4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/config.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
39 #include <asm/system.h>
40 #include <asm/uaccess.h>
43 #include "delegation.h"
45 #define NFSDBG_FACILITY NFSDBG_VFS
46 #define NFS_PARANOIA 1
48 /* Maximum number of readahead requests
49 * FIXME: this should really be a sysctl so that users may tune it to suit
50 * their needs. People that do NFS over a slow network, might for
51 * instance want to reduce it to something closer to 1 for improved
52 * interactive response.
54 #define NFS_MAX_READAHEAD (RPC_DEF_SLOT_TABLE - 1)
56 static void nfs_invalidate_inode(struct inode *);
57 static int nfs_update_inode(struct inode *, struct nfs_fattr *, unsigned long);
59 static struct inode *nfs_alloc_inode(struct super_block *sb);
60 static void nfs_destroy_inode(struct inode *);
61 static int nfs_write_inode(struct inode *,int);
62 static void nfs_delete_inode(struct inode *);
63 static void nfs_clear_inode(struct inode *);
64 static void nfs_umount_begin(struct super_block *);
65 static int nfs_statfs(struct super_block *, struct kstatfs *);
66 static int nfs_show_options(struct seq_file *, struct vfsmount *);
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 sb->s_maxbytes = fsinfo.maxfilesize;
327 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
328 sb->s_maxbytes = MAX_LFS_FILESIZE;
330 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
331 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
333 /* We're airborne Set socket buffersize */
334 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
336 /* Yargs. It didn't work out. */
338 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
339 if (!IS_ERR(root_inode))
341 return no_root_error;
345 * Create an RPC client handle.
347 static struct rpc_clnt *
348 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
350 struct rpc_timeout timeparms;
351 struct rpc_xprt *xprt = NULL;
352 struct rpc_clnt *clnt = NULL;
353 int tcp = (data->flags & NFS_MOUNT_TCP);
355 /* Initialize timeout values */
356 timeparms.to_initval = data->timeo * HZ / 10;
357 timeparms.to_retries = data->retrans;
358 timeparms.to_maxval = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
359 timeparms.to_exponential = 1;
361 if (!timeparms.to_initval)
362 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
363 if (!timeparms.to_retries)
364 timeparms.to_retries = 5;
366 /* create transport and client */
367 xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
368 &server->addr, &timeparms);
370 dprintk("%s: cannot create RPC transport. Error = %ld\n",
371 __FUNCTION__, PTR_ERR(xprt));
372 return (struct rpc_clnt *)xprt;
374 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
375 server->rpc_ops->version, data->pseudoflavor);
377 dprintk("%s: cannot create RPC client. Error = %ld\n",
378 __FUNCTION__, PTR_ERR(xprt));
383 clnt->cl_softrtry = 1;
393 * The way this works is that the mount process passes a structure
394 * in the data argument which contains the server's IP address
395 * and the root file handle obtained from the server's mount
396 * daemon. We stash these away in the private superblock fields.
399 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
401 struct nfs_server *server;
402 rpc_authflavor_t authflavor;
405 sb->s_blocksize_bits = 0;
408 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
410 server->rsize = nfs_block_size(data->rsize, NULL);
412 server->wsize = nfs_block_size(data->wsize, NULL);
413 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
415 server->acregmin = data->acregmin*HZ;
416 server->acregmax = data->acregmax*HZ;
417 server->acdirmin = data->acdirmin*HZ;
418 server->acdirmax = data->acdirmax*HZ;
420 /* Start lockd here, before we might error out */
421 if (!(server->flags & NFS_MOUNT_NONLM))
424 server->namelen = data->namlen;
425 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
426 if (!server->hostname)
428 strcpy(server->hostname, data->hostname);
430 /* Check NFS protocol revision and initialize RPC op vector
431 * and file handle pool. */
433 if (server->flags & NFS_MOUNT_VER3) {
434 server->rpc_ops = &nfs_v3_clientops;
435 server->caps |= NFS_CAP_READDIRPLUS;
437 server->rpc_ops = &nfs_v2_clientops;
440 server->rpc_ops = &nfs_v2_clientops;
443 /* Fill in pseudoflavor for mount version < 5 */
444 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
445 data->pseudoflavor = RPC_AUTH_UNIX;
446 authflavor = data->pseudoflavor; /* save for sb_init() */
447 /* XXX maybe we want to add a server->pseudoflavor field */
449 /* Create RPC client handles */
450 server->client = nfs_create_client(server, data);
451 if (IS_ERR(server->client))
452 return PTR_ERR(server->client);
453 /* RFC 2623, sec 2.3.2 */
454 if (authflavor != RPC_AUTH_UNIX) {
455 server->client_sys = rpc_clone_client(server->client);
456 if (IS_ERR(server->client_sys))
457 return PTR_ERR(server->client_sys);
458 if (!rpcauth_create(RPC_AUTH_UNIX, server->client_sys))
461 atomic_inc(&server->client->cl_count);
462 server->client_sys = server->client;
465 if (server->flags & NFS_MOUNT_VER3) {
466 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
467 server->namelen = NFS3_MAXNAMLEN;
470 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
471 server->namelen = NFS2_MAXNAMLEN;
474 sb->s_op = &nfs_sops;
475 return nfs_sb_init(sb, authflavor);
479 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
481 struct nfs_server *server = NFS_SB(sb);
482 unsigned char blockbits;
483 unsigned long blockres;
484 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
485 struct nfs_fattr fattr;
486 struct nfs_fsstat res = {
493 error = server->rpc_ops->statfs(server, rootfh, &res);
494 buf->f_type = NFS_SUPER_MAGIC;
499 * Current versions of glibc do not correctly handle the
500 * case where f_frsize != f_bsize. Eventually we want to
501 * report the value of wtmult in this field.
503 buf->f_frsize = sb->s_blocksize;
506 * On most *nix systems, f_blocks, f_bfree, and f_bavail
507 * are reported in units of f_frsize. Linux hasn't had
508 * an f_frsize field in its statfs struct until recently,
509 * thus historically Linux's sys_statfs reports these
510 * fields in units of f_bsize.
512 buf->f_bsize = sb->s_blocksize;
513 blockbits = sb->s_blocksize_bits;
514 blockres = (1 << blockbits) - 1;
515 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
516 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
517 buf->f_bavail = (res.abytes + blockres) >> blockbits;
519 buf->f_files = res.tfiles;
520 buf->f_ffree = res.afiles;
522 buf->f_namelen = server->namelen;
529 printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
530 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
535 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
537 static struct proc_nfs_info {
542 { NFS_MOUNT_SOFT, ",soft", ",hard" },
543 { NFS_MOUNT_INTR, ",intr", "" },
544 { NFS_MOUNT_POSIX, ",posix", "" },
545 { NFS_MOUNT_TCP, ",tcp", ",udp" },
546 { NFS_MOUNT_NOCTO, ",nocto", "" },
547 { NFS_MOUNT_NOAC, ",noac", "" },
548 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
551 struct proc_nfs_info *nfs_infop;
552 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
554 seq_printf(m, ",v%d", nfss->rpc_ops->version);
555 seq_printf(m, ",rsize=%d", nfss->rsize);
556 seq_printf(m, ",wsize=%d", nfss->wsize);
557 if (nfss->acregmin != 3*HZ)
558 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
559 if (nfss->acregmax != 60*HZ)
560 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
561 if (nfss->acdirmin != 30*HZ)
562 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
563 if (nfss->acdirmax != 60*HZ)
564 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
565 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
566 if (nfss->flags & nfs_infop->flag)
567 seq_puts(m, nfs_infop->str);
569 seq_puts(m, nfs_infop->nostr);
571 seq_puts(m, ",addr=");
572 seq_escape(m, nfss->hostname, " \t\n\\");
577 * Invalidate the local caches
580 nfs_zap_caches(struct inode *inode)
582 struct nfs_inode *nfsi = NFS_I(inode);
583 int mode = inode->i_mode;
585 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
586 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
588 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
589 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
590 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS;
592 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS;
596 * Invalidate, but do not unhash, the inode
599 nfs_invalidate_inode(struct inode *inode)
601 umode_t save_mode = inode->i_mode;
603 make_bad_inode(inode);
604 inode->i_mode = save_mode;
605 nfs_zap_caches(inode);
608 struct nfs_find_desc {
610 struct nfs_fattr *fattr;
614 * In NFSv3 we can have 64bit inode numbers. In order to support
615 * this, and re-exported directories (also seen in NFSv2)
616 * we are forced to allow 2 different inodes to have the same
620 nfs_find_actor(struct inode *inode, void *opaque)
622 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
623 struct nfs_fh *fh = desc->fh;
624 struct nfs_fattr *fattr = desc->fattr;
626 if (NFS_FILEID(inode) != fattr->fileid)
628 if (nfs_compare_fh(NFS_FH(inode), fh))
630 if (is_bad_inode(inode) || NFS_STALE(inode))
636 nfs_init_locked(struct inode *inode, void *opaque)
638 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
639 struct nfs_fattr *fattr = desc->fattr;
641 NFS_FILEID(inode) = fattr->fileid;
642 nfs_copy_fh(NFS_FH(inode), desc->fh);
646 /* Don't use READDIRPLUS on directories that we believe are too large */
647 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
650 * This is our front-end to iget that looks up inodes by file handle
651 * instead of inode number.
654 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
656 struct nfs_find_desc desc = {
660 struct inode *inode = NULL;
663 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
667 printk("NFS: Buggy server - nlink == 0!\n");
671 hash = nfs_fattr_to_ino_t(fattr);
673 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
676 if (inode->i_state & I_NEW) {
677 struct nfs_inode *nfsi = NFS_I(inode);
679 /* We set i_ino for the few things that still rely on it,
683 /* We can't support update_atime(), since the server will reset it */
684 inode->i_flags |= S_NOATIME|S_NOCMTIME;
685 inode->i_mode = fattr->mode;
686 /* Why so? Because we want revalidate for devices/FIFOs, and
687 * that's precisely what we have in nfs_file_inode_operations.
689 inode->i_op = &nfs_file_inode_operations;
690 if (S_ISREG(inode->i_mode)) {
691 inode->i_fop = &nfs_file_operations;
692 inode->i_data.a_ops = &nfs_file_aops;
693 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
694 } else if (S_ISDIR(inode->i_mode)) {
695 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
696 inode->i_fop = &nfs_dir_operations;
697 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
698 && fattr->size <= NFS_LIMIT_READDIRPLUS)
699 NFS_FLAGS(inode) |= NFS_INO_ADVISE_RDPLUS;
700 } else if (S_ISLNK(inode->i_mode))
701 inode->i_op = &nfs_symlink_inode_operations;
703 init_special_inode(inode, inode->i_mode, fattr->rdev);
705 nfsi->read_cache_jiffies = fattr->timestamp;
706 inode->i_atime = fattr->atime;
707 inode->i_mtime = fattr->mtime;
708 inode->i_ctime = fattr->ctime;
709 if (fattr->valid & NFS_ATTR_FATTR_V4)
710 nfsi->change_attr = fattr->change_attr;
711 inode->i_size = nfs_size_to_loff_t(fattr->size);
712 inode->i_nlink = fattr->nlink;
713 inode->i_uid = fattr->uid;
714 inode->i_gid = fattr->gid;
715 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
717 * report the blocks in 512byte units
719 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
720 inode->i_blksize = inode->i_sb->s_blocksize;
722 inode->i_blocks = fattr->du.nfs2.blocks;
723 inode->i_blksize = fattr->du.nfs2.blocksize;
725 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
726 nfsi->attrtimeo_timestamp = jiffies;
727 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
728 nfsi->cache_access.cred = NULL;
730 unlock_new_inode(inode);
732 nfs_refresh_inode(inode, fattr);
733 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
735 (long long)NFS_FILEID(inode),
736 atomic_read(&inode->i_count));
742 printk("nfs_fhget: iget failed\n");
746 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
749 nfs_setattr(struct dentry *dentry, struct iattr *attr)
751 struct inode *inode = dentry->d_inode;
752 struct nfs_fattr fattr;
755 if (attr->ia_valid & ATTR_SIZE) {
756 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
757 attr->ia_valid &= ~ATTR_SIZE;
760 /* Optimization: if the end result is no change, don't RPC */
761 attr->ia_valid &= NFS_VALID_ATTRS;
762 if (attr->ia_valid == 0)
766 nfs_begin_data_update(inode);
767 /* Write all dirty data if we're changing file permissions or size */
768 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
769 if (filemap_fdatawrite(inode->i_mapping) == 0)
770 filemap_fdatawait(inode->i_mapping);
773 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
775 nfs_refresh_inode(inode, &fattr);
776 if ((attr->ia_valid & ATTR_MODE) != 0) {
778 mode = inode->i_mode & ~S_IALLUGO;
779 mode |= attr->ia_mode & S_IALLUGO;
780 inode->i_mode = mode;
782 if ((attr->ia_valid & ATTR_UID) != 0)
783 inode->i_uid = attr->ia_uid;
784 if ((attr->ia_valid & ATTR_GID) != 0)
785 inode->i_gid = attr->ia_gid;
786 if ((attr->ia_valid & ATTR_SIZE) != 0) {
787 inode->i_size = attr->ia_size;
788 vmtruncate(inode, attr->ia_size);
791 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
792 NFS_FLAGS(inode) |= NFS_INO_INVALID_ACCESS;
793 nfs_end_data_update(inode);
799 * Wait for the inode to get unlocked.
800 * (Used for NFS_INO_LOCKED and NFS_INO_REVALIDATING).
803 nfs_wait_on_inode(struct inode *inode, int flag)
805 struct rpc_clnt *clnt = NFS_CLIENT(inode);
806 struct nfs_inode *nfsi = NFS_I(inode);
809 if (!(NFS_FLAGS(inode) & flag))
811 atomic_inc(&inode->i_count);
812 error = nfs_wait_event(clnt, nfsi->nfs_i_wait,
813 !(NFS_FLAGS(inode) & flag));
818 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
820 struct inode *inode = dentry->d_inode;
821 struct nfs_inode *nfsi = NFS_I(inode);
822 int need_atime = nfsi->flags & NFS_INO_INVALID_ATIME;
825 if (__IS_FLG(inode, MS_NOATIME))
827 else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
829 /* We may force a getattr if the user cares about atime */
831 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
833 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
835 generic_fillattr(inode, stat);
839 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
841 struct nfs_open_context *ctx;
843 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
845 atomic_set(&ctx->count, 1);
846 ctx->dentry = dget(dentry);
847 ctx->cred = get_rpccred(cred);
849 ctx->lockowner = current->files;
851 init_waitqueue_head(&ctx->waitq);
856 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
859 atomic_inc(&ctx->count);
863 void put_nfs_open_context(struct nfs_open_context *ctx)
865 if (atomic_dec_and_test(&ctx->count)) {
866 if (!list_empty(&ctx->list)) {
867 struct inode *inode = ctx->dentry->d_inode;
868 spin_lock(&inode->i_lock);
869 list_del(&ctx->list);
870 spin_unlock(&inode->i_lock);
872 if (ctx->state != NULL)
873 nfs4_close_state(ctx->state, ctx->mode);
874 if (ctx->cred != NULL)
875 put_rpccred(ctx->cred);
882 * Ensure that mmap has a recent RPC credential for use when writing out
885 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
887 struct inode *inode = filp->f_dentry->d_inode;
888 struct nfs_inode *nfsi = NFS_I(inode);
890 filp->private_data = get_nfs_open_context(ctx);
891 spin_lock(&inode->i_lock);
892 list_add(&ctx->list, &nfsi->open_files);
893 spin_unlock(&inode->i_lock);
896 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
898 struct nfs_inode *nfsi = NFS_I(inode);
899 struct nfs_open_context *pos, *ctx = NULL;
901 spin_lock(&inode->i_lock);
902 list_for_each_entry(pos, &nfsi->open_files, list) {
903 if ((pos->mode & mode) == mode) {
904 ctx = get_nfs_open_context(pos);
908 spin_unlock(&inode->i_lock);
912 void nfs_file_clear_open_context(struct file *filp)
914 struct inode *inode = filp->f_dentry->d_inode;
915 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
918 filp->private_data = NULL;
919 spin_lock(&inode->i_lock);
920 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
921 spin_unlock(&inode->i_lock);
922 put_nfs_open_context(ctx);
927 * These allocate and release file read/write context information.
929 int nfs_open(struct inode *inode, struct file *filp)
931 struct nfs_open_context *ctx;
932 struct rpc_cred *cred;
934 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
936 return PTR_ERR(cred);
937 ctx = alloc_nfs_open_context(filp->f_dentry, cred);
941 ctx->mode = filp->f_mode;
942 nfs_file_set_open_context(filp, ctx);
943 put_nfs_open_context(ctx);
944 if ((filp->f_mode & FMODE_WRITE) != 0)
945 nfs_begin_data_update(inode);
949 int nfs_release(struct inode *inode, struct file *filp)
951 if ((filp->f_mode & FMODE_WRITE) != 0)
952 nfs_end_data_update(inode);
953 nfs_file_clear_open_context(filp);
958 * This function is called whenever some part of NFS notices that
959 * the cached attributes have to be refreshed.
962 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
964 int status = -ESTALE;
965 struct nfs_fattr fattr;
966 struct nfs_inode *nfsi = NFS_I(inode);
967 unsigned long verifier;
970 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
971 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
974 if (!inode || is_bad_inode(inode))
976 if (NFS_STALE(inode))
979 while (NFS_REVALIDATING(inode)) {
980 status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
983 if (NFS_ATTRTIMEO(inode) == 0)
985 if (NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME))
987 status = NFS_STALE(inode) ? -ESTALE : 0;
990 NFS_FLAGS(inode) |= NFS_INO_REVALIDATING;
992 /* Protect against RPC races by saving the change attribute */
993 verifier = nfs_save_change_attribute(inode);
994 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
996 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
998 (long long)NFS_FILEID(inode), status);
999 if (status == -ESTALE) {
1000 nfs_zap_caches(inode);
1001 if (!S_ISDIR(inode->i_mode))
1002 NFS_FLAGS(inode) |= NFS_INO_STALE;
1007 status = nfs_update_inode(inode, &fattr, verifier);
1009 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1011 (long long)NFS_FILEID(inode), status);
1014 flags = nfsi->flags;
1016 * We may need to keep the attributes marked as invalid if
1017 * we raced with nfs_end_attr_update().
1019 if (verifier == nfsi->cache_change_attribute)
1020 nfsi->flags &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1021 /* Do the page cache invalidation */
1022 if (flags & NFS_INO_INVALID_DATA) {
1023 if (S_ISREG(inode->i_mode)) {
1024 if (filemap_fdatawrite(inode->i_mapping) == 0)
1025 filemap_fdatawait(inode->i_mapping);
1028 nfsi->flags &= ~NFS_INO_INVALID_DATA;
1029 invalidate_inode_pages2(inode->i_mapping);
1030 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
1031 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1033 (long long)NFS_FILEID(inode));
1034 /* This ensures we revalidate dentries */
1035 nfsi->cache_change_attribute++;
1037 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1039 (long long)NFS_FILEID(inode));
1042 NFS_FLAGS(inode) &= ~NFS_INO_REVALIDATING;
1043 wake_up(&nfsi->nfs_i_wait);
1049 int nfs_attribute_timeout(struct inode *inode)
1051 struct nfs_inode *nfsi = NFS_I(inode);
1053 if (nfs_have_delegation(inode, FMODE_READ))
1055 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1059 * nfs_revalidate_inode - Revalidate the inode attributes
1060 * @server - pointer to nfs_server struct
1061 * @inode - pointer to inode struct
1063 * Updates inode attribute information by retrieving the data from the server.
1065 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1067 if (!(NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1068 && !nfs_attribute_timeout(inode))
1069 return NFS_STALE(inode) ? -ESTALE : 0;
1070 return __nfs_revalidate_inode(server, inode);
1074 * nfs_begin_data_update
1075 * @inode - pointer to inode
1076 * Declare that a set of operations will update file data on the server
1078 void nfs_begin_data_update(struct inode *inode)
1080 atomic_inc(&NFS_I(inode)->data_updates);
1084 * nfs_end_data_update
1085 * @inode - pointer to inode
1086 * Declare end of the operations that will update file data
1087 * This will mark the inode as immediately needing revalidation
1088 * of its attribute cache.
1090 void nfs_end_data_update(struct inode *inode)
1092 struct nfs_inode *nfsi = NFS_I(inode);
1094 if (!nfs_have_delegation(inode, FMODE_READ)) {
1095 /* Mark the attribute cache for revalidation */
1096 nfsi->flags |= NFS_INO_INVALID_ATTR;
1097 /* Directories and symlinks: invalidate page cache too */
1098 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1099 nfsi->flags |= NFS_INO_INVALID_DATA;
1101 nfsi->cache_change_attribute ++;
1102 atomic_dec(&nfsi->data_updates);
1106 * nfs_end_data_update_defer
1107 * @inode - pointer to inode
1108 * Declare end of the operations that will update file data
1109 * This will defer marking the inode as needing revalidation
1110 * unless there are no other pending updates.
1112 void nfs_end_data_update_defer(struct inode *inode)
1114 struct nfs_inode *nfsi = NFS_I(inode);
1116 if (atomic_dec_and_test(&nfsi->data_updates)) {
1117 /* Mark the attribute cache for revalidation */
1118 nfsi->flags |= NFS_INO_INVALID_ATTR;
1119 /* Directories and symlinks: invalidate page cache too */
1120 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1121 nfsi->flags |= NFS_INO_INVALID_DATA;
1122 nfsi->cache_change_attribute ++;
1127 * nfs_refresh_inode - verify consistency of the inode attribute cache
1128 * @inode - pointer to inode
1129 * @fattr - updated attributes
1131 * Verifies the attribute cache. If we have just changed the attributes,
1132 * so that fattr carries weak cache consistency data, then it may
1133 * also update the ctime/mtime/change_attribute.
1135 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1137 struct nfs_inode *nfsi = NFS_I(inode);
1138 loff_t cur_size, new_isize;
1141 /* Do we hold a delegation? */
1142 if (nfs_have_delegation(inode, FMODE_READ))
1145 /* Are we in the process of updating data on the server? */
1146 data_unstable = nfs_caches_unstable(inode);
1148 if (fattr->valid & NFS_ATTR_FATTR_V4) {
1149 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1150 && nfsi->change_attr == fattr->pre_change_attr)
1151 nfsi->change_attr = fattr->change_attr;
1152 if (!data_unstable && nfsi->change_attr != fattr->change_attr)
1153 nfsi->flags |= NFS_INO_INVALID_ATTR;
1156 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1159 /* Has the inode gone and changed behind our back? */
1160 if (nfsi->fileid != fattr->fileid
1161 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1164 cur_size = i_size_read(inode);
1165 new_isize = nfs_size_to_loff_t(fattr->size);
1167 /* If we have atomic WCC data, we may update some attributes */
1168 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1169 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1170 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1171 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1172 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1175 /* Verify a few of the more important attributes */
1176 if (!data_unstable) {
1177 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)
1178 || cur_size != new_isize)
1179 nfsi->flags |= NFS_INO_INVALID_ATTR;
1180 } else if (S_ISREG(inode->i_mode) && new_isize > cur_size)
1181 nfsi->flags |= NFS_INO_INVALID_ATTR;
1183 /* Have any file permissions changed? */
1184 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1185 || inode->i_uid != fattr->uid
1186 || inode->i_gid != fattr->gid)
1187 nfsi->flags |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1189 /* Has the link count changed? */
1190 if (inode->i_nlink != fattr->nlink)
1191 nfsi->flags |= NFS_INO_INVALID_ATTR;
1193 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1194 nfsi->flags |= NFS_INO_INVALID_ATIME;
1196 nfsi->read_cache_jiffies = fattr->timestamp;
1201 * Many nfs protocol calls return the new file attributes after
1202 * an operation. Here we update the inode to reflect the state
1203 * of the server's inode.
1205 * This is a bit tricky because we have to make sure all dirty pages
1206 * have been sent off to the server before calling invalidate_inode_pages.
1207 * To make sure no other process adds more write requests while we try
1208 * our best to flush them, we make them sleep during the attribute refresh.
1210 * A very similar scenario holds for the dir cache.
1212 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1214 struct nfs_inode *nfsi = NFS_I(inode);
1217 unsigned int invalid = 0;
1221 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1222 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1223 atomic_read(&inode->i_count), fattr->valid);
1225 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1228 if (nfsi->fileid != fattr->fileid) {
1229 printk(KERN_ERR "%s: inode number mismatch\n"
1230 "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1232 inode->i_sb->s_id, (long long)nfsi->fileid,
1233 inode->i_sb->s_id, (long long)fattr->fileid);
1238 * Make sure the inode's type hasn't changed.
1240 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1244 * Update the read time so we don't revalidate too often.
1246 nfsi->read_cache_jiffies = fattr->timestamp;
1248 /* Are we racing with known updates of the metadata on the server? */
1249 data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1251 /* Check if the file size agrees */
1252 new_size = fattr->size;
1253 new_isize = nfs_size_to_loff_t(fattr->size);
1254 cur_isize = i_size_read(inode);
1255 if (cur_isize != new_size) {
1256 #ifdef NFS_DEBUG_VERBOSE
1257 printk(KERN_DEBUG "NFS: isize change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1260 * If we have pending writebacks, things can get
1263 if (S_ISREG(inode->i_mode) && data_unstable) {
1264 if (new_isize > cur_isize) {
1265 inode->i_size = new_isize;
1266 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1269 inode->i_size = new_isize;
1270 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1275 * Note: we don't check inode->i_mtime since pipes etc.
1276 * can change this value in VFS without requiring a
1277 * cache revalidation.
1279 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1280 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1281 #ifdef NFS_DEBUG_VERBOSE
1282 printk(KERN_DEBUG "NFS: mtime change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1285 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1288 if ((fattr->valid & NFS_ATTR_FATTR_V4)
1289 && nfsi->change_attr != fattr->change_attr) {
1290 #ifdef NFS_DEBUG_VERBOSE
1291 printk(KERN_DEBUG "NFS: change_attr change on %s/%ld\n",
1292 inode->i_sb->s_id, inode->i_ino);
1294 nfsi->change_attr = fattr->change_attr;
1296 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS;
1299 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1300 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1302 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1303 inode->i_uid != fattr->uid ||
1304 inode->i_gid != fattr->gid)
1305 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS;
1307 inode->i_mode = fattr->mode;
1308 inode->i_nlink = fattr->nlink;
1309 inode->i_uid = fattr->uid;
1310 inode->i_gid = fattr->gid;
1312 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1314 * report the blocks in 512byte units
1316 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1317 inode->i_blksize = inode->i_sb->s_blocksize;
1319 inode->i_blocks = fattr->du.nfs2.blocks;
1320 inode->i_blksize = fattr->du.nfs2.blocksize;
1323 /* Update attrtimeo value if we're out of the unstable period */
1324 if (invalid & NFS_INO_INVALID_ATTR) {
1325 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1326 nfsi->attrtimeo_timestamp = jiffies;
1327 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1328 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1329 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1330 nfsi->attrtimeo_timestamp = jiffies;
1332 /* Don't invalidate the data if we were to blame */
1333 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1334 || S_ISLNK(inode->i_mode)))
1335 invalid &= ~NFS_INO_INVALID_DATA;
1336 if (!nfs_have_delegation(inode, FMODE_READ))
1337 nfsi->flags |= invalid;
1342 * Big trouble! The inode has become a different object.
1345 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1346 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1349 * No need to worry about unhashing the dentry, as the
1350 * lookup validation will know that the inode is bad.
1351 * (But we fall through to invalidate the caches.)
1353 nfs_invalidate_inode(inode);
1355 NFS_FLAGS(inode) |= NFS_INO_STALE;
1360 * File system information
1363 static int nfs_set_super(struct super_block *s, void *data)
1365 s->s_fs_info = data;
1366 return set_anon_super(s, data);
1369 static int nfs_compare_super(struct super_block *sb, void *data)
1371 struct nfs_server *server = data;
1372 struct nfs_server *old = NFS_SB(sb);
1374 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1376 if (old->addr.sin_port != server->addr.sin_port)
1378 return !nfs_compare_fh(&old->fh, &server->fh);
1381 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1382 int flags, const char *dev_name, void *raw_data)
1385 struct nfs_server *server = NULL;
1386 struct super_block *s;
1387 struct nfs_fh *root;
1388 struct nfs_mount_data *data = raw_data;
1390 s = ERR_PTR(-EINVAL);
1392 dprintk("%s: missing data argument\n", __FUNCTION__);
1395 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1396 dprintk("%s: bad mount version\n", __FUNCTION__);
1399 switch (data->version) {
1405 if (data->flags & NFS_MOUNT_VER3) {
1406 dprintk("%s: mount structure version %d does not support NFSv3\n",
1411 data->root.size = NFS2_FHSIZE;
1412 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1414 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1415 dprintk("%s: mount structure version %d does not support strong security\n",
1421 memset(data->context, 0, sizeof(data->context));
1423 #ifndef CONFIG_NFS_V3
1424 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1425 s = ERR_PTR(-EPROTONOSUPPORT);
1426 if (data->flags & NFS_MOUNT_VER3) {
1427 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1430 #endif /* CONFIG_NFS_V3 */
1432 s = ERR_PTR(-ENOMEM);
1433 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1436 memset(server, 0, sizeof(struct nfs_server));
1437 /* Zero out the NFS state stuff */
1438 init_nfsv4_state(server);
1441 if (data->flags & NFS_MOUNT_VER3)
1442 root->size = data->root.size;
1444 root->size = NFS2_FHSIZE;
1445 s = ERR_PTR(-EINVAL);
1446 if (root->size > sizeof(root->data)) {
1447 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1450 memcpy(root->data, data->root.data, root->size);
1452 /* We now require that the mount process passes the remote address */
1453 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1454 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1455 dprintk("%s: mount program didn't pass remote address!\n",
1460 /* Fire up rpciod if not yet running */
1461 s = ERR_PTR(rpciod_up());
1463 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1464 __FUNCTION__, PTR_ERR(s));
1468 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1469 if (IS_ERR(s) || s->s_root)
1470 goto out_rpciod_down;
1474 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1476 up_write(&s->s_umount);
1477 deactivate_super(s);
1478 return ERR_PTR(error);
1480 s->s_flags |= MS_ACTIVE;
1489 static void nfs_kill_super(struct super_block *s)
1491 struct nfs_server *server = NFS_SB(s);
1495 if (server->client != NULL && !IS_ERR(server->client))
1496 rpc_shutdown_client(server->client);
1497 if (server->client_sys != NULL && !IS_ERR(server->client_sys))
1498 rpc_shutdown_client(server->client_sys);
1500 if (!(server->flags & NFS_MOUNT_NONLM))
1501 lockd_down(); /* release rpc.lockd */
1503 rpciod_down(); /* release rpciod */
1505 if (server->hostname != NULL)
1506 kfree(server->hostname);
1510 static struct file_system_type nfs_fs_type = {
1511 .owner = THIS_MODULE,
1513 .get_sb = nfs_get_sb,
1514 .kill_sb = nfs_kill_super,
1515 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1518 #ifdef CONFIG_NFS_V4
1520 static void nfs4_clear_inode(struct inode *);
1523 static struct super_operations nfs4_sops = {
1524 .alloc_inode = nfs_alloc_inode,
1525 .destroy_inode = nfs_destroy_inode,
1526 .write_inode = nfs_write_inode,
1527 .delete_inode = nfs_delete_inode,
1528 .statfs = nfs_statfs,
1529 .clear_inode = nfs4_clear_inode,
1530 .umount_begin = nfs_umount_begin,
1531 .show_options = nfs_show_options,
1535 * Clean out any remaining NFSv4 state that might be left over due
1536 * to open() calls that passed nfs_atomic_lookup, but failed to call
1539 static void nfs4_clear_inode(struct inode *inode)
1541 struct nfs_inode *nfsi = NFS_I(inode);
1543 /* If we are holding a delegation, return it! */
1544 if (nfsi->delegation != NULL)
1545 nfs_inode_return_delegation(inode);
1546 /* First call standard NFS clear_inode() code */
1547 nfs_clear_inode(inode);
1548 /* Now clear out any remaining state */
1549 while (!list_empty(&nfsi->open_states)) {
1550 struct nfs4_state *state;
1552 state = list_entry(nfsi->open_states.next,
1555 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1558 (long long)NFS_FILEID(inode),
1560 BUG_ON(atomic_read(&state->count) != 1);
1561 nfs4_close_state(state, state->state);
1566 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1568 struct nfs_server *server;
1569 struct nfs4_client *clp = NULL;
1570 struct rpc_xprt *xprt = NULL;
1571 struct rpc_clnt *clnt = NULL;
1572 struct rpc_timeout timeparms;
1573 rpc_authflavor_t authflavour;
1574 int proto, err = -EIO;
1576 sb->s_blocksize_bits = 0;
1577 sb->s_blocksize = 0;
1578 server = NFS_SB(sb);
1579 if (data->rsize != 0)
1580 server->rsize = nfs_block_size(data->rsize, NULL);
1581 if (data->wsize != 0)
1582 server->wsize = nfs_block_size(data->wsize, NULL);
1583 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1584 server->caps = NFS_CAP_ATOMIC_OPEN;
1586 server->acregmin = data->acregmin*HZ;
1587 server->acregmax = data->acregmax*HZ;
1588 server->acdirmin = data->acdirmin*HZ;
1589 server->acdirmax = data->acdirmax*HZ;
1591 server->rpc_ops = &nfs_v4_clientops;
1592 /* Initialize timeout values */
1594 timeparms.to_initval = data->timeo * HZ / 10;
1595 timeparms.to_retries = data->retrans;
1596 timeparms.to_exponential = 1;
1597 if (!timeparms.to_retries)
1598 timeparms.to_retries = 5;
1600 proto = data->proto;
1601 /* Which IP protocol do we use? */
1604 timeparms.to_maxval = RPC_MAX_TCP_TIMEOUT;
1605 if (!timeparms.to_initval)
1606 timeparms.to_initval = 600 * HZ / 10;
1609 timeparms.to_maxval = RPC_MAX_UDP_TIMEOUT;
1610 if (!timeparms.to_initval)
1611 timeparms.to_initval = 11 * HZ / 10;
1617 clp = nfs4_get_client(&server->addr.sin_addr);
1619 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1623 /* Now create transport and client */
1624 authflavour = RPC_AUTH_UNIX;
1625 if (data->auth_flavourlen != 0) {
1626 if (data->auth_flavourlen != 1) {
1627 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1628 __FUNCTION__, data->auth_flavourlen);
1632 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1638 down_write(&clp->cl_sem);
1639 if (clp->cl_rpcclient == NULL) {
1640 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1642 up_write(&clp->cl_sem);
1643 err = PTR_ERR(xprt);
1644 dprintk("%s: cannot create RPC transport. Error = %d\n",
1648 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1649 server->rpc_ops->version, authflavour);
1651 up_write(&clp->cl_sem);
1652 err = PTR_ERR(clnt);
1653 dprintk("%s: cannot create RPC client. Error = %d\n",
1658 clnt->cl_softrtry = 1;
1659 clnt->cl_chatty = 1;
1660 clp->cl_rpcclient = clnt;
1661 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1662 if (IS_ERR(clp->cl_cred)) {
1663 up_write(&clp->cl_sem);
1664 err = PTR_ERR(clp->cl_cred);
1665 clp->cl_cred = NULL;
1668 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1671 if (list_empty(&clp->cl_superblocks)) {
1672 err = nfs4_init_client(clp);
1674 up_write(&clp->cl_sem);
1678 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1679 clnt = rpc_clone_client(clp->cl_rpcclient);
1681 server->nfs4_state = clp;
1682 up_write(&clp->cl_sem);
1686 err = PTR_ERR(clnt);
1687 dprintk("%s: cannot create RPC client. Error = %d\n",
1692 server->client = clnt;
1694 if (server->nfs4_state->cl_idmap == NULL) {
1695 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1699 if (clnt->cl_auth->au_flavor != authflavour) {
1700 if (rpcauth_create(authflavour, clnt) == NULL) {
1701 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1706 sb->s_time_gran = 1;
1708 sb->s_op = &nfs4_sops;
1709 err = nfs_sb_init(sb, authflavour);
1714 nfs4_put_client(clp);
1718 static int nfs4_compare_super(struct super_block *sb, void *data)
1720 struct nfs_server *server = data;
1721 struct nfs_server *old = NFS_SB(sb);
1723 if (strcmp(server->hostname, old->hostname) != 0)
1725 if (strcmp(server->mnt_path, old->mnt_path) != 0)
1731 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1736 return ERR_PTR(-EINVAL);
1737 if (src->len < maxlen)
1740 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1742 return ERR_PTR(-ENOMEM);
1744 if (copy_from_user(dst, src->data, maxlen)) {
1747 return ERR_PTR(-EFAULT);
1753 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1754 int flags, const char *dev_name, void *raw_data)
1757 struct nfs_server *server;
1758 struct super_block *s;
1759 struct nfs4_mount_data *data = raw_data;
1763 dprintk("%s: missing data argument\n", __FUNCTION__);
1764 return ERR_PTR(-EINVAL);
1766 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1767 dprintk("%s: bad mount version\n", __FUNCTION__);
1768 return ERR_PTR(-EINVAL);
1771 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1773 return ERR_PTR(-ENOMEM);
1774 memset(server, 0, sizeof(struct nfs_server));
1775 /* Zero out the NFS state stuff */
1776 init_nfsv4_state(server);
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 dprintk("%s: mount program didn't pass remote IP address!\n",
1806 s = ERR_PTR(-EINVAL);
1810 /* Fire up rpciod if not yet running */
1811 s = ERR_PTR(rpciod_up());
1813 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1814 __FUNCTION__, PTR_ERR(s));
1818 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1820 if (IS_ERR(s) || s->s_root)
1825 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1827 up_write(&s->s_umount);
1828 deactivate_super(s);
1829 return ERR_PTR(error);
1831 s->s_flags |= MS_ACTIVE;
1834 s = (struct super_block *)p;
1836 if (server->mnt_path)
1837 kfree(server->mnt_path);
1838 if (server->hostname)
1839 kfree(server->hostname);
1844 static void nfs4_kill_super(struct super_block *sb)
1846 struct nfs_server *server = NFS_SB(sb);
1848 nfs_return_all_delegations(sb);
1849 kill_anon_super(sb);
1851 nfs4_renewd_prepare_shutdown(server);
1853 if (server->client != NULL && !IS_ERR(server->client))
1854 rpc_shutdown_client(server->client);
1855 rpciod_down(); /* release rpciod */
1857 destroy_nfsv4_state(server);
1859 if (server->hostname != NULL)
1860 kfree(server->hostname);
1864 static struct file_system_type nfs4_fs_type = {
1865 .owner = THIS_MODULE,
1867 .get_sb = nfs4_get_sb,
1868 .kill_sb = nfs4_kill_super,
1869 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1872 #define nfs4_init_once(nfsi) \
1874 INIT_LIST_HEAD(&(nfsi)->open_states); \
1875 nfsi->delegation = NULL; \
1876 nfsi->delegation_state = 0; \
1877 init_rwsem(&nfsi->rwsem); \
1879 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1880 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1882 #define nfs4_init_once(nfsi) \
1884 #define register_nfs4fs() (0)
1885 #define unregister_nfs4fs()
1888 extern int nfs_init_nfspagecache(void);
1889 extern void nfs_destroy_nfspagecache(void);
1890 extern int nfs_init_readpagecache(void);
1891 extern void nfs_destroy_readpagecache(void);
1892 extern int nfs_init_writepagecache(void);
1893 extern void nfs_destroy_writepagecache(void);
1894 #ifdef CONFIG_NFS_DIRECTIO
1895 extern int nfs_init_directcache(void);
1896 extern void nfs_destroy_directcache(void);
1899 static kmem_cache_t * nfs_inode_cachep;
1901 static struct inode *nfs_alloc_inode(struct super_block *sb)
1903 struct nfs_inode *nfsi;
1904 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
1908 return &nfsi->vfs_inode;
1911 static void nfs_destroy_inode(struct inode *inode)
1913 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1916 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
1918 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1920 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1921 SLAB_CTOR_CONSTRUCTOR) {
1922 inode_init_once(&nfsi->vfs_inode);
1923 spin_lock_init(&nfsi->req_lock);
1924 INIT_LIST_HEAD(&nfsi->dirty);
1925 INIT_LIST_HEAD(&nfsi->commit);
1926 INIT_LIST_HEAD(&nfsi->open_files);
1927 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1928 atomic_set(&nfsi->data_updates, 0);
1932 init_waitqueue_head(&nfsi->nfs_i_wait);
1933 nfs4_init_once(nfsi);
1937 static int nfs_init_inodecache(void)
1939 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1940 sizeof(struct nfs_inode),
1941 0, SLAB_RECLAIM_ACCOUNT,
1943 if (nfs_inode_cachep == NULL)
1949 static void nfs_destroy_inodecache(void)
1951 if (kmem_cache_destroy(nfs_inode_cachep))
1952 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
1958 static int __init init_nfs_fs(void)
1962 err = nfs_init_nfspagecache();
1966 err = nfs_init_inodecache();
1970 err = nfs_init_readpagecache();
1974 err = nfs_init_writepagecache();
1978 #ifdef CONFIG_NFS_DIRECTIO
1979 err = nfs_init_directcache();
1984 #ifdef CONFIG_PROC_FS
1985 rpc_proc_register(&nfs_rpcstat);
1987 err = register_filesystem(&nfs_fs_type);
1990 if ((err = register_nfs4fs()) != 0)
1994 #ifdef CONFIG_PROC_FS
1995 rpc_proc_unregister("nfs");
1997 nfs_destroy_writepagecache();
1998 #ifdef CONFIG_NFS_DIRECTIO
2000 nfs_destroy_directcache();
2003 nfs_destroy_readpagecache();
2005 nfs_destroy_inodecache();
2007 nfs_destroy_nfspagecache();
2012 static void __exit exit_nfs_fs(void)
2014 #ifdef CONFIG_NFS_DIRECTIO
2015 nfs_destroy_directcache();
2017 nfs_destroy_writepagecache();
2018 nfs_destroy_readpagecache();
2019 nfs_destroy_inodecache();
2020 nfs_destroy_nfspagecache();
2021 #ifdef CONFIG_PROC_FS
2022 rpc_proc_unregister("nfs");
2024 unregister_filesystem(&nfs_fs_type);
2025 unregister_nfs4fs();
2028 /* Not quite true; I just maintain it */
2029 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2030 MODULE_LICENSE("GPL");
2032 module_init(init_nfs_fs)
2033 module_exit(exit_nfs_fs)