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, flags);
145 nfs_delete_inode(struct inode * inode)
147 dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
149 truncate_inode_pages(&inode->i_data, 0);
153 * The following should never happen...
155 if (nfs_have_writebacks(inode)) {
156 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
163 nfs_clear_inode(struct inode *inode)
165 struct nfs_inode *nfsi = NFS_I(inode);
166 struct rpc_cred *cred;
169 BUG_ON (!list_empty(&nfsi->open_files));
170 nfs_zap_acl_cache(inode);
171 cred = nfsi->cache_access.cred;
174 BUG_ON(atomic_read(&nfsi->data_updates) != 0);
178 nfs_umount_begin(struct super_block *sb)
180 struct rpc_clnt *rpc = NFS_SB(sb)->client;
182 /* -EIO all pending I/O */
184 rpc_killall_tasks(rpc);
185 rpc = NFS_SB(sb)->client_acl;
187 rpc_killall_tasks(rpc);
191 static inline unsigned long
192 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
194 /* make sure blocksize is a power of two */
195 if ((bsize & (bsize - 1)) || nrbitsp) {
196 unsigned char nrbits;
198 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
209 * Calculate the number of 512byte blocks used.
211 static inline unsigned long
212 nfs_calc_block_size(u64 tsize)
214 loff_t used = (tsize + 511) >> 9;
215 return (used > ULONG_MAX) ? ULONG_MAX : used;
219 * Compute and set NFS server blocksize
221 static inline unsigned long
222 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
225 bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
226 else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
227 bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
229 return nfs_block_bits(bsize, nrbitsp);
233 * Obtain the root inode of the file system.
235 static struct inode *
236 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
238 struct nfs_server *server = NFS_SB(sb);
242 error = server->rpc_ops->getroot(server, rootfh, fsinfo);
244 dprintk("nfs_get_root: getattr error = %d\n", -error);
245 return ERR_PTR(error);
248 rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
250 return ERR_PTR(-ENOMEM);
255 * Do NFS version-independent mount processing, and sanity checking
258 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
260 struct nfs_server *server;
261 struct inode *root_inode;
262 struct nfs_fattr fattr;
263 struct nfs_fsinfo fsinfo = {
266 struct nfs_pathconf pathinfo = {
269 int no_root_error = 0;
270 unsigned long max_rpc_payload;
272 /* We probably want something more informative here */
273 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
277 sb->s_magic = NFS_SUPER_MAGIC;
279 root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
280 /* Did getting the root inode fail? */
281 if (IS_ERR(root_inode)) {
282 no_root_error = PTR_ERR(root_inode);
285 sb->s_root = d_alloc_root(root_inode);
287 no_root_error = -ENOMEM;
290 sb->s_root->d_op = server->rpc_ops->dentry_ops;
292 /* Get some general file system info */
293 if (server->namelen == 0 &&
294 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
295 server->namelen = pathinfo.max_namelen;
296 /* Work out a lot of parameters */
297 if (server->rsize == 0)
298 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
299 if (server->wsize == 0)
300 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
302 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
303 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
304 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
305 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
307 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
308 if (server->rsize > max_rpc_payload)
309 server->rsize = max_rpc_payload;
310 if (server->wsize > max_rpc_payload)
311 server->wsize = max_rpc_payload;
313 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
314 if (server->rpages > NFS_READ_MAXIOV) {
315 server->rpages = NFS_READ_MAXIOV;
316 server->rsize = server->rpages << PAGE_CACHE_SHIFT;
319 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
320 if (server->wpages > NFS_WRITE_MAXIOV) {
321 server->wpages = NFS_WRITE_MAXIOV;
322 server->wsize = server->wpages << PAGE_CACHE_SHIFT;
325 if (sb->s_blocksize == 0)
326 sb->s_blocksize = nfs_block_bits(server->wsize,
327 &sb->s_blocksize_bits);
328 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
330 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
331 if (server->dtsize > PAGE_CACHE_SIZE)
332 server->dtsize = PAGE_CACHE_SIZE;
333 if (server->dtsize > server->rsize)
334 server->dtsize = server->rsize;
336 if (server->flags & NFS_MOUNT_NOAC) {
337 server->acregmin = server->acregmax = 0;
338 server->acdirmin = server->acdirmax = 0;
339 sb->s_flags |= MS_SYNCHRONOUS;
341 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
343 sb->s_maxbytes = fsinfo.maxfilesize;
344 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
345 sb->s_maxbytes = MAX_LFS_FILESIZE;
347 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
348 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
350 /* We're airborne Set socket buffersize */
351 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
353 /* Yargs. It didn't work out. */
355 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
356 if (!IS_ERR(root_inode))
358 return no_root_error;
361 static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
363 to->to_initval = timeo * HZ / 10;
364 to->to_retries = retrans;
371 to->to_initval = 60 * HZ;
372 if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
373 to->to_initval = NFS_MAX_TCP_TIMEOUT;
374 to->to_increment = to->to_initval;
375 to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
376 to->to_exponential = 0;
381 to->to_initval = 11 * HZ / 10;
382 if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
383 to->to_initval = NFS_MAX_UDP_TIMEOUT;
384 to->to_maxval = NFS_MAX_UDP_TIMEOUT;
385 to->to_exponential = 1;
391 * Create an RPC client handle.
393 static struct rpc_clnt *
394 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
396 struct rpc_timeout timeparms;
397 struct rpc_xprt *xprt = NULL;
398 struct rpc_clnt *clnt = NULL;
399 int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
401 nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
403 /* create transport and client */
404 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
406 dprintk("%s: cannot create RPC transport. Error = %ld\n",
407 __FUNCTION__, PTR_ERR(xprt));
408 return (struct rpc_clnt *)xprt;
410 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
411 server->rpc_ops->version, data->pseudoflavor);
413 dprintk("%s: cannot create RPC client. Error = %ld\n",
414 __FUNCTION__, PTR_ERR(xprt));
419 clnt->cl_softrtry = 1;
429 * The way this works is that the mount process passes a structure
430 * in the data argument which contains the server's IP address
431 * and the root file handle obtained from the server's mount
432 * daemon. We stash these away in the private superblock fields.
435 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
437 struct nfs_server *server;
438 rpc_authflavor_t authflavor;
441 sb->s_blocksize_bits = 0;
444 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
446 server->rsize = nfs_block_size(data->rsize, NULL);
448 server->wsize = nfs_block_size(data->wsize, NULL);
449 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
451 server->acregmin = data->acregmin*HZ;
452 server->acregmax = data->acregmax*HZ;
453 server->acdirmin = data->acdirmin*HZ;
454 server->acdirmax = data->acdirmax*HZ;
456 /* Start lockd here, before we might error out */
457 if (!(server->flags & NFS_MOUNT_NONLM))
460 server->namelen = data->namlen;
461 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
462 if (!server->hostname)
464 strcpy(server->hostname, data->hostname);
466 /* Check NFS protocol revision and initialize RPC op vector
467 * and file handle pool. */
469 if (server->flags & NFS_MOUNT_VER3) {
470 server->rpc_ops = &nfs_v3_clientops;
471 server->caps |= NFS_CAP_READDIRPLUS;
473 server->rpc_ops = &nfs_v2_clientops;
476 server->rpc_ops = &nfs_v2_clientops;
479 /* Fill in pseudoflavor for mount version < 5 */
480 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
481 data->pseudoflavor = RPC_AUTH_UNIX;
482 authflavor = data->pseudoflavor; /* save for sb_init() */
483 /* XXX maybe we want to add a server->pseudoflavor field */
485 /* Create RPC client handles */
486 server->client = nfs_create_client(server, data);
487 if (IS_ERR(server->client))
488 return PTR_ERR(server->client);
489 /* RFC 2623, sec 2.3.2 */
490 if (authflavor != RPC_AUTH_UNIX) {
491 struct rpc_auth *auth;
493 server->client_sys = rpc_clone_client(server->client);
494 if (IS_ERR(server->client_sys))
495 return PTR_ERR(server->client_sys);
496 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
498 return PTR_ERR(auth);
500 atomic_inc(&server->client->cl_count);
501 server->client_sys = server->client;
503 if (server->flags & NFS_MOUNT_VER3) {
504 #ifdef CONFIG_NFS_V3_ACL
505 if (!(server->flags & NFS_MOUNT_NOACL)) {
506 server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
507 /* No errors! Assume that Sun nfsacls are supported */
508 if (!IS_ERR(server->client_acl))
509 server->caps |= NFS_CAP_ACLS;
512 server->flags &= ~NFS_MOUNT_NOACL;
513 #endif /* CONFIG_NFS_V3_ACL */
515 * The VFS shouldn't apply the umask to mode bits. We will
516 * do so ourselves when necessary.
518 sb->s_flags |= MS_POSIXACL;
519 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
520 server->namelen = NFS3_MAXNAMLEN;
523 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
524 server->namelen = NFS2_MAXNAMLEN;
527 sb->s_op = &nfs_sops;
528 return nfs_sb_init(sb, authflavor);
532 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
534 struct nfs_server *server = NFS_SB(sb);
535 unsigned char blockbits;
536 unsigned long blockres;
537 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
538 struct nfs_fattr fattr;
539 struct nfs_fsstat res = {
546 error = server->rpc_ops->statfs(server, rootfh, &res);
547 buf->f_type = NFS_SUPER_MAGIC;
552 * Current versions of glibc do not correctly handle the
553 * case where f_frsize != f_bsize. Eventually we want to
554 * report the value of wtmult in this field.
556 buf->f_frsize = sb->s_blocksize;
559 * On most *nix systems, f_blocks, f_bfree, and f_bavail
560 * are reported in units of f_frsize. Linux hasn't had
561 * an f_frsize field in its statfs struct until recently,
562 * thus historically Linux's sys_statfs reports these
563 * fields in units of f_bsize.
565 buf->f_bsize = sb->s_blocksize;
566 blockbits = sb->s_blocksize_bits;
567 blockres = (1 << blockbits) - 1;
568 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
569 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
570 buf->f_bavail = (res.abytes + blockres) >> blockbits;
572 buf->f_files = res.tfiles;
573 buf->f_ffree = res.afiles;
575 buf->f_namelen = server->namelen;
582 printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
583 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
588 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
590 static struct proc_nfs_info {
595 { NFS_MOUNT_SOFT, ",soft", ",hard" },
596 { NFS_MOUNT_INTR, ",intr", "" },
597 { NFS_MOUNT_POSIX, ",posix", "" },
598 { NFS_MOUNT_NOCTO, ",nocto", "" },
599 { NFS_MOUNT_NOAC, ",noac", "" },
600 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
601 { NFS_MOUNT_NOACL, ",noacl", "" },
604 struct proc_nfs_info *nfs_infop;
605 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
609 seq_printf(m, ",v%d", nfss->rpc_ops->version);
610 seq_printf(m, ",rsize=%d", nfss->rsize);
611 seq_printf(m, ",wsize=%d", nfss->wsize);
612 if (nfss->acregmin != 3*HZ)
613 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
614 if (nfss->acregmax != 60*HZ)
615 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
616 if (nfss->acdirmin != 30*HZ)
617 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
618 if (nfss->acdirmax != 60*HZ)
619 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
620 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
621 if (nfss->flags & nfs_infop->flag)
622 seq_puts(m, nfs_infop->str);
624 seq_puts(m, nfs_infop->nostr);
626 switch (nfss->client->cl_xprt->prot) {
634 snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
637 seq_printf(m, ",proto=%s", proto);
638 seq_puts(m, ",addr=");
639 seq_escape(m, nfss->hostname, " \t\n\\");
644 * Invalidate the local caches
647 nfs_zap_caches(struct inode *inode)
649 struct nfs_inode *nfsi = NFS_I(inode);
650 int mode = inode->i_mode;
652 spin_lock(&inode->i_lock);
654 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
655 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
657 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
658 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
659 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
661 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
663 spin_unlock(&inode->i_lock);
666 static void nfs_zap_acl_cache(struct inode *inode)
668 void (*clear_acl_cache)(struct inode *);
670 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
671 if (clear_acl_cache != NULL)
672 clear_acl_cache(inode);
673 spin_lock(&inode->i_lock);
674 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
675 spin_unlock(&inode->i_lock);
679 * Invalidate, but do not unhash, the inode
682 nfs_invalidate_inode(struct inode *inode)
684 umode_t save_mode = inode->i_mode;
686 make_bad_inode(inode);
687 inode->i_mode = save_mode;
688 nfs_zap_caches(inode);
691 struct nfs_find_desc {
693 struct nfs_fattr *fattr;
697 * In NFSv3 we can have 64bit inode numbers. In order to support
698 * this, and re-exported directories (also seen in NFSv2)
699 * we are forced to allow 2 different inodes to have the same
703 nfs_find_actor(struct inode *inode, void *opaque)
705 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
706 struct nfs_fh *fh = desc->fh;
707 struct nfs_fattr *fattr = desc->fattr;
709 if (NFS_FILEID(inode) != fattr->fileid)
711 if (nfs_compare_fh(NFS_FH(inode), fh))
713 if (is_bad_inode(inode) || NFS_STALE(inode))
719 nfs_init_locked(struct inode *inode, void *opaque)
721 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
722 struct nfs_fattr *fattr = desc->fattr;
724 NFS_FILEID(inode) = fattr->fileid;
725 nfs_copy_fh(NFS_FH(inode), desc->fh);
729 /* Don't use READDIRPLUS on directories that we believe are too large */
730 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
733 * This is our front-end to iget that looks up inodes by file handle
734 * instead of inode number.
737 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
739 struct nfs_find_desc desc = {
743 struct inode *inode = NULL;
746 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
750 printk("NFS: Buggy server - nlink == 0!\n");
754 hash = nfs_fattr_to_ino_t(fattr);
756 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
759 if (inode->i_state & I_NEW) {
760 struct nfs_inode *nfsi = NFS_I(inode);
762 /* We set i_ino for the few things that still rely on it,
766 /* We can't support update_atime(), since the server will reset it */
767 inode->i_flags |= S_NOATIME|S_NOCMTIME;
768 inode->i_mode = fattr->mode;
769 /* Why so? Because we want revalidate for devices/FIFOs, and
770 * that's precisely what we have in nfs_file_inode_operations.
772 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
773 if (S_ISREG(inode->i_mode)) {
774 inode->i_fop = &nfs_file_operations;
775 inode->i_data.a_ops = &nfs_file_aops;
776 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
777 } else if (S_ISDIR(inode->i_mode)) {
778 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
779 inode->i_fop = &nfs_dir_operations;
780 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
781 && fattr->size <= NFS_LIMIT_READDIRPLUS)
782 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
783 } else if (S_ISLNK(inode->i_mode))
784 inode->i_op = &nfs_symlink_inode_operations;
786 init_special_inode(inode, inode->i_mode, fattr->rdev);
788 nfsi->read_cache_jiffies = fattr->timestamp;
789 inode->i_atime = fattr->atime;
790 inode->i_mtime = fattr->mtime;
791 inode->i_ctime = fattr->ctime;
792 if (fattr->valid & NFS_ATTR_FATTR_V4)
793 nfsi->change_attr = fattr->change_attr;
794 inode->i_size = nfs_size_to_loff_t(fattr->size);
795 inode->i_nlink = fattr->nlink;
796 inode->i_uid = fattr->uid;
797 inode->i_gid = fattr->gid;
798 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
800 * report the blocks in 512byte units
802 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
803 inode->i_blksize = inode->i_sb->s_blocksize;
805 inode->i_blocks = fattr->du.nfs2.blocks;
806 inode->i_blksize = fattr->du.nfs2.blocksize;
808 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
809 nfsi->attrtimeo_timestamp = jiffies;
810 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
811 nfsi->cache_access.cred = NULL;
813 unlock_new_inode(inode);
815 nfs_refresh_inode(inode, fattr);
816 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
818 (long long)NFS_FILEID(inode),
819 atomic_read(&inode->i_count));
825 printk("nfs_fhget: iget failed\n");
829 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
832 nfs_setattr(struct dentry *dentry, struct iattr *attr)
834 struct inode *inode = dentry->d_inode;
835 struct nfs_fattr fattr;
838 if (attr->ia_valid & ATTR_SIZE) {
839 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
840 attr->ia_valid &= ~ATTR_SIZE;
843 /* Optimization: if the end result is no change, don't RPC */
844 attr->ia_valid &= NFS_VALID_ATTRS;
845 if (attr->ia_valid == 0)
849 nfs_begin_data_update(inode);
850 /* Write all dirty data if we're changing file permissions or size */
851 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
852 if (filemap_fdatawrite(inode->i_mapping) == 0)
853 filemap_fdatawait(inode->i_mapping);
856 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
858 nfs_refresh_inode(inode, &fattr);
859 nfs_end_data_update(inode);
865 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
866 * @inode: pointer to struct inode
867 * @attr: pointer to struct iattr
869 * Note: we do this in the *proc.c in order to ensure that
870 * it works for things like exclusive creates too.
872 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
874 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
875 if ((attr->ia_valid & ATTR_MODE) != 0) {
876 int mode = attr->ia_mode & S_IALLUGO;
877 mode |= inode->i_mode & ~S_IALLUGO;
878 inode->i_mode = mode;
880 if ((attr->ia_valid & ATTR_UID) != 0)
881 inode->i_uid = attr->ia_uid;
882 if ((attr->ia_valid & ATTR_GID) != 0)
883 inode->i_gid = attr->ia_gid;
884 spin_lock(&inode->i_lock);
885 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
886 spin_unlock(&inode->i_lock);
888 if ((attr->ia_valid & ATTR_SIZE) != 0) {
889 inode->i_size = attr->ia_size;
890 vmtruncate(inode, attr->ia_size);
894 static int nfs_wait_schedule(void *word)
896 if (signal_pending(current))
903 * Wait for the inode to get unlocked.
905 static int nfs_wait_on_inode(struct inode *inode)
907 struct rpc_clnt *clnt = NFS_CLIENT(inode);
908 struct nfs_inode *nfsi = NFS_I(inode);
912 atomic_inc(&inode->i_count);
913 rpc_clnt_sigmask(clnt, &oldmask);
914 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
915 nfs_wait_schedule, TASK_INTERRUPTIBLE);
916 rpc_clnt_sigunmask(clnt, &oldmask);
922 static void nfs_wake_up_inode(struct inode *inode)
924 struct nfs_inode *nfsi = NFS_I(inode);
926 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
927 smp_mb__after_clear_bit();
928 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
931 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
933 struct inode *inode = dentry->d_inode;
934 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
937 if (__IS_FLG(inode, MS_NOATIME))
939 else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
941 /* We may force a getattr if the user cares about atime */
943 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
945 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
947 generic_fillattr(inode, stat);
951 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
953 struct nfs_open_context *ctx;
955 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
957 atomic_set(&ctx->count, 1);
958 ctx->dentry = dget(dentry);
959 ctx->cred = get_rpccred(cred);
961 ctx->lockowner = current->files;
968 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
971 atomic_inc(&ctx->count);
975 void put_nfs_open_context(struct nfs_open_context *ctx)
977 if (atomic_dec_and_test(&ctx->count)) {
978 if (!list_empty(&ctx->list)) {
979 struct inode *inode = ctx->dentry->d_inode;
980 spin_lock(&inode->i_lock);
981 list_del(&ctx->list);
982 spin_unlock(&inode->i_lock);
984 if (ctx->state != NULL)
985 nfs4_close_state(ctx->state, ctx->mode);
986 if (ctx->cred != NULL)
987 put_rpccred(ctx->cred);
994 * Ensure that mmap has a recent RPC credential for use when writing out
997 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
999 struct inode *inode = filp->f_dentry->d_inode;
1000 struct nfs_inode *nfsi = NFS_I(inode);
1002 filp->private_data = get_nfs_open_context(ctx);
1003 spin_lock(&inode->i_lock);
1004 list_add(&ctx->list, &nfsi->open_files);
1005 spin_unlock(&inode->i_lock);
1008 struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
1010 struct nfs_inode *nfsi = NFS_I(inode);
1011 struct nfs_open_context *pos, *ctx = NULL;
1013 spin_lock(&inode->i_lock);
1014 list_for_each_entry(pos, &nfsi->open_files, list) {
1015 if ((pos->mode & mode) == mode) {
1016 ctx = get_nfs_open_context(pos);
1020 spin_unlock(&inode->i_lock);
1024 void nfs_file_clear_open_context(struct file *filp)
1026 struct inode *inode = filp->f_dentry->d_inode;
1027 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
1030 filp->private_data = NULL;
1031 spin_lock(&inode->i_lock);
1032 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1033 spin_unlock(&inode->i_lock);
1034 put_nfs_open_context(ctx);
1039 * These allocate and release file read/write context information.
1041 int nfs_open(struct inode *inode, struct file *filp)
1043 struct nfs_open_context *ctx;
1044 struct rpc_cred *cred;
1046 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1048 return PTR_ERR(cred);
1049 ctx = alloc_nfs_open_context(filp->f_dentry, cred);
1053 ctx->mode = filp->f_mode;
1054 nfs_file_set_open_context(filp, ctx);
1055 put_nfs_open_context(ctx);
1056 if ((filp->f_mode & FMODE_WRITE) != 0)
1057 nfs_begin_data_update(inode);
1061 int nfs_release(struct inode *inode, struct file *filp)
1063 if ((filp->f_mode & FMODE_WRITE) != 0)
1064 nfs_end_data_update(inode);
1065 nfs_file_clear_open_context(filp);
1070 * This function is called whenever some part of NFS notices that
1071 * the cached attributes have to be refreshed.
1074 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1076 int status = -ESTALE;
1077 struct nfs_fattr fattr;
1078 struct nfs_inode *nfsi = NFS_I(inode);
1079 unsigned long verifier;
1080 unsigned long cache_validity;
1082 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1083 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1086 if (!inode || is_bad_inode(inode))
1088 if (NFS_STALE(inode))
1091 status = nfs_wait_on_inode(inode);
1094 if (NFS_STALE(inode)) {
1096 /* Do we trust the cached ESTALE? */
1097 if (NFS_ATTRTIMEO(inode) != 0) {
1098 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1105 /* Protect against RPC races by saving the change attribute */
1106 verifier = nfs_save_change_attribute(inode);
1107 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1109 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1111 (long long)NFS_FILEID(inode), status);
1112 if (status == -ESTALE) {
1113 nfs_zap_caches(inode);
1114 if (!S_ISDIR(inode->i_mode))
1115 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1120 status = nfs_update_inode(inode, &fattr, verifier);
1122 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1124 (long long)NFS_FILEID(inode), status);
1127 spin_lock(&inode->i_lock);
1128 cache_validity = nfsi->cache_validity;
1129 nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1132 * We may need to keep the attributes marked as invalid if
1133 * we raced with nfs_end_attr_update().
1135 if (verifier == nfsi->cache_change_attribute)
1136 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1137 spin_unlock(&inode->i_lock);
1139 nfs_revalidate_mapping(inode, inode->i_mapping);
1141 if (cache_validity & NFS_INO_INVALID_ACL)
1142 nfs_zap_acl_cache(inode);
1144 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1146 (long long)NFS_FILEID(inode));
1149 nfs_wake_up_inode(inode);
1156 int nfs_attribute_timeout(struct inode *inode)
1158 struct nfs_inode *nfsi = NFS_I(inode);
1160 if (nfs_have_delegation(inode, FMODE_READ))
1162 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1166 * nfs_revalidate_inode - Revalidate the inode attributes
1167 * @server - pointer to nfs_server struct
1168 * @inode - pointer to inode struct
1170 * Updates inode attribute information by retrieving the data from the server.
1172 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1174 if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1175 && !nfs_attribute_timeout(inode))
1176 return NFS_STALE(inode) ? -ESTALE : 0;
1177 return __nfs_revalidate_inode(server, inode);
1181 * nfs_revalidate_mapping - Revalidate the pagecache
1182 * @inode - pointer to host inode
1183 * @mapping - pointer to mapping
1185 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1187 struct nfs_inode *nfsi = NFS_I(inode);
1189 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1190 if (S_ISREG(inode->i_mode)) {
1191 if (filemap_fdatawrite(mapping) == 0)
1192 filemap_fdatawait(mapping);
1195 invalidate_inode_pages2(mapping);
1197 spin_lock(&inode->i_lock);
1198 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1199 if (S_ISDIR(inode->i_mode)) {
1200 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1201 /* This ensures we revalidate child dentries */
1202 nfsi->cache_change_attribute++;
1204 spin_unlock(&inode->i_lock);
1206 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1208 (long long)NFS_FILEID(inode));
1213 * nfs_begin_data_update
1214 * @inode - pointer to inode
1215 * Declare that a set of operations will update file data on the server
1217 void nfs_begin_data_update(struct inode *inode)
1219 atomic_inc(&NFS_I(inode)->data_updates);
1223 * nfs_end_data_update
1224 * @inode - pointer to inode
1225 * Declare end of the operations that will update file data
1226 * This will mark the inode as immediately needing revalidation
1227 * of its attribute cache.
1229 void nfs_end_data_update(struct inode *inode)
1231 struct nfs_inode *nfsi = NFS_I(inode);
1233 if (!nfs_have_delegation(inode, FMODE_READ)) {
1234 /* Mark the attribute cache for revalidation */
1235 spin_lock(&inode->i_lock);
1236 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1237 /* Directories and symlinks: invalidate page cache too */
1238 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1239 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1240 spin_unlock(&inode->i_lock);
1242 nfsi->cache_change_attribute ++;
1243 atomic_dec(&nfsi->data_updates);
1247 * nfs_refresh_inode - verify consistency of the inode attribute cache
1248 * @inode - pointer to inode
1249 * @fattr - updated attributes
1251 * Verifies the attribute cache. If we have just changed the attributes,
1252 * so that fattr carries weak cache consistency data, then it may
1253 * also update the ctime/mtime/change_attribute.
1255 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1257 struct nfs_inode *nfsi = NFS_I(inode);
1258 loff_t cur_size, new_isize;
1261 /* Do we hold a delegation? */
1262 if (nfs_have_delegation(inode, FMODE_READ))
1265 spin_lock(&inode->i_lock);
1267 /* Are we in the process of updating data on the server? */
1268 data_unstable = nfs_caches_unstable(inode);
1270 if (fattr->valid & NFS_ATTR_FATTR_V4) {
1271 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1272 && nfsi->change_attr == fattr->pre_change_attr)
1273 nfsi->change_attr = fattr->change_attr;
1274 if (nfsi->change_attr != fattr->change_attr) {
1275 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1277 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1281 if ((fattr->valid & NFS_ATTR_FATTR) == 0) {
1282 spin_unlock(&inode->i_lock);
1286 /* Has the inode gone and changed behind our back? */
1287 if (nfsi->fileid != fattr->fileid
1288 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1289 spin_unlock(&inode->i_lock);
1293 cur_size = i_size_read(inode);
1294 new_isize = nfs_size_to_loff_t(fattr->size);
1296 /* If we have atomic WCC data, we may update some attributes */
1297 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1298 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
1299 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1300 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
1301 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1304 /* Verify a few of the more important attributes */
1305 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1306 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1308 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1310 if (cur_size != new_isize) {
1311 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1312 if (nfsi->npages == 0)
1313 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1316 /* Have any file permissions changed? */
1317 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1318 || inode->i_uid != fattr->uid
1319 || inode->i_gid != fattr->gid)
1320 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1322 /* Has the link count changed? */
1323 if (inode->i_nlink != fattr->nlink)
1324 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1326 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1327 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1329 nfsi->read_cache_jiffies = fattr->timestamp;
1330 spin_unlock(&inode->i_lock);
1335 * Many nfs protocol calls return the new file attributes after
1336 * an operation. Here we update the inode to reflect the state
1337 * of the server's inode.
1339 * This is a bit tricky because we have to make sure all dirty pages
1340 * have been sent off to the server before calling invalidate_inode_pages.
1341 * To make sure no other process adds more write requests while we try
1342 * our best to flush them, we make them sleep during the attribute refresh.
1344 * A very similar scenario holds for the dir cache.
1346 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
1348 struct nfs_inode *nfsi = NFS_I(inode);
1349 loff_t cur_isize, new_isize;
1350 unsigned int invalid = 0;
1353 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1354 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1355 atomic_read(&inode->i_count), fattr->valid);
1357 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1360 if (nfsi->fileid != fattr->fileid) {
1361 printk(KERN_ERR "%s: inode number mismatch\n"
1362 "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
1364 inode->i_sb->s_id, (long long)nfsi->fileid,
1365 inode->i_sb->s_id, (long long)fattr->fileid);
1369 spin_lock(&inode->i_lock);
1372 * Make sure the inode's type hasn't changed.
1374 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1375 spin_unlock(&inode->i_lock);
1380 * Update the read time so we don't revalidate too often.
1382 nfsi->read_cache_jiffies = fattr->timestamp;
1384 /* Are we racing with known updates of the metadata on the server? */
1385 data_unstable = ! nfs_verify_change_attribute(inode, verifier);
1387 /* Check if our cached file size is stale */
1388 new_isize = nfs_size_to_loff_t(fattr->size);
1389 cur_isize = i_size_read(inode);
1390 if (new_isize != cur_isize) {
1391 /* Do we perhaps have any outstanding writes? */
1392 if (nfsi->npages == 0) {
1393 /* No, but did we race with nfs_end_data_update()? */
1394 if (verifier == nfsi->cache_change_attribute) {
1395 inode->i_size = new_isize;
1396 invalid |= NFS_INO_INVALID_DATA;
1398 invalid |= NFS_INO_INVALID_ATTR;
1399 } else if (new_isize > cur_isize) {
1400 inode->i_size = new_isize;
1401 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1403 dprintk("NFS: isize change on server for file %s/%ld\n",
1404 inode->i_sb->s_id, inode->i_ino);
1407 /* Check if the mtime agrees */
1408 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1409 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1410 dprintk("NFS: mtime change on server for file %s/%ld\n",
1411 inode->i_sb->s_id, inode->i_ino);
1413 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1416 if ((fattr->valid & NFS_ATTR_FATTR_V4)
1417 && nfsi->change_attr != fattr->change_attr) {
1418 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1419 inode->i_sb->s_id, inode->i_ino);
1420 nfsi->change_attr = fattr->change_attr;
1422 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1425 /* If ctime has changed we should definitely clear access+acl caches */
1426 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1428 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1429 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1431 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1433 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1434 inode->i_uid != fattr->uid ||
1435 inode->i_gid != fattr->gid)
1436 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1438 inode->i_mode = fattr->mode;
1439 inode->i_nlink = fattr->nlink;
1440 inode->i_uid = fattr->uid;
1441 inode->i_gid = fattr->gid;
1443 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1445 * report the blocks in 512byte units
1447 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1448 inode->i_blksize = inode->i_sb->s_blocksize;
1450 inode->i_blocks = fattr->du.nfs2.blocks;
1451 inode->i_blksize = fattr->du.nfs2.blocksize;
1454 /* Update attrtimeo value if we're out of the unstable period */
1455 if (invalid & NFS_INO_INVALID_ATTR) {
1456 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1457 nfsi->attrtimeo_timestamp = jiffies;
1458 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1459 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1460 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1461 nfsi->attrtimeo_timestamp = jiffies;
1463 /* Don't invalidate the data if we were to blame */
1464 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1465 || S_ISLNK(inode->i_mode)))
1466 invalid &= ~NFS_INO_INVALID_DATA;
1467 if (!nfs_have_delegation(inode, FMODE_READ))
1468 nfsi->cache_validity |= invalid;
1470 spin_unlock(&inode->i_lock);
1474 * Big trouble! The inode has become a different object.
1477 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1478 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1481 * No need to worry about unhashing the dentry, as the
1482 * lookup validation will know that the inode is bad.
1483 * (But we fall through to invalidate the caches.)
1485 nfs_invalidate_inode(inode);
1487 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1492 * File system information
1495 static int nfs_set_super(struct super_block *s, void *data)
1497 s->s_fs_info = data;
1498 return set_anon_super(s, data);
1501 static int nfs_compare_super(struct super_block *sb, void *data)
1503 struct nfs_server *server = data;
1504 struct nfs_server *old = NFS_SB(sb);
1506 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1508 if (old->addr.sin_port != server->addr.sin_port)
1510 return !nfs_compare_fh(&old->fh, &server->fh);
1513 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1514 int flags, const char *dev_name, void *raw_data)
1517 struct nfs_server *server = NULL;
1518 struct super_block *s;
1519 struct nfs_fh *root;
1520 struct nfs_mount_data *data = raw_data;
1522 s = ERR_PTR(-EINVAL);
1524 dprintk("%s: missing data argument\n", __FUNCTION__);
1527 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1528 dprintk("%s: bad mount version\n", __FUNCTION__);
1531 switch (data->version) {
1537 if (data->flags & NFS_MOUNT_VER3) {
1538 dprintk("%s: mount structure version %d does not support NFSv3\n",
1543 data->root.size = NFS2_FHSIZE;
1544 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1546 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1547 dprintk("%s: mount structure version %d does not support strong security\n",
1553 memset(data->context, 0, sizeof(data->context));
1555 #ifndef CONFIG_NFS_V3
1556 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1557 s = ERR_PTR(-EPROTONOSUPPORT);
1558 if (data->flags & NFS_MOUNT_VER3) {
1559 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1562 #endif /* CONFIG_NFS_V3 */
1564 s = ERR_PTR(-ENOMEM);
1565 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1568 memset(server, 0, sizeof(struct nfs_server));
1569 /* Zero out the NFS state stuff */
1570 init_nfsv4_state(server);
1571 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1574 if (data->flags & NFS_MOUNT_VER3)
1575 root->size = data->root.size;
1577 root->size = NFS2_FHSIZE;
1578 s = ERR_PTR(-EINVAL);
1579 if (root->size > sizeof(root->data)) {
1580 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1583 memcpy(root->data, data->root.data, root->size);
1585 /* We now require that the mount process passes the remote address */
1586 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1587 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1588 dprintk("%s: mount program didn't pass remote address!\n",
1593 /* Fire up rpciod if not yet running */
1594 s = ERR_PTR(rpciod_up());
1596 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1597 __FUNCTION__, PTR_ERR(s));
1601 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1602 if (IS_ERR(s) || s->s_root)
1603 goto out_rpciod_down;
1607 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1609 up_write(&s->s_umount);
1610 deactivate_super(s);
1611 return ERR_PTR(error);
1613 s->s_flags |= MS_ACTIVE;
1622 static void nfs_kill_super(struct super_block *s)
1624 struct nfs_server *server = NFS_SB(s);
1628 if (!IS_ERR(server->client))
1629 rpc_shutdown_client(server->client);
1630 if (!IS_ERR(server->client_sys))
1631 rpc_shutdown_client(server->client_sys);
1632 if (!IS_ERR(server->client_acl))
1633 rpc_shutdown_client(server->client_acl);
1635 if (!(server->flags & NFS_MOUNT_NONLM))
1636 lockd_down(); /* release rpc.lockd */
1638 rpciod_down(); /* release rpciod */
1640 if (server->hostname != NULL)
1641 kfree(server->hostname);
1645 static struct file_system_type nfs_fs_type = {
1646 .owner = THIS_MODULE,
1648 .get_sb = nfs_get_sb,
1649 .kill_sb = nfs_kill_super,
1650 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1653 #ifdef CONFIG_NFS_V4
1655 static void nfs4_clear_inode(struct inode *);
1658 static struct super_operations nfs4_sops = {
1659 .alloc_inode = nfs_alloc_inode,
1660 .destroy_inode = nfs_destroy_inode,
1661 .write_inode = nfs_write_inode,
1662 .delete_inode = nfs_delete_inode,
1663 .statfs = nfs_statfs,
1664 .clear_inode = nfs4_clear_inode,
1665 .umount_begin = nfs_umount_begin,
1666 .show_options = nfs_show_options,
1670 * Clean out any remaining NFSv4 state that might be left over due
1671 * to open() calls that passed nfs_atomic_lookup, but failed to call
1674 static void nfs4_clear_inode(struct inode *inode)
1676 struct nfs_inode *nfsi = NFS_I(inode);
1678 /* If we are holding a delegation, return it! */
1679 if (nfsi->delegation != NULL)
1680 nfs_inode_return_delegation(inode);
1681 /* First call standard NFS clear_inode() code */
1682 nfs_clear_inode(inode);
1683 /* Now clear out any remaining state */
1684 while (!list_empty(&nfsi->open_states)) {
1685 struct nfs4_state *state;
1687 state = list_entry(nfsi->open_states.next,
1690 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1693 (long long)NFS_FILEID(inode),
1695 BUG_ON(atomic_read(&state->count) != 1);
1696 nfs4_close_state(state, state->state);
1701 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1703 struct nfs_server *server;
1704 struct nfs4_client *clp = NULL;
1705 struct rpc_xprt *xprt = NULL;
1706 struct rpc_clnt *clnt = NULL;
1707 struct rpc_timeout timeparms;
1708 rpc_authflavor_t authflavour;
1711 sb->s_blocksize_bits = 0;
1712 sb->s_blocksize = 0;
1713 server = NFS_SB(sb);
1714 if (data->rsize != 0)
1715 server->rsize = nfs_block_size(data->rsize, NULL);
1716 if (data->wsize != 0)
1717 server->wsize = nfs_block_size(data->wsize, NULL);
1718 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1719 server->caps = NFS_CAP_ATOMIC_OPEN;
1721 server->acregmin = data->acregmin*HZ;
1722 server->acregmax = data->acregmax*HZ;
1723 server->acdirmin = data->acdirmin*HZ;
1724 server->acdirmax = data->acdirmax*HZ;
1726 server->rpc_ops = &nfs_v4_clientops;
1728 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
1730 clp = nfs4_get_client(&server->addr.sin_addr);
1732 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1736 /* Now create transport and client */
1737 authflavour = RPC_AUTH_UNIX;
1738 if (data->auth_flavourlen != 0) {
1739 if (data->auth_flavourlen != 1) {
1740 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1741 __FUNCTION__, data->auth_flavourlen);
1745 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1751 down_write(&clp->cl_sem);
1752 if (IS_ERR(clp->cl_rpcclient)) {
1753 xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
1755 up_write(&clp->cl_sem);
1756 err = PTR_ERR(xprt);
1757 dprintk("%s: cannot create RPC transport. Error = %d\n",
1761 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1762 server->rpc_ops->version, authflavour);
1764 up_write(&clp->cl_sem);
1765 err = PTR_ERR(clnt);
1766 dprintk("%s: cannot create RPC client. Error = %d\n",
1771 clnt->cl_softrtry = 1;
1772 clnt->cl_chatty = 1;
1773 clp->cl_rpcclient = clnt;
1774 clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
1775 if (IS_ERR(clp->cl_cred)) {
1776 up_write(&clp->cl_sem);
1777 err = PTR_ERR(clp->cl_cred);
1778 clp->cl_cred = NULL;
1781 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1784 if (list_empty(&clp->cl_superblocks)) {
1785 err = nfs4_init_client(clp);
1787 up_write(&clp->cl_sem);
1791 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1792 clnt = rpc_clone_client(clp->cl_rpcclient);
1794 server->nfs4_state = clp;
1795 up_write(&clp->cl_sem);
1799 err = PTR_ERR(clnt);
1800 dprintk("%s: cannot create RPC client. Error = %d\n",
1805 server->client = clnt;
1807 if (server->nfs4_state->cl_idmap == NULL) {
1808 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1812 if (clnt->cl_auth->au_flavor != authflavour) {
1813 struct rpc_auth *auth;
1815 auth = rpcauth_create(authflavour, clnt);
1817 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1818 return PTR_ERR(auth);
1822 sb->s_time_gran = 1;
1824 sb->s_op = &nfs4_sops;
1825 err = nfs_sb_init(sb, authflavour);
1830 nfs4_put_client(clp);
1834 static int nfs4_compare_super(struct super_block *sb, void *data)
1836 struct nfs_server *server = data;
1837 struct nfs_server *old = NFS_SB(sb);
1839 if (strcmp(server->hostname, old->hostname) != 0)
1841 if (strcmp(server->mnt_path, old->mnt_path) != 0)
1847 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1852 return ERR_PTR(-EINVAL);
1853 if (src->len < maxlen)
1856 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1858 return ERR_PTR(-ENOMEM);
1860 if (copy_from_user(dst, src->data, maxlen)) {
1863 return ERR_PTR(-EFAULT);
1869 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1870 int flags, const char *dev_name, void *raw_data)
1873 struct nfs_server *server;
1874 struct super_block *s;
1875 struct nfs4_mount_data *data = raw_data;
1879 dprintk("%s: missing data argument\n", __FUNCTION__);
1880 return ERR_PTR(-EINVAL);
1882 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1883 dprintk("%s: bad mount version\n", __FUNCTION__);
1884 return ERR_PTR(-EINVAL);
1887 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1889 return ERR_PTR(-ENOMEM);
1890 memset(server, 0, sizeof(struct nfs_server));
1891 /* Zero out the NFS state stuff */
1892 init_nfsv4_state(server);
1893 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1895 p = nfs_copy_user_string(NULL, &data->hostname, 256);
1898 server->hostname = p;
1900 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1903 server->mnt_path = p;
1905 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1906 sizeof(server->ip_addr) - 1);
1910 /* We now require that the mount process passes the remote address */
1911 if (data->host_addrlen != sizeof(server->addr)) {
1912 s = ERR_PTR(-EINVAL);
1915 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1916 s = ERR_PTR(-EFAULT);
1919 if (server->addr.sin_family != AF_INET ||
1920 server->addr.sin_addr.s_addr == INADDR_ANY) {
1921 dprintk("%s: mount program didn't pass remote IP address!\n",
1923 s = ERR_PTR(-EINVAL);
1927 /* Fire up rpciod if not yet running */
1928 s = ERR_PTR(rpciod_up());
1930 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1931 __FUNCTION__, PTR_ERR(s));
1935 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1937 if (IS_ERR(s) || s->s_root)
1942 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1944 up_write(&s->s_umount);
1945 deactivate_super(s);
1946 return ERR_PTR(error);
1948 s->s_flags |= MS_ACTIVE;
1951 s = (struct super_block *)p;
1953 if (server->mnt_path)
1954 kfree(server->mnt_path);
1955 if (server->hostname)
1956 kfree(server->hostname);
1961 static void nfs4_kill_super(struct super_block *sb)
1963 struct nfs_server *server = NFS_SB(sb);
1965 nfs_return_all_delegations(sb);
1966 kill_anon_super(sb);
1968 nfs4_renewd_prepare_shutdown(server);
1970 if (server->client != NULL && !IS_ERR(server->client))
1971 rpc_shutdown_client(server->client);
1972 rpciod_down(); /* release rpciod */
1974 destroy_nfsv4_state(server);
1976 if (server->hostname != NULL)
1977 kfree(server->hostname);
1981 static struct file_system_type nfs4_fs_type = {
1982 .owner = THIS_MODULE,
1984 .get_sb = nfs4_get_sb,
1985 .kill_sb = nfs4_kill_super,
1986 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1989 #define nfs4_init_once(nfsi) \
1991 INIT_LIST_HEAD(&(nfsi)->open_states); \
1992 nfsi->delegation = NULL; \
1993 nfsi->delegation_state = 0; \
1994 init_rwsem(&nfsi->rwsem); \
1996 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
1997 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
1999 #define nfs4_init_once(nfsi) \
2001 #define register_nfs4fs() (0)
2002 #define unregister_nfs4fs()
2005 extern int nfs_init_nfspagecache(void);
2006 extern void nfs_destroy_nfspagecache(void);
2007 extern int nfs_init_readpagecache(void);
2008 extern void nfs_destroy_readpagecache(void);
2009 extern int nfs_init_writepagecache(void);
2010 extern void nfs_destroy_writepagecache(void);
2011 #ifdef CONFIG_NFS_DIRECTIO
2012 extern int nfs_init_directcache(void);
2013 extern void nfs_destroy_directcache(void);
2016 static kmem_cache_t * nfs_inode_cachep;
2018 static struct inode *nfs_alloc_inode(struct super_block *sb)
2020 struct nfs_inode *nfsi;
2021 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2025 nfsi->cache_validity = 0UL;
2026 #ifdef CONFIG_NFS_V3_ACL
2027 nfsi->acl_access = ERR_PTR(-EAGAIN);
2028 nfsi->acl_default = ERR_PTR(-EAGAIN);
2030 #ifdef CONFIG_NFS_V4
2031 nfsi->nfs4_acl = NULL;
2032 #endif /* CONFIG_NFS_V4 */
2033 return &nfsi->vfs_inode;
2036 static void nfs_destroy_inode(struct inode *inode)
2038 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2041 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2043 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2045 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2046 SLAB_CTOR_CONSTRUCTOR) {
2047 inode_init_once(&nfsi->vfs_inode);
2048 spin_lock_init(&nfsi->req_lock);
2049 INIT_LIST_HEAD(&nfsi->dirty);
2050 INIT_LIST_HEAD(&nfsi->commit);
2051 INIT_LIST_HEAD(&nfsi->open_files);
2052 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2053 atomic_set(&nfsi->data_updates, 0);
2057 nfs4_init_once(nfsi);
2061 static int nfs_init_inodecache(void)
2063 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2064 sizeof(struct nfs_inode),
2065 0, SLAB_RECLAIM_ACCOUNT,
2067 if (nfs_inode_cachep == NULL)
2073 static void nfs_destroy_inodecache(void)
2075 if (kmem_cache_destroy(nfs_inode_cachep))
2076 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2082 static int __init init_nfs_fs(void)
2086 err = nfs_init_nfspagecache();
2090 err = nfs_init_inodecache();
2094 err = nfs_init_readpagecache();
2098 err = nfs_init_writepagecache();
2102 #ifdef CONFIG_NFS_DIRECTIO
2103 err = nfs_init_directcache();
2108 #ifdef CONFIG_PROC_FS
2109 rpc_proc_register(&nfs_rpcstat);
2111 err = register_filesystem(&nfs_fs_type);
2114 if ((err = register_nfs4fs()) != 0)
2118 #ifdef CONFIG_PROC_FS
2119 rpc_proc_unregister("nfs");
2121 nfs_destroy_writepagecache();
2122 #ifdef CONFIG_NFS_DIRECTIO
2124 nfs_destroy_directcache();
2127 nfs_destroy_readpagecache();
2129 nfs_destroy_inodecache();
2131 nfs_destroy_nfspagecache();
2136 static void __exit exit_nfs_fs(void)
2138 #ifdef CONFIG_NFS_DIRECTIO
2139 nfs_destroy_directcache();
2141 nfs_destroy_writepagecache();
2142 nfs_destroy_readpagecache();
2143 nfs_destroy_inodecache();
2144 nfs_destroy_nfspagecache();
2145 #ifdef CONFIG_PROC_FS
2146 rpc_proc_unregister("nfs");
2148 unregister_filesystem(&nfs_fs_type);
2149 unregister_nfs4fs();
2152 /* Not quite true; I just maintain it */
2153 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2154 MODULE_LICENSE("GPL");
2156 module_init(init_nfs_fs)
2157 module_exit(exit_nfs_fs)