File: /usr/src/linux/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/sched.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/locks.h>
27 #include <linux/unistd.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/sunrpc/stats.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/nfs_flushd.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/smp_lock.h>
35
36 #include <asm/system.h>
37 #include <asm/uaccess.h>
38
39 #define CONFIG_NFS_SNAPSHOT 1
40 #define NFSDBG_FACILITY NFSDBG_VFS
41 #define NFS_PARANOIA 1
42
43 static struct inode * __nfs_fhget(struct super_block *, struct nfs_fh *, struct nfs_fattr *);
44 void nfs_zap_caches(struct inode *);
45 static void nfs_invalidate_inode(struct inode *);
46
47 static void nfs_read_inode(struct inode *);
48 static void nfs_write_inode(struct inode *,int);
49 static void nfs_delete_inode(struct inode *);
50 static void nfs_put_super(struct super_block *);
51 static void nfs_clear_inode(struct inode *);
52 static void nfs_umount_begin(struct super_block *);
53 static int nfs_statfs(struct super_block *, struct statfs *);
54
55 static struct super_operations nfs_sops = {
56 read_inode: nfs_read_inode,
57 write_inode: nfs_write_inode,
58 delete_inode: nfs_delete_inode,
59 put_super: nfs_put_super,
60 statfs: nfs_statfs,
61 clear_inode: nfs_clear_inode,
62 umount_begin: nfs_umount_begin,
63 };
64
65 /*
66 * RPC cruft for NFS
67 */
68 struct rpc_stat nfs_rpcstat = { &nfs_program };
69 static struct rpc_version * nfs_version[] = {
70 NULL,
71 NULL,
72 &nfs_version2,
73 #ifdef CONFIG_NFS_V3
74 &nfs_version3,
75 #endif
76 };
77
78 struct rpc_program nfs_program = {
79 "nfs",
80 NFS_PROGRAM,
81 sizeof(nfs_version) / sizeof(nfs_version[0]),
82 nfs_version,
83 &nfs_rpcstat,
84 };
85
86 static inline unsigned long
87 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
88 {
89 return nfs_fileid_to_ino_t(fattr->fileid);
90 }
91
92 /*
93 * The "read_inode" function doesn't actually do anything:
94 * the real data is filled in later in nfs_fhget. Here we
95 * just mark the cache times invalid, and zero out i_mode
96 * (the latter makes "nfs_refresh_inode" do the right thing
97 * wrt pipe inodes)
98 */
99 static void
100 nfs_read_inode(struct inode * inode)
101 {
102 inode->i_blksize = inode->i_sb->s_blocksize;
103 inode->i_mode = 0;
104 inode->i_rdev = 0;
105 /* We can't support UPDATE_ATIME(), since the server will reset it */
106 inode->i_flags |= S_NOATIME;
107 NFS_FILEID(inode) = 0;
108 NFS_FSID(inode) = 0;
109 NFS_FLAGS(inode) = 0;
110 INIT_LIST_HEAD(&inode->u.nfs_i.read);
111 INIT_LIST_HEAD(&inode->u.nfs_i.dirty);
112 INIT_LIST_HEAD(&inode->u.nfs_i.commit);
113 INIT_LIST_HEAD(&inode->u.nfs_i.writeback);
114 inode->u.nfs_i.nread = 0;
115 inode->u.nfs_i.ndirty = 0;
116 inode->u.nfs_i.ncommit = 0;
117 inode->u.nfs_i.npages = 0;
118 NFS_CACHEINV(inode);
119 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
120 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
121 }
122
123 static void
124 nfs_write_inode(struct inode *inode, int sync)
125 {
126 int flags = sync ? FLUSH_WAIT : 0;
127
128 nfs_sync_file(inode, NULL, 0, 0, flags);
129 }
130
131 static void
132 nfs_delete_inode(struct inode * inode)
133 {
134 dprintk("NFS: delete_inode(%x/%ld)\n", inode->i_dev, inode->i_ino);
135
136 /*
137 * The following can never actually happen...
138 */
139 if (nfs_have_writebacks(inode) || nfs_have_read(inode)) {
140 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
141 }
142
143 clear_inode(inode);
144 }
145
146 /*
147 * For the moment, the only task for the NFS clear_inode method is to
148 * release the mmap credential
149 */
150 static void
151 nfs_clear_inode(struct inode *inode)
152 {
153 struct rpc_cred *cred = NFS_I(inode)->mm_cred;
154
155 if (cred)
156 put_rpccred(cred);
157 }
158
159 void
160 nfs_put_super(struct super_block *sb)
161 {
162 struct nfs_server *server = &sb->u.nfs_sb.s_server;
163 struct rpc_clnt *rpc;
164
165 /*
166 * First get rid of the request flushing daemon.
167 * Relies on rpc_shutdown_client() waiting on all
168 * client tasks to finish.
169 */
170 nfs_reqlist_exit(server);
171
172 if ((rpc = server->client) != NULL)
173 rpc_shutdown_client(rpc);
174
175 nfs_reqlist_free(server);
176
177 if (!(server->flags & NFS_MOUNT_NONLM))
178 lockd_down(); /* release rpc.lockd */
179 rpciod_down(); /* release rpciod */
180
181 kfree(server->hostname);
182 }
183
184 void
185 nfs_umount_begin(struct super_block *sb)
186 {
187 struct nfs_server *server = &sb->u.nfs_sb.s_server;
188 struct rpc_clnt *rpc;
189
190 /* -EIO all pending I/O */
191 if ((rpc = server->client) != NULL)
192 rpc_killall_tasks(rpc);
193 }
194
195
196 static inline unsigned long
197 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
198 {
199 /* make sure blocksize is a power of two */
200 if ((bsize & (bsize - 1)) || nrbitsp) {
201 unsigned char nrbits;
202
203 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
204 ;
205 bsize = 1 << nrbits;
206 if (nrbitsp)
207 *nrbitsp = nrbits;
208 }
209
210 return bsize;
211 }
212
213 /*
214 * Calculate the number of 512byte blocks used.
215 */
216 static inline unsigned long
217 nfs_calc_block_size(u64 tsize)
218 {
219 loff_t used = (tsize + 511) >> 9;
220 return (used > ULONG_MAX) ? ULONG_MAX : used;
221 }
222
223 /*
224 * Compute and set NFS server blocksize
225 */
226 static inline unsigned long
227 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
228 {
229 if (bsize < 1024)
230 bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
231 else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
232 bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
233
234 return nfs_block_bits(bsize, nrbitsp);
235 }
236
237 /*
238 * Obtain the root inode of the file system.
239 */
240 static struct inode *
241 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh)
242 {
243 struct nfs_server *server = &sb->u.nfs_sb.s_server;
244 struct nfs_fattr fattr;
245 struct inode *inode;
246 int error;
247
248 if ((error = server->rpc_ops->getroot(server, rootfh, &fattr)) < 0) {
249 printk(KERN_NOTICE "nfs_get_root: getattr error = %d\n", -error);
250 return NULL;
251 }
252
253 inode = __nfs_fhget(sb, rootfh, &fattr);
254 return inode;
255 }
256
257 /*
258 * The way this works is that the mount process passes a structure
259 * in the data argument which contains the server's IP address
260 * and the root file handle obtained from the server's mount
261 * daemon. We stash these away in the private superblock fields.
262 */
263 struct super_block *
264 nfs_read_super(struct super_block *sb, void *raw_data, int silent)
265 {
266 struct nfs_mount_data *data = (struct nfs_mount_data *) raw_data;
267 struct nfs_server *server;
268 struct rpc_xprt *xprt = NULL;
269 struct rpc_clnt *clnt = NULL;
270 struct nfs_fh *root = &data->root, fh;
271 struct inode *root_inode = NULL;
272 unsigned int authflavor;
273 struct sockaddr_in srvaddr;
274 struct rpc_timeout timeparms;
275 struct nfs_fsinfo fsinfo;
276 int tcp, version, maxlen;
277
278 memset(&sb->u.nfs_sb, 0, sizeof(sb->u.nfs_sb));
279 if (!data)
280 goto out_miss_args;
281
282 memset(&fh, 0, sizeof(fh));
283 if (data->version != NFS_MOUNT_VERSION) {
284 printk("nfs warning: mount version %s than kernel\n",
285 data->version < NFS_MOUNT_VERSION ? "older" : "newer");
286 if (data->version < 2)
287 data->namlen = 0;
288 if (data->version < 3)
289 data->bsize = 0;
290 if (data->version < 4) {
291 data->flags &= ~NFS_MOUNT_VER3;
292 root = &fh;
293 root->size = NFS2_FHSIZE;
294 memcpy(root->data, data->old_root.data, NFS2_FHSIZE);
295 }
296 }
297
298 /* We now require that the mount process passes the remote address */
299 memcpy(&srvaddr, &data->addr, sizeof(srvaddr));
300 if (srvaddr.sin_addr.s_addr == INADDR_ANY)
301 goto out_no_remote;
302
303 sb->s_magic = NFS_SUPER_MAGIC;
304 sb->s_op = &nfs_sops;
305 sb->s_blocksize_bits = 0;
306 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
307 server = &sb->u.nfs_sb.s_server;
308 server->rsize = nfs_block_size(data->rsize, NULL);
309 server->wsize = nfs_block_size(data->wsize, NULL);
310 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
311
312 if (data->flags & NFS_MOUNT_NOAC) {
313 data->acregmin = data->acregmax = 0;
314 data->acdirmin = data->acdirmax = 0;
315 }
316 server->acregmin = data->acregmin*HZ;
317 server->acregmax = data->acregmax*HZ;
318 server->acdirmin = data->acdirmin*HZ;
319 server->acdirmax = data->acdirmax*HZ;
320
321 server->namelen = data->namlen;
322 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
323 if (!server->hostname)
324 goto out_unlock;
325 strcpy(server->hostname, data->hostname);
326
327 nfsv3_try_again:
328 /* Check NFS protocol revision and initialize RPC op vector
329 * and file handle pool. */
330 if (data->flags & NFS_MOUNT_VER3) {
331 #ifdef CONFIG_NFS_V3
332 server->rpc_ops = &nfs_v3_clientops;
333 version = 3;
334 if (data->version < 4) {
335 printk(KERN_NOTICE "NFS: NFSv3 not supported by mount program.\n");
336 goto out_unlock;
337 }
338 #else
339 printk(KERN_NOTICE "NFS: NFSv3 not supported.\n");
340 goto out_unlock;
341 #endif
342 } else {
343 server->rpc_ops = &nfs_v2_clientops;
344 version = 2;
345 }
346
347 /* Which protocol do we use? */
348 tcp = (data->flags & NFS_MOUNT_TCP);
349
350 /* Initialize timeout values */
351 timeparms.to_initval = data->timeo * HZ / 10;
352 timeparms.to_retries = data->retrans;
353 timeparms.to_maxval = tcp? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
354 timeparms.to_exponential = 1;
355
356 if (!timeparms.to_initval)
357 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
358 if (!timeparms.to_retries)
359 timeparms.to_retries = 5;
360
361 /* Now create transport and client */
362 xprt = xprt_create_proto(tcp? IPPROTO_TCP : IPPROTO_UDP,
363 &srvaddr, &timeparms);
364 if (xprt == NULL)
365 goto out_no_xprt;
366
367 /* Choose authentication flavor */
368 authflavor = RPC_AUTH_UNIX;
369 if (data->flags & NFS_MOUNT_SECURE)
370 authflavor = RPC_AUTH_DES;
371 else if (data->flags & NFS_MOUNT_KERBEROS)
372 authflavor = RPC_AUTH_KRB;
373
374 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
375 version, authflavor);
376 if (clnt == NULL)
377 goto out_no_client;
378
379 clnt->cl_intr = (data->flags & NFS_MOUNT_INTR)? 1 : 0;
380 clnt->cl_softrtry = (data->flags & NFS_MOUNT_SOFT)? 1 : 0;
381 clnt->cl_droppriv = (data->flags & NFS_MOUNT_BROKEN_SUID) ? 1 : 0;
382 clnt->cl_chatty = 1;
383 server->client = clnt;
384
385 /* Fire up rpciod if not yet running */
386 if (rpciod_up() != 0)
387 goto out_no_iod;
388
389 /*
390 * Keep the super block locked while we try to get
391 * the root fh attributes.
392 */
393 /* Did getting the root inode fail? */
394 if (!(root_inode = nfs_get_root(sb, root))
395 && (data->flags & NFS_MOUNT_VER3)) {
396 data->flags &= ~NFS_MOUNT_VER3;
397 rpciod_down();
398 rpc_shutdown_client(server->client);
399 goto nfsv3_try_again;
400 }
401
402 if (!root_inode)
403 goto out_no_root;
404 sb->s_root = d_alloc_root(root_inode);
405 if (!sb->s_root)
406 goto out_no_root;
407
408 sb->s_root->d_op = &nfs_dentry_operations;
409
410 /* Get some general file system info */
411 if (server->rpc_ops->statfs(server, root, &fsinfo) >= 0) {
412 if (server->namelen == 0)
413 server->namelen = fsinfo.namelen;
414 } else {
415 printk(KERN_NOTICE "NFS: cannot retrieve file system info.\n");
416 goto out_no_root;
417 }
418
419 /* Work out a lot of parameters */
420 if (data->rsize == 0)
421 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
422 if (data->wsize == 0)
423 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
424 /* NFSv3: we don't have bsize, but rather rtmult and wtmult... */
425 if (!fsinfo.bsize)
426 fsinfo.bsize = (fsinfo.rtmult>fsinfo.wtmult) ? fsinfo.rtmult : fsinfo.wtmult;
427 /* Also make sure we don't go below rsize/wsize since
428 * RPC calls are expensive */
429 if (fsinfo.bsize < server->rsize)
430 fsinfo.bsize = server->rsize;
431 if (fsinfo.bsize < server->wsize)
432 fsinfo.bsize = server->wsize;
433
434 if (data->bsize == 0)
435 sb->s_blocksize = nfs_block_bits(fsinfo.bsize, &sb->s_blocksize_bits);
436 if (server->rsize > fsinfo.rtmax)
437 server->rsize = fsinfo.rtmax;
438 if (server->wsize > fsinfo.wtmax)
439 server->wsize = fsinfo.wtmax;
440
441 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
442 if (server->rpages > NFS_READ_MAXIOV) {
443 server->rpages = NFS_READ_MAXIOV;
444 server->rsize = server->rpages << PAGE_CACHE_SHIFT;
445 }
446
447 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
448 if (server->wpages > NFS_WRITE_MAXIOV) {
449 server->wpages = NFS_WRITE_MAXIOV;
450 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
451 }
452
453 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
454 if (server->dtsize > PAGE_CACHE_SIZE)
455 server->dtsize = PAGE_CACHE_SIZE;
456 if (server->dtsize > server->rsize)
457 server->dtsize = server->rsize;
458
459 maxlen = (version == 2) ? NFS2_MAXNAMLEN : NFS3_MAXNAMLEN;
460
461 if (server->namelen == 0 || server->namelen > maxlen)
462 server->namelen = maxlen;
463
464 sb->s_maxbytes = fsinfo.maxfilesize;
465
466 /* Fire up the writeback cache */
467 if (nfs_reqlist_alloc(server) < 0) {
468 printk(KERN_NOTICE "NFS: cannot initialize writeback cache.\n");
469 goto failure_kill_reqlist;
470 }
471
472 /* We're airborne */
473
474 /* Check whether to start the lockd process */
475 if (!(server->flags & NFS_MOUNT_NONLM))
476 lockd_up();
477 return sb;
478
479 /* Yargs. It didn't work out. */
480 failure_kill_reqlist:
481 nfs_reqlist_exit(server);
482 out_no_root:
483 printk("nfs_read_super: get root inode failed\n");
484 iput(root_inode);
485 rpciod_down();
486 goto out_shutdown;
487
488 out_no_iod:
489 printk(KERN_WARNING "NFS: couldn't start rpciod!\n");
490 out_shutdown:
491 rpc_shutdown_client(server->client);
492 goto out_free_host;
493
494 out_no_client:
495 printk(KERN_WARNING "NFS: cannot create RPC client.\n");
496 xprt_destroy(xprt);
497 goto out_free_host;
498
499 out_no_xprt:
500 printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
501
502 out_free_host:
503 nfs_reqlist_free(server);
504 kfree(server->hostname);
505 out_unlock:
506 goto out_fail;
507
508 out_no_remote:
509 printk("NFS: mount program didn't pass remote address!\n");
510 goto out_fail;
511
512 out_miss_args:
513 printk("nfs_read_super: missing data argument\n");
514
515 out_fail:
516 return NULL;
517 }
518
519 static int
520 nfs_statfs(struct super_block *sb, struct statfs *buf)
521 {
522 struct nfs_server *server = &sb->u.nfs_sb.s_server;
523 unsigned char blockbits;
524 unsigned long blockres;
525 struct nfs_fsinfo res;
526 int error;
527
528 error = server->rpc_ops->statfs(server, NFS_FH(sb->s_root->d_inode), &res);
529 buf->f_type = NFS_SUPER_MAGIC;
530 if (error < 0)
531 goto out_err;
532
533 if (res.bsize == 0)
534 res.bsize = sb->s_blocksize;
535 buf->f_bsize = nfs_block_bits(res.bsize, &blockbits);
536 blockres = (1 << blockbits) - 1;
537 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
538 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
539 buf->f_bavail = (res.abytes + blockres) >> blockbits;
540 buf->f_files = res.tfiles;
541 buf->f_ffree = res.afiles;
542 if (res.namelen == 0 || res.namelen > server->namelen)
543 res.namelen = server->namelen;
544 buf->f_namelen = res.namelen;
545 return 0;
546 out_err:
547 printk("nfs_statfs: statfs error = %d\n", -error);
548 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
549 return 0;
550 }
551
552 /*
553 * Invalidate the local caches
554 */
555 void
556 nfs_zap_caches(struct inode *inode)
557 {
558 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
559 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
560
561 invalidate_inode_pages(inode);
562
563 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
564 NFS_CACHEINV(inode);
565 }
566
567 /*
568 * Invalidate, but do not unhash, the inode
569 */
570 static void
571 nfs_invalidate_inode(struct inode *inode)
572 {
573 umode_t save_mode = inode->i_mode;
574
575 make_bad_inode(inode);
576 inode->i_mode = save_mode;
577 nfs_zap_caches(inode);
578 }
579
580 /*
581 * Fill in inode information from the fattr.
582 */
583 static void
584 nfs_fill_inode(struct inode *inode, struct nfs_fh *fh, struct nfs_fattr *fattr)
585 {
586 /*
587 * Check whether the mode has been set, as we only want to
588 * do this once. (We don't allow inodes to change types.)
589 */
590 if (inode->i_mode == 0) {
591 NFS_FILEID(inode) = fattr->fileid;
592 NFS_FSID(inode) = fattr->fsid;
593 inode->i_mode = fattr->mode;
594 /* Why so? Because we want revalidate for devices/FIFOs, and
595 * that's precisely what we have in nfs_file_inode_operations.
596 */
597 inode->i_op = &nfs_file_inode_operations;
598 if (S_ISREG(inode->i_mode)) {
599 inode->i_fop = &nfs_file_operations;
600 inode->i_data.a_ops = &nfs_file_aops;
601 } else if (S_ISDIR(inode->i_mode)) {
602 inode->i_op = &nfs_dir_inode_operations;
603 inode->i_fop = &nfs_dir_operations;
604 } else if (S_ISLNK(inode->i_mode))
605 inode->i_op = &nfs_symlink_inode_operations;
606 else
607 init_special_inode(inode, inode->i_mode, fattr->rdev);
608 /*
609 * Preset the size and mtime, as there's no need
610 * to invalidate the caches.
611 */
612 inode->i_size = nfs_size_to_loff_t(fattr->size);
613 inode->i_mtime = nfs_time_to_secs(fattr->mtime);
614 inode->i_atime = nfs_time_to_secs(fattr->atime);
615 inode->i_ctime = nfs_time_to_secs(fattr->ctime);
616 NFS_CACHE_CTIME(inode) = fattr->ctime;
617 NFS_CACHE_MTIME(inode) = fattr->mtime;
618 NFS_CACHE_ISIZE(inode) = fattr->size;
619 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
620 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
621 memcpy(&inode->u.nfs_i.fh, fh, sizeof(inode->u.nfs_i.fh));
622 }
623 nfs_refresh_inode(inode, fattr);
624 }
625
626 struct nfs_find_desc {
627 struct nfs_fh *fh;
628 struct nfs_fattr *fattr;
629 };
630
631 /*
632 * In NFSv3 we can have 64bit inode numbers. In order to support
633 * this, and re-exported directories (also seen in NFSv2)
634 * we are forced to allow 2 different inodes to have the same
635 * i_ino.
636 */
637 static int
638 nfs_find_actor(struct inode *inode, unsigned long ino, void *opaque)
639 {
640 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
641 struct nfs_fh *fh = desc->fh;
642 struct nfs_fattr *fattr = desc->fattr;
643
644 if (NFS_FSID(inode) != fattr->fsid)
645 return 0;
646 if (NFS_FILEID(inode) != fattr->fileid)
647 return 0;
648 if (memcmp(&inode->u.nfs_i.fh, fh, sizeof(inode->u.nfs_i.fh)) != 0)
649 return 0;
650 return 1;
651 }
652
653 int
654 nfs_inode_is_stale(struct inode *inode, struct nfs_fh *fh, struct nfs_fattr *fattr)
655 {
656 /* Empty inodes are not stale */
657 if (!inode->i_mode)
658 return 0;
659
660 if ((fattr->mode & S_IFMT) != (inode->i_mode & S_IFMT))
661 return 1;
662
663 if (is_bad_inode(inode) || NFS_STALE(inode))
664 return 1;
665
666 /* Has the filehandle changed? If so is the old one stale? */
667 if (memcmp(&inode->u.nfs_i.fh, fh, sizeof(inode->u.nfs_i.fh)) != 0 &&
668 __nfs_revalidate_inode(NFS_SERVER(inode),inode) == -ESTALE)
669 return 1;
670
671 return 0;
672 }
673
674 /*
675 * This is our own version of iget that looks up inodes by file handle
676 * instead of inode number. We use this technique instead of using
677 * the vfs read_inode function because there is no way to pass the
678 * file handle or current attributes into the read_inode function.
679 *
680 */
681 struct inode *
682 nfs_fhget(struct dentry *dentry, struct nfs_fh *fhandle,
683 struct nfs_fattr *fattr)
684 {
685 struct super_block *sb = dentry->d_sb;
686
687 dprintk("NFS: nfs_fhget(%s/%s fileid=%Ld)\n",
688 dentry->d_parent->d_name.name, dentry->d_name.name,
689 (long long)fattr->fileid);
690 return __nfs_fhget(sb, fhandle, fattr);
691 }
692
693 /*
694 * Look up the inode by super block and fattr->fileid.
695 */
696 static struct inode *
697 __nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
698 {
699 struct nfs_find_desc desc = { fh, fattr };
700 struct inode *inode = NULL;
701 unsigned long ino;
702
703 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
704 goto out_no_inode;
705
706 if (!fattr->nlink) {
707 printk("NFS: Buggy server - nlink == 0!\n");
708 goto out_no_inode;
709 }
710
711 ino = nfs_fattr_to_ino_t(fattr);
712
713 if (!(inode = iget4(sb, ino, nfs_find_actor, &desc)))
714 goto out_no_inode;
715
716 nfs_fill_inode(inode, fh, fattr);
717 dprintk("NFS: __nfs_fhget(%x/%Ld ct=%d)\n",
718 inode->i_dev, (long long)NFS_FILEID(inode),
719 atomic_read(&inode->i_count));
720
721 out:
722 return inode;
723
724 out_no_inode:
725 printk("__nfs_fhget: iget failed\n");
726 goto out;
727 }
728
729 int
730 nfs_notify_change(struct dentry *dentry, struct iattr *attr)
731 {
732 struct inode *inode = dentry->d_inode;
733 struct nfs_fattr fattr;
734 int error;
735
736 /*
737 * Make sure the inode is up-to-date.
738 */
739 error = nfs_revalidate(dentry);
740 if (error) {
741 #ifdef NFS_PARANOIA
742 printk("nfs_notify_change: revalidate failed, error=%d\n", error);
743 #endif
744 goto out;
745 }
746
747 if (!S_ISREG(inode->i_mode))
748 attr->ia_valid &= ~ATTR_SIZE;
749
750 error = nfs_wb_all(inode);
751 if (error)
752 goto out;
753
754 error = NFS_PROTO(inode)->setattr(inode, &fattr, attr);
755 if (error)
756 goto out;
757 /*
758 * If we changed the size or mtime, update the inode
759 * now to avoid invalidating the page cache.
760 */
761 if (attr->ia_valid & ATTR_SIZE) {
762 if (attr->ia_size != fattr.size)
763 printk("nfs_notify_change: attr=%Ld, fattr=%Ld??\n",
764 (long long) attr->ia_size, (long long)fattr.size);
765 vmtruncate(inode, attr->ia_size);
766 }
767
768 /*
769 * If we changed the size or mtime, update the inode
770 * now to avoid invalidating the page cache.
771 */
772 if (!(fattr.valid & NFS_ATTR_WCC)) {
773 fattr.pre_size = NFS_CACHE_ISIZE(inode);
774 fattr.pre_mtime = NFS_CACHE_MTIME(inode);
775 fattr.pre_ctime = NFS_CACHE_CTIME(inode);
776 fattr.valid |= NFS_ATTR_WCC;
777 }
778 error = nfs_refresh_inode(inode, &fattr);
779 out:
780 return error;
781 }
782
783 /*
784 * Wait for the inode to get unlocked.
785 * (Used for NFS_INO_LOCKED and NFS_INO_REVALIDATING).
786 */
787 int
788 nfs_wait_on_inode(struct inode *inode, int flag)
789 {
790 struct rpc_clnt *clnt = NFS_CLIENT(inode);
791 int error;
792 if (!(NFS_FLAGS(inode) & flag))
793 return 0;
794 atomic_inc(&inode->i_count);
795 error = nfs_wait_event(clnt, inode->i_wait, !(NFS_FLAGS(inode) & flag));
796 iput(inode);
797 return error;
798 }
799
800 /*
801 * Externally visible revalidation function
802 */
803 int
804 nfs_revalidate(struct dentry *dentry)
805 {
806 struct inode *inode = dentry->d_inode;
807 return nfs_revalidate_inode(NFS_SERVER(inode), inode);
808 }
809
810 /*
811 * Ensure that mmap has a recent RPC credential for use when writing out
812 * shared pages
813 */
814 static inline void
815 nfs_set_mmcred(struct inode *inode, struct rpc_cred *cred)
816 {
817 struct rpc_cred **p = &NFS_I(inode)->mm_cred,
818 *oldcred = *p;
819
820 *p = get_rpccred(cred);
821 if (oldcred)
822 put_rpccred(oldcred);
823 }
824
825 /*
826 * These are probably going to contain hooks for
827 * allocating and releasing RPC credentials for
828 * the file. I'll have to think about Tronds patch
829 * a bit more..
830 */
831 int nfs_open(struct inode *inode, struct file *filp)
832 {
833 struct rpc_auth *auth;
834 struct rpc_cred *cred;
835
836 lock_kernel();
837 auth = NFS_CLIENT(inode)->cl_auth;
838 cred = rpcauth_lookupcred(auth, 0);
839 filp->private_data = cred;
840 if (filp->f_mode & FMODE_WRITE)
841 nfs_set_mmcred(inode, cred);
842 unlock_kernel();
843 return 0;
844 }
845
846 int nfs_release(struct inode *inode, struct file *filp)
847 {
848 struct rpc_cred *cred;
849
850 lock_kernel();
851 cred = nfs_file_cred(filp);
852 if (cred)
853 put_rpccred(cred);
854 unlock_kernel();
855 return 0;
856 }
857
858 /*
859 * This function is called whenever some part of NFS notices that
860 * the cached attributes have to be refreshed.
861 */
862 int
863 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
864 {
865 int status = -ESTALE;
866 struct nfs_fattr fattr;
867
868 dfprintk(PAGECACHE, "NFS: revalidating (%x/%Ld)\n",
869 inode->i_dev, (long long)NFS_FILEID(inode));
870
871 lock_kernel();
872 if (!inode || is_bad_inode(inode))
873 goto out_nowait;
874 if (NFS_STALE(inode) && inode != inode->i_sb->s_root->d_inode)
875 goto out_nowait;
876
877 while (NFS_REVALIDATING(inode)) {
878 status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
879 if (status < 0)
880 goto out_nowait;
881 if (time_before(jiffies,NFS_READTIME(inode)+NFS_ATTRTIMEO(inode))) {
882 status = NFS_STALE(inode) ? -ESTALE : 0;
883 goto out_nowait;
884 }
885 }
886 NFS_FLAGS(inode) |= NFS_INO_REVALIDATING;
887
888 status = NFS_PROTO(inode)->getattr(inode, &fattr);
889 if (status) {
890 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%x/%Ld) getattr failed, error=%d\n",
891 inode->i_dev, (long long)NFS_FILEID(inode), status);
892 if (status == -ESTALE) {
893 NFS_FLAGS(inode) |= NFS_INO_STALE;
894 if (inode != inode->i_sb->s_root->d_inode)
895 remove_inode_hash(inode);
896 }
897 goto out;
898 }
899
900 status = nfs_refresh_inode(inode, &fattr);
901 if (status) {
902 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%x/%Ld) refresh failed, error=%d\n",
903 inode->i_dev, (long long)NFS_FILEID(inode), status);
904 goto out;
905 }
906 dfprintk(PAGECACHE, "NFS: (%x/%Ld) revalidation complete\n",
907 inode->i_dev, (long long)NFS_FILEID(inode));
908
909 NFS_FLAGS(inode) &= ~NFS_INO_STALE;
910 out:
911 NFS_FLAGS(inode) &= ~NFS_INO_REVALIDATING;
912 wake_up(&inode->i_wait);
913 out_nowait:
914 unlock_kernel();
915 return status;
916 }
917
918 /*
919 * Many nfs protocol calls return the new file attributes after
920 * an operation. Here we update the inode to reflect the state
921 * of the server's inode.
922 *
923 * This is a bit tricky because we have to make sure all dirty pages
924 * have been sent off to the server before calling invalidate_inode_pages.
925 * To make sure no other process adds more write requests while we try
926 * our best to flush them, we make them sleep during the attribute refresh.
927 *
928 * A very similar scenario holds for the dir cache.
929 */
930 int
931 __nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
932 {
933 __u64 new_size, new_mtime;
934 loff_t new_isize;
935 int invalid = 0;
936
937 dfprintk(VFS, "NFS: refresh_inode(%x/%ld ct=%d info=0x%x)\n",
938 inode->i_dev, inode->i_ino,
939 atomic_read(&inode->i_count), fattr->valid);
940
941 if (NFS_FSID(inode) != fattr->fsid ||
942 NFS_FILEID(inode) != fattr->fileid) {
943 printk(KERN_ERR "nfs_refresh_inode: inode number mismatch\n"
944 "expected (0x%Lx/0x%Lx), got (0x%Lx/0x%Lx)\n",
945 (long long)NFS_FSID(inode), (long long)NFS_FILEID(inode),
946 (long long)fattr->fsid, (long long)fattr->fileid);
947 goto out_err;
948 }
949
950 /*
951 * Make sure the inode's type hasn't changed.
952 */
953 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
954 goto out_changed;
955
956 new_mtime = fattr->mtime;
957 new_size = fattr->size;
958 new_isize = nfs_size_to_loff_t(fattr->size);
959
960 /*
961 * Update the read time so we don't revalidate too often.
962 */
963 NFS_READTIME(inode) = jiffies;
964
965 /*
966 * Note: NFS_CACHE_ISIZE(inode) reflects the state of the cache.
967 * NOT inode->i_size!!!
968 */
969 if (NFS_CACHE_ISIZE(inode) != new_size) {
970 #ifdef NFS_DEBUG_VERBOSE
971 printk(KERN_DEBUG "NFS: isize change on %x/%ld\n", inode->i_dev, inode->i_ino);
972 #endif
973 invalid = 1;
974 }
975
976 /*
977 * Note: we don't check inode->i_mtime since pipes etc.
978 * can change this value in VFS without requiring a
979 * cache revalidation.
980 */
981 if (NFS_CACHE_MTIME(inode) != new_mtime) {
982 #ifdef NFS_DEBUG_VERBOSE
983 printk(KERN_DEBUG "NFS: mtime change on %x/%ld\n", inode->i_dev, inode->i_ino);
984 #endif
985 invalid = 1;
986 }
987
988 /* Check Weak Cache Consistency data.
989 * If size and mtime match the pre-operation values, we can
990 * assume that any attribute changes were caused by our NFS
991 * operation, so there's no need to invalidate the caches.
992 */
993 if ((fattr->valid & NFS_ATTR_WCC)
994 && NFS_CACHE_ISIZE(inode) == fattr->pre_size
995 && NFS_CACHE_MTIME(inode) == fattr->pre_mtime) {
996 invalid = 0;
997 }
998
999 /*
1000 * If we have pending writebacks, things can get
1001 * messy.
1002 */
1003 if (nfs_have_writebacks(inode) && new_isize < inode->i_size)
1004 new_isize = inode->i_size;
1005
1006 NFS_CACHE_CTIME(inode) = fattr->ctime;
1007 inode->i_ctime = nfs_time_to_secs(fattr->ctime);
1008
1009 inode->i_atime = nfs_time_to_secs(fattr->atime);
1010
1011 NFS_CACHE_MTIME(inode) = new_mtime;
1012 inode->i_mtime = nfs_time_to_secs(new_mtime);
1013
1014 NFS_CACHE_ISIZE(inode) = new_size;
1015 inode->i_size = new_isize;
1016
1017 inode->i_mode = fattr->mode;
1018 inode->i_nlink = fattr->nlink;
1019 inode->i_uid = fattr->uid;
1020 inode->i_gid = fattr->gid;
1021
1022 if (fattr->valid & NFS_ATTR_FATTR_V3) {
1023 /*
1024 * report the blocks in 512byte units
1025 */
1026 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1027 inode->i_blksize = inode->i_sb->s_blocksize;
1028 } else {
1029 inode->i_blocks = fattr->du.nfs2.blocks;
1030 inode->i_blksize = fattr->du.nfs2.blocksize;
1031 }
1032 inode->i_rdev = 0;
1033 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1034 inode->i_rdev = to_kdev_t(fattr->rdev);
1035
1036 /* Update attrtimeo value */
1037 if (!invalid && time_after(jiffies, NFS_ATTRTIMEO_UPDATE(inode)+NFS_ATTRTIMEO(inode))) {
1038 if ((NFS_ATTRTIMEO(inode) <<= 1) > NFS_MAXATTRTIMEO(inode))
1039 NFS_ATTRTIMEO(inode) = NFS_MAXATTRTIMEO(inode);
1040 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
1041 }
1042
1043 if (invalid)
1044 nfs_zap_caches(inode);
1045 return 0;
1046
1047 out_changed:
1048 /*
1049 * Big trouble! The inode has become a different object.
1050 */
1051 #ifdef NFS_PARANOIA
1052 printk(KERN_DEBUG "nfs_refresh_inode: inode %ld mode changed, %07o to %07o\n",
1053 inode->i_ino, inode->i_mode, fattr->mode);
1054 #endif
1055 /*
1056 * No need to worry about unhashing the dentry, as the
1057 * lookup validation will know that the inode is bad.
1058 * (But we fall through to invalidate the caches.)
1059 */
1060 nfs_invalidate_inode(inode);
1061 out_err:
1062 return -EIO;
1063 }
1064
1065 /*
1066 * File system information
1067 */
1068 static DECLARE_FSTYPE(nfs_fs_type, "nfs", nfs_read_super, FS_ODD_RENAME);
1069
1070 extern int nfs_init_nfspagecache(void);
1071 extern void nfs_destroy_nfspagecache(void);
1072 extern int nfs_init_readpagecache(void);
1073 extern int nfs_destroy_readpagecache(void);
1074
1075 /*
1076 * Initialize NFS
1077 */
1078 static int __init init_nfs_fs(void)
1079 {
1080 int err;
1081
1082 err = nfs_init_nfspagecache();
1083 if (err)
1084 return err;
1085
1086 err = nfs_init_readpagecache();
1087 if (err)
1088 return err;
1089
1090 #ifdef CONFIG_PROC_FS
1091 rpc_proc_register(&nfs_rpcstat);
1092 #endif
1093 return register_filesystem(&nfs_fs_type);
1094 }
1095
1096 static void __exit exit_nfs_fs(void)
1097 {
1098 nfs_destroy_readpagecache();
1099 nfs_destroy_nfspagecache();
1100 #ifdef CONFIG_PROC_FS
1101 rpc_proc_unregister("nfs");
1102 #endif
1103 unregister_filesystem(&nfs_fs_type);
1104 }
1105
1106 EXPORT_NO_SYMBOLS;
1107 /* Not quite true; I just maintain it */
1108 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1109
1110 module_init(init_nfs_fs)
1111 module_exit(exit_nfs_fs)
1112