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 */
494 * The VFS shouldn't apply the umask to mode bits. We will
495 * do so ourselves when necessary.
497 sb->s_flags |= MS_POSIXACL;
498 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
499 server->namelen = NFS3_MAXNAMLEN;
502 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
503 server->namelen = NFS2_MAXNAMLEN;
506 sb->s_op = &nfs_sops;
507 return nfs_sb_init(sb, authflavor);
511 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
513 struct nfs_server *server = NFS_SB(sb);
514 unsigned char blockbits;
515 unsigned long blockres;
516 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
517 struct nfs_fattr fattr;
518 struct nfs_fsstat res = {
525 error = server->rpc_ops->statfs(server, rootfh, &res);
526 buf->f_type = NFS_SUPER_MAGIC;
531 * Current versions of glibc do not correctly handle the
532 * case where f_frsize != f_bsize. Eventually we want to
533 * report the value of wtmult in this field.
535 buf->f_frsize = sb->s_blocksize;
538 * On most *nix systems, f_blocks, f_bfree, and f_bavail
539 * are reported in units of f_frsize. Linux hasn't had
540 * an f_frsize field in its statfs struct until recently,
541 * thus historically Linux's sys_statfs reports these
542 * fields in units of f_bsize.
544 buf->f_bsize = sb->s_blocksize;
545 blockbits = sb->s_blocksize_bits;
546 blockres = (1 << blockbits) - 1;
547 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
548 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
549 buf->f_bavail = (res.abytes + blockres) >> blockbits;
551 buf->f_files = res.tfiles;
552 buf->f_ffree = res.afiles;
554 buf->f_namelen = server->namelen;
561 printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
562 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
567 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
569 static struct proc_nfs_info {
574 { NFS_MOUNT_SOFT, ",soft", ",hard" },
575 { NFS_MOUNT_INTR, ",intr", "" },
576 { NFS_MOUNT_POSIX, ",posix", "" },
577 { NFS_MOUNT_TCP, ",tcp", ",udp" },
578 { NFS_MOUNT_NOCTO, ",nocto", "" },
579 { NFS_MOUNT_NOAC, ",noac", "" },
580 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
581 { NFS_MOUNT_NOACL, ",noacl", "" },
584 struct proc_nfs_info *nfs_infop;
585 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
587 seq_printf(m, ",v%d", nfss->rpc_ops->version);
588 seq_printf(m, ",rsize=%d", nfss->rsize);
589 seq_printf(m, ",wsize=%d", nfss->wsize);
590 if (nfss->acregmin != 3*HZ)
591 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
592 if (nfss->acregmax != 60*HZ)
593 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
594 if (nfss->acdirmin != 30*HZ)
595 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
596 if (nfss->acdirmax != 60*HZ)
597 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
598 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
599 if (nfss->flags & nfs_infop->flag)
600 seq_puts(m, nfs_infop->str);
602 seq_puts(m, nfs_infop->nostr);
604 seq_puts(m, ",addr=");
605 seq_escape(m, nfss->hostname, " \t\n\\");
610 * Invalidate the local caches
613 nfs_zap_caches(struct inode *inode)
615 struct nfs_inode *nfsi = NFS_I(inode);
616 int mode = inode->i_mode;
618 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
619 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
621 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
622 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
623 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
625 nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
628 static void nfs_zap_acl_cache(struct inode *inode)
630 void (*clear_acl_cache)(struct inode *);
632 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
633 if (clear_acl_cache != NULL)
634 clear_acl_cache(inode);
635 NFS_I(inode)->flags &= ~NFS_INO_INVALID_ACL;
639 * Invalidate, but do not unhash, the inode
642 nfs_invalidate_inode(struct inode *inode)
644 umode_t save_mode = inode->i_mode;
646 make_bad_inode(inode);
647 inode->i_mode = save_mode;
648 nfs_zap_caches(inode);
651 struct nfs_find_desc {
653 struct nfs_fattr *fattr;
657 * In NFSv3 we can have 64bit inode numbers. In order to support
658 * this, and re-exported directories (also seen in NFSv2)
659 * we are forced to allow 2 different inodes to have the same
663 nfs_find_actor(struct inode *inode, void *opaque)
665 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
666 struct nfs_fh *fh = desc->fh;
667 struct nfs_fattr *fattr = desc->fattr;
669 if (NFS_FILEID(inode) != fattr->fileid)
671 if (nfs_compare_fh(NFS_FH(inode), fh))
673 if (is_bad_inode(inode) || NFS_STALE(inode))
679 nfs_init_locked(struct inode *inode, void *opaque)
681 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
682 struct nfs_fattr *fattr = desc->fattr;
684 NFS_FILEID(inode) = fattr->fileid;
685 nfs_copy_fh(NFS_FH(inode), desc->fh);
689 /* Don't use READDIRPLUS on directories that we believe are too large */
690 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
693 * This is our front-end to iget that looks up inodes by file handle
694 * instead of inode number.
697 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
699 struct nfs_find_desc desc = {
703 struct inode *inode = NULL;
706 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
710 printk("NFS: Buggy server - nlink == 0!\n");
714 hash = nfs_fattr_to_ino_t(fattr);
716 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
719 if (inode->i_state & I_NEW) {
720 struct nfs_inode *nfsi = NFS_I(inode);
722 /* We set i_ino for the few things that still rely on it,
726 /* We can't support update_atime(), since the server will reset it */
727 inode->i_flags |= S_NOATIME|S_NOCMTIME;
728 inode->i_mode = fattr->mode;
729 /* Why so? Because we want revalidate for devices/FIFOs, and
730 * that's precisely what we have in nfs_file_inode_operations.
732 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
733 if (S_ISREG(inode->i_mode)) {
734 inode->i_fop = &nfs_file_operations;
735 inode->i_data.a_ops = &nfs_file_aops;
736 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
737 } else if (S_ISDIR(inode->i_mode)) {
738 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
739 inode->i_fop = &nfs_dir_operations;
740 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
741 && fattr->size <= NFS_LIMIT_READDIRPLUS)
742 NFS_FLAGS(inode) |= NFS_INO_ADVISE_RDPLUS;
743 } else if (S_ISLNK(inode->i_mode))
744 inode->i_op = &nfs_symlink_inode_operations;
746 init_special_inode(inode, inode->i_mode, fattr->rdev);
748 nfsi->read_cache_jiffies = fattr->timestamp;
749 inode->i_atime = fattr->atime;
750 inode->i_mtime = fattr->mtime;
751 inode->i_ctime = fattr->ctime;
752 if (fattr->valid & NFS_ATTR_FATTR_V4)
753 nfsi->change_attr = fattr->change_attr;
754 inode->i_size = nfs_size_to_loff_t(fattr->size);
755 inode->i_nlink = fattr->nlink;
756 inode->i_uid = fattr->uid;
757 inode->i_gid = fattr->gid;
758 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
760 * report the blocks in 512byte units
762 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
763 inode->i_blksize = inode->i_sb->s_blocksize;
765 inode->i_blocks = fattr->du.nfs2.blocks;
766 inode->i_blksize = fattr->du.nfs2.blocksize;
768 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
769 nfsi->attrtimeo_timestamp = jiffies;
770 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
771 nfsi->cache_access.cred = NULL;
773 unlock_new_inode(inode);
775 nfs_refresh_inode(inode, fattr);
776 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
778 (long long)NFS_FILEID(inode),
779 atomic_read(&inode->i_count));
785 printk("nfs_fhget: iget failed\n");
789 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
792 nfs_setattr(struct dentry *dentry, struct iattr *attr)
794 struct inode *inode = dentry->d_inode;
795 struct nfs_fattr fattr;
798 if (attr->ia_valid & ATTR_SIZE) {
799 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
800 attr->ia_valid &= ~ATTR_SIZE;
803 /* Optimization: if the end result is no change, don't RPC */
804 attr->ia_valid &= NFS_VALID_ATTRS;
805 if (attr->ia_valid == 0)
809 nfs_begin_data_update(inode);
810 /* Write all dirty data if we're changing file permissions or size */
811 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
812 if (filemap_fdatawrite(inode->i_mapping) == 0)
813 filemap_fdatawait(inode->i_mapping);
816 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
818 nfs_refresh_inode(inode, &fattr);
819 if ((attr->ia_valid & ATTR_MODE) != 0) {
821 mode = inode->i_mode & ~S_IALLUGO;
822 mode |= attr->ia_mode & S_IALLUGO;
823 inode->i_mode = mode;
825 if ((attr->ia_valid & ATTR_UID) != 0)
826 inode->i_uid = attr->ia_uid;
827 if ((attr->ia_valid & ATTR_GID) != 0)
828 inode->i_gid = attr->ia_gid;
829 if ((attr->ia_valid & ATTR_SIZE) != 0) {
830 inode->i_size = attr->ia_size;
831 vmtruncate(inode, attr->ia_size);
834 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
835 NFS_FLAGS(inode) |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
836 nfs_end_data_update(inode);
842 * Wait for the inode to get unlocked.
843 * (Used for NFS_INO_LOCKED and NFS_INO_REVALIDATING).
846 nfs_wait_on_inode(struct inode *inode, int flag)
848 struct rpc_clnt *clnt = NFS_CLIENT(inode);
849 struct nfs_inode *nfsi = NFS_I(inode);
852 if (!(NFS_FLAGS(inode) & flag))
854 atomic_inc(&inode->i_count);
855 error = nfs_wait_event(clnt, nfsi->nfs_i_wait,
856 !(NFS_FLAGS(inode) & flag));
861 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
863 struct inode *inode = dentry->d_inode;
864 struct nfs_inode *nfsi = NFS_I(inode);
865 int need_atime = nfsi->flags & NFS_INO_INVALID_ATIME;
868 if (__IS_FLG(inode, MS_NOATIME))
870 else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
872 /* We may force a getattr if the user cares about atime */
874 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
876 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
878 generic_fillattr(inode, stat);
882 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
884 struct nfs_open_context *ctx;
886 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
888 atomic_set(&ctx->count, 1);
889 ctx->dentry = dget(dentry);
890 ctx->cred = get_rpccred(cred);
892 ctx->lockowner = current->files;
898 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
901 atomic_inc(&ctx->count);
905 void put_nfs_open_context(struct nfs_open_context *ctx)
907 if (atomic_dec_and_test(&ctx->count)) {
908 if (!list_empty(&ctx->list)) {
909 struct inode *inode = ctx->dentry->d_inode;
910 spin_lock(&inode->i_lock);
911 list_del(&ctx->list);
912 spin_unlock(&inode->i_lock);
914 if (ctx->state != NULL)
915 nfs4_close_state(ctx->state, ctx->mode);
916 if (ctx->cred != NULL)
917 put_rpccred(ctx->cred);
924 * Ensure that mmap has a recent RPC credential for use when writing out
927 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
929 struct inode *inode = filp->f_dentry->d_inode;
930 struct nfs_inode *nfsi = NFS_I(inode);
932 filp->private_data = get_nfs_open_context(ctx);
933 spin_lock(&inode->i_lock);
934 list_add(&ctx->list, &nfsi->open_files);
935 spin_unlock(&inode->i_lock);
938 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
940 struct nfs_inode *nfsi = NFS_I(inode);
941 struct nfs_open_context *pos, *ctx = NULL;
943 spin_lock(&inode->i_lock);
944 list_for_each_entry(pos, &nfsi->open_files, list) {
945 if ((pos->mode & mode) == mode) {
946 ctx = get_nfs_open_context(pos);
950 spin_unlock(&inode->i_lock);
954 void nfs_file_clear_open_context(struct file *filp)
956 struct inode *inode = filp->f_dentry->d_inode;
957 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
960 filp->private_data = NULL;
961 spin_lock(&inode->i_lock);
962 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
963 spin_unlock(&inode->i_lock);
964 put_nfs_open_context(ctx);
969 * These allocate and release file read/write context information.
971 int nfs_open(struct inode *inode, struct file *filp)
973 struct nfs_open_context *ctx;
974 struct rpc_cred *cred;
976 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
978 return PTR_ERR(cred);
979 ctx = alloc_nfs_open_context(filp->f_dentry, cred);
983 ctx->mode = filp->f_mode;
984 nfs_file_set_open_context(filp, ctx);
985 put_nfs_open_context(ctx);
986 if ((filp->f_mode & FMODE_WRITE) != 0)
987 nfs_begin_data_update(inode);
991 int nfs_release(struct inode *inode, struct file *filp)
993 if ((filp->f_mode & FMODE_WRITE) != 0)
994 nfs_end_data_update(inode);
995 nfs_file_clear_open_context(filp);
1000 * This function is called whenever some part of NFS notices that
1001 * the cached attributes have to be refreshed.
1004 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1006 int status = -ESTALE;
1007 struct nfs_fattr fattr;
1008 struct nfs_inode *nfsi = NFS_I(inode);
1009 unsigned long verifier;
1012 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1013 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1016 if (!inode || is_bad_inode(inode))
1018 if (NFS_STALE(inode))
1021 while (NFS_REVALIDATING(inode)) {
1022 status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
1025 if (NFS_ATTRTIMEO(inode) == 0)
1027 if (NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME))
1029 status = NFS_STALE(inode) ? -ESTALE : 0;
1032 NFS_FLAGS(inode) |= NFS_INO_REVALIDATING;
1034 /* Protect against RPC races by saving the change attribute */
1035 verifier = nfs_save_change_attribute(inode);
1036 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1038 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1040 (long long)NFS_FILEID(inode), status);
1041 if (status == -ESTALE) {
1042 nfs_zap_caches(inode);
1043 if (!S_ISDIR(inode->i_mode))
1044 NFS_FLAGS(inode) |= NFS_INO_STALE;
1049 status = nfs_update_inode(inode, &fattr, verifier);
1051 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1053 (long long)NFS_FILEID(inode), status);
1056 flags = nfsi->flags;
1058 * We may need to keep the attributes marked as invalid if
1059 * we raced with nfs_end_attr_update().
1061 if (verifier == nfsi->cache_change_attribute)
1062 nfsi->flags &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1063 /* Do the page cache invalidation */
1064 if (flags & NFS_INO_INVALID_DATA) {
1065 if (S_ISREG(inode->i_mode)) {
1066 if (filemap_fdatawrite(inode->i_mapping) == 0)
1067 filemap_fdatawait(inode->i_mapping);
1070 nfsi->flags &= ~NFS_INO_INVALID_DATA;
1071 invalidate_inode_pages2(inode->i_mapping);
1072 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
1073 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1075 (long long)NFS_FILEID(inode));
1076 /* This ensures we revalidate dentries */
1077 nfsi->cache_change_attribute++;
1079 if (flags & NFS_INO_INVALID_ACL)
1080 nfs_zap_acl_cache(inode);
1081 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1083 (long long)NFS_FILEID(inode));
1086 NFS_FLAGS(inode) &= ~NFS_INO_REVALIDATING;
1087 wake_up(&nfsi->nfs_i_wait);
1093 int nfs_attribute_timeout(struct inode *inode)
1095 struct nfs_inode *nfsi = NFS_I(inode);
1097 if (nfs_have_delegation(inode, FMODE_READ))
1099 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1103 * nfs_revalidate_inode - Revalidate the inode attributes
1104 * @server - pointer to nfs_server struct
1105 * @inode - pointer to inode struct
1107 * Updates inode attribute information by retrieving the data from the server.
1109 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1111 if (!(NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1112 && !nfs_attribute_timeout(inode))
1113 return NFS_STALE(inode) ? -ESTALE : 0;
1114 return __nfs_revalidate_inode(server, inode);
1118 * nfs_begin_data_update
1119 * @inode - pointer to inode
1120 * Declare that a set of operations will update file data on the server
1122 void nfs_begin_data_update(struct inode *inode)
1124 atomic_inc(&NFS_I(inode)->data_updates);
1128 * nfs_end_data_update
1129 * @inode - pointer to inode
1130 * Declare end of the operations that will update file data
1131 * This will mark the inode as immediately needing revalidation
1132 * of its attribute cache.
1134 void nfs_end_data_update(struct inode *inode)
1136 struct nfs_inode *nfsi = NFS_I(inode);
1138 if (!nfs_have_delegation(inode, FMODE_READ)) {
1139 /* Mark the attribute cache for revalidation */
1140 nfsi->flags |= NFS_INO_INVALID_ATTR;
1141 /* Directories and symlinks: invalidate page cache too */
1142 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1143 nfsi->flags |= NFS_INO_INVALID_DATA;
1145 nfsi->cache_change_attribute ++;
1146 atomic_dec(&nfsi->data_updates);
1150 * nfs_end_data_update_defer
1151 * @inode - pointer to inode
1152 * Declare end of the operations that will update file data
1153 * This will defer marking the inode as needing revalidation
1154 * unless there are no other pending updates.
1156 void nfs_end_data_update_defer(struct inode *inode)
1158 struct nfs_inode *nfsi = NFS_I(inode);
1160 if (atomic_dec_and_test(&nfsi->data_updates)) {
1161 /* Mark the attribute cache for revalidation */
1162 nfsi->flags |= NFS_INO_INVALID_ATTR;
1163 /* Directories and symlinks: invalidate page cache too */
1164 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1165 nfsi->flags |= NFS_INO_INVALID_DATA;
1166 nfsi->cache_change_attribute ++;
1171 * nfs_refresh_inode - verify consistency of the inode attribute cache
1172 * @inode - pointer to inode
1173 * @fattr - updated attributes
1175 * Verifies the attribute cache. If we have just changed the attributes,
1176 * so that fattr carries weak cache consistency data, then it may
1177 * also update the ctime/mtime/change_attribute.
1179 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1181 struct nfs_inode *nfsi = NFS_I(inode);
1182 loff_t cur_size, new_isize;
1185 /* Do we hold a delegation? */
1186 if (nfs_have_delegation(inode, FMODE_READ))
1189 /* Are we in the process of updating data on the server? */
1190 data_unstable = nfs_caches_unstable(inode);
1192 if (fattr->valid & NFS_ATTR_FATTR_V4) {
1193 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1194 && nfsi->change_attr == fattr->pre_change_attr)
1195 nfsi->change_attr = fattr->change_attr;
1196 if (!data_unstable && nfsi->change_attr != fattr->change_attr)
1197 nfsi->flags |= NFS_INO_INVALID_ATTR;
1200 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1203 /* Has the inode gone and changed behind our back? */
1204 if (nfsi->fileid != fattr->fileid
1205 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1208 cur_size = i_size_read(inode);
1209 new_isize = nfs_size_to_loff_t(fattr->size);
1211 /* If we have atomic WCC data, we may update some attributes */
1212 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1213 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1214 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1215 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1216 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1219 /* Verify a few of the more important attributes */
1220 if (!data_unstable) {
1221 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)
1222 || cur_size != new_isize)
1223 nfsi->flags |= NFS_INO_INVALID_ATTR;
1224 } else if (S_ISREG(inode->i_mode) && new_isize > cur_size)
1225 nfsi->flags |= NFS_INO_INVALID_ATTR;
1227 /* Have any file permissions changed? */
1228 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1229 || inode->i_uid != fattr->uid
1230 || inode->i_gid != fattr->gid)
1231 nfsi->flags |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1233 /* Has the link count changed? */
1234 if (inode->i_nlink != fattr->nlink)
1235 nfsi->flags |= NFS_INO_INVALID_ATTR;
1237 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1238 nfsi->flags |= NFS_INO_INVALID_ATIME;
1240 nfsi->read_cache_jiffies = fattr->timestamp;
1245 * Many nfs protocol calls return the new file attributes after
1246 * an operation. Here we update the inode to reflect the state
1247 * of the server's inode.
1249 * This is a bit tricky because we have to make sure all dirty pages
1250 * have been sent off to the server before calling invalidate_inode_pages.
1251 * To make sure no other process adds more write requests while we try
1252 * our best to flush them, we make them sleep during the attribute refresh.
1254 * A very similar scenario holds for the dir cache.
1256 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1258 struct nfs_inode *nfsi = NFS_I(inode);
1261 unsigned int invalid = 0;
1265 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1266 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1267 atomic_read(&inode->i_count), fattr->valid);
1269 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1272 if (nfsi->fileid != fattr->fileid) {
1273 printk(KERN_ERR "%s: inode number mismatch\n"
1274 "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1276 inode->i_sb->s_id, (long long)nfsi->fileid,
1277 inode->i_sb->s_id, (long long)fattr->fileid);
1282 * Make sure the inode's type hasn't changed.
1284 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1288 * Update the read time so we don't revalidate too often.
1290 nfsi->read_cache_jiffies = fattr->timestamp;
1292 /* Are we racing with known updates of the metadata on the server? */
1293 data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1295 /* Check if the file size agrees */
1296 new_size = fattr->size;
1297 new_isize = nfs_size_to_loff_t(fattr->size);
1298 cur_isize = i_size_read(inode);
1299 if (cur_isize != new_size) {
1300 #ifdef NFS_DEBUG_VERBOSE
1301 printk(KERN_DEBUG "NFS: isize change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1304 * If we have pending writebacks, things can get
1307 if (S_ISREG(inode->i_mode) && data_unstable) {
1308 if (new_isize > cur_isize) {
1309 inode->i_size = new_isize;
1310 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1313 inode->i_size = new_isize;
1314 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1319 * Note: we don't check inode->i_mtime since pipes etc.
1320 * can change this value in VFS without requiring a
1321 * cache revalidation.
1323 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1324 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1325 #ifdef NFS_DEBUG_VERBOSE
1326 printk(KERN_DEBUG "NFS: mtime change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
1329 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1332 if ((fattr->valid & NFS_ATTR_FATTR_V4)
1333 && nfsi->change_attr != fattr->change_attr) {
1334 #ifdef NFS_DEBUG_VERBOSE
1335 printk(KERN_DEBUG "NFS: change_attr change on %s/%ld\n",
1336 inode->i_sb->s_id, inode->i_ino);
1338 nfsi->change_attr = fattr->change_attr;
1340 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1343 /* If ctime has changed we should definitely clear access+acl caches */
1344 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1346 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1347 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1349 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1351 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1352 inode->i_uid != fattr->uid ||
1353 inode->i_gid != fattr->gid)
1354 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1356 inode->i_mode = fattr->mode;
1357 inode->i_nlink = fattr->nlink;
1358 inode->i_uid = fattr->uid;
1359 inode->i_gid = fattr->gid;
1361 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1363 * report the blocks in 512byte units
1365 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1366 inode->i_blksize = inode->i_sb->s_blocksize;
1368 inode->i_blocks = fattr->du.nfs2.blocks;
1369 inode->i_blksize = fattr->du.nfs2.blocksize;
1372 /* Update attrtimeo value if we're out of the unstable period */
1373 if (invalid & NFS_INO_INVALID_ATTR) {
1374 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1375 nfsi->attrtimeo_timestamp = jiffies;
1376 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1377 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1378 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1379 nfsi->attrtimeo_timestamp = jiffies;
1381 /* Don't invalidate the data if we were to blame */
1382 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1383 || S_ISLNK(inode->i_mode)))
1384 invalid &= ~NFS_INO_INVALID_DATA;
1385 if (!nfs_have_delegation(inode, FMODE_READ))
1386 nfsi->flags |= invalid;
1391 * Big trouble! The inode has become a different object.
1394 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1395 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1398 * No need to worry about unhashing the dentry, as the
1399 * lookup validation will know that the inode is bad.
1400 * (But we fall through to invalidate the caches.)
1402 nfs_invalidate_inode(inode);
1404 NFS_FLAGS(inode) |= NFS_INO_STALE;
1409 * File system information
1412 static int nfs_set_super(struct super_block *s, void *data)
1414 s->s_fs_info = data;
1415 return set_anon_super(s, data);
1418 static int nfs_compare_super(struct super_block *sb, void *data)
1420 struct nfs_server *server = data;
1421 struct nfs_server *old = NFS_SB(sb);
1423 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1425 if (old->addr.sin_port != server->addr.sin_port)
1427 return !nfs_compare_fh(&old->fh, &server->fh);
1430 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1431 int flags, const char *dev_name, void *raw_data)
1434 struct nfs_server *server = NULL;
1435 struct super_block *s;
1436 struct nfs_fh *root;
1437 struct nfs_mount_data *data = raw_data;
1439 s = ERR_PTR(-EINVAL);
1441 dprintk("%s: missing data argument\n", __FUNCTION__);
1444 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1445 dprintk("%s: bad mount version\n", __FUNCTION__);
1448 switch (data->version) {
1454 if (data->flags & NFS_MOUNT_VER3) {
1455 dprintk("%s: mount structure version %d does not support NFSv3\n",
1460 data->root.size = NFS2_FHSIZE;
1461 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1463 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1464 dprintk("%s: mount structure version %d does not support strong security\n",
1470 memset(data->context, 0, sizeof(data->context));
1472 #ifndef CONFIG_NFS_V3
1473 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1474 s = ERR_PTR(-EPROTONOSUPPORT);
1475 if (data->flags & NFS_MOUNT_VER3) {
1476 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1479 #endif /* CONFIG_NFS_V3 */
1481 s = ERR_PTR(-ENOMEM);
1482 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1485 memset(server, 0, sizeof(struct nfs_server));
1486 /* Zero out the NFS state stuff */
1487 init_nfsv4_state(server);
1488 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1491 if (data->flags & NFS_MOUNT_VER3)
1492 root->size = data->root.size;
1494 root->size = NFS2_FHSIZE;
1495 s = ERR_PTR(-EINVAL);
1496 if (root->size > sizeof(root->data)) {
1497 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1500 memcpy(root->data, data->root.data, root->size);
1502 /* We now require that the mount process passes the remote address */
1503 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1504 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1505 dprintk("%s: mount program didn't pass remote address!\n",
1510 /* Fire up rpciod if not yet running */
1511 s = ERR_PTR(rpciod_up());
1513 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1514 __FUNCTION__, PTR_ERR(s));
1518 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1519 if (IS_ERR(s) || s->s_root)
1520 goto out_rpciod_down;
1524 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1526 up_write(&s->s_umount);
1527 deactivate_super(s);
1528 return ERR_PTR(error);
1530 s->s_flags |= MS_ACTIVE;
1539 static void nfs_kill_super(struct super_block *s)
1541 struct nfs_server *server = NFS_SB(s);
1545 if (!IS_ERR(server->client))
1546 rpc_shutdown_client(server->client);
1547 if (!IS_ERR(server->client_sys))
1548 rpc_shutdown_client(server->client_sys);
1549 if (!IS_ERR(server->client_acl))
1550 rpc_shutdown_client(server->client_acl);
1552 if (!(server->flags & NFS_MOUNT_NONLM))
1553 lockd_down(); /* release rpc.lockd */
1555 rpciod_down(); /* release rpciod */
1557 if (server->hostname != NULL)
1558 kfree(server->hostname);
1562 static struct file_system_type nfs_fs_type = {
1563 .owner = THIS_MODULE,
1565 .get_sb = nfs_get_sb,
1566 .kill_sb = nfs_kill_super,
1567 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1570 #ifdef CONFIG_NFS_V4
1572 static void nfs4_clear_inode(struct inode *);
1575 static struct super_operations nfs4_sops = {
1576 .alloc_inode = nfs_alloc_inode,
1577 .destroy_inode = nfs_destroy_inode,
1578 .write_inode = nfs_write_inode,
1579 .delete_inode = nfs_delete_inode,
1580 .statfs = nfs_statfs,
1581 .clear_inode = nfs4_clear_inode,
1582 .umount_begin = nfs_umount_begin,
1583 .show_options = nfs_show_options,
1587 * Clean out any remaining NFSv4 state that might be left over due
1588 * to open() calls that passed nfs_atomic_lookup, but failed to call
1591 static void nfs4_clear_inode(struct inode *inode)
1593 struct nfs_inode *nfsi = NFS_I(inode);
1595 /* If we are holding a delegation, return it! */
1596 if (nfsi->delegation != NULL)
1597 nfs_inode_return_delegation(inode);
1598 /* First call standard NFS clear_inode() code */
1599 nfs_clear_inode(inode);
1600 /* Now clear out any remaining state */
1601 while (!list_empty(&nfsi->open_states)) {
1602 struct nfs4_state *state;
1604 state = list_entry(nfsi->open_states.next,
1607 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1610 (long long)NFS_FILEID(inode),
1612 BUG_ON(atomic_read(&state->count) != 1);
1613 nfs4_close_state(state, state->state);
1618 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1620 struct nfs_server *server;
1621 struct nfs4_client *clp = NULL;
1622 struct rpc_xprt *xprt = NULL;
1623 struct rpc_clnt *clnt = NULL;
1624 struct rpc_timeout timeparms;
1625 rpc_authflavor_t authflavour;
1626 int proto, err = -EIO;
1628 sb->s_blocksize_bits = 0;
1629 sb->s_blocksize = 0;
1630 server = NFS_SB(sb);
1631 if (data->rsize != 0)
1632 server->rsize = nfs_block_size(data->rsize, NULL);
1633 if (data->wsize != 0)
1634 server->wsize = nfs_block_size(data->wsize, NULL);
1635 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1636 server->caps = NFS_CAP_ATOMIC_OPEN;
1638 server->acregmin = data->acregmin*HZ;
1639 server->acregmax = data->acregmax*HZ;
1640 server->acdirmin = data->acdirmin*HZ;
1641 server->acdirmax = data->acdirmax*HZ;
1643 server->rpc_ops = &nfs_v4_clientops;
1644 /* Initialize timeout values */
1646 timeparms.to_initval = data->timeo * HZ / 10;
1647 timeparms.to_retries = data->retrans;
1648 timeparms.to_exponential = 1;
1649 if (!timeparms.to_retries)
1650 timeparms.to_retries = 5;
1652 proto = data->proto;
1653 /* Which IP protocol do we use? */
1656 timeparms.to_maxval = RPC_MAX_TCP_TIMEOUT;
1657 if (!timeparms.to_initval)
1658 timeparms.to_initval = 600 * HZ / 10;
1661 timeparms.to_maxval = RPC_MAX_UDP_TIMEOUT;
1662 if (!timeparms.to_initval)
1663 timeparms.to_initval = 11 * HZ / 10;
1669 clp = nfs4_get_client(&server->addr.sin_addr);
1671 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1675 /* Now create transport and client */
1676 authflavour = RPC_AUTH_UNIX;
1677 if (data->auth_flavourlen != 0) {
1678 if (data->auth_flavourlen != 1) {
1679 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1680 __FUNCTION__, data->auth_flavourlen);
1684 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1690 down_write(&clp->cl_sem);
1691 if (IS_ERR(clp->cl_rpcclient)) {
1692 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
1694 up_write(&clp->cl_sem);
1695 err = PTR_ERR(xprt);
1696 dprintk("%s: cannot create RPC transport. Error = %d\n",
1700 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1701 server->rpc_ops->version, authflavour);
1703 up_write(&clp->cl_sem);
1704 err = PTR_ERR(clnt);
1705 dprintk("%s: cannot create RPC client. Error = %d\n",
1710 clnt->cl_softrtry = 1;
1711 clnt->cl_chatty = 1;
1712 clp->cl_rpcclient = clnt;
1713 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1714 if (IS_ERR(clp->cl_cred)) {
1715 up_write(&clp->cl_sem);
1716 err = PTR_ERR(clp->cl_cred);
1717 clp->cl_cred = NULL;
1720 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1723 if (list_empty(&clp->cl_superblocks)) {
1724 err = nfs4_init_client(clp);
1726 up_write(&clp->cl_sem);
1730 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1731 clnt = rpc_clone_client(clp->cl_rpcclient);
1733 server->nfs4_state = clp;
1734 up_write(&clp->cl_sem);
1738 err = PTR_ERR(clnt);
1739 dprintk("%s: cannot create RPC client. Error = %d\n",
1744 server->client = clnt;
1746 if (server->nfs4_state->cl_idmap == NULL) {
1747 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1751 if (clnt->cl_auth->au_flavor != authflavour) {
1752 struct rpc_auth *auth;
1754 auth = rpcauth_create(authflavour, clnt);
1756 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1757 return PTR_ERR(auth);
1761 sb->s_time_gran = 1;
1763 sb->s_op = &nfs4_sops;
1764 err = nfs_sb_init(sb, authflavour);
1769 nfs4_put_client(clp);
1773 static int nfs4_compare_super(struct super_block *sb, void *data)
1775 struct nfs_server *server = data;
1776 struct nfs_server *old = NFS_SB(sb);
1778 if (strcmp(server->hostname, old->hostname) != 0)
1780 if (strcmp(server->mnt_path, old->mnt_path) != 0)
1786 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1791 return ERR_PTR(-EINVAL);
1792 if (src->len < maxlen)
1795 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1797 return ERR_PTR(-ENOMEM);
1799 if (copy_from_user(dst, src->data, maxlen)) {
1802 return ERR_PTR(-EFAULT);
1808 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1809 int flags, const char *dev_name, void *raw_data)
1812 struct nfs_server *server;
1813 struct super_block *s;
1814 struct nfs4_mount_data *data = raw_data;
1818 dprintk("%s: missing data argument\n", __FUNCTION__);
1819 return ERR_PTR(-EINVAL);
1821 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1822 dprintk("%s: bad mount version\n", __FUNCTION__);
1823 return ERR_PTR(-EINVAL);
1826 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1828 return ERR_PTR(-ENOMEM);
1829 memset(server, 0, sizeof(struct nfs_server));
1830 /* Zero out the NFS state stuff */
1831 init_nfsv4_state(server);
1832 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1834 p = nfs_copy_user_string(NULL, &data->hostname, 256);
1837 server->hostname = p;
1839 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1842 server->mnt_path = p;
1844 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1845 sizeof(server->ip_addr) - 1);
1849 /* We now require that the mount process passes the remote address */
1850 if (data->host_addrlen != sizeof(server->addr)) {
1851 s = ERR_PTR(-EINVAL);
1854 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1855 s = ERR_PTR(-EFAULT);
1858 if (server->addr.sin_family != AF_INET ||
1859 server->addr.sin_addr.s_addr == INADDR_ANY) {
1860 dprintk("%s: mount program didn't pass remote IP address!\n",
1862 s = ERR_PTR(-EINVAL);
1866 /* Fire up rpciod if not yet running */
1867 s = ERR_PTR(rpciod_up());
1869 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1870 __FUNCTION__, PTR_ERR(s));
1874 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1876 if (IS_ERR(s) || s->s_root)
1881 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1883 up_write(&s->s_umount);
1884 deactivate_super(s);
1885 return ERR_PTR(error);
1887 s->s_flags |= MS_ACTIVE;
1890 s = (struct super_block *)p;
1892 if (server->mnt_path)
1893 kfree(server->mnt_path);
1894 if (server->hostname)
1895 kfree(server->hostname);
1900 static void nfs4_kill_super(struct super_block *sb)
1902 struct nfs_server *server = NFS_SB(sb);
1904 nfs_return_all_delegations(sb);
1905 kill_anon_super(sb);
1907 nfs4_renewd_prepare_shutdown(server);
1909 if (server->client != NULL && !IS_ERR(server->client))
1910 rpc_shutdown_client(server->client);
1911 rpciod_down(); /* release rpciod */
1913 destroy_nfsv4_state(server);
1915 if (server->hostname != NULL)
1916 kfree(server->hostname);
1920 static struct file_system_type nfs4_fs_type = {
1921 .owner = THIS_MODULE,
1923 .get_sb = nfs4_get_sb,
1924 .kill_sb = nfs4_kill_super,
1925 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1928 #define nfs4_init_once(nfsi) \
1930 INIT_LIST_HEAD(&(nfsi)->open_states); \
1931 nfsi->delegation = NULL; \
1932 nfsi->delegation_state = 0; \
1933 init_rwsem(&nfsi->rwsem); \
1935 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1936 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1938 #define nfs4_init_once(nfsi) \
1940 #define register_nfs4fs() (0)
1941 #define unregister_nfs4fs()
1944 extern int nfs_init_nfspagecache(void);
1945 extern void nfs_destroy_nfspagecache(void);
1946 extern int nfs_init_readpagecache(void);
1947 extern void nfs_destroy_readpagecache(void);
1948 extern int nfs_init_writepagecache(void);
1949 extern void nfs_destroy_writepagecache(void);
1950 #ifdef CONFIG_NFS_DIRECTIO
1951 extern int nfs_init_directcache(void);
1952 extern void nfs_destroy_directcache(void);
1955 static kmem_cache_t * nfs_inode_cachep;
1957 static struct inode *nfs_alloc_inode(struct super_block *sb)
1959 struct nfs_inode *nfsi;
1960 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
1964 #ifdef CONFIG_NFS_V4
1965 nfsi->nfs4_acl = NULL;
1966 #endif /* CONFIG_NFS_V4 */
1967 return &nfsi->vfs_inode;
1970 static void nfs_destroy_inode(struct inode *inode)
1972 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1975 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
1977 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1979 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1980 SLAB_CTOR_CONSTRUCTOR) {
1981 inode_init_once(&nfsi->vfs_inode);
1982 spin_lock_init(&nfsi->req_lock);
1983 INIT_LIST_HEAD(&nfsi->dirty);
1984 INIT_LIST_HEAD(&nfsi->commit);
1985 INIT_LIST_HEAD(&nfsi->open_files);
1986 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1987 atomic_set(&nfsi->data_updates, 0);
1991 init_waitqueue_head(&nfsi->nfs_i_wait);
1992 nfs4_init_once(nfsi);
1996 static int nfs_init_inodecache(void)
1998 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1999 sizeof(struct nfs_inode),
2000 0, SLAB_RECLAIM_ACCOUNT,
2002 if (nfs_inode_cachep == NULL)
2008 static void nfs_destroy_inodecache(void)
2010 if (kmem_cache_destroy(nfs_inode_cachep))
2011 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2017 static int __init init_nfs_fs(void)
2021 err = nfs_init_nfspagecache();
2025 err = nfs_init_inodecache();
2029 err = nfs_init_readpagecache();
2033 err = nfs_init_writepagecache();
2037 #ifdef CONFIG_NFS_DIRECTIO
2038 err = nfs_init_directcache();
2043 #ifdef CONFIG_PROC_FS
2044 rpc_proc_register(&nfs_rpcstat);
2046 err = register_filesystem(&nfs_fs_type);
2049 if ((err = register_nfs4fs()) != 0)
2053 #ifdef CONFIG_PROC_FS
2054 rpc_proc_unregister("nfs");
2056 nfs_destroy_writepagecache();
2057 #ifdef CONFIG_NFS_DIRECTIO
2059 nfs_destroy_directcache();
2062 nfs_destroy_readpagecache();
2064 nfs_destroy_inodecache();
2066 nfs_destroy_nfspagecache();
2071 static void __exit exit_nfs_fs(void)
2073 #ifdef CONFIG_NFS_DIRECTIO
2074 nfs_destroy_directcache();
2076 nfs_destroy_writepagecache();
2077 nfs_destroy_readpagecache();
2078 nfs_destroy_inodecache();
2079 nfs_destroy_nfspagecache();
2080 #ifdef CONFIG_PROC_FS
2081 rpc_proc_unregister("nfs");
2083 unregister_filesystem(&nfs_fs_type);
2084 unregister_nfs4fs();
2087 /* Not quite true; I just maintain it */
2088 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2089 MODULE_LICENSE("GPL");
2091 module_init(init_nfs_fs)
2092 module_exit(exit_nfs_fs)