4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/config.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
38 #include <linux/vserver/xid.h>
40 #include <asm/system.h>
41 #include <asm/uaccess.h>
43 #include "delegation.h"
45 #define NFSDBG_FACILITY NFSDBG_VFS
46 #define NFS_PARANOIA 1
48 /* Maximum number of readahead requests
49 * FIXME: this should really be a sysctl so that users may tune it to suit
50 * their needs. People that do NFS over a slow network, might for
51 * instance want to reduce it to something closer to 1 for improved
52 * interactive response.
54 #define NFS_MAX_READAHEAD (RPC_DEF_SLOT_TABLE - 1)
56 static void nfs_invalidate_inode(struct inode *);
57 static int nfs_update_inode(struct inode *, struct nfs_fattr *, unsigned long);
59 static struct inode *nfs_alloc_inode(struct super_block *sb);
60 static void nfs_destroy_inode(struct inode *);
61 static int nfs_write_inode(struct inode *,int);
62 static void nfs_delete_inode(struct inode *);
63 static void nfs_clear_inode(struct inode *);
64 static void nfs_umount_begin(struct super_block *);
65 static int nfs_statfs(struct super_block *, struct kstatfs *);
66 static int nfs_show_options(struct seq_file *, struct vfsmount *);
68 static struct super_operations nfs_sops = {
69 .alloc_inode = nfs_alloc_inode,
70 .destroy_inode = nfs_destroy_inode,
71 .write_inode = nfs_write_inode,
72 .delete_inode = nfs_delete_inode,
74 .clear_inode = nfs_clear_inode,
75 .umount_begin = nfs_umount_begin,
76 .show_options = nfs_show_options,
82 struct rpc_stat nfs_rpcstat = {
83 .program = &nfs_program
85 static struct rpc_version * nfs_version[] = {
89 #if defined(CONFIG_NFS_V3)
91 #elif defined(CONFIG_NFS_V4)
94 #if defined(CONFIG_NFS_V4)
99 struct rpc_program nfs_program = {
101 .number = NFS_PROGRAM,
102 .nrvers = sizeof(nfs_version) / sizeof(nfs_version[0]),
103 .version = nfs_version,
104 .stats = &nfs_rpcstat,
105 .pipe_dir_name = "/nfs",
108 static inline unsigned long
109 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
111 return nfs_fileid_to_ino_t(fattr->fileid);
115 nfs_write_inode(struct inode *inode, int sync)
117 int flags = sync ? FLUSH_WAIT : 0;
120 ret = nfs_commit_inode(inode, 0, 0, flags);
127 nfs_delete_inode(struct inode * inode)
129 dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
133 * The following should never happen...
135 if (nfs_have_writebacks(inode)) {
136 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
143 * For the moment, the only task for the NFS clear_inode method is to
144 * release the mmap credential
147 nfs_clear_inode(struct inode *inode)
149 struct nfs_inode *nfsi = NFS_I(inode);
150 struct rpc_cred *cred;
153 BUG_ON (!list_empty(&nfsi->open_files));
154 cred = nfsi->cache_access.cred;
157 BUG_ON(atomic_read(&nfsi->data_updates) != 0);
161 nfs_umount_begin(struct super_block *sb)
163 struct nfs_server *server = NFS_SB(sb);
164 struct rpc_clnt *rpc;
166 /* -EIO all pending I/O */
167 if ((rpc = server->client) != NULL)
168 rpc_killall_tasks(rpc);
172 static inline unsigned long
173 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
175 /* make sure blocksize is a power of two */
176 if ((bsize & (bsize - 1)) || nrbitsp) {
177 unsigned char nrbits;
179 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
190 * Calculate the number of 512byte blocks used.
192 static inline unsigned long
193 nfs_calc_block_size(u64 tsize)
195 loff_t used = (tsize + 511) >> 9;
196 return (used > ULONG_MAX) ? ULONG_MAX : used;
200 * Compute and set NFS server blocksize
202 static inline unsigned long
203 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
206 bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
207 else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
208 bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
210 return nfs_block_bits(bsize, nrbitsp);
214 * Obtain the root inode of the file system.
216 static struct inode *
217 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
219 struct nfs_server *server = NFS_SB(sb);
223 error = server->rpc_ops->getroot(server, rootfh, fsinfo);
225 dprintk("nfs_get_root: getattr error = %d\n", -error);
226 return ERR_PTR(error);
229 rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
231 return ERR_PTR(-ENOMEM);
236 * Do NFS version-independent mount processing, and sanity checking
239 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
241 struct nfs_server *server;
242 struct inode *root_inode;
243 struct nfs_fattr fattr;
244 struct nfs_fsinfo fsinfo = {
247 struct nfs_pathconf pathinfo = {
250 int no_root_error = 0;
252 /* We probably want something more informative here */
253 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
257 sb->s_magic = NFS_SUPER_MAGIC;
259 root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
260 /* Did getting the root inode fail? */
261 if (IS_ERR(root_inode)) {
262 no_root_error = PTR_ERR(root_inode);
265 sb->s_root = d_alloc_root(root_inode);
267 no_root_error = -ENOMEM;
270 sb->s_root->d_op = server->rpc_ops->dentry_ops;
272 /* Get some general file system info */
273 if (server->namelen == 0 &&
274 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
275 server->namelen = pathinfo.max_namelen;
276 /* Work out a lot of parameters */
277 if (server->rsize == 0)
278 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
279 if (server->wsize == 0)
280 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
282 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
283 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
284 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
285 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
287 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
288 if (server->rpages > NFS_READ_MAXIOV) {
289 server->rpages = NFS_READ_MAXIOV;
290 server->rsize = server->rpages << PAGE_CACHE_SHIFT;
293 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
294 if (server->wpages > NFS_WRITE_MAXIOV) {
295 server->wpages = NFS_WRITE_MAXIOV;
296 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
299 if (sb->s_blocksize == 0)
300 sb->s_blocksize = nfs_block_bits(server->wsize,
301 &sb->s_blocksize_bits);
302 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
304 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
305 if (server->dtsize > PAGE_CACHE_SIZE)
306 server->dtsize = PAGE_CACHE_SIZE;
307 if (server->dtsize > server->rsize)
308 server->dtsize = server->rsize;
310 if (server->flags & NFS_MOUNT_NOAC) {
311 server->acregmin = server->acregmax = 0;
312 server->acdirmin = server->acdirmax = 0;
313 sb->s_flags |= MS_SYNCHRONOUS;
315 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
317 if (server->flags & NFS_MOUNT_TAGXID)
318 sb->s_flags |= MS_TAGXID;
320 sb->s_maxbytes = fsinfo.maxfilesize;
321 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
322 sb->s_maxbytes = MAX_LFS_FILESIZE;
324 /* We're airborne Set socket buffersize */
325 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
327 /* Yargs. It didn't work out. */
329 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
330 if (!IS_ERR(root_inode))
332 return no_root_error;
336 * Create an RPC client handle.
338 static struct rpc_clnt *
339 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
341 struct rpc_timeout timeparms;
342 struct rpc_xprt *xprt = NULL;
343 struct rpc_clnt *clnt = NULL;
344 int tcp = (data->flags & NFS_MOUNT_TCP);
346 /* Initialize timeout values */
347 timeparms.to_initval = data->timeo * HZ / 10;
348 timeparms.to_retries = data->retrans;
349 timeparms.to_maxval = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
350 timeparms.to_exponential = 1;
352 if (!timeparms.to_initval)
353 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
354 if (!timeparms.to_retries)
355 timeparms.to_retries = 5;
357 /* create transport and client */
358 xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
359 &server->addr, &timeparms);
361 printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
362 return (struct rpc_clnt *)xprt;
364 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
365 server->rpc_ops->version, data->pseudoflavor);
367 printk(KERN_WARNING "NFS: cannot create RPC client.\n");
371 clnt->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
372 clnt->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
373 clnt->cl_droppriv = (server->flags & NFS_MOUNT_BROKEN_SUID) ? 1 : 0;
374 clnt->cl_tagxid = (server->flags & NFS_MOUNT_TAGXID) ? 1 : 0;
385 * The way this works is that the mount process passes a structure
386 * in the data argument which contains the server's IP address
387 * and the root file handle obtained from the server's mount
388 * daemon. We stash these away in the private superblock fields.
391 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
393 struct nfs_server *server;
394 rpc_authflavor_t authflavor;
397 sb->s_blocksize_bits = 0;
400 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
402 server->rsize = nfs_block_size(data->rsize, NULL);
404 server->wsize = nfs_block_size(data->wsize, NULL);
405 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
407 server->acregmin = data->acregmin*HZ;
408 server->acregmax = data->acregmax*HZ;
409 server->acdirmin = data->acdirmin*HZ;
410 server->acdirmax = data->acdirmax*HZ;
412 /* Start lockd here, before we might error out */
413 if (!(server->flags & NFS_MOUNT_NONLM))
416 server->namelen = data->namlen;
417 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
418 if (!server->hostname)
420 strcpy(server->hostname, data->hostname);
422 /* Check NFS protocol revision and initialize RPC op vector
423 * and file handle pool. */
424 if (server->flags & NFS_MOUNT_VER3) {
426 server->rpc_ops = &nfs_v3_clientops;
427 server->caps |= NFS_CAP_READDIRPLUS;
428 if (data->version < 4) {
429 printk(KERN_NOTICE "NFS: NFSv3 not supported by mount program.\n");
433 printk(KERN_NOTICE "NFS: NFSv3 not supported.\n");
437 server->rpc_ops = &nfs_v2_clientops;
440 /* Fill in pseudoflavor for mount version < 5 */
441 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
442 data->pseudoflavor = RPC_AUTH_UNIX;
443 authflavor = data->pseudoflavor; /* save for sb_init() */
444 /* XXX maybe we want to add a server->pseudoflavor field */
446 /* Create RPC client handles */
447 server->client = nfs_create_client(server, data);
448 if (IS_ERR(server->client))
449 return PTR_ERR(server->client);
450 /* RFC 2623, sec 2.3.2 */
451 if (authflavor != RPC_AUTH_UNIX) {
452 server->client_sys = rpc_clone_client(server->client);
453 if (IS_ERR(server->client_sys))
454 return PTR_ERR(server->client_sys);
455 if (!rpcauth_create(RPC_AUTH_UNIX, server->client_sys))
458 atomic_inc(&server->client->cl_count);
459 server->client_sys = server->client;
462 if (server->flags & NFS_MOUNT_VER3) {
463 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
464 server->namelen = NFS3_MAXNAMLEN;
466 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
467 server->namelen = NFS2_MAXNAMLEN;
470 sb->s_op = &nfs_sops;
471 return nfs_sb_init(sb, authflavor);
475 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
477 struct nfs_server *server = NFS_SB(sb);
478 unsigned char blockbits;
479 unsigned long blockres;
480 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
481 struct nfs_fattr fattr;
482 struct nfs_fsstat res = {
489 error = server->rpc_ops->statfs(server, rootfh, &res);
490 buf->f_type = NFS_SUPER_MAGIC;
494 buf->f_frsize = server->wtmult;
495 buf->f_bsize = sb->s_blocksize;
496 blockbits = sb->s_blocksize_bits;
497 blockres = (1 << blockbits) - 1;
498 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
499 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
500 buf->f_bavail = (res.abytes + blockres) >> blockbits;
501 buf->f_files = res.tfiles;
502 buf->f_ffree = res.afiles;
504 buf->f_namelen = server->namelen;
511 printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
512 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
517 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
519 static struct proc_nfs_info {
524 { NFS_MOUNT_SOFT, ",soft", ",hard" },
525 { NFS_MOUNT_INTR, ",intr", "" },
526 { NFS_MOUNT_POSIX, ",posix", "" },
527 { NFS_MOUNT_TCP, ",tcp", ",udp" },
528 { NFS_MOUNT_NOCTO, ",nocto", "" },
529 { NFS_MOUNT_NOAC, ",noac", "" },
530 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
531 { NFS_MOUNT_BROKEN_SUID, ",broken_suid", "" },
532 { NFS_MOUNT_TAGXID, ",tagxid", "" },
535 struct proc_nfs_info *nfs_infop;
536 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
538 seq_printf(m, ",v%d", nfss->rpc_ops->version);
539 seq_printf(m, ",rsize=%d", nfss->rsize);
540 seq_printf(m, ",wsize=%d", nfss->wsize);
541 if (nfss->acregmin != 3*HZ)
542 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
543 if (nfss->acregmax != 60*HZ)
544 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
545 if (nfss->acdirmin != 30*HZ)
546 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
547 if (nfss->acdirmax != 60*HZ)
548 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
549 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
550 if (nfss->flags & nfs_infop->flag)
551 seq_puts(m, nfs_infop->str);
553 seq_puts(m, nfs_infop->nostr);
555 seq_puts(m, ",addr=");
556 seq_escape(m, nfss->hostname, " \t\n\\");
561 * Invalidate the local caches
564 nfs_zap_caches(struct inode *inode)
566 struct nfs_inode *nfsi = NFS_I(inode);
567 int mode = inode->i_mode;
569 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
570 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
572 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
573 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
574 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
576 nfsi->flags |= NFS_INO_INVALID_ATTR;
580 * Invalidate, but do not unhash, the inode
583 nfs_invalidate_inode(struct inode *inode)
585 umode_t save_mode = inode->i_mode;
587 make_bad_inode(inode);
588 inode->i_mode = save_mode;
589 nfs_zap_caches(inode);
592 struct nfs_find_desc {
594 struct nfs_fattr *fattr;
598 * In NFSv3 we can have 64bit inode numbers. In order to support
599 * this, and re-exported directories (also seen in NFSv2)
600 * we are forced to allow 2 different inodes to have the same
604 nfs_find_actor(struct inode *inode, void *opaque)
606 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
607 struct nfs_fh *fh = desc->fh;
608 struct nfs_fattr *fattr = desc->fattr;
610 if (NFS_FILEID(inode) != fattr->fileid)
612 if (nfs_compare_fh(NFS_FH(inode), fh))
614 if (is_bad_inode(inode))
620 nfs_init_locked(struct inode *inode, void *opaque)
622 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
623 struct nfs_fattr *fattr = desc->fattr;
625 NFS_FILEID(inode) = fattr->fileid;
626 nfs_copy_fh(NFS_FH(inode), desc->fh);
630 /* Don't use READDIRPLUS on directories that we believe are too large */
631 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
634 * This is our front-end to iget that looks up inodes by file handle
635 * instead of inode number.
638 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
640 struct nfs_find_desc desc = {
644 struct inode *inode = NULL;
647 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
651 printk("NFS: Buggy server - nlink == 0!\n");
655 hash = nfs_fattr_to_ino_t(fattr);
657 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
660 if (inode->i_state & I_NEW) {
661 struct nfs_inode *nfsi = NFS_I(inode);
663 /* We set i_ino for the few things that still rely on it,
667 /* We can't support update_atime(), since the server will reset it */
668 inode->i_flags |= S_NOATIME|S_NOCMTIME;
669 inode->i_mode = fattr->mode;
670 /* Why so? Because we want revalidate for devices/FIFOs, and
671 * that's precisely what we have in nfs_file_inode_operations.
673 inode->i_op = &nfs_file_inode_operations;
674 if (S_ISREG(inode->i_mode)) {
675 inode->i_fop = &nfs_file_operations;
676 inode->i_data.a_ops = &nfs_file_aops;
677 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
678 } else if (S_ISDIR(inode->i_mode)) {
679 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
680 inode->i_fop = &nfs_dir_operations;
681 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
682 && fattr->size <= NFS_LIMIT_READDIRPLUS)
683 NFS_FLAGS(inode) |= NFS_INO_ADVISE_RDPLUS;
684 } else if (S_ISLNK(inode->i_mode))
685 inode->i_op = &nfs_symlink_inode_operations;
687 init_special_inode(inode, inode->i_mode, fattr->rdev);
689 nfsi->read_cache_jiffies = fattr->timestamp;
690 inode->i_atime = fattr->atime;
691 inode->i_mtime = fattr->mtime;
692 inode->i_ctime = fattr->ctime;
693 if (fattr->valid & NFS_ATTR_FATTR_V4)
694 nfsi->change_attr = fattr->change_attr;
695 inode->i_size = nfs_size_to_loff_t(fattr->size);
696 inode->i_nlink = fattr->nlink;
697 inode->i_uid = INOXID_UID(XID_TAG(inode), fattr->uid, fattr->gid);
698 inode->i_gid = INOXID_GID(XID_TAG(inode), fattr->uid, fattr->gid);
699 inode->i_xid = INOXID_XID(XID_TAG(inode), fattr->uid, fattr->gid, 0);
700 /* maybe fattr->xid someday */
701 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
703 * report the blocks in 512byte units
705 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
706 inode->i_blksize = inode->i_sb->s_blocksize;
708 inode->i_blocks = fattr->du.nfs2.blocks;
709 inode->i_blksize = fattr->du.nfs2.blocksize;
711 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
712 nfsi->attrtimeo_timestamp = jiffies;
713 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
714 nfsi->cache_access.cred = NULL;
716 unlock_new_inode(inode);
718 nfs_refresh_inode(inode, fattr);
719 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
721 (long long)NFS_FILEID(inode),
722 atomic_read(&inode->i_count));
728 make_bad_inode(inode);
733 printk("nfs_fhget: iget failed\n");
737 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
740 nfs_setattr(struct dentry *dentry, struct iattr *attr)
742 struct inode *inode = dentry->d_inode;
743 struct nfs_fattr fattr;
746 if (attr->ia_valid & ATTR_SIZE) {
747 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
748 attr->ia_valid &= ~ATTR_SIZE;
751 /* Optimization: if the end result is no change, don't RPC */
752 attr->ia_valid &= NFS_VALID_ATTRS;
753 if (attr->ia_valid == 0)
757 nfs_begin_data_update(inode);
758 /* Write all dirty data if we're changing file permissions or size */
759 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
760 if (filemap_fdatawrite(inode->i_mapping) == 0)
761 filemap_fdatawait(inode->i_mapping);
764 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
766 nfs_refresh_inode(inode, &fattr);
767 if ((attr->ia_valid & ATTR_MODE) != 0) {
769 mode = inode->i_mode & ~S_IALLUGO;
770 mode |= attr->ia_mode & S_IALLUGO;
771 inode->i_mode = mode;
773 if ((attr->ia_valid & ATTR_UID) != 0)
774 inode->i_uid = attr->ia_uid;
775 if ((attr->ia_valid & ATTR_GID) != 0)
776 inode->i_gid = attr->ia_gid;
777 if ((attr->ia_valid & ATTR_XID) != 0)
778 inode->i_xid = attr->ia_xid;
779 if ((attr->ia_valid & ATTR_SIZE) != 0) {
780 inode->i_size = attr->ia_size;
781 vmtruncate(inode, attr->ia_size);
784 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
785 struct rpc_cred **cred = &NFS_I(inode)->cache_access.cred;
791 nfs_end_data_update(inode);
797 * Wait for the inode to get unlocked.
798 * (Used for NFS_INO_LOCKED and NFS_INO_REVALIDATING).
801 nfs_wait_on_inode(struct inode *inode, int flag)
803 struct rpc_clnt *clnt = NFS_CLIENT(inode);
804 struct nfs_inode *nfsi = NFS_I(inode);
807 if (!(NFS_FLAGS(inode) & flag))
809 atomic_inc(&inode->i_count);
810 error = nfs_wait_event(clnt, nfsi->nfs_i_wait,
811 !(NFS_FLAGS(inode) & flag));
816 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
818 struct inode *inode = dentry->d_inode;
819 struct nfs_inode *nfsi = NFS_I(inode);
820 int need_atime = nfsi->flags & NFS_INO_INVALID_ATIME;
823 if (__IS_FLG(inode, MS_NOATIME))
825 else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
827 /* We may force a getattr if the user cares about atime */
829 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
831 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
833 generic_fillattr(inode, stat);
837 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
839 struct nfs_open_context *ctx;
841 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
843 atomic_set(&ctx->count, 1);
844 ctx->dentry = dget(dentry);
845 ctx->cred = get_rpccred(cred);
847 ctx->lockowner = current->files;
849 init_waitqueue_head(&ctx->waitq);
854 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
857 atomic_inc(&ctx->count);
861 void put_nfs_open_context(struct nfs_open_context *ctx)
863 if (atomic_dec_and_test(&ctx->count)) {
864 if (ctx->state != NULL)
865 nfs4_close_state(ctx->state, ctx->mode);
866 if (ctx->cred != NULL)
867 put_rpccred(ctx->cred);
874 * Ensure that mmap has a recent RPC credential for use when writing out
877 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
879 struct inode *inode = filp->f_dentry->d_inode;
880 struct nfs_inode *nfsi = NFS_I(inode);
882 filp->private_data = get_nfs_open_context(ctx);
883 spin_lock(&inode->i_lock);
884 list_add(&ctx->list, &nfsi->open_files);
885 spin_unlock(&inode->i_lock);
888 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
890 struct nfs_inode *nfsi = NFS_I(inode);
891 struct nfs_open_context *pos, *ctx = NULL;
893 spin_lock(&inode->i_lock);
894 list_for_each_entry(pos, &nfsi->open_files, list) {
895 if ((pos->mode & mode) == mode) {
896 ctx = get_nfs_open_context(pos);
900 spin_unlock(&inode->i_lock);
904 void nfs_file_clear_open_context(struct file *filp)
906 struct inode *inode = filp->f_dentry->d_inode;
907 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
910 filp->private_data = NULL;
911 spin_lock(&inode->i_lock);
912 list_del(&ctx->list);
913 spin_unlock(&inode->i_lock);
914 put_nfs_open_context(ctx);
919 * These allocate and release file read/write context information.
921 int nfs_open(struct inode *inode, struct file *filp)
923 struct nfs_open_context *ctx;
924 struct rpc_cred *cred;
926 if ((cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0)) == NULL)
928 ctx = alloc_nfs_open_context(filp->f_dentry, cred);
932 ctx->mode = filp->f_mode;
933 nfs_file_set_open_context(filp, ctx);
934 put_nfs_open_context(ctx);
935 if ((filp->f_mode & FMODE_WRITE) != 0)
936 nfs_begin_data_update(inode);
940 int nfs_release(struct inode *inode, struct file *filp)
942 if ((filp->f_mode & FMODE_WRITE) != 0)
943 nfs_end_data_update(inode);
944 nfs_file_clear_open_context(filp);
949 * This function is called whenever some part of NFS notices that
950 * the cached attributes have to be refreshed.
953 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
955 int status = -ESTALE;
956 struct nfs_fattr fattr;
957 struct nfs_inode *nfsi = NFS_I(inode);
958 unsigned long verifier;
961 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
962 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
965 if (!inode || is_bad_inode(inode))
967 if (NFS_STALE(inode) && inode != inode->i_sb->s_root->d_inode)
970 while (NFS_REVALIDATING(inode)) {
971 status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
974 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOAC)
976 if (NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME))
978 status = NFS_STALE(inode) ? -ESTALE : 0;
981 NFS_FLAGS(inode) |= NFS_INO_REVALIDATING;
983 /* Protect against RPC races by saving the change attribute */
984 verifier = nfs_save_change_attribute(inode);
985 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
987 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
989 (long long)NFS_FILEID(inode), status);
990 if (status == -ESTALE) {
991 NFS_FLAGS(inode) |= NFS_INO_STALE;
992 if (inode != inode->i_sb->s_root->d_inode)
993 remove_inode_hash(inode);
998 status = nfs_update_inode(inode, &fattr, verifier);
1000 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1002 (long long)NFS_FILEID(inode), status);
1005 flags = nfsi->flags;
1007 * We may need to keep the attributes marked as invalid if
1008 * we raced with nfs_end_attr_update().
1010 if (verifier == nfsi->cache_change_attribute)
1011 nfsi->flags &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1012 /* Do the page cache invalidation */
1013 if (flags & NFS_INO_INVALID_DATA) {
1014 if (S_ISREG(inode->i_mode)) {
1015 if (filemap_fdatawrite(inode->i_mapping) == 0)
1016 filemap_fdatawait(inode->i_mapping);
1019 nfsi->flags &= ~NFS_INO_INVALID_DATA;
1020 invalidate_inode_pages2(inode->i_mapping);
1021 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
1022 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1024 (long long)NFS_FILEID(inode));
1025 /* This ensures we revalidate dentries */
1026 nfsi->cache_change_attribute++;
1028 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1030 (long long)NFS_FILEID(inode));
1032 NFS_FLAGS(inode) &= ~NFS_INO_STALE;
1034 NFS_FLAGS(inode) &= ~NFS_INO_REVALIDATING;
1035 wake_up(&nfsi->nfs_i_wait);
1041 int nfs_attribute_timeout(struct inode *inode)
1043 struct nfs_inode *nfsi = NFS_I(inode);
1045 if (nfs_have_delegation(inode, FMODE_READ))
1047 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1051 * nfs_revalidate_inode - Revalidate the inode attributes
1052 * @server - pointer to nfs_server struct
1053 * @inode - pointer to inode struct
1055 * Updates inode attribute information by retrieving the data from the server.
1057 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1059 if (!(NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1060 && !nfs_attribute_timeout(inode))
1061 return NFS_STALE(inode) ? -ESTALE : 0;
1062 return __nfs_revalidate_inode(server, inode);
1066 * nfs_begin_data_update
1067 * @inode - pointer to inode
1068 * Declare that a set of operations will update file data on the server
1070 void nfs_begin_data_update(struct inode *inode)
1072 atomic_inc(&NFS_I(inode)->data_updates);
1076 * nfs_end_data_update
1077 * @inode - pointer to inode
1078 * Declare end of the operations that will update file data
1079 * This will mark the inode as immediately needing revalidation
1080 * of its attribute cache.
1082 void nfs_end_data_update(struct inode *inode)
1084 struct nfs_inode *nfsi = NFS_I(inode);
1086 if (!nfs_have_delegation(inode, FMODE_READ)) {
1087 /* Mark the attribute cache for revalidation */
1088 nfsi->flags |= NFS_INO_INVALID_ATTR;
1089 /* Directories and symlinks: invalidate page cache too */
1090 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1091 nfsi->flags |= NFS_INO_INVALID_DATA;
1093 nfsi->cache_change_attribute ++;
1094 atomic_dec(&nfsi->data_updates);
1098 * nfs_end_data_update_defer
1099 * @inode - pointer to inode
1100 * Declare end of the operations that will update file data
1101 * This will defer marking the inode as needing revalidation
1102 * unless there are no other pending updates.
1104 void nfs_end_data_update_defer(struct inode *inode)
1106 struct nfs_inode *nfsi = NFS_I(inode);
1108 if (atomic_dec_and_test(&nfsi->data_updates)) {
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;
1114 nfsi->cache_change_attribute ++;
1119 * nfs_refresh_inode - verify consistency of the inode attribute cache
1120 * @inode - pointer to inode
1121 * @fattr - updated attributes
1123 * Verifies the attribute cache. If we have just changed the attributes,
1124 * so that fattr carries weak cache consistency data, then it may
1125 * also update the ctime/mtime/change_attribute.
1127 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1129 struct nfs_inode *nfsi = NFS_I(inode);
1130 loff_t cur_size, new_isize;
1136 /* Do we hold a delegation? */
1137 if (nfs_have_delegation(inode, FMODE_READ))
1140 /* Are we in the process of updating data on the server? */
1141 data_unstable = nfs_caches_unstable(inode);
1143 if (fattr->valid & NFS_ATTR_FATTR_V4) {
1144 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1145 && nfsi->change_attr == fattr->pre_change_attr)
1146 nfsi->change_attr = fattr->change_attr;
1147 if (!data_unstable && nfsi->change_attr != fattr->change_attr)
1148 nfsi->flags |= NFS_INO_INVALID_ATTR;
1151 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1154 /* Has the inode gone and changed behind our back? */
1155 if (nfsi->fileid != fattr->fileid
1156 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1159 cur_size = i_size_read(inode);
1160 new_isize = nfs_size_to_loff_t(fattr->size);
1162 /* If we have atomic WCC data, we may update some attributes */
1163 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1164 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1165 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1166 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1167 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1170 /* Verify a few of the more important attributes */
1171 if (!data_unstable) {
1172 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)
1173 || cur_size != new_isize)
1174 nfsi->flags |= NFS_INO_INVALID_ATTR;
1175 } else if (S_ISREG(inode->i_mode) && new_isize > cur_size)
1176 nfsi->flags |= NFS_INO_INVALID_ATTR;
1178 uid = INOXID_UID(XID_TAG(inode), fattr->uid, fattr->gid);
1179 gid = INOXID_GID(XID_TAG(inode), fattr->uid, fattr->gid);
1180 xid = INOXID_XID(XID_TAG(inode), fattr->uid, fattr->gid, 0);
1182 /* Have any file permissions changed? */
1183 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1184 || inode->i_uid != uid
1185 || inode->i_gid != gid
1186 || inode->i_xid != xid)
1187 nfsi->flags |= NFS_INO_INVALID_ATTR;
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;
1224 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1225 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1226 atomic_read(&inode->i_count), fattr->valid);
1228 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1231 if (nfsi->fileid != fattr->fileid) {
1232 printk(KERN_ERR "%s: inode number mismatch\n"
1233 "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1235 inode->i_sb->s_id, (long long)nfsi->fileid,
1236 inode->i_sb->s_id, (long long)fattr->fileid);
1241 * Make sure the inode's type hasn't changed.
1243 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1247 * Update the read time so we don't revalidate too often.
1249 nfsi->read_cache_jiffies = fattr->timestamp;
1251 /* Are we racing with known updates of the metadata on the server? */
1252 data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1254 /* Check if the file size agrees */
1255 new_size = fattr->size;
1256 new_isize = nfs_size_to_loff_t(fattr->size);
1257 cur_isize = i_size_read(inode);
1258 if (cur_isize != new_size) {
1259 #ifdef NFS_DEBUG_VERBOSE
1260 printk(KERN_DEBUG "NFS: isize change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1263 * If we have pending writebacks, things can get
1266 if (S_ISREG(inode->i_mode) && data_unstable) {
1267 if (new_isize > cur_isize) {
1268 inode->i_size = new_isize;
1269 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1272 inode->i_size = new_isize;
1273 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1278 * Note: we don't check inode->i_mtime since pipes etc.
1279 * can change this value in VFS without requiring a
1280 * cache revalidation.
1282 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1283 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1284 #ifdef NFS_DEBUG_VERBOSE
1285 printk(KERN_DEBUG "NFS: mtime change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1288 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1291 if ((fattr->valid & NFS_ATTR_FATTR_V4)
1292 && nfsi->change_attr != fattr->change_attr) {
1293 #ifdef NFS_DEBUG_VERBOSE
1294 printk(KERN_DEBUG "NFS: change_attr change on %s/%ld\n",
1295 inode->i_sb->s_id, inode->i_ino);
1297 nfsi->change_attr = fattr->change_attr;
1299 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1302 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1303 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1305 uid = INOXID_UID(XID_TAG(inode), fattr->uid, fattr->gid);
1306 gid = INOXID_GID(XID_TAG(inode), fattr->uid, fattr->gid);
1307 xid = INOXID_XID(XID_TAG(inode), fattr->uid, fattr->gid, 0);
1309 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1310 inode->i_uid != uid ||
1311 inode->i_gid != gid ||
1312 inode->i_xid != xid) {
1313 struct rpc_cred **cred = &NFS_I(inode)->cache_access.cred;
1318 invalid |= NFS_INO_INVALID_ATTR;
1321 inode->i_mode = fattr->mode;
1322 inode->i_nlink = fattr->nlink;
1327 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1329 * report the blocks in 512byte units
1331 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1332 inode->i_blksize = inode->i_sb->s_blocksize;
1334 inode->i_blocks = fattr->du.nfs2.blocks;
1335 inode->i_blksize = fattr->du.nfs2.blocksize;
1338 /* Update attrtimeo value if we're out of the unstable period */
1339 if (invalid & NFS_INO_INVALID_ATTR) {
1340 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1341 nfsi->attrtimeo_timestamp = jiffies;
1342 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1343 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1344 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1345 nfsi->attrtimeo_timestamp = jiffies;
1347 /* Don't invalidate the data if we were to blame */
1348 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1349 || S_ISLNK(inode->i_mode)))
1350 invalid &= ~NFS_INO_INVALID_DATA;
1351 if (!nfs_have_delegation(inode, FMODE_READ))
1352 nfsi->flags |= invalid;
1357 * Big trouble! The inode has become a different object.
1360 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1361 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1364 * No need to worry about unhashing the dentry, as the
1365 * lookup validation will know that the inode is bad.
1366 * (But we fall through to invalidate the caches.)
1368 nfs_invalidate_inode(inode);
1374 * File system information
1377 static int nfs_set_super(struct super_block *s, void *data)
1379 s->s_fs_info = data;
1380 return set_anon_super(s, data);
1383 static int nfs_compare_super(struct super_block *sb, void *data)
1385 struct nfs_server *server = data;
1386 struct nfs_server *old = NFS_SB(sb);
1388 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1390 if (old->addr.sin_port != server->addr.sin_port)
1392 return !nfs_compare_fh(&old->fh, &server->fh);
1395 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1396 int flags, const char *dev_name, void *raw_data)
1399 struct nfs_server *server;
1400 struct super_block *s;
1401 struct nfs_fh *root;
1402 struct nfs_mount_data *data = raw_data;
1405 printk("nfs_read_super: missing data argument\n");
1406 return ERR_PTR(-EINVAL);
1409 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1411 return ERR_PTR(-ENOMEM);
1412 memset(server, 0, sizeof(struct nfs_server));
1413 /* Zero out the NFS state stuff */
1414 init_nfsv4_state(server);
1416 if (data->version != NFS_MOUNT_VERSION) {
1417 printk("nfs warning: mount version %s than kernel\n",
1418 data->version < NFS_MOUNT_VERSION ? "older" : "newer");
1419 if (data->version < 2)
1421 if (data->version < 3)
1423 if (data->version < 4) {
1424 data->flags &= ~NFS_MOUNT_VER3;
1425 data->root.size = NFS2_FHSIZE;
1426 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1428 if (data->version < 5)
1429 data->flags &= ~NFS_MOUNT_SECFLAVOUR;
1433 if (data->flags & NFS_MOUNT_VER3)
1434 root->size = data->root.size;
1436 root->size = NFS2_FHSIZE;
1437 if (root->size > sizeof(root->data)) {
1438 printk("nfs_get_sb: invalid root filehandle\n");
1440 return ERR_PTR(-EINVAL);
1442 memcpy(root->data, data->root.data, root->size);
1444 /* We now require that the mount process passes the remote address */
1445 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1446 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1447 printk("NFS: mount program didn't pass remote address!\n");
1449 return ERR_PTR(-EINVAL);
1452 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1454 if (IS_ERR(s) || s->s_root) {
1461 /* Fire up rpciod if not yet running */
1462 if (rpciod_up() != 0) {
1463 printk(KERN_WARNING "NFS: couldn't start rpciod!\n");
1465 return ERR_PTR(-EIO);
1468 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1470 up_write(&s->s_umount);
1471 deactivate_super(s);
1472 return ERR_PTR(error);
1474 s->s_flags |= MS_ACTIVE;
1478 static void nfs_kill_super(struct super_block *s)
1480 struct nfs_server *server = NFS_SB(s);
1484 nfs4_renewd_prepare_shutdown(server);
1486 if (server->client != NULL && !IS_ERR(server->client))
1487 rpc_shutdown_client(server->client);
1488 if (server->client_sys != NULL && !IS_ERR(server->client_sys))
1489 rpc_shutdown_client(server->client_sys);
1491 if (!(server->flags & NFS_MOUNT_NONLM))
1492 lockd_down(); /* release rpc.lockd */
1494 rpciod_down(); /* release rpciod */
1496 destroy_nfsv4_state(server);
1498 if (server->hostname != NULL)
1499 kfree(server->hostname);
1503 static struct file_system_type nfs_fs_type = {
1504 .owner = THIS_MODULE,
1506 .get_sb = nfs_get_sb,
1507 .kill_sb = nfs_kill_super,
1508 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1511 #ifdef CONFIG_NFS_V4
1513 static void nfs4_clear_inode(struct inode *);
1516 static struct super_operations nfs4_sops = {
1517 .alloc_inode = nfs_alloc_inode,
1518 .destroy_inode = nfs_destroy_inode,
1519 .write_inode = nfs_write_inode,
1520 .delete_inode = nfs_delete_inode,
1521 .statfs = nfs_statfs,
1522 .clear_inode = nfs4_clear_inode,
1523 .umount_begin = nfs_umount_begin,
1524 .show_options = nfs_show_options,
1528 * Clean out any remaining NFSv4 state that might be left over due
1529 * to open() calls that passed nfs_atomic_lookup, but failed to call
1532 static void nfs4_clear_inode(struct inode *inode)
1534 struct nfs_inode *nfsi = NFS_I(inode);
1536 /* If we are holding a delegation, return it! */
1537 if (nfsi->delegation != NULL)
1538 nfs_inode_return_delegation(inode);
1539 /* First call standard NFS clear_inode() code */
1540 nfs_clear_inode(inode);
1541 /* Now clear out any remaining state */
1542 while (!list_empty(&nfsi->open_states)) {
1543 struct nfs4_state *state;
1545 state = list_entry(nfsi->open_states.next,
1548 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1551 (long long)NFS_FILEID(inode),
1553 BUG_ON(atomic_read(&state->count) != 1);
1554 nfs4_close_state(state, state->state);
1559 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1561 struct nfs_server *server;
1562 struct nfs4_client *clp = NULL;
1563 struct rpc_xprt *xprt = NULL;
1564 struct rpc_clnt *clnt = NULL;
1565 struct rpc_timeout timeparms;
1566 rpc_authflavor_t authflavour;
1567 int proto, err = -EIO;
1569 sb->s_blocksize_bits = 0;
1570 sb->s_blocksize = 0;
1571 server = NFS_SB(sb);
1572 if (data->rsize != 0)
1573 server->rsize = nfs_block_size(data->rsize, NULL);
1574 if (data->wsize != 0)
1575 server->wsize = nfs_block_size(data->wsize, NULL);
1576 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1578 /* NFSv4 doesn't use NLM locking */
1579 server->flags |= NFS_MOUNT_NONLM;
1581 server->acregmin = data->acregmin*HZ;
1582 server->acregmax = data->acregmax*HZ;
1583 server->acdirmin = data->acdirmin*HZ;
1584 server->acdirmax = data->acdirmax*HZ;
1586 server->rpc_ops = &nfs_v4_clientops;
1587 /* Initialize timeout values */
1589 timeparms.to_initval = data->timeo * HZ / 10;
1590 timeparms.to_retries = data->retrans;
1591 timeparms.to_exponential = 1;
1592 if (!timeparms.to_retries)
1593 timeparms.to_retries = 5;
1595 proto = data->proto;
1596 /* Which IP protocol do we use? */
1599 timeparms.to_maxval = RPC_MAX_TCP_TIMEOUT;
1600 if (!timeparms.to_initval)
1601 timeparms.to_initval = 600 * HZ / 10;
1604 timeparms.to_maxval = RPC_MAX_UDP_TIMEOUT;
1605 if (!timeparms.to_initval)
1606 timeparms.to_initval = 11 * HZ / 10;
1612 clp = nfs4_get_client(&server->addr.sin_addr);
1614 printk(KERN_WARNING "NFS: failed to create NFS4 client.\n");
1618 /* Now create transport and client */
1619 authflavour = RPC_AUTH_UNIX;
1620 if (data->auth_flavourlen != 0) {
1621 if (data->auth_flavourlen > 1)
1622 printk(KERN_INFO "NFS: cannot yet deal with multiple auth flavours.\n");
1623 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1629 down_write(&clp->cl_sem);
1630 if (clp->cl_rpcclient == NULL) {
1631 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1633 up_write(&clp->cl_sem);
1634 printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
1635 err = PTR_ERR(xprt);
1638 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1639 server->rpc_ops->version, authflavour);
1641 up_write(&clp->cl_sem);
1642 printk(KERN_WARNING "NFS: cannot create RPC client.\n");
1644 err = PTR_ERR(clnt);
1647 clnt->cl_chatty = 1;
1648 clp->cl_rpcclient = clnt;
1649 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1650 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1653 if (list_empty(&clp->cl_superblocks)) {
1654 err = nfs4_init_client(clp);
1656 up_write(&clp->cl_sem);
1660 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1661 clnt = rpc_clone_client(clp->cl_rpcclient);
1663 server->nfs4_state = clp;
1664 up_write(&clp->cl_sem);
1668 printk(KERN_WARNING "NFS: cannot create RPC client.\n");
1669 return PTR_ERR(clnt);
1672 clnt->cl_intr = (server->flags & NFS4_MOUNT_INTR) ? 1 : 0;
1673 clnt->cl_softrtry = (server->flags & NFS4_MOUNT_SOFT) ? 1 : 0;
1674 server->client = clnt;
1676 if (server->nfs4_state->cl_idmap == NULL) {
1677 printk(KERN_WARNING "NFS: failed to create idmapper.\n");
1681 if (clnt->cl_auth->au_flavor != authflavour) {
1682 if (rpcauth_create(authflavour, clnt) == NULL) {
1683 printk(KERN_WARNING "NFS: couldn't create credcache!\n");
1688 sb->s_op = &nfs4_sops;
1689 err = nfs_sb_init(sb, authflavour);
1694 nfs4_put_client(clp);
1698 static int nfs4_compare_super(struct super_block *sb, void *data)
1700 struct nfs_server *server = data;
1701 struct nfs_server *old = NFS_SB(sb);
1703 if (strcmp(server->hostname, old->hostname) != 0)
1705 if (strcmp(server->mnt_path, old->mnt_path) != 0)
1711 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1716 return ERR_PTR(-EINVAL);
1717 if (src->len < maxlen)
1720 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1722 return ERR_PTR(-ENOMEM);
1724 if (copy_from_user(dst, src->data, maxlen)) {
1727 return ERR_PTR(-EFAULT);
1733 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1734 int flags, const char *dev_name, void *raw_data)
1737 struct nfs_server *server;
1738 struct super_block *s;
1739 struct nfs4_mount_data *data = raw_data;
1743 printk("nfs_read_super: missing data argument\n");
1744 return ERR_PTR(-EINVAL);
1747 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1749 return ERR_PTR(-ENOMEM);
1750 memset(server, 0, sizeof(struct nfs_server));
1751 /* Zero out the NFS state stuff */
1752 init_nfsv4_state(server);
1754 if (data->version != NFS4_MOUNT_VERSION) {
1755 printk("nfs warning: mount version %s than kernel\n",
1756 data->version < NFS4_MOUNT_VERSION ? "older" : "newer");
1759 p = nfs_copy_user_string(NULL, &data->hostname, 256);
1762 server->hostname = p;
1764 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1767 server->mnt_path = p;
1769 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1770 sizeof(server->ip_addr) - 1);
1774 /* We now require that the mount process passes the remote address */
1775 if (data->host_addrlen != sizeof(server->addr)) {
1776 s = ERR_PTR(-EINVAL);
1779 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1780 s = ERR_PTR(-EFAULT);
1783 if (server->addr.sin_family != AF_INET ||
1784 server->addr.sin_addr.s_addr == INADDR_ANY) {
1785 printk("NFS: mount program didn't pass remote IP address!\n");
1786 s = ERR_PTR(-EINVAL);
1790 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1792 if (IS_ERR(s) || s->s_root)
1797 /* Fire up rpciod if not yet running */
1798 if (rpciod_up() != 0) {
1799 printk(KERN_WARNING "NFS: couldn't start rpciod!\n");
1804 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1806 up_write(&s->s_umount);
1807 deactivate_super(s);
1808 return ERR_PTR(error);
1810 s->s_flags |= MS_ACTIVE;
1813 s = (struct super_block *)p;
1815 if (server->mnt_path)
1816 kfree(server->mnt_path);
1817 if (server->hostname)
1818 kfree(server->hostname);
1823 static void nfs4_kill_super(struct super_block *sb)
1825 nfs_return_all_delegations(sb);
1829 static struct file_system_type nfs4_fs_type = {
1830 .owner = THIS_MODULE,
1832 .get_sb = nfs4_get_sb,
1833 .kill_sb = nfs4_kill_super,
1834 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1837 #define nfs4_init_once(nfsi) \
1839 INIT_LIST_HEAD(&(nfsi)->open_states); \
1840 nfsi->delegation = NULL; \
1841 nfsi->delegation_state = 0; \
1842 init_rwsem(&nfsi->rwsem); \
1844 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1845 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1847 #define nfs4_init_once(nfsi) \
1849 #define register_nfs4fs() (0)
1850 #define unregister_nfs4fs()
1853 extern int nfs_init_nfspagecache(void);
1854 extern void nfs_destroy_nfspagecache(void);
1855 extern int nfs_init_readpagecache(void);
1856 extern int nfs_destroy_readpagecache(void);
1857 extern int nfs_init_writepagecache(void);
1858 extern int nfs_destroy_writepagecache(void);
1860 static kmem_cache_t * nfs_inode_cachep;
1862 static struct inode *nfs_alloc_inode(struct super_block *sb)
1864 struct nfs_inode *nfsi;
1865 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
1869 return &nfsi->vfs_inode;
1872 static void nfs_destroy_inode(struct inode *inode)
1874 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1877 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
1879 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1881 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1882 SLAB_CTOR_CONSTRUCTOR) {
1883 inode_init_once(&nfsi->vfs_inode);
1884 spin_lock_init(&nfsi->req_lock);
1885 INIT_LIST_HEAD(&nfsi->dirty);
1886 INIT_LIST_HEAD(&nfsi->commit);
1887 INIT_LIST_HEAD(&nfsi->open_files);
1888 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1889 atomic_set(&nfsi->data_updates, 0);
1893 init_waitqueue_head(&nfsi->nfs_i_wait);
1894 nfs4_init_once(nfsi);
1898 int nfs_init_inodecache(void)
1900 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1901 sizeof(struct nfs_inode),
1902 0, SLAB_RECLAIM_ACCOUNT,
1904 if (nfs_inode_cachep == NULL)
1910 void nfs_destroy_inodecache(void)
1912 if (kmem_cache_destroy(nfs_inode_cachep))
1913 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
1919 static int __init init_nfs_fs(void)
1923 err = nfs_init_nfspagecache();
1927 err = nfs_init_inodecache();
1931 err = nfs_init_readpagecache();
1935 err = nfs_init_writepagecache();
1939 #ifdef CONFIG_PROC_FS
1940 rpc_proc_register(&nfs_rpcstat);
1942 err = register_filesystem(&nfs_fs_type);
1945 if ((err = register_nfs4fs()) != 0)
1949 rpc_proc_unregister("nfs");
1950 nfs_destroy_writepagecache();
1952 nfs_destroy_readpagecache();
1954 nfs_destroy_inodecache();
1956 nfs_destroy_nfspagecache();
1961 static void __exit exit_nfs_fs(void)
1963 nfs_destroy_writepagecache();
1964 nfs_destroy_readpagecache();
1965 nfs_destroy_inodecache();
1966 nfs_destroy_nfspagecache();
1967 #ifdef CONFIG_PROC_FS
1968 rpc_proc_unregister("nfs");
1970 unregister_filesystem(&nfs_fs_type);
1971 unregister_nfs4fs();
1974 /* Not quite true; I just maintain it */
1975 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1976 MODULE_LICENSE("GPL");
1978 module_init(init_nfs_fs)
1979 module_exit(exit_nfs_fs)