4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/config.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
20 #include <linux/time.h>
21 #include <linux/kernel.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>
39 #include <asm/system.h>
40 #include <asm/uaccess.h>
43 #include "delegation.h"
45 #define NFSDBG_FACILITY NFSDBG_VFS
46 #define NFS_PARANOIA 1
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.
54 #define NFS_MAX_READAHEAD (RPC_DEF_SLOT_TABLE - 1)
56 static void nfs_invalidate_inode(struct inode *);
57 static int nfs_update_inode(struct inode *, struct nfs_fattr *, unsigned long);
59 static struct inode *nfs_alloc_inode(struct super_block *sb);
60 static void nfs_destroy_inode(struct inode *);
61 static int nfs_write_inode(struct inode *,int);
62 static void nfs_delete_inode(struct inode *);
63 static void nfs_clear_inode(struct inode *);
64 static void nfs_umount_begin(struct super_block *);
65 static int nfs_statfs(struct super_block *, struct kstatfs *);
66 static int nfs_show_options(struct seq_file *, struct vfsmount *);
67 static void nfs_zap_acl_cache(struct inode *);
69 static struct rpc_program nfs_program;
71 static struct super_operations nfs_sops = {
72 .alloc_inode = nfs_alloc_inode,
73 .destroy_inode = nfs_destroy_inode,
74 .write_inode = nfs_write_inode,
75 .delete_inode = nfs_delete_inode,
77 .clear_inode = nfs_clear_inode,
78 .umount_begin = nfs_umount_begin,
79 .show_options = nfs_show_options,
85 static struct rpc_stat nfs_rpcstat = {
86 .program = &nfs_program
88 static struct rpc_version * nfs_version[] = {
92 #if defined(CONFIG_NFS_V3)
94 #elif defined(CONFIG_NFS_V4)
97 #if defined(CONFIG_NFS_V4)
102 static struct rpc_program nfs_program = {
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",
111 #ifdef CONFIG_NFS_V3_ACL
112 static struct rpc_stat nfsacl_rpcstat = { &nfsacl_program };
113 static struct rpc_version * nfsacl_version[] = {
114 [3] = &nfsacl_version3,
117 struct rpc_program nfsacl_program = {
119 .number = NFS_ACL_PROGRAM,
120 .nrvers = sizeof(nfsacl_version) / sizeof(nfsacl_version[0]),
121 .version = nfsacl_version,
122 .stats = &nfsacl_rpcstat,
124 #endif /* CONFIG_NFS_V3_ACL */
126 static inline unsigned long
127 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
129 return nfs_fileid_to_ino_t(fattr->fileid);
133 nfs_write_inode(struct inode *inode, int sync)
135 int flags = sync ? FLUSH_WAIT : 0;
138 ret = nfs_commit_inode(inode, 0, 0, flags);
145 nfs_delete_inode(struct inode * inode)
147 dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
151 * The following should never happen...
153 if (nfs_have_writebacks(inode)) {
154 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
161 nfs_clear_inode(struct inode *inode)
163 struct nfs_inode *nfsi = NFS_I(inode);
164 struct rpc_cred *cred;
167 BUG_ON (!list_empty(&nfsi->open_files));
168 nfs_zap_acl_cache(inode);
169 cred = nfsi->cache_access.cred;
172 BUG_ON(atomic_read(&nfsi->data_updates) != 0);
176 nfs_umount_begin(struct super_block *sb)
178 struct rpc_clnt *rpc = NFS_SB(sb)->client;
180 /* -EIO all pending I/O */
182 rpc_killall_tasks(rpc);
183 rpc = NFS_SB(sb)->client_acl;
185 rpc_killall_tasks(rpc);
189 static inline unsigned long
190 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
192 /* make sure blocksize is a power of two */
193 if ((bsize & (bsize - 1)) || nrbitsp) {
194 unsigned char nrbits;
196 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
207 * Calculate the number of 512byte blocks used.
209 static inline unsigned long
210 nfs_calc_block_size(u64 tsize)
212 loff_t used = (tsize + 511) >> 9;
213 return (used > ULONG_MAX) ? ULONG_MAX : used;
217 * Compute and set NFS server blocksize
219 static inline unsigned long
220 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
223 bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
224 else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
225 bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
227 return nfs_block_bits(bsize, nrbitsp);
231 * Obtain the root inode of the file system.
233 static struct inode *
234 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
236 struct nfs_server *server = NFS_SB(sb);
240 error = server->rpc_ops->getroot(server, rootfh, fsinfo);
242 dprintk("nfs_get_root: getattr error = %d\n", -error);
243 return ERR_PTR(error);
246 rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
248 return ERR_PTR(-ENOMEM);
253 * Do NFS version-independent mount processing, and sanity checking
256 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
258 struct nfs_server *server;
259 struct inode *root_inode;
260 struct nfs_fattr fattr;
261 struct nfs_fsinfo fsinfo = {
264 struct nfs_pathconf pathinfo = {
267 int no_root_error = 0;
268 unsigned long max_rpc_payload;
270 /* We probably want something more informative here */
271 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
275 sb->s_magic = NFS_SUPER_MAGIC;
277 root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
278 /* Did getting the root inode fail? */
279 if (IS_ERR(root_inode)) {
280 no_root_error = PTR_ERR(root_inode);
283 sb->s_root = d_alloc_root(root_inode);
285 no_root_error = -ENOMEM;
288 sb->s_root->d_op = server->rpc_ops->dentry_ops;
290 /* Get some general file system info */
291 if (server->namelen == 0 &&
292 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
293 server->namelen = pathinfo.max_namelen;
294 /* Work out a lot of parameters */
295 if (server->rsize == 0)
296 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
297 if (server->wsize == 0)
298 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
300 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
301 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
302 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
303 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
305 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
306 if (server->rsize > max_rpc_payload)
307 server->rsize = max_rpc_payload;
308 if (server->wsize > max_rpc_payload)
309 server->wsize = max_rpc_payload;
311 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
312 if (server->rpages > NFS_READ_MAXIOV) {
313 server->rpages = NFS_READ_MAXIOV;
314 server->rsize = server->rpages << PAGE_CACHE_SHIFT;
317 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
318 if (server->wpages > NFS_WRITE_MAXIOV) {
319 server->wpages = NFS_WRITE_MAXIOV;
320 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
323 if (sb->s_blocksize == 0)
324 sb->s_blocksize = nfs_block_bits(server->wsize,
325 &sb->s_blocksize_bits);
326 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
328 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
329 if (server->dtsize > PAGE_CACHE_SIZE)
330 server->dtsize = PAGE_CACHE_SIZE;
331 if (server->dtsize > server->rsize)
332 server->dtsize = server->rsize;
334 if (server->flags & NFS_MOUNT_NOAC) {
335 server->acregmin = server->acregmax = 0;
336 server->acdirmin = server->acdirmax = 0;
337 sb->s_flags |= MS_SYNCHRONOUS;
339 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
341 sb->s_maxbytes = fsinfo.maxfilesize;
342 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
343 sb->s_maxbytes = MAX_LFS_FILESIZE;
345 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
346 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
348 /* We're airborne Set socket buffersize */
349 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
351 /* Yargs. It didn't work out. */
353 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
354 if (!IS_ERR(root_inode))
356 return no_root_error;
360 * Create an RPC client handle.
362 static struct rpc_clnt *
363 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
365 struct rpc_timeout timeparms;
366 struct rpc_xprt *xprt = NULL;
367 struct rpc_clnt *clnt = NULL;
368 int tcp = (data->flags & NFS_MOUNT_TCP);
370 /* Initialize timeout values */
371 timeparms.to_initval = data->timeo * HZ / 10;
372 timeparms.to_retries = data->retrans;
373 timeparms.to_maxval = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
374 timeparms.to_exponential = 1;
376 if (!timeparms.to_initval)
377 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
378 if (!timeparms.to_retries)
379 timeparms.to_retries = 5;
381 /* create transport and client */
382 xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
383 &server->addr, &timeparms);
385 dprintk("%s: cannot create RPC transport. Error = %ld\n",
386 __FUNCTION__, PTR_ERR(xprt));
387 return (struct rpc_clnt *)xprt;
389 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
390 server->rpc_ops->version, data->pseudoflavor);
392 dprintk("%s: cannot create RPC client. Error = %ld\n",
393 __FUNCTION__, PTR_ERR(xprt));
398 clnt->cl_softrtry = 1;
408 * The way this works is that the mount process passes a structure
409 * in the data argument which contains the server's IP address
410 * and the root file handle obtained from the server's mount
411 * daemon. We stash these away in the private superblock fields.
414 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
416 struct nfs_server *server;
417 rpc_authflavor_t authflavor;
420 sb->s_blocksize_bits = 0;
423 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
425 server->rsize = nfs_block_size(data->rsize, NULL);
427 server->wsize = nfs_block_size(data->wsize, NULL);
428 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
430 server->acregmin = data->acregmin*HZ;
431 server->acregmax = data->acregmax*HZ;
432 server->acdirmin = data->acdirmin*HZ;
433 server->acdirmax = data->acdirmax*HZ;
435 /* Start lockd here, before we might error out */
436 if (!(server->flags & NFS_MOUNT_NONLM))
439 server->namelen = data->namlen;
440 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
441 if (!server->hostname)
443 strcpy(server->hostname, data->hostname);
445 /* Check NFS protocol revision and initialize RPC op vector
446 * and file handle pool. */
448 if (server->flags & NFS_MOUNT_VER3) {
449 server->rpc_ops = &nfs_v3_clientops;
450 server->caps |= NFS_CAP_READDIRPLUS;
452 server->rpc_ops = &nfs_v2_clientops;
455 server->rpc_ops = &nfs_v2_clientops;
458 /* Fill in pseudoflavor for mount version < 5 */
459 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
460 data->pseudoflavor = RPC_AUTH_UNIX;
461 authflavor = data->pseudoflavor; /* save for sb_init() */
462 /* XXX maybe we want to add a server->pseudoflavor field */
464 /* Create RPC client handles */
465 server->client = nfs_create_client(server, data);
466 if (IS_ERR(server->client))
467 return PTR_ERR(server->client);
468 /* RFC 2623, sec 2.3.2 */
469 if (authflavor != RPC_AUTH_UNIX) {
470 struct rpc_auth *auth;
472 server->client_sys = rpc_clone_client(server->client);
473 if (IS_ERR(server->client_sys))
474 return PTR_ERR(server->client_sys);
475 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
477 return PTR_ERR(auth);
479 atomic_inc(&server->client->cl_count);
480 server->client_sys = server->client;
482 if (server->flags & NFS_MOUNT_VER3) {
483 #ifdef CONFIG_NFS_V3_ACL
484 if (!(server->flags & NFS_MOUNT_NOACL)) {
485 server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
486 /* No errors! Assume that Sun nfsacls are supported */
487 if (!IS_ERR(server->client_acl))
488 server->caps |= NFS_CAP_ACLS;
491 server->flags &= ~NFS_MOUNT_NOACL;
492 #endif /* CONFIG_NFS_V3_ACL */
494 * The VFS shouldn't apply the umask to mode bits. We will
495 * do so ourselves when necessary.
497 sb->s_flags |= MS_POSIXACL;
498 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
499 server->namelen = NFS3_MAXNAMLEN;
502 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
503 server->namelen = NFS2_MAXNAMLEN;
506 sb->s_op = &nfs_sops;
507 return nfs_sb_init(sb, authflavor);
511 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
513 struct nfs_server *server = NFS_SB(sb);
514 unsigned char blockbits;
515 unsigned long blockres;
516 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
517 struct nfs_fattr fattr;
518 struct nfs_fsstat res = {
525 error = server->rpc_ops->statfs(server, rootfh, &res);
526 buf->f_type = NFS_SUPER_MAGIC;
531 * Current versions of glibc do not correctly handle the
532 * case where f_frsize != f_bsize. Eventually we want to
533 * report the value of wtmult in this field.
535 buf->f_frsize = sb->s_blocksize;
538 * On most *nix systems, f_blocks, f_bfree, and f_bavail
539 * are reported in units of f_frsize. Linux hasn't had
540 * an f_frsize field in its statfs struct until recently,
541 * thus historically Linux's sys_statfs reports these
542 * fields in units of f_bsize.
544 buf->f_bsize = sb->s_blocksize;
545 blockbits = sb->s_blocksize_bits;
546 blockres = (1 << blockbits) - 1;
547 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
548 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
549 buf->f_bavail = (res.abytes + blockres) >> blockbits;
551 buf->f_files = res.tfiles;
552 buf->f_ffree = res.afiles;
554 buf->f_namelen = server->namelen;
561 printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
562 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
567 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
569 static struct proc_nfs_info {
574 { NFS_MOUNT_SOFT, ",soft", ",hard" },
575 { NFS_MOUNT_INTR, ",intr", "" },
576 { NFS_MOUNT_POSIX, ",posix", "" },
577 { NFS_MOUNT_TCP, ",tcp", ",udp" },
578 { NFS_MOUNT_NOCTO, ",nocto", "" },
579 { NFS_MOUNT_NOAC, ",noac", "" },
580 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
581 { NFS_MOUNT_NOACL, ",noacl", "" },
584 struct proc_nfs_info *nfs_infop;
585 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
587 seq_printf(m, ",v%d", nfss->rpc_ops->version);
588 seq_printf(m, ",rsize=%d", nfss->rsize);
589 seq_printf(m, ",wsize=%d", nfss->wsize);
590 if (nfss->acregmin != 3*HZ)
591 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
592 if (nfss->acregmax != 60*HZ)
593 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
594 if (nfss->acdirmin != 30*HZ)
595 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
596 if (nfss->acdirmax != 60*HZ)
597 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
598 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
599 if (nfss->flags & nfs_infop->flag)
600 seq_puts(m, nfs_infop->str);
602 seq_puts(m, nfs_infop->nostr);
604 seq_puts(m, ",addr=");
605 seq_escape(m, nfss->hostname, " \t\n\\");
610 * Invalidate the local caches
613 nfs_zap_caches(struct inode *inode)
615 struct nfs_inode *nfsi = NFS_I(inode);
616 int mode = inode->i_mode;
618 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
619 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
621 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
622 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
623 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
625 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
628 static void nfs_zap_acl_cache(struct inode *inode)
630 void (*clear_acl_cache)(struct inode *);
632 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
633 if (clear_acl_cache != NULL)
634 clear_acl_cache(inode);
635 NFS_I(inode)->flags &= ~NFS_INO_INVALID_ACL;
639 * Invalidate, but do not unhash, the inode
642 nfs_invalidate_inode(struct inode *inode)
644 umode_t save_mode = inode->i_mode;
646 make_bad_inode(inode);
647 inode->i_mode = save_mode;
648 nfs_zap_caches(inode);
651 struct nfs_find_desc {
653 struct nfs_fattr *fattr;
657 * In NFSv3 we can have 64bit inode numbers. In order to support
658 * this, and re-exported directories (also seen in NFSv2)
659 * we are forced to allow 2 different inodes to have the same
663 nfs_find_actor(struct inode *inode, void *opaque)
665 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
666 struct nfs_fh *fh = desc->fh;
667 struct nfs_fattr *fattr = desc->fattr;
669 if (NFS_FILEID(inode) != fattr->fileid)
671 if (nfs_compare_fh(NFS_FH(inode), fh))
673 if (is_bad_inode(inode) || NFS_STALE(inode))
679 nfs_init_locked(struct inode *inode, void *opaque)
681 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
682 struct nfs_fattr *fattr = desc->fattr;
684 NFS_FILEID(inode) = fattr->fileid;
685 nfs_copy_fh(NFS_FH(inode), desc->fh);
689 /* Don't use READDIRPLUS on directories that we believe are too large */
690 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
693 * This is our front-end to iget that looks up inodes by file handle
694 * instead of inode number.
697 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
699 struct nfs_find_desc desc = {
703 struct inode *inode = NULL;
706 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
710 printk("NFS: Buggy server - nlink == 0!\n");
714 hash = nfs_fattr_to_ino_t(fattr);
716 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
719 if (inode->i_state & I_NEW) {
720 struct nfs_inode *nfsi = NFS_I(inode);
722 /* We set i_ino for the few things that still rely on it,
726 /* We can't support update_atime(), since the server will reset it */
727 inode->i_flags |= S_NOATIME|S_NOCMTIME;
728 inode->i_mode = fattr->mode;
729 /* Why so? Because we want revalidate for devices/FIFOs, and
730 * that's precisely what we have in nfs_file_inode_operations.
732 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
733 if (S_ISREG(inode->i_mode)) {
734 inode->i_fop = &nfs_file_operations;
735 inode->i_data.a_ops = &nfs_file_aops;
736 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
737 } else if (S_ISDIR(inode->i_mode)) {
738 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
739 inode->i_fop = &nfs_dir_operations;
740 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
741 && fattr->size <= NFS_LIMIT_READDIRPLUS)
742 NFS_FLAGS(inode) |= NFS_INO_ADVISE_RDPLUS;
743 } else if (S_ISLNK(inode->i_mode))
744 inode->i_op = &nfs_symlink_inode_operations;
746 init_special_inode(inode, inode->i_mode, fattr->rdev);
748 nfsi->read_cache_jiffies = fattr->timestamp;
749 inode->i_atime = fattr->atime;
750 inode->i_mtime = fattr->mtime;
751 inode->i_ctime = fattr->ctime;
752 if (fattr->valid & NFS_ATTR_FATTR_V4)
753 nfsi->change_attr = fattr->change_attr;
754 inode->i_size = nfs_size_to_loff_t(fattr->size);
755 inode->i_nlink = fattr->nlink;
756 inode->i_uid = fattr->uid;
757 inode->i_gid = fattr->gid;
758 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
760 * report the blocks in 512byte units
762 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
763 inode->i_blksize = inode->i_sb->s_blocksize;
765 inode->i_blocks = fattr->du.nfs2.blocks;
766 inode->i_blksize = fattr->du.nfs2.blocksize;
768 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
769 nfsi->attrtimeo_timestamp = jiffies;
770 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
771 nfsi->cache_access.cred = NULL;
773 unlock_new_inode(inode);
775 nfs_refresh_inode(inode, fattr);
776 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
778 (long long)NFS_FILEID(inode),
779 atomic_read(&inode->i_count));
785 printk("nfs_fhget: iget failed\n");
789 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
792 nfs_setattr(struct dentry *dentry, struct iattr *attr)
794 struct inode *inode = dentry->d_inode;
795 struct nfs_fattr fattr;
798 if (attr->ia_valid & ATTR_SIZE) {
799 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
800 attr->ia_valid &= ~ATTR_SIZE;
803 /* Optimization: if the end result is no change, don't RPC */
804 attr->ia_valid &= NFS_VALID_ATTRS;
805 if (attr->ia_valid == 0)
809 nfs_begin_data_update(inode);
810 /* Write all dirty data if we're changing file permissions or size */
811 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
812 if (filemap_fdatawrite(inode->i_mapping) == 0)
813 filemap_fdatawait(inode->i_mapping);
816 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
818 nfs_refresh_inode(inode, &fattr);
819 if ((attr->ia_valid & ATTR_MODE) != 0) {
821 mode = inode->i_mode & ~S_IALLUGO;
822 mode |= attr->ia_mode & S_IALLUGO;
823 inode->i_mode = mode;
825 if ((attr->ia_valid & ATTR_UID) != 0)
826 inode->i_uid = attr->ia_uid;
827 if ((attr->ia_valid & ATTR_GID) != 0)
828 inode->i_gid = attr->ia_gid;
829 if ((attr->ia_valid & ATTR_SIZE) != 0) {
830 inode->i_size = attr->ia_size;
831 vmtruncate(inode, attr->ia_size);
834 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
835 NFS_FLAGS(inode) |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
836 nfs_end_data_update(inode);
842 * Wait for the inode to get unlocked.
843 * (Used for NFS_INO_LOCKED and NFS_INO_REVALIDATING).
846 nfs_wait_on_inode(struct inode *inode, int flag)
848 struct rpc_clnt *clnt = NFS_CLIENT(inode);
849 struct nfs_inode *nfsi = NFS_I(inode);
852 if (!(NFS_FLAGS(inode) & flag))
854 atomic_inc(&inode->i_count);
855 error = nfs_wait_event(clnt, nfsi->nfs_i_wait,
856 !(NFS_FLAGS(inode) & flag));
861 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
863 struct inode *inode = dentry->d_inode;
864 struct nfs_inode *nfsi = NFS_I(inode);
865 int need_atime = nfsi->flags & NFS_INO_INVALID_ATIME;
868 if (__IS_FLG(inode, MS_NOATIME))
870 else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
872 /* We may force a getattr if the user cares about atime */
874 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
876 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
878 generic_fillattr(inode, stat);
882 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
884 struct nfs_open_context *ctx;
886 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
888 atomic_set(&ctx->count, 1);
889 ctx->dentry = dget(dentry);
890 ctx->cred = get_rpccred(cred);
892 ctx->lockowner = current->files;
899 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
902 atomic_inc(&ctx->count);
906 void put_nfs_open_context(struct nfs_open_context *ctx)
908 if (atomic_dec_and_test(&ctx->count)) {
909 if (!list_empty(&ctx->list)) {
910 struct inode *inode = ctx->dentry->d_inode;
911 spin_lock(&inode->i_lock);
912 list_del(&ctx->list);
913 spin_unlock(&inode->i_lock);
915 if (ctx->state != NULL)
916 nfs4_close_state(ctx->state, ctx->mode);
917 if (ctx->cred != NULL)
918 put_rpccred(ctx->cred);
925 * Ensure that mmap has a recent RPC credential for use when writing out
928 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
930 struct inode *inode = filp->f_dentry->d_inode;
931 struct nfs_inode *nfsi = NFS_I(inode);
933 filp->private_data = get_nfs_open_context(ctx);
934 spin_lock(&inode->i_lock);
935 list_add(&ctx->list, &nfsi->open_files);
936 spin_unlock(&inode->i_lock);
939 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
941 struct nfs_inode *nfsi = NFS_I(inode);
942 struct nfs_open_context *pos, *ctx = NULL;
944 spin_lock(&inode->i_lock);
945 list_for_each_entry(pos, &nfsi->open_files, list) {
946 if ((pos->mode & mode) == mode) {
947 ctx = get_nfs_open_context(pos);
951 spin_unlock(&inode->i_lock);
955 void nfs_file_clear_open_context(struct file *filp)
957 struct inode *inode = filp->f_dentry->d_inode;
958 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
961 filp->private_data = NULL;
962 spin_lock(&inode->i_lock);
963 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
964 spin_unlock(&inode->i_lock);
965 put_nfs_open_context(ctx);
970 * These allocate and release file read/write context information.
972 int nfs_open(struct inode *inode, struct file *filp)
974 struct nfs_open_context *ctx;
975 struct rpc_cred *cred;
977 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
979 return PTR_ERR(cred);
980 ctx = alloc_nfs_open_context(filp->f_dentry, cred);
984 ctx->mode = filp->f_mode;
985 nfs_file_set_open_context(filp, ctx);
986 put_nfs_open_context(ctx);
987 if ((filp->f_mode & FMODE_WRITE) != 0)
988 nfs_begin_data_update(inode);
992 int nfs_release(struct inode *inode, struct file *filp)
994 if ((filp->f_mode & FMODE_WRITE) != 0)
995 nfs_end_data_update(inode);
996 nfs_file_clear_open_context(filp);
1001 * This function is called whenever some part of NFS notices that
1002 * the cached attributes have to be refreshed.
1005 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1007 int status = -ESTALE;
1008 struct nfs_fattr fattr;
1009 struct nfs_inode *nfsi = NFS_I(inode);
1010 unsigned long verifier;
1013 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1014 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1017 if (!inode || is_bad_inode(inode))
1019 if (NFS_STALE(inode))
1022 while (NFS_REVALIDATING(inode)) {
1023 status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
1026 if (NFS_ATTRTIMEO(inode) == 0)
1028 if (NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME))
1030 status = NFS_STALE(inode) ? -ESTALE : 0;
1033 NFS_FLAGS(inode) |= NFS_INO_REVALIDATING;
1035 /* Protect against RPC races by saving the change attribute */
1036 verifier = nfs_save_change_attribute(inode);
1037 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1039 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1041 (long long)NFS_FILEID(inode), status);
1042 if (status == -ESTALE) {
1043 nfs_zap_caches(inode);
1044 if (!S_ISDIR(inode->i_mode))
1045 NFS_FLAGS(inode) |= NFS_INO_STALE;
1050 status = nfs_update_inode(inode, &fattr, verifier);
1052 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1054 (long long)NFS_FILEID(inode), status);
1057 flags = nfsi->flags;
1059 * We may need to keep the attributes marked as invalid if
1060 * we raced with nfs_end_attr_update().
1062 if (verifier == nfsi->cache_change_attribute)
1063 nfsi->flags &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1064 /* Do the page cache invalidation */
1065 nfs_revalidate_mapping(inode, inode->i_mapping);
1066 if (flags & NFS_INO_INVALID_ACL)
1067 nfs_zap_acl_cache(inode);
1068 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1070 (long long)NFS_FILEID(inode));
1073 NFS_FLAGS(inode) &= ~NFS_INO_REVALIDATING;
1074 wake_up(&nfsi->nfs_i_wait);
1080 int nfs_attribute_timeout(struct inode *inode)
1082 struct nfs_inode *nfsi = NFS_I(inode);
1084 if (nfs_have_delegation(inode, FMODE_READ))
1086 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1090 * nfs_revalidate_inode - Revalidate the inode attributes
1091 * @server - pointer to nfs_server struct
1092 * @inode - pointer to inode struct
1094 * Updates inode attribute information by retrieving the data from the server.
1096 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1098 if (!(NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1099 && !nfs_attribute_timeout(inode))
1100 return NFS_STALE(inode) ? -ESTALE : 0;
1101 return __nfs_revalidate_inode(server, inode);
1105 * nfs_revalidate_mapping - Revalidate the pagecache
1106 * @inode - pointer to host inode
1107 * @mapping - pointer to mapping
1109 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1111 struct nfs_inode *nfsi = NFS_I(inode);
1113 if (nfsi->flags & NFS_INO_INVALID_DATA) {
1114 if (S_ISREG(inode->i_mode)) {
1115 if (filemap_fdatawrite(mapping) == 0)
1116 filemap_fdatawait(mapping);
1119 invalidate_inode_pages2(mapping);
1120 nfsi->flags &= ~NFS_INO_INVALID_DATA;
1121 if (S_ISDIR(inode->i_mode)) {
1122 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1123 /* This ensures we revalidate child dentries */
1124 nfsi->cache_change_attribute++;
1126 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1128 (long long)NFS_FILEID(inode));
1133 * nfs_begin_data_update
1134 * @inode - pointer to inode
1135 * Declare that a set of operations will update file data on the server
1137 void nfs_begin_data_update(struct inode *inode)
1139 atomic_inc(&NFS_I(inode)->data_updates);
1143 * nfs_end_data_update
1144 * @inode - pointer to inode
1145 * Declare end of the operations that will update file data
1146 * This will mark the inode as immediately needing revalidation
1147 * of its attribute cache.
1149 void nfs_end_data_update(struct inode *inode)
1151 struct nfs_inode *nfsi = NFS_I(inode);
1153 if (!nfs_have_delegation(inode, FMODE_READ)) {
1154 /* Mark the attribute cache for revalidation */
1155 nfsi->flags |= NFS_INO_INVALID_ATTR;
1156 /* Directories and symlinks: invalidate page cache too */
1157 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1158 nfsi->flags |= NFS_INO_INVALID_DATA;
1160 nfsi->cache_change_attribute ++;
1161 atomic_dec(&nfsi->data_updates);
1165 * nfs_refresh_inode - verify consistency of the inode attribute cache
1166 * @inode - pointer to inode
1167 * @fattr - updated attributes
1169 * Verifies the attribute cache. If we have just changed the attributes,
1170 * so that fattr carries weak cache consistency data, then it may
1171 * also update the ctime/mtime/change_attribute.
1173 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1175 struct nfs_inode *nfsi = NFS_I(inode);
1176 loff_t cur_size, new_isize;
1179 /* Do we hold a delegation? */
1180 if (nfs_have_delegation(inode, FMODE_READ))
1183 /* Are we in the process of updating data on the server? */
1184 data_unstable = nfs_caches_unstable(inode);
1186 if (fattr->valid & NFS_ATTR_FATTR_V4) {
1187 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1188 && nfsi->change_attr == fattr->pre_change_attr)
1189 nfsi->change_attr = fattr->change_attr;
1190 if (!data_unstable && nfsi->change_attr != fattr->change_attr)
1191 nfsi->flags |= NFS_INO_INVALID_ATTR;
1194 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1197 /* Has the inode gone and changed behind our back? */
1198 if (nfsi->fileid != fattr->fileid
1199 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1202 cur_size = i_size_read(inode);
1203 new_isize = nfs_size_to_loff_t(fattr->size);
1205 /* If we have atomic WCC data, we may update some attributes */
1206 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1207 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1208 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1209 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1210 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1213 /* Verify a few of the more important attributes */
1214 if (!data_unstable) {
1215 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)
1216 || cur_size != new_isize)
1217 nfsi->flags |= NFS_INO_INVALID_ATTR;
1218 } else if (new_isize != cur_size && nfsi->npages == 0)
1219 nfsi->flags |= NFS_INO_INVALID_ATTR;
1221 /* Have any file permissions changed? */
1222 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1223 || inode->i_uid != fattr->uid
1224 || inode->i_gid != fattr->gid)
1225 nfsi->flags |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1227 /* Has the link count changed? */
1228 if (inode->i_nlink != fattr->nlink)
1229 nfsi->flags |= NFS_INO_INVALID_ATTR;
1231 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1232 nfsi->flags |= NFS_INO_INVALID_ATIME;
1234 nfsi->read_cache_jiffies = fattr->timestamp;
1239 * Many nfs protocol calls return the new file attributes after
1240 * an operation. Here we update the inode to reflect the state
1241 * of the server's inode.
1243 * This is a bit tricky because we have to make sure all dirty pages
1244 * have been sent off to the server before calling invalidate_inode_pages.
1245 * To make sure no other process adds more write requests while we try
1246 * our best to flush them, we make them sleep during the attribute refresh.
1248 * A very similar scenario holds for the dir cache.
1250 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1252 struct nfs_inode *nfsi = NFS_I(inode);
1253 loff_t cur_isize, new_isize;
1254 unsigned int invalid = 0;
1257 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1258 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1259 atomic_read(&inode->i_count), fattr->valid);
1261 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1264 if (nfsi->fileid != fattr->fileid) {
1265 printk(KERN_ERR "%s: inode number mismatch\n"
1266 "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1268 inode->i_sb->s_id, (long long)nfsi->fileid,
1269 inode->i_sb->s_id, (long long)fattr->fileid);
1274 * Make sure the inode's type hasn't changed.
1276 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1280 * Update the read time so we don't revalidate too often.
1282 nfsi->read_cache_jiffies = fattr->timestamp;
1284 /* Are we racing with known updates of the metadata on the server? */
1285 data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1287 /* Check if our cached file size is stale */
1288 new_isize = nfs_size_to_loff_t(fattr->size);
1289 cur_isize = i_size_read(inode);
1290 if (new_isize != cur_isize) {
1291 /* Do we perhaps have any outstanding writes? */
1292 if (nfsi->npages == 0) {
1293 /* No, but did we race with nfs_end_data_update()? */
1294 if (verifier == nfsi->cache_change_attribute) {
1295 inode->i_size = new_isize;
1296 invalid |= NFS_INO_INVALID_DATA;
1298 invalid |= NFS_INO_INVALID_ATTR;
1299 } else if (new_isize > cur_isize) {
1300 inode->i_size = new_isize;
1301 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1303 dprintk("NFS: isize change on server for file %s/%ld\n",
1304 inode->i_sb->s_id, inode->i_ino);
1307 /* Check if the mtime agrees */
1308 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1309 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1310 dprintk("NFS: mtime change on server for file %s/%ld\n",
1311 inode->i_sb->s_id, inode->i_ino);
1313 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1316 if ((fattr->valid & NFS_ATTR_FATTR_V4)
1317 && nfsi->change_attr != fattr->change_attr) {
1318 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1319 inode->i_sb->s_id, inode->i_ino);
1320 nfsi->change_attr = fattr->change_attr;
1322 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1325 /* If ctime has changed we should definitely clear access+acl caches */
1326 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1328 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1329 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1331 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1333 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1334 inode->i_uid != fattr->uid ||
1335 inode->i_gid != fattr->gid)
1336 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1338 inode->i_mode = fattr->mode;
1339 inode->i_nlink = fattr->nlink;
1340 inode->i_uid = fattr->uid;
1341 inode->i_gid = fattr->gid;
1343 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1345 * report the blocks in 512byte units
1347 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1348 inode->i_blksize = inode->i_sb->s_blocksize;
1350 inode->i_blocks = fattr->du.nfs2.blocks;
1351 inode->i_blksize = fattr->du.nfs2.blocksize;
1354 /* Update attrtimeo value if we're out of the unstable period */
1355 if (invalid & NFS_INO_INVALID_ATTR) {
1356 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1357 nfsi->attrtimeo_timestamp = jiffies;
1358 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1359 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1360 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1361 nfsi->attrtimeo_timestamp = jiffies;
1363 /* Don't invalidate the data if we were to blame */
1364 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1365 || S_ISLNK(inode->i_mode)))
1366 invalid &= ~NFS_INO_INVALID_DATA;
1367 if (!nfs_have_delegation(inode, FMODE_READ))
1368 nfsi->flags |= invalid;
1373 * Big trouble! The inode has become a different object.
1376 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1377 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1380 * No need to worry about unhashing the dentry, as the
1381 * lookup validation will know that the inode is bad.
1382 * (But we fall through to invalidate the caches.)
1384 nfs_invalidate_inode(inode);
1386 NFS_FLAGS(inode) |= NFS_INO_STALE;
1391 * File system information
1394 static int nfs_set_super(struct super_block *s, void *data)
1396 s->s_fs_info = data;
1397 return set_anon_super(s, data);
1400 static int nfs_compare_super(struct super_block *sb, void *data)
1402 struct nfs_server *server = data;
1403 struct nfs_server *old = NFS_SB(sb);
1405 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1407 if (old->addr.sin_port != server->addr.sin_port)
1409 return !nfs_compare_fh(&old->fh, &server->fh);
1412 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1413 int flags, const char *dev_name, void *raw_data)
1416 struct nfs_server *server = NULL;
1417 struct super_block *s;
1418 struct nfs_fh *root;
1419 struct nfs_mount_data *data = raw_data;
1421 s = ERR_PTR(-EINVAL);
1423 dprintk("%s: missing data argument\n", __FUNCTION__);
1426 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1427 dprintk("%s: bad mount version\n", __FUNCTION__);
1430 switch (data->version) {
1436 if (data->flags & NFS_MOUNT_VER3) {
1437 dprintk("%s: mount structure version %d does not support NFSv3\n",
1442 data->root.size = NFS2_FHSIZE;
1443 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1445 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1446 dprintk("%s: mount structure version %d does not support strong security\n",
1452 memset(data->context, 0, sizeof(data->context));
1454 #ifndef CONFIG_NFS_V3
1455 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1456 s = ERR_PTR(-EPROTONOSUPPORT);
1457 if (data->flags & NFS_MOUNT_VER3) {
1458 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1461 #endif /* CONFIG_NFS_V3 */
1463 s = ERR_PTR(-ENOMEM);
1464 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1467 memset(server, 0, sizeof(struct nfs_server));
1468 /* Zero out the NFS state stuff */
1469 init_nfsv4_state(server);
1470 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1473 if (data->flags & NFS_MOUNT_VER3)
1474 root->size = data->root.size;
1476 root->size = NFS2_FHSIZE;
1477 s = ERR_PTR(-EINVAL);
1478 if (root->size > sizeof(root->data)) {
1479 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1482 memcpy(root->data, data->root.data, root->size);
1484 /* We now require that the mount process passes the remote address */
1485 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1486 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1487 dprintk("%s: mount program didn't pass remote address!\n",
1492 /* Fire up rpciod if not yet running */
1493 s = ERR_PTR(rpciod_up());
1495 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1496 __FUNCTION__, PTR_ERR(s));
1500 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1501 if (IS_ERR(s) || s->s_root)
1502 goto out_rpciod_down;
1506 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1508 up_write(&s->s_umount);
1509 deactivate_super(s);
1510 return ERR_PTR(error);
1512 s->s_flags |= MS_ACTIVE;
1521 static void nfs_kill_super(struct super_block *s)
1523 struct nfs_server *server = NFS_SB(s);
1527 if (!IS_ERR(server->client))
1528 rpc_shutdown_client(server->client);
1529 if (!IS_ERR(server->client_sys))
1530 rpc_shutdown_client(server->client_sys);
1531 if (!IS_ERR(server->client_acl))
1532 rpc_shutdown_client(server->client_acl);
1534 if (!(server->flags & NFS_MOUNT_NONLM))
1535 lockd_down(); /* release rpc.lockd */
1537 rpciod_down(); /* release rpciod */
1539 if (server->hostname != NULL)
1540 kfree(server->hostname);
1544 static struct file_system_type nfs_fs_type = {
1545 .owner = THIS_MODULE,
1547 .get_sb = nfs_get_sb,
1548 .kill_sb = nfs_kill_super,
1549 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1552 #ifdef CONFIG_NFS_V4
1554 static void nfs4_clear_inode(struct inode *);
1557 static struct super_operations nfs4_sops = {
1558 .alloc_inode = nfs_alloc_inode,
1559 .destroy_inode = nfs_destroy_inode,
1560 .write_inode = nfs_write_inode,
1561 .delete_inode = nfs_delete_inode,
1562 .statfs = nfs_statfs,
1563 .clear_inode = nfs4_clear_inode,
1564 .umount_begin = nfs_umount_begin,
1565 .show_options = nfs_show_options,
1569 * Clean out any remaining NFSv4 state that might be left over due
1570 * to open() calls that passed nfs_atomic_lookup, but failed to call
1573 static void nfs4_clear_inode(struct inode *inode)
1575 struct nfs_inode *nfsi = NFS_I(inode);
1577 /* If we are holding a delegation, return it! */
1578 if (nfsi->delegation != NULL)
1579 nfs_inode_return_delegation(inode);
1580 /* First call standard NFS clear_inode() code */
1581 nfs_clear_inode(inode);
1582 /* Now clear out any remaining state */
1583 while (!list_empty(&nfsi->open_states)) {
1584 struct nfs4_state *state;
1586 state = list_entry(nfsi->open_states.next,
1589 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1592 (long long)NFS_FILEID(inode),
1594 BUG_ON(atomic_read(&state->count) != 1);
1595 nfs4_close_state(state, state->state);
1600 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1602 struct nfs_server *server;
1603 struct nfs4_client *clp = NULL;
1604 struct rpc_xprt *xprt = NULL;
1605 struct rpc_clnt *clnt = NULL;
1606 struct rpc_timeout timeparms;
1607 rpc_authflavor_t authflavour;
1608 int proto, err = -EIO;
1610 sb->s_blocksize_bits = 0;
1611 sb->s_blocksize = 0;
1612 server = NFS_SB(sb);
1613 if (data->rsize != 0)
1614 server->rsize = nfs_block_size(data->rsize, NULL);
1615 if (data->wsize != 0)
1616 server->wsize = nfs_block_size(data->wsize, NULL);
1617 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1618 server->caps = NFS_CAP_ATOMIC_OPEN;
1620 server->acregmin = data->acregmin*HZ;
1621 server->acregmax = data->acregmax*HZ;
1622 server->acdirmin = data->acdirmin*HZ;
1623 server->acdirmax = data->acdirmax*HZ;
1625 server->rpc_ops = &nfs_v4_clientops;
1626 /* Initialize timeout values */
1628 timeparms.to_initval = data->timeo * HZ / 10;
1629 timeparms.to_retries = data->retrans;
1630 timeparms.to_exponential = 1;
1631 if (!timeparms.to_retries)
1632 timeparms.to_retries = 5;
1634 proto = data->proto;
1635 /* Which IP protocol do we use? */
1638 timeparms.to_maxval = RPC_MAX_TCP_TIMEOUT;
1639 if (!timeparms.to_initval)
1640 timeparms.to_initval = 600 * HZ / 10;
1643 timeparms.to_maxval = RPC_MAX_UDP_TIMEOUT;
1644 if (!timeparms.to_initval)
1645 timeparms.to_initval = 11 * HZ / 10;
1651 clp = nfs4_get_client(&server->addr.sin_addr);
1653 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1657 /* Now create transport and client */
1658 authflavour = RPC_AUTH_UNIX;
1659 if (data->auth_flavourlen != 0) {
1660 if (data->auth_flavourlen != 1) {
1661 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1662 __FUNCTION__, data->auth_flavourlen);
1666 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1672 down_write(&clp->cl_sem);
1673 if (IS_ERR(clp->cl_rpcclient)) {
1674 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1676 up_write(&clp->cl_sem);
1677 err = PTR_ERR(xprt);
1678 dprintk("%s: cannot create RPC transport. Error = %d\n",
1682 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1683 server->rpc_ops->version, authflavour);
1685 up_write(&clp->cl_sem);
1686 err = PTR_ERR(clnt);
1687 dprintk("%s: cannot create RPC client. Error = %d\n",
1692 clnt->cl_softrtry = 1;
1693 clnt->cl_chatty = 1;
1694 clp->cl_rpcclient = clnt;
1695 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1696 if (IS_ERR(clp->cl_cred)) {
1697 up_write(&clp->cl_sem);
1698 err = PTR_ERR(clp->cl_cred);
1699 clp->cl_cred = NULL;
1702 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1705 if (list_empty(&clp->cl_superblocks)) {
1706 err = nfs4_init_client(clp);
1708 up_write(&clp->cl_sem);
1712 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1713 clnt = rpc_clone_client(clp->cl_rpcclient);
1715 server->nfs4_state = clp;
1716 up_write(&clp->cl_sem);
1720 err = PTR_ERR(clnt);
1721 dprintk("%s: cannot create RPC client. Error = %d\n",
1726 server->client = clnt;
1728 if (server->nfs4_state->cl_idmap == NULL) {
1729 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1733 if (clnt->cl_auth->au_flavor != authflavour) {
1734 struct rpc_auth *auth;
1736 auth = rpcauth_create(authflavour, clnt);
1738 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1739 return PTR_ERR(auth);
1743 sb->s_time_gran = 1;
1745 sb->s_op = &nfs4_sops;
1746 err = nfs_sb_init(sb, authflavour);
1751 nfs4_put_client(clp);
1755 static int nfs4_compare_super(struct super_block *sb, void *data)
1757 struct nfs_server *server = data;
1758 struct nfs_server *old = NFS_SB(sb);
1760 if (strcmp(server->hostname, old->hostname) != 0)
1762 if (strcmp(server->mnt_path, old->mnt_path) != 0)
1768 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1773 return ERR_PTR(-EINVAL);
1774 if (src->len < maxlen)
1777 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1779 return ERR_PTR(-ENOMEM);
1781 if (copy_from_user(dst, src->data, maxlen)) {
1784 return ERR_PTR(-EFAULT);
1790 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1791 int flags, const char *dev_name, void *raw_data)
1794 struct nfs_server *server;
1795 struct super_block *s;
1796 struct nfs4_mount_data *data = raw_data;
1800 dprintk("%s: missing data argument\n", __FUNCTION__);
1801 return ERR_PTR(-EINVAL);
1803 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1804 dprintk("%s: bad mount version\n", __FUNCTION__);
1805 return ERR_PTR(-EINVAL);
1808 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1810 return ERR_PTR(-ENOMEM);
1811 memset(server, 0, sizeof(struct nfs_server));
1812 /* Zero out the NFS state stuff */
1813 init_nfsv4_state(server);
1814 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1816 p = nfs_copy_user_string(NULL, &data->hostname, 256);
1819 server->hostname = p;
1821 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1824 server->mnt_path = p;
1826 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1827 sizeof(server->ip_addr) - 1);
1831 /* We now require that the mount process passes the remote address */
1832 if (data->host_addrlen != sizeof(server->addr)) {
1833 s = ERR_PTR(-EINVAL);
1836 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1837 s = ERR_PTR(-EFAULT);
1840 if (server->addr.sin_family != AF_INET ||
1841 server->addr.sin_addr.s_addr == INADDR_ANY) {
1842 dprintk("%s: mount program didn't pass remote IP address!\n",
1844 s = ERR_PTR(-EINVAL);
1848 /* Fire up rpciod if not yet running */
1849 s = ERR_PTR(rpciod_up());
1851 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1852 __FUNCTION__, PTR_ERR(s));
1856 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1858 if (IS_ERR(s) || s->s_root)
1863 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1865 up_write(&s->s_umount);
1866 deactivate_super(s);
1867 return ERR_PTR(error);
1869 s->s_flags |= MS_ACTIVE;
1872 s = (struct super_block *)p;
1874 if (server->mnt_path)
1875 kfree(server->mnt_path);
1876 if (server->hostname)
1877 kfree(server->hostname);
1882 static void nfs4_kill_super(struct super_block *sb)
1884 struct nfs_server *server = NFS_SB(sb);
1886 nfs_return_all_delegations(sb);
1887 kill_anon_super(sb);
1889 nfs4_renewd_prepare_shutdown(server);
1891 if (server->client != NULL && !IS_ERR(server->client))
1892 rpc_shutdown_client(server->client);
1893 rpciod_down(); /* release rpciod */
1895 destroy_nfsv4_state(server);
1897 if (server->hostname != NULL)
1898 kfree(server->hostname);
1902 static struct file_system_type nfs4_fs_type = {
1903 .owner = THIS_MODULE,
1905 .get_sb = nfs4_get_sb,
1906 .kill_sb = nfs4_kill_super,
1907 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1910 #define nfs4_init_once(nfsi) \
1912 INIT_LIST_HEAD(&(nfsi)->open_states); \
1913 nfsi->delegation = NULL; \
1914 nfsi->delegation_state = 0; \
1915 init_rwsem(&nfsi->rwsem); \
1917 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1918 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1920 #define nfs4_init_once(nfsi) \
1922 #define register_nfs4fs() (0)
1923 #define unregister_nfs4fs()
1926 extern int nfs_init_nfspagecache(void);
1927 extern void nfs_destroy_nfspagecache(void);
1928 extern int nfs_init_readpagecache(void);
1929 extern void nfs_destroy_readpagecache(void);
1930 extern int nfs_init_writepagecache(void);
1931 extern void nfs_destroy_writepagecache(void);
1932 #ifdef CONFIG_NFS_DIRECTIO
1933 extern int nfs_init_directcache(void);
1934 extern void nfs_destroy_directcache(void);
1937 static kmem_cache_t * nfs_inode_cachep;
1939 static struct inode *nfs_alloc_inode(struct super_block *sb)
1941 struct nfs_inode *nfsi;
1942 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
1946 #ifdef CONFIG_NFS_V3_ACL
1947 nfsi->acl_access = ERR_PTR(-EAGAIN);
1948 nfsi->acl_default = ERR_PTR(-EAGAIN);
1950 #ifdef CONFIG_NFS_V4
1951 nfsi->nfs4_acl = NULL;
1952 #endif /* CONFIG_NFS_V4 */
1953 return &nfsi->vfs_inode;
1956 static void nfs_destroy_inode(struct inode *inode)
1958 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1961 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
1963 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1965 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1966 SLAB_CTOR_CONSTRUCTOR) {
1967 inode_init_once(&nfsi->vfs_inode);
1968 spin_lock_init(&nfsi->req_lock);
1969 INIT_LIST_HEAD(&nfsi->dirty);
1970 INIT_LIST_HEAD(&nfsi->commit);
1971 INIT_LIST_HEAD(&nfsi->open_files);
1972 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1973 atomic_set(&nfsi->data_updates, 0);
1977 init_waitqueue_head(&nfsi->nfs_i_wait);
1978 nfs4_init_once(nfsi);
1982 static int nfs_init_inodecache(void)
1984 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1985 sizeof(struct nfs_inode),
1986 0, SLAB_RECLAIM_ACCOUNT,
1988 if (nfs_inode_cachep == NULL)
1994 static void nfs_destroy_inodecache(void)
1996 if (kmem_cache_destroy(nfs_inode_cachep))
1997 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2003 static int __init init_nfs_fs(void)
2007 err = nfs_init_nfspagecache();
2011 err = nfs_init_inodecache();
2015 err = nfs_init_readpagecache();
2019 err = nfs_init_writepagecache();
2023 #ifdef CONFIG_NFS_DIRECTIO
2024 err = nfs_init_directcache();
2029 #ifdef CONFIG_PROC_FS
2030 rpc_proc_register(&nfs_rpcstat);
2032 err = register_filesystem(&nfs_fs_type);
2035 if ((err = register_nfs4fs()) != 0)
2039 #ifdef CONFIG_PROC_FS
2040 rpc_proc_unregister("nfs");
2042 nfs_destroy_writepagecache();
2043 #ifdef CONFIG_NFS_DIRECTIO
2045 nfs_destroy_directcache();
2048 nfs_destroy_readpagecache();
2050 nfs_destroy_inodecache();
2052 nfs_destroy_nfspagecache();
2057 static void __exit exit_nfs_fs(void)
2059 #ifdef CONFIG_NFS_DIRECTIO
2060 nfs_destroy_directcache();
2062 nfs_destroy_writepagecache();
2063 nfs_destroy_readpagecache();
2064 nfs_destroy_inodecache();
2065 nfs_destroy_nfspagecache();
2066 #ifdef CONFIG_PROC_FS
2067 rpc_proc_unregister("nfs");
2069 unregister_filesystem(&nfs_fs_type);
2070 unregister_nfs4fs();
2073 /* Not quite true; I just maintain it */
2074 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2075 MODULE_LICENSE("GPL");
2077 module_init(init_nfs_fs)
2078 module_exit(exit_nfs_fs)