode->i_fop = &nfs_file_operations;
735                         inode->i_data.a_ops = &nfs_file_aops;
736                         inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
737                 } else if (S_ISDIR(inode->i_mode)) {
738                         inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
739                         inode->i_fop = &nfs_dir_operations;
740                         if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
741                             && fattr->size <= NFS_LIMIT_READDIRPLUS)
742                                 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
743                 } else if (S_ISLNK(inode->i_mode))
744                         inode->i_op = &nfs_symlink_inode_operations;
745                 else
746                         init_special_inode(inode, inode->i_mode, fattr->rdev);
747
748                 nfsi->read_cache_jiffies = fattr->timestamp;
749                 inode->i_atime = fattr->atime;
750                 inode->i_mtime = fattr->mtime;
751                 inode->i_ctime = fattr->ctime;
752                 if (fattr->valid & NFS_ATTR_FATTR_V4)
753                         nfsi->change_attr = fattr->change_attr;
754                 inode->i_size = nfs_size_to_loff_t(fattr->size);
755                 inode->i_nlink = fattr->nlink;
756                 inode->i_uid = fattr->uid;
757                 inode->i_gid = fattr->gid;
758                 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
759                         /*
760                          * report the blocks in 512byte units
761                          */
762                         inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
763                         inode->i_blksize = inode->i_sb->s_blocksize;
764                 } else {
765                         inode->i_blocks = fattr->du.nfs2.blocks;
766                         inode->i_blksize = fattr->du.nfs2.blocksize;
767                 }
768                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
769                 nfsi->attrtimeo_timestamp = jiffies;
770                 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
771                 nfsi->cache_access.cred = NULL;
772
773                 unlock_new_inode(inode);
774         } else
775                 nfs_refresh_inode(inode, fattr);
776         dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
777                 inode->i_sb->s_id,
778                 (long long)NFS_FILEID(inode),
779                 atomic_read(&inode->i_count));
780
781 out:
782         return inode;
783
784 out_no_inode:
785         printk("nfs_fhget: iget failed\n");
786         goto out;
787 }
788
789 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
790
791 int
792 nfs_setattr(struct dentry *dentry, struct iattr *attr)
793 {
794         struct inode *inode = dentry->d_inode;
795         struct nfs_fattr fattr;
796         int error;
797
798         if (attr->ia_valid & ATTR_SIZE) {
799                 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
800                         attr->ia_valid &= ~ATTR_SIZE;
801         }
802
803         /* Optimization: if the end result is no change, don't RPC */
804         attr->ia_valid &= NFS_VALID_ATTRS;
805         if (attr->ia_valid == 0)
806                 return 0;
807
808         lock_kernel();
809         nfs_begin_data_update(inode);
810         /* Write all dirty data if we're changing file permissions or size */
811         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
812                 if (filemap_fdatawrite(inode->i_mapping) == 0)
813                         filemap_fdatawait(inode->i_mapping);
814                 nfs_wb_all(inode);
815         }
816         error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
817         if (error == 0)
818                 nfs_refresh_inode(inode, &fattr);
819         nfs_end_data_update(inode);
820         unlock_kernel();
821         return error;
822 }
823
824 /**
825  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
826  * @inode: pointer to struct inode
827  * @attr: pointer to struct iattr
828  *
829  * Note: we do this in the *proc.c in order to ensure that
830  *       it works for things like exclusive creates too.
831  */
832 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
833 {
834         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
835                 if ((attr->ia_valid & ATTR_MODE) != 0) {
836                         int mode = attr->ia_mode & S_IALLUGO;
837                         mode |= inode->i_mode & ~S_IALLUGO;
838                         inode->i_mode = mode;
839                 }
840                 if ((attr->ia_valid & ATTR_UID) != 0)
841                         inode->i_uid = attr->ia_uid;
842                 if ((attr->ia_valid & ATTR_GID) != 0)
843                         inode->i_gid = attr->ia_gid;
844                 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
845         }
846         if ((attr->ia_valid & ATTR_SIZE) != 0) {
847                 inode->i_size = attr->ia_size;
848                 vmtruncate(inode, attr->ia_size);
849         }
850 }
851
852 static int nfs_wait_schedule(void *word)
853 {
854         if (signal_pending(current))
855                 return -ERESTARTSYS;
856         schedule();
857         return 0;
858 }
859
860 /*
861  * Wait for the inode to get unlocked.
862  */
863 static int nfs_wait_on_inode(struct inode *inode)
864 {
865         struct rpc_clnt *clnt = NFS_CLIENT(inode);
866         struct nfs_inode *nfsi = NFS_I(inode);
867         sigset_t oldmask;
868         int error;
869
870         atomic_inc(&inode->i_count);
871         rpc_clnt_sigmask(clnt, &oldmask);
872         error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
873                                         nfs_wait_schedule, TASK_INTERRUPTIBLE);
874         rpc_clnt_sigunmask(clnt, &oldmask);
875         iput(inode);
876
877         return error;
878 }
879
880 static void nfs_wake_up_inode(struct inode *inode)
881 {
882         struct nfs_inode *nfsi = NFS_I(inode);
883
884         clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
885         smp_mb__after_clear_bit();
886         wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
887 }
888
889 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
890 {
891         struct inode *inode = dentry->d_inode;
892         int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
893         int err;
894
895         if (__IS_FLG(inode, MS_NOATIME))
896                 need_atime = 0;
897         else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
898                 need_atime = 0;
899         /* We may force a getattr if the user cares about atime */
900         if (need_atime)
901                 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
902         else
903                 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
904         if (!err)
905                 generic_fillattr(inode, stat);
906         return err;
907 }
908
909 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
910 {
911         struct nfs_open_context *ctx;
912
913         ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
914         if (ctx != NULL) {
915                 atomic_set(&ctx->count, 1);
916                 ctx->dentry = dget(dentry);
917                 ctx->cred = get_rpccred(cred);
918                 ctx->state = NULL;
919                 ctx->lockowner = current->files;
920                 ctx->error = 0;
921                 ctx->dir_cookie = 0;
922         }
923         return ctx;
924 }
925
926 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
927 {
928         if (ctx != NULL)
929                 atomic_inc(&ctx->count);
930         return ctx;
931 }
932
933 void put_nfs_open_context(struct nfs_open_context *ctx)
934 {
935         if (atomic_dec_and_test(&ctx->count)) {
936                 if (!list_empty(&ctx->list)) {
937                         struct inode *inode = ctx->dentry->d_inode;
938                         spin_lock(&inode->i_lock);
939                         list_del(&ctx->list);
940                         spin_unlock(&inode->i_lock);
941                 }
942                 if (ctx->state != NULL)
943                         nfs4_close_state(ctx->state, ctx->mode);
944                 if (ctx->cred != NULL)
945                         put_rpccred(ctx->cred);
946                 dput(ctx->dentry);
947                 kfree(ctx);
948         }
949 }
950
951 /*
952  * Ensure that mmap has a recent RPC credential for use when writing out
953  * shared pages
954  */
955 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
956 {
957         struct inode *inode = filp->f_dentry->d_inode;
958         struct nfs_inode *nfsi = NFS_I(inode);
959
960         filp->private_data = get_nfs_open_context(ctx);
961         spin_lock(&inode->i_lock);
962         list_add(&ctx->list, &nfsi->open_files);
963         spin_unlock(&inode->i_lock);
964 }
965
966 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
967 {
968         struct nfs_inode *nfsi = NFS_I(inode);
969         struct nfs_open_context *pos, *ctx = NULL;
970
971         spin_lock(&inode->i_lock);
972         list_for_each_entry(pos, &nfsi->open_files, list) {
973                 if ((pos->mode & mode) == mode) {
974                         ctx = get_nfs_open_context(pos);
975                         break;
976                 }
977         }
978         spin_unlock(&inode->i_lock);
979         return ctx;
980 }
981
982 void nfs_file_clear_open_context(struct file *filp)
983 {
984         struct inode *inode = filp->f_dentry->d_inode;
985         struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
986
987         if (ctx) {
988                 filp->private_data = NULL;
989                 spin_lock(&inode->i_lock);
990                 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
991                 spin_unlock(&inode->i_lock);
992                 put_nfs_open_context(ctx);
993         }
994 }
995
996 /*
997  * These allocate and release file read/write context information.
998  */
999 int nfs_open(struct inode *inode, struct file *filp)
1000 {
1001         struct nfs_open_context *ctx;
1002         struct rpc_cred *cred;
1003
1004         cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1005         if (IS_ERR(cred))
1006                 return PTR_ERR(cred);
1007         ctx = alloc_nfs_open_context(filp->f_dentry, cred);
1008         put_rpccred(cred);
1009         if (ctx == NULL)
1010                 return -ENOMEM;
1011         ctx->mode = filp->f_mode;
1012         nfs_file_set_open_context(filp, ctx);
1013         put_nfs_open_context(ctx);
1014         if ((filp->f_mode & FMODE_WRITE) != 0)
1015                 nfs_begin_data_update(inode);
1016         return 0;
1017 }
1018
1019 int nfs_release(struct inode *inode, struct file *filp)
1020 {
1021         if ((filp->f_mode & FMODE_WRITE) != 0)
1022                 nfs_end_data_update(inode);
1023         nfs_file_clear_open_context(filp);
1024         return 0;
1025 }
1026
1027 /*
1028  * This function is called whenever some part of NFS notices that
1029  * the cached attributes have to be refreshed.
1030  */
1031 int
1032 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1033 {
1034         int              status = -ESTALE;
1035         struct nfs_fattr fattr;
1036         struct nfs_inode *nfsi = NFS_I(inode);
1037         unsigned long verifier;
1038         unsigned long cache_validity;
1039
1040         dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1041                 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1042
1043         lock_kernel();
1044         if (!inode || is_bad_inode(inode))
1045                 goto out_nowait;
1046         if (NFS_STALE(inode))
1047                 goto out_nowait;
1048
1049         status = nfs_wait_on_inode(inode);
1050         if (status < 0)
1051                 goto out;
1052         if (NFS_STALE(inode)) {
1053                 status = -ESTALE;
1054                 /* Do we trust the cached ESTALE? */
1055                 if (NFS_ATTRTIMEO(inode) != 0) {
1056                         if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1057                                 /* no */
1058                         } else
1059                                 goto out;
1060                 }
1061         }
1062
1063         /* Protect against RPC races by saving the change attribute */
1064         verifier = nfs_save_change_attribute(inode);
1065         status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1066         if (status != 0) {
1067                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1068                          inode->i_sb->s_id,
1069                          (long long)NFS_FILEID(inode), status);
1070                 if (status == -ESTALE) {
1071                         nfs_zap_caches(inode);
1072                         if (!S_ISDIR(inode->i_mode))
1073                                 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1074                 }
1075                 goto out;
1076         }
1077
1078         status = nfs_update_inode(inode, &fattr, verifier);
1079         if (status) {
1080                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1081                          inode->i_sb->s_id,
1082                          (long long)NFS_FILEID(inode), status);
1083                 goto out;
1084         }
1085         cache_validity = nfsi->cache_validity;
1086         nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1087
1088         /*
1089          * We may need to keep the attributes marked as invalid if
1090          * we raced with nfs_end_attr_update().
1091          */
1092         if (verifier == nfsi->cache_change_attribute)
1093                 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1094
1095         nfs_revalidate_mapping(inode, inode->i_mapping);
1096
1097         if (cache_validity & NFS_INO_INVALID_ACL)
1098                 nfs_zap_acl_cache(inode);
1099
1100         dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1101                 inode->i_sb->s_id,
1102                 (long long)NFS_FILEID(inode));
1103
1104  out:
1105         nfs_wake_up_inode(inode);
1106
1107  out_nowait:
1108         unlock_kernel();
1109         return status;
1110 }
1111
1112 int nfs_attribute_timeout(struct inode *inode)
1113 {
1114         struct nfs_inode *nfsi = NFS_I(inode);
1115
1116         if (nfs_have_delegation(inode, FMODE_READ))
1117                 return 0;
1118         return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1119 }
1120
1121 /**
1122  * nfs_revalidate_inode - Revalidate the inode attributes
1123  * @server - pointer to nfs_server struct
1124  * @inode - pointer to inode struct
1125  *
1126  * Updates inode attribute information by retrieving the data from the server.
1127  */
1128 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1129 {
1130         if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1131                         && !nfs_attribute_timeout(inode))
1132                 return NFS_STALE(inode) ? -ESTALE : 0;
1133         return __nfs_revalidate_inode(server, inode);
1134 }
1135
1136 /**
1137  * nfs_revalidate_mapping - Revalidate the pagecache
1138  * @inode - pointer to host inode
1139  * @mapping - pointer to mapping
1140  */
1141 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1142 {
1143         struct nfs_inode *nfsi = NFS_I(inode);
1144
1145         if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1146                 if (S_ISREG(inode->i_mode)) {
1147                         if (filemap_fdatawrite(mapping) == 0)
1148                                 filemap_fdatawait(mapping);
1149                         nfs_wb_all(inode);
1150                 }
1151                 invalidate_inode_pages2(mapping);
1152                 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1153                 if (S_ISDIR(inode->i_mode)) {
1154                         memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1155                         /* This ensures we revalidate child dentries */
1156                         nfsi->cache_change_attribute++;
1157                 }
1158                 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1159                                 inode->i_sb->s_id,
1160                                 (long long)NFS_FILEID(inode));
1161         }
1162 }
1163
1164 /**
1165  * nfs_begin_data_update
1166  * @inode - pointer to inode
1167  * Declare that a set of operations will update file data on the server
1168  */
1169 void nfs_begin_data_update(struct inode *inode)
1170 {
1171         atomic_inc(&NFS_I(inode)->data_updates);
1172 }
1173
1174 /**
1175  * nfs_end_data_update
1176  * @inode - pointer to inode
1177  * Declare end of the operations that will update file data
1178  * This will mark the inode as immediately needing revalidation
1179  * of its attribute cache.
1180  */
1181 void nfs_end_data_update(struct inode *inode)
1182 {
1183         struct nfs_inode *nfsi = NFS_I(inode);
1184
1185         if (!nfs_have_delegation(inode, FMODE_READ)) {
1186                 /* Mark the attribute cache for revalidation */
1187                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1188                 /* Directories and symlinks: invalidate page cache too */
1189                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1190                         nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1191         }
1192         nfsi->cache_change_attribute ++;
1193         atomic_dec(&nfsi->data_updates);
1194 }
1195
1196 /**
1197  * nfs_refresh_inode - verify consistency of the inode attribute cache
1198  * @inode - pointer to inode
1199  * @fattr - updated attributes
1200  *
1201  * Verifies the attribute cache. If we have just changed the attributes,
1202  * so that fattr carries weak cache consistency data, then it may
1203  * also update the ctime/mtime/change_attribute.
1204  */
1205 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1206 {
1207         struct nfs_inode *nfsi = NFS_I(inode);
1208         loff_t cur_size, new_isize;
1209         int data_unstable;
1210
1211         /* Do we hold a delegation? */
1212         if (nfs_have_delegation(inode, FMODE_READ))
1213                 return 0;
1214
1215         /* Are we in the process of updating data on the server? */
1216         data_unstable = nfs_caches_unstable(inode);
1217
1218         if (fattr->valid & NFS_ATTR_FATTR_V4) {
1219                 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1220                                 && nfsi->change_attr == fattr->pre_change_attr)
1221                         nfsi->change_attr = fattr->change_attr;
1222                 if (nfsi->change_attr != fattr->change_attr) {
1223                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1224                         if (!data_unstable)
1225                                 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1226                 }
1227         }
1228
1229         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1230                 return 0;
1231
1232         /* Has the inode gone and changed behind our back? */
1233         if (nfsi->fileid != fattr->fileid
1234                         || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1235                 return -EIO;
1236
1237         cur_size = i_size_read(inode);
1238         new_isize = nfs_size_to_loff_t(fattr->size);
1239
1240         /* If we have atomic WCC data, we may update some attributes */
1241         if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1242                 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1243                         memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1244                 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1245                         memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1246         }
1247
1248         /* Verify a few of the more important attributes */
1249         if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1250                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1251                 if (!data_unstable)
1252                         nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1253         }
1254         if (cur_size != new_isize) {
1255                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1256                 if (nfsi->npages == 0)
1257                         nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1258         }
1259
1260         /* Have any file permissions changed? */
1261         if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1262                         || inode->i_uid != fattr->uid
1263                         || inode->i_gid != fattr->gid)
1264                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1265
1266         /* Has the link count changed? */
1267         if (inode->i_nlink != fattr->nlink)
1268                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1269
1270         if (!timespec_equal(&inode->i_atime, &fattr->atime))
1271                 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1272
1273         nfsi->read_cache_jiffies = fattr->timestamp;
1274         return 0;
1275 }
1276
1277 /*
1278  * Many nfs protocol calls return the new file attributes after
1279  * an operation.  Here we update the inode to reflect the state
1280  * of the server's inode.
1281  *
1282  * This is a bit tricky because we have to make sure all dirty pages
1283  * have been sent off to the server before calling invalidate_inode_pages.
1284  * To make sure no other process adds more write requests while we try
1285  * our best to flush them, we make them sleep during the attribute refresh.
1286  *
1287  * A very similar scenario holds for the dir cache.
1288  */
1289 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1290 {
1291         struct nfs_inode *nfsi = NFS_I(inode);
1292         loff_t cur_isize, new_isize;
1293         unsigned int    invalid = 0;
1294         int data_unstable;
1295
1296         dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1297                         __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1298                         atomic_read(&inode->i_count), fattr->valid);
1299
1300         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1301                 return 0;
1302
1303         if (nfsi->fileid != fattr->fileid) {
1304                 printk(KERN_ERR "%s: inode number mismatch\n"
1305                        "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1306                        __FUNCTION__,
1307                        inode->i_sb->s_id, (long long)nfsi->fileid,
1308                        inode->i_sb->s_id, (long long)fattr->fileid);
1309                 goto out_err;
1310         }
1311
1312         /*
1313          * Make sure the inode's type hasn't changed.
1314          */
1315         if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1316                 goto out_changed;
1317
1318         /*
1319          * Update the read time so we don't revalidate too often.
1320          */
1321         nfsi->read_cache_jiffies = fattr->timestamp;
1322
1323         /* Are we racing with known updates of the metadata on the server? */
1324         data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1325
1326         /* Check if our cached file size is stale */
1327         new_isize = nfs_size_to_loff_t(fattr->size);
1328         cur_isize = i_size_read(inode);
1329         if (new_isize != cur_isize) {
1330                 /* Do we perhaps have any outstanding writes? */
1331                 if (nfsi->npages == 0) {
1332                         /* No, but did we race with nfs_end_data_update()? */
1333                         if (verifier  ==  nfsi->cache_change_attribute) {
1334                                 inode->i_size = new_isize;
1335                                 invalid |= NFS_INO_INVALID_DATA;
1336                         }
1337                         invalid |= NFS_INO_INVALID_ATTR;
1338                 } else if (new_isize > cur_isize) {
1339                         inode->i_size = new_isize;
1340                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1341                 }
1342                 dprintk("NFS: isize change on server for file %s/%ld\n",
1343                                 inode->i_sb->s_id, inode->i_ino);
1344         }
1345
1346         /* Check if the mtime agrees */
1347         if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1348                 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1349                 dprintk("NFS: mtime change on server for file %s/%ld\n",
1350                                 inode->i_sb->s_id, inode->i_ino);
1351                 if (!data_unstable)
1352                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1353         }
1354
1355         if ((fattr->valid & NFS_ATTR_FATTR_V4)
1356             && nfsi->change_attr != fattr->change_attr) {
1357                 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1358                        inode->i_sb->s_id, inode->i_ino);
1359                 nfsi->change_attr = fattr->change_attr;
1360                 if (!data_unstable)
1361                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1362         }
1363
1364         /* If ctime has changed we should definitely clear access+acl caches */
1365         if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1366                 if (!data_unstable)
1367                         invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1368                 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1369         }
1370         memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1371
1372         if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1373             inode->i_uid != fattr->uid ||
1374             inode->i_gid != fattr->gid)
1375                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1376
1377         inode->i_mode = fattr->mode;
1378         inode->i_nlink = fattr->nlink;
1379         inode->i_uid = fattr->uid;
1380         inode->i_gid = fattr->gid;
1381
1382         if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1383                 /*
1384                  * report the blocks in 512byte units
1385                  */
1386                 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1387                 inode->i_blksize = inode->i_sb->s_blocksize;
1388         } else {
1389                 inode->i_blocks = fattr->du.nfs2.blocks;
1390                 inode->i_blksize = fattr->du.nfs2.blocksize;
1391         }
1392
1393         /* Update attrtimeo value if we're out of the unstable period */
1394         if (invalid & NFS_INO_INVALID_ATTR) {
1395                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1396                 nfsi->attrtimeo_timestamp = jiffies;
1397         } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1398                 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1399                         nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1400                 nfsi->attrtimeo_timestamp = jiffies;
1401         }
1402         /* Don't invalidate the data if we were to blame */
1403         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1404                                 || S_ISLNK(inode->i_mode)))
1405                 invalid &= ~NFS_INO_INVALID_DATA;
1406         if (!nfs_have_delegation(inode, FMODE_READ))
1407                 nfsi->cache_validity |= invalid;
1408
1409         return 0;
1410  out_changed:
1411         /*
1412          * Big trouble! The inode has become a different object.
1413          */
1414 #ifdef NFS_PARANOIA
1415         printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1416                         __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1417 #endif
1418         /*
1419          * No need to worry about unhashing the dentry, as the
1420          * lookup validation will know that the inode is bad.
1421          * (But we fall through to invalidate the caches.)
1422          */
1423         nfs_invalidate_inode(inode);
1424  out_err:
1425         set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1426         return -ESTALE;
1427 }
1428
1429 /*
1430  * File system information
1431  */
1432
1433 static int nfs_set_super(struct super_block *s, void *data)
1434 {
1435         s->s_fs_info = data;
1436         return set_anon_super(s, data);
1437 }
1438  
1439 static int nfs_compare_super(struct super_block *sb, void *data)
1440 {
1441         struct nfs_server *server = data;
1442         struct nfs_server *old = NFS_SB(sb);
1443
1444         if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1445                 return 0;
1446         if (old->addr.sin_port != server->addr.sin_port)
1447                 return 0;
1448         return !nfs_compare_fh(&old->fh, &server->fh);
1449 }
1450
1451 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1452         int flags, const char *dev_name, void *raw_data)
1453 {
1454         int error;
1455         struct nfs_server *server = NULL;
1456         struct super_block *s;
1457         struct nfs_fh *root;
1458         struct nfs_mount_data *data = raw_data;
1459
1460         s = ERR_PTR(-EINVAL);
1461         if (data == NULL) {
1462                 dprintk("%s: missing data argument\n", __FUNCTION__);
1463                 goto out_err;
1464         }
1465         if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1466                 dprintk("%s: bad mount version\n", __FUNCTION__);
1467                 goto out_err;
1468         }
1469         switch (data->version) {
1470                 case 1:
1471                         data->namlen = 0;
1472                 case 2:
1473                         data->bsize  = 0;
1474                 case 3:
1475                         if (data->flags & NFS_MOUNT_VER3) {
1476                                 dprintk("%s: mount structure version %d does not support NFSv3\n",
1477                                                 __FUNCTION__,
1478                                                 data->version);
1479                                 goto out_err;
1480                         }
1481                         data->root.size = NFS2_FHSIZE;
1482                         memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1483                 case 4:
1484                         if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1485                                 dprintk("%s: mount structure version %d does not support strong security\n",
1486                                                 __FUNCTION__,
1487                                                 data->version);
1488                                 goto out_err;
1489                         }
1490                 case 5:
1491                         memset(data->context, 0, sizeof(data->context));
1492         }
1493 #ifndef CONFIG_NFS_V3
1494         /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1495         s = ERR_PTR(-EPROTONOSUPPORT);
1496         if (data->flags & NFS_MOUNT_VER3) {
1497                 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1498                 goto out_err;
1499         }
1500 #endif /* CONFIG_NFS_V3 */
1501
1502         s = ERR_PTR(-ENOMEM);
1503         server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1504         if (!server)
1505                 goto out_err;
1506         memset(server, 0, sizeof(struct nfs_server));
1507         /* Zero out the NFS state stuff */
1508         init_nfsv4_state(server);
1509         server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1510
1511         root = &server->fh;
1512         if (data->flags & NFS_MOUNT_VER3)
1513                 root->size = data->root.size;
1514         else
1515                 root->size = NFS2_FHSIZE;
1516         s = ERR_PTR(-EINVAL);
1517         if (root->size > sizeof(root->data)) {
1518                 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1519                 goto out_err;
1520         }
1521         memcpy(root->data, data->root.data, root->size);
1522
1523         /* We now require that the mount process passes the remote address */
1524         memcpy(&server->addr, &data->addr, sizeof(server->addr));
1525         if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1526                 dprintk("%s: mount program didn't pass remote address!\n",
1527                                 __FUNCTION__);
1528                 goto out_err;
1529         }
1530
1531         /* Fire up rpciod if not yet running */
1532         s = ERR_PTR(rpciod_up());
1533         if (IS_ERR(s)) {
1534                 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1535                                 __FUNCTION__, PTR_ERR(s));
1536                 goto out_err;
1537         }
1538
1539         s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1540         if (IS_ERR(s) || s->s_root)
1541                 goto out_rpciod_down;
1542
1543         s->s_flags = flags;
1544
1545         error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1546         if (error) {
1547                 up_write(&s->s_umount);
1548                 deactivate_super(s);
1549                 return ERR_PTR(error);
1550         }
1551         s->s_flags |= MS_ACTIVE;
1552         return s;
1553 out_rpciod_down:
1554         rpciod_down();
1555 out_err:
1556         kfree(server);
1557         return s;
1558 }
1559
1560 static void nfs_kill_super(struct super_block *s)
1561 {
1562         struct nfs_server *server = NFS_SB(s);
1563
1564         kill_anon_super(s);
1565
1566         if (!IS_ERR(server->client))
1567                 rpc_shutdown_client(server->client);
1568         if (!IS_ERR(server->client_sys))
1569                 rpc_shutdown_client(server->client_sys);
1570         if (!IS_ERR(server->client_acl))
1571                 rpc_shutdown_client(server->client_acl);
1572
1573         if (!(server->flags & NFS_MOUNT_NONLM))
1574                 lockd_down();   /* release rpc.lockd */
1575
1576         rpciod_down();          /* release rpciod */
1577
1578         if (server->hostname != NULL)
1579                 kfree(server->hostname);
1580         kfree(server);
1581 }
1582
1583 static struct file_system_type nfs_fs_type = {
1584         .owner          = THIS_MODULE,
1585         .name           = "nfs",
1586         .get_sb         = nfs_get_sb,
1587         .kill_sb        = nfs_kill_super,
1588         .fs_flags       = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1589 };
1590
1591 #ifdef CONFIG_NFS_V4
1592
1593 static void nfs4_clear_inode(struct inode *);
1594
1595
1596 static struct super_operations nfs4_sops = { 
1597         .alloc_inode    = nfs_alloc_inode,
1598         .destroy_inode  = nfs_destroy_inode,
1599         .write_inode    = nfs_write_inode,
1600         .delete_inode   = nfs_delete_inode,
1601         .statfs         = nfs_statfs,
1602         .clear_inode    = nfs4_clear_inode,
1603         .umount_begin   = nfs_umount_begin,
1604         .show_options   = nfs_show_options,
1605 };
1606
1607 /*
1608  * Clean out any remaining NFSv4 state that might be left over due
1609  * to open() calls that passed nfs_atomic_lookup, but failed to call
1610  * nfs_open().
1611  */
1612 static void nfs4_clear_inode(struct inode *inode)
1613 {
1614         struct nfs_inode *nfsi = NFS_I(inode);
1615
1616         /* If we are holding a delegation, return it! */
1617         if (nfsi->delegation != NULL)
1618                 nfs_inode_return_delegation(inode);
1619         /* First call standard NFS clear_inode() code */
1620         nfs_clear_inode(inode);
1621         /* Now clear out any remaining state */
1622         while (!list_empty(&nfsi->open_states)) {
1623                 struct nfs4_state *state;
1624                 
1625                 state = list_entry(nfsi->open_states.next,
1626                                 struct nfs4_state,
1627                                 inode_states);
1628                 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1629                                 __FUNCTION__,
1630                                 inode->i_sb->s_id,
1631                                 (long long)NFS_FILEID(inode),
1632                                 state);
1633                 BUG_ON(atomic_read(&state->count) != 1);
1634                 nfs4_close_state(state, state->state);
1635         }
1636 }
1637
1638
1639 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1640 {
1641         struct nfs_server *server;
1642         struct nfs4_client *clp = NULL;
1643         struct rpc_xprt *xprt = NULL;
1644         struct rpc_clnt *clnt = NULL;
1645         struct rpc_timeout timeparms;
1646         rpc_authflavor_t authflavour;
1647         int proto, err = -EIO;
1648
1649         sb->s_blocksize_bits = 0;
1650         sb->s_blocksize = 0;
1651         server = NFS_SB(sb);
1652         if (data->rsize != 0)
1653                 server->rsize = nfs_block_size(data->rsize, NULL);
1654         if (data->wsize != 0)
1655                 server->wsize = nfs_block_size(data->wsize, NULL);
1656         server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1657         server->caps = NFS_CAP_ATOMIC_OPEN;
1658
1659         server->acregmin = data->acregmin*HZ;
1660         server->acregmax = data->acregmax*HZ;
1661         server->acdirmin = data->acdirmin*HZ;
1662         server->acdirmax = data->acdirmax*HZ;
1663
1664         server->rpc_ops = &nfs_v4_clientops;
1665         /* Initialize timeout values */
1666
1667         timeparms.to_initval = data->timeo * HZ / 10;
1668         timeparms.to_retries = data->retrans;
1669         timeparms.to_exponential = 1;
1670         if (!timeparms.to_retries)
1671                 timeparms.to_retries = 5;
1672
1673         proto = data->proto;
1674         /* Which IP protocol do we use? */
1675         switch (proto) {
1676         case IPPROTO_TCP:
1677                 timeparms.to_maxval  = RPC_MAX_TCP_TIMEOUT;
1678                 if (!timeparms.to_initval)
1679                         timeparms.to_initval = 600 * HZ / 10;
1680                 break;
1681         case IPPROTO_UDP:
1682                 timeparms.to_maxval  = RPC_MAX_UDP_TIMEOUT;
1683                 if (!timeparms.to_initval)
1684                         timeparms.to_initval = 11 * HZ / 10;
1685                 break;
1686         default:
1687                 return -EINVAL;
1688         }
1689
1690         clp = nfs4_get_client(&server->addr.sin_addr);
1691         if (!clp) {
1692                 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1693                 return -EIO;
1694         }
1695
1696         /* Now create transport and client */
1697         authflavour = RPC_AUTH_UNIX;
1698         if (data->auth_flavourlen != 0) {
1699                 if (data->auth_flavourlen != 1) {
1700                         dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1701                                         __FUNCTION__, data->auth_flavourlen);
1702                         err = -EINVAL;
1703                         goto out_fail;
1704                 }
1705                 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1706                         err = -EFAULT;
1707                         goto out_fail;
1708                 }
1709         }
1710
1711         down_write(&clp->cl_sem);
1712         if (IS_ERR(clp->cl_rpcclient)) {
1713                 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1714                 if (IS_ERR(xprt)) {
1715                         up_write(&clp->cl_sem);
1716                         err = PTR_ERR(xprt);
1717                         dprintk("%s: cannot create RPC transport. Error = %d\n",
1718                                         __FUNCTION__, err);
1719                         goto out_fail;
1720                 }
1721                 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1722                                 server->rpc_ops->version, authflavour);
1723                 if (IS_ERR(clnt)) {
1724                         up_write(&clp->cl_sem);
1725                         err = PTR_ERR(clnt);
1726                         dprintk("%s: cannot create RPC client. Error = %d\n",
1727                                         __FUNCTION__, err);
1728                         goto out_fail;
1729                 }
1730                 clnt->cl_intr     = 1;
1731                 clnt->cl_softrtry = 1;
1732                 clnt->cl_chatty   = 1;
1733                 clp->cl_rpcclient = clnt;
1734                 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1735                 if (IS_ERR(clp->cl_cred)) {
1736                         up_write(&clp->cl_sem);
1737                         err = PTR_ERR(clp->cl_cred);
1738                         clp->cl_cred = NULL;
1739                         goto out_fail;
1740                 }
1741                 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1742                 nfs_idmap_new(clp);
1743         }
1744         if (list_empty(&clp->cl_superblocks)) {
1745                 err = nfs4_init_client(clp);
1746                 if (err != 0) {
1747                         up_write(&clp->cl_sem);
1748                         goto out_fail;
1749                 }
1750         }
1751         list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1752         clnt = rpc_clone_client(clp->cl_rpcclient);
1753         if (!IS_ERR(clnt))
1754                         server->nfs4_state = clp;
1755         up_write(&clp->cl_sem);
1756         clp = NULL;
1757
1758         if (IS_ERR(clnt)) {
1759                 err = PTR_ERR(clnt);
1760                 dprintk("%s: cannot create RPC client. Error = %d\n",
1761                                 __FUNCTION__, err);
1762                 return err;
1763         }
1764
1765         server->client    = clnt;
1766
1767         if (server->nfs4_state->cl_idmap == NULL) {
1768                 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1769                 return -ENOMEM;
1770         }
1771
1772         if (clnt->cl_auth->au_flavor != authflavour) {
1773                 struct rpc_auth *auth;
1774
1775                 auth = rpcauth_create(authflavour, clnt);
1776                 if (IS_ERR(auth)) {
1777                         dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1778                         return PTR_ERR(auth);
1779                 }
1780         }
1781
1782         sb->s_time_gran = 1;
1783
1784         sb->s_op = &nfs4_sops;
1785         err = nfs_sb_init(sb, authflavour);
1786         if (err == 0)
1787                 return 0;
1788 out_fail:
1789         if (clp)
1790                 nfs4_put_client(clp);
1791         return err;
1792 }
1793
1794 static int nfs4_compare_super(struct super_block *sb, void *data)
1795 {
1796         struct nfs_server *server = data;
1797         struct nfs_server *old = NFS_SB(sb);
1798
1799         if (strcmp(server->hostname, old->hostname) != 0)
1800                 return 0;
1801         if (strcmp(server->mnt_path, old->mnt_path) != 0)
1802                 return 0;
1803         return 1;
1804 }
1805
1806 static void *
1807 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1808 {
1809         void *p = NULL;
1810
1811         if (!src->len)
1812                 return ERR_PTR(-EINVAL);
1813         if (src->len < maxlen)
1814                 maxlen = src->len;
1815         if (dst == NULL) {
1816                 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1817                 if (p == NULL)
1818                         return ERR_PTR(-ENOMEM);
1819         }
1820         if (copy_from_user(dst, src->data, maxlen)) {
1821                 if (p != NULL)
1822                         kfree(p);
1823                 return ERR_PTR(-EFAULT);
1824         }
1825         dst[maxlen] = '\0';
1826         return dst;
1827 }
1828
1829 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1830         int flags, const char *dev_name, void *raw_data)
1831 {
1832         int error;
1833         struct nfs_server *server;
1834         struct super_block *s;
1835         struct nfs4_mount_data *data = raw_data;
1836         void *p;
1837
1838         if (data == NULL) {
1839                 dprintk("%s: missing data argument\n", __FUNCTION__);
1840                 return ERR_PTR(-EINVAL);
1841         }
1842         if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1843                 dprintk("%s: bad mount version\n", __FUNCTION__);
1844                 return ERR_PTR(-EINVAL);
1845         }
1846
1847         server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1848         if (!server)
1849                 return ERR_PTR(-ENOMEM);
1850         memset(server, 0, sizeof(struct nfs_server));
1851         /* Zero out the NFS state stuff */
1852         init_nfsv4_state(server);
1853         server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1854
1855         p = nfs_copy_user_string(NULL, &data->hostname, 256);
1856         if (IS_ERR(p))
1857                 goto out_err;
1858         server->hostname = p;
1859
1860         p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1861         if (IS_ERR(p))
1862                 goto out_err;
1863         server->mnt_path = p;
1864
1865         p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1866                         sizeof(server->ip_addr) - 1);
1867         if (IS_ERR(p))
1868                 goto out_err;
1869
1870         /* We now require that the mount process passes the remote address */
1871         if (data->host_addrlen != sizeof(server->addr)) {
1872                 s = ERR_PTR(-EINVAL);
1873                 goto out_free;
1874         }
1875         if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1876                 s = ERR_PTR(-EFAULT);
1877                 goto out_free;
1878         }
1879         if (server->addr.sin_family != AF_INET ||
1880             server->addr.sin_addr.s_addr == INADDR_ANY) {
1881                 dprintk("%s: mount program didn't pass remote IP address!\n",
1882                                 __FUNCTION__);
1883                 s = ERR_PTR(-EINVAL);
1884                 goto out_free;
1885         }
1886
1887         /* Fire up rpciod if not yet running */
1888         s = ERR_PTR(rpciod_up());
1889         if (IS_ERR(s)) {
1890                 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1891                                 __FUNCTION__, PTR_ERR(s));
1892                 goto out_free;
1893         }
1894
1895         s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1896
1897         if (IS_ERR(s) || s->s_root)
1898                 goto out_free;
1899
1900         s->s_flags = flags;
1901
1902         error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1903         if (error) {
1904                 up_write(&s->s_umount);
1905                 deactivate_super(s);
1906                 return ERR_PTR(error);
1907         }
1908         s->s_flags |= MS_ACTIVE;
1909         return s;
1910 out_err:
1911         s = (struct super_block *)p;
1912 out_free:
1913         if (server->mnt_path)
1914                 kfree(server->mnt_path);
1915         if (server->hostname)
1916                 kfree(server->hostname);
1917         kfree(server);
1918         return s;
1919 }
1920
1921 static void nfs4_kill_super(struct super_block *sb)
1922 {
1923         struct nfs_server *server = NFS_SB(sb);
1924
1925         nfs_return_all_delegations(sb);
1926         kill_anon_super(sb);
1927
1928         nfs4_renewd_prepare_shutdown(server);
1929
1930         if (server->client != NULL && !IS_ERR(server->client))
1931                 rpc_shutdown_client(server->client);
1932         rpciod_down();          /* release rpciod */
1933
1934         destroy_nfsv4_state(server);
1935
1936         if (server->hostname != NULL)
1937                 kfree(server->hostname);
1938         kfree(server);
1939 }
1940
1941 static struct file_system_type nfs4_fs_type = {
1942         .owner          = THIS_MODULE,
1943         .name           = "nfs4",
1944         .get_sb         = nfs4_get_sb,
1945         .kill_sb        = nfs4_kill_super,
1946         .fs_flags       = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1947 };
1948
1949 #define nfs4_init_once(nfsi) \
1950         do { \
1951                 INIT_LIST_HEAD(&(nfsi)->open_states); \
1952                 nfsi->delegation = NULL; \
1953                 nfsi->delegation_state = 0; \
1954                 init_rwsem(&nfsi->rwsem); \
1955         } while(0)
1956 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1957 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1958 #else
1959 #define nfs4_init_once(nfsi) \
1960         do { } while (0)
1961 #define register_nfs4fs() (0)
1962 #define unregister_nfs4fs()
1963 #endif
1964
1965 extern int nfs_init_nfspagecache(void);
1966 extern void nfs_destroy_nfspagecache(void);
1967 extern int nfs_init_readpagecache(void);
1968 extern void nfs_destroy_readpagecache(void);
1969 extern int nfs_init_writepagecache(void);
1970 extern void nfs_destroy_writepagecache(void);
1971 #ifdef CONFIG_NFS_DIRECTIO
1972 extern int nfs_init_directcache(void);
1973 extern void nfs_destroy_directcache(void);
1974 #endif
1975
1976 static kmem_cache_t * nfs_inode_cachep;
1977
1978 static struct inode *nfs_alloc_inode(struct super_block *sb)
1979 {
1980         struct nfs_inode *nfsi;
1981         nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
1982         if (!nfsi)
1983                 return NULL;
1984         nfsi->flags = 0UL;
1985         nfsi->cache_validity = 0UL;
1986 #ifdef CONFIG_NFS_V3_ACL
1987         nfsi->acl_access = ERR_PTR(-EAGAIN);
1988         nfsi->acl_default = ERR_PTR(-EAGAIN);
1989 #endif
1990 #ifdef CONFIG_NFS_V4
1991         nfsi->nfs4_acl = NULL;
1992 #endif /* CONFIG_NFS_V4 */
1993         return &nfsi->vfs_inode;
1994 }
1995
1996 static void nfs_destroy_inode(struct inode *inode)
1997 {
1998         kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1999 }
2000
2001 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2002 {
2003         struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2004
2005         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2006             SLAB_CTOR_CONSTRUCTOR) {
2007                 inode_init_once(&nfsi->vfs_inode);
2008                 spin_lock_init(&nfsi->req_lock);
2009                 INIT_LIST_HEAD(&nfsi->dirty);
2010                 INIT_LIST_HEAD(&nfsi->commit);
2011                 INIT_LIST_HEAD(&nfsi->open_files);
2012                 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2013                 atomic_set(&nfsi->data_updates, 0);
2014                 nfsi->ndirty = 0;
2015                 nfsi->ncommit = 0;
2016                 nfsi->npages = 0;
2017                 nfs4_init_once(nfsi);
2018         }
2019 }
2020  
2021 static int nfs_init_inodecache(void)
2022 {
2023         nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2024                                              sizeof(struct nfs_inode),
2025                                              0, SLAB_RECLAIM_ACCOUNT,
2026                                              init_once, NULL);
2027         if (nfs_inode_cachep == NULL)
2028                 return -ENOMEM;
2029
2030         return 0;
2031 }
2032
2033 static void nfs_destroy_inodecache(void)
2034 {
2035         if (kmem_cache_destroy(nfs_inode_cachep))
2036                 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2037 }
2038
2039 /*
2040  * Initialize NFS
2041  */
2042 static int __init init_nfs_fs(void)
2043 {
2044         int err;
2045
2046         err = nfs_init_nfspagecache();
2047         if (err)
2048                 goto out4;
2049
2050         err = nfs_init_inodecache();
2051         if (err)
2052                 goto out3;
2053
2054         err = nfs_init_readpagecache();
2055         if (err)
2056                 goto out2;
2057
2058         err = nfs_init_writepagecache();
2059         if (err)
2060                 goto out1;
2061
2062 #ifdef CONFIG_NFS_DIRECTIO
2063         err = nfs_init_directcache();
2064         if (err)
2065                 goto out0;
2066 #endif
2067
2068 #ifdef CONFIG_PROC_FS
2069         rpc_proc_register(&nfs_rpcstat);
2070 #endif
2071         err = register_filesystem(&nfs_fs_type);
2072         if (err)
2073                 goto out;
2074         if ((err = register_nfs4fs()) != 0)
2075                 goto out;
2076         return 0;
2077 out:
2078 #ifdef CONFIG_PROC_FS
2079         rpc_proc_unregister("nfs");
2080 #endif
2081         nfs_destroy_writepagecache();
2082 #ifdef CONFIG_NFS_DIRECTIO
2083 out0:
2084         nfs_destroy_directcache();
2085 #endif
2086 out1:
2087         nfs_destroy_readpagecache();
2088 out2:
2089         nfs_destroy_inodecache();
2090 out3:
2091         nfs_destroy_nfspagecache();
2092 out4:
2093         return err;
2094 }
2095
2096 static void __exit exit_nfs_fs(void)
2097 {
2098 #ifdef CONFIG_NFS_DIRECTIO
2099         nfs_destroy_directcache();
2100 #endif
2101         nfs_destroy_writepagecache();
2102         nfs_destroy_readpagecache();
2103         nfs_destroy_inodecache();
2104         nfs_destroy_nfspagecache();
2105 #ifdef CONFIG_PROC_FS
2106         rpc_proc_unregister("nfs");
2107 #endif
2108         unregister_filesystem(&nfs_fs_type);
2109         unregister_nfs4fs();
2110 }
2111
2112 /* Not quite true; I just maintain it */
2113 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2114 MODULE_LICENSE("GPL");
2115
2116 module_init(init_nfs_fs)
2117 module_exit(exit_nfs_fs)