4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/config.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
39 #include <asm/system.h>
40 #include <asm/uaccess.h>
43 #include "delegation.h"
45 #define NFSDBG_FACILITY NFSDBG_VFS
46 #define NFS_PARANOIA 1
48 /* Maximum number of readahead requests
49 * FIXME: this should really be a sysctl so that users may tune it to suit
50 * their needs. People that do NFS over a slow network, might for
51 * instance want to reduce it to something closer to 1 for improved
52 * interactive response.
54 #define NFS_MAX_READAHEAD (RPC_DEF_SLOT_TABLE - 1)
56 static void nfs_invalidate_inode(struct inode *);
57 static int nfs_update_inode(struct inode *, struct nfs_fattr *, unsigned long);
59 static struct inode *nfs_alloc_inode(struct super_block *sb);
60 static void nfs_destroy_inode(struct inode *);
61 static int nfs_write_inode(struct inode *,int);
62 static void nfs_delete_inode(struct inode *);
63 static void nfs_clear_inode(struct inode *);
64 static void nfs_umount_begin(struct super_block *);
65 static int nfs_statfs(struct super_block *, struct kstatfs *);
66 static int nfs_show_options(struct seq_file *, struct vfsmount *);
67 static void nfs_zap_acl_cache(struct inode *);
69 static struct rpc_program nfs_program;
71 static struct super_operations nfs_sops = {
72 .alloc_inode = nfs_alloc_inode,
73 .destroy_inode = nfs_destroy_inode,
74 .write_inode = nfs_write_inode,
75 .delete_inode = nfs_delete_inode,
77 .clear_inode = nfs_clear_inode,
78 .umount_begin = nfs_umount_begin,
79 .show_options = nfs_show_options,
85 static struct rpc_stat nfs_rpcstat = {
86 .program = &nfs_program
88 static struct rpc_version * nfs_version[] = {
92 #if defined(CONFIG_NFS_V3)
94 #elif defined(CONFIG_NFS_V4)
97 #if defined(CONFIG_NFS_V4)
102 static struct rpc_program nfs_program = {
104 .number = NFS_PROGRAM,
105 .nrvers = sizeof(nfs_version) / sizeof(nfs_version[0]),
106 .version = nfs_version,
107 .stats = &nfs_rpcstat,
108 .pipe_dir_name = "/nfs",
111 #ifdef CONFIG_NFS_V3_ACL
112 static struct rpc_stat nfsacl_rpcstat = { &nfsacl_program };
113 static struct rpc_version * nfsacl_version[] = {
114 [3] = &nfsacl_version3,
117 struct rpc_program nfsacl_program = {
119 .number = NFS_ACL_PROGRAM,
120 .nrvers = sizeof(nfsacl_version) / sizeof(nfsacl_version[0]),
121 .version = nfsacl_version,
122 .stats = &nfsacl_rpcstat,
124 #endif /* CONFIG_NFS_V3_ACL */
126 static inline unsigned long
127 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
129 return nfs_fileid_to_ino_t(fattr->fileid);
133 nfs_write_inode(struct inode *inode, int sync)
135 int flags = sync ? FLUSH_WAIT : 0;
138 ret = nfs_commit_inode(inode, 0, 0, flags);
145 nfs_delete_inode(struct inode * inode)
147 dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
151 * The following should never happen...
153 if (nfs_have_writebacks(inode)) {
154 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
161 nfs_clear_inode(struct inode *inode)
163 struct nfs_inode *nfsi = NFS_I(inode);
164 struct rpc_cred *cred;
167 BUG_ON (!list_empty(&nfsi->open_files));
168 nfs_zap_acl_cache(inode);
169 cred = nfsi->cache_access.cred;
172 BUG_ON(atomic_read(&nfsi->data_updates) != 0);
176 nfs_umount_begin(struct super_block *sb)
178 struct rpc_clnt *rpc = NFS_SB(sb)->client;
180 /* -EIO all pending I/O */
182 rpc_killall_tasks(rpc);
183 rpc = NFS_SB(sb)->client_acl;
185 rpc_killall_tasks(rpc);
189 static inline unsigned long
190 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
192 /* make sure blocksize is a power of two */
193 if ((bsize & (bsize - 1)) || nrbitsp) {
194 unsigned char nrbits;
196 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
207 * Calculate the number of 512byte blocks used.
209 static inline unsigned long
210 nfs_calc_block_size(u64 tsize)
212 loff_t used = (tsize + 511) >> 9;
213 return (used > ULONG_MAX) ? ULONG_MAX : used;
217 * Compute and set NFS server blocksize
219 static inline unsigned long
220 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
223 bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
224 else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
225 bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
227 return nfs_block_bits(bsize, nrbitsp);
231 * Obtain the root inode of the file system.
233 static struct inode *
234 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
236 struct nfs_server *server = NFS_SB(sb);
240 error = server->rpc_ops->getroot(server, rootfh, fsinfo);
242 dprintk("nfs_get_root: getattr error = %d\n", -error);
243 return ERR_PTR(error);
246 rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
248 return ERR_PTR(-ENOMEM);
253 * Do NFS version-independent mount processing, and sanity checking
256 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
258 struct nfs_server *server;
259 struct inode *root_inode;
260 struct nfs_fattr fattr;
261 struct nfs_fsinfo fsinfo = {
264 struct nfs_pathconf pathinfo = {
267 int no_root_error = 0;
268 unsigned long max_rpc_payload;
270 /* We probably want something more informative here */
271 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
275 sb->s_magic = NFS_SUPER_MAGIC;
277 root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
278 /* Did getting the root inode fail? */
279 if (IS_ERR(root_inode)) {
280 no_root_error = PTR_ERR(root_inode);
283 sb->s_root = d_alloc_root(root_inode);
285 no_root_error = -ENOMEM;
288 sb->s_root->d_op = server->rpc_ops->dentry_ops;
290 /* Get some general file system info */
291 if (server->namelen == 0 &&
292 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
293 server->namelen = pathinfo.max_namelen;
294 /* Work out a lot of parameters */
295 if (server->rsize == 0)
296 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
297 if (server->wsize == 0)
298 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
300 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
301 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
302 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
303 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
305 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
306 if (server->rsize > max_rpc_payload)
307 server->rsize = max_rpc_payload;
308 if (server->wsize > max_rpc_payload)
309 server->wsize = max_rpc_payload;
311 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
312 if (server->rpages > NFS_READ_MAXIOV) {
313 server->rpages = NFS_READ_MAXIOV;
314 server->rsize = server->rpages << PAGE_CACHE_SHIFT;
317 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
318 if (server->wpages > NFS_WRITE_MAXIOV) {
319 server->wpages = NFS_WRITE_MAXIOV;
320 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
323 if (sb->s_blocksize == 0)
324 sb->s_blocksize = nfs_block_bits(server->wsize,
325 &sb->s_blocksize_bits);
326 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
328 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
329 if (server->dtsize > PAGE_CACHE_SIZE)
330 server->dtsize = PAGE_CACHE_SIZE;
331 if (server->dtsize > server->rsize)
332 server->dtsize = server->rsize;
334 if (server->flags & NFS_MOUNT_NOAC) {
335 server->acregmin = server->acregmax = 0;
336 server->acdirmin = server->acdirmax = 0;
337 sb->s_flags |= MS_SYNCHRONOUS;
339 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
341 sb->s_maxbytes = fsinfo.maxfilesize;
342 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
343 sb->s_maxbytes = MAX_LFS_FILESIZE;
345 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
346 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
348 /* We're airborne Set socket buffersize */
349 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
351 /* Yargs. It didn't work out. */
353 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
354 if (!IS_ERR(root_inode))
356 return no_root_error;
360 * Create an RPC client handle.
362 static struct rpc_clnt *
363 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
365 struct rpc_timeout timeparms;
366 struct rpc_xprt *xprt = NULL;
367 struct rpc_clnt *clnt = NULL;
368 int tcp = (data->flags & NFS_MOUNT_TCP);
370 /* Initialize timeout values */
371 timeparms.to_initval = data->timeo * HZ / 10;
372 timeparms.to_retries = data->retrans;
373 timeparms.to_maxval = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
374 timeparms.to_exponential = 1;
376 if (!timeparms.to_initval)
377 timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
378 if (!timeparms.to_retries)
379 timeparms.to_retries = 5;
381 /* create transport and client */
382 xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
383 &server->addr, &timeparms);
385 dprintk("%s: cannot create RPC transport. Error = %ld\n",
386 __FUNCTION__, PTR_ERR(xprt));
387 return (struct rpc_clnt *)xprt;
389 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
390 server->rpc_ops->version, data->pseudoflavor);
392 dprintk("%s: cannot create RPC client. Error = %ld\n",
393 __FUNCTION__, PTR_ERR(xprt));
398 clnt->cl_softrtry = 1;
408 * The way this works is that the mount process passes a structure
409 * in the data argument which contains the server's IP address
410 * and the root file handle obtained from the server's mount
411 * daemon. We stash these away in the private superblock fields.
414 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
416 struct nfs_server *server;
417 rpc_authflavor_t authflavor;
420 sb->s_blocksize_bits = 0;
423 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
425 server->rsize = nfs_block_size(data->rsize, NULL);
427 server->wsize = nfs_block_size(data->wsize, NULL);
428 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
430 server->acregmin = data->acregmin*HZ;
431 server->acregmax = data->acregmax*HZ;
432 server->acdirmin = data->acdirmin*HZ;
433 server->acdirmax = data->acdirmax*HZ;
435 /* Start lockd here, before we might error out */
436 if (!(server->flags & NFS_MOUNT_NONLM))
439 server->namelen = data->namlen;
440 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
441 if (!server->hostname)
443 strcpy(server->hostname, data->hostname);
445 /* Check NFS protocol revision and initialize RPC op vector
446 * and file handle pool. */
448 if (server->flags & NFS_MOUNT_VER3) {
449 server->rpc_ops = &nfs_v3_clientops;
450 server->caps |= NFS_CAP_READDIRPLUS;
452 server->rpc_ops = &nfs_v2_clientops;
455 server->rpc_ops = &nfs_v2_clientops;
458 /* Fill in pseudoflavor for mount version < 5 */
459 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
460 data->pseudoflavor = RPC_AUTH_UNIX;
461 authflavor = data->pseudoflavor; /* save for sb_init() */
462 /* XXX maybe we want to add a server->pseudoflavor field */
464 /* Create RPC client handles */
465 server->client = nfs_create_client(server, data);
466 if (IS_ERR(server->client))
467 return PTR_ERR(server->client);
468 /* RFC 2623, sec 2.3.2 */
469 if (authflavor != RPC_AUTH_UNIX) {
470 struct rpc_auth *auth;
472 server->client_sys = rpc_clone_client(server->client);
473 if (IS_ERR(server->client_sys))
474 return PTR_ERR(server->client_sys);
475 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
477 return PTR_ERR(auth);
479 atomic_inc(&server->client->cl_count);
480 server->client_sys = server->client;
482 if (server->flags & NFS_MOUNT_VER3) {
483 #ifdef CONFIG_NFS_V3_ACL
484 if (!(server->flags & NFS_MOUNT_NOACL)) {
485 server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
486 /* No errors! Assume that Sun nfsacls are supported */
487 if (!IS_ERR(server->client_acl))
488 server->caps |= NFS_CAP_ACLS;
491 server->flags &= ~NFS_MOUNT_NOACL;
492 #endif /* CONFIG_NFS_V3_ACL */
493 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
494 server->namelen = NFS3_MAXNAMLEN;
497 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
498 server->namelen = NFS2_MAXNAMLEN;
501 sb->s_op = &nfs_sops;
502 return nfs_sb_init(sb, authflavor);
506 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
508 struct nfs_server *server = NFS_SB(sb);
509 unsigned char blockbits;
510 unsigned long blockres;
511 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
512 struct nfs_fattr fattr;
513 struct nfs_fsstat res = {
520 error = server->rpc_ops->statfs(server, rootfh, &res);
521 buf->f_type = NFS_SUPER_MAGIC;
526 * Current versions of glibc do not correctly handle the
527 * case where f_frsize != f_bsize. Eventually we want to
528 * report the value of wtmult in this field.
530 buf->f_frsize = sb->s_blocksize;
533 * On most *nix systems, f_blocks, f_bfree, and f_bavail
534 * are reported in units of f_frsize. Linux hasn't had
535 * an f_frsize field in its statfs struct until recently,
536 * thus historically Linux's sys_statfs reports these
537 * fields in units of f_bsize.
539 buf->f_bsize = sb->s_blocksize;
540 blockbits = sb->s_blocksize_bits;
541 blockres = (1 << blockbits) - 1;
542 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
543 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
544 buf->f_bavail = (res.abytes + blockres) >> blockbits;
546 buf->f_files = res.tfiles;
547 buf->f_ffree = res.afiles;
549 buf->f_namelen = server->namelen;
556 printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
557 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
562 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
564 static struct proc_nfs_info {
569 { NFS_MOUNT_SOFT, ",soft", ",hard" },
570 { NFS_MOUNT_INTR, ",intr", "" },
571 { NFS_MOUNT_POSIX, ",posix", "" },
572 { NFS_MOUNT_TCP, ",tcp", ",udp" },
573 { NFS_MOUNT_NOCTO, ",nocto", "" },
574 { NFS_MOUNT_NOAC, ",noac", "" },
575 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
576 { NFS_MOUNT_NOACL, ",noacl", "" },
579 struct proc_nfs_info *nfs_infop;
580 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
582 seq_printf(m, ",v%d", nfss->rpc_ops->version);
583 seq_printf(m, ",rsize=%d", nfss->rsize);
584 seq_printf(m, ",wsize=%d", nfss->wsize);
585 if (nfss->acregmin != 3*HZ)
586 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
587 if (nfss->acregmax != 60*HZ)
588 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
589 if (nfss->acdirmin != 30*HZ)
590 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
591 if (nfss->acdirmax != 60*HZ)
592 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
593 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
594 if (nfss->flags & nfs_infop->flag)
595 seq_puts(m, nfs_infop->str);
597 seq_puts(m, nfs_infop->nostr);
599 seq_puts(m, ",addr=");
600 seq_escape(m, nfss->hostname, " \t\n\\");
605 * Invalidate the local caches
608 nfs_zap_caches(struct inode *inode)
610 struct nfs_inode *nfsi = NFS_I(inode);
611 int mode = inode->i_mode;
613 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
614 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
616 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
617 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
618 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
620 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
623 static void nfs_zap_acl_cache(struct inode *inode)
625 void (*clear_acl_cache)(struct inode *);
627 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
628 if (clear_acl_cache != NULL)
629 clear_acl_cache(inode);
630 NFS_I(inode)->flags &= ~NFS_INO_INVALID_ACL;
634 * Invalidate, but do not unhash, the inode
637 nfs_invalidate_inode(struct inode *inode)
639 umode_t save_mode = inode->i_mode;
641 make_bad_inode(inode);
642 inode->i_mode = save_mode;
643 nfs_zap_caches(inode);
646 struct nfs_find_desc {
648 struct nfs_fattr *fattr;
652 * In NFSv3 we can have 64bit inode numbers. In order to support
653 * this, and re-exported directories (also seen in NFSv2)
654 * we are forced to allow 2 different inodes to have the same
658 nfs_find_actor(struct inode *inode, void *opaque)
660 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
661 struct nfs_fh *fh = desc->fh;
662 struct nfs_fattr *fattr = desc->fattr;
664 if (NFS_FILEID(inode) != fattr->fileid)
666 if (nfs_compare_fh(NFS_FH(inode), fh))
668 if (is_bad_inode(inode) || NFS_STALE(inode))
674 nfs_init_locked(struct inode *inode, void *opaque)
676 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
677 struct nfs_fattr *fattr = desc->fattr;
679 NFS_FILEID(inode) = fattr->fileid;
680 nfs_copy_fh(NFS_FH(inode), desc->fh);
684 /* Don't use READDIRPLUS on directories that we believe are too large */
685 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
688 * This is our front-end to iget that looks up inodes by file handle
689 * instead of inode number.
692 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
694 struct nfs_find_desc desc = {
698 struct inode *inode = NULL;
701 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
705 printk("NFS: Buggy server - nlink == 0!\n");
709 hash = nfs_fattr_to_ino_t(fattr);
711 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
714 if (inode->i_state & I_NEW) {
715 struct nfs_inode *nfsi = NFS_I(inode);
717 /* We set i_ino for the few things that still rely on it,
721 /* We can't support update_atime(), since the server will reset it */
722 inode->i_flags |= S_NOATIME|S_NOCMTIME;
723 inode->i_mode = fattr->mode;
724 /* Why so? Because we want revalidate for devices/FIFOs, and
725 * that's precisely what we have in nfs_file_inode_operations.
727 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
728 if (S_ISREG(inode->i_mode)) {
729 inode->i_fop = &nfs_file_operations;
730 inode->i_data.a_ops = &nfs_file_aops;
731 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
732 } else if (S_ISDIR(inode->i_mode)) {
733 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
734 inode->i_fop = &nfs_dir_operations;
735 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
736 && fattr->size <= NFS_LIMIT_READDIRPLUS)
737 NFS_FLAGS(inode) |= NFS_INO_ADVISE_RDPLUS;
738 } else if (S_ISLNK(inode->i_mode))
739 inode->i_op = &nfs_symlink_inode_operations;
741 init_special_inode(inode, inode->i_mode, fattr->rdev);
743 nfsi->read_cache_jiffies = fattr->timestamp;
744 inode->i_atime = fattr->atime;
745 inode->i_mtime = fattr->mtime;
746 inode->i_ctime = fattr->ctime;
747 if (fattr->valid & NFS_ATTR_FATTR_V4)
748 nfsi->change_attr = fattr->change_attr;
749 inode->i_size = nfs_size_to_loff_t(fattr->size);
750 inode->i_nlink = fattr->nlink;
751 inode->i_uid = fattr->uid;
752 inode->i_gid = fattr->gid;
753 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
755 * report the blocks in 512byte units
757 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
758 inode->i_blksize = inode->i_sb->s_blocksize;
760 inode->i_blocks = fattr->du.nfs2.blocks;
761 inode->i_blksize = fattr->du.nfs2.blocksize;
763 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
764 nfsi->attrtimeo_timestamp = jiffies;
765 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
766 nfsi->cache_access.cred = NULL;
768 unlock_new_inode(inode);
770 nfs_refresh_inode(inode, fattr);
771 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
773 (long long)NFS_FILEID(inode),
774 atomic_read(&inode->i_count));
780 printk("nfs_fhget: iget failed\n");
784 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
787 nfs_setattr(struct dentry *dentry, struct iattr *attr)
789 struct inode *inode = dentry->d_inode;
790 struct nfs_fattr fattr;
793 if (attr->ia_valid & ATTR_SIZE) {
794 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
795 attr->ia_valid &= ~ATTR_SIZE;
798 /* Optimization: if the end result is no change, don't RPC */
799 attr->ia_valid &= NFS_VALID_ATTRS;
800 if (attr->ia_valid == 0)
804 nfs_begin_data_update(inode);
805 /* Write all dirty data if we're changing file permissions or size */
806 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
807 if (filemap_fdatawrite(inode->i_mapping) == 0)
808 filemap_fdatawait(inode->i_mapping);
811 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
813 nfs_refresh_inode(inode, &fattr);
814 if ((attr->ia_valid & ATTR_MODE) != 0) {
816 mode = inode->i_mode & ~S_IALLUGO;
817 mode |= attr->ia_mode & S_IALLUGO;
818 inode->i_mode = mode;
820 if ((attr->ia_valid & ATTR_UID) != 0)
821 inode->i_uid = attr->ia_uid;
822 if ((attr->ia_valid & ATTR_GID) != 0)
823 inode->i_gid = attr->ia_gid;
824 if ((attr->ia_valid & ATTR_SIZE) != 0) {
825 inode->i_size = attr->ia_size;
826 vmtruncate(inode, attr->ia_size);
829 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
830 NFS_FLAGS(inode) |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
831 nfs_end_data_update(inode);
837 * Wait for the inode to get unlocked.
838 * (Used for NFS_INO_LOCKED and NFS_INO_REVALIDATING).
841 nfs_wait_on_inode(struct inode *inode, int flag)
843 struct rpc_clnt *clnt = NFS_CLIENT(inode);
844 struct nfs_inode *nfsi = NFS_I(inode);
847 if (!(NFS_FLAGS(inode) & flag))
849 atomic_inc(&inode->i_count);
850 error = nfs_wait_event(clnt, nfsi->nfs_i_wait,
851 !(NFS_FLAGS(inode) & flag));
856 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
858 struct inode *inode = dentry->d_inode;
859 struct nfs_inode *nfsi = NFS_I(inode);
860 int need_atime = nfsi->flags & NFS_INO_INVALID_ATIME;
863 if (__IS_FLG(inode, MS_NOATIME))
865 else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
867 /* We may force a getattr if the user cares about atime */
869 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
871 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
873 generic_fillattr(inode, stat);
877 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
879 struct nfs_open_context *ctx;
881 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
883 atomic_set(&ctx->count, 1);
884 ctx->dentry = dget(dentry);
885 ctx->cred = get_rpccred(cred);
887 ctx->lockowner = current->files;
893 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
896 atomic_inc(&ctx->count);
900 void put_nfs_open_context(struct nfs_open_context *ctx)
902 if (atomic_dec_and_test(&ctx->count)) {
903 if (!list_empty(&ctx->list)) {
904 struct inode *inode = ctx->dentry->d_inode;
905 spin_lock(&inode->i_lock);
906 list_del(&ctx->list);
907 spin_unlock(&inode->i_lock);
909 if (ctx->state != NULL)
910 nfs4_close_state(ctx->state, ctx->mode);
911 if (ctx->cred != NULL)
912 put_rpccred(ctx->cred);
919 * Ensure that mmap has a recent RPC credential for use when writing out
922 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
924 struct inode *inode = filp->f_dentry->d_inode;
925 struct nfs_inode *nfsi = NFS_I(inode);
927 filp->private_data = get_nfs_open_context(ctx);
928 spin_lock(&inode->i_lock);
929 list_add(&ctx->list, &nfsi->open_files);
930 spin_unlock(&inode->i_lock);
933 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
935 struct nfs_inode *nfsi = NFS_I(inode);
936 struct nfs_open_context *pos, *ctx = NULL;
938 spin_lock(&inode->i_lock);
939 list_for_each_entry(pos, &nfsi->open_files, list) {
940 if ((pos->mode & mode) == mode) {
941 ctx = get_nfs_open_context(pos);
945 spin_unlock(&inode->i_lock);
949 void nfs_file_clear_open_context(struct file *filp)
951 struct inode *inode = filp->f_dentry->d_inode;
952 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
955 filp->private_data = NULL;
956 spin_lock(&inode->i_lock);
957 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
958 spin_unlock(&inode->i_lock);
959 put_nfs_open_context(ctx);
964 * These allocate and release file read/write context information.
966 int nfs_open(struct inode *inode, struct file *filp)
968 struct nfs_open_context *ctx;
969 struct rpc_cred *cred;
971 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
973 return PTR_ERR(cred);
974 ctx = alloc_nfs_open_context(filp->f_dentry, cred);
978 ctx->mode = filp->f_mode;
979 nfs_file_set_open_context(filp, ctx);
980 put_nfs_open_context(ctx);
981 if ((filp->f_mode & FMODE_WRITE) != 0)
982 nfs_begin_data_update(inode);
986 int nfs_release(struct inode *inode, struct file *filp)
988 if ((filp->f_mode & FMODE_WRITE) != 0)
989 nfs_end_data_update(inode);
990 nfs_file_clear_open_context(filp);
995 * This function is called whenever some part of NFS notices that
996 * the cached attributes have to be refreshed.
999 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1001 int status = -ESTALE;
1002 struct nfs_fattr fattr;
1003 struct nfs_inode *nfsi = NFS_I(inode);
1004 unsigned long verifier;
1007 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1008 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1011 if (!inode || is_bad_inode(inode))
1013 if (NFS_STALE(inode))
1016 while (NFS_REVALIDATING(inode)) {
1017 status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
1020 if (NFS_ATTRTIMEO(inode) == 0)
1022 if (NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME))
1024 status = NFS_STALE(inode) ? -ESTALE : 0;
1027 NFS_FLAGS(inode) |= NFS_INO_REVALIDATING;
1029 /* Protect against RPC races by saving the change attribute */
1030 verifier = nfs_save_change_attribute(inode);
1031 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1033 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1035 (long long)NFS_FILEID(inode), status);
1036 if (status == -ESTALE) {
1037 nfs_zap_caches(inode);
1038 if (!S_ISDIR(inode->i_mode))
1039 NFS_FLAGS(inode) |= NFS_INO_STALE;
1044 status = nfs_update_inode(inode, &fattr, verifier);
1046 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1048 (long long)NFS_FILEID(inode), status);
1051 flags = nfsi->flags;
1053 * We may need to keep the attributes marked as invalid if
1054 * we raced with nfs_end_attr_update().
1056 if (verifier == nfsi->cache_change_attribute)
1057 nfsi->flags &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1058 /* Do the page cache invalidation */
1059 if (flags & NFS_INO_INVALID_DATA) {
1060 if (S_ISREG(inode->i_mode)) {
1061 if (filemap_fdatawrite(inode->i_mapping) == 0)
1062 filemap_fdatawait(inode->i_mapping);
1065 nfsi->flags &= ~NFS_INO_INVALID_DATA;
1066 invalidate_inode_pages2(inode->i_mapping);
1067 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
1068 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1070 (long long)NFS_FILEID(inode));
1071 /* This ensures we revalidate dentries */
1072 nfsi->cache_change_attribute++;
1074 if (flags & NFS_INO_INVALID_ACL)
1075 nfs_zap_acl_cache(inode);
1076 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1078 (long long)NFS_FILEID(inode));
1081 NFS_FLAGS(inode) &= ~NFS_INO_REVALIDATING;
1082 wake_up(&nfsi->nfs_i_wait);
1088 int nfs_attribute_timeout(struct inode *inode)
1090 struct nfs_inode *nfsi = NFS_I(inode);
1092 if (nfs_have_delegation(inode, FMODE_READ))
1094 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1098 * nfs_revalidate_inode - Revalidate the inode attributes
1099 * @server - pointer to nfs_server struct
1100 * @inode - pointer to inode struct
1102 * Updates inode attribute information by retrieving the data from the server.
1104 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1106 if (!(NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1107 && !nfs_attribute_timeout(inode))
1108 return NFS_STALE(inode) ? -ESTALE : 0;
1109 return __nfs_revalidate_inode(server, inode);
1113 * nfs_begin_data_update
1114 * @inode - pointer to inode
1115 * Declare that a set of operations will update file data on the server
1117 void nfs_begin_data_update(struct inode *inode)
1119 atomic_inc(&NFS_I(inode)->data_updates);
1123 * nfs_end_data_update
1124 * @inode - pointer to inode
1125 * Declare end of the operations that will update file data
1126 * This will mark the inode as immediately needing revalidation
1127 * of its attribute cache.
1129 void nfs_end_data_update(struct inode *inode)
1131 struct nfs_inode *nfsi = NFS_I(inode);
1133 if (!nfs_have_delegation(inode, FMODE_READ)) {
1134 /* Mark the attribute cache for revalidation */
1135 nfsi->flags |= NFS_INO_INVALID_ATTR;
1136 /* Directories and symlinks: invalidate page cache too */
1137 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1138 nfsi->flags |= NFS_INO_INVALID_DATA;
1140 nfsi->cache_change_attribute ++;
1141 atomic_dec(&nfsi->data_updates);
1145 * nfs_end_data_update_defer
1146 * @inode - pointer to inode
1147 * Declare end of the operations that will update file data
1148 * This will defer marking the inode as needing revalidation
1149 * unless there are no other pending updates.
1151 void nfs_end_data_update_defer(struct inode *inode)
1153 struct nfs_inode *nfsi = NFS_I(inode);
1155 if (atomic_dec_and_test(&nfsi->data_updates)) {
1156 /* Mark the attribute cache for revalidation */
1157 nfsi->flags |= NFS_INO_INVALID_ATTR;
1158 /* Directories and symlinks: invalidate page cache too */
1159 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1160 nfsi->flags |= NFS_INO_INVALID_DATA;
1161 nfsi->cache_change_attribute ++;
1166 * nfs_refresh_inode - verify consistency of the inode attribute cache
1167 * @inode - pointer to inode
1168 * @fattr - updated attributes
1170 * Verifies the attribute cache. If we have just changed the attributes,
1171 * so that fattr carries weak cache consistency data, then it may
1172 * also update the ctime/mtime/change_attribute.
1174 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1176 struct nfs_inode *nfsi = NFS_I(inode);
1177 loff_t cur_size, new_isize;
1180 /* Do we hold a delegation? */
1181 if (nfs_have_delegation(inode, FMODE_READ))
1184 /* Are we in the process of updating data on the server? */
1185 data_unstable = nfs_caches_unstable(inode);
1187 if (fattr->valid & NFS_ATTR_FATTR_V4) {
1188 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1189 && nfsi->change_attr == fattr->pre_change_attr)
1190 nfsi->change_attr = fattr->change_attr;
1191 if (!data_unstable && nfsi->change_attr != fattr->change_attr)
1192 nfsi->flags |= NFS_INO_INVALID_ATTR;
1195 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1198 /* Has the inode gone and changed behind our back? */
1199 if (nfsi->fileid != fattr->fileid
1200 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1203 cur_size = i_size_read(inode);
1204 new_isize = nfs_size_to_loff_t(fattr->size);
1206 /* If we have atomic WCC data, we may update some attributes */
1207 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1208 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1209 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1210 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1211 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1214 /* Verify a few of the more important attributes */
1215 if (!data_unstable) {
1216 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)
1217 || cur_size != new_isize)
1218 nfsi->flags |= NFS_INO_INVALID_ATTR;
1219 } else if (S_ISREG(inode->i_mode) && new_isize > cur_size)
1220 nfsi->flags |= NFS_INO_INVALID_ATTR;
1222 /* Have any file permissions changed? */
1223 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1224 || inode->i_uid != fattr->uid
1225 || inode->i_gid != fattr->gid)
1226 nfsi->flags |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1228 /* Has the link count changed? */
1229 if (inode->i_nlink != fattr->nlink)
1230 nfsi->flags |= NFS_INO_INVALID_ATTR;
1232 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1233 nfsi->flags |= NFS_INO_INVALID_ATIME;
1235 nfsi->read_cache_jiffies = fattr->timestamp;
1240 * Many nfs protocol calls return the new file attributes after
1241 * an operation. Here we update the inode to reflect the state
1242 * of the server's inode.
1244 * This is a bit tricky because we have to make sure all dirty pages
1245 * have been sent off to the server before calling invalidate_inode_pages.
1246 * To make sure no other process adds more write requests while we try
1247 * our best to flush them, we make them sleep during the attribute refresh.
1249 * A very similar scenario holds for the dir cache.
1251 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1253 struct nfs_inode *nfsi = NFS_I(inode);
1256 unsigned int invalid = 0;
1260 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1261 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1262 atomic_read(&inode->i_count), fattr->valid);
1264 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1267 if (nfsi->fileid != fattr->fileid) {
1268 printk(KERN_ERR "%s: inode number mismatch\n"
1269 "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1271 inode->i_sb->s_id, (long long)nfsi->fileid,
1272 inode->i_sb->s_id, (long long)fattr->fileid);
1277 * Make sure the inode's type hasn't changed.
1279 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1283 * Update the read time so we don't revalidate too often.
1285 nfsi->read_cache_jiffies = fattr->timestamp;
1287 /* Are we racing with known updates of the metadata on the server? */
1288 data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1290 /* Check if the file size agrees */
1291 new_size = fattr->size;
1292 new_isize = nfs_size_to_loff_t(fattr->size);
1293 cur_isize = i_size_read(inode);
1294 if (cur_isize != new_size) {
1295 #ifdef NFS_DEBUG_VERBOSE
1296 printk(KERN_DEBUG "NFS: isize change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1299 * If we have pending writebacks, things can get
1302 if (S_ISREG(inode->i_mode) && data_unstable) {
1303 if (new_isize > cur_isize) {
1304 inode->i_size = new_isize;
1305 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1308 inode->i_size = new_isize;
1309 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1314 * Note: we don't check inode->i_mtime since pipes etc.
1315 * can change this value in VFS without requiring a
1316 * cache revalidation.
1318 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1319 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1320 #ifdef NFS_DEBUG_VERBOSE
1321 printk(KERN_DEBUG "NFS: mtime change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1324 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1327 if ((fattr->valid & NFS_ATTR_FATTR_V4)
1328 && nfsi->change_attr != fattr->change_attr) {
1329 #ifdef NFS_DEBUG_VERBOSE
1330 printk(KERN_DEBUG "NFS: change_attr change on %s/%ld\n",
1331 inode->i_sb->s_id, inode->i_ino);
1333 nfsi->change_attr = fattr->change_attr;
1335 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1338 /* If ctime has changed we should definitely clear access+acl caches */
1339 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1341 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1342 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1344 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1346 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1347 inode->i_uid != fattr->uid ||
1348 inode->i_gid != fattr->gid)
1349 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1351 inode->i_mode = fattr->mode;
1352 inode->i_nlink = fattr->nlink;
1353 inode->i_uid = fattr->uid;
1354 inode->i_gid = fattr->gid;
1356 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1358 * report the blocks in 512byte units
1360 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1361 inode->i_blksize = inode->i_sb->s_blocksize;
1363 inode->i_blocks = fattr->du.nfs2.blocks;
1364 inode->i_blksize = fattr->du.nfs2.blocksize;
1367 /* Update attrtimeo value if we're out of the unstable period */
1368 if (invalid & NFS_INO_INVALID_ATTR) {
1369 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1370 nfsi->attrtimeo_timestamp = jiffies;
1371 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1372 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1373 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1374 nfsi->attrtimeo_timestamp = jiffies;
1376 /* Don't invalidate the data if we were to blame */
1377 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1378 || S_ISLNK(inode->i_mode)))
1379 invalid &= ~NFS_INO_INVALID_DATA;
1380 if (!nfs_have_delegation(inode, FMODE_READ))
1381 nfsi->flags |= invalid;
1386 * Big trouble! The inode has become a different object.
1389 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1390 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1393 * No need to worry about unhashing the dentry, as the
1394 * lookup validation will know that the inode is bad.
1395 * (But we fall through to invalidate the caches.)
1397 nfs_invalidate_inode(inode);
1399 NFS_FLAGS(inode) |= NFS_INO_STALE;
1404 * File system information
1407 static int nfs_set_super(struct super_block *s, void *data)
1409 s->s_fs_info = data;
1410 return set_anon_super(s, data);
1413 static int nfs_compare_super(struct super_block *sb, void *data)
1415 struct nfs_server *server = data;
1416 struct nfs_server *old = NFS_SB(sb);
1418 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1420 if (old->addr.sin_port != server->addr.sin_port)
1422 return !nfs_compare_fh(&old->fh, &server->fh);
1425 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1426 int flags, const char *dev_name, void *raw_data)
1429 struct nfs_server *server = NULL;
1430 struct super_block *s;
1431 struct nfs_fh *root;
1432 struct nfs_mount_data *data = raw_data;
1434 s = ERR_PTR(-EINVAL);
1436 dprintk("%s: missing data argument\n", __FUNCTION__);
1439 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1440 dprintk("%s: bad mount version\n", __FUNCTION__);
1443 switch (data->version) {
1449 if (data->flags & NFS_MOUNT_VER3) {
1450 dprintk("%s: mount structure version %d does not support NFSv3\n",
1455 data->root.size = NFS2_FHSIZE;
1456 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1458 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1459 dprintk("%s: mount structure version %d does not support strong security\n",
1465 memset(data->context, 0, sizeof(data->context));
1467 #ifndef CONFIG_NFS_V3
1468 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1469 s = ERR_PTR(-EPROTONOSUPPORT);
1470 if (data->flags & NFS_MOUNT_VER3) {
1471 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1474 #endif /* CONFIG_NFS_V3 */
1476 s = ERR_PTR(-ENOMEM);
1477 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1480 memset(server, 0, sizeof(struct nfs_server));
1481 /* Zero out the NFS state stuff */
1482 init_nfsv4_state(server);
1483 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1486 if (data->flags & NFS_MOUNT_VER3)
1487 root->size = data->root.size;
1489 root->size = NFS2_FHSIZE;
1490 s = ERR_PTR(-EINVAL);
1491 if (root->size > sizeof(root->data)) {
1492 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1495 memcpy(root->data, data->root.data, root->size);
1497 /* We now require that the mount process passes the remote address */
1498 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1499 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1500 dprintk("%s: mount program didn't pass remote address!\n",
1505 /* Fire up rpciod if not yet running */
1506 s = ERR_PTR(rpciod_up());
1508 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1509 __FUNCTION__, PTR_ERR(s));
1513 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1514 if (IS_ERR(s) || s->s_root)
1515 goto out_rpciod_down;
1519 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1521 up_write(&s->s_umount);
1522 deactivate_super(s);
1523 return ERR_PTR(error);
1525 s->s_flags |= MS_ACTIVE;
1534 static void nfs_kill_super(struct super_block *s)
1536 struct nfs_server *server = NFS_SB(s);
1540 if (!IS_ERR(server->client))
1541 rpc_shutdown_client(server->client);
1542 if (!IS_ERR(server->client_sys))
1543 rpc_shutdown_client(server->client_sys);
1544 if (!IS_ERR(server->client_acl))
1545 rpc_shutdown_client(server->client_acl);
1547 if (!(server->flags & NFS_MOUNT_NONLM))
1548 lockd_down(); /* release rpc.lockd */
1550 rpciod_down(); /* release rpciod */
1552 if (server->hostname != NULL)
1553 kfree(server->hostname);
1557 static struct file_system_type nfs_fs_type = {
1558 .owner = THIS_MODULE,
1560 .get_sb = nfs_get_sb,
1561 .kill_sb = nfs_kill_super,
1562 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1565 #ifdef CONFIG_NFS_V4
1567 static void nfs4_clear_inode(struct inode *);
1570 static struct super_operations nfs4_sops = {
1571 .alloc_inode = nfs_alloc_inode,
1572 .destroy_inode = nfs_destroy_inode,
1573 .write_inode = nfs_write_inode,
1574 .delete_inode = nfs_delete_inode,
1575 .statfs = nfs_statfs,
1576 .clear_inode = nfs4_clear_inode,
1577 .umount_begin = nfs_umount_begin,
1578 .show_options = nfs_show_options,
1582 * Clean out any remaining NFSv4 state that might be left over due
1583 * to open() calls that passed nfs_atomic_lookup, but failed to call
1586 static void nfs4_clear_inode(struct inode *inode)
1588 struct nfs_inode *nfsi = NFS_I(inode);
1590 /* If we are holding a delegation, return it! */
1591 if (nfsi->delegation != NULL)
1592 nfs_inode_return_delegation(inode);
1593 /* First call standard NFS clear_inode() code */
1594 nfs_clear_inode(inode);
1595 /* Now clear out any remaining state */
1596 while (!list_empty(&nfsi->open_states)) {
1597 struct nfs4_state *state;
1599 state = list_entry(nfsi->open_states.next,
1602 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1605 (long long)NFS_FILEID(inode),
1607 BUG_ON(atomic_read(&state->count) != 1);
1608 nfs4_close_state(state, state->state);
1613 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1615 struct nfs_server *server;
1616 struct nfs4_client *clp = NULL;
1617 struct rpc_xprt *xprt = NULL;
1618 struct rpc_clnt *clnt = NULL;
1619 struct rpc_timeout timeparms;
1620 rpc_authflavor_t authflavour;
1621 int proto, err = -EIO;
1623 sb->s_blocksize_bits = 0;
1624 sb->s_blocksize = 0;
1625 server = NFS_SB(sb);
1626 if (data->rsize != 0)
1627 server->rsize = nfs_block_size(data->rsize, NULL);
1628 if (data->wsize != 0)
1629 server->wsize = nfs_block_size(data->wsize, NULL);
1630 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1631 server->caps = NFS_CAP_ATOMIC_OPEN;
1633 server->acregmin = data->acregmin*HZ;
1634 server->acregmax = data->acregmax*HZ;
1635 server->acdirmin = data->acdirmin*HZ;
1636 server->acdirmax = data->acdirmax*HZ;
1638 server->rpc_ops = &nfs_v4_clientops;
1639 /* Initialize timeout values */
1641 timeparms.to_initval = data->timeo * HZ / 10;
1642 timeparms.to_retries = data->retrans;
1643 timeparms.to_exponential = 1;
1644 if (!timeparms.to_retries)
1645 timeparms.to_retries = 5;
1647 proto = data->proto;
1648 /* Which IP protocol do we use? */
1651 timeparms.to_maxval = RPC_MAX_TCP_TIMEOUT;
1652 if (!timeparms.to_initval)
1653 timeparms.to_initval = 600 * HZ / 10;
1656 timeparms.to_maxval = RPC_MAX_UDP_TIMEOUT;
1657 if (!timeparms.to_initval)
1658 timeparms.to_initval = 11 * HZ / 10;
1664 clp = nfs4_get_client(&server->addr.sin_addr);
1666 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1670 /* Now create transport and client */
1671 authflavour = RPC_AUTH_UNIX;
1672 if (data->auth_flavourlen != 0) {
1673 if (data->auth_flavourlen != 1) {
1674 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1675 __FUNCTION__, data->auth_flavourlen);
1679 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1685 down_write(&clp->cl_sem);
1686 if (IS_ERR(clp->cl_rpcclient)) {
1687 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1689 up_write(&clp->cl_sem);
1690 err = PTR_ERR(xprt);
1691 dprintk("%s: cannot create RPC transport. Error = %d\n",
1695 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1696 server->rpc_ops->version, authflavour);
1698 up_write(&clp->cl_sem);
1699 err = PTR_ERR(clnt);
1700 dprintk("%s: cannot create RPC client. Error = %d\n",
1705 clnt->cl_softrtry = 1;
1706 clnt->cl_chatty = 1;
1707 clp->cl_rpcclient = clnt;
1708 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1709 if (IS_ERR(clp->cl_cred)) {
1710 up_write(&clp->cl_sem);
1711 err = PTR_ERR(clp->cl_cred);
1712 clp->cl_cred = NULL;
1715 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1718 if (list_empty(&clp->cl_superblocks)) {
1719 err = nfs4_init_client(clp);
1721 up_write(&clp->cl_sem);
1725 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1726 clnt = rpc_clone_client(clp->cl_rpcclient);
1728 server->nfs4_state = clp;
1729 up_write(&clp->cl_sem);
1733 err = PTR_ERR(clnt);
1734 dprintk("%s: cannot create RPC client. Error = %d\n",
1739 server->client = clnt;
1741 if (server->nfs4_state->cl_idmap == NULL) {
1742 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1746 if (clnt->cl_auth->au_flavor != authflavour) {
1747 struct rpc_auth *auth;
1749 auth = rpcauth_create(authflavour, clnt);
1751 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1752 return PTR_ERR(auth);
1756 sb->s_time_gran = 1;
1758 sb->s_op = &nfs4_sops;
1759 err = nfs_sb_init(sb, authflavour);
1764 nfs4_put_client(clp);
1768 static int nfs4_compare_super(struct super_block *sb, void *data)
1770 struct nfs_server *server = data;
1771 struct nfs_server *old = NFS_SB(sb);
1773 if (strcmp(server->hostname, old->hostname) != 0)
1775 if (strcmp(server->mnt_path, old->mnt_path) != 0)
1781 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1786 return ERR_PTR(-EINVAL);
1787 if (src->len < maxlen)
1790 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1792 return ERR_PTR(-ENOMEM);
1794 if (copy_from_user(dst, src->data, maxlen)) {
1797 return ERR_PTR(-EFAULT);
1803 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1804 int flags, const char *dev_name, void *raw_data)
1807 struct nfs_server *server;
1808 struct super_block *s;
1809 struct nfs4_mount_data *data = raw_data;
1813 dprintk("%s: missing data argument\n", __FUNCTION__);
1814 return ERR_PTR(-EINVAL);
1816 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1817 dprintk("%s: bad mount version\n", __FUNCTION__);
1818 return ERR_PTR(-EINVAL);
1821 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1823 return ERR_PTR(-ENOMEM);
1824 memset(server, 0, sizeof(struct nfs_server));
1825 /* Zero out the NFS state stuff */
1826 init_nfsv4_state(server);
1827 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1829 p = nfs_copy_user_string(NULL, &data->hostname, 256);
1832 server->hostname = p;
1834 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1837 server->mnt_path = p;
1839 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1840 sizeof(server->ip_addr) - 1);
1844 /* We now require that the mount process passes the remote address */
1845 if (data->host_addrlen != sizeof(server->addr)) {
1846 s = ERR_PTR(-EINVAL);
1849 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1850 s = ERR_PTR(-EFAULT);
1853 if (server->addr.sin_family != AF_INET ||
1854 server->addr.sin_addr.s_addr == INADDR_ANY) {
1855 dprintk("%s: mount program didn't pass remote IP address!\n",
1857 s = ERR_PTR(-EINVAL);
1861 /* Fire up rpciod if not yet running */
1862 s = ERR_PTR(rpciod_up());
1864 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1865 __FUNCTION__, PTR_ERR(s));
1869 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1871 if (IS_ERR(s) || s->s_root)
1876 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1878 up_write(&s->s_umount);
1879 deactivate_super(s);
1880 return ERR_PTR(error);
1882 s->s_flags |= MS_ACTIVE;
1885 s = (struct super_block *)p;
1887 if (server->mnt_path)
1888 kfree(server->mnt_path);
1889 if (server->hostname)
1890 kfree(server->hostname);
1895 static void nfs4_kill_super(struct super_block *sb)
1897 struct nfs_server *server = NFS_SB(sb);
1899 nfs_return_all_delegations(sb);
1900 kill_anon_super(sb);
1902 nfs4_renewd_prepare_shutdown(server);
1904 if (server->client != NULL && !IS_ERR(server->client))
1905 rpc_shutdown_client(server->client);
1906 rpciod_down(); /* release rpciod */
1908 destroy_nfsv4_state(server);
1910 if (server->hostname != NULL)
1911 kfree(server->hostname);
1915 static struct file_system_type nfs4_fs_type = {
1916 .owner = THIS_MODULE,
1918 .get_sb = nfs4_get_sb,
1919 .kill_sb = nfs4_kill_super,
1920 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1923 #define nfs4_init_once(nfsi) \
1925 INIT_LIST_HEAD(&(nfsi)->open_states); \
1926 nfsi->delegation = NULL; \
1927 nfsi->delegation_state = 0; \
1928 init_rwsem(&nfsi->rwsem); \
1930 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1931 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1933 #define nfs4_init_once(nfsi) \
1935 #define register_nfs4fs() (0)
1936 #define unregister_nfs4fs()
1939 extern int nfs_init_nfspagecache(void);
1940 extern void nfs_destroy_nfspagecache(void);
1941 extern int nfs_init_readpagecache(void);
1942 extern void nfs_destroy_readpagecache(void);
1943 extern int nfs_init_writepagecache(void);
1944 extern void nfs_destroy_writepagecache(void);
1945 #ifdef CONFIG_NFS_DIRECTIO
1946 extern int nfs_init_directcache(void);
1947 extern void nfs_destroy_directcache(void);
1950 static kmem_cache_t * nfs_inode_cachep;
1952 static struct inode *nfs_alloc_inode(struct super_block *sb)
1954 struct nfs_inode *nfsi;
1955 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
1959 #ifdef CONFIG_NFS_V4
1960 nfsi->nfs4_acl = NULL;
1961 #endif /* CONFIG_NFS_V4 */
1962 return &nfsi->vfs_inode;
1965 static void nfs_destroy_inode(struct inode *inode)
1967 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1970 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
1972 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1974 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1975 SLAB_CTOR_CONSTRUCTOR) {
1976 inode_init_once(&nfsi->vfs_inode);
1977 spin_lock_init(&nfsi->req_lock);
1978 INIT_LIST_HEAD(&nfsi->dirty);
1979 INIT_LIST_HEAD(&nfsi->commit);
1980 INIT_LIST_HEAD(&nfsi->open_files);
1981 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1982 atomic_set(&nfsi->data_updates, 0);
1986 init_waitqueue_head(&nfsi->nfs_i_wait);
1987 nfs4_init_once(nfsi);
1991 static int nfs_init_inodecache(void)
1993 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1994 sizeof(struct nfs_inode),
1995 0, SLAB_RECLAIM_ACCOUNT,
1997 if (nfs_inode_cachep == NULL)
2003 static void nfs_destroy_inodecache(void)
2005 if (kmem_cache_destroy(nfs_inode_cachep))
2006 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2012 static int __init init_nfs_fs(void)
2016 err = nfs_init_nfspagecache();
2020 err = nfs_init_inodecache();
2024 err = nfs_init_readpagecache();
2028 err = nfs_init_writepagecache();
2032 #ifdef CONFIG_NFS_DIRECTIO
2033 err = nfs_init_directcache();
2038 #ifdef CONFIG_PROC_FS
2039 rpc_proc_register(&nfs_rpcstat);
2041 err = register_filesystem(&nfs_fs_type);
2044 if ((err = register_nfs4fs()) != 0)
2048 #ifdef CONFIG_PROC_FS
2049 rpc_proc_unregister("nfs");
2051 nfs_destroy_writepagecache();
2052 #ifdef CONFIG_NFS_DIRECTIO
2054 nfs_destroy_directcache();
2057 nfs_destroy_readpagecache();
2059 nfs_destroy_inodecache();
2061 nfs_destroy_nfspagecache();
2066 static void __exit exit_nfs_fs(void)
2068 #ifdef CONFIG_NFS_DIRECTIO
2069 nfs_destroy_directcache();
2071 nfs_destroy_writepagecache();
2072 nfs_destroy_readpagecache();
2073 nfs_destroy_inodecache();
2074 nfs_destroy_nfspagecache();
2075 #ifdef CONFIG_PROC_FS
2076 rpc_proc_unregister("nfs");
2078 unregister_filesystem(&nfs_fs_type);
2079 unregister_nfs4fs();
2082 /* Not quite true; I just maintain it */
2083 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2084 MODULE_LICENSE("GPL");
2086 module_init(init_nfs_fs)
2087 module_exit(exit_nfs_fs)