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 rpc_program nfs_program;
70 static struct super_operations nfs_sops = {
71 .alloc_inode = nfs_alloc_inode,
72 .destroy_inode = nfs_destroy_inode,
73 .write_inode = nfs_write_inode,
74 .delete_inode = nfs_delete_inode,
76 .clear_inode = nfs_clear_inode,
77 .umount_begin = nfs_umount_begin,
78 .show_options = nfs_show_options,
84 static struct rpc_stat nfs_rpcstat = {
85 .program = &nfs_program
87 static struct rpc_version * nfs_version[] = {
91 #if defined(CONFIG_NFS_V3)
93 #elif defined(CONFIG_NFS_V4)
96 #if defined(CONFIG_NFS_V4)
101 static struct rpc_program nfs_program = {
103 .number = NFS_PROGRAM,
104 .nrvers = sizeof(nfs_version) / sizeof(nfs_version[0]),
105 .version = nfs_version,
106 .stats = &nfs_rpcstat,
107 .pipe_dir_name = "/nfs",
110 static inline unsigned long
111 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
113 return nfs_fileid_to_ino_t(fattr->fileid);
117 nfs_write_inode(struct inode *inode, int sync)
119 int flags = sync ? FLUSH_WAIT : 0;
122 ret = nfs_commit_inode(inode, 0, 0, flags);
129 nfs_delete_inode(struct inode * inode)
131 dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
135 * The following should never happen...
137 if (nfs_have_writebacks(inode)) {
138 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
145 * For the moment, the only task for the NFS clear_inode method is to
146 * release the mmap credential
149 nfs_clear_inode(struct inode *inode)
151 struct nfs_inode *nfsi = NFS_I(inode);
152 struct rpc_cred *cred;
155 BUG_ON (!list_empty(&nfsi->open_files));
156 cred = nfsi->cache_access.cred;
159 BUG_ON(atomic_read(&nfsi->data_updates) != 0);
163 nfs_umount_begin(struct super_block *sb)
165 struct nfs_server *server = NFS_SB(sb);
166 struct rpc_clnt *rpc;
168 /* -EIO all pending I/O */
169 if ((rpc = server->client) != NULL)
170 rpc_killall_tasks(rpc);
174 static inline unsigned long
175 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
177 /* make sure blocksize is a power of two */
178 if ((bsize & (bsize - 1)) || nrbitsp) {
179 unsigned char nrbits;
181 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
192 * Calculate the number of 512byte blocks used.
194 static inline unsigned long
195 nfs_calc_block_size(u64 tsize)
197 loff_t used = (tsize + 511) >> 9;
198 return (used > ULONG_MAX) ? ULONG_MAX : used;
202 * Compute and set NFS server blocksize
204 static inline unsigned long
205 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
208 bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
209 else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
210 bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
212 return nfs_block_bits(bsize, nrbitsp);
216 * Obtain the root inode of the file system.
218 static struct inode *
219 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
221 struct nfs_server *server = NFS_SB(sb);
225 error = server->rpc_ops->getroot(server, rootfh, fsinfo);
227 dprintk("nfs_get_root: getattr error = %d\n", -error);
228 return ERR_PTR(error);
231 rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
233 return ERR_PTR(-ENOMEM);
238 * Do NFS version-independent mount processing, and sanity checking
241 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
243 struct nfs_server *server;
244 struct inode *root_inode;
245 struct nfs_fattr fattr;
246 struct nfs_fsinfo fsinfo = {
249 struct nfs_pathconf pathinfo = {
252 int no_root_error = 0;
253 unsigned long max_rpc_payload;
255 /* We probably want something more informative here */
256 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
260 sb->s_magic = NFS_SUPER_MAGIC;
262 root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
263 /* Did getting the root inode fail? */
264 if (IS_ERR(root_inode)) {
265 no_root_error = PTR_ERR(root_inode);
268 sb->s_root = d_alloc_root(root_inode);
270 no_root_error = -ENOMEM;
273 sb->s_root->d_op = server->rpc_ops->dentry_ops;
275 /* Get some general file system info */
276 if (server->namelen == 0 &&
277 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
278 server->namelen = pathinfo.max_namelen;
279 /* Work out a lot of parameters */
280 if (server->rsize == 0)
281 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
282 if (server->wsize == 0)
283 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
285 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
286 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
287 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
288 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
290 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
291 if (server->rsize > max_rpc_payload)
292 server->rsize = max_rpc_payload;
293 if (server->wsize > max_rpc_payload)
294 server->wsize = max_rpc_payload;
296 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
297 if (server->rpages > NFS_READ_MAXIOV) {
298 server->rpages = NFS_READ_MAXIOV;
299 server->rsize = server->rpages << PAGE_CACHE_SHIFT;
302 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
303 if (server->wpages > NFS_WRITE_MAXIOV) {
304 server->wpages = NFS_WRITE_MAXIOV;
305 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
308 if (sb->s_blocksize == 0)
309 sb->s_blocksize = nfs_block_bits(server->wsize,
310 &sb->s_blocksize_bits);
311 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
313 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
314 if (server->dtsize > PAGE_CACHE_SIZE)
315 server->dtsize = PAGE_CACHE_SIZE;
316 if (server->dtsize > server->rsize)
317 server->dtsize = server->rsize;
319 if (server->flags & NFS_MOUNT_NOAC) {
320 server->acregmin = server->acregmax = 0;
321 server->acdirmin = server->acdirmax = 0;
322 sb->s_flags |= MS_SYNCHRONOUS;
324 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
326 if (server->flags & NFS_MOUNT_TAGXID)
327 sb->s_flags |= MS_TAGXID;
329 sb->s_maxbytes = fsinfo.maxfilesize;
330 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
331 sb->s_maxbytes = MAX_LFS_FILESIZE;
333 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
334 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
335 server->client->cl_tagxid = (server->flags & NFS_MOUNT_TAGXID) ? 1 : 0;
337 /* We're airborne Set socket buffersize */
338 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
340 /* Yargs. It didn't work out. */
342 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
343 if (!IS_ERR(root_inode))
345 return no_root_error;
349 * Create an RPC client handle.
351 static struct rpc_clnt *
352 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
354 struct rpc_timeout timeparms;
355 struct rpc_xprt *xprt = NULL;
356 struct rpc_clnt *clnt = NULL;
357 int tcp = (data->flags & NFS_MOUNT_TCP);
359 /* Initialize timeout values */
360 timeparms.to_initval = data->timeo * HZ / 10;
361 timeparms.to_retries = data->retrans;
362 timeparms.to_maxval = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
363 timeparms.to_exponential = 1;
365 if (!timeparms.to_initval)
366 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
367 if (!timeparms.to_retries)
368 timeparms.to_retries = 5;
370 /* create transport and client */
371 xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
372 &server->addr, &timeparms);
374 printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
375 return (struct rpc_clnt *)xprt;
377 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
378 server->rpc_ops->version, data->pseudoflavor);
380 printk(KERN_WARNING "NFS: cannot create RPC client.\n");
385 clnt->cl_softrtry = 1;
397 * The way this works is that the mount process passes a structure
398 * in the data argument which contains the server's IP address
399 * and the root file handle obtained from the server's mount
400 * daemon. We stash these away in the private superblock fields.
403 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
405 struct nfs_server *server;
406 rpc_authflavor_t authflavor;
409 sb->s_blocksize_bits = 0;
412 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
414 server->rsize = nfs_block_size(data->rsize, NULL);
416 server->wsize = nfs_block_size(data->wsize, NULL);
417 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
419 server->acregmin = data->acregmin*HZ;
420 server->acregmax = data->acregmax*HZ;
421 server->acdirmin = data->acdirmin*HZ;
422 server->acdirmax = data->acdirmax*HZ;
424 /* Start lockd here, before we might error out */
425 if (!(server->flags & NFS_MOUNT_NONLM))
428 server->namelen = data->namlen;
429 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
430 if (!server->hostname)
432 strcpy(server->hostname, data->hostname);
434 /* Check NFS protocol revision and initialize RPC op vector
435 * and file handle pool. */
436 if (server->flags & NFS_MOUNT_VER3) {
438 server->rpc_ops = &nfs_v3_clientops;
439 server->caps |= NFS_CAP_READDIRPLUS;
440 if (data->version < 4) {
441 printk(KERN_NOTICE "NFS: NFSv3 not supported by mount program.\n");
445 printk(KERN_NOTICE "NFS: NFSv3 not supported.\n");
449 server->rpc_ops = &nfs_v2_clientops;
452 /* Fill in pseudoflavor for mount version < 5 */
453 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
454 data->pseudoflavor = RPC_AUTH_UNIX;
455 authflavor = data->pseudoflavor; /* save for sb_init() */
456 /* XXX maybe we want to add a server->pseudoflavor field */
458 /* Create RPC client handles */
459 server->client = nfs_create_client(server, data);
460 if (IS_ERR(server->client))
461 return PTR_ERR(server->client);
462 /* RFC 2623, sec 2.3.2 */
463 if (authflavor != RPC_AUTH_UNIX) {
464 server->client_sys = rpc_clone_client(server->client);
465 if (IS_ERR(server->client_sys))
466 return PTR_ERR(server->client_sys);
467 if (!rpcauth_create(RPC_AUTH_UNIX, server->client_sys))
470 atomic_inc(&server->client->cl_count);
471 server->client_sys = server->client;
474 if (server->flags & NFS_MOUNT_VER3) {
475 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
476 server->namelen = NFS3_MAXNAMLEN;
479 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
480 server->namelen = NFS2_MAXNAMLEN;
483 sb->s_op = &nfs_sops;
484 return nfs_sb_init(sb, authflavor);
488 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
490 struct nfs_server *server = NFS_SB(sb);
491 unsigned char blockbits;
492 unsigned long blockres;
493 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
494 struct nfs_fattr fattr;
495 struct nfs_fsstat res = {
502 error = server->rpc_ops->statfs(server, rootfh, &res);
503 buf->f_type = NFS_SUPER_MAGIC;
508 * Current versions of glibc do not correctly handle the
509 * case where f_frsize != f_bsize. Eventually we want to
510 * report the value of wtmult in this field.
512 buf->f_frsize = sb->s_blocksize;
515 * On most *nix systems, f_blocks, f_bfree, and f_bavail
516 * are reported in units of f_frsize. Linux hasn't had
517 * an f_frsize field in its statfs struct until recently,
518 * thus historically Linux's sys_statfs reports these
519 * fields in units of f_bsize.
521 buf->f_bsize = sb->s_blocksize;
522 blockbits = sb->s_blocksize_bits;
523 blockres = (1 << blockbits) - 1;
524 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
525 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
526 buf->f_bavail = (res.abytes + blockres) >> blockbits;
528 buf->f_files = res.tfiles;
529 buf->f_ffree = res.afiles;
531 buf->f_namelen = server->namelen;
538 printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
539 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
544 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
546 static struct proc_nfs_info {
551 { NFS_MOUNT_SOFT, ",soft", ",hard" },
552 { NFS_MOUNT_INTR, ",intr", "" },
553 { NFS_MOUNT_POSIX, ",posix", "" },
554 { NFS_MOUNT_TCP, ",tcp", ",udp" },
555 { NFS_MOUNT_NOCTO, ",nocto", "" },
556 { NFS_MOUNT_NOAC, ",noac", "" },
557 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
558 { NFS_MOUNT_TAGXID, ",tagxid", "" },
561 struct proc_nfs_info *nfs_infop;
562 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
564 seq_printf(m, ",v%d", nfss->rpc_ops->version);
565 seq_printf(m, ",rsize=%d", nfss->rsize);
566 seq_printf(m, ",wsize=%d", nfss->wsize);
567 if (nfss->acregmin != 3*HZ)
568 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
569 if (nfss->acregmax != 60*HZ)
570 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
571 if (nfss->acdirmin != 30*HZ)
572 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
573 if (nfss->acdirmax != 60*HZ)
574 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
575 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
576 if (nfss->flags & nfs_infop->flag)
577 seq_puts(m, nfs_infop->str);
579 seq_puts(m, nfs_infop->nostr);
581 seq_puts(m, ",addr=");
582 seq_escape(m, nfss->hostname, " \t\n\\");
587 * Invalidate the local caches
590 nfs_zap_caches(struct inode *inode)
592 struct nfs_inode *nfsi = NFS_I(inode);
593 int mode = inode->i_mode;
595 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
596 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
598 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
599 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
600 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS;
602 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS;
606 * Invalidate, but do not unhash, the inode
609 nfs_invalidate_inode(struct inode *inode)
611 umode_t save_mode = inode->i_mode;
613 make_bad_inode(inode);
614 inode->i_mode = save_mode;
615 nfs_zap_caches(inode);
618 struct nfs_find_desc {
620 struct nfs_fattr *fattr;
624 * In NFSv3 we can have 64bit inode numbers. In order to support
625 * this, and re-exported directories (also seen in NFSv2)
626 * we are forced to allow 2 different inodes to have the same
630 nfs_find_actor(struct inode *inode, void *opaque)
632 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
633 struct nfs_fh *fh = desc->fh;
634 struct nfs_fattr *fattr = desc->fattr;
636 if (NFS_FILEID(inode) != fattr->fileid)
638 if (nfs_compare_fh(NFS_FH(inode), fh))
640 if (is_bad_inode(inode) || NFS_STALE(inode))
646 nfs_init_locked(struct inode *inode, void *opaque)
648 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
649 struct nfs_fattr *fattr = desc->fattr;
651 NFS_FILEID(inode) = fattr->fileid;
652 nfs_copy_fh(NFS_FH(inode), desc->fh);
656 /* Don't use READDIRPLUS on directories that we believe are too large */
657 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
660 * This is our front-end to iget that looks up inodes by file handle
661 * instead of inode number.
664 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
666 struct nfs_find_desc desc = {
670 struct inode *inode = NULL;
673 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
677 printk("NFS: Buggy server - nlink == 0!\n");
681 hash = nfs_fattr_to_ino_t(fattr);
683 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
686 if (inode->i_state & I_NEW) {
687 struct nfs_inode *nfsi = NFS_I(inode);
689 /* We set i_ino for the few things that still rely on it,
693 /* We can't support update_atime(), since the server will reset it */
694 inode->i_flags |= S_NOATIME|S_NOCMTIME;
695 inode->i_mode = fattr->mode;
696 /* Why so? Because we want revalidate for devices/FIFOs, and
697 * that's precisely what we have in nfs_file_inode_operations.
699 inode->i_op = &nfs_file_inode_operations;
700 if (S_ISREG(inode->i_mode)) {
701 inode->i_fop = &nfs_file_operations;
702 inode->i_data.a_ops = &nfs_file_aops;
703 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
704 } else if (S_ISDIR(inode->i_mode)) {
705 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
706 inode->i_fop = &nfs_dir_operations;
707 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
708 && fattr->size <= NFS_LIMIT_READDIRPLUS)
709 NFS_FLAGS(inode) |= NFS_INO_ADVISE_RDPLUS;
710 } else if (S_ISLNK(inode->i_mode))
711 inode->i_op = &nfs_symlink_inode_operations;
713 init_special_inode(inode, inode->i_mode, fattr->rdev);
715 nfsi->read_cache_jiffies = fattr->timestamp;
716 inode->i_atime = fattr->atime;
717 inode->i_mtime = fattr->mtime;
718 inode->i_ctime = fattr->ctime;
719 if (fattr->valid & NFS_ATTR_FATTR_V4)
720 nfsi->change_attr = fattr->change_attr;
721 inode->i_size = nfs_size_to_loff_t(fattr->size);
722 inode->i_nlink = fattr->nlink;
723 inode->i_uid = INOXID_UID(XID_TAG(inode), fattr->uid, fattr->gid);
724 inode->i_gid = INOXID_GID(XID_TAG(inode), fattr->uid, fattr->gid);
725 inode->i_xid = INOXID_XID(XID_TAG(inode), fattr->uid, fattr->gid, 0);
726 /* maybe fattr->xid someday */
727 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
729 * report the blocks in 512byte units
731 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
732 inode->i_blksize = inode->i_sb->s_blocksize;
734 inode->i_blocks = fattr->du.nfs2.blocks;
735 inode->i_blksize = fattr->du.nfs2.blocksize;
737 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
738 nfsi->attrtimeo_timestamp = jiffies;
739 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
740 nfsi->cache_access.cred = NULL;
742 unlock_new_inode(inode);
744 nfs_refresh_inode(inode, fattr);
745 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
747 (long long)NFS_FILEID(inode),
748 atomic_read(&inode->i_count));
754 printk("nfs_fhget: iget failed\n");
758 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
761 nfs_setattr(struct dentry *dentry, struct iattr *attr)
763 struct inode *inode = dentry->d_inode;
764 struct nfs_fattr fattr;
767 if (attr->ia_valid & ATTR_SIZE) {
768 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
769 attr->ia_valid &= ~ATTR_SIZE;
772 /* Optimization: if the end result is no change, don't RPC */
773 attr->ia_valid &= NFS_VALID_ATTRS;
774 if (attr->ia_valid == 0)
778 nfs_begin_data_update(inode);
779 /* Write all dirty data if we're changing file permissions or size */
780 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
781 if (filemap_fdatawrite(inode->i_mapping) == 0)
782 filemap_fdatawait(inode->i_mapping);
785 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
787 nfs_refresh_inode(inode, &fattr);
788 if ((attr->ia_valid & ATTR_MODE) != 0) {
790 mode = inode->i_mode & ~S_IALLUGO;
791 mode |= attr->ia_mode & S_IALLUGO;
792 inode->i_mode = mode;
794 if ((attr->ia_valid & ATTR_UID) != 0)
795 inode->i_uid = attr->ia_uid;
796 if ((attr->ia_valid & ATTR_GID) != 0)
797 inode->i_gid = attr->ia_gid;
798 if ((attr->ia_valid & ATTR_XID) != 0)
799 inode->i_xid = attr->ia_xid;
800 if ((attr->ia_valid & ATTR_SIZE) != 0) {
801 inode->i_size = attr->ia_size;
802 vmtruncate(inode, attr->ia_size);
805 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
806 NFS_FLAGS(inode) |= NFS_INO_INVALID_ACCESS;
807 nfs_end_data_update(inode);
813 * Wait for the inode to get unlocked.
814 * (Used for NFS_INO_LOCKED and NFS_INO_REVALIDATING).
817 nfs_wait_on_inode(struct inode *inode, int flag)
819 struct rpc_clnt *clnt = NFS_CLIENT(inode);
820 struct nfs_inode *nfsi = NFS_I(inode);
823 if (!(NFS_FLAGS(inode) & flag))
825 atomic_inc(&inode->i_count);
826 error = nfs_wait_event(clnt, nfsi->nfs_i_wait,
827 !(NFS_FLAGS(inode) & flag));
832 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
834 struct inode *inode = dentry->d_inode;
835 struct nfs_inode *nfsi = NFS_I(inode);
836 int need_atime = nfsi->flags & NFS_INO_INVALID_ATIME;
839 if (__IS_FLG(inode, MS_NOATIME))
841 else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
843 /* We may force a getattr if the user cares about atime */
845 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
847 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
849 generic_fillattr(inode, stat);
853 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
855 struct nfs_open_context *ctx;
857 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
859 atomic_set(&ctx->count, 1);
860 ctx->dentry = dget(dentry);
861 ctx->cred = get_rpccred(cred);
863 ctx->lockowner = current->files;
865 init_waitqueue_head(&ctx->waitq);
870 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
873 atomic_inc(&ctx->count);
877 void put_nfs_open_context(struct nfs_open_context *ctx)
879 if (atomic_dec_and_test(&ctx->count)) {
880 if (!list_empty(&ctx->list)) {
881 struct inode *inode = ctx->dentry->d_inode;
882 spin_lock(&inode->i_lock);
883 list_del(&ctx->list);
884 spin_unlock(&inode->i_lock);
886 if (ctx->state != NULL)
887 nfs4_close_state(ctx->state, ctx->mode);
888 if (ctx->cred != NULL)
889 put_rpccred(ctx->cred);
896 * Ensure that mmap has a recent RPC credential for use when writing out
899 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
901 struct inode *inode = filp->f_dentry->d_inode;
902 struct nfs_inode *nfsi = NFS_I(inode);
904 filp->private_data = get_nfs_open_context(ctx);
905 spin_lock(&inode->i_lock);
906 list_add(&ctx->list, &nfsi->open_files);
907 spin_unlock(&inode->i_lock);
910 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
912 struct nfs_inode *nfsi = NFS_I(inode);
913 struct nfs_open_context *pos, *ctx = NULL;
915 spin_lock(&inode->i_lock);
916 list_for_each_entry(pos, &nfsi->open_files, list) {
917 if ((pos->mode & mode) == mode) {
918 ctx = get_nfs_open_context(pos);
922 spin_unlock(&inode->i_lock);
926 void nfs_file_clear_open_context(struct file *filp)
928 struct inode *inode = filp->f_dentry->d_inode;
929 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
932 filp->private_data = NULL;
933 spin_lock(&inode->i_lock);
934 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
935 spin_unlock(&inode->i_lock);
936 put_nfs_open_context(ctx);
941 * These allocate and release file read/write context information.
943 int nfs_open(struct inode *inode, struct file *filp)
945 struct nfs_open_context *ctx;
946 struct rpc_cred *cred;
948 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
950 return PTR_ERR(cred);
951 ctx = alloc_nfs_open_context(filp->f_dentry, cred);
955 ctx->mode = filp->f_mode;
956 nfs_file_set_open_context(filp, ctx);
957 put_nfs_open_context(ctx);
958 if ((filp->f_mode & FMODE_WRITE) != 0)
959 nfs_begin_data_update(inode);
963 int nfs_release(struct inode *inode, struct file *filp)
965 if ((filp->f_mode & FMODE_WRITE) != 0)
966 nfs_end_data_update(inode);
967 nfs_file_clear_open_context(filp);
972 * This function is called whenever some part of NFS notices that
973 * the cached attributes have to be refreshed.
976 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
978 int status = -ESTALE;
979 struct nfs_fattr fattr;
980 struct nfs_inode *nfsi = NFS_I(inode);
981 unsigned long verifier;
984 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
985 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
988 if (!inode || is_bad_inode(inode))
990 if (NFS_STALE(inode))
993 while (NFS_REVALIDATING(inode)) {
994 status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
997 if (NFS_ATTRTIMEO(inode) == 0)
999 if (NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME))
1001 status = NFS_STALE(inode) ? -ESTALE : 0;
1004 NFS_FLAGS(inode) |= NFS_INO_REVALIDATING;
1006 /* Protect against RPC races by saving the change attribute */
1007 verifier = nfs_save_change_attribute(inode);
1008 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1010 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1012 (long long)NFS_FILEID(inode), status);
1013 if (status == -ESTALE) {
1014 nfs_zap_caches(inode);
1015 if (!S_ISDIR(inode->i_mode))
1016 NFS_FLAGS(inode) |= NFS_INO_STALE;
1021 status = nfs_update_inode(inode, &fattr, verifier);
1023 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1025 (long long)NFS_FILEID(inode), status);
1028 flags = nfsi->flags;
1030 * We may need to keep the attributes marked as invalid if
1031 * we raced with nfs_end_attr_update().
1033 if (verifier == nfsi->cache_change_attribute)
1034 nfsi->flags &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1035 /* Do the page cache invalidation */
1036 if (flags & NFS_INO_INVALID_DATA) {
1037 if (S_ISREG(inode->i_mode)) {
1038 if (filemap_fdatawrite(inode->i_mapping) == 0)
1039 filemap_fdatawait(inode->i_mapping);
1042 nfsi->flags &= ~NFS_INO_INVALID_DATA;
1043 invalidate_inode_pages2(inode->i_mapping);
1044 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
1045 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1047 (long long)NFS_FILEID(inode));
1048 /* This ensures we revalidate dentries */
1049 nfsi->cache_change_attribute++;
1051 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1053 (long long)NFS_FILEID(inode));
1056 NFS_FLAGS(inode) &= ~NFS_INO_REVALIDATING;
1057 wake_up(&nfsi->nfs_i_wait);
1063 int nfs_attribute_timeout(struct inode *inode)
1065 struct nfs_inode *nfsi = NFS_I(inode);
1067 if (nfs_have_delegation(inode, FMODE_READ))
1069 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1073 * nfs_revalidate_inode - Revalidate the inode attributes
1074 * @server - pointer to nfs_server struct
1075 * @inode - pointer to inode struct
1077 * Updates inode attribute information by retrieving the data from the server.
1079 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1081 if (!(NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1082 && !nfs_attribute_timeout(inode))
1083 return NFS_STALE(inode) ? -ESTALE : 0;
1084 return __nfs_revalidate_inode(server, inode);
1088 * nfs_begin_data_update
1089 * @inode - pointer to inode
1090 * Declare that a set of operations will update file data on the server
1092 void nfs_begin_data_update(struct inode *inode)
1094 atomic_inc(&NFS_I(inode)->data_updates);
1098 * nfs_end_data_update
1099 * @inode - pointer to inode
1100 * Declare end of the operations that will update file data
1101 * This will mark the inode as immediately needing revalidation
1102 * of its attribute cache.
1104 void nfs_end_data_update(struct inode *inode)
1106 struct nfs_inode *nfsi = NFS_I(inode);
1108 if (!nfs_have_delegation(inode, FMODE_READ)) {
1109 /* Mark the attribute cache for revalidation */
1110 nfsi->flags |= NFS_INO_INVALID_ATTR;
1111 /* Directories and symlinks: invalidate page cache too */
1112 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1113 nfsi->flags |= NFS_INO_INVALID_DATA;
1115 nfsi->cache_change_attribute ++;
1116 atomic_dec(&nfsi->data_updates);
1120 * nfs_end_data_update_defer
1121 * @inode - pointer to inode
1122 * Declare end of the operations that will update file data
1123 * This will defer marking the inode as needing revalidation
1124 * unless there are no other pending updates.
1126 void nfs_end_data_update_defer(struct inode *inode)
1128 struct nfs_inode *nfsi = NFS_I(inode);
1130 if (atomic_dec_and_test(&nfsi->data_updates)) {
1131 /* Mark the attribute cache for revalidation */
1132 nfsi->flags |= NFS_INO_INVALID_ATTR;
1133 /* Directories and symlinks: invalidate page cache too */
1134 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1135 nfsi->flags |= NFS_INO_INVALID_DATA;
1136 nfsi->cache_change_attribute ++;
1141 * nfs_refresh_inode - verify consistency of the inode attribute cache
1142 * @inode - pointer to inode
1143 * @fattr - updated attributes
1145 * Verifies the attribute cache. If we have just changed the attributes,
1146 * so that fattr carries weak cache consistency data, then it may
1147 * also update the ctime/mtime/change_attribute.
1149 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1151 struct nfs_inode *nfsi = NFS_I(inode);
1152 loff_t cur_size, new_isize;
1158 /* Do we hold a delegation? */
1159 if (nfs_have_delegation(inode, FMODE_READ))
1162 /* Are we in the process of updating data on the server? */
1163 data_unstable = nfs_caches_unstable(inode);
1165 if (fattr->valid & NFS_ATTR_FATTR_V4) {
1166 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1167 && nfsi->change_attr == fattr->pre_change_attr)
1168 nfsi->change_attr = fattr->change_attr;
1169 if (!data_unstable && nfsi->change_attr != fattr->change_attr)
1170 nfsi->flags |= NFS_INO_INVALID_ATTR;
1173 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1176 /* Has the inode gone and changed behind our back? */
1177 if (nfsi->fileid != fattr->fileid
1178 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1181 cur_size = i_size_read(inode);
1182 new_isize = nfs_size_to_loff_t(fattr->size);
1184 /* If we have atomic WCC data, we may update some attributes */
1185 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1186 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1187 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1188 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1189 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1192 /* Verify a few of the more important attributes */
1193 if (!data_unstable) {
1194 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)
1195 || cur_size != new_isize)
1196 nfsi->flags |= NFS_INO_INVALID_ATTR;
1197 } else if (S_ISREG(inode->i_mode) && new_isize > cur_size)
1198 nfsi->flags |= NFS_INO_INVALID_ATTR;
1200 uid = INOXID_UID(XID_TAG(inode), fattr->uid, fattr->gid);
1201 gid = INOXID_GID(XID_TAG(inode), fattr->uid, fattr->gid);
1202 xid = INOXID_XID(XID_TAG(inode), fattr->uid, fattr->gid, 0);
1204 /* Have any file permissions changed? */
1205 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1206 || inode->i_uid != uid
1207 || inode->i_gid != gid
1208 || inode->i_xid != xid)
1209 nfsi->flags |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1211 /* Has the link count changed? */
1212 if (inode->i_nlink != fattr->nlink)
1213 nfsi->flags |= NFS_INO_INVALID_ATTR;
1215 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1216 nfsi->flags |= NFS_INO_INVALID_ATIME;
1218 nfsi->read_cache_jiffies = fattr->timestamp;
1223 * Many nfs protocol calls return the new file attributes after
1224 * an operation. Here we update the inode to reflect the state
1225 * of the server's inode.
1227 * This is a bit tricky because we have to make sure all dirty pages
1228 * have been sent off to the server before calling invalidate_inode_pages.
1229 * To make sure no other process adds more write requests while we try
1230 * our best to flush them, we make them sleep during the attribute refresh.
1232 * A very similar scenario holds for the dir cache.
1234 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1236 struct nfs_inode *nfsi = NFS_I(inode);
1239 unsigned int invalid = 0;
1246 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1247 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1248 atomic_read(&inode->i_count), fattr->valid);
1250 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1253 if (nfsi->fileid != fattr->fileid) {
1254 printk(KERN_ERR "%s: inode number mismatch\n"
1255 "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1257 inode->i_sb->s_id, (long long)nfsi->fileid,
1258 inode->i_sb->s_id, (long long)fattr->fileid);
1263 * Make sure the inode's type hasn't changed.
1265 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1269 * Update the read time so we don't revalidate too often.
1271 nfsi->read_cache_jiffies = fattr->timestamp;
1273 /* Are we racing with known updates of the metadata on the server? */
1274 data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1276 /* Check if the file size agrees */
1277 new_size = fattr->size;
1278 new_isize = nfs_size_to_loff_t(fattr->size);
1279 cur_isize = i_size_read(inode);
1280 if (cur_isize != new_size) {
1281 #ifdef NFS_DEBUG_VERBOSE
1282 printk(KERN_DEBUG "NFS: isize change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1285 * If we have pending writebacks, things can get
1288 if (S_ISREG(inode->i_mode) && data_unstable) {
1289 if (new_isize > cur_isize) {
1290 inode->i_size = new_isize;
1291 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1294 inode->i_size = new_isize;
1295 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1300 * Note: we don't check inode->i_mtime since pipes etc.
1301 * can change this value in VFS without requiring a
1302 * cache revalidation.
1304 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1305 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1306 #ifdef NFS_DEBUG_VERBOSE
1307 printk(KERN_DEBUG "NFS: mtime change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1310 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1313 if ((fattr->valid & NFS_ATTR_FATTR_V4)
1314 && nfsi->change_attr != fattr->change_attr) {
1315 #ifdef NFS_DEBUG_VERBOSE
1316 printk(KERN_DEBUG "NFS: change_attr change on %s/%ld\n",
1317 inode->i_sb->s_id, inode->i_ino);
1319 nfsi->change_attr = fattr->change_attr;
1321 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS;
1324 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1325 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1327 uid = INOXID_UID(XID_TAG(inode), fattr->uid, fattr->gid);
1328 gid = INOXID_GID(XID_TAG(inode), fattr->uid, fattr->gid);
1329 xid = INOXID_XID(XID_TAG(inode), fattr->uid, fattr->gid, 0);
1331 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1332 inode->i_uid != uid ||
1333 inode->i_gid != gid ||
1334 inode->i_xid != xid)
1335 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS;
1337 inode->i_mode = fattr->mode;
1338 inode->i_nlink = fattr->nlink;
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;
1417 struct super_block *s;
1418 struct nfs_fh *root;
1419 struct nfs_mount_data *data = raw_data;
1422 printk("nfs_read_super: missing data argument\n");
1423 return ERR_PTR(-EINVAL);
1426 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1428 return ERR_PTR(-ENOMEM);
1429 memset(server, 0, sizeof(struct nfs_server));
1430 /* Zero out the NFS state stuff */
1431 init_nfsv4_state(server);
1433 if (data->version != NFS_MOUNT_VERSION) {
1434 printk("nfs warning: mount version %s than kernel\n",
1435 data->version < NFS_MOUNT_VERSION ? "older" : "newer");
1436 if (data->version < 2)
1438 if (data->version < 3)
1440 if (data->version < 4) {
1441 data->flags &= ~NFS_MOUNT_VER3;
1442 data->root.size = NFS2_FHSIZE;
1443 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1445 if (data->version < 5)
1446 data->flags &= ~NFS_MOUNT_SECFLAVOUR;
1450 if (data->flags & NFS_MOUNT_VER3)
1451 root->size = data->root.size;
1453 root->size = NFS2_FHSIZE;
1454 if (root->size > sizeof(root->data)) {
1455 printk("nfs_get_sb: invalid root filehandle\n");
1457 return ERR_PTR(-EINVAL);
1459 memcpy(root->data, data->root.data, root->size);
1461 /* We now require that the mount process passes the remote address */
1462 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1463 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1464 printk("NFS: mount program didn't pass remote address!\n");
1466 return ERR_PTR(-EINVAL);
1469 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1471 if (IS_ERR(s) || s->s_root) {
1478 /* Fire up rpciod if not yet running */
1479 if (rpciod_up() != 0) {
1480 printk(KERN_WARNING "NFS: couldn't start rpciod!\n");
1482 return ERR_PTR(-EIO);
1485 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1487 up_write(&s->s_umount);
1488 deactivate_super(s);
1489 return ERR_PTR(error);
1491 s->s_flags |= MS_ACTIVE;
1495 static void nfs_kill_super(struct super_block *s)
1497 struct nfs_server *server = NFS_SB(s);
1501 if (server->client != NULL && !IS_ERR(server->client))
1502 rpc_shutdown_client(server->client);
1503 if (server->client_sys != NULL && !IS_ERR(server->client_sys))
1504 rpc_shutdown_client(server->client_sys);
1506 if (!(server->flags & NFS_MOUNT_NONLM))
1507 lockd_down(); /* release rpc.lockd */
1509 rpciod_down(); /* release rpciod */
1511 if (server->hostname != NULL)
1512 kfree(server->hostname);
1516 static struct file_system_type nfs_fs_type = {
1517 .owner = THIS_MODULE,
1519 .get_sb = nfs_get_sb,
1520 .kill_sb = nfs_kill_super,
1521 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1524 #ifdef CONFIG_NFS_V4
1526 static void nfs4_clear_inode(struct inode *);
1529 static struct super_operations nfs4_sops = {
1530 .alloc_inode = nfs_alloc_inode,
1531 .destroy_inode = nfs_destroy_inode,
1532 .write_inode = nfs_write_inode,
1533 .delete_inode = nfs_delete_inode,
1534 .statfs = nfs_statfs,
1535 .clear_inode = nfs4_clear_inode,
1536 .umount_begin = nfs_umount_begin,
1537 .show_options = nfs_show_options,
1541 * Clean out any remaining NFSv4 state that might be left over due
1542 * to open() calls that passed nfs_atomic_lookup, but failed to call
1545 static void nfs4_clear_inode(struct inode *inode)
1547 struct nfs_inode *nfsi = NFS_I(inode);
1549 /* If we are holding a delegation, return it! */
1550 if (nfsi->delegation != NULL)
1551 nfs_inode_return_delegation(inode);
1552 /* First call standard NFS clear_inode() code */
1553 nfs_clear_inode(inode);
1554 /* Now clear out any remaining state */
1555 while (!list_empty(&nfsi->open_states)) {
1556 struct nfs4_state *state;
1558 state = list_entry(nfsi->open_states.next,
1561 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1564 (long long)NFS_FILEID(inode),
1566 BUG_ON(atomic_read(&state->count) != 1);
1567 nfs4_close_state(state, state->state);
1572 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1574 struct nfs_server *server;
1575 struct nfs4_client *clp = NULL;
1576 struct rpc_xprt *xprt = NULL;
1577 struct rpc_clnt *clnt = NULL;
1578 struct rpc_timeout timeparms;
1579 rpc_authflavor_t authflavour;
1580 int proto, err = -EIO;
1582 sb->s_blocksize_bits = 0;
1583 sb->s_blocksize = 0;
1584 server = NFS_SB(sb);
1585 if (data->rsize != 0)
1586 server->rsize = nfs_block_size(data->rsize, NULL);
1587 if (data->wsize != 0)
1588 server->wsize = nfs_block_size(data->wsize, NULL);
1589 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1590 server->caps = NFS_CAP_ATOMIC_OPEN;
1592 server->acregmin = data->acregmin*HZ;
1593 server->acregmax = data->acregmax*HZ;
1594 server->acdirmin = data->acdirmin*HZ;
1595 server->acdirmax = data->acdirmax*HZ;
1597 server->rpc_ops = &nfs_v4_clientops;
1598 /* Initialize timeout values */
1600 timeparms.to_initval = data->timeo * HZ / 10;
1601 timeparms.to_retries = data->retrans;
1602 timeparms.to_exponential = 1;
1603 if (!timeparms.to_retries)
1604 timeparms.to_retries = 5;
1606 proto = data->proto;
1607 /* Which IP protocol do we use? */
1610 timeparms.to_maxval = RPC_MAX_TCP_TIMEOUT;
1611 if (!timeparms.to_initval)
1612 timeparms.to_initval = 600 * HZ / 10;
1615 timeparms.to_maxval = RPC_MAX_UDP_TIMEOUT;
1616 if (!timeparms.to_initval)
1617 timeparms.to_initval = 11 * HZ / 10;
1623 clp = nfs4_get_client(&server->addr.sin_addr);
1625 printk(KERN_WARNING "NFS: failed to create NFS4 client.\n");
1629 /* Now create transport and client */
1630 authflavour = RPC_AUTH_UNIX;
1631 if (data->auth_flavourlen != 0) {
1632 if (data->auth_flavourlen > 1)
1633 printk(KERN_INFO "NFS: cannot yet deal with multiple auth flavours.\n");
1634 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1640 down_write(&clp->cl_sem);
1641 if (clp->cl_rpcclient == NULL) {
1642 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1644 up_write(&clp->cl_sem);
1645 printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
1646 err = PTR_ERR(xprt);
1649 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1650 server->rpc_ops->version, authflavour);
1652 up_write(&clp->cl_sem);
1653 printk(KERN_WARNING "NFS: cannot create RPC client.\n");
1655 err = PTR_ERR(clnt);
1659 clnt->cl_softrtry = 1;
1660 clnt->cl_chatty = 1;
1661 clp->cl_rpcclient = clnt;
1662 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1663 if (IS_ERR(clp->cl_cred)) {
1664 up_write(&clp->cl_sem);
1665 err = PTR_ERR(clp->cl_cred);
1666 clp->cl_cred = NULL;
1669 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1672 if (list_empty(&clp->cl_superblocks)) {
1673 err = nfs4_init_client(clp);
1675 up_write(&clp->cl_sem);
1679 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1680 clnt = rpc_clone_client(clp->cl_rpcclient);
1682 server->nfs4_state = clp;
1683 up_write(&clp->cl_sem);
1687 printk(KERN_WARNING "NFS: cannot create RPC client.\n");
1688 return PTR_ERR(clnt);
1691 server->client = clnt;
1693 if (server->nfs4_state->cl_idmap == NULL) {
1694 printk(KERN_WARNING "NFS: failed to create idmapper.\n");
1698 if (clnt->cl_auth->au_flavor != authflavour) {
1699 if (rpcauth_create(authflavour, clnt) == NULL) {
1700 printk(KERN_WARNING "NFS: couldn't create credcache!\n");
1705 sb->s_time_gran = 1;
1707 sb->s_op = &nfs4_sops;
1708 err = nfs_sb_init(sb, authflavour);
1713 nfs4_put_client(clp);
1717 static int nfs4_compare_super(struct super_block *sb, void *data)
1719 struct nfs_server *server = data;
1720 struct nfs_server *old = NFS_SB(sb);
1722 if (strcmp(server->hostname, old->hostname) != 0)
1724 if (strcmp(server->mnt_path, old->mnt_path) != 0)
1730 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1735 return ERR_PTR(-EINVAL);
1736 if (src->len < maxlen)
1739 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1741 return ERR_PTR(-ENOMEM);
1743 if (copy_from_user(dst, src->data, maxlen)) {
1746 return ERR_PTR(-EFAULT);
1752 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1753 int flags, const char *dev_name, void *raw_data)
1756 struct nfs_server *server;
1757 struct super_block *s;
1758 struct nfs4_mount_data *data = raw_data;
1762 printk("nfs_read_super: missing data argument\n");
1763 return ERR_PTR(-EINVAL);
1766 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1768 return ERR_PTR(-ENOMEM);
1769 memset(server, 0, sizeof(struct nfs_server));
1770 /* Zero out the NFS state stuff */
1771 init_nfsv4_state(server);
1773 if (data->version != NFS4_MOUNT_VERSION) {
1774 printk("nfs warning: mount version %s than kernel\n",
1775 data->version < NFS4_MOUNT_VERSION ? "older" : "newer");
1778 p = nfs_copy_user_string(NULL, &data->hostname, 256);
1781 server->hostname = p;
1783 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1786 server->mnt_path = p;
1788 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1789 sizeof(server->ip_addr) - 1);
1793 /* We now require that the mount process passes the remote address */
1794 if (data->host_addrlen != sizeof(server->addr)) {
1795 s = ERR_PTR(-EINVAL);
1798 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1799 s = ERR_PTR(-EFAULT);
1802 if (server->addr.sin_family != AF_INET ||
1803 server->addr.sin_addr.s_addr == INADDR_ANY) {
1804 printk("NFS: mount program didn't pass remote IP address!\n");
1805 s = ERR_PTR(-EINVAL);
1809 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1811 if (IS_ERR(s) || s->s_root)
1816 /* Fire up rpciod if not yet running */
1817 if (rpciod_up() != 0) {
1818 printk(KERN_WARNING "NFS: couldn't start rpciod!\n");
1823 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1825 up_write(&s->s_umount);
1826 deactivate_super(s);
1827 return ERR_PTR(error);
1829 s->s_flags |= MS_ACTIVE;
1832 s = (struct super_block *)p;
1834 if (server->mnt_path)
1835 kfree(server->mnt_path);
1836 if (server->hostname)
1837 kfree(server->hostname);
1842 static void nfs4_kill_super(struct super_block *sb)
1844 struct nfs_server *server = NFS_SB(sb);
1846 nfs_return_all_delegations(sb);
1847 kill_anon_super(sb);
1849 nfs4_renewd_prepare_shutdown(server);
1851 if (server->client != NULL && !IS_ERR(server->client))
1852 rpc_shutdown_client(server->client);
1853 rpciod_down(); /* release rpciod */
1855 destroy_nfsv4_state(server);
1857 if (server->hostname != NULL)
1858 kfree(server->hostname);
1862 static struct file_system_type nfs4_fs_type = {
1863 .owner = THIS_MODULE,
1865 .get_sb = nfs4_get_sb,
1866 .kill_sb = nfs4_kill_super,
1867 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1870 #define nfs4_init_once(nfsi) \
1872 INIT_LIST_HEAD(&(nfsi)->open_states); \
1873 nfsi->delegation = NULL; \
1874 nfsi->delegation_state = 0; \
1875 init_rwsem(&nfsi->rwsem); \
1877 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1878 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1880 #define nfs4_init_once(nfsi) \
1882 #define register_nfs4fs() (0)
1883 #define unregister_nfs4fs()
1886 extern int nfs_init_nfspagecache(void);
1887 extern void nfs_destroy_nfspagecache(void);
1888 extern int nfs_init_readpagecache(void);
1889 extern void nfs_destroy_readpagecache(void);
1890 extern int nfs_init_writepagecache(void);
1891 extern void nfs_destroy_writepagecache(void);
1892 #ifdef CONFIG_NFS_DIRECTIO
1893 extern int nfs_init_directcache(void);
1894 extern void nfs_destroy_directcache(void);
1897 static kmem_cache_t * nfs_inode_cachep;
1899 static struct inode *nfs_alloc_inode(struct super_block *sb)
1901 struct nfs_inode *nfsi;
1902 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
1906 return &nfsi->vfs_inode;
1909 static void nfs_destroy_inode(struct inode *inode)
1911 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1914 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
1916 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1918 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1919 SLAB_CTOR_CONSTRUCTOR) {
1920 inode_init_once(&nfsi->vfs_inode);
1921 spin_lock_init(&nfsi->req_lock);
1922 INIT_LIST_HEAD(&nfsi->dirty);
1923 INIT_LIST_HEAD(&nfsi->commit);
1924 INIT_LIST_HEAD(&nfsi->open_files);
1925 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1926 atomic_set(&nfsi->data_updates, 0);
1930 init_waitqueue_head(&nfsi->nfs_i_wait);
1931 nfs4_init_once(nfsi);
1935 static int nfs_init_inodecache(void)
1937 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1938 sizeof(struct nfs_inode),
1939 0, SLAB_RECLAIM_ACCOUNT,
1941 if (nfs_inode_cachep == NULL)
1947 static void nfs_destroy_inodecache(void)
1949 if (kmem_cache_destroy(nfs_inode_cachep))
1950 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
1956 static int __init init_nfs_fs(void)
1960 err = nfs_init_nfspagecache();
1964 err = nfs_init_inodecache();
1968 err = nfs_init_readpagecache();
1972 err = nfs_init_writepagecache();
1976 #ifdef CONFIG_NFS_DIRECTIO
1977 err = nfs_init_directcache();
1982 #ifdef CONFIG_PROC_FS
1983 rpc_proc_register(&nfs_rpcstat);
1985 err = register_filesystem(&nfs_fs_type);
1988 if ((err = register_nfs4fs()) != 0)
1992 #ifdef CONFIG_PROC_FS
1993 rpc_proc_unregister("nfs");
1995 nfs_destroy_writepagecache();
1996 #ifdef CONFIG_NFS_DIRECTIO
1998 nfs_destroy_directcache();
2001 nfs_destroy_readpagecache();
2003 nfs_destroy_inodecache();
2005 nfs_destroy_nfspagecache();
2010 static void __exit exit_nfs_fs(void)
2012 #ifdef CONFIG_NFS_DIRECTIO
2013 nfs_destroy_directcache();
2015 nfs_destroy_writepagecache();
2016 nfs_destroy_readpagecache();
2017 nfs_destroy_inodecache();
2018 nfs_destroy_nfspagecache();
2019 #ifdef CONFIG_PROC_FS
2020 rpc_proc_unregister("nfs");
2022 unregister_filesystem(&nfs_fs_type);
2023 unregister_nfs4fs();
2026 /* Not quite true; I just maintain it */
2027 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2028 MODULE_LICENSE("GPL");
2030 module_init(init_nfs_fs)
2031 module_exit(exit_nfs_fs)