projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
NFS: Clean up weak cache consistency code
[linux-2.6-block.git] / fs / nfs / inode.c
CommitLineData
1da177e4
LT		 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
4ce79717 42#include "nfs4_fs.h"
1da177e4
LT		 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
56static void nfs_invalidate_inode(struct inode *);
24aa1fe6 57static int nfs_update_inode(struct inode *, struct nfs_fattr *);
1da177e4
LT		 58
59static struct inode *nfs_alloc_inode(struct super_block *sb);
60static void nfs_destroy_inode(struct inode *);
61static int nfs_write_inode(struct inode *,int);
62static void nfs_delete_inode(struct inode *);
63static void nfs_clear_inode(struct inode *);
64static void nfs_umount_begin(struct super_block *);
65static int nfs_statfs(struct super_block *, struct kstatfs *);
66static int nfs_show_options(struct seq_file *, struct vfsmount *);
ada70d94 67static void nfs_zap_acl_cache(struct inode *);
1da177e4
LT		 68
69static struct rpc_program nfs_program;
70
71static 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 */
85static struct rpc_stat nfs_rpcstat = {
86 .program = &nfs_program
87};
88static 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
102static 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
b7fa0554
AG		 111#ifdef CONFIG_NFS_V3_ACL
112static struct rpc_stat nfsacl_rpcstat = { &nfsacl_program };
113static struct rpc_version * nfsacl_version[] = {
114 [3] = &nfsacl_version3,
115};
116
117struct 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
1da177e4
LT		 126static inline unsigned long
127nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
128{
129 return nfs_fileid_to_ino_t(fattr->fileid);
130}
131
132static int
133nfs_write_inode(struct inode *inode, int sync)
134{
135 int flags = sync ? FLUSH_WAIT : 0;
136 int ret;
137
3da28eb1 138 ret = nfs_commit_inode(inode, flags);
1da177e4
LT		 139 if (ret < 0)
140 return ret;
141 return 0;
142}
143
144static void
145nfs_delete_inode(struct inode * inode)
146{
147 dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
148
fef26658
MF		 149 truncate_inode_pages(&inode->i_data, 0);
150
1da177e4
LT		 151 nfs_wb_all(inode);
152 /*
153 * The following should never happen...
154 */
155 if (nfs_have_writebacks(inode)) {
156 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
157 }
158
159 clear_inode(inode);
160}
161
1da177e4
LT		 162static void
163nfs_clear_inode(struct inode *inode)
164{
165 struct nfs_inode *nfsi = NFS_I(inode);
166 struct rpc_cred *cred;
167
168 nfs_wb_all(inode);
169 BUG_ON (!list_empty(&nfsi->open_files));
ada70d94 170 nfs_zap_acl_cache(inode);
1da177e4
LT		 171 cred = nfsi->cache_access.cred;
172 if (cred)
173 put_rpccred(cred);
174 BUG_ON(atomic_read(&nfsi->data_updates) != 0);
175}
176
177void
178nfs_umount_begin(struct super_block *sb)
179{
6a19275a 180 struct rpc_clnt *rpc = NFS_SB(sb)->client;
1da177e4
LT		 181
182 /* -EIO all pending I/O */
6a19275a 183 if (!IS_ERR(rpc))
1da177e4 184 rpc_killall_tasks(rpc);
b7fa0554
AG		 185 rpc = NFS_SB(sb)->client_acl;
186 if (!IS_ERR(rpc))
187 rpc_killall_tasks(rpc);
1da177e4
LT		 188}
189
190
191static inline unsigned long
192nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
193{
194 /* make sure blocksize is a power of two */
195 if ((bsize & (bsize - 1)) || nrbitsp) {
196 unsigned char nrbits;
197
198 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
199 ;
200 bsize = 1 << nrbits;
201 if (nrbitsp)
202 *nrbitsp = nrbits;
203 }
204
205 return bsize;
206}
207
208/*
209 * Calculate the number of 512byte blocks used.
210 */
211static inline unsigned long
212nfs_calc_block_size(u64 tsize)
213{
214 loff_t used = (tsize + 511) >> 9;
215 return (used > ULONG_MAX) ? ULONG_MAX : used;
216}
217
218/*
219 * Compute and set NFS server blocksize
220 */
221static inline unsigned long
222nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
223{
40859d7e
CL		 224 if (bsize < NFS_MIN_FILE_IO_SIZE)
225 bsize = NFS_DEF_FILE_IO_SIZE;
226 else if (bsize >= NFS_MAX_FILE_IO_SIZE)
227 bsize = NFS_MAX_FILE_IO_SIZE;
1da177e4
LT		 228
229 return nfs_block_bits(bsize, nrbitsp);
230}
231
232/*
233 * Obtain the root inode of the file system.
234 */
235static struct inode *
236nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
237{
238 struct nfs_server *server = NFS_SB(sb);
239 struct inode *rooti;
240 int error;
241
242 error = server->rpc_ops->getroot(server, rootfh, fsinfo);
243 if (error < 0) {
244 dprintk("nfs_get_root: getattr error = %d\n", -error);
245 return ERR_PTR(error);
246 }
247
248 rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
249 if (!rooti)
250 return ERR_PTR(-ENOMEM);
251 return rooti;
252}
253
254/*
255 * Do NFS version-independent mount processing, and sanity checking
256 */
257static int
258nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
259{
260 struct nfs_server *server;
261 struct inode *root_inode;
262 struct nfs_fattr fattr;
263 struct nfs_fsinfo fsinfo = {
264 .fattr = &fattr,
265 };
266 struct nfs_pathconf pathinfo = {
267 .fattr = &fattr,
268 };
269 int no_root_error = 0;
270 unsigned long max_rpc_payload;
271
272 /* We probably want something more informative here */
273 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
274
275 server = NFS_SB(sb);
276
277 sb->s_magic = NFS_SUPER_MAGIC;
278
279 root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
280 /* Did getting the root inode fail? */
281 if (IS_ERR(root_inode)) {
282 no_root_error = PTR_ERR(root_inode);
283 goto out_no_root;
284 }
285 sb->s_root = d_alloc_root(root_inode);
286 if (!sb->s_root) {
287 no_root_error = -ENOMEM;
288 goto out_no_root;
289 }
290 sb->s_root->d_op = server->rpc_ops->dentry_ops;
291
292 /* Get some general file system info */
293 if (server->namelen == 0 &&
294 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
295 server->namelen = pathinfo.max_namelen;
296 /* Work out a lot of parameters */
297 if (server->rsize == 0)
298 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
299 if (server->wsize == 0)
300 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
301
302 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
303 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
304 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
305 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
306
307 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
308 if (server->rsize > max_rpc_payload)
309 server->rsize = max_rpc_payload;
40859d7e
CL		 310 if (server->rsize > NFS_MAX_FILE_IO_SIZE)
311 server->rsize = NFS_MAX_FILE_IO_SIZE;
1da177e4 312 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1da177e4 313
40859d7e
CL		 314 if (server->wsize > max_rpc_payload)
315 server->wsize = max_rpc_payload;
316 if (server->wsize > NFS_MAX_FILE_IO_SIZE)
317 server->wsize = NFS_MAX_FILE_IO_SIZE;
1da177e4 318 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1da177e4
LT		 319
320 if (sb->s_blocksize == 0)
321 sb->s_blocksize = nfs_block_bits(server->wsize,
322 &sb->s_blocksize_bits);
323 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
324
325 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
326 if (server->dtsize > PAGE_CACHE_SIZE)
327 server->dtsize = PAGE_CACHE_SIZE;
328 if (server->dtsize > server->rsize)
329 server->dtsize = server->rsize;
330
331 if (server->flags & NFS_MOUNT_NOAC) {
332 server->acregmin = server->acregmax = 0;
333 server->acdirmin = server->acdirmax = 0;
334 sb->s_flags |= MS_SYNCHRONOUS;
335 }
336 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
337
338 sb->s_maxbytes = fsinfo.maxfilesize;
339 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
340 sb->s_maxbytes = MAX_LFS_FILESIZE;
341
342 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
343 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
344
345 /* We're airborne Set socket buffersize */
346 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
347 return 0;
348 /* Yargs. It didn't work out. */
349out_no_root:
350 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
351 if (!IS_ERR(root_inode))
352 iput(root_inode);
353 return no_root_error;
354}
355
eab5c084
CL		 356static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
357{
358 to->to_initval = timeo * HZ / 10;
359 to->to_retries = retrans;
360 if (!to->to_retries)
361 to->to_retries = 2;
362
363 switch (proto) {
364 case IPPROTO_TCP:
365 if (!to->to_initval)
366 to->to_initval = 60 * HZ;
03bf4b70
CL		 367 if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
368 to->to_initval = NFS_MAX_TCP_TIMEOUT;
eab5c084
CL		 369 to->to_increment = to->to_initval;
370 to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
371 to->to_exponential = 0;
372 break;
373 case IPPROTO_UDP:
374 default:
375 if (!to->to_initval)
376 to->to_initval = 11 * HZ / 10;
03bf4b70
CL		 377 if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
378 to->to_initval = NFS_MAX_UDP_TIMEOUT;
379 to->to_maxval = NFS_MAX_UDP_TIMEOUT;
eab5c084
CL		 380 to->to_exponential = 1;
381 break;
382 }
383}
384
1da177e4
LT		 385/*
386 * Create an RPC client handle.
387 */
388static struct rpc_clnt *
389nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
390{
391 struct rpc_timeout timeparms;
392 struct rpc_xprt *xprt = NULL;
393 struct rpc_clnt *clnt = NULL;
eab5c084 394 int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
1da177e4 395
eab5c084 396 nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
1da177e4
LT		 397
398 /* create transport and client */
eab5c084 399 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1da177e4 400 if (IS_ERR(xprt)) {
9085bbcb
TM		 401 dprintk("%s: cannot create RPC transport. Error = %ld\n",
402 __FUNCTION__, PTR_ERR(xprt));
1da177e4
LT		 403 return (struct rpc_clnt *)xprt;
404 }
405 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
406 server->rpc_ops->version, data->pseudoflavor);
407 if (IS_ERR(clnt)) {
9085bbcb
TM		 408 dprintk("%s: cannot create RPC client. Error = %ld\n",
409 __FUNCTION__, PTR_ERR(xprt));
1da177e4
LT		 410 goto out_fail;
411 }
412
413 clnt->cl_intr = 1;
414 clnt->cl_softrtry = 1;
415 clnt->cl_chatty = 1;
416
417 return clnt;
418
419out_fail:
1da177e4
LT		 420 return clnt;
421}
422
423/*
424 * The way this works is that the mount process passes a structure
425 * in the data argument which contains the server's IP address
426 * and the root file handle obtained from the server's mount
427 * daemon. We stash these away in the private superblock fields.
428 */
429static int
430nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
431{
432 struct nfs_server *server;
433 rpc_authflavor_t authflavor;
434
435 server = NFS_SB(sb);
436 sb->s_blocksize_bits = 0;
437 sb->s_blocksize = 0;
438 if (data->bsize)
439 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
440 if (data->rsize)
441 server->rsize = nfs_block_size(data->rsize, NULL);
442 if (data->wsize)
443 server->wsize = nfs_block_size(data->wsize, NULL);
444 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
445
446 server->acregmin = data->acregmin*HZ;
447 server->acregmax = data->acregmax*HZ;
448 server->acdirmin = data->acdirmin*HZ;
449 server->acdirmax = data->acdirmax*HZ;
450
451 /* Start lockd here, before we might error out */
452 if (!(server->flags & NFS_MOUNT_NONLM))
453 lockd_up();
454
455 server->namelen = data->namlen;
456 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
457 if (!server->hostname)
458 return -ENOMEM;
459 strcpy(server->hostname, data->hostname);
460
461 /* Check NFS protocol revision and initialize RPC op vector
462 * and file handle pool. */
1da177e4 463#ifdef CONFIG_NFS_V3
9085bbcb 464 if (server->flags & NFS_MOUNT_VER3) {
1da177e4
LT		 465 server->rpc_ops = &nfs_v3_clientops;
466 server->caps |= NFS_CAP_READDIRPLUS;
1da177e4
LT		 467 } else {
468 server->rpc_ops = &nfs_v2_clientops;
469 }
9085bbcb
TM		 470#else
471 server->rpc_ops = &nfs_v2_clientops;
472#endif
1da177e4
LT		 473
474 /* Fill in pseudoflavor for mount version < 5 */
475 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
476 data->pseudoflavor = RPC_AUTH_UNIX;
477 authflavor = data->pseudoflavor; /* save for sb_init() */
478 /* XXX maybe we want to add a server->pseudoflavor field */
479
480 /* Create RPC client handles */
481 server->client = nfs_create_client(server, data);
482 if (IS_ERR(server->client))
483 return PTR_ERR(server->client);
484 /* RFC 2623, sec 2.3.2 */
485 if (authflavor != RPC_AUTH_UNIX) {
6a19275a
BF		 486 struct rpc_auth *auth;
487
1da177e4
LT		 488 server->client_sys = rpc_clone_client(server->client);
489 if (IS_ERR(server->client_sys))
490 return PTR_ERR(server->client_sys);
6a19275a
BF		 491 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
492 if (IS_ERR(auth))
493 return PTR_ERR(auth);
1da177e4
LT		 494 } else {
495 atomic_inc(&server->client->cl_count);
496 server->client_sys = server->client;
497 }
1da177e4 498 if (server->flags & NFS_MOUNT_VER3) {
b7fa0554
AG		 499#ifdef CONFIG_NFS_V3_ACL
500 if (!(server->flags & NFS_MOUNT_NOACL)) {
501 server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
502 /* No errors! Assume that Sun nfsacls are supported */
503 if (!IS_ERR(server->client_acl))
504 server->caps |= NFS_CAP_ACLS;
505 }
506#else
507 server->flags &= ~NFS_MOUNT_NOACL;
508#endif /* CONFIG_NFS_V3_ACL */
055ffbea
AG		 509 /*
510 * The VFS shouldn't apply the umask to mode bits. We will
511 * do so ourselves when necessary.
512 */
513 sb->s_flags |= MS_POSIXACL;
1da177e4
LT		 514 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
515 server->namelen = NFS3_MAXNAMLEN;
516 sb->s_time_gran = 1;
517 } else {
518 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
519 server->namelen = NFS2_MAXNAMLEN;
520 }
521
522 sb->s_op = &nfs_sops;
523 return nfs_sb_init(sb, authflavor);
524}
525
526static int
527nfs_statfs(struct super_block *sb, struct kstatfs *buf)
528{
529 struct nfs_server *server = NFS_SB(sb);
530 unsigned char blockbits;
531 unsigned long blockres;
532 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
533 struct nfs_fattr fattr;
534 struct nfs_fsstat res = {
535 .fattr = &fattr,
536 };
537 int error;
538
539 lock_kernel();
540
541 error = server->rpc_ops->statfs(server, rootfh, &res);
542 buf->f_type = NFS_SUPER_MAGIC;
543 if (error < 0)
544 goto out_err;
545
546 /*
547 * Current versions of glibc do not correctly handle the
548 * case where f_frsize != f_bsize. Eventually we want to
549 * report the value of wtmult in this field.
550 */
551 buf->f_frsize = sb->s_blocksize;
552
553 /*
554 * On most *nix systems, f_blocks, f_bfree, and f_bavail
555 * are reported in units of f_frsize. Linux hasn't had
556 * an f_frsize field in its statfs struct until recently,
557 * thus historically Linux's sys_statfs reports these
558 * fields in units of f_bsize.
559 */
560 buf->f_bsize = sb->s_blocksize;
561 blockbits = sb->s_blocksize_bits;
562 blockres = (1 << blockbits) - 1;
563 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
564 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
565 buf->f_bavail = (res.abytes + blockres) >> blockbits;
566
567 buf->f_files = res.tfiles;
568 buf->f_ffree = res.afiles;
569
570 buf->f_namelen = server->namelen;
571 out:
572 unlock_kernel();
1da177e4
LT		 573 return 0;
574
575 out_err:
dc20f803 576 dprintk("%s: statfs error = %d\n", __FUNCTION__, -error);
1da177e4
LT		 577 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
578 goto out;
579
580}
581
582static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
583{
584 static struct proc_nfs_info {
585 int flag;
586 char *str;
587 char *nostr;
588 } nfs_info[] = {
589 { NFS_MOUNT_SOFT, ",soft", ",hard" },
590 { NFS_MOUNT_INTR, ",intr", "" },
591 { NFS_MOUNT_POSIX, ",posix", "" },
1da177e4
LT		 592 { NFS_MOUNT_NOCTO, ",nocto", "" },
593 { NFS_MOUNT_NOAC, ",noac", "" },
594 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
b7fa0554 595 { NFS_MOUNT_NOACL, ",noacl", "" },
1da177e4
LT		 596 { 0, NULL, NULL }
597 };
598 struct proc_nfs_info *nfs_infop;
599 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
3063d8a1
TM		 600 char buf[12];
601 char *proto;
1da177e4
LT		 602
603 seq_printf(m, ",v%d", nfss->rpc_ops->version);
604 seq_printf(m, ",rsize=%d", nfss->rsize);
605 seq_printf(m, ",wsize=%d", nfss->wsize);
606 if (nfss->acregmin != 3*HZ)
607 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
608 if (nfss->acregmax != 60*HZ)
609 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
610 if (nfss->acdirmin != 30*HZ)
611 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
612 if (nfss->acdirmax != 60*HZ)
613 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
614 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
615 if (nfss->flags & nfs_infop->flag)
616 seq_puts(m, nfs_infop->str);
617 else
618 seq_puts(m, nfs_infop->nostr);
619 }
3063d8a1
TM		 620 switch (nfss->client->cl_xprt->prot) {
621 case IPPROTO_TCP:
622 proto = "tcp";
623 break;
624 case IPPROTO_UDP:
625 proto = "udp";
626 break;
627 default:
628 snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
629 proto = buf;
630 }
631 seq_printf(m, ",proto=%s", proto);
1da177e4
LT		 632 seq_puts(m, ",addr=");
633 seq_escape(m, nfss->hostname, " \t\n\\");
634 return 0;
635}
636
29884df0
TM		 637/**
638 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
639 */
640int nfs_sync_mapping(struct address_space *mapping)
641{
642 int ret;
643
644 if (mapping->nrpages == 0)
645 return 0;
646 unmap_mapping_range(mapping, 0, 0, 0);
647 ret = filemap_fdatawrite(mapping);
648 if (ret != 0)
649 goto out;
650 ret = filemap_fdatawait(mapping);
651 if (ret != 0)
652 goto out;
653 ret = nfs_wb_all(mapping->host);
654out:
655 return ret;
656}
657
1da177e4
LT		 658/*
659 * Invalidate the local caches
660 */
b37b03b7 661static void nfs_zap_caches_locked(struct inode *inode)
1da177e4
LT		 662{
663 struct nfs_inode *nfsi = NFS_I(inode);
664 int mode = inode->i_mode;
665
666 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
667 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
668
669 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
670 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
55296809 671 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
1da177e4 672 else
55296809 673 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
b37b03b7 674}
dc59250c 675
b37b03b7
TM		 676void nfs_zap_caches(struct inode *inode)
677{
678 spin_lock(&inode->i_lock);
679 nfs_zap_caches_locked(inode);
dc59250c 680 spin_unlock(&inode->i_lock);
ada70d94
TM		 681}
682
683static void nfs_zap_acl_cache(struct inode *inode)
684{
685 void (*clear_acl_cache)(struct inode *);
686
687 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
688 if (clear_acl_cache != NULL)
689 clear_acl_cache(inode);
dc59250c 690 spin_lock(&inode->i_lock);
55296809 691 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
dc59250c 692 spin_unlock(&inode->i_lock);
1da177e4
LT		 693}
694
695/*
b37b03b7
TM		 696 * Invalidate, but do not unhash, the inode.
697 * NB: must be called with inode->i_lock held!
1da177e4 698 */
b37b03b7 699static void nfs_invalidate_inode(struct inode *inode)
1da177e4 700{
b37b03b7
TM		 701 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
702 nfs_zap_caches_locked(inode);
1da177e4
LT		 703}
704
705struct nfs_find_desc {
706 struct nfs_fh *fh;
707 struct nfs_fattr *fattr;
708};
709
710/*
711 * In NFSv3 we can have 64bit inode numbers. In order to support
712 * this, and re-exported directories (also seen in NFSv2)
713 * we are forced to allow 2 different inodes to have the same
714 * i_ino.
715 */
716static int
717nfs_find_actor(struct inode *inode, void *opaque)
718{
719 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
720 struct nfs_fh *fh = desc->fh;
721 struct nfs_fattr *fattr = desc->fattr;
722
723 if (NFS_FILEID(inode) != fattr->fileid)
724 return 0;
725 if (nfs_compare_fh(NFS_FH(inode), fh))
726 return 0;
727 if (is_bad_inode(inode) || NFS_STALE(inode))
728 return 0;
729 return 1;
730}
731
732static int
733nfs_init_locked(struct inode *inode, void *opaque)
734{
735 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
736 struct nfs_fattr *fattr = desc->fattr;
737
738 NFS_FILEID(inode) = fattr->fileid;
739 nfs_copy_fh(NFS_FH(inode), desc->fh);
740 return 0;
741}
742
743/* Don't use READDIRPLUS on directories that we believe are too large */
744#define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
745
746/*
747 * This is our front-end to iget that looks up inodes by file handle
748 * instead of inode number.
749 */
750struct inode *
751nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
752{
753 struct nfs_find_desc desc = {
754 .fh = fh,
755 .fattr = fattr
756 };
757 struct inode *inode = NULL;
758 unsigned long hash;
759
760 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
761 goto out_no_inode;
762
763 if (!fattr->nlink) {
764 printk("NFS: Buggy server - nlink == 0!\n");
765 goto out_no_inode;
766 }
767
768 hash = nfs_fattr_to_ino_t(fattr);
769
770 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
771 goto out_no_inode;
772
773 if (inode->i_state & I_NEW) {
774 struct nfs_inode *nfsi = NFS_I(inode);
775
776 /* We set i_ino for the few things that still rely on it,
777 * such as stat(2) */
778 inode->i_ino = hash;
779
780 /* We can't support update_atime(), since the server will reset it */
781 inode->i_flags |= S_NOATIME|S_NOCMTIME;
782 inode->i_mode = fattr->mode;
783 /* Why so? Because we want revalidate for devices/FIFOs, and
784 * that's precisely what we have in nfs_file_inode_operations.
785 */
92cfc62c 786 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
1da177e4
LT		 787 if (S_ISREG(inode->i_mode)) {
788 inode->i_fop = &nfs_file_operations;
789 inode->i_data.a_ops = &nfs_file_aops;
790 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
791 } else if (S_ISDIR(inode->i_mode)) {
792 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
793 inode->i_fop = &nfs_dir_operations;
794 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
795 && fattr->size <= NFS_LIMIT_READDIRPLUS)
412d582e 796 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
1da177e4
LT		 797 } else if (S_ISLNK(inode->i_mode))
798 inode->i_op = &nfs_symlink_inode_operations;
799 else
800 init_special_inode(inode, inode->i_mode, fattr->rdev);
801
33801147
TM		 802 nfsi->read_cache_jiffies = fattr->time_start;
803 nfsi->last_updated = jiffies;
1da177e4
LT		 804 inode->i_atime = fattr->atime;
805 inode->i_mtime = fattr->mtime;
806 inode->i_ctime = fattr->ctime;
807 if (fattr->valid & NFS_ATTR_FATTR_V4)
808 nfsi->change_attr = fattr->change_attr;
809 inode->i_size = nfs_size_to_loff_t(fattr->size);
810 inode->i_nlink = fattr->nlink;
811 inode->i_uid = fattr->uid;
812 inode->i_gid = fattr->gid;
813 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
814 /*
815 * report the blocks in 512byte units
816 */
817 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
818 inode->i_blksize = inode->i_sb->s_blocksize;
819 } else {
820 inode->i_blocks = fattr->du.nfs2.blocks;
821 inode->i_blksize = fattr->du.nfs2.blocksize;
822 }
823 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
824 nfsi->attrtimeo_timestamp = jiffies;
825 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
826 nfsi->cache_access.cred = NULL;
827
828 unlock_new_inode(inode);
829 } else
830 nfs_refresh_inode(inode, fattr);
831 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
832 inode->i_sb->s_id,
833 (long long)NFS_FILEID(inode),
834 atomic_read(&inode->i_count));
835
836out:
837 return inode;
838
839out_no_inode:
840 printk("nfs_fhget: iget failed\n");
841 goto out;
842}
843
844#define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
845
846int
847nfs_setattr(struct dentry *dentry, struct iattr *attr)
848{
849 struct inode *inode = dentry->d_inode;
850 struct nfs_fattr fattr;
851 int error;
852
853 if (attr->ia_valid & ATTR_SIZE) {
854 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
855 attr->ia_valid &= ~ATTR_SIZE;
856 }
857
858 /* Optimization: if the end result is no change, don't RPC */
859 attr->ia_valid &= NFS_VALID_ATTRS;
860 if (attr->ia_valid == 0)
861 return 0;
862
863 lock_kernel();
864 nfs_begin_data_update(inode);
865 /* Write all dirty data if we're changing file permissions or size */
866 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
867 if (filemap_fdatawrite(inode->i_mapping) == 0)
868 filemap_fdatawait(inode->i_mapping);
869 nfs_wb_all(inode);
870 }
642ac549
TM		 871 /*
872 * Return any delegations if we're going to change ACLs
873 */
874 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
875 nfs_inode_return_delegation(inode);
1da177e4 876 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
65e4308d 877 if (error == 0)
1da177e4 878 nfs_refresh_inode(inode, &fattr);
65e4308d
TM		 879 nfs_end_data_update(inode);
880 unlock_kernel();
881 return error;
882}
883
884/**
885 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
886 * @inode: pointer to struct inode
887 * @attr: pointer to struct iattr
888 *
889 * Note: we do this in the *proc.c in order to ensure that
890 * it works for things like exclusive creates too.
891 */
892void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
893{
894 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
1da177e4 895 if ((attr->ia_valid & ATTR_MODE) != 0) {
65e4308d
TM		 896 int mode = attr->ia_mode & S_IALLUGO;
897 mode |= inode->i_mode & ~S_IALLUGO;
1da177e4
LT		 898 inode->i_mode = mode;
899 }
900 if ((attr->ia_valid & ATTR_UID) != 0)
901 inode->i_uid = attr->ia_uid;
902 if ((attr->ia_valid & ATTR_GID) != 0)
903 inode->i_gid = attr->ia_gid;
dc59250c 904 spin_lock(&inode->i_lock);
55296809 905 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
dc59250c 906 spin_unlock(&inode->i_lock);
65e4308d
TM		 907 }
908 if ((attr->ia_valid & ATTR_SIZE) != 0) {
909 inode->i_size = attr->ia_size;
910 vmtruncate(inode, attr->ia_size);
911 }
1da177e4
LT		 912}
913
412d582e
CL		 914static int nfs_wait_schedule(void *word)
915{
916 if (signal_pending(current))
917 return -ERESTARTSYS;
918 schedule();
919 return 0;
920}
921
1da177e4
LT		 922/*
923 * Wait for the inode to get unlocked.
1da177e4 924 */
412d582e 925static int nfs_wait_on_inode(struct inode *inode)
1da177e4
LT		 926{
927 struct rpc_clnt *clnt = NFS_CLIENT(inode);
928 struct nfs_inode *nfsi = NFS_I(inode);
412d582e 929 sigset_t oldmask;
1da177e4 930 int error;
412d582e 931
412d582e
CL		 932 rpc_clnt_sigmask(clnt, &oldmask);
933 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
934 nfs_wait_schedule, TASK_INTERRUPTIBLE);
935 rpc_clnt_sigunmask(clnt, &oldmask);
412d582e 936
1da177e4
LT		 937 return error;
938}
939
412d582e
CL		 940static void nfs_wake_up_inode(struct inode *inode)
941{
942 struct nfs_inode *nfsi = NFS_I(inode);
943
944 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
945 smp_mb__after_clear_bit();
946 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
947}
948
1da177e4
LT		 949int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
950{
951 struct inode *inode = dentry->d_inode;
55296809 952 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
1da177e4
LT		 953 int err;
954
70b9ecbd
TM		 955 /* Flush out writes to the server in order to update c/mtime */
956 nfs_sync_inode(inode, 0, 0, FLUSH_WAIT|FLUSH_NOCOMMIT);
1da177e4
LT		 957 if (__IS_FLG(inode, MS_NOATIME))
958 need_atime = 0;
959 else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
960 need_atime = 0;
961 /* We may force a getattr if the user cares about atime */
962 if (need_atime)
963 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
964 else
965 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
966 if (!err)
967 generic_fillattr(inode, stat);
968 return err;
969}
970
971struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
972{
973 struct nfs_open_context *ctx;
974
975 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
976 if (ctx != NULL) {
977 atomic_set(&ctx->count, 1);
978 ctx->dentry = dget(dentry);
979 ctx->cred = get_rpccred(cred);
980 ctx->state = NULL;
981 ctx->lockowner = current->files;
982 ctx->error = 0;
00a92642 983 ctx->dir_cookie = 0;
1da177e4
LT		 984 }
985 return ctx;
986}
987
988struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
989{
990 if (ctx != NULL)
991 atomic_inc(&ctx->count);
992 return ctx;
993}
994
995void put_nfs_open_context(struct nfs_open_context *ctx)
996{
997 if (atomic_dec_and_test(&ctx->count)) {
998 if (!list_empty(&ctx->list)) {
999 struct inode *inode = ctx->dentry->d_inode;
1000 spin_lock(&inode->i_lock);
1001 list_del(&ctx->list);
1002 spin_unlock(&inode->i_lock);
1003 }
1004 if (ctx->state != NULL)
1005 nfs4_close_state(ctx->state, ctx->mode);
1006 if (ctx->cred != NULL)
1007 put_rpccred(ctx->cred);
1008 dput(ctx->dentry);
1009 kfree(ctx);
1010 }
1011}
1012
1013/*
1014 * Ensure that mmap has a recent RPC credential for use when writing out
1015 * shared pages
1016 */
1017void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1018{
1019 struct inode *inode = filp->f_dentry->d_inode;
1020 struct nfs_inode *nfsi = NFS_I(inode);
1021
1022 filp->private_data = get_nfs_open_context(ctx);
1023 spin_lock(&inode->i_lock);
1024 list_add(&ctx->list, &nfsi->open_files);
1025 spin_unlock(&inode->i_lock);
1026}
1027
d530838b
TM		 1028/*
1029 * Given an inode, search for an open context with the desired characteristics
1030 */
1031struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
1da177e4
LT		 1032{
1033 struct nfs_inode *nfsi = NFS_I(inode);
1034 struct nfs_open_context *pos, *ctx = NULL;
1035
1036 spin_lock(&inode->i_lock);
1037 list_for_each_entry(pos, &nfsi->open_files, list) {
d530838b
TM		 1038 if (cred != NULL && pos->cred != cred)
1039 continue;
1da177e4
LT		 1040 if ((pos->mode & mode) == mode) {
1041 ctx = get_nfs_open_context(pos);
1042 break;
1043 }
1044 }
1045 spin_unlock(&inode->i_lock);
1046 return ctx;
1047}
1048
1049void nfs_file_clear_open_context(struct file *filp)
1050{
1051 struct inode *inode = filp->f_dentry->d_inode;
1052 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
1053
1054 if (ctx) {
1055 filp->private_data = NULL;
1056 spin_lock(&inode->i_lock);
1057 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1058 spin_unlock(&inode->i_lock);
1059 put_nfs_open_context(ctx);
1060 }
1061}
1062
1063/*
1064 * These allocate and release file read/write context information.
1065 */
1066int nfs_open(struct inode *inode, struct file *filp)
1067{
1068 struct nfs_open_context *ctx;
1069 struct rpc_cred *cred;
1070
1071 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1072 if (IS_ERR(cred))
1073 return PTR_ERR(cred);
1074 ctx = alloc_nfs_open_context(filp->f_dentry, cred);
1075 put_rpccred(cred);
1076 if (ctx == NULL)
1077 return -ENOMEM;
1078 ctx->mode = filp->f_mode;
1079 nfs_file_set_open_context(filp, ctx);
1080 put_nfs_open_context(ctx);
1da177e4
LT		 1081 return 0;
1082}
1083
1084int nfs_release(struct inode *inode, struct file *filp)
1085{
1da177e4
LT		 1086 nfs_file_clear_open_context(filp);
1087 return 0;
1088}
1089
1090/*
1091 * This function is called whenever some part of NFS notices that
1092 * the cached attributes have to be refreshed.
1093 */
1094int
1095__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1096{
1097 int status = -ESTALE;
1098 struct nfs_fattr fattr;
1099 struct nfs_inode *nfsi = NFS_I(inode);
1da177e4
LT		 1100
1101 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1102 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1103
1104 lock_kernel();
1105 if (!inode || is_bad_inode(inode))
1106 goto out_nowait;
1107 if (NFS_STALE(inode))
1108 goto out_nowait;
1109
412d582e
CL		 1110 status = nfs_wait_on_inode(inode);
1111 if (status < 0)
1112 goto out;
1113 if (NFS_STALE(inode)) {
1114 status = -ESTALE;
1115 /* Do we trust the cached ESTALE? */
1116 if (NFS_ATTRTIMEO(inode) != 0) {
1117 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1118 /* no */
1119 } else
1120 goto out;
1121 }
1da177e4 1122 }
1da177e4 1123
1da177e4
LT		 1124 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1125 if (status != 0) {
1126 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1127 inode->i_sb->s_id,
1128 (long long)NFS_FILEID(inode), status);
1129 if (status == -ESTALE) {
1130 nfs_zap_caches(inode);
1131 if (!S_ISDIR(inode->i_mode))
412d582e 1132 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1da177e4
LT		 1133 }
1134 goto out;
1135 }
1136
33801147 1137 spin_lock(&inode->i_lock);
24aa1fe6 1138 status = nfs_update_inode(inode, &fattr);
1da177e4 1139 if (status) {
33801147 1140 spin_unlock(&inode->i_lock);
1da177e4
LT		 1141 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1142 inode->i_sb->s_id,
1143 (long long)NFS_FILEID(inode), status);
1144 goto out;
1145 }
dc59250c 1146 spin_unlock(&inode->i_lock);
55296809 1147
7d52e862 1148 nfs_revalidate_mapping(inode, inode->i_mapping);
55296809 1149
24aa1fe6 1150 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
ada70d94 1151 nfs_zap_acl_cache(inode);
55296809 1152
1da177e4
LT		 1153 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1154 inode->i_sb->s_id,
1155 (long long)NFS_FILEID(inode));
1156
412d582e
CL		 1157 out:
1158 nfs_wake_up_inode(inode);
1159
1da177e4
LT		 1160 out_nowait:
1161 unlock_kernel();
1162 return status;
1163}
1164
1165int nfs_attribute_timeout(struct inode *inode)
1166{
1167 struct nfs_inode *nfsi = NFS_I(inode);
1168
1169 if (nfs_have_delegation(inode, FMODE_READ))
1170 return 0;
1171 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1172}
1173
1174/**
1175 * nfs_revalidate_inode - Revalidate the inode attributes
1176 * @server - pointer to nfs_server struct
1177 * @inode - pointer to inode struct
1178 *
1179 * Updates inode attribute information by retrieving the data from the server.
1180 */
1181int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1182{
55296809 1183 if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1da177e4
LT		 1184 && !nfs_attribute_timeout(inode))
1185 return NFS_STALE(inode) ? -ESTALE : 0;
1186 return __nfs_revalidate_inode(server, inode);
1187}
1188
7d52e862
TM		 1189/**
1190 * nfs_revalidate_mapping - Revalidate the pagecache
1191 * @inode - pointer to host inode
1192 * @mapping - pointer to mapping
1193 */
1194void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1195{
1196 struct nfs_inode *nfsi = NFS_I(inode);
1197
55296809 1198 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
29884df0
TM		 1199 if (S_ISREG(inode->i_mode))
1200 nfs_sync_mapping(mapping);
7d52e862 1201 invalidate_inode_pages2(mapping);
dc59250c
CL		 1202
1203 spin_lock(&inode->i_lock);
55296809 1204 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
7d52e862
TM		 1205 if (S_ISDIR(inode->i_mode)) {
1206 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1207 /* This ensures we revalidate child dentries */
913a70fc 1208 nfsi->cache_change_attribute = jiffies;
7d52e862 1209 }
dc59250c
CL		 1210 spin_unlock(&inode->i_lock);
1211
7d52e862
TM		 1212 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1213 inode->i_sb->s_id,
1214 (long long)NFS_FILEID(inode));
1215 }
1216}
1217
1da177e4
LT		 1218/**
1219 * nfs_begin_data_update
1220 * @inode - pointer to inode
1221 * Declare that a set of operations will update file data on the server
1222 */
1223void nfs_begin_data_update(struct inode *inode)
1224{
1225 atomic_inc(&NFS_I(inode)->data_updates);
1226}
1227
1228/**
1229 * nfs_end_data_update
1230 * @inode - pointer to inode
1231 * Declare end of the operations that will update file data
1232 * This will mark the inode as immediately needing revalidation
1233 * of its attribute cache.
1234 */
1235void nfs_end_data_update(struct inode *inode)
1236{
1237 struct nfs_inode *nfsi = NFS_I(inode);
1238
1239 if (!nfs_have_delegation(inode, FMODE_READ)) {
decf491f
TM		 1240 /* Directories and symlinks: invalidate page cache */
1241 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
1242 spin_lock(&inode->i_lock);
55296809 1243 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
decf491f
TM		 1244 spin_unlock(&inode->i_lock);
1245 }
1da177e4 1246 }
913a70fc 1247 nfsi->cache_change_attribute = jiffies;
1da177e4
LT		 1248 atomic_dec(&nfsi->data_updates);
1249}
1250
a895b4a1
TM		 1251static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1252{
1253 struct nfs_inode *nfsi = NFS_I(inode);
1254
1255 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1256 && nfsi->change_attr == fattr->pre_change_attr) {
1257 nfsi->change_attr = fattr->change_attr;
1258 nfsi->cache_change_attribute = jiffies;
1259 }
1260
1261 /* If we have atomic WCC data, we may update some attributes */
1262 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1263 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1264 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1265 nfsi->cache_change_attribute = jiffies;
1266 }
1267 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1268 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1269 nfsi->cache_change_attribute = jiffies;
1270 }
1271 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
1272 inode->i_size = fattr->size;
1273 nfsi->cache_change_attribute = jiffies;
1274 }
1275 }
1276}
1277
1da177e4 1278/**
33801147 1279 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1da177e4
LT		 1280 * @inode - pointer to inode
1281 * @fattr - updated attributes
1282 *
1283 * Verifies the attribute cache. If we have just changed the attributes,
1284 * so that fattr carries weak cache consistency data, then it may
1285 * also update the ctime/mtime/change_attribute.
1286 */
33801147 1287static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1da177e4
LT		 1288{
1289 struct nfs_inode *nfsi = NFS_I(inode);
1290 loff_t cur_size, new_isize;
1291 int data_unstable;
1292
dc59250c 1293
a895b4a1
TM		 1294 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1295 return 0;
1296
1da177e4
LT		 1297 /* Are we in the process of updating data on the server? */
1298 data_unstable = nfs_caches_unstable(inode);
1299
a895b4a1
TM		 1300 /* Do atomic weak cache consistency updates */
1301 nfs_wcc_update_inode(inode, fattr);
1da177e4 1302
a895b4a1
TM		 1303 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1304 nfsi->change_attr != fattr->change_attr) {
1305 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1306 if (!data_unstable)
1307 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
dc59250c 1308 }
1da177e4
LT		 1309
1310 /* Has the inode gone and changed behind our back? */
1311 if (nfsi->fileid != fattr->fileid
dc59250c 1312 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1da177e4 1313 return -EIO;
dc59250c 1314 }
1da177e4
LT		 1315
1316 cur_size = i_size_read(inode);
1317 new_isize = nfs_size_to_loff_t(fattr->size);
1318
1da177e4 1319 /* Verify a few of the more important attributes */
fe51beec 1320 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
55296809 1321 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
fe51beec 1322 if (!data_unstable)
55296809 1323 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
fe51beec
TM		 1324 }
1325 if (cur_size != new_isize) {
55296809 1326 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
fe51beec 1327 if (nfsi->npages == 0)
55296809 1328 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
fe51beec 1329 }
1da177e4
LT		 1330
1331 /* Have any file permissions changed? */
1332 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1333 || inode->i_uid != fattr->uid
1334 || inode->i_gid != fattr->gid)
55296809 1335 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1da177e4
LT		 1336
1337 /* Has the link count changed? */
1338 if (inode->i_nlink != fattr->nlink)
55296809 1339 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1da177e4
LT		 1340
1341 if (!timespec_equal(&inode->i_atime, &fattr->atime))
55296809 1342 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1da177e4 1343
33801147 1344 nfsi->read_cache_jiffies = fattr->time_start;
1da177e4
LT		 1345 return 0;
1346}
1347
33801147
TM		 1348/**
1349 * nfs_refresh_inode - try to update the inode attribute cache
1350 * @inode - pointer to inode
1351 * @fattr - updated attributes
1352 *
1353 * Check that an RPC call that returned attributes has not overlapped with
1354 * other recent updates of the inode metadata, then decide whether it is
1355 * safe to do a full update of the inode attributes, or whether just to
1356 * call nfs_check_inode_attributes.
1357 */
1358int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1359{
1360 struct nfs_inode *nfsi = NFS_I(inode);
1361 int status;
1362
1363 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1364 return 0;
1365 spin_lock(&inode->i_lock);
1366 nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
33801147 1367 if (time_after(fattr->time_start, nfsi->last_updated))
24aa1fe6 1368 status = nfs_update_inode(inode, fattr);
33801147
TM		 1369 else
1370 status = nfs_check_inode_attributes(inode, fattr);
1371
1372 spin_unlock(&inode->i_lock);
1373 return status;
1374}
1375
decf491f
TM		 1376/**
1377 * nfs_post_op_update_inode - try to update the inode attribute cache
1378 * @inode - pointer to inode
1379 * @fattr - updated attributes
1380 *
1381 * After an operation that has changed the inode metadata, mark the
1382 * attribute cache as being invalid, then try to update it.
1383 */
1384int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1385{
1386 struct nfs_inode *nfsi = NFS_I(inode);
1387 int status = 0;
1388
1389 spin_lock(&inode->i_lock);
1390 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
1391 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1392 goto out;
1393 }
24aa1fe6 1394 status = nfs_update_inode(inode, fattr);
decf491f
TM		 1395out:
1396 spin_unlock(&inode->i_lock);
1397 return status;
1398}
1399
1da177e4
LT		 1400/*
1401 * Many nfs protocol calls return the new file attributes after
1402 * an operation. Here we update the inode to reflect the state
1403 * of the server's inode.
1404 *
1405 * This is a bit tricky because we have to make sure all dirty pages
1406 * have been sent off to the server before calling invalidate_inode_pages.
1407 * To make sure no other process adds more write requests while we try
1408 * our best to flush them, we make them sleep during the attribute refresh.
1409 *
1410 * A very similar scenario holds for the dir cache.
1411 */
24aa1fe6 1412static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1da177e4
LT		 1413{
1414 struct nfs_inode *nfsi = NFS_I(inode);
951a143b 1415 loff_t cur_isize, new_isize;
1da177e4 1416 unsigned int invalid = 0;
24aa1fe6 1417 int data_stable;
1da177e4
LT		 1418
1419 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1420 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1421 atomic_read(&inode->i_count), fattr->valid);
1422
1423 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1424 return 0;
1425
325cfed9
CL		 1426 if (nfsi->fileid != fattr->fileid)
1427 goto out_fileid;
1da177e4
LT		 1428
1429 /*
1430 * Make sure the inode's type hasn't changed.
1431 */
33801147 1432 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1da177e4
LT		 1433 goto out_changed;
1434
1435 /*
1436 * Update the read time so we don't revalidate too often.
1437 */
33801147
TM		 1438 nfsi->read_cache_jiffies = fattr->time_start;
1439 nfsi->last_updated = jiffies;
1da177e4
LT		 1440
1441 /* Are we racing with known updates of the metadata on the server? */
24aa1fe6
TM		 1442 data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
1443 if (data_stable)
1444 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1da177e4 1445
a895b4a1
TM		 1446 /* Do atomic weak cache consistency updates */
1447 nfs_wcc_update_inode(inode, fattr);
1448
951a143b 1449 /* Check if our cached file size is stale */
1da177e4
LT		 1450 new_isize = nfs_size_to_loff_t(fattr->size);
1451 cur_isize = i_size_read(inode);
951a143b
TM		 1452 if (new_isize != cur_isize) {
1453 /* Do we perhaps have any outstanding writes? */
1454 if (nfsi->npages == 0) {
1455 /* No, but did we race with nfs_end_data_update()? */
24aa1fe6 1456 if (data_stable) {
1da177e4 1457 inode->i_size = new_isize;
951a143b 1458 invalid |= NFS_INO_INVALID_DATA;
1da177e4 1459 }
951a143b
TM		 1460 invalid |= NFS_INO_INVALID_ATTR;
1461 } else if (new_isize > cur_isize) {
1da177e4
LT		 1462 inode->i_size = new_isize;
1463 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1464 }
24aa1fe6 1465 nfsi->cache_change_attribute = jiffies;
951a143b
TM		 1466 dprintk("NFS: isize change on server for file %s/%ld\n",
1467 inode->i_sb->s_id, inode->i_ino);
1da177e4
LT		 1468 }
1469
951a143b 1470 /* Check if the mtime agrees */
1da177e4
LT		 1471 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1472 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
951a143b
TM		 1473 dprintk("NFS: mtime change on server for file %s/%ld\n",
1474 inode->i_sb->s_id, inode->i_ino);
24aa1fe6
TM		 1475 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1476 nfsi->cache_change_attribute = jiffies;
1da177e4
LT		 1477 }
1478
1479 if ((fattr->valid & NFS_ATTR_FATTR_V4)
1480 && nfsi->change_attr != fattr->change_attr) {
951a143b 1481 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1da177e4 1482 inode->i_sb->s_id, inode->i_ino);
1da177e4 1483 nfsi->change_attr = fattr->change_attr;
24aa1fe6
TM		 1484 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1485 nfsi->cache_change_attribute = jiffies;
1da177e4
LT		 1486 }
1487
ada70d94
TM		 1488 /* If ctime has changed we should definitely clear access+acl caches */
1489 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
24aa1fe6 1490 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
ada70d94 1491 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
24aa1fe6 1492 nfsi->cache_change_attribute = jiffies;
ada70d94 1493 }
1da177e4
LT		 1494 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1495
1496 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1497 inode->i_uid != fattr->uid ||
1498 inode->i_gid != fattr->gid)
ada70d94 1499 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1da177e4
LT		 1500
1501 inode->i_mode = fattr->mode;
1502 inode->i_nlink = fattr->nlink;
1503 inode->i_uid = fattr->uid;
1504 inode->i_gid = fattr->gid;
1505
1506 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1507 /*
1508 * report the blocks in 512byte units
1509 */
1510 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1511 inode->i_blksize = inode->i_sb->s_blocksize;
1512 } else {
1513 inode->i_blocks = fattr->du.nfs2.blocks;
1514 inode->i_blksize = fattr->du.nfs2.blocksize;
1515 }
1516
1517 /* Update attrtimeo value if we're out of the unstable period */
1518 if (invalid & NFS_INO_INVALID_ATTR) {
1519 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1520 nfsi->attrtimeo_timestamp = jiffies;
1521 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1522 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1523 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1524 nfsi->attrtimeo_timestamp = jiffies;
1525 }
1526 /* Don't invalidate the data if we were to blame */
1527 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1528 || S_ISLNK(inode->i_mode)))
1529 invalid &= ~NFS_INO_INVALID_DATA;
24aa1fe6
TM		 1530 if (data_stable)
1531 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1da177e4 1532 if (!nfs_have_delegation(inode, FMODE_READ))
55296809 1533 nfsi->cache_validity |= invalid;
1da177e4
LT		 1534
1535 return 0;
1536 out_changed:
1537 /*
1538 * Big trouble! The inode has become a different object.
1539 */
1540#ifdef NFS_PARANOIA
1541 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1542 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1543#endif
b37b03b7 1544 out_err:
1da177e4
LT		 1545 /*
1546 * No need to worry about unhashing the dentry, as the
1547 * lookup validation will know that the inode is bad.
1548 * (But we fall through to invalidate the caches.)
1549 */
1550 nfs_invalidate_inode(inode);
1da177e4 1551 return -ESTALE;
325cfed9
CL		 1552
1553 out_fileid:
1554 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1555 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1556 NFS_SERVER(inode)->hostname, inode->i_sb->s_id,
1557 (long long)nfsi->fileid, (long long)fattr->fileid);
1558 goto out_err;
1da177e4
LT		 1559}
1560
1561/*
1562 * File system information
1563 */
1564
1565static int nfs_set_super(struct super_block *s, void *data)
1566{
1567 s->s_fs_info = data;
1568 return set_anon_super(s, data);
1569}
1570
1571static int nfs_compare_super(struct super_block *sb, void *data)
1572{
1573 struct nfs_server *server = data;
1574 struct nfs_server *old = NFS_SB(sb);
1575
1576 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1577 return 0;
1578 if (old->addr.sin_port != server->addr.sin_port)
1579 return 0;
1580 return !nfs_compare_fh(&old->fh, &server->fh);
1581}
1582
1583static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1584 int flags, const char *dev_name, void *raw_data)
1585{
1586 int error;
9085bbcb 1587 struct nfs_server *server = NULL;
1da177e4
LT		 1588 struct super_block *s;
1589 struct nfs_fh *root;
1590 struct nfs_mount_data *data = raw_data;
1591
9085bbcb
TM		 1592 s = ERR_PTR(-EINVAL);
1593 if (data == NULL) {
1594 dprintk("%s: missing data argument\n", __FUNCTION__);
1595 goto out_err;
1596 }
1597 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1598 dprintk("%s: bad mount version\n", __FUNCTION__);
1599 goto out_err;
1da177e4 1600 }
9085bbcb
TM		 1601 switch (data->version) {
1602 case 1:
1603 data->namlen = 0;
1604 case 2:
1605 data->bsize = 0;
1606 case 3:
1607 if (data->flags & NFS_MOUNT_VER3) {
1608 dprintk("%s: mount structure version %d does not support NFSv3\n",
1609 __FUNCTION__,
1610 data->version);
1611 goto out_err;
1612 }
1613 data->root.size = NFS2_FHSIZE;
1614 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1615 case 4:
1616 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1617 dprintk("%s: mount structure version %d does not support strong security\n",
1618 __FUNCTION__,
1619 data->version);
1620 goto out_err;
1621 }
1622 case 5:
1623 memset(data->context, 0, sizeof(data->context));
1624 }
1625#ifndef CONFIG_NFS_V3
1626 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1627 s = ERR_PTR(-EPROTONOSUPPORT);
1628 if (data->flags & NFS_MOUNT_VER3) {
1629 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1630 goto out_err;
1631 }
1632#endif /* CONFIG_NFS_V3 */
1da177e4 1633
9085bbcb 1634 s = ERR_PTR(-ENOMEM);
1da177e4
LT		 1635 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1636 if (!server)
9085bbcb 1637 goto out_err;
1da177e4
LT		 1638 memset(server, 0, sizeof(struct nfs_server));
1639 /* Zero out the NFS state stuff */
1640 init_nfsv4_state(server);
b7fa0554 1641 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1da177e4 1642
1da177e4
LT		 1643 root = &server->fh;
1644 if (data->flags & NFS_MOUNT_VER3)
1645 root->size = data->root.size;
1646 else
1647 root->size = NFS2_FHSIZE;
9085bbcb 1648 s = ERR_PTR(-EINVAL);
1da177e4 1649 if (root->size > sizeof(root->data)) {
9085bbcb
TM		 1650 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1651 goto out_err;
1da177e4
LT		 1652 }
1653 memcpy(root->data, data->root.data, root->size);
1654
1655 /* We now require that the mount process passes the remote address */
1656 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1657 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
9085bbcb
TM		 1658 dprintk("%s: mount program didn't pass remote address!\n",
1659 __FUNCTION__);
1660 goto out_err;
1da177e4
LT		 1661 }
1662
9085bbcb
TM		 1663 /* Fire up rpciod if not yet running */
1664 s = ERR_PTR(rpciod_up());
1665 if (IS_ERR(s)) {
1666 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1667 __FUNCTION__, PTR_ERR(s));
1668 goto out_err;
1da177e4
LT		 1669 }
1670
9085bbcb
TM		 1671 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1672 if (IS_ERR(s) || s->s_root)
1673 goto out_rpciod_down;
1da177e4 1674
9085bbcb 1675 s->s_flags = flags;
1da177e4
LT		 1676
1677 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1678 if (error) {
1679 up_write(&s->s_umount);
1680 deactivate_super(s);
1681 return ERR_PTR(error);
1682 }
1683 s->s_flags |= MS_ACTIVE;
1684 return s;
9085bbcb
TM		 1685out_rpciod_down:
1686 rpciod_down();
1687out_err:
1688 kfree(server);
1689 return s;
1da177e4
LT		 1690}
1691
1692static void nfs_kill_super(struct super_block *s)
1693{
1694 struct nfs_server *server = NFS_SB(s);
1695
1696 kill_anon_super(s);
1697
6a19275a 1698 if (!IS_ERR(server->client))
1da177e4 1699 rpc_shutdown_client(server->client);
6a19275a 1700 if (!IS_ERR(server->client_sys))
1da177e4 1701 rpc_shutdown_client(server->client_sys);
b7fa0554
AG		 1702 if (!IS_ERR(server->client_acl))
1703 rpc_shutdown_client(server->client_acl);
1da177e4
LT		 1704
1705 if (!(server->flags & NFS_MOUNT_NONLM))
1706 lockd_down(); /* release rpc.lockd */
1707
1708 rpciod_down(); /* release rpciod */
1709
f99d49ad 1710 kfree(server->hostname);
1da177e4
LT		 1711 kfree(server);
1712}
1713
1714static struct file_system_type nfs_fs_type = {
1715 .owner = THIS_MODULE,
1716 .name = "nfs",
1717 .get_sb = nfs_get_sb,
1718 .kill_sb = nfs_kill_super,
1719 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1720};
1721
1722#ifdef CONFIG_NFS_V4
1723
1724static void nfs4_clear_inode(struct inode *);
1725
1726
1727static struct super_operations nfs4_sops = {
1728 .alloc_inode = nfs_alloc_inode,
1729 .destroy_inode = nfs_destroy_inode,
1730 .write_inode = nfs_write_inode,
1731 .delete_inode = nfs_delete_inode,
1732 .statfs = nfs_statfs,
1733 .clear_inode = nfs4_clear_inode,
1734 .umount_begin = nfs_umount_begin,
1735 .show_options = nfs_show_options,
1736};
1737
1738/*
1739 * Clean out any remaining NFSv4 state that might be left over due
1740 * to open() calls that passed nfs_atomic_lookup, but failed to call
1741 * nfs_open().
1742 */
1743static void nfs4_clear_inode(struct inode *inode)
1744{
1745 struct nfs_inode *nfsi = NFS_I(inode);
1746
1747 /* If we are holding a delegation, return it! */
cae7a073 1748 nfs_inode_return_delegation(inode);
1da177e4
LT		 1749 /* First call standard NFS clear_inode() code */
1750 nfs_clear_inode(inode);
1751 /* Now clear out any remaining state */
1752 while (!list_empty(&nfsi->open_states)) {
1753 struct nfs4_state *state;
1754
1755 state = list_entry(nfsi->open_states.next,
1756 struct nfs4_state,
1757 inode_states);
1758 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1759 __FUNCTION__,
1760 inode->i_sb->s_id,
1761 (long long)NFS_FILEID(inode),
1762 state);
1763 BUG_ON(atomic_read(&state->count) != 1);
1764 nfs4_close_state(state, state->state);
1765 }
1766}
1767
1768
1769static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1770{
1771 struct nfs_server *server;
1772 struct nfs4_client *clp = NULL;
1773 struct rpc_xprt *xprt = NULL;
1774 struct rpc_clnt *clnt = NULL;
1775 struct rpc_timeout timeparms;
1776 rpc_authflavor_t authflavour;
eab5c084 1777 int err = -EIO;
1da177e4
LT		 1778
1779 sb->s_blocksize_bits = 0;
1780 sb->s_blocksize = 0;
1781 server = NFS_SB(sb);
1782 if (data->rsize != 0)
1783 server->rsize = nfs_block_size(data->rsize, NULL);
1784 if (data->wsize != 0)
1785 server->wsize = nfs_block_size(data->wsize, NULL);
1786 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1787 server->caps = NFS_CAP_ATOMIC_OPEN;
1788
1789 server->acregmin = data->acregmin*HZ;
1790 server->acregmax = data->acregmax*HZ;
1791 server->acdirmin = data->acdirmin*HZ;
1792 server->acdirmax = data->acdirmax*HZ;
1793
1794 server->rpc_ops = &nfs_v4_clientops;
1da177e4 1795
eab5c084 1796 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
1da177e4
LT		 1797
1798 clp = nfs4_get_client(&server->addr.sin_addr);
1799 if (!clp) {
9085bbcb 1800 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1da177e4
LT		 1801 return -EIO;
1802 }
1803
1804 /* Now create transport and client */
1805 authflavour = RPC_AUTH_UNIX;
1806 if (data->auth_flavourlen != 0) {
9085bbcb
TM		 1807 if (data->auth_flavourlen != 1) {
1808 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1809 __FUNCTION__, data->auth_flavourlen);
1810 err = -EINVAL;
1811 goto out_fail;
1812 }
1da177e4
LT		 1813 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1814 err = -EFAULT;
1815 goto out_fail;
1816 }
1817 }
1818
1819 down_write(&clp->cl_sem);
6a19275a 1820 if (IS_ERR(clp->cl_rpcclient)) {
eab5c084 1821 xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
1da177e4
LT		 1822 if (IS_ERR(xprt)) {
1823 up_write(&clp->cl_sem);
1da177e4 1824 err = PTR_ERR(xprt);
9085bbcb
TM		 1825 dprintk("%s: cannot create RPC transport. Error = %d\n",
1826 __FUNCTION__, err);
1da177e4
LT		 1827 goto out_fail;
1828 }
1829 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1830 server->rpc_ops->version, authflavour);
1831 if (IS_ERR(clnt)) {
1832 up_write(&clp->cl_sem);
1da177e4 1833 err = PTR_ERR(clnt);
9085bbcb
TM		 1834 dprintk("%s: cannot create RPC client. Error = %d\n",
1835 __FUNCTION__, err);
1da177e4
LT		 1836 goto out_fail;
1837 }
1838 clnt->cl_intr = 1;
1839 clnt->cl_softrtry = 1;
1840 clnt->cl_chatty = 1;
1841 clp->cl_rpcclient = clnt;
1da177e4
LT		 1842 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1843 nfs_idmap_new(clp);
1844 }
1da177e4
LT		 1845 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1846 clnt = rpc_clone_client(clp->cl_rpcclient);
1847 if (!IS_ERR(clnt))
1848 server->nfs4_state = clp;
1849 up_write(&clp->cl_sem);
1850 clp = NULL;
1851
1852 if (IS_ERR(clnt)) {
9085bbcb
TM		 1853 err = PTR_ERR(clnt);
1854 dprintk("%s: cannot create RPC client. Error = %d\n",
1855 __FUNCTION__, err);
1856 return err;
1da177e4
LT		 1857 }
1858
1859 server->client = clnt;
1860
1861 if (server->nfs4_state->cl_idmap == NULL) {
9085bbcb 1862 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1da177e4
LT		 1863 return -ENOMEM;
1864 }
1865
1866 if (clnt->cl_auth->au_flavor != authflavour) {
6a19275a
BF		 1867 struct rpc_auth *auth;
1868
1869 auth = rpcauth_create(authflavour, clnt);
1870 if (IS_ERR(auth)) {
9085bbcb 1871 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
6a19275a 1872 return PTR_ERR(auth);
1da177e4
LT		 1873 }
1874 }
1875
1876 sb->s_time_gran = 1;
1877
1878 sb->s_op = &nfs4_sops;
1879 err = nfs_sb_init(sb, authflavour);
1880 if (err == 0)
1881 return 0;
1882out_fail:
1883 if (clp)
1884 nfs4_put_client(clp);
1885 return err;
1886}
1887
1888static int nfs4_compare_super(struct super_block *sb, void *data)
1889{
1890 struct nfs_server *server = data;
1891 struct nfs_server *old = NFS_SB(sb);
1892
1893 if (strcmp(server->hostname, old->hostname) != 0)
1894 return 0;
1895 if (strcmp(server->mnt_path, old->mnt_path) != 0)
1896 return 0;
1897 return 1;
1898}
1899
1900static void *
1901nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1902{
1903 void *p = NULL;
1904
1905 if (!src->len)
1906 return ERR_PTR(-EINVAL);
1907 if (src->len < maxlen)
1908 maxlen = src->len;
1909 if (dst == NULL) {
1910 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1911 if (p == NULL)
1912 return ERR_PTR(-ENOMEM);
1913 }
1914 if (copy_from_user(dst, src->data, maxlen)) {
f99d49ad 1915 kfree(p);
1da177e4
LT		 1916 return ERR_PTR(-EFAULT);
1917 }
1918 dst[maxlen] = '\0';
1919 return dst;
1920}
1921
1922static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1923 int flags, const char *dev_name, void *raw_data)
1924{
1925 int error;
1926 struct nfs_server *server;
1927 struct super_block *s;
1928 struct nfs4_mount_data *data = raw_data;
1929 void *p;
1930
9085bbcb
TM		 1931 if (data == NULL) {
1932 dprintk("%s: missing data argument\n", __FUNCTION__);
1933 return ERR_PTR(-EINVAL);
1934 }
1935 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1936 dprintk("%s: bad mount version\n", __FUNCTION__);
1da177e4
LT		 1937 return ERR_PTR(-EINVAL);
1938 }
1939
1940 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1941 if (!server)
1942 return ERR_PTR(-ENOMEM);
1943 memset(server, 0, sizeof(struct nfs_server));
1944 /* Zero out the NFS state stuff */
1945 init_nfsv4_state(server);
b7fa0554 1946 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1da177e4 1947
1da177e4
LT		 1948 p = nfs_copy_user_string(NULL, &data->hostname, 256);
1949 if (IS_ERR(p))
1950 goto out_err;
1951 server->hostname = p;
1952
1953 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1954 if (IS_ERR(p))
1955 goto out_err;
1956 server->mnt_path = p;
1957
1958 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1959 sizeof(server->ip_addr) - 1);
1960 if (IS_ERR(p))
1961 goto out_err;
1962
1963 /* We now require that the mount process passes the remote address */
1964 if (data->host_addrlen != sizeof(server->addr)) {
1965 s = ERR_PTR(-EINVAL);
1966 goto out_free;
1967 }
1968 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1969 s = ERR_PTR(-EFAULT);
1970 goto out_free;
1971 }
1972 if (server->addr.sin_family != AF_INET ||
1973 server->addr.sin_addr.s_addr == INADDR_ANY) {
9085bbcb
TM		 1974 dprintk("%s: mount program didn't pass remote IP address!\n",
1975 __FUNCTION__);
1da177e4
LT		 1976 s = ERR_PTR(-EINVAL);
1977 goto out_free;
1978 }
1979
9085bbcb
TM		 1980 /* Fire up rpciod if not yet running */
1981 s = ERR_PTR(rpciod_up());
1982 if (IS_ERR(s)) {
1983 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1984 __FUNCTION__, PTR_ERR(s));
1985 goto out_free;
1986 }
1987
1da177e4
LT		 1988 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1989
1990 if (IS_ERR(s) || s->s_root)
1991 goto out_free;
1992
1993 s->s_flags = flags;
1994
1da177e4
LT		 1995 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1996 if (error) {
1997 up_write(&s->s_umount);
1998 deactivate_super(s);
1999 return ERR_PTR(error);
2000 }
2001 s->s_flags |= MS_ACTIVE;
2002 return s;
2003out_err:
2004 s = (struct super_block *)p;
2005out_free:
f99d49ad
JJ		 2006 kfree(server->mnt_path);
2007 kfree(server->hostname);
1da177e4
LT		 2008 kfree(server);
2009 return s;
2010}
2011
2012static void nfs4_kill_super(struct super_block *sb)
2013{
2014 struct nfs_server *server = NFS_SB(sb);
2015
2016 nfs_return_all_delegations(sb);
2017 kill_anon_super(sb);
2018
2019 nfs4_renewd_prepare_shutdown(server);
2020
2021 if (server->client != NULL && !IS_ERR(server->client))
2022 rpc_shutdown_client(server->client);
2023 rpciod_down(); /* release rpciod */
2024
2025 destroy_nfsv4_state(server);
2026
f99d49ad 2027 kfree(server->hostname);
1da177e4
LT		 2028 kfree(server);
2029}
2030
2031static struct file_system_type nfs4_fs_type = {
2032 .owner = THIS_MODULE,
2033 .name = "nfs4",
2034 .get_sb = nfs4_get_sb,
2035 .kill_sb = nfs4_kill_super,
2036 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2037};
2038
2039#define nfs4_init_once(nfsi) \
2040 do { \
2041 INIT_LIST_HEAD(&(nfsi)->open_states); \
2042 nfsi->delegation = NULL; \
2043 nfsi->delegation_state = 0; \
2044 init_rwsem(&nfsi->rwsem); \
2045 } while(0)
2046#define register_nfs4fs() register_filesystem(&nfs4_fs_type)
2047#define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
2048#else
2049#define nfs4_init_once(nfsi) \
2050 do { } while (0)
2051#define register_nfs4fs() (0)
2052#define unregister_nfs4fs()
2053#endif
2054
2055extern int nfs_init_nfspagecache(void);
2056extern void nfs_destroy_nfspagecache(void);
2057extern int nfs_init_readpagecache(void);
2058extern void nfs_destroy_readpagecache(void);
2059extern int nfs_init_writepagecache(void);
2060extern void nfs_destroy_writepagecache(void);
2061#ifdef CONFIG_NFS_DIRECTIO
2062extern int nfs_init_directcache(void);
2063extern void nfs_destroy_directcache(void);
2064#endif
2065
2066static kmem_cache_t * nfs_inode_cachep;
2067
2068static struct inode *nfs_alloc_inode(struct super_block *sb)
2069{
2070 struct nfs_inode *nfsi;
2071 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2072 if (!nfsi)
2073 return NULL;
55296809
CL		 2074 nfsi->flags = 0UL;
2075 nfsi->cache_validity = 0UL;
223db122 2076 nfsi->cache_change_attribute = jiffies;
458818ed
TM		 2077#ifdef CONFIG_NFS_V3_ACL
2078 nfsi->acl_access = ERR_PTR(-EAGAIN);
2079 nfsi->acl_default = ERR_PTR(-EAGAIN);
2080#endif
e50a1c2e
BF		 2081#ifdef CONFIG_NFS_V4
2082 nfsi->nfs4_acl = NULL;
2083#endif /* CONFIG_NFS_V4 */
1da177e4
LT		 2084 return &nfsi->vfs_inode;
2085}
2086
2087static void nfs_destroy_inode(struct inode *inode)
2088{
2089 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2090}
2091
2092static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2093{
2094 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2095
2096 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2097 SLAB_CTOR_CONSTRUCTOR) {
2098 inode_init_once(&nfsi->vfs_inode);
2099 spin_lock_init(&nfsi->req_lock);
2100 INIT_LIST_HEAD(&nfsi->dirty);
2101 INIT_LIST_HEAD(&nfsi->commit);
2102 INIT_LIST_HEAD(&nfsi->open_files);
2103 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2104 atomic_set(&nfsi->data_updates, 0);
2105 nfsi->ndirty = 0;
2106 nfsi->ncommit = 0;
2107 nfsi->npages = 0;
1da177e4
LT		 2108 nfs4_init_once(nfsi);
2109 }
2110}
2111
75c96f85 2112static int nfs_init_inodecache(void)
1da177e4
LT		 2113{
2114 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2115 sizeof(struct nfs_inode),
2116 0, SLAB_RECLAIM_ACCOUNT,
2117 init_once, NULL);
2118 if (nfs_inode_cachep == NULL)
2119 return -ENOMEM;
2120
2121 return 0;
2122}
2123
75c96f85 2124static void nfs_destroy_inodecache(void)
1da177e4
LT		 2125{
2126 if (kmem_cache_destroy(nfs_inode_cachep))
2127 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2128}
2129
2130/*
2131 * Initialize NFS
2132 */
2133static int __init init_nfs_fs(void)
2134{
2135 int err;
2136
2137 err = nfs_init_nfspagecache();
2138 if (err)
2139 goto out4;
2140
2141 err = nfs_init_inodecache();
2142 if (err)
2143 goto out3;
2144
2145 err = nfs_init_readpagecache();
2146 if (err)
2147 goto out2;
2148
2149 err = nfs_init_writepagecache();
2150 if (err)
2151 goto out1;
2152
2153#ifdef CONFIG_NFS_DIRECTIO
2154 err = nfs_init_directcache();
2155 if (err)
2156 goto out0;
2157#endif
2158
2159#ifdef CONFIG_PROC_FS
2160 rpc_proc_register(&nfs_rpcstat);
2161#endif
2162 err = register_filesystem(&nfs_fs_type);
2163 if (err)
2164 goto out;
2165 if ((err = register_nfs4fs()) != 0)
2166 goto out;
2167 return 0;
2168out:
2169#ifdef CONFIG_PROC_FS
2170 rpc_proc_unregister("nfs");
2171#endif
1da177e4 2172#ifdef CONFIG_NFS_DIRECTIO
1da177e4 2173 nfs_destroy_directcache();
6b59a754 2174out0:
1da177e4 2175#endif
6b59a754 2176 nfs_destroy_writepagecache();
1da177e4
LT		 2177out1:
2178 nfs_destroy_readpagecache();
2179out2:
2180 nfs_destroy_inodecache();
2181out3:
2182 nfs_destroy_nfspagecache();
2183out4:
2184 return err;
2185}
2186
2187static void __exit exit_nfs_fs(void)
2188{
2189#ifdef CONFIG_NFS_DIRECTIO
2190 nfs_destroy_directcache();
2191#endif
2192 nfs_destroy_writepagecache();
2193 nfs_destroy_readpagecache();
2194 nfs_destroy_inodecache();
2195 nfs_destroy_nfspagecache();
2196#ifdef CONFIG_PROC_FS
2197 rpc_proc_unregister("nfs");
2198#endif
2199 unregister_filesystem(&nfs_fs_type);
2200 unregister_nfs4fs();
2201}
2202
2203/* Not quite true; I just maintain it */
2204MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2205MODULE_LICENSE("GPL");
2206
2207module_init(init_nfs_fs)
2208module_exit(exit_nfs_fs)
Linux 6.x block layer and io_uring tree(s)