Pull swiotlb-size into release branch
[linux-2.6] / fs / nfs / inode.c
1 /*
2  *  linux/fs/nfs/inode.c
3  *
4  *  Copyright (C) 1992  Rick Sladkey
5  *
6  *  nfs inode and superblock handling functions
7  *
8  *  Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some
9  *  experimental NFS changes. Modularisation taken straight from SYS5 fs.
10  *
11  *  Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12  *  J.S.Peatfield@damtp.cam.ac.uk
13  *
14  */
15
16 #include <linux/config.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
19
20 #include <linux/time.h>
21 #include <linux/kernel.h>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
38
39 #include <asm/system.h>
40 #include <asm/uaccess.h>
41
42 #include "nfs4_fs.h"
43 #include "delegation.h"
44
45 #define NFSDBG_FACILITY         NFSDBG_VFS
46 #define NFS_PARANOIA 1
47
48 /* Maximum number of readahead requests
49  * FIXME: this should really be a sysctl so that users may tune it to suit
50  *        their needs. People that do NFS over a slow network, might for
51  *        instance want to reduce it to something closer to 1 for improved
52  *        interactive response.
53  */
54 #define NFS_MAX_READAHEAD       (RPC_DEF_SLOT_TABLE - 1)
55
56 static void nfs_invalidate_inode(struct inode *);
57 static int nfs_update_inode(struct inode *, struct nfs_fattr *, unsigned long);
58
59 static struct inode *nfs_alloc_inode(struct super_block *sb);
60 static void nfs_destroy_inode(struct inode *);
61 static int nfs_write_inode(struct inode *,int);
62 static void nfs_delete_inode(struct inode *);
63 static void nfs_clear_inode(struct inode *);
64 static void nfs_umount_begin(struct super_block *);
65 static int  nfs_statfs(struct super_block *, struct kstatfs *);
66 static int  nfs_show_options(struct seq_file *, struct vfsmount *);
67 static void nfs_zap_acl_cache(struct inode *);
68
69 static struct rpc_program       nfs_program;
70
71 static struct super_operations nfs_sops = { 
72         .alloc_inode    = nfs_alloc_inode,
73         .destroy_inode  = nfs_destroy_inode,
74         .write_inode    = nfs_write_inode,
75         .delete_inode   = nfs_delete_inode,
76         .statfs         = nfs_statfs,
77         .clear_inode    = nfs_clear_inode,
78         .umount_begin   = nfs_umount_begin,
79         .show_options   = nfs_show_options,
80 };
81
82 /*
83  * RPC cruft for NFS
84  */
85 static struct rpc_stat          nfs_rpcstat = {
86         .program                = &nfs_program
87 };
88 static struct rpc_version *     nfs_version[] = {
89         NULL,
90         NULL,
91         &nfs_version2,
92 #if defined(CONFIG_NFS_V3)
93         &nfs_version3,
94 #elif defined(CONFIG_NFS_V4)
95         NULL,
96 #endif
97 #if defined(CONFIG_NFS_V4)
98         &nfs_version4,
99 #endif
100 };
101
102 static struct rpc_program       nfs_program = {
103         .name                   = "nfs",
104         .number                 = NFS_PROGRAM,
105         .nrvers                 = sizeof(nfs_version) / sizeof(nfs_version[0]),
106         .version                = nfs_version,
107         .stats                  = &nfs_rpcstat,
108         .pipe_dir_name          = "/nfs",
109 };
110
111 #ifdef CONFIG_NFS_V3_ACL
112 static struct rpc_stat          nfsacl_rpcstat = { &nfsacl_program };
113 static struct rpc_version *     nfsacl_version[] = {
114         [3]                     = &nfsacl_version3,
115 };
116
117 struct rpc_program              nfsacl_program = {
118         .name =                 "nfsacl",
119         .number =               NFS_ACL_PROGRAM,
120         .nrvers =               sizeof(nfsacl_version) / sizeof(nfsacl_version[0]),
121         .version =              nfsacl_version,
122         .stats =                &nfsacl_rpcstat,
123 };
124 #endif  /* CONFIG_NFS_V3_ACL */
125
126 static inline unsigned long
127 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
128 {
129         return nfs_fileid_to_ino_t(fattr->fileid);
130 }
131
132 static int
133 nfs_write_inode(struct inode *inode, int sync)
134 {
135         int flags = sync ? FLUSH_WAIT : 0;
136         int ret;
137
138         ret = nfs_commit_inode(inode, flags);
139         if (ret < 0)
140                 return ret;
141         return 0;
142 }
143
144 static void
145 nfs_delete_inode(struct inode * inode)
146 {
147         dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
148
149         nfs_wb_all(inode);
150         /*
151          * The following should never happen...
152          */
153         if (nfs_have_writebacks(inode)) {
154                 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
155         }
156
157         clear_inode(inode);
158 }
159
160 static void
161 nfs_clear_inode(struct inode *inode)
162 {
163         struct nfs_inode *nfsi = NFS_I(inode);
164         struct rpc_cred *cred;
165
166         nfs_wb_all(inode);
167         BUG_ON (!list_empty(&nfsi->open_files));
168         nfs_zap_acl_cache(inode);
169         cred = nfsi->cache_access.cred;
170         if (cred)
171                 put_rpccred(cred);
172         BUG_ON(atomic_read(&nfsi->data_updates) != 0);
173 }
174
175 void
176 nfs_umount_begin(struct super_block *sb)
177 {
178         struct rpc_clnt *rpc = NFS_SB(sb)->client;
179
180         /* -EIO all pending I/O */
181         if (!IS_ERR(rpc))
182                 rpc_killall_tasks(rpc);
183         rpc = NFS_SB(sb)->client_acl;
184         if (!IS_ERR(rpc))
185                 rpc_killall_tasks(rpc);
186 }
187
188
189 static inline unsigned long
190 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
191 {
192         /* make sure blocksize is a power of two */
193         if ((bsize & (bsize - 1)) || nrbitsp) {
194                 unsigned char   nrbits;
195
196                 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
197                         ;
198                 bsize = 1 << nrbits;
199                 if (nrbitsp)
200                         *nrbitsp = nrbits;
201         }
202
203         return bsize;
204 }
205
206 /*
207  * Calculate the number of 512byte blocks used.
208  */
209 static inline unsigned long
210 nfs_calc_block_size(u64 tsize)
211 {
212         loff_t used = (tsize + 511) >> 9;
213         return (used > ULONG_MAX) ? ULONG_MAX : used;
214 }
215
216 /*
217  * Compute and set NFS server blocksize
218  */
219 static inline unsigned long
220 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
221 {
222         if (bsize < 1024)
223                 bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
224         else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
225                 bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
226
227         return nfs_block_bits(bsize, nrbitsp);
228 }
229
230 /*
231  * Obtain the root inode of the file system.
232  */
233 static struct inode *
234 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
235 {
236         struct nfs_server       *server = NFS_SB(sb);
237         struct inode *rooti;
238         int                     error;
239
240         error = server->rpc_ops->getroot(server, rootfh, fsinfo);
241         if (error < 0) {
242                 dprintk("nfs_get_root: getattr error = %d\n", -error);
243                 return ERR_PTR(error);
244         }
245
246         rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
247         if (!rooti)
248                 return ERR_PTR(-ENOMEM);
249         return rooti;
250 }
251
252 /*
253  * Do NFS version-independent mount processing, and sanity checking
254  */
255 static int
256 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
257 {
258         struct nfs_server       *server;
259         struct inode            *root_inode;
260         struct nfs_fattr        fattr;
261         struct nfs_fsinfo       fsinfo = {
262                                         .fattr = &fattr,
263                                 };
264         struct nfs_pathconf pathinfo = {
265                         .fattr = &fattr,
266         };
267         int no_root_error = 0;
268         unsigned long max_rpc_payload;
269
270         /* We probably want something more informative here */
271         snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
272
273         server = NFS_SB(sb);
274
275         sb->s_magic      = NFS_SUPER_MAGIC;
276
277         root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
278         /* Did getting the root inode fail? */
279         if (IS_ERR(root_inode)) {
280                 no_root_error = PTR_ERR(root_inode);
281                 goto out_no_root;
282         }
283         sb->s_root = d_alloc_root(root_inode);
284         if (!sb->s_root) {
285                 no_root_error = -ENOMEM;
286                 goto out_no_root;
287         }
288         sb->s_root->d_op = server->rpc_ops->dentry_ops;
289
290         /* Get some general file system info */
291         if (server->namelen == 0 &&
292             server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
293                 server->namelen = pathinfo.max_namelen;
294         /* Work out a lot of parameters */
295         if (server->rsize == 0)
296                 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
297         if (server->wsize == 0)
298                 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
299
300         if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
301                 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
302         if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
303                 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
304
305         max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
306         if (server->rsize > max_rpc_payload)
307                 server->rsize = max_rpc_payload;
308         if (server->wsize > max_rpc_payload)
309                 server->wsize = max_rpc_payload;
310
311         server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
312         if (server->rpages > NFS_READ_MAXIOV) {
313                 server->rpages = NFS_READ_MAXIOV;
314                 server->rsize = server->rpages << PAGE_CACHE_SHIFT;
315         }
316
317         server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
318         if (server->wpages > NFS_WRITE_MAXIOV) {
319                 server->wpages = NFS_WRITE_MAXIOV;
320                 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
321         }
322
323         if (sb->s_blocksize == 0)
324                 sb->s_blocksize = nfs_block_bits(server->wsize,
325                                                          &sb->s_blocksize_bits);
326         server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
327
328         server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
329         if (server->dtsize > PAGE_CACHE_SIZE)
330                 server->dtsize = PAGE_CACHE_SIZE;
331         if (server->dtsize > server->rsize)
332                 server->dtsize = server->rsize;
333
334         if (server->flags & NFS_MOUNT_NOAC) {
335                 server->acregmin = server->acregmax = 0;
336                 server->acdirmin = server->acdirmax = 0;
337                 sb->s_flags |= MS_SYNCHRONOUS;
338         }
339         server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
340
341         sb->s_maxbytes = fsinfo.maxfilesize;
342         if (sb->s_maxbytes > MAX_LFS_FILESIZE) 
343                 sb->s_maxbytes = MAX_LFS_FILESIZE; 
344
345         server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
346         server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
347
348         /* We're airborne Set socket buffersize */
349         rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
350         return 0;
351         /* Yargs. It didn't work out. */
352 out_no_root:
353         dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
354         if (!IS_ERR(root_inode))
355                 iput(root_inode);
356         return no_root_error;
357 }
358
359 /*
360  * Create an RPC client handle.
361  */
362 static struct rpc_clnt *
363 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
364 {
365         struct rpc_timeout      timeparms;
366         struct rpc_xprt         *xprt = NULL;
367         struct rpc_clnt         *clnt = NULL;
368         int                     tcp   = (data->flags & NFS_MOUNT_TCP);
369
370         /* Initialize timeout values */
371         timeparms.to_initval = data->timeo * HZ / 10;
372         timeparms.to_retries = data->retrans;
373         timeparms.to_maxval  = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
374         timeparms.to_exponential = 1;
375
376         if (!timeparms.to_initval)
377                 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
378         if (!timeparms.to_retries)
379                 timeparms.to_retries = 5;
380
381         /* create transport and client */
382         xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
383                                  &server->addr, &timeparms);
384         if (IS_ERR(xprt)) {
385                 dprintk("%s: cannot create RPC transport. Error = %ld\n",
386                                 __FUNCTION__, PTR_ERR(xprt));
387                 return (struct rpc_clnt *)xprt;
388         }
389         clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
390                                  server->rpc_ops->version, data->pseudoflavor);
391         if (IS_ERR(clnt)) {
392                 dprintk("%s: cannot create RPC client. Error = %ld\n",
393                                 __FUNCTION__, PTR_ERR(xprt));
394                 goto out_fail;
395         }
396
397         clnt->cl_intr     = 1;
398         clnt->cl_softrtry = 1;
399         clnt->cl_chatty   = 1;
400
401         return clnt;
402
403 out_fail:
404         return clnt;
405 }
406
407 /*
408  * The way this works is that the mount process passes a structure
409  * in the data argument which contains the server's IP address
410  * and the root file handle obtained from the server's mount
411  * daemon. We stash these away in the private superblock fields.
412  */
413 static int
414 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
415 {
416         struct nfs_server       *server;
417         rpc_authflavor_t        authflavor;
418
419         server           = NFS_SB(sb);
420         sb->s_blocksize_bits = 0;
421         sb->s_blocksize = 0;
422         if (data->bsize)
423                 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
424         if (data->rsize)
425                 server->rsize = nfs_block_size(data->rsize, NULL);
426         if (data->wsize)
427                 server->wsize = nfs_block_size(data->wsize, NULL);
428         server->flags    = data->flags & NFS_MOUNT_FLAGMASK;
429
430         server->acregmin = data->acregmin*HZ;
431         server->acregmax = data->acregmax*HZ;
432         server->acdirmin = data->acdirmin*HZ;
433         server->acdirmax = data->acdirmax*HZ;
434
435         /* Start lockd here, before we might error out */
436         if (!(server->flags & NFS_MOUNT_NONLM))
437                 lockd_up();
438
439         server->namelen  = data->namlen;
440         server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
441         if (!server->hostname)
442                 return -ENOMEM;
443         strcpy(server->hostname, data->hostname);
444
445         /* Check NFS protocol revision and initialize RPC op vector
446          * and file handle pool. */
447 #ifdef CONFIG_NFS_V3
448         if (server->flags & NFS_MOUNT_VER3) {
449                 server->rpc_ops = &nfs_v3_clientops;
450                 server->caps |= NFS_CAP_READDIRPLUS;
451         } else {
452                 server->rpc_ops = &nfs_v2_clientops;
453         }
454 #else
455         server->rpc_ops = &nfs_v2_clientops;
456 #endif
457
458         /* Fill in pseudoflavor for mount version < 5 */
459         if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
460                 data->pseudoflavor = RPC_AUTH_UNIX;
461         authflavor = data->pseudoflavor;        /* save for sb_init() */
462         /* XXX maybe we want to add a server->pseudoflavor field */
463
464         /* Create RPC client handles */
465         server->client = nfs_create_client(server, data);
466         if (IS_ERR(server->client))
467                 return PTR_ERR(server->client);
468         /* RFC 2623, sec 2.3.2 */
469         if (authflavor != RPC_AUTH_UNIX) {
470                 struct rpc_auth *auth;
471
472                 server->client_sys = rpc_clone_client(server->client);
473                 if (IS_ERR(server->client_sys))
474                         return PTR_ERR(server->client_sys);
475                 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
476                 if (IS_ERR(auth))
477                         return PTR_ERR(auth);
478         } else {
479                 atomic_inc(&server->client->cl_count);
480                 server->client_sys = server->client;
481         }
482         if (server->flags & NFS_MOUNT_VER3) {
483 #ifdef CONFIG_NFS_V3_ACL
484                 if (!(server->flags & NFS_MOUNT_NOACL)) {
485                         server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
486                         /* No errors! Assume that Sun nfsacls are supported */
487                         if (!IS_ERR(server->client_acl))
488                                 server->caps |= NFS_CAP_ACLS;
489                 }
490 #else
491                 server->flags &= ~NFS_MOUNT_NOACL;
492 #endif /* CONFIG_NFS_V3_ACL */
493                 /*
494                  * The VFS shouldn't apply the umask to mode bits. We will
495                  * do so ourselves when necessary.
496                  */
497                 sb->s_flags |= MS_POSIXACL;
498                 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
499                         server->namelen = NFS3_MAXNAMLEN;
500                 sb->s_time_gran = 1;
501         } else {
502                 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
503                         server->namelen = NFS2_MAXNAMLEN;
504         }
505
506         sb->s_op = &nfs_sops;
507         return nfs_sb_init(sb, authflavor);
508 }
509
510 static int
511 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
512 {
513         struct nfs_server *server = NFS_SB(sb);
514         unsigned char blockbits;
515         unsigned long blockres;
516         struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
517         struct nfs_fattr fattr;
518         struct nfs_fsstat res = {
519                         .fattr = &fattr,
520         };
521         int error;
522
523         lock_kernel();
524
525         error = server->rpc_ops->statfs(server, rootfh, &res);
526         buf->f_type = NFS_SUPER_MAGIC;
527         if (error < 0)
528                 goto out_err;
529
530         /*
531          * Current versions of glibc do not correctly handle the
532          * case where f_frsize != f_bsize.  Eventually we want to
533          * report the value of wtmult in this field.
534          */
535         buf->f_frsize = sb->s_blocksize;
536
537         /*
538          * On most *nix systems, f_blocks, f_bfree, and f_bavail
539          * are reported in units of f_frsize.  Linux hasn't had
540          * an f_frsize field in its statfs struct until recently,
541          * thus historically Linux's sys_statfs reports these
542          * fields in units of f_bsize.
543          */
544         buf->f_bsize = sb->s_blocksize;
545         blockbits = sb->s_blocksize_bits;
546         blockres = (1 << blockbits) - 1;
547         buf->f_blocks = (res.tbytes + blockres) >> blockbits;
548         buf->f_bfree = (res.fbytes + blockres) >> blockbits;
549         buf->f_bavail = (res.abytes + blockres) >> blockbits;
550
551         buf->f_files = res.tfiles;
552         buf->f_ffree = res.afiles;
553
554         buf->f_namelen = server->namelen;
555  out:
556         unlock_kernel();
557
558         return 0;
559
560  out_err:
561         printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
562         buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
563         goto out;
564
565 }
566
567 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
568 {
569         static struct proc_nfs_info {
570                 int flag;
571                 char *str;
572                 char *nostr;
573         } nfs_info[] = {
574                 { NFS_MOUNT_SOFT, ",soft", ",hard" },
575                 { NFS_MOUNT_INTR, ",intr", "" },
576                 { NFS_MOUNT_POSIX, ",posix", "" },
577                 { NFS_MOUNT_TCP, ",tcp", ",udp" },
578                 { NFS_MOUNT_NOCTO, ",nocto", "" },
579                 { NFS_MOUNT_NOAC, ",noac", "" },
580                 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
581                 { NFS_MOUNT_NOACL, ",noacl", "" },
582                 { 0, NULL, NULL }
583         };
584         struct proc_nfs_info *nfs_infop;
585         struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
586
587         seq_printf(m, ",v%d", nfss->rpc_ops->version);
588         seq_printf(m, ",rsize=%d", nfss->rsize);
589         seq_printf(m, ",wsize=%d", nfss->wsize);
590         if (nfss->acregmin != 3*HZ)
591                 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
592         if (nfss->acregmax != 60*HZ)
593                 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
594         if (nfss->acdirmin != 30*HZ)
595                 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
596         if (nfss->acdirmax != 60*HZ)
597                 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
598         for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
599                 if (nfss->flags & nfs_infop->flag)
600                         seq_puts(m, nfs_infop->str);
601                 else
602                         seq_puts(m, nfs_infop->nostr);
603         }
604         seq_puts(m, ",addr=");
605         seq_escape(m, nfss->hostname, " \t\n\\");
606         return 0;
607 }
608
609 /*
610  * Invalidate the local caches
611  */
612 void
613 nfs_zap_caches(struct inode *inode)
614 {
615         struct nfs_inode *nfsi = NFS_I(inode);
616         int mode = inode->i_mode;
617
618         spin_lock(&inode->i_lock);
619
620         NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
621         NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
622
623         memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
624         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
625                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
626         else
627                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
628
629         spin_unlock(&inode->i_lock);
630 }
631
632 static void nfs_zap_acl_cache(struct inode *inode)
633 {
634         void (*clear_acl_cache)(struct inode *);
635
636         clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
637         if (clear_acl_cache != NULL)
638                 clear_acl_cache(inode);
639         spin_lock(&inode->i_lock);
640         NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
641         spin_unlock(&inode->i_lock);
642 }
643
644 /*
645  * Invalidate, but do not unhash, the inode
646  */
647 static void
648 nfs_invalidate_inode(struct inode *inode)
649 {
650         umode_t save_mode = inode->i_mode;
651
652         make_bad_inode(inode);
653         inode->i_mode = save_mode;
654         nfs_zap_caches(inode);
655 }
656
657 struct nfs_find_desc {
658         struct nfs_fh           *fh;
659         struct nfs_fattr        *fattr;
660 };
661
662 /*
663  * In NFSv3 we can have 64bit inode numbers. In order to support
664  * this, and re-exported directories (also seen in NFSv2)
665  * we are forced to allow 2 different inodes to have the same
666  * i_ino.
667  */
668 static int
669 nfs_find_actor(struct inode *inode, void *opaque)
670 {
671         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
672         struct nfs_fh           *fh = desc->fh;
673         struct nfs_fattr        *fattr = desc->fattr;
674
675         if (NFS_FILEID(inode) != fattr->fileid)
676                 return 0;
677         if (nfs_compare_fh(NFS_FH(inode), fh))
678                 return 0;
679         if (is_bad_inode(inode) || NFS_STALE(inode))
680                 return 0;
681         return 1;
682 }
683
684 static int
685 nfs_init_locked(struct inode *inode, void *opaque)
686 {
687         struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
688         struct nfs_fattr        *fattr = desc->fattr;
689
690         NFS_FILEID(inode) = fattr->fileid;
691         nfs_copy_fh(NFS_FH(inode), desc->fh);
692         return 0;
693 }
694
695 /* Don't use READDIRPLUS on directories that we believe are too large */
696 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
697
698 /*
699  * This is our front-end to iget that looks up inodes by file handle
700  * instead of inode number.
701  */
702 struct inode *
703 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
704 {
705         struct nfs_find_desc desc = {
706                 .fh     = fh,
707                 .fattr  = fattr
708         };
709         struct inode *inode = NULL;
710         unsigned long hash;
711
712         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
713                 goto out_no_inode;
714
715         if (!fattr->nlink) {
716                 printk("NFS: Buggy server - nlink == 0!\n");
717                 goto out_no_inode;
718         }
719
720         hash = nfs_fattr_to_ino_t(fattr);
721
722         if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
723                 goto out_no_inode;
724
725         if (inode->i_state & I_NEW) {
726                 struct nfs_inode *nfsi = NFS_I(inode);
727
728                 /* We set i_ino for the few things that still rely on it,
729                  * such as stat(2) */
730                 inode->i_ino = hash;
731
732                 /* We can't support update_atime(), since the server will reset it */
733                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
734                 inode->i_mode = fattr->mode;
735                 /* Why so? Because we want revalidate for devices/FIFOs, and
736                  * that's precisely what we have in nfs_file_inode_operations.
737                  */
738                 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
739                 if (S_ISREG(inode->i_mode)) {
740                         inode->i_fop = &nfs_file_operations;
741                         inode->i_data.a_ops = &nfs_file_aops;
742                         inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
743                 } else if (S_ISDIR(inode->i_mode)) {
744                         inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
745                         inode->i_fop = &nfs_dir_operations;
746                         if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
747                             && fattr->size <= NFS_LIMIT_READDIRPLUS)
748                                 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
749                 } else if (S_ISLNK(inode->i_mode))
750                         inode->i_op = &nfs_symlink_inode_operations;
751                 else
752                         init_special_inode(inode, inode->i_mode, fattr->rdev);
753
754                 nfsi->read_cache_jiffies = fattr->timestamp;
755                 inode->i_atime = fattr->atime;
756                 inode->i_mtime = fattr->mtime;
757                 inode->i_ctime = fattr->ctime;
758                 if (fattr->valid & NFS_ATTR_FATTR_V4)
759                         nfsi->change_attr = fattr->change_attr;
760                 inode->i_size = nfs_size_to_loff_t(fattr->size);
761                 inode->i_nlink = fattr->nlink;
762                 inode->i_uid = fattr->uid;
763                 inode->i_gid = fattr->gid;
764                 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
765                         /*
766                          * report the blocks in 512byte units
767                          */
768                         inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
769                         inode->i_blksize = inode->i_sb->s_blocksize;
770                 } else {
771                         inode->i_blocks = fattr->du.nfs2.blocks;
772                         inode->i_blksize = fattr->du.nfs2.blocksize;
773                 }
774                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
775                 nfsi->attrtimeo_timestamp = jiffies;
776                 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
777                 nfsi->cache_access.cred = NULL;
778
779                 unlock_new_inode(inode);
780         } else
781                 nfs_refresh_inode(inode, fattr);
782         dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
783                 inode->i_sb->s_id,
784                 (long long)NFS_FILEID(inode),
785                 atomic_read(&inode->i_count));
786
787 out:
788         return inode;
789
790 out_no_inode:
791         printk("nfs_fhget: iget failed\n");
792         goto out;
793 }
794
795 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
796
797 int
798 nfs_setattr(struct dentry *dentry, struct iattr *attr)
799 {
800         struct inode *inode = dentry->d_inode;
801         struct nfs_fattr fattr;
802         int error;
803
804         if (attr->ia_valid & ATTR_SIZE) {
805                 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
806                         attr->ia_valid &= ~ATTR_SIZE;
807         }
808
809         /* Optimization: if the end result is no change, don't RPC */
810         attr->ia_valid &= NFS_VALID_ATTRS;
811         if (attr->ia_valid == 0)
812                 return 0;
813
814         lock_kernel();
815         nfs_begin_data_update(inode);
816         /* Write all dirty data if we're changing file permissions or size */
817         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
818                 if (filemap_fdatawrite(inode->i_mapping) == 0)
819                         filemap_fdatawait(inode->i_mapping);
820                 nfs_wb_all(inode);
821         }
822         error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
823         if (error == 0)
824                 nfs_refresh_inode(inode, &fattr);
825         nfs_end_data_update(inode);
826         unlock_kernel();
827         return error;
828 }
829
830 /**
831  * nfs_setattr_update_inode - Update inode metadata after a setattr call.
832  * @inode: pointer to struct inode
833  * @attr: pointer to struct iattr
834  *
835  * Note: we do this in the *proc.c in order to ensure that
836  *       it works for things like exclusive creates too.
837  */
838 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
839 {
840         if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
841                 if ((attr->ia_valid & ATTR_MODE) != 0) {
842                         int mode = attr->ia_mode & S_IALLUGO;
843                         mode |= inode->i_mode & ~S_IALLUGO;
844                         inode->i_mode = mode;
845                 }
846                 if ((attr->ia_valid & ATTR_UID) != 0)
847                         inode->i_uid = attr->ia_uid;
848                 if ((attr->ia_valid & ATTR_GID) != 0)
849                         inode->i_gid = attr->ia_gid;
850                 spin_lock(&inode->i_lock);
851                 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
852                 spin_unlock(&inode->i_lock);
853         }
854         if ((attr->ia_valid & ATTR_SIZE) != 0) {
855                 inode->i_size = attr->ia_size;
856                 vmtruncate(inode, attr->ia_size);
857         }
858 }
859
860 static int nfs_wait_schedule(void *word)
861 {
862         if (signal_pending(current))
863                 return -ERESTARTSYS;
864         schedule();
865         return 0;
866 }
867
868 /*
869  * Wait for the inode to get unlocked.
870  */
871 static int nfs_wait_on_inode(struct inode *inode)
872 {
873         struct rpc_clnt *clnt = NFS_CLIENT(inode);
874         struct nfs_inode *nfsi = NFS_I(inode);
875         sigset_t oldmask;
876         int error;
877
878         atomic_inc(&inode->i_count);
879         rpc_clnt_sigmask(clnt, &oldmask);
880         error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
881                                         nfs_wait_schedule, TASK_INTERRUPTIBLE);
882         rpc_clnt_sigunmask(clnt, &oldmask);
883         iput(inode);
884
885         return error;
886 }
887
888 static void nfs_wake_up_inode(struct inode *inode)
889 {
890         struct nfs_inode *nfsi = NFS_I(inode);
891
892         clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
893         smp_mb__after_clear_bit();
894         wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
895 }
896
897 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
898 {
899         struct inode *inode = dentry->d_inode;
900         int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
901         int err;
902
903         if (__IS_FLG(inode, MS_NOATIME))
904                 need_atime = 0;
905         else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
906                 need_atime = 0;
907         /* We may force a getattr if the user cares about atime */
908         if (need_atime)
909                 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
910         else
911                 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
912         if (!err)
913                 generic_fillattr(inode, stat);
914         return err;
915 }
916
917 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
918 {
919         struct nfs_open_context *ctx;
920
921         ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
922         if (ctx != NULL) {
923                 atomic_set(&ctx->count, 1);
924                 ctx->dentry = dget(dentry);
925                 ctx->cred = get_rpccred(cred);
926                 ctx->state = NULL;
927                 ctx->lockowner = current->files;
928                 ctx->error = 0;
929                 ctx->dir_cookie = 0;
930         }
931         return ctx;
932 }
933
934 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
935 {
936         if (ctx != NULL)
937                 atomic_inc(&ctx->count);
938         return ctx;
939 }
940
941 void put_nfs_open_context(struct nfs_open_context *ctx)
942 {
943         if (atomic_dec_and_test(&ctx->count)) {
944                 if (!list_empty(&ctx->list)) {
945                         struct inode *inode = ctx->dentry->d_inode;
946                         spin_lock(&inode->i_lock);
947                         list_del(&ctx->list);
948                         spin_unlock(&inode->i_lock);
949                 }
950                 if (ctx->state != NULL)
951                         nfs4_close_state(ctx->state, ctx->mode);
952                 if (ctx->cred != NULL)
953                         put_rpccred(ctx->cred);
954                 dput(ctx->dentry);
955                 kfree(ctx);
956         }
957 }
958
959 /*
960  * Ensure that mmap has a recent RPC credential for use when writing out
961  * shared pages
962  */
963 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
964 {
965         struct inode *inode = filp->f_dentry->d_inode;
966         struct nfs_inode *nfsi = NFS_I(inode);
967
968         filp->private_data = get_nfs_open_context(ctx);
969         spin_lock(&inode->i_lock);
970         list_add(&ctx->list, &nfsi->open_files);
971         spin_unlock(&inode->i_lock);
972 }
973
974 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
975 {
976         struct nfs_inode *nfsi = NFS_I(inode);
977         struct nfs_open_context *pos, *ctx = NULL;
978
979         spin_lock(&inode->i_lock);
980         list_for_each_entry(pos, &nfsi->open_files, list) {
981                 if ((pos->mode & mode) == mode) {
982                         ctx = get_nfs_open_context(pos);
983                         break;
984                 }
985         }
986         spin_unlock(&inode->i_lock);
987         return ctx;
988 }
989
990 void nfs_file_clear_open_context(struct file *filp)
991 {
992         struct inode *inode = filp->f_dentry->d_inode;
993         struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
994
995         if (ctx) {
996                 filp->private_data = NULL;
997                 spin_lock(&inode->i_lock);
998                 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
999                 spin_unlock(&inode->i_lock);
1000                 put_nfs_open_context(ctx);
1001         }
1002 }
1003
1004 /*
1005  * These allocate and release file read/write context information.
1006  */
1007 int nfs_open(struct inode *inode, struct file *filp)
1008 {
1009         struct nfs_open_context *ctx;
1010         struct rpc_cred *cred;
1011
1012         cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1013         if (IS_ERR(cred))
1014                 return PTR_ERR(cred);
1015         ctx = alloc_nfs_open_context(filp->f_dentry, cred);
1016         put_rpccred(cred);
1017         if (ctx == NULL)
1018                 return -ENOMEM;
1019         ctx->mode = filp->f_mode;
1020         nfs_file_set_open_context(filp, ctx);
1021         put_nfs_open_context(ctx);
1022         if ((filp->f_mode & FMODE_WRITE) != 0)
1023                 nfs_begin_data_update(inode);
1024         return 0;
1025 }
1026
1027 int nfs_release(struct inode *inode, struct file *filp)
1028 {
1029         if ((filp->f_mode & FMODE_WRITE) != 0)
1030                 nfs_end_data_update(inode);
1031         nfs_file_clear_open_context(filp);
1032         return 0;
1033 }
1034
1035 /*
1036  * This function is called whenever some part of NFS notices that
1037  * the cached attributes have to be refreshed.
1038  */
1039 int
1040 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1041 {
1042         int              status = -ESTALE;
1043         struct nfs_fattr fattr;
1044         struct nfs_inode *nfsi = NFS_I(inode);
1045         unsigned long verifier;
1046         unsigned long cache_validity;
1047
1048         dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1049                 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1050
1051         lock_kernel();
1052         if (!inode || is_bad_inode(inode))
1053                 goto out_nowait;
1054         if (NFS_STALE(inode))
1055                 goto out_nowait;
1056
1057         status = nfs_wait_on_inode(inode);
1058         if (status < 0)
1059                 goto out;
1060         if (NFS_STALE(inode)) {
1061                 status = -ESTALE;
1062                 /* Do we trust the cached ESTALE? */
1063                 if (NFS_ATTRTIMEO(inode) != 0) {
1064                         if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1065                                 /* no */
1066                         } else
1067                                 goto out;
1068                 }
1069         }
1070
1071         /* Protect against RPC races by saving the change attribute */
1072         verifier = nfs_save_change_attribute(inode);
1073         status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1074         if (status != 0) {
1075                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1076                          inode->i_sb->s_id,
1077                          (long long)NFS_FILEID(inode), status);
1078                 if (status == -ESTALE) {
1079                         nfs_zap_caches(inode);
1080                         if (!S_ISDIR(inode->i_mode))
1081                                 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1082                 }
1083                 goto out;
1084         }
1085
1086         status = nfs_update_inode(inode, &fattr, verifier);
1087         if (status) {
1088                 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1089                          inode->i_sb->s_id,
1090                          (long long)NFS_FILEID(inode), status);
1091                 goto out;
1092         }
1093         spin_lock(&inode->i_lock);
1094         cache_validity = nfsi->cache_validity;
1095         nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1096
1097         /*
1098          * We may need to keep the attributes marked as invalid if
1099          * we raced with nfs_end_attr_update().
1100          */
1101         if (verifier == nfsi->cache_change_attribute)
1102                 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1103         spin_unlock(&inode->i_lock);
1104
1105         nfs_revalidate_mapping(inode, inode->i_mapping);
1106
1107         if (cache_validity & NFS_INO_INVALID_ACL)
1108                 nfs_zap_acl_cache(inode);
1109
1110         dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1111                 inode->i_sb->s_id,
1112                 (long long)NFS_FILEID(inode));
1113
1114  out:
1115         nfs_wake_up_inode(inode);
1116
1117  out_nowait:
1118         unlock_kernel();
1119         return status;
1120 }
1121
1122 int nfs_attribute_timeout(struct inode *inode)
1123 {
1124         struct nfs_inode *nfsi = NFS_I(inode);
1125
1126         if (nfs_have_delegation(inode, FMODE_READ))
1127                 return 0;
1128         return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1129 }
1130
1131 /**
1132  * nfs_revalidate_inode - Revalidate the inode attributes
1133  * @server - pointer to nfs_server struct
1134  * @inode - pointer to inode struct
1135  *
1136  * Updates inode attribute information by retrieving the data from the server.
1137  */
1138 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1139 {
1140         if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1141                         && !nfs_attribute_timeout(inode))
1142                 return NFS_STALE(inode) ? -ESTALE : 0;
1143         return __nfs_revalidate_inode(server, inode);
1144 }
1145
1146 /**
1147  * nfs_revalidate_mapping - Revalidate the pagecache
1148  * @inode - pointer to host inode
1149  * @mapping - pointer to mapping
1150  */
1151 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1152 {
1153         struct nfs_inode *nfsi = NFS_I(inode);
1154
1155         if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1156                 if (S_ISREG(inode->i_mode)) {
1157                         if (filemap_fdatawrite(mapping) == 0)
1158                                 filemap_fdatawait(mapping);
1159                         nfs_wb_all(inode);
1160                 }
1161                 invalidate_inode_pages2(mapping);
1162
1163                 spin_lock(&inode->i_lock);
1164                 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1165                 if (S_ISDIR(inode->i_mode)) {
1166                         memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1167                         /* This ensures we revalidate child dentries */
1168                         nfsi->cache_change_attribute++;
1169                 }
1170                 spin_unlock(&inode->i_lock);
1171
1172                 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1173                                 inode->i_sb->s_id,
1174                                 (long long)NFS_FILEID(inode));
1175         }
1176 }
1177
1178 /**
1179  * nfs_begin_data_update
1180  * @inode - pointer to inode
1181  * Declare that a set of operations will update file data on the server
1182  */
1183 void nfs_begin_data_update(struct inode *inode)
1184 {
1185         atomic_inc(&NFS_I(inode)->data_updates);
1186 }
1187
1188 /**
1189  * nfs_end_data_update
1190  * @inode - pointer to inode
1191  * Declare end of the operations that will update file data
1192  * This will mark the inode as immediately needing revalidation
1193  * of its attribute cache.
1194  */
1195 void nfs_end_data_update(struct inode *inode)
1196 {
1197         struct nfs_inode *nfsi = NFS_I(inode);
1198
1199         if (!nfs_have_delegation(inode, FMODE_READ)) {
1200                 /* Mark the attribute cache for revalidation */
1201                 spin_lock(&inode->i_lock);
1202                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1203                 /* Directories and symlinks: invalidate page cache too */
1204                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1205                         nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1206                 spin_unlock(&inode->i_lock);
1207         }
1208         nfsi->cache_change_attribute ++;
1209         atomic_dec(&nfsi->data_updates);
1210 }
1211
1212 /**
1213  * nfs_refresh_inode - verify consistency of the inode attribute cache
1214  * @inode - pointer to inode
1215  * @fattr - updated attributes
1216  *
1217  * Verifies the attribute cache. If we have just changed the attributes,
1218  * so that fattr carries weak cache consistency data, then it may
1219  * also update the ctime/mtime/change_attribute.
1220  */
1221 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1222 {
1223         struct nfs_inode *nfsi = NFS_I(inode);
1224         loff_t cur_size, new_isize;
1225         int data_unstable;
1226
1227         /* Do we hold a delegation? */
1228         if (nfs_have_delegation(inode, FMODE_READ))
1229                 return 0;
1230
1231         spin_lock(&inode->i_lock);
1232
1233         /* Are we in the process of updating data on the server? */
1234         data_unstable = nfs_caches_unstable(inode);
1235
1236         if (fattr->valid & NFS_ATTR_FATTR_V4) {
1237                 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1238                                 && nfsi->change_attr == fattr->pre_change_attr)
1239                         nfsi->change_attr = fattr->change_attr;
1240                 if (nfsi->change_attr != fattr->change_attr) {
1241                         nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1242                         if (!data_unstable)
1243                                 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1244                 }
1245         }
1246
1247         if ((fattr->valid & NFS_ATTR_FATTR) == 0) {
1248                 spin_unlock(&inode->i_lock);
1249                 return 0;
1250         }
1251
1252         /* Has the inode gone and changed behind our back? */
1253         if (nfsi->fileid != fattr->fileid
1254                         || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1255                 spin_unlock(&inode->i_lock);
1256                 return -EIO;
1257         }
1258
1259         cur_size = i_size_read(inode);
1260         new_isize = nfs_size_to_loff_t(fattr->size);
1261
1262         /* If we have atomic WCC data, we may update some attributes */
1263         if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1264                 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1265                         memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1266                 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1267                         memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1268         }
1269
1270         /* Verify a few of the more important attributes */
1271         if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1272                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1273                 if (!data_unstable)
1274                         nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1275         }
1276         if (cur_size != new_isize) {
1277                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1278                 if (nfsi->npages == 0)
1279                         nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1280         }
1281
1282         /* Have any file permissions changed? */
1283         if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1284                         || inode->i_uid != fattr->uid
1285                         || inode->i_gid != fattr->gid)
1286                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1287
1288         /* Has the link count changed? */
1289         if (inode->i_nlink != fattr->nlink)
1290                 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1291
1292         if (!timespec_equal(&inode->i_atime, &fattr->atime))
1293                 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1294
1295         nfsi->read_cache_jiffies = fattr->timestamp;
1296         spin_unlock(&inode->i_lock);
1297         return 0;
1298 }
1299
1300 /*
1301  * Many nfs protocol calls return the new file attributes after
1302  * an operation.  Here we update the inode to reflect the state
1303  * of the server's inode.
1304  *
1305  * This is a bit tricky because we have to make sure all dirty pages
1306  * have been sent off to the server before calling invalidate_inode_pages.
1307  * To make sure no other process adds more write requests while we try
1308  * our best to flush them, we make them sleep during the attribute refresh.
1309  *
1310  * A very similar scenario holds for the dir cache.
1311  */
1312 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1313 {
1314         struct nfs_inode *nfsi = NFS_I(inode);
1315         loff_t cur_isize, new_isize;
1316         unsigned int    invalid = 0;
1317         int data_unstable;
1318
1319         dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1320                         __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1321                         atomic_read(&inode->i_count), fattr->valid);
1322
1323         if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1324                 return 0;
1325
1326         if (nfsi->fileid != fattr->fileid) {
1327                 printk(KERN_ERR "%s: inode number mismatch\n"
1328                        "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1329                        __FUNCTION__,
1330                        inode->i_sb->s_id, (long long)nfsi->fileid,
1331                        inode->i_sb->s_id, (long long)fattr->fileid);
1332                 goto out_err;
1333         }
1334
1335         spin_lock(&inode->i_lock);
1336
1337         /*
1338          * Make sure the inode's type hasn't changed.
1339          */
1340         if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1341                 spin_unlock(&inode->i_lock);
1342                 goto out_changed;
1343         }
1344
1345         /*
1346          * Update the read time so we don't revalidate too often.
1347          */
1348         nfsi->read_cache_jiffies = fattr->timestamp;
1349
1350         /* Are we racing with known updates of the metadata on the server? */
1351         data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1352
1353         /* Check if our cached file size is stale */
1354         new_isize = nfs_size_to_loff_t(fattr->size);
1355         cur_isize = i_size_read(inode);
1356         if (new_isize != cur_isize) {
1357                 /* Do we perhaps have any outstanding writes? */
1358                 if (nfsi->npages == 0) {
1359                         /* No, but did we race with nfs_end_data_update()? */
1360                         if (verifier  ==  nfsi->cache_change_attribute) {
1361                                 inode->i_size = new_isize;
1362                                 invalid |= NFS_INO_INVALID_DATA;
1363                         }
1364                         invalid |= NFS_INO_INVALID_ATTR;
1365                 } else if (new_isize > cur_isize) {
1366                         inode->i_size = new_isize;
1367                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1368                 }
1369                 dprintk("NFS: isize change on server for file %s/%ld\n",
1370                                 inode->i_sb->s_id, inode->i_ino);
1371         }
1372
1373         /* Check if the mtime agrees */
1374         if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1375                 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1376                 dprintk("NFS: mtime change on server for file %s/%ld\n",
1377                                 inode->i_sb->s_id, inode->i_ino);
1378                 if (!data_unstable)
1379                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1380         }
1381
1382         if ((fattr->valid & NFS_ATTR_FATTR_V4)
1383             && nfsi->change_attr != fattr->change_attr) {
1384                 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1385                        inode->i_sb->s_id, inode->i_ino);
1386                 nfsi->change_attr = fattr->change_attr;
1387                 if (!data_unstable)
1388                         invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1389         }
1390
1391         /* If ctime has changed we should definitely clear access+acl caches */
1392         if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1393                 if (!data_unstable)
1394                         invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1395                 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1396         }
1397         memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1398
1399         if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1400             inode->i_uid != fattr->uid ||
1401             inode->i_gid != fattr->gid)
1402                 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1403
1404         inode->i_mode = fattr->mode;
1405         inode->i_nlink = fattr->nlink;
1406         inode->i_uid = fattr->uid;
1407         inode->i_gid = fattr->gid;
1408
1409         if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1410                 /*
1411                  * report the blocks in 512byte units
1412                  */
1413                 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1414                 inode->i_blksize = inode->i_sb->s_blocksize;
1415         } else {
1416                 inode->i_blocks = fattr->du.nfs2.blocks;
1417                 inode->i_blksize = fattr->du.nfs2.blocksize;
1418         }
1419
1420         /* Update attrtimeo value if we're out of the unstable period */
1421         if (invalid & NFS_INO_INVALID_ATTR) {
1422                 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1423                 nfsi->attrtimeo_timestamp = jiffies;
1424         } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1425                 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1426                         nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1427                 nfsi->attrtimeo_timestamp = jiffies;
1428         }
1429         /* Don't invalidate the data if we were to blame */
1430         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1431                                 || S_ISLNK(inode->i_mode)))
1432                 invalid &= ~NFS_INO_INVALID_DATA;
1433         if (!nfs_have_delegation(inode, FMODE_READ))
1434                 nfsi->cache_validity |= invalid;
1435
1436         spin_unlock(&inode->i_lock);
1437         return 0;
1438  out_changed:
1439         /*
1440          * Big trouble! The inode has become a different object.
1441          */
1442 #ifdef NFS_PARANOIA
1443         printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1444                         __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1445 #endif
1446         /*
1447          * No need to worry about unhashing the dentry, as the
1448          * lookup validation will know that the inode is bad.
1449          * (But we fall through to invalidate the caches.)
1450          */
1451         nfs_invalidate_inode(inode);
1452  out_err:
1453         set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1454         return -ESTALE;
1455 }
1456
1457 /*
1458  * File system information
1459  */
1460
1461 static int nfs_set_super(struct super_block *s, void *data)
1462 {
1463         s->s_fs_info = data;
1464         return set_anon_super(s, data);
1465 }
1466  
1467 static int nfs_compare_super(struct super_block *sb, void *data)
1468 {
1469         struct nfs_server *server = data;
1470         struct nfs_server *old = NFS_SB(sb);
1471
1472         if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1473                 return 0;
1474         if (old->addr.sin_port != server->addr.sin_port)
1475                 return 0;
1476         return !nfs_compare_fh(&old->fh, &server->fh);
1477 }
1478
1479 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1480         int flags, const char *dev_name, void *raw_data)
1481 {
1482         int error;
1483         struct nfs_server *server = NULL;
1484         struct super_block *s;
1485         struct nfs_fh *root;
1486         struct nfs_mount_data *data = raw_data;
1487
1488         s = ERR_PTR(-EINVAL);
1489         if (data == NULL) {
1490                 dprintk("%s: missing data argument\n", __FUNCTION__);
1491                 goto out_err;
1492         }
1493         if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1494                 dprintk("%s: bad mount version\n", __FUNCTION__);
1495                 goto out_err;
1496         }
1497         switch (data->version) {
1498                 case 1:
1499                         data->namlen = 0;
1500                 case 2:
1501                         data->bsize  = 0;
1502                 case 3:
1503                         if (data->flags & NFS_MOUNT_VER3) {
1504                                 dprintk("%s: mount structure version %d does not support NFSv3\n",
1505                                                 __FUNCTION__,
1506                                                 data->version);
1507                                 goto out_err;
1508                         }
1509                         data->root.size = NFS2_FHSIZE;
1510                         memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1511                 case 4:
1512                         if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1513                                 dprintk("%s: mount structure version %d does not support strong security\n",
1514                                                 __FUNCTION__,
1515                                                 data->version);
1516                                 goto out_err;
1517                         }
1518                 case 5:
1519                         memset(data->context, 0, sizeof(data->context));
1520         }
1521 #ifndef CONFIG_NFS_V3
1522         /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1523         s = ERR_PTR(-EPROTONOSUPPORT);
1524         if (data->flags & NFS_MOUNT_VER3) {
1525                 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1526                 goto out_err;
1527         }
1528 #endif /* CONFIG_NFS_V3 */
1529
1530         s = ERR_PTR(-ENOMEM);
1531         server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1532         if (!server)
1533                 goto out_err;
1534         memset(server, 0, sizeof(struct nfs_server));
1535         /* Zero out the NFS state stuff */
1536         init_nfsv4_state(server);
1537         server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1538
1539         root = &server->fh;
1540         if (data->flags & NFS_MOUNT_VER3)
1541                 root->size = data->root.size;
1542         else
1543                 root->size = NFS2_FHSIZE;
1544         s = ERR_PTR(-EINVAL);
1545         if (root->size > sizeof(root->data)) {
1546                 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1547                 goto out_err;
1548         }
1549         memcpy(root->data, data->root.data, root->size);
1550
1551         /* We now require that the mount process passes the remote address */
1552         memcpy(&server->addr, &data->addr, sizeof(server->addr));
1553         if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1554                 dprintk("%s: mount program didn't pass remote address!\n",
1555                                 __FUNCTION__);
1556                 goto out_err;
1557         }
1558
1559         /* Fire up rpciod if not yet running */
1560         s = ERR_PTR(rpciod_up());
1561         if (IS_ERR(s)) {
1562                 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1563                                 __FUNCTION__, PTR_ERR(s));
1564                 goto out_err;
1565         }
1566
1567         s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1568         if (IS_ERR(s) || s->s_root)
1569                 goto out_rpciod_down;
1570
1571         s->s_flags = flags;
1572
1573         error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1574         if (error) {
1575                 up_write(&s->s_umount);
1576                 deactivate_super(s);
1577                 return ERR_PTR(error);
1578         }
1579         s->s_flags |= MS_ACTIVE;
1580         return s;
1581 out_rpciod_down:
1582         rpciod_down();
1583 out_err:
1584         kfree(server);
1585         return s;
1586 }
1587
1588 static void nfs_kill_super(struct super_block *s)
1589 {
1590         struct nfs_server *server = NFS_SB(s);
1591
1592         kill_anon_super(s);
1593
1594         if (!IS_ERR(server->client))
1595                 rpc_shutdown_client(server->client);
1596         if (!IS_ERR(server->client_sys))
1597                 rpc_shutdown_client(server->client_sys);
1598         if (!IS_ERR(server->client_acl))
1599                 rpc_shutdown_client(server->client_acl);
1600
1601         if (!(server->flags & NFS_MOUNT_NONLM))
1602                 lockd_down();   /* release rpc.lockd */
1603
1604         rpciod_down();          /* release rpciod */
1605
1606         if (server->hostname != NULL)
1607                 kfree(server->hostname);
1608         kfree(server);
1609 }
1610
1611 static struct file_system_type nfs_fs_type = {
1612         .owner          = THIS_MODULE,
1613         .name           = "nfs",
1614         .get_sb         = nfs_get_sb,
1615         .kill_sb        = nfs_kill_super,
1616         .fs_flags       = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1617 };
1618
1619 #ifdef CONFIG_NFS_V4
1620
1621 static void nfs4_clear_inode(struct inode *);
1622
1623
1624 static struct super_operations nfs4_sops = { 
1625         .alloc_inode    = nfs_alloc_inode,
1626         .destroy_inode  = nfs_destroy_inode,
1627         .write_inode    = nfs_write_inode,
1628         .delete_inode   = nfs_delete_inode,
1629         .statfs         = nfs_statfs,
1630         .clear_inode    = nfs4_clear_inode,
1631         .umount_begin   = nfs_umount_begin,
1632         .show_options   = nfs_show_options,
1633 };
1634
1635 /*
1636  * Clean out any remaining NFSv4 state that might be left over due
1637  * to open() calls that passed nfs_atomic_lookup, but failed to call
1638  * nfs_open().
1639  */
1640 static void nfs4_clear_inode(struct inode *inode)
1641 {
1642         struct nfs_inode *nfsi = NFS_I(inode);
1643
1644         /* If we are holding a delegation, return it! */
1645         if (nfsi->delegation != NULL)
1646                 nfs_inode_return_delegation(inode);
1647         /* First call standard NFS clear_inode() code */
1648         nfs_clear_inode(inode);
1649         /* Now clear out any remaining state */
1650         while (!list_empty(&nfsi->open_states)) {
1651                 struct nfs4_state *state;
1652                 
1653                 state = list_entry(nfsi->open_states.next,
1654                                 struct nfs4_state,
1655                                 inode_states);
1656                 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1657                                 __FUNCTION__,
1658                                 inode->i_sb->s_id,
1659                                 (long long)NFS_FILEID(inode),
1660                                 state);
1661                 BUG_ON(atomic_read(&state->count) != 1);
1662                 nfs4_close_state(state, state->state);
1663         }
1664 }
1665
1666
1667 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1668 {
1669         struct nfs_server *server;
1670         struct nfs4_client *clp = NULL;
1671         struct rpc_xprt *xprt = NULL;
1672         struct rpc_clnt *clnt = NULL;
1673         struct rpc_timeout timeparms;
1674         rpc_authflavor_t authflavour;
1675         int proto, err = -EIO;
1676
1677         sb->s_blocksize_bits = 0;
1678         sb->s_blocksize = 0;
1679         server = NFS_SB(sb);
1680         if (data->rsize != 0)
1681                 server->rsize = nfs_block_size(data->rsize, NULL);
1682         if (data->wsize != 0)
1683                 server->wsize = nfs_block_size(data->wsize, NULL);
1684         server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1685         server->caps = NFS_CAP_ATOMIC_OPEN;
1686
1687         server->acregmin = data->acregmin*HZ;
1688         server->acregmax = data->acregmax*HZ;
1689         server->acdirmin = data->acdirmin*HZ;
1690         server->acdirmax = data->acdirmax*HZ;
1691
1692         server->rpc_ops = &nfs_v4_clientops;
1693         /* Initialize timeout values */
1694
1695         timeparms.to_initval = data->timeo * HZ / 10;
1696         timeparms.to_retries = data->retrans;
1697         timeparms.to_exponential = 1;
1698         if (!timeparms.to_retries)
1699                 timeparms.to_retries = 5;
1700
1701         proto = data->proto;
1702         /* Which IP protocol do we use? */
1703         switch (proto) {
1704         case IPPROTO_TCP:
1705                 timeparms.to_maxval  = RPC_MAX_TCP_TIMEOUT;
1706                 if (!timeparms.to_initval)
1707                         timeparms.to_initval = 600 * HZ / 10;
1708                 break;
1709         case IPPROTO_UDP:
1710                 timeparms.to_maxval  = RPC_MAX_UDP_TIMEOUT;
1711                 if (!timeparms.to_initval)
1712                         timeparms.to_initval = 11 * HZ / 10;
1713                 break;
1714         default:
1715                 return -EINVAL;
1716         }
1717
1718         clp = nfs4_get_client(&server->addr.sin_addr);
1719         if (!clp) {
1720                 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1721                 return -EIO;
1722         }
1723
1724         /* Now create transport and client */
1725         authflavour = RPC_AUTH_UNIX;
1726         if (data->auth_flavourlen != 0) {
1727                 if (data->auth_flavourlen != 1) {
1728                         dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1729                                         __FUNCTION__, data->auth_flavourlen);
1730                         err = -EINVAL;
1731                         goto out_fail;
1732                 }
1733                 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1734                         err = -EFAULT;
1735                         goto out_fail;
1736                 }
1737         }
1738
1739         down_write(&clp->cl_sem);
1740         if (IS_ERR(clp->cl_rpcclient)) {
1741                 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1742                 if (IS_ERR(xprt)) {
1743                         up_write(&clp->cl_sem);
1744                         err = PTR_ERR(xprt);
1745                         dprintk("%s: cannot create RPC transport. Error = %d\n",
1746                                         __FUNCTION__, err);
1747                         goto out_fail;
1748                 }
1749                 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1750                                 server->rpc_ops->version, authflavour);
1751                 if (IS_ERR(clnt)) {
1752                         up_write(&clp->cl_sem);
1753                         err = PTR_ERR(clnt);
1754                         dprintk("%s: cannot create RPC client. Error = %d\n",
1755                                         __FUNCTION__, err);
1756                         goto out_fail;
1757                 }
1758                 clnt->cl_intr     = 1;
1759                 clnt->cl_softrtry = 1;
1760                 clnt->cl_chatty   = 1;
1761                 clp->cl_rpcclient = clnt;
1762                 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1763                 if (IS_ERR(clp->cl_cred)) {
1764                         up_write(&clp->cl_sem);
1765                         err = PTR_ERR(clp->cl_cred);
1766                         clp->cl_cred = NULL;
1767                         goto out_fail;
1768                 }
1769                 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1770                 nfs_idmap_new(clp);
1771         }
1772         if (list_empty(&clp->cl_superblocks)) {
1773                 err = nfs4_init_client(clp);
1774                 if (err != 0) {
1775                         up_write(&clp->cl_sem);
1776                         goto out_fail;
1777                 }
1778         }
1779         list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1780         clnt = rpc_clone_client(clp->cl_rpcclient);
1781         if (!IS_ERR(clnt))
1782                         server->nfs4_state = clp;
1783         up_write(&clp->cl_sem);
1784         clp = NULL;
1785
1786         if (IS_ERR(clnt)) {
1787                 err = PTR_ERR(clnt);
1788                 dprintk("%s: cannot create RPC client. Error = %d\n",
1789                                 __FUNCTION__, err);
1790                 return err;
1791         }
1792
1793         server->client    = clnt;
1794
1795         if (server->nfs4_state->cl_idmap == NULL) {
1796                 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1797                 return -ENOMEM;
1798         }
1799
1800         if (clnt->cl_auth->au_flavor != authflavour) {
1801                 struct rpc_auth *auth;
1802
1803                 auth = rpcauth_create(authflavour, clnt);
1804                 if (IS_ERR(auth)) {
1805                         dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1806                         return PTR_ERR(auth);
1807                 }
1808         }
1809
1810         sb->s_time_gran = 1;
1811
1812         sb->s_op = &nfs4_sops;
1813         err = nfs_sb_init(sb, authflavour);
1814         if (err == 0)
1815                 return 0;
1816 out_fail:
1817         if (clp)
1818                 nfs4_put_client(clp);
1819         return err;
1820 }
1821
1822 static int nfs4_compare_super(struct super_block *sb, void *data)
1823 {
1824         struct nfs_server *server = data;
1825         struct nfs_server *old = NFS_SB(sb);
1826
1827         if (strcmp(server->hostname, old->hostname) != 0)
1828                 return 0;
1829         if (strcmp(server->mnt_path, old->mnt_path) != 0)
1830                 return 0;
1831         return 1;
1832 }
1833
1834 static void *
1835 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1836 {
1837         void *p = NULL;
1838
1839         if (!src->len)
1840                 return ERR_PTR(-EINVAL);
1841         if (src->len < maxlen)
1842                 maxlen = src->len;
1843         if (dst == NULL) {
1844                 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1845                 if (p == NULL)
1846                         return ERR_PTR(-ENOMEM);
1847         }
1848         if (copy_from_user(dst, src->data, maxlen)) {
1849                 if (p != NULL)
1850                         kfree(p);
1851                 return ERR_PTR(-EFAULT);
1852         }
1853         dst[maxlen] = '\0';
1854         return dst;
1855 }
1856
1857 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1858         int flags, const char *dev_name, void *raw_data)
1859 {
1860         int error;
1861         struct nfs_server *server;
1862         struct super_block *s;
1863         struct nfs4_mount_data *data = raw_data;
1864         void *p;
1865
1866         if (data == NULL) {
1867                 dprintk("%s: missing data argument\n", __FUNCTION__);
1868                 return ERR_PTR(-EINVAL);
1869         }
1870         if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1871                 dprintk("%s: bad mount version\n", __FUNCTION__);
1872                 return ERR_PTR(-EINVAL);
1873         }
1874
1875         server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1876         if (!server)
1877                 return ERR_PTR(-ENOMEM);
1878         memset(server, 0, sizeof(struct nfs_server));
1879         /* Zero out the NFS state stuff */
1880         init_nfsv4_state(server);
1881         server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1882
1883         p = nfs_copy_user_string(NULL, &data->hostname, 256);
1884         if (IS_ERR(p))
1885                 goto out_err;
1886         server->hostname = p;
1887
1888         p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1889         if (IS_ERR(p))
1890                 goto out_err;
1891         server->mnt_path = p;
1892
1893         p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1894                         sizeof(server->ip_addr) - 1);
1895         if (IS_ERR(p))
1896                 goto out_err;
1897
1898         /* We now require that the mount process passes the remote address */
1899         if (data->host_addrlen != sizeof(server->addr)) {
1900                 s = ERR_PTR(-EINVAL);
1901                 goto out_free;
1902         }
1903         if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1904                 s = ERR_PTR(-EFAULT);
1905                 goto out_free;
1906         }
1907         if (server->addr.sin_family != AF_INET ||
1908             server->addr.sin_addr.s_addr == INADDR_ANY) {
1909                 dprintk("%s: mount program didn't pass remote IP address!\n",
1910                                 __FUNCTION__);
1911                 s = ERR_PTR(-EINVAL);
1912                 goto out_free;
1913         }
1914
1915         /* Fire up rpciod if not yet running */
1916         s = ERR_PTR(rpciod_up());
1917         if (IS_ERR(s)) {
1918                 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1919                                 __FUNCTION__, PTR_ERR(s));
1920                 goto out_free;
1921         }
1922
1923         s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1924
1925         if (IS_ERR(s) || s->s_root)
1926                 goto out_free;
1927
1928         s->s_flags = flags;
1929
1930         error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1931         if (error) {
1932                 up_write(&s->s_umount);
1933                 deactivate_super(s);
1934                 return ERR_PTR(error);
1935         }
1936         s->s_flags |= MS_ACTIVE;
1937         return s;
1938 out_err:
1939         s = (struct super_block *)p;
1940 out_free:
1941         if (server->mnt_path)
1942                 kfree(server->mnt_path);
1943         if (server->hostname)
1944                 kfree(server->hostname);
1945         kfree(server);
1946         return s;
1947 }
1948
1949 static void nfs4_kill_super(struct super_block *sb)
1950 {
1951         struct nfs_server *server = NFS_SB(sb);
1952
1953         nfs_return_all_delegations(sb);
1954         kill_anon_super(sb);
1955
1956         nfs4_renewd_prepare_shutdown(server);
1957
1958         if (server->client != NULL && !IS_ERR(server->client))
1959                 rpc_shutdown_client(server->client);
1960         rpciod_down();          /* release rpciod */
1961
1962         destroy_nfsv4_state(server);
1963
1964         if (server->hostname != NULL)
1965                 kfree(server->hostname);
1966         kfree(server);
1967 }
1968
1969 static struct file_system_type nfs4_fs_type = {
1970         .owner          = THIS_MODULE,
1971         .name           = "nfs4",
1972         .get_sb         = nfs4_get_sb,
1973         .kill_sb        = nfs4_kill_super,
1974         .fs_flags       = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1975 };
1976
1977 #define nfs4_init_once(nfsi) \
1978         do { \
1979                 INIT_LIST_HEAD(&(nfsi)->open_states); \
1980                 nfsi->delegation = NULL; \
1981                 nfsi->delegation_state = 0; \
1982                 init_rwsem(&nfsi->rwsem); \
1983         } while(0)
1984 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1985 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1986 #else
1987 #define nfs4_init_once(nfsi) \
1988         do { } while (0)
1989 #define register_nfs4fs() (0)
1990 #define unregister_nfs4fs()
1991 #endif
1992
1993 extern int nfs_init_nfspagecache(void);
1994 extern void nfs_destroy_nfspagecache(void);
1995 extern int nfs_init_readpagecache(void);
1996 extern void nfs_destroy_readpagecache(void);
1997 extern int nfs_init_writepagecache(void);
1998 extern void nfs_destroy_writepagecache(void);
1999 #ifdef CONFIG_NFS_DIRECTIO
2000 extern int nfs_init_directcache(void);
2001 extern void nfs_destroy_directcache(void);
2002 #endif
2003
2004 static kmem_cache_t * nfs_inode_cachep;
2005
2006 static struct inode *nfs_alloc_inode(struct super_block *sb)
2007 {
2008         struct nfs_inode *nfsi;
2009         nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2010         if (!nfsi)
2011                 return NULL;
2012         nfsi->flags = 0UL;
2013         nfsi->cache_validity = 0UL;
2014 #ifdef CONFIG_NFS_V3_ACL
2015         nfsi->acl_access = ERR_PTR(-EAGAIN);
2016         nfsi->acl_default = ERR_PTR(-EAGAIN);
2017 #endif
2018 #ifdef CONFIG_NFS_V4
2019         nfsi->nfs4_acl = NULL;
2020 #endif /* CONFIG_NFS_V4 */
2021         return &nfsi->vfs_inode;
2022 }
2023
2024 static void nfs_destroy_inode(struct inode *inode)
2025 {
2026         kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2027 }
2028
2029 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2030 {
2031         struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2032
2033         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2034             SLAB_CTOR_CONSTRUCTOR) {
2035                 inode_init_once(&nfsi->vfs_inode);
2036                 spin_lock_init(&nfsi->req_lock);
2037                 INIT_LIST_HEAD(&nfsi->dirty);
2038                 INIT_LIST_HEAD(&nfsi->commit);
2039                 INIT_LIST_HEAD(&nfsi->open_files);
2040                 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2041                 atomic_set(&nfsi->data_updates, 0);
2042                 nfsi->ndirty = 0;
2043                 nfsi->ncommit = 0;
2044                 nfsi->npages = 0;
2045                 nfs4_init_once(nfsi);
2046         }
2047 }
2048  
2049 static int nfs_init_inodecache(void)
2050 {
2051         nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2052                                              sizeof(struct nfs_inode),
2053                                              0, SLAB_RECLAIM_ACCOUNT,
2054                                              init_once, NULL);
2055         if (nfs_inode_cachep == NULL)
2056                 return -ENOMEM;
2057
2058         return 0;
2059 }
2060
2061 static void nfs_destroy_inodecache(void)
2062 {
2063         if (kmem_cache_destroy(nfs_inode_cachep))
2064                 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2065 }
2066
2067 /*
2068  * Initialize NFS
2069  */
2070 static int __init init_nfs_fs(void)
2071 {
2072         int err;
2073
2074         err = nfs_init_nfspagecache();
2075         if (err)
2076                 goto out4;
2077
2078         err = nfs_init_inodecache();
2079         if (err)
2080                 goto out3;
2081
2082         err = nfs_init_readpagecache();
2083         if (err)
2084                 goto out2;
2085
2086         err = nfs_init_writepagecache();
2087         if (err)
2088                 goto out1;
2089
2090 #ifdef CONFIG_NFS_DIRECTIO
2091         err = nfs_init_directcache();
2092         if (err)
2093                 goto out0;
2094 #endif
2095
2096 #ifdef CONFIG_PROC_FS
2097         rpc_proc_register(&nfs_rpcstat);
2098 #endif
2099         err = register_filesystem(&nfs_fs_type);
2100         if (err)
2101                 goto out;
2102         if ((err = register_nfs4fs()) != 0)
2103                 goto out;
2104         return 0;
2105 out:
2106 #ifdef CONFIG_PROC_FS
2107         rpc_proc_unregister("nfs");
2108 #endif
2109         nfs_destroy_writepagecache();
2110 #ifdef CONFIG_NFS_DIRECTIO
2111 out0:
2112         nfs_destroy_directcache();
2113 #endif
2114 out1:
2115         nfs_destroy_readpagecache();
2116 out2:
2117         nfs_destroy_inodecache();
2118 out3:
2119         nfs_destroy_nfspagecache();
2120 out4:
2121         return err;
2122 }
2123
2124 static void __exit exit_nfs_fs(void)
2125 {
2126 #ifdef CONFIG_NFS_DIRECTIO
2127         nfs_destroy_directcache();
2128 #endif
2129         nfs_destroy_writepagecache();
2130         nfs_destroy_readpagecache();
2131         nfs_destroy_inodecache();
2132         nfs_destroy_nfspagecache();
2133 #ifdef CONFIG_PROC_FS
2134         rpc_proc_unregister("nfs");
2135 #endif
2136         unregister_filesystem(&nfs_fs_type);
2137         unregister_nfs4fs();
2138 }
2139
2140 /* Not quite true; I just maintain it */
2141 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2142 MODULE_LICENSE("GPL");
2143
2144 module_init(init_nfs_fs)
2145 module_exit(exit_nfs_fs)