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>
38 #include <linux/vserver/xid.h>
40 #include <asm/system.h>
41 #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 *);
68 static struct super_operations nfs_sops = {
69 .alloc_inode = nfs_alloc_inode,
70 .destroy_inode = nfs_destroy_inode,
71 .write_inode = nfs_write_inode,
72 .delete_inode = nfs_delete_inode,
74 .clear_inode = nfs_clear_inode,
75 .umount_begin = nfs_umount_begin,
76 .show_options = nfs_show_options,
82 struct rpc_stat nfs_rpcstat = {
83 .program = &nfs_program
85 static struct rpc_version * nfs_version[] = {
89 #if defined(CONFIG_NFS_V3)
91 #elif defined(CONFIG_NFS_V4)
94 #if defined(CONFIG_NFS_V4)
99 struct rpc_program nfs_program = {
101 .number = NFS_PROGRAM,
102 .nrvers = sizeof(nfs_version) / sizeof(nfs_version[0]),
103 .version = nfs_version,
104 .stats = &nfs_rpcstat,
105 .pipe_dir_name = "/nfs",
108 static inline unsigned long
109 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
111 return nfs_fileid_to_ino_t(fattr->fileid);
115 nfs_write_inode(struct inode *inode, int sync)
117 int flags = sync ? FLUSH_WAIT : 0;
120 ret = nfs_commit_inode(inode, 0, 0, flags);
127 nfs_delete_inode(struct inode * inode)
129 dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
133 * The following should never happen...
135 if (nfs_have_writebacks(inode)) {
136 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
143 * For the moment, the only task for the NFS clear_inode method is to
144 * release the mmap credential
147 nfs_clear_inode(struct inode *inode)
149 struct nfs_inode *nfsi = NFS_I(inode);
150 struct rpc_cred *cred;
153 BUG_ON (!list_empty(&nfsi->open_files));
154 cred = nfsi->cache_access.cred;
157 BUG_ON(atomic_read(&nfsi->data_updates) != 0);
161 nfs_umount_begin(struct super_block *sb)
163 struct nfs_server *server = NFS_SB(sb);
164 struct rpc_clnt *rpc;
166 /* -EIO all pending I/O */
167 if ((rpc = server->client) != NULL)
168 rpc_killall_tasks(rpc);
172 static inline unsigned long
173 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
175 /* make sure blocksize is a power of two */
176 if ((bsize & (bsize - 1)) || nrbitsp) {
177 unsigned char nrbits;
179 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
190 * Calculate the number of 512byte blocks used.
192 static inline unsigned long
193 nfs_calc_block_size(u64 tsize)
195 loff_t used = (tsize + 511) >> 9;
196 return (used > ULONG_MAX) ? ULONG_MAX : used;
200 * Compute and set NFS server blocksize
202 static inline unsigned long
203 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
206 bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
207 else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
208 bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
210 return nfs_block_bits(bsize, nrbitsp);
214 * Obtain the root inode of the file system.
216 static struct inode *
217 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
219 struct nfs_server *server = NFS_SB(sb);
223 error = server->rpc_ops->getroot(server, rootfh, fsinfo);
225 dprintk("nfs_get_root: getattr error = %d\n", -error);
226 return ERR_PTR(error);
229 rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
231 return ERR_PTR(-ENOMEM);
236 * Do NFS version-independent mount processing, and sanity checking
239 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
241 struct nfs_server *server;
242 struct inode *root_inode;
243 struct nfs_fattr fattr;
244 struct nfs_fsinfo fsinfo = {
247 struct nfs_pathconf pathinfo = {
250 int no_root_error = 0;
252 /* We probably want something more informative here */
253 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
257 sb->s_magic = NFS_SUPER_MAGIC;
259 root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
260 /* Did getting the root inode fail? */
261 if (IS_ERR(root_inode)) {
262 no_root_error = PTR_ERR(root_inode);
265 sb->s_root = d_alloc_root(root_inode);
267 no_root_error = -ENOMEM;
270 sb->s_root->d_op = server->rpc_ops->dentry_ops;
272 /* Get some general file system info */
273 if (server->namelen == 0 &&
274 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
275 server->namelen = pathinfo.max_namelen;
276 /* Work out a lot of parameters */
277 if (server->rsize == 0)
278 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
279 if (server->wsize == 0)
280 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
282 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
283 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
284 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
285 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
287 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
288 if (server->rpages > NFS_READ_MAXIOV) {
289 server->rpages = NFS_READ_MAXIOV;
290 server->rsize = server->rpages << PAGE_CACHE_SHIFT;
293 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
294 if (server->wpages > NFS_WRITE_MAXIOV) {
295 server->wpages = NFS_WRITE_MAXIOV;
296 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
299 if (sb->s_blocksize == 0)
300 sb->s_blocksize = nfs_block_bits(server->wsize,
301 &sb->s_blocksize_bits);
302 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
304 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
305 if (server->dtsize > PAGE_CACHE_SIZE)
306 server->dtsize = PAGE_CACHE_SIZE;
307 if (server->dtsize > server->rsize)
308 server->dtsize = server->rsize;
310 if (server->flags & NFS_MOUNT_NOAC) {
311 server->acregmin = server->acregmax = 0;
312 server->acdirmin = server->acdirmax = 0;
313 sb->s_flags |= MS_SYNCHRONOUS;
315 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
317 if (server->flags & NFS_MOUNT_TAGXID)
318 sb->s_flags |= MS_TAGXID;
320 sb->s_maxbytes = fsinfo.maxfilesize;
321 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
322 sb->s_maxbytes = MAX_LFS_FILESIZE;
324 /* We're airborne Set socket buffersize */
325 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
327 /* Yargs. It didn't work out. */
329 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
330 if (!IS_ERR(root_inode))
332 return no_root_error;
336 * Create an RPC client handle.
338 static struct rpc_clnt *
339 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
341 struct rpc_timeout timeparms;
342 struct rpc_xprt *xprt = NULL;
343 struct rpc_clnt *clnt = NULL;
344 int tcp = (data->flags & NFS_MOUNT_TCP);
346 /* Initialize timeout values */
347 timeparms.to_initval = data->timeo * HZ / 10;
348 timeparms.to_retries = data->retrans;
349 timeparms.to_maxval = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
350 timeparms.to_exponential = 1;
352 if (!timeparms.to_initval)
353 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
354 if (!timeparms.to_retries)
355 timeparms.to_retries = 5;
357 /* create transport and client */
358 xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
359 &server->addr, &timeparms);
361 printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
362 return (struct rpc_clnt *)xprt;
364 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
365 server->rpc_ops->version, data->pseudoflavor);
367 printk(KERN_WARNING "NFS: cannot create RPC client.\n");
371 clnt->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
372 clnt->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
373 clnt->cl_droppriv = (server->flags & NFS_MOUNT_BROKEN_SUID) ? 1 : 0;
374 clnt->cl_tagxid = (server->flags & NFS_MOUNT_TAGXID) ? 1 : 0;
385 * The way this works is that the mount process passes a structure
386 * in the data argument which contains the server's IP address
387 * and the root file handle obtained from the server's mount
388 * daemon. We stash these away in the private superblock fields.
391 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
393 struct nfs_server *server;
394 rpc_authflavor_t authflavor;
397 sb->s_blocksize_bits = 0;
400 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
402 server->rsize = nfs_block_size(data->rsize, NULL);
404 server->wsize = nfs_block_size(data->wsize, NULL);
405 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
407 server->acregmin = data->acregmin*HZ;
408 server->acregmax = data->acregmax*HZ;
409 server->acdirmin = data->acdirmin*HZ;
410 server->acdirmax = data->acdirmax*HZ;
412 /* Start lockd here, before we might error out */
413 if (!(server->flags & NFS_MOUNT_NONLM))
416 server->namelen = data->namlen;
417 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
418 if (!server->hostname)
420 strcpy(server->hostname, data->hostname);
422 /* Check NFS protocol revision and initialize RPC op vector
423 * and file handle pool. */
424 if (server->flags & NFS_MOUNT_VER3) {
426 server->rpc_ops = &nfs_v3_clientops;
427 server->caps |= NFS_CAP_READDIRPLUS;
428 if (data->version < 4) {
429 printk(KERN_NOTICE "NFS: NFSv3 not supported by mount program.\n");
433 printk(KERN_NOTICE "NFS: NFSv3 not supported.\n");
437 server->rpc_ops = &nfs_v2_clientops;
440 /* Fill in pseudoflavor for mount version < 5 */
441 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
442 data->pseudoflavor = RPC_AUTH_UNIX;
443 authflavor = data->pseudoflavor; /* save for sb_init() */
444 /* XXX maybe we want to add a server->pseudoflavor field */
446 /* Create RPC client handles */
447 server->client = nfs_create_client(server, data);
448 if (IS_ERR(server->client))
449 return PTR_ERR(server->client);
450 /* RFC 2623, sec 2.3.2 */
451 if (authflavor != RPC_AUTH_UNIX) {
452 server->client_sys = rpc_clone_client(server->client);
453 if (IS_ERR(server->client_sys))
454 return PTR_ERR(server->client_sys);
455 if (!rpcauth_create(RPC_AUTH_UNIX, server->client_sys))
458 atomic_inc(&server->client->cl_count);
459 server->client_sys = server->client;
462 if (server->flags & NFS_MOUNT_VER3) {
463 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
464 server->namelen = NFS3_MAXNAMLEN;
467 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
468 server->namelen = NFS2_MAXNAMLEN;
471 sb->s_op = &nfs_sops;
472 return nfs_sb_init(sb, authflavor);
476 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
478 struct nfs_server *server = NFS_SB(sb);
479 unsigned char blockbits;
480 unsigned long blockres;
481 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
482 struct nfs_fattr fattr;
483 struct nfs_fsstat res = {
490 error = server->rpc_ops->statfs(server, rootfh, &res);
491 buf->f_type = NFS_SUPER_MAGIC;
496 * Current versions of glibc do not correctly handle the
497 * case where f_frsize != f_bsize. Eventually we want to
498 * report the value of wtmult in this field.
500 buf->f_frsize = sb->s_blocksize;
503 * On most *nix systems, f_blocks, f_bfree, and f_bavail
504 * are reported in units of f_frsize. Linux hasn't had
505 * an f_frsize field in its statfs struct until recently,
506 * thus historically Linux's sys_statfs reports these
507 * fields in units of f_bsize.
509 buf->f_bsize = sb->s_blocksize;
510 blockbits = sb->s_blocksize_bits;
511 blockres = (1 << blockbits) - 1;
512 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
513 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
514 buf->f_bavail = (res.abytes + blockres) >> blockbits;
516 buf->f_files = res.tfiles;
517 buf->f_ffree = res.afiles;
519 buf->f_namelen = server->namelen;
526 printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
527 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
532 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
534 static struct proc_nfs_info {
539 { NFS_MOUNT_SOFT, ",soft", ",hard" },
540 { NFS_MOUNT_INTR, ",intr", "" },
541 { NFS_MOUNT_POSIX, ",posix", "" },
542 { NFS_MOUNT_TCP, ",tcp", ",udp" },
543 { NFS_MOUNT_NOCTO, ",nocto", "" },
544 { NFS_MOUNT_NOAC, ",noac", "" },
545 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
546 { NFS_MOUNT_BROKEN_SUID, ",broken_suid", "" },
547 { NFS_MOUNT_TAGXID, ",tagxid", "" },
550 struct proc_nfs_info *nfs_infop;
551 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
553 seq_printf(m, ",v%d", nfss->rpc_ops->version);
554 seq_printf(m, ",rsize=%d", nfss->rsize);
555 seq_printf(m, ",wsize=%d", nfss->wsize);
556 if (nfss->acregmin != 3*HZ)
557 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
558 if (nfss->acregmax != 60*HZ)
559 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
560 if (nfss->acdirmin != 30*HZ)
561 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
562 if (nfss->acdirmax != 60*HZ)
563 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
564 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
565 if (nfss->flags & nfs_infop->flag)
566 seq_puts(m, nfs_infop->str);
568 seq_puts(m, nfs_infop->nostr);
570 seq_puts(m, ",addr=");
571 seq_escape(m, nfss->hostname, " \t\n\\");
576 * Invalidate the local caches
579 nfs_zap_caches(struct inode *inode)
581 struct nfs_inode *nfsi = NFS_I(inode);
582 int mode = inode->i_mode;
584 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
585 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
587 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
588 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
589 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS;
591 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS;
595 * Invalidate, but do not unhash, the inode
598 nfs_invalidate_inode(struct inode *inode)
600 umode_t save_mode = inode->i_mode;
602 make_bad_inode(inode);
603 inode->i_mode = save_mode;
604 nfs_zap_caches(inode);
607 struct nfs_find_desc {
609 struct nfs_fattr *fattr;
613 * In NFSv3 we can have 64bit inode numbers. In order to support
614 * this, and re-exported directories (also seen in NFSv2)
615 * we are forced to allow 2 different inodes to have the same
619 nfs_find_actor(struct inode *inode, void *opaque)
621 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
622 struct nfs_fh *fh = desc->fh;
623 struct nfs_fattr *fattr = desc->fattr;
625 if (NFS_FILEID(inode) != fattr->fileid)
627 if (nfs_compare_fh(NFS_FH(inode), fh))
629 if (is_bad_inode(inode) || NFS_STALE(inode))
635 nfs_init_locked(struct inode *inode, void *opaque)
637 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
638 struct nfs_fattr *fattr = desc->fattr;
640 NFS_FILEID(inode) = fattr->fileid;
641 nfs_copy_fh(NFS_FH(inode), desc->fh);
645 /* Don't use READDIRPLUS on directories that we believe are too large */
646 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
649 * This is our front-end to iget that looks up inodes by file handle
650 * instead of inode number.
653 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
655 struct nfs_find_desc desc = {
659 struct inode *inode = NULL;
662 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
666 printk("NFS: Buggy server - nlink == 0!\n");
670 hash = nfs_fattr_to_ino_t(fattr);
672 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
675 if (inode->i_state & I_NEW) {
676 struct nfs_inode *nfsi = NFS_I(inode);
678 /* We set i_ino for the few things that still rely on it,
682 /* We can't support update_atime(), since the server will reset it */
683 inode->i_flags |= S_NOATIME|S_NOCMTIME;
684 inode->i_mode = fattr->mode;
685 /* Why so? Because we want revalidate for devices/FIFOs, and
686 * that's precisely what we have in nfs_file_inode_operations.
688 inode->i_op = &nfs_file_inode_operations;
689 if (S_ISREG(inode->i_mode)) {
690 inode->i_fop = &nfs_file_operations;
691 inode->i_data.a_ops = &nfs_file_aops;
692 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
693 } else if (S_ISDIR(inode->i_mode)) {
694 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
695 inode->i_fop = &nfs_dir_operations;
696 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
697 && fattr->size <= NFS_LIMIT_READDIRPLUS)
698 NFS_FLAGS(inode) |= NFS_INO_ADVISE_RDPLUS;
699 } else if (S_ISLNK(inode->i_mode))
700 inode->i_op = &nfs_symlink_inode_operations;
702 init_special_inode(inode, inode->i_mode, fattr->rdev);
704 nfsi->read_cache_jiffies = fattr->timestamp;
705 inode->i_atime = fattr->atime;
706 inode->i_mtime = fattr->mtime;
707 inode->i_ctime = fattr->ctime;
708 if (fattr->valid & NFS_ATTR_FATTR_V4)
709 nfsi->change_attr = fattr->change_attr;
710 inode->i_size = nfs_size_to_loff_t(fattr->size);
711 inode->i_nlink = fattr->nlink;
712 inode->i_uid = INOXID_UID(XID_TAG(inode), fattr->uid, fattr->gid);
713 inode->i_gid = INOXID_GID(XID_TAG(inode), fattr->uid, fattr->gid);
714 inode->i_xid = INOXID_XID(XID_TAG(inode), fattr->uid, fattr->gid, 0);
715 /* maybe fattr->xid someday */
716 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
718 * report the blocks in 512byte units
720 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
721 inode->i_blksize = inode->i_sb->s_blocksize;
723 inode->i_blocks = fattr->du.nfs2.blocks;
724 inode->i_blksize = fattr->du.nfs2.blocksize;
726 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
727 nfsi->attrtimeo_timestamp = jiffies;
728 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
729 nfsi->cache_access.cred = NULL;
731 unlock_new_inode(inode);
733 nfs_refresh_inode(inode, fattr);
734 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
736 (long long)NFS_FILEID(inode),
737 atomic_read(&inode->i_count));
743 make_bad_inode(inode);
748 printk("nfs_fhget: iget failed\n");
752 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
755 nfs_setattr(struct dentry *dentry, struct iattr *attr)
757 struct inode *inode = dentry->d_inode;
758 struct nfs_fattr fattr;
761 if (attr->ia_valid & ATTR_SIZE) {
762 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
763 attr->ia_valid &= ~ATTR_SIZE;
766 /* Optimization: if the end result is no change, don't RPC */
767 attr->ia_valid &= NFS_VALID_ATTRS;
768 if (attr->ia_valid == 0)
772 nfs_begin_data_update(inode);
773 /* Write all dirty data if we're changing file permissions or size */
774 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
775 if (filemap_fdatawrite(inode->i_mapping) == 0)
776 filemap_fdatawait(inode->i_mapping);
779 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
781 nfs_refresh_inode(inode, &fattr);
782 if ((attr->ia_valid & ATTR_MODE) != 0) {
784 mode = inode->i_mode & ~S_IALLUGO;
785 mode |= attr->ia_mode & S_IALLUGO;
786 inode->i_mode = mode;
788 if ((attr->ia_valid & ATTR_UID) != 0)
789 inode->i_uid = attr->ia_uid;
790 if ((attr->ia_valid & ATTR_GID) != 0)
791 inode->i_gid = attr->ia_gid;
792 if ((attr->ia_valid & ATTR_XID) != 0)
793 inode->i_xid = attr->ia_xid;
794 if ((attr->ia_valid & ATTR_SIZE) != 0) {
795 inode->i_size = attr->ia_size;
796 vmtruncate(inode, attr->ia_size);
799 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
800 NFS_FLAGS(inode) |= NFS_INO_INVALID_ACCESS;
801 nfs_end_data_update(inode);
807 * Wait for the inode to get unlocked.
808 * (Used for NFS_INO_LOCKED and NFS_INO_REVALIDATING).
811 nfs_wait_on_inode(struct inode *inode, int flag)
813 struct rpc_clnt *clnt = NFS_CLIENT(inode);
814 struct nfs_inode *nfsi = NFS_I(inode);
817 if (!(NFS_FLAGS(inode) & flag))
819 atomic_inc(&inode->i_count);
820 error = nfs_wait_event(clnt, nfsi->nfs_i_wait,
821 !(NFS_FLAGS(inode) & flag));
826 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
828 struct inode *inode = dentry->d_inode;
829 struct nfs_inode *nfsi = NFS_I(inode);
830 int need_atime = nfsi->flags & NFS_INO_INVALID_ATIME;
833 if (__IS_FLG(inode, MS_NOATIME))
835 else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
837 /* We may force a getattr if the user cares about atime */
839 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
841 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
843 generic_fillattr(inode, stat);
847 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
849 struct nfs_open_context *ctx;
851 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
853 atomic_set(&ctx->count, 1);
854 ctx->dentry = dget(dentry);
855 ctx->cred = get_rpccred(cred);
857 ctx->lockowner = current->files;
859 init_waitqueue_head(&ctx->waitq);
864 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
867 atomic_inc(&ctx->count);
871 void put_nfs_open_context(struct nfs_open_context *ctx)
873 if (atomic_dec_and_test(&ctx->count)) {
874 if (ctx->state != NULL)
875 nfs4_close_state(ctx->state, ctx->mode);
876 if (ctx->cred != NULL)
877 put_rpccred(ctx->cred);
884 * Ensure that mmap has a recent RPC credential for use when writing out
887 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
889 struct inode *inode = filp->f_dentry->d_inode;
890 struct nfs_inode *nfsi = NFS_I(inode);
892 filp->private_data = get_nfs_open_context(ctx);
893 spin_lock(&inode->i_lock);
894 list_add(&ctx->list, &nfsi->open_files);
895 spin_unlock(&inode->i_lock);
898 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
900 struct nfs_inode *nfsi = NFS_I(inode);
901 struct nfs_open_context *pos, *ctx = NULL;
903 spin_lock(&inode->i_lock);
904 list_for_each_entry(pos, &nfsi->open_files, list) {
905 if ((pos->mode & mode) == mode) {
906 ctx = get_nfs_open_context(pos);
910 spin_unlock(&inode->i_lock);
914 void nfs_file_clear_open_context(struct file *filp)
916 struct inode *inode = filp->f_dentry->d_inode;
917 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
920 filp->private_data = NULL;
921 spin_lock(&inode->i_lock);
922 list_del(&ctx->list);
923 spin_unlock(&inode->i_lock);
924 put_nfs_open_context(ctx);
929 * These allocate and release file read/write context information.
931 int nfs_open(struct inode *inode, struct file *filp)
933 struct nfs_open_context *ctx;
934 struct rpc_cred *cred;
936 if ((cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0)) == NULL)
938 ctx = alloc_nfs_open_context(filp->f_dentry, cred);
942 ctx->mode = filp->f_mode;
943 nfs_file_set_open_context(filp, ctx);
944 put_nfs_open_context(ctx);
945 if ((filp->f_mode & FMODE_WRITE) != 0)
946 nfs_begin_data_update(inode);
950 int nfs_release(struct inode *inode, struct file *filp)
952 if ((filp->f_mode & FMODE_WRITE) != 0)
953 nfs_end_data_update(inode);
954 nfs_file_clear_open_context(filp);
959 * This function is called whenever some part of NFS notices that
960 * the cached attributes have to be refreshed.
963 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
965 int status = -ESTALE;
966 struct nfs_fattr fattr;
967 struct nfs_inode *nfsi = NFS_I(inode);
968 unsigned long verifier;
971 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
972 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
975 if (!inode || is_bad_inode(inode))
977 if (NFS_STALE(inode))
980 while (NFS_REVALIDATING(inode)) {
981 status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
984 if (NFS_ATTRTIMEO(inode) == 0)
986 if (NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME))
988 status = NFS_STALE(inode) ? -ESTALE : 0;
991 NFS_FLAGS(inode) |= NFS_INO_REVALIDATING;
993 /* Protect against RPC races by saving the change attribute */
994 verifier = nfs_save_change_attribute(inode);
995 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
997 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
999 (long long)NFS_FILEID(inode), status);
1000 if (status == -ESTALE) {
1001 nfs_zap_caches(inode);
1002 if (!S_ISDIR(inode->i_mode))
1003 NFS_FLAGS(inode) |= NFS_INO_STALE;
1008 status = nfs_update_inode(inode, &fattr, verifier);
1010 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1012 (long long)NFS_FILEID(inode), status);
1015 flags = nfsi->flags;
1017 * We may need to keep the attributes marked as invalid if
1018 * we raced with nfs_end_attr_update().
1020 if (verifier == nfsi->cache_change_attribute)
1021 nfsi->flags &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1022 /* Do the page cache invalidation */
1023 if (flags & NFS_INO_INVALID_DATA) {
1024 if (S_ISREG(inode->i_mode)) {
1025 if (filemap_fdatawrite(inode->i_mapping) == 0)
1026 filemap_fdatawait(inode->i_mapping);
1029 nfsi->flags &= ~NFS_INO_INVALID_DATA;
1030 invalidate_inode_pages2(inode->i_mapping);
1031 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
1032 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1034 (long long)NFS_FILEID(inode));
1035 /* This ensures we revalidate dentries */
1036 nfsi->cache_change_attribute++;
1038 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1040 (long long)NFS_FILEID(inode));
1043 NFS_FLAGS(inode) &= ~NFS_INO_REVALIDATING;
1044 wake_up(&nfsi->nfs_i_wait);
1050 int nfs_attribute_timeout(struct inode *inode)
1052 struct nfs_inode *nfsi = NFS_I(inode);
1054 if (nfs_have_delegation(inode, FMODE_READ))
1056 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1060 * nfs_revalidate_inode - Revalidate the inode attributes
1061 * @server - pointer to nfs_server struct
1062 * @inode - pointer to inode struct
1064 * Updates inode attribute information by retrieving the data from the server.
1066 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1068 if (!(NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1069 && !nfs_attribute_timeout(inode))
1070 return NFS_STALE(inode) ? -ESTALE : 0;
1071 return __nfs_revalidate_inode(server, inode);
1075 * nfs_begin_data_update
1076 * @inode - pointer to inode
1077 * Declare that a set of operations will update file data on the server
1079 void nfs_begin_data_update(struct inode *inode)
1081 atomic_inc(&NFS_I(inode)->data_updates);
1085 * nfs_end_data_update
1086 * @inode - pointer to inode
1087 * Declare end of the operations that will update file data
1088 * This will mark the inode as immediately needing revalidation
1089 * of its attribute cache.
1091 void nfs_end_data_update(struct inode *inode)
1093 struct nfs_inode *nfsi = NFS_I(inode);
1095 if (!nfs_have_delegation(inode, FMODE_READ)) {
1096 /* Mark the attribute cache for revalidation */
1097 nfsi->flags |= NFS_INO_INVALID_ATTR;
1098 /* Directories and symlinks: invalidate page cache too */
1099 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1100 nfsi->flags |= NFS_INO_INVALID_DATA;
1102 nfsi->cache_change_attribute ++;
1103 atomic_dec(&nfsi->data_updates);
1107 * nfs_end_data_update_defer
1108 * @inode - pointer to inode
1109 * Declare end of the operations that will update file data
1110 * This will defer marking the inode as needing revalidation
1111 * unless there are no other pending updates.
1113 void nfs_end_data_update_defer(struct inode *inode)
1115 struct nfs_inode *nfsi = NFS_I(inode);
1117 if (atomic_dec_and_test(&nfsi->data_updates)) {
1118 /* Mark the attribute cache for revalidation */
1119 nfsi->flags |= NFS_INO_INVALID_ATTR;
1120 /* Directories and symlinks: invalidate page cache too */
1121 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1122 nfsi->flags |= NFS_INO_INVALID_DATA;
1123 nfsi->cache_change_attribute ++;
1128 * nfs_refresh_inode - verify consistency of the inode attribute cache
1129 * @inode - pointer to inode
1130 * @fattr - updated attributes
1132 * Verifies the attribute cache. If we have just changed the attributes,
1133 * so that fattr carries weak cache consistency data, then it may
1134 * also update the ctime/mtime/change_attribute.
1136 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1138 struct nfs_inode *nfsi = NFS_I(inode);
1139 loff_t cur_size, new_isize;
1145 /* Do we hold a delegation? */
1146 if (nfs_have_delegation(inode, FMODE_READ))
1149 /* Are we in the process of updating data on the server? */
1150 data_unstable = nfs_caches_unstable(inode);
1152 if (fattr->valid & NFS_ATTR_FATTR_V4) {
1153 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1154 && nfsi->change_attr == fattr->pre_change_attr)
1155 nfsi->change_attr = fattr->change_attr;
1156 if (!data_unstable && nfsi->change_attr != fattr->change_attr)
1157 nfsi->flags |= NFS_INO_INVALID_ATTR;
1160 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1163 /* Has the inode gone and changed behind our back? */
1164 if (nfsi->fileid != fattr->fileid
1165 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1168 cur_size = i_size_read(inode);
1169 new_isize = nfs_size_to_loff_t(fattr->size);
1171 /* If we have atomic WCC data, we may update some attributes */
1172 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1173 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1174 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1175 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1176 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1179 /* Verify a few of the more important attributes */
1180 if (!data_unstable) {
1181 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)
1182 || cur_size != new_isize)
1183 nfsi->flags |= NFS_INO_INVALID_ATTR;
1184 } else if (S_ISREG(inode->i_mode) && new_isize > cur_size)
1185 nfsi->flags |= NFS_INO_INVALID_ATTR;
1187 uid = INOXID_UID(XID_TAG(inode), fattr->uid, fattr->gid);
1188 gid = INOXID_GID(XID_TAG(inode), fattr->uid, fattr->gid);
1189 xid = INOXID_XID(XID_TAG(inode), fattr->uid, fattr->gid, 0);
1191 /* Have any file permissions changed? */
1192 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1193 || inode->i_uid != uid
1194 || inode->i_gid != gid
1195 || inode->i_xid != xid)
1196 nfsi->flags |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1198 /* Has the link count changed? */
1199 if (inode->i_nlink != fattr->nlink)
1200 nfsi->flags |= NFS_INO_INVALID_ATTR;
1202 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1203 nfsi->flags |= NFS_INO_INVALID_ATIME;
1205 nfsi->read_cache_jiffies = fattr->timestamp;
1210 * Many nfs protocol calls return the new file attributes after
1211 * an operation. Here we update the inode to reflect the state
1212 * of the server's inode.
1214 * This is a bit tricky because we have to make sure all dirty pages
1215 * have been sent off to the server before calling invalidate_inode_pages.
1216 * To make sure no other process adds more write requests while we try
1217 * our best to flush them, we make them sleep during the attribute refresh.
1219 * A very similar scenario holds for the dir cache.
1221 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1223 struct nfs_inode *nfsi = NFS_I(inode);
1226 unsigned int invalid = 0;
1233 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1234 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1235 atomic_read(&inode->i_count), fattr->valid);
1237 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1240 if (nfsi->fileid != fattr->fileid) {
1241 printk(KERN_ERR "%s: inode number mismatch\n"
1242 "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1244 inode->i_sb->s_id, (long long)nfsi->fileid,
1245 inode->i_sb->s_id, (long long)fattr->fileid);
1250 * Make sure the inode's type hasn't changed.
1252 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1256 * Update the read time so we don't revalidate too often.
1258 nfsi->read_cache_jiffies = fattr->timestamp;
1260 /* Are we racing with known updates of the metadata on the server? */
1261 data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1263 /* Check if the file size agrees */
1264 new_size = fattr->size;
1265 new_isize = nfs_size_to_loff_t(fattr->size);
1266 cur_isize = i_size_read(inode);
1267 if (cur_isize != new_size) {
1268 #ifdef NFS_DEBUG_VERBOSE
1269 printk(KERN_DEBUG "NFS: isize change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1272 * If we have pending writebacks, things can get
1275 if (S_ISREG(inode->i_mode) && data_unstable) {
1276 if (new_isize > cur_isize) {
1277 inode->i_size = new_isize;
1278 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1281 inode->i_size = new_isize;
1282 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1287 * Note: we don't check inode->i_mtime since pipes etc.
1288 * can change this value in VFS without requiring a
1289 * cache revalidation.
1291 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1292 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1293 #ifdef NFS_DEBUG_VERBOSE
1294 printk(KERN_DEBUG "NFS: mtime change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1297 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1300 if ((fattr->valid & NFS_ATTR_FATTR_V4)
1301 && nfsi->change_attr != fattr->change_attr) {
1302 #ifdef NFS_DEBUG_VERBOSE
1303 printk(KERN_DEBUG "NFS: change_attr change on %s/%ld\n",
1304 inode->i_sb->s_id, inode->i_ino);
1306 nfsi->change_attr = fattr->change_attr;
1308 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS;
1311 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1312 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1314 uid = INOXID_UID(XID_TAG(inode), fattr->uid, fattr->gid);
1315 gid = INOXID_GID(XID_TAG(inode), fattr->uid, fattr->gid);
1316 xid = INOXID_XID(XID_TAG(inode), fattr->uid, fattr->gid, 0);
1318 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1319 inode->i_uid != uid ||
1320 inode->i_gid != gid ||
1321 inode->i_xid != xid)
1322 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS;
1324 inode->i_mode = fattr->mode;
1325 inode->i_nlink = fattr->nlink;
1330 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1332 * report the blocks in 512byte units
1334 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1335 inode->i_blksize = inode->i_sb->s_blocksize;
1337 inode->i_blocks = fattr->du.nfs2.blocks;
1338 inode->i_blksize = fattr->du.nfs2.blocksize;
1341 /* Update attrtimeo value if we're out of the unstable period */
1342 if (invalid & NFS_INO_INVALID_ATTR) {
1343 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1344 nfsi->attrtimeo_timestamp = jiffies;
1345 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1346 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1347 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1348 nfsi->attrtimeo_timestamp = jiffies;
1350 /* Don't invalidate the data if we were to blame */
1351 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1352 || S_ISLNK(inode->i_mode)))
1353 invalid &= ~NFS_INO_INVALID_DATA;
1354 if (!nfs_have_delegation(inode, FMODE_READ))
1355 nfsi->flags |= invalid;
1360 * Big trouble! The inode has become a different object.
1363 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1364 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1367 * No need to worry about unhashing the dentry, as the
1368 * lookup validation will know that the inode is bad.
1369 * (But we fall through to invalidate the caches.)
1371 nfs_invalidate_inode(inode);
1373 NFS_FLAGS(inode) |= NFS_INO_STALE;
1378 * File system information
1381 static int nfs_set_super(struct super_block *s, void *data)
1383 s->s_fs_info = data;
1384 return set_anon_super(s, data);
1387 static int nfs_compare_super(struct super_block *sb, void *data)
1389 struct nfs_server *server = data;
1390 struct nfs_server *old = NFS_SB(sb);
1392 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1394 if (old->addr.sin_port != server->addr.sin_port)
1396 return !nfs_compare_fh(&old->fh, &server->fh);
1399 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1400 int flags, const char *dev_name, void *raw_data)
1403 struct nfs_server *server;
1404 struct super_block *s;
1405 struct nfs_fh *root;
1406 struct nfs_mount_data *data = raw_data;
1409 printk("nfs_read_super: missing data argument\n");
1410 return ERR_PTR(-EINVAL);
1413 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1415 return ERR_PTR(-ENOMEM);
1416 memset(server, 0, sizeof(struct nfs_server));
1417 /* Zero out the NFS state stuff */
1418 init_nfsv4_state(server);
1420 if (data->version != NFS_MOUNT_VERSION) {
1421 printk("nfs warning: mount version %s than kernel\n",
1422 data->version < NFS_MOUNT_VERSION ? "older" : "newer");
1423 if (data->version < 2)
1425 if (data->version < 3)
1427 if (data->version < 4) {
1428 data->flags &= ~NFS_MOUNT_VER3;
1429 data->root.size = NFS2_FHSIZE;
1430 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1432 if (data->version < 5)
1433 data->flags &= ~NFS_MOUNT_SECFLAVOUR;
1437 if (data->flags & NFS_MOUNT_VER3)
1438 root->size = data->root.size;
1440 root->size = NFS2_FHSIZE;
1441 if (root->size > sizeof(root->data)) {
1442 printk("nfs_get_sb: invalid root filehandle\n");
1444 return ERR_PTR(-EINVAL);
1446 memcpy(root->data, data->root.data, root->size);
1448 /* We now require that the mount process passes the remote address */
1449 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1450 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1451 printk("NFS: mount program didn't pass remote address!\n");
1453 return ERR_PTR(-EINVAL);
1456 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1458 if (IS_ERR(s) || s->s_root) {
1465 /* Fire up rpciod if not yet running */
1466 if (rpciod_up() != 0) {
1467 printk(KERN_WARNING "NFS: couldn't start rpciod!\n");
1469 return ERR_PTR(-EIO);
1472 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1474 up_write(&s->s_umount);
1475 deactivate_super(s);
1476 return ERR_PTR(error);
1478 s->s_flags |= MS_ACTIVE;
1482 static void nfs_kill_super(struct super_block *s)
1484 struct nfs_server *server = NFS_SB(s);
1488 if (server->client != NULL && !IS_ERR(server->client))
1489 rpc_shutdown_client(server->client);
1490 if (server->client_sys != NULL && !IS_ERR(server->client_sys))
1491 rpc_shutdown_client(server->client_sys);
1493 if (!(server->flags & NFS_MOUNT_NONLM))
1494 lockd_down(); /* release rpc.lockd */
1496 rpciod_down(); /* release rpciod */
1498 if (server->hostname != NULL)
1499 kfree(server->hostname);
1503 static struct file_system_type nfs_fs_type = {
1504 .owner = THIS_MODULE,
1506 .get_sb = nfs_get_sb,
1507 .kill_sb = nfs_kill_super,
1508 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1511 #ifdef CONFIG_NFS_V4
1513 static void nfs4_clear_inode(struct inode *);
1516 static struct super_operations nfs4_sops = {
1517 .alloc_inode = nfs_alloc_inode,
1518 .destroy_inode = nfs_destroy_inode,
1519 .write_inode = nfs_write_inode,
1520 .delete_inode = nfs_delete_inode,
1521 .statfs = nfs_statfs,
1522 .clear_inode = nfs4_clear_inode,
1523 .umount_begin = nfs_umount_begin,
1524 .show_options = nfs_show_options,
1528 * Clean out any remaining NFSv4 state that might be left over due
1529 * to open() calls that passed nfs_atomic_lookup, but failed to call
1532 static void nfs4_clear_inode(struct inode *inode)
1534 struct nfs_inode *nfsi = NFS_I(inode);
1536 /* If we are holding a delegation, return it! */
1537 if (nfsi->delegation != NULL)
1538 nfs_inode_return_delegation(inode);
1539 /* First call standard NFS clear_inode() code */
1540 nfs_clear_inode(inode);
1541 /* Now clear out any remaining state */
1542 while (!list_empty(&nfsi->open_states)) {
1543 struct nfs4_state *state;
1545 state = list_entry(nfsi->open_states.next,
1548 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1551 (long long)NFS_FILEID(inode),
1553 BUG_ON(atomic_read(&state->count) != 1);
1554 nfs4_close_state(state, state->state);
1559 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1561 struct nfs_server *server;
1562 struct nfs4_client *clp = NULL;
1563 struct rpc_xprt *xprt = NULL;
1564 struct rpc_clnt *clnt = NULL;
1565 struct rpc_timeout timeparms;
1566 rpc_authflavor_t authflavour;
1567 int proto, err = -EIO;
1569 sb->s_blocksize_bits = 0;
1570 sb->s_blocksize = 0;
1571 server = NFS_SB(sb);
1572 if (data->rsize != 0)
1573 server->rsize = nfs_block_size(data->rsize, NULL);
1574 if (data->wsize != 0)
1575 server->wsize = nfs_block_size(data->wsize, NULL);
1576 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1578 server->acregmin = data->acregmin*HZ;
1579 server->acregmax = data->acregmax*HZ;
1580 server->acdirmin = data->acdirmin*HZ;
1581 server->acdirmax = data->acdirmax*HZ;
1583 server->rpc_ops = &nfs_v4_clientops;
1584 /* Initialize timeout values */
1586 timeparms.to_initval = data->timeo * HZ / 10;
1587 timeparms.to_retries = data->retrans;
1588 timeparms.to_exponential = 1;
1589 if (!timeparms.to_retries)
1590 timeparms.to_retries = 5;
1592 proto = data->proto;
1593 /* Which IP protocol do we use? */
1596 timeparms.to_maxval = RPC_MAX_TCP_TIMEOUT;
1597 if (!timeparms.to_initval)
1598 timeparms.to_initval = 600 * HZ / 10;
1601 timeparms.to_maxval = RPC_MAX_UDP_TIMEOUT;
1602 if (!timeparms.to_initval)
1603 timeparms.to_initval = 11 * HZ / 10;
1609 clp = nfs4_get_client(&server->addr.sin_addr);
1611 printk(KERN_WARNING "NFS: failed to create NFS4 client.\n");
1615 /* Now create transport and client */
1616 authflavour = RPC_AUTH_UNIX;
1617 if (data->auth_flavourlen != 0) {
1618 if (data->auth_flavourlen > 1)
1619 printk(KERN_INFO "NFS: cannot yet deal with multiple auth flavours.\n");
1620 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1626 down_write(&clp->cl_sem);
1627 if (clp->cl_rpcclient == NULL) {
1628 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1630 up_write(&clp->cl_sem);
1631 printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
1632 err = PTR_ERR(xprt);
1635 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1636 server->rpc_ops->version, authflavour);
1638 up_write(&clp->cl_sem);
1639 printk(KERN_WARNING "NFS: cannot create RPC client.\n");
1641 err = PTR_ERR(clnt);
1644 clnt->cl_chatty = 1;
1645 clp->cl_rpcclient = clnt;
1646 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1647 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1650 if (list_empty(&clp->cl_superblocks)) {
1651 err = nfs4_init_client(clp);
1653 up_write(&clp->cl_sem);
1657 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1658 clnt = rpc_clone_client(clp->cl_rpcclient);
1660 server->nfs4_state = clp;
1661 up_write(&clp->cl_sem);
1665 printk(KERN_WARNING "NFS: cannot create RPC client.\n");
1666 return PTR_ERR(clnt);
1669 clnt->cl_intr = (server->flags & NFS4_MOUNT_INTR) ? 1 : 0;
1670 clnt->cl_softrtry = (server->flags & NFS4_MOUNT_SOFT) ? 1 : 0;
1671 server->client = clnt;
1673 if (server->nfs4_state->cl_idmap == NULL) {
1674 printk(KERN_WARNING "NFS: failed to create idmapper.\n");
1678 if (clnt->cl_auth->au_flavor != authflavour) {
1679 if (rpcauth_create(authflavour, clnt) == NULL) {
1680 printk(KERN_WARNING "NFS: couldn't create credcache!\n");
1685 sb->s_time_gran = 1;
1687 sb->s_op = &nfs4_sops;
1688 err = nfs_sb_init(sb, authflavour);
1693 nfs4_put_client(clp);
1697 static int nfs4_compare_super(struct super_block *sb, void *data)
1699 struct nfs_server *server = data;
1700 struct nfs_server *old = NFS_SB(sb);
1702 if (strcmp(server->hostname, old->hostname) != 0)
1704 if (strcmp(server->mnt_path, old->mnt_path) != 0)
1710 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1715 return ERR_PTR(-EINVAL);
1716 if (src->len < maxlen)
1719 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1721 return ERR_PTR(-ENOMEM);
1723 if (copy_from_user(dst, src->data, maxlen)) {
1726 return ERR_PTR(-EFAULT);
1732 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1733 int flags, const char *dev_name, void *raw_data)
1736 struct nfs_server *server;
1737 struct super_block *s;
1738 struct nfs4_mount_data *data = raw_data;
1742 printk("nfs_read_super: missing data argument\n");
1743 return ERR_PTR(-EINVAL);
1746 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1748 return ERR_PTR(-ENOMEM);
1749 memset(server, 0, sizeof(struct nfs_server));
1750 /* Zero out the NFS state stuff */
1751 init_nfsv4_state(server);
1753 if (data->version != NFS4_MOUNT_VERSION) {
1754 printk("nfs warning: mount version %s than kernel\n",
1755 data->version < NFS4_MOUNT_VERSION ? "older" : "newer");
1758 p = nfs_copy_user_string(NULL, &data->hostname, 256);
1761 server->hostname = p;
1763 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1766 server->mnt_path = p;
1768 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1769 sizeof(server->ip_addr) - 1);
1773 /* We now require that the mount process passes the remote address */
1774 if (data->host_addrlen != sizeof(server->addr)) {
1775 s = ERR_PTR(-EINVAL);
1778 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1779 s = ERR_PTR(-EFAULT);
1782 if (server->addr.sin_family != AF_INET ||
1783 server->addr.sin_addr.s_addr == INADDR_ANY) {
1784 printk("NFS: mount program didn't pass remote IP address!\n");
1785 s = ERR_PTR(-EINVAL);
1789 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1791 if (IS_ERR(s) || s->s_root)
1796 /* Fire up rpciod if not yet running */
1797 if (rpciod_up() != 0) {
1798 printk(KERN_WARNING "NFS: couldn't start rpciod!\n");
1803 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1805 up_write(&s->s_umount);
1806 deactivate_super(s);
1807 return ERR_PTR(error);
1809 s->s_flags |= MS_ACTIVE;
1812 s = (struct super_block *)p;
1814 if (server->mnt_path)
1815 kfree(server->mnt_path);
1816 if (server->hostname)
1817 kfree(server->hostname);
1822 static void nfs4_kill_super(struct super_block *sb)
1824 struct nfs_server *server = NFS_SB(sb);
1826 nfs_return_all_delegations(sb);
1827 kill_anon_super(sb);
1829 nfs4_renewd_prepare_shutdown(server);
1831 if (server->client != NULL && !IS_ERR(server->client))
1832 rpc_shutdown_client(server->client);
1833 rpciod_down(); /* release rpciod */
1835 destroy_nfsv4_state(server);
1837 if (server->hostname != NULL)
1838 kfree(server->hostname);
1842 static struct file_system_type nfs4_fs_type = {
1843 .owner = THIS_MODULE,
1845 .get_sb = nfs4_get_sb,
1846 .kill_sb = nfs4_kill_super,
1847 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1850 #define nfs4_init_once(nfsi) \
1852 INIT_LIST_HEAD(&(nfsi)->open_states); \
1853 nfsi->delegation = NULL; \
1854 nfsi->delegation_state = 0; \
1855 init_rwsem(&nfsi->rwsem); \
1857 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1858 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1860 #define nfs4_init_once(nfsi) \
1862 #define register_nfs4fs() (0)
1863 #define unregister_nfs4fs()
1866 extern int nfs_init_nfspagecache(void);
1867 extern void nfs_destroy_nfspagecache(void);
1868 extern int nfs_init_readpagecache(void);
1869 extern void nfs_destroy_readpagecache(void);
1870 extern int nfs_init_writepagecache(void);
1871 extern void nfs_destroy_writepagecache(void);
1872 #ifdef CONFIG_NFS_DIRECTIO
1873 extern int nfs_init_directcache(void);
1874 extern void nfs_destroy_directcache(void);
1877 static kmem_cache_t * nfs_inode_cachep;
1879 static struct inode *nfs_alloc_inode(struct super_block *sb)
1881 struct nfs_inode *nfsi;
1882 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
1886 return &nfsi->vfs_inode;
1889 static void nfs_destroy_inode(struct inode *inode)
1891 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1894 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
1896 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1898 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1899 SLAB_CTOR_CONSTRUCTOR) {
1900 inode_init_once(&nfsi->vfs_inode);
1901 spin_lock_init(&nfsi->req_lock);
1902 INIT_LIST_HEAD(&nfsi->dirty);
1903 INIT_LIST_HEAD(&nfsi->commit);
1904 INIT_LIST_HEAD(&nfsi->open_files);
1905 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1906 atomic_set(&nfsi->data_updates, 0);
1910 init_waitqueue_head(&nfsi->nfs_i_wait);
1911 nfs4_init_once(nfsi);
1915 int nfs_init_inodecache(void)
1917 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1918 sizeof(struct nfs_inode),
1919 0, SLAB_RECLAIM_ACCOUNT,
1921 if (nfs_inode_cachep == NULL)
1927 void nfs_destroy_inodecache(void)
1929 if (kmem_cache_destroy(nfs_inode_cachep))
1930 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
1936 static int __init init_nfs_fs(void)
1940 err = nfs_init_nfspagecache();
1944 err = nfs_init_inodecache();
1948 err = nfs_init_readpagecache();
1952 err = nfs_init_writepagecache();
1956 #ifdef CONFIG_NFS_DIRECTIO
1957 err = nfs_init_directcache();
1962 #ifdef CONFIG_PROC_FS
1963 rpc_proc_register(&nfs_rpcstat);
1965 err = register_filesystem(&nfs_fs_type);
1968 if ((err = register_nfs4fs()) != 0)
1972 #ifdef CONFIG_PROC_FS
1973 rpc_proc_unregister("nfs");
1975 nfs_destroy_writepagecache();
1976 #ifdef CONFIG_NFS_DIRECTIO
1978 nfs_destroy_directcache();
1981 nfs_destroy_readpagecache();
1983 nfs_destroy_inodecache();
1985 nfs_destroy_nfspagecache();
1990 static void __exit exit_nfs_fs(void)
1992 #ifdef CONFIG_NFS_DIRECTIO
1993 nfs_destroy_directcache();
1995 nfs_destroy_writepagecache();
1996 nfs_destroy_readpagecache();
1997 nfs_destroy_inodecache();
1998 nfs_destroy_nfspagecache();
1999 #ifdef CONFIG_PROC_FS
2000 rpc_proc_unregister("nfs");
2002 unregister_filesystem(&nfs_fs_type);
2003 unregister_nfs4fs();
2006 /* Not quite true; I just maintain it */
2007 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2008 MODULE_LICENSE("GPL");
2010 module_init(init_nfs_fs)
2011 module_exit(exit_nfs_fs)