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>
44 #include "delegation.h"
46 #define NFSDBG_FACILITY NFSDBG_VFS
47 #define NFS_PARANOIA 1
49 /* Maximum number of readahead requests
50 * FIXME: this should really be a sysctl so that users may tune it to suit
51 * their needs. People that do NFS over a slow network, might for
52 * instance want to reduce it to something closer to 1 for improved
53 * interactive response.
55 #define NFS_MAX_READAHEAD (RPC_DEF_SLOT_TABLE - 1)
57 static void nfs_invalidate_inode(struct inode *);
58 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
60 static struct inode *nfs_alloc_inode(struct super_block *sb);
61 static void nfs_destroy_inode(struct inode *);
62 static int nfs_write_inode(struct inode *,int);
63 static void nfs_delete_inode(struct inode *);
64 static void nfs_clear_inode(struct inode *);
65 static void nfs_umount_begin(struct super_block *);
66 static int nfs_statfs(struct super_block *, struct kstatfs *);
67 static int nfs_show_options(struct seq_file *, struct vfsmount *);
68 static void nfs_zap_acl_cache(struct inode *);
70 static struct rpc_program nfs_program;
72 static struct super_operations nfs_sops = {
73 .alloc_inode = nfs_alloc_inode,
74 .destroy_inode = nfs_destroy_inode,
75 .write_inode = nfs_write_inode,
76 .delete_inode = nfs_delete_inode,
78 .clear_inode = nfs_clear_inode,
79 .umount_begin = nfs_umount_begin,
80 .show_options = nfs_show_options,
86 static struct rpc_stat nfs_rpcstat = {
87 .program = &nfs_program
89 static struct rpc_version * nfs_version[] = {
93 #if defined(CONFIG_NFS_V3)
95 #elif defined(CONFIG_NFS_V4)
98 #if defined(CONFIG_NFS_V4)
103 static struct rpc_program nfs_program = {
105 .number = NFS_PROGRAM,
106 .nrvers = ARRAY_SIZE(nfs_version),
107 .version = nfs_version,
108 .stats = &nfs_rpcstat,
109 .pipe_dir_name = "/nfs",
112 #ifdef CONFIG_NFS_V3_ACL
113 static struct rpc_stat nfsacl_rpcstat = { &nfsacl_program };
114 static struct rpc_version * nfsacl_version[] = {
115 [3] = &nfsacl_version3,
118 struct rpc_program nfsacl_program = {
120 .number = NFS_ACL_PROGRAM,
121 .nrvers = ARRAY_SIZE(nfsacl_version),
122 .version = nfsacl_version,
123 .stats = &nfsacl_rpcstat,
125 #endif /* CONFIG_NFS_V3_ACL */
127 static inline unsigned long
128 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
130 return nfs_fileid_to_ino_t(fattr->fileid);
134 nfs_write_inode(struct inode *inode, int sync)
136 int flags = sync ? FLUSH_WAIT : 0;
139 ret = nfs_commit_inode(inode, flags);
146 nfs_delete_inode(struct inode * inode)
148 dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
150 truncate_inode_pages(&inode->i_data, 0);
154 * The following should never happen...
156 if (nfs_have_writebacks(inode)) {
157 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
164 nfs_clear_inode(struct inode *inode)
166 struct nfs_inode *nfsi = NFS_I(inode);
167 struct rpc_cred *cred;
170 BUG_ON (!list_empty(&nfsi->open_files));
171 nfs_zap_acl_cache(inode);
172 cred = nfsi->cache_access.cred;
175 BUG_ON(atomic_read(&nfsi->data_updates) != 0);
179 nfs_umount_begin(struct super_block *sb)
181 struct rpc_clnt *rpc = NFS_SB(sb)->client;
183 /* -EIO all pending I/O */
185 rpc_killall_tasks(rpc);
186 rpc = NFS_SB(sb)->client_acl;
188 rpc_killall_tasks(rpc);
192 static inline unsigned long
193 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
195 /* make sure blocksize is a power of two */
196 if ((bsize & (bsize - 1)) || nrbitsp) {
197 unsigned char nrbits;
199 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
210 * Calculate the number of 512byte blocks used.
212 static inline unsigned long
213 nfs_calc_block_size(u64 tsize)
215 loff_t used = (tsize + 511) >> 9;
216 return (used > ULONG_MAX) ? ULONG_MAX : used;
220 * Compute and set NFS server blocksize
222 static inline unsigned long
223 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
225 if (bsize < NFS_MIN_FILE_IO_SIZE)
226 bsize = NFS_DEF_FILE_IO_SIZE;
227 else if (bsize >= NFS_MAX_FILE_IO_SIZE)
228 bsize = NFS_MAX_FILE_IO_SIZE;
230 return nfs_block_bits(bsize, nrbitsp);
234 * Obtain the root inode of the file system.
236 static struct inode *
237 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
239 struct nfs_server *server = NFS_SB(sb);
243 error = server->rpc_ops->getroot(server, rootfh, fsinfo);
245 dprintk("nfs_get_root: getattr error = %d\n", -error);
246 return ERR_PTR(error);
249 rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
251 return ERR_PTR(-ENOMEM);
256 * Do NFS version-independent mount processing, and sanity checking
259 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
261 struct nfs_server *server;
262 struct inode *root_inode;
263 struct nfs_fattr fattr;
264 struct nfs_fsinfo fsinfo = {
267 struct nfs_pathconf pathinfo = {
270 int no_root_error = 0;
271 unsigned long max_rpc_payload;
273 /* We probably want something more informative here */
274 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
278 sb->s_magic = NFS_SUPER_MAGIC;
280 root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
281 /* Did getting the root inode fail? */
282 if (IS_ERR(root_inode)) {
283 no_root_error = PTR_ERR(root_inode);
286 sb->s_root = d_alloc_root(root_inode);
288 no_root_error = -ENOMEM;
291 sb->s_root->d_op = server->rpc_ops->dentry_ops;
293 /* Get some general file system info */
294 if (server->namelen == 0 &&
295 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
296 server->namelen = pathinfo.max_namelen;
297 /* Work out a lot of parameters */
298 if (server->rsize == 0)
299 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
300 if (server->wsize == 0)
301 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
303 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
304 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
305 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
306 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
308 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
309 if (server->rsize > max_rpc_payload)
310 server->rsize = max_rpc_payload;
311 if (server->rsize > NFS_MAX_FILE_IO_SIZE)
312 server->rsize = NFS_MAX_FILE_IO_SIZE;
313 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
315 if (server->wsize > max_rpc_payload)
316 server->wsize = max_rpc_payload;
317 if (server->wsize > NFS_MAX_FILE_IO_SIZE)
318 server->wsize = NFS_MAX_FILE_IO_SIZE;
319 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
321 if (sb->s_blocksize == 0)
322 sb->s_blocksize = nfs_block_bits(server->wsize,
323 &sb->s_blocksize_bits);
324 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
326 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
327 if (server->dtsize > PAGE_CACHE_SIZE)
328 server->dtsize = PAGE_CACHE_SIZE;
329 if (server->dtsize > server->rsize)
330 server->dtsize = server->rsize;
332 if (server->flags & NFS_MOUNT_NOAC) {
333 server->acregmin = server->acregmax = 0;
334 server->acdirmin = server->acdirmax = 0;
335 sb->s_flags |= MS_SYNCHRONOUS;
337 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
339 sb->s_maxbytes = fsinfo.maxfilesize;
340 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
341 sb->s_maxbytes = MAX_LFS_FILESIZE;
343 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
344 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
346 /* We're airborne Set socket buffersize */
347 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
349 /* Yargs. It didn't work out. */
351 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
352 if (!IS_ERR(root_inode))
354 return no_root_error;
357 static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
359 to->to_initval = timeo * HZ / 10;
360 to->to_retries = retrans;
367 to->to_initval = 60 * HZ;
368 if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
369 to->to_initval = NFS_MAX_TCP_TIMEOUT;
370 to->to_increment = to->to_initval;
371 to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
372 to->to_exponential = 0;
377 to->to_initval = 11 * HZ / 10;
378 if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
379 to->to_initval = NFS_MAX_UDP_TIMEOUT;
380 to->to_maxval = NFS_MAX_UDP_TIMEOUT;
381 to->to_exponential = 1;
387 * Create an RPC client handle.
389 static struct rpc_clnt *
390 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
392 struct rpc_timeout timeparms;
393 struct rpc_xprt *xprt = NULL;
394 struct rpc_clnt *clnt = NULL;
395 int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
397 nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
399 /* create transport and client */
400 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
402 dprintk("%s: cannot create RPC transport. Error = %ld\n",
403 __FUNCTION__, PTR_ERR(xprt));
404 return (struct rpc_clnt *)xprt;
406 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
407 server->rpc_ops->version, data->pseudoflavor);
409 dprintk("%s: cannot create RPC client. Error = %ld\n",
410 __FUNCTION__, PTR_ERR(xprt));
415 clnt->cl_softrtry = 1;
424 * The way this works is that the mount process passes a structure
425 * in the data argument which contains the server's IP address
426 * and the root file handle obtained from the server's mount
427 * daemon. We stash these away in the private superblock fields.
430 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
432 struct nfs_server *server;
433 rpc_authflavor_t authflavor;
436 sb->s_blocksize_bits = 0;
439 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
441 server->rsize = nfs_block_size(data->rsize, NULL);
443 server->wsize = nfs_block_size(data->wsize, NULL);
444 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
446 server->acregmin = data->acregmin*HZ;
447 server->acregmax = data->acregmax*HZ;
448 server->acdirmin = data->acdirmin*HZ;
449 server->acdirmax = data->acdirmax*HZ;
451 /* Start lockd here, before we might error out */
452 if (!(server->flags & NFS_MOUNT_NONLM))
455 server->namelen = data->namlen;
456 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
457 if (!server->hostname)
459 strcpy(server->hostname, data->hostname);
461 /* Check NFS protocol revision and initialize RPC op vector
462 * and file handle pool. */
464 if (server->flags & NFS_MOUNT_VER3) {
465 server->rpc_ops = &nfs_v3_clientops;
466 server->caps |= NFS_CAP_READDIRPLUS;
468 server->rpc_ops = &nfs_v2_clientops;
471 server->rpc_ops = &nfs_v2_clientops;
474 /* Fill in pseudoflavor for mount version < 5 */
475 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
476 data->pseudoflavor = RPC_AUTH_UNIX;
477 authflavor = data->pseudoflavor; /* save for sb_init() */
478 /* XXX maybe we want to add a server->pseudoflavor field */
480 /* Create RPC client handles */
481 server->client = nfs_create_client(server, data);
482 if (IS_ERR(server->client))
483 return PTR_ERR(server->client);
484 /* RFC 2623, sec 2.3.2 */
485 if (authflavor != RPC_AUTH_UNIX) {
486 struct rpc_auth *auth;
488 server->client_sys = rpc_clone_client(server->client);
489 if (IS_ERR(server->client_sys))
490 return PTR_ERR(server->client_sys);
491 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
493 return PTR_ERR(auth);
495 atomic_inc(&server->client->cl_count);
496 server->client_sys = server->client;
498 if (server->flags & NFS_MOUNT_VER3) {
499 #ifdef CONFIG_NFS_V3_ACL
500 if (!(server->flags & NFS_MOUNT_NOACL)) {
501 server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
502 /* No errors! Assume that Sun nfsacls are supported */
503 if (!IS_ERR(server->client_acl))
504 server->caps |= NFS_CAP_ACLS;
507 server->flags &= ~NFS_MOUNT_NOACL;
508 #endif /* CONFIG_NFS_V3_ACL */
510 * The VFS shouldn't apply the umask to mode bits. We will
511 * do so ourselves when necessary.
513 sb->s_flags |= MS_POSIXACL;
514 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
515 server->namelen = NFS3_MAXNAMLEN;
518 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
519 server->namelen = NFS2_MAXNAMLEN;
522 sb->s_op = &nfs_sops;
523 return nfs_sb_init(sb, authflavor);
527 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
529 struct nfs_server *server = NFS_SB(sb);
530 unsigned char blockbits;
531 unsigned long blockres;
532 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
533 struct nfs_fattr fattr;
534 struct nfs_fsstat res = {
541 error = server->rpc_ops->statfs(server, rootfh, &res);
542 buf->f_type = NFS_SUPER_MAGIC;
547 * Current versions of glibc do not correctly handle the
548 * case where f_frsize != f_bsize. Eventually we want to
549 * report the value of wtmult in this field.
551 buf->f_frsize = sb->s_blocksize;
554 * On most *nix systems, f_blocks, f_bfree, and f_bavail
555 * are reported in units of f_frsize. Linux hasn't had
556 * an f_frsize field in its statfs struct until recently,
557 * thus historically Linux's sys_statfs reports these
558 * fields in units of f_bsize.
560 buf->f_bsize = sb->s_blocksize;
561 blockbits = sb->s_blocksize_bits;
562 blockres = (1 << blockbits) - 1;
563 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
564 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
565 buf->f_bavail = (res.abytes + blockres) >> blockbits;
567 buf->f_files = res.tfiles;
568 buf->f_ffree = res.afiles;
570 buf->f_namelen = server->namelen;
576 dprintk("%s: statfs error = %d\n", __FUNCTION__, -error);
577 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
582 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
584 static struct proc_nfs_info {
589 { NFS_MOUNT_SOFT, ",soft", ",hard" },
590 { NFS_MOUNT_INTR, ",intr", "" },
591 { NFS_MOUNT_POSIX, ",posix", "" },
592 { NFS_MOUNT_NOCTO, ",nocto", "" },
593 { NFS_MOUNT_NOAC, ",noac", "" },
594 { NFS_MOUNT_NONLM, ",nolock", ",lock" },
595 { NFS_MOUNT_NOACL, ",noacl", "" },
598 struct proc_nfs_info *nfs_infop;
599 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
603 seq_printf(m, ",v%d", nfss->rpc_ops->version);
604 seq_printf(m, ",rsize=%d", nfss->rsize);
605 seq_printf(m, ",wsize=%d", nfss->wsize);
606 if (nfss->acregmin != 3*HZ)
607 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
608 if (nfss->acregmax != 60*HZ)
609 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
610 if (nfss->acdirmin != 30*HZ)
611 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
612 if (nfss->acdirmax != 60*HZ)
613 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
614 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
615 if (nfss->flags & nfs_infop->flag)
616 seq_puts(m, nfs_infop->str);
618 seq_puts(m, nfs_infop->nostr);
620 switch (nfss->client->cl_xprt->prot) {
628 snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
631 seq_printf(m, ",proto=%s", proto);
632 seq_puts(m, ",addr=");
633 seq_escape(m, nfss->hostname, " \t\n\\");
638 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
640 int nfs_sync_mapping(struct address_space *mapping)
644 if (mapping->nrpages == 0)
646 unmap_mapping_range(mapping, 0, 0, 0);
647 ret = filemap_write_and_wait(mapping);
650 ret = nfs_wb_all(mapping->host);
656 * Invalidate the local caches
658 static void nfs_zap_caches_locked(struct inode *inode)
660 struct nfs_inode *nfsi = NFS_I(inode);
661 int mode = inode->i_mode;
663 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
664 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
666 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
667 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
668 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
670 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
673 void nfs_zap_caches(struct inode *inode)
675 spin_lock(&inode->i_lock);
676 nfs_zap_caches_locked(inode);
677 spin_unlock(&inode->i_lock);
680 static void nfs_zap_acl_cache(struct inode *inode)
682 void (*clear_acl_cache)(struct inode *);
684 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
685 if (clear_acl_cache != NULL)
686 clear_acl_cache(inode);
687 spin_lock(&inode->i_lock);
688 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
689 spin_unlock(&inode->i_lock);
693 * Invalidate, but do not unhash, the inode.
694 * NB: must be called with inode->i_lock held!
696 static void nfs_invalidate_inode(struct inode *inode)
698 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
699 nfs_zap_caches_locked(inode);
702 struct nfs_find_desc {
704 struct nfs_fattr *fattr;
708 * In NFSv3 we can have 64bit inode numbers. In order to support
709 * this, and re-exported directories (also seen in NFSv2)
710 * we are forced to allow 2 different inodes to have the same
714 nfs_find_actor(struct inode *inode, void *opaque)
716 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
717 struct nfs_fh *fh = desc->fh;
718 struct nfs_fattr *fattr = desc->fattr;
720 if (NFS_FILEID(inode) != fattr->fileid)
722 if (nfs_compare_fh(NFS_FH(inode), fh))
724 if (is_bad_inode(inode) || NFS_STALE(inode))
730 nfs_init_locked(struct inode *inode, void *opaque)
732 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
733 struct nfs_fattr *fattr = desc->fattr;
735 NFS_FILEID(inode) = fattr->fileid;
736 nfs_copy_fh(NFS_FH(inode), desc->fh);
740 /* Don't use READDIRPLUS on directories that we believe are too large */
741 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
744 * This is our front-end to iget that looks up inodes by file handle
745 * instead of inode number.
748 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
750 struct nfs_find_desc desc = {
754 struct inode *inode = NULL;
757 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
761 printk("NFS: Buggy server - nlink == 0!\n");
765 hash = nfs_fattr_to_ino_t(fattr);
767 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
770 if (inode->i_state & I_NEW) {
771 struct nfs_inode *nfsi = NFS_I(inode);
773 /* We set i_ino for the few things that still rely on it,
777 /* We can't support update_atime(), since the server will reset it */
778 inode->i_flags |= S_NOATIME|S_NOCMTIME;
779 inode->i_mode = fattr->mode;
780 /* Why so? Because we want revalidate for devices/FIFOs, and
781 * that's precisely what we have in nfs_file_inode_operations.
783 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
784 if (S_ISREG(inode->i_mode)) {
785 inode->i_fop = &nfs_file_operations;
786 inode->i_data.a_ops = &nfs_file_aops;
787 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
788 } else if (S_ISDIR(inode->i_mode)) {
789 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
790 inode->i_fop = &nfs_dir_operations;
791 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
792 && fattr->size <= NFS_LIMIT_READDIRPLUS)
793 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
794 } else if (S_ISLNK(inode->i_mode))
795 inode->i_op = &nfs_symlink_inode_operations;
797 init_special_inode(inode, inode->i_mode, fattr->rdev);
799 nfsi->read_cache_jiffies = fattr->time_start;
800 nfsi->last_updated = jiffies;
801 inode->i_atime = fattr->atime;
802 inode->i_mtime = fattr->mtime;
803 inode->i_ctime = fattr->ctime;
804 if (fattr->valid & NFS_ATTR_FATTR_V4)
805 nfsi->change_attr = fattr->change_attr;
806 inode->i_size = nfs_size_to_loff_t(fattr->size);
807 inode->i_nlink = fattr->nlink;
808 inode->i_uid = fattr->uid;
809 inode->i_gid = fattr->gid;
810 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
812 * report the blocks in 512byte units
814 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
815 inode->i_blksize = inode->i_sb->s_blocksize;
817 inode->i_blocks = fattr->du.nfs2.blocks;
818 inode->i_blksize = fattr->du.nfs2.blocksize;
820 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
821 nfsi->attrtimeo_timestamp = jiffies;
822 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
823 nfsi->cache_access.cred = NULL;
825 unlock_new_inode(inode);
827 nfs_refresh_inode(inode, fattr);
828 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
830 (long long)NFS_FILEID(inode),
831 atomic_read(&inode->i_count));
837 printk("nfs_fhget: iget failed\n");
841 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
844 nfs_setattr(struct dentry *dentry, struct iattr *attr)
846 struct inode *inode = dentry->d_inode;
847 struct nfs_fattr fattr;
850 if (attr->ia_valid & ATTR_SIZE) {
851 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
852 attr->ia_valid &= ~ATTR_SIZE;
855 /* Optimization: if the end result is no change, don't RPC */
856 attr->ia_valid &= NFS_VALID_ATTRS;
857 if (attr->ia_valid == 0)
861 nfs_begin_data_update(inode);
862 /* Write all dirty data if we're changing file permissions or size */
863 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
864 filemap_write_and_wait(inode->i_mapping);
868 * Return any delegations if we're going to change ACLs
870 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
871 nfs_inode_return_delegation(inode);
872 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
874 nfs_refresh_inode(inode, &fattr);
875 nfs_end_data_update(inode);
881 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
882 * @inode: pointer to struct inode
883 * @attr: pointer to struct iattr
885 * Note: we do this in the *proc.c in order to ensure that
886 * it works for things like exclusive creates too.
888 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
890 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
891 if ((attr->ia_valid & ATTR_MODE) != 0) {
892 int mode = attr->ia_mode & S_IALLUGO;
893 mode |= inode->i_mode & ~S_IALLUGO;
894 inode->i_mode = mode;
896 if ((attr->ia_valid & ATTR_UID) != 0)
897 inode->i_uid = attr->ia_uid;
898 if ((attr->ia_valid & ATTR_GID) != 0)
899 inode->i_gid = attr->ia_gid;
900 spin_lock(&inode->i_lock);
901 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
902 spin_unlock(&inode->i_lock);
904 if ((attr->ia_valid & ATTR_SIZE) != 0) {
905 inode->i_size = attr->ia_size;
906 vmtruncate(inode, attr->ia_size);
910 static int nfs_wait_schedule(void *word)
912 if (signal_pending(current))
919 * Wait for the inode to get unlocked.
921 static int nfs_wait_on_inode(struct inode *inode)
923 struct rpc_clnt *clnt = NFS_CLIENT(inode);
924 struct nfs_inode *nfsi = NFS_I(inode);
928 rpc_clnt_sigmask(clnt, &oldmask);
929 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
930 nfs_wait_schedule, TASK_INTERRUPTIBLE);
931 rpc_clnt_sigunmask(clnt, &oldmask);
936 static void nfs_wake_up_inode(struct inode *inode)
938 struct nfs_inode *nfsi = NFS_I(inode);
940 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
941 smp_mb__after_clear_bit();
942 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
945 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
947 struct inode *inode = dentry->d_inode;
948 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
951 /* Flush out writes to the server in order to update c/mtime */
952 nfs_sync_inode(inode, 0, 0, FLUSH_WAIT|FLUSH_NOCOMMIT);
955 * We may force a getattr if the user cares about atime.
957 * Note that we only have to check the vfsmount flags here:
958 * - NFS always sets S_NOATIME by so checking it would give a
960 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
961 * no point in checking those.
963 if ((mnt->mnt_flags & MNT_NOATIME) ||
964 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
968 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
970 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
972 generic_fillattr(inode, stat);
976 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
978 struct nfs_open_context *ctx;
980 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
982 atomic_set(&ctx->count, 1);
983 ctx->dentry = dget(dentry);
984 ctx->cred = get_rpccred(cred);
986 ctx->lockowner = current->files;
993 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
996 atomic_inc(&ctx->count);
1000 void put_nfs_open_context(struct nfs_open_context *ctx)
1002 if (atomic_dec_and_test(&ctx->count)) {
1003 if (!list_empty(&ctx->list)) {
1004 struct inode *inode = ctx->dentry->d_inode;
1005 spin_lock(&inode->i_lock);
1006 list_del(&ctx->list);
1007 spin_unlock(&inode->i_lock);
1009 if (ctx->state != NULL)
1010 nfs4_close_state(ctx->state, ctx->mode);
1011 if (ctx->cred != NULL)
1012 put_rpccred(ctx->cred);
1019 * Ensure that mmap has a recent RPC credential for use when writing out
1022 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1024 struct inode *inode = filp->f_dentry->d_inode;
1025 struct nfs_inode *nfsi = NFS_I(inode);
1027 filp->private_data = get_nfs_open_context(ctx);
1028 spin_lock(&inode->i_lock);
1029 list_add(&ctx->list, &nfsi->open_files);
1030 spin_unlock(&inode->i_lock);
1034 * Given an inode, search for an open context with the desired characteristics
1036 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
1038 struct nfs_inode *nfsi = NFS_I(inode);
1039 struct nfs_open_context *pos, *ctx = NULL;
1041 spin_lock(&inode->i_lock);
1042 list_for_each_entry(pos, &nfsi->open_files, list) {
1043 if (cred != NULL && pos->cred != cred)
1045 if ((pos->mode & mode) == mode) {
1046 ctx = get_nfs_open_context(pos);
1050 spin_unlock(&inode->i_lock);
1054 void nfs_file_clear_open_context(struct file *filp)
1056 struct inode *inode = filp->f_dentry->d_inode;
1057 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
1060 filp->private_data = NULL;
1061 spin_lock(&inode->i_lock);
1062 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1063 spin_unlock(&inode->i_lock);
1064 put_nfs_open_context(ctx);
1069 * These allocate and release file read/write context information.
1071 int nfs_open(struct inode *inode, struct file *filp)
1073 struct nfs_open_context *ctx;
1074 struct rpc_cred *cred;
1076 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1078 return PTR_ERR(cred);
1079 ctx = alloc_nfs_open_context(filp->f_dentry, cred);
1083 ctx->mode = filp->f_mode;
1084 nfs_file_set_open_context(filp, ctx);
1085 put_nfs_open_context(ctx);
1089 int nfs_release(struct inode *inode, struct file *filp)
1091 nfs_file_clear_open_context(filp);
1096 * This function is called whenever some part of NFS notices that
1097 * the cached attributes have to be refreshed.
1100 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1102 int status = -ESTALE;
1103 struct nfs_fattr fattr;
1104 struct nfs_inode *nfsi = NFS_I(inode);
1106 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1107 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1110 if (!inode || is_bad_inode(inode))
1112 if (NFS_STALE(inode))
1115 status = nfs_wait_on_inode(inode);
1118 if (NFS_STALE(inode)) {
1120 /* Do we trust the cached ESTALE? */
1121 if (NFS_ATTRTIMEO(inode) != 0) {
1122 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1129 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1131 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1133 (long long)NFS_FILEID(inode), status);
1134 if (status == -ESTALE) {
1135 nfs_zap_caches(inode);
1136 if (!S_ISDIR(inode->i_mode))
1137 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1142 spin_lock(&inode->i_lock);
1143 status = nfs_update_inode(inode, &fattr);
1145 spin_unlock(&inode->i_lock);
1146 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1148 (long long)NFS_FILEID(inode), status);
1151 spin_unlock(&inode->i_lock);
1153 nfs_revalidate_mapping(inode, inode->i_mapping);
1155 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1156 nfs_zap_acl_cache(inode);
1158 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1160 (long long)NFS_FILEID(inode));
1163 nfs_wake_up_inode(inode);
1170 int nfs_attribute_timeout(struct inode *inode)
1172 struct nfs_inode *nfsi = NFS_I(inode);
1174 if (nfs_have_delegation(inode, FMODE_READ))
1176 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1180 * nfs_revalidate_inode - Revalidate the inode attributes
1181 * @server - pointer to nfs_server struct
1182 * @inode - pointer to inode struct
1184 * Updates inode attribute information by retrieving the data from the server.
1186 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1188 if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1189 && !nfs_attribute_timeout(inode))
1190 return NFS_STALE(inode) ? -ESTALE : 0;
1191 return __nfs_revalidate_inode(server, inode);
1195 * nfs_revalidate_mapping - Revalidate the pagecache
1196 * @inode - pointer to host inode
1197 * @mapping - pointer to mapping
1199 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1201 struct nfs_inode *nfsi = NFS_I(inode);
1203 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1204 if (S_ISREG(inode->i_mode))
1205 nfs_sync_mapping(mapping);
1206 invalidate_inode_pages2(mapping);
1208 spin_lock(&inode->i_lock);
1209 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1210 if (S_ISDIR(inode->i_mode)) {
1211 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1212 /* This ensures we revalidate child dentries */
1213 nfsi->cache_change_attribute = jiffies;
1215 spin_unlock(&inode->i_lock);
1217 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1219 (long long)NFS_FILEID(inode));
1224 * nfs_begin_data_update
1225 * @inode - pointer to inode
1226 * Declare that a set of operations will update file data on the server
1228 void nfs_begin_data_update(struct inode *inode)
1230 atomic_inc(&NFS_I(inode)->data_updates);
1234 * nfs_end_data_update
1235 * @inode - pointer to inode
1236 * Declare end of the operations that will update file data
1237 * This will mark the inode as immediately needing revalidation
1238 * of its attribute cache.
1240 void nfs_end_data_update(struct inode *inode)
1242 struct nfs_inode *nfsi = NFS_I(inode);
1244 if (!nfs_have_delegation(inode, FMODE_READ)) {
1245 /* Directories and symlinks: invalidate page cache */
1246 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
1247 spin_lock(&inode->i_lock);
1248 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1249 spin_unlock(&inode->i_lock);
1252 nfsi->cache_change_attribute = jiffies;
1253 atomic_dec(&nfsi->data_updates);
1256 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1258 struct nfs_inode *nfsi = NFS_I(inode);
1260 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1261 && nfsi->change_attr == fattr->pre_change_attr) {
1262 nfsi->change_attr = fattr->change_attr;
1263 nfsi->cache_change_attribute = jiffies;
1266 /* If we have atomic WCC data, we may update some attributes */
1267 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1268 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1269 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1270 nfsi->cache_change_attribute = jiffies;
1272 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1273 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1274 nfsi->cache_change_attribute = jiffies;
1276 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
1277 inode->i_size = fattr->size;
1278 nfsi->cache_change_attribute = jiffies;
1284 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1285 * @inode - pointer to inode
1286 * @fattr - updated attributes
1288 * Verifies the attribute cache. If we have just changed the attributes,
1289 * so that fattr carries weak cache consistency data, then it may
1290 * also update the ctime/mtime/change_attribute.
1292 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1294 struct nfs_inode *nfsi = NFS_I(inode);
1295 loff_t cur_size, new_isize;
1299 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1302 /* Are we in the process of updating data on the server? */
1303 data_unstable = nfs_caches_unstable(inode);
1305 /* Do atomic weak cache consistency updates */
1306 nfs_wcc_update_inode(inode, fattr);
1308 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1309 nfsi->change_attr != fattr->change_attr) {
1310 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1312 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1315 /* Has the inode gone and changed behind our back? */
1316 if (nfsi->fileid != fattr->fileid
1317 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1321 cur_size = i_size_read(inode);
1322 new_isize = nfs_size_to_loff_t(fattr->size);
1324 /* Verify a few of the more important attributes */
1325 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1326 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1328 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1330 if (cur_size != new_isize) {
1331 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1332 if (nfsi->npages == 0)
1333 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1336 /* Have any file permissions changed? */
1337 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1338 || inode->i_uid != fattr->uid
1339 || inode->i_gid != fattr->gid)
1340 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1342 /* Has the link count changed? */
1343 if (inode->i_nlink != fattr->nlink)
1344 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1346 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1347 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1349 nfsi->read_cache_jiffies = fattr->time_start;
1354 * nfs_refresh_inode - try to update the inode attribute cache
1355 * @inode - pointer to inode
1356 * @fattr - updated attributes
1358 * Check that an RPC call that returned attributes has not overlapped with
1359 * other recent updates of the inode metadata, then decide whether it is
1360 * safe to do a full update of the inode attributes, or whether just to
1361 * call nfs_check_inode_attributes.
1363 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1365 struct nfs_inode *nfsi = NFS_I(inode);
1368 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1370 spin_lock(&inode->i_lock);
1371 nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1372 if (time_after(fattr->time_start, nfsi->last_updated))
1373 status = nfs_update_inode(inode, fattr);
1375 status = nfs_check_inode_attributes(inode, fattr);
1377 spin_unlock(&inode->i_lock);
1382 * nfs_post_op_update_inode - try to update the inode attribute cache
1383 * @inode - pointer to inode
1384 * @fattr - updated attributes
1386 * After an operation that has changed the inode metadata, mark the
1387 * attribute cache as being invalid, then try to update it.
1389 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1391 struct nfs_inode *nfsi = NFS_I(inode);
1394 spin_lock(&inode->i_lock);
1395 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
1396 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1399 status = nfs_update_inode(inode, fattr);
1401 spin_unlock(&inode->i_lock);
1406 * Many nfs protocol calls return the new file attributes after
1407 * an operation. Here we update the inode to reflect the state
1408 * of the server's inode.
1410 * This is a bit tricky because we have to make sure all dirty pages
1411 * have been sent off to the server before calling invalidate_inode_pages.
1412 * To make sure no other process adds more write requests while we try
1413 * our best to flush them, we make them sleep during the attribute refresh.
1415 * A very similar scenario holds for the dir cache.
1417 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1419 struct nfs_inode *nfsi = NFS_I(inode);
1420 loff_t cur_isize, new_isize;
1421 unsigned int invalid = 0;
1424 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1425 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1426 atomic_read(&inode->i_count), fattr->valid);
1428 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1431 if (nfsi->fileid != fattr->fileid)
1435 * Make sure the inode's type hasn't changed.
1437 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1441 * Update the read time so we don't revalidate too often.
1443 nfsi->read_cache_jiffies = fattr->time_start;
1444 nfsi->last_updated = jiffies;
1446 /* Are we racing with known updates of the metadata on the server? */
1447 data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
1449 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1451 /* Do atomic weak cache consistency updates */
1452 nfs_wcc_update_inode(inode, fattr);
1454 /* Check if our cached file size is stale */
1455 new_isize = nfs_size_to_loff_t(fattr->size);
1456 cur_isize = i_size_read(inode);
1457 if (new_isize != cur_isize) {
1458 /* Do we perhaps have any outstanding writes? */
1459 if (nfsi->npages == 0) {
1460 /* No, but did we race with nfs_end_data_update()? */
1462 inode->i_size = new_isize;
1463 invalid |= NFS_INO_INVALID_DATA;
1465 invalid |= NFS_INO_INVALID_ATTR;
1466 } else if (new_isize > cur_isize) {
1467 inode->i_size = new_isize;
1468 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1470 nfsi->cache_change_attribute = jiffies;
1471 dprintk("NFS: isize change on server for file %s/%ld\n",
1472 inode->i_sb->s_id, inode->i_ino);
1475 /* Check if the mtime agrees */
1476 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1477 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1478 dprintk("NFS: mtime change on server for file %s/%ld\n",
1479 inode->i_sb->s_id, inode->i_ino);
1480 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1481 nfsi->cache_change_attribute = jiffies;
1484 if ((fattr->valid & NFS_ATTR_FATTR_V4)
1485 && nfsi->change_attr != fattr->change_attr) {
1486 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1487 inode->i_sb->s_id, inode->i_ino);
1488 nfsi->change_attr = fattr->change_attr;
1489 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1490 nfsi->cache_change_attribute = jiffies;
1493 /* If ctime has changed we should definitely clear access+acl caches */
1494 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1495 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1496 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1497 nfsi->cache_change_attribute = jiffies;
1499 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1501 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1502 inode->i_uid != fattr->uid ||
1503 inode->i_gid != fattr->gid)
1504 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1506 inode->i_mode = fattr->mode;
1507 inode->i_nlink = fattr->nlink;
1508 inode->i_uid = fattr->uid;
1509 inode->i_gid = fattr->gid;
1511 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1513 * report the blocks in 512byte units
1515 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1516 inode->i_blksize = inode->i_sb->s_blocksize;
1518 inode->i_blocks = fattr->du.nfs2.blocks;
1519 inode->i_blksize = fattr->du.nfs2.blocksize;
1522 /* Update attrtimeo value if we're out of the unstable period */
1523 if (invalid & NFS_INO_INVALID_ATTR) {
1524 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1525 nfsi->attrtimeo_timestamp = jiffies;
1526 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1527 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1528 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1529 nfsi->attrtimeo_timestamp = jiffies;
1531 /* Don't invalidate the data if we were to blame */
1532 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1533 || S_ISLNK(inode->i_mode)))
1534 invalid &= ~NFS_INO_INVALID_DATA;
1536 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1537 if (!nfs_have_delegation(inode, FMODE_READ))
1538 nfsi->cache_validity |= invalid;
1543 * Big trouble! The inode has become a different object.
1546 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1547 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1551 * No need to worry about unhashing the dentry, as the
1552 * lookup validation will know that the inode is bad.
1553 * (But we fall through to invalidate the caches.)
1555 nfs_invalidate_inode(inode);
1559 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1560 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1561 NFS_SERVER(inode)->hostname, inode->i_sb->s_id,
1562 (long long)nfsi->fileid, (long long)fattr->fileid);
1567 * File system information
1570 static int nfs_set_super(struct super_block *s, void *data)
1572 s->s_fs_info = data;
1573 return set_anon_super(s, data);
1576 static int nfs_compare_super(struct super_block *sb, void *data)
1578 struct nfs_server *server = data;
1579 struct nfs_server *old = NFS_SB(sb);
1581 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1583 if (old->addr.sin_port != server->addr.sin_port)
1585 return !nfs_compare_fh(&old->fh, &server->fh);
1588 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1589 int flags, const char *dev_name, void *raw_data)
1592 struct nfs_server *server = NULL;
1593 struct super_block *s;
1594 struct nfs_fh *root;
1595 struct nfs_mount_data *data = raw_data;
1597 s = ERR_PTR(-EINVAL);
1599 dprintk("%s: missing data argument\n", __FUNCTION__);
1602 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1603 dprintk("%s: bad mount version\n", __FUNCTION__);
1606 switch (data->version) {
1612 if (data->flags & NFS_MOUNT_VER3) {
1613 dprintk("%s: mount structure version %d does not support NFSv3\n",
1618 data->root.size = NFS2_FHSIZE;
1619 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1621 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1622 dprintk("%s: mount structure version %d does not support strong security\n",
1628 memset(data->context, 0, sizeof(data->context));
1630 #ifndef CONFIG_NFS_V3
1631 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1632 s = ERR_PTR(-EPROTONOSUPPORT);
1633 if (data->flags & NFS_MOUNT_VER3) {
1634 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1637 #endif /* CONFIG_NFS_V3 */
1639 s = ERR_PTR(-ENOMEM);
1640 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1643 memset(server, 0, sizeof(struct nfs_server));
1644 /* Zero out the NFS state stuff */
1645 init_nfsv4_state(server);
1646 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1649 if (data->flags & NFS_MOUNT_VER3)
1650 root->size = data->root.size;
1652 root->size = NFS2_FHSIZE;
1653 s = ERR_PTR(-EINVAL);
1654 if (root->size > sizeof(root->data)) {
1655 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1658 memcpy(root->data, data->root.data, root->size);
1660 /* We now require that the mount process passes the remote address */
1661 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1662 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1663 dprintk("%s: mount program didn't pass remote address!\n",
1668 /* Fire up rpciod if not yet running */
1669 s = ERR_PTR(rpciod_up());
1671 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1672 __FUNCTION__, PTR_ERR(s));
1676 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1677 if (IS_ERR(s) || s->s_root)
1678 goto out_rpciod_down;
1682 error = nfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1684 up_write(&s->s_umount);
1685 deactivate_super(s);
1686 return ERR_PTR(error);
1688 s->s_flags |= MS_ACTIVE;
1697 static void nfs_kill_super(struct super_block *s)
1699 struct nfs_server *server = NFS_SB(s);
1703 if (!IS_ERR(server->client))
1704 rpc_shutdown_client(server->client);
1705 if (!IS_ERR(server->client_sys))
1706 rpc_shutdown_client(server->client_sys);
1707 if (!IS_ERR(server->client_acl))
1708 rpc_shutdown_client(server->client_acl);
1710 if (!(server->flags & NFS_MOUNT_NONLM))
1711 lockd_down(); /* release rpc.lockd */
1713 rpciod_down(); /* release rpciod */
1715 kfree(server->hostname);
1719 static struct file_system_type nfs_fs_type = {
1720 .owner = THIS_MODULE,
1722 .get_sb = nfs_get_sb,
1723 .kill_sb = nfs_kill_super,
1724 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1727 #ifdef CONFIG_NFS_V4
1729 static void nfs4_clear_inode(struct inode *);
1732 static struct super_operations nfs4_sops = {
1733 .alloc_inode = nfs_alloc_inode,
1734 .destroy_inode = nfs_destroy_inode,
1735 .write_inode = nfs_write_inode,
1736 .delete_inode = nfs_delete_inode,
1737 .statfs = nfs_statfs,
1738 .clear_inode = nfs4_clear_inode,
1739 .umount_begin = nfs_umount_begin,
1740 .show_options = nfs_show_options,
1744 * Clean out any remaining NFSv4 state that might be left over due
1745 * to open() calls that passed nfs_atomic_lookup, but failed to call
1748 static void nfs4_clear_inode(struct inode *inode)
1750 struct nfs_inode *nfsi = NFS_I(inode);
1752 /* If we are holding a delegation, return it! */
1753 nfs_inode_return_delegation(inode);
1754 /* First call standard NFS clear_inode() code */
1755 nfs_clear_inode(inode);
1756 /* Now clear out any remaining state */
1757 while (!list_empty(&nfsi->open_states)) {
1758 struct nfs4_state *state;
1760 state = list_entry(nfsi->open_states.next,
1763 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1766 (long long)NFS_FILEID(inode),
1768 BUG_ON(atomic_read(&state->count) != 1);
1769 nfs4_close_state(state, state->state);
1774 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1776 struct nfs_server *server;
1777 struct nfs4_client *clp = NULL;
1778 struct rpc_xprt *xprt = NULL;
1779 struct rpc_clnt *clnt = NULL;
1780 struct rpc_timeout timeparms;
1781 rpc_authflavor_t authflavour;
1784 sb->s_blocksize_bits = 0;
1785 sb->s_blocksize = 0;
1786 server = NFS_SB(sb);
1787 if (data->rsize != 0)
1788 server->rsize = nfs_block_size(data->rsize, NULL);
1789 if (data->wsize != 0)
1790 server->wsize = nfs_block_size(data->wsize, NULL);
1791 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1792 server->caps = NFS_CAP_ATOMIC_OPEN;
1794 server->acregmin = data->acregmin*HZ;
1795 server->acregmax = data->acregmax*HZ;
1796 server->acdirmin = data->acdirmin*HZ;
1797 server->acdirmax = data->acdirmax*HZ;
1799 server->rpc_ops = &nfs_v4_clientops;
1801 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
1803 clp = nfs4_get_client(&server->addr.sin_addr);
1805 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1809 /* Now create transport and client */
1810 authflavour = RPC_AUTH_UNIX;
1811 if (data->auth_flavourlen != 0) {
1812 if (data->auth_flavourlen != 1) {
1813 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1814 __FUNCTION__, data->auth_flavourlen);
1818 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1824 down_write(&clp->cl_sem);
1825 if (IS_ERR(clp->cl_rpcclient)) {
1826 xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
1828 up_write(&clp->cl_sem);
1829 err = PTR_ERR(xprt);
1830 dprintk("%s: cannot create RPC transport. Error = %d\n",
1834 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1835 server->rpc_ops->version, authflavour);
1837 up_write(&clp->cl_sem);
1838 err = PTR_ERR(clnt);
1839 dprintk("%s: cannot create RPC client. Error = %d\n",
1844 clnt->cl_softrtry = 1;
1845 clp->cl_rpcclient = clnt;
1846 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1849 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1850 clnt = rpc_clone_client(clp->cl_rpcclient);
1852 server->nfs4_state = clp;
1853 up_write(&clp->cl_sem);
1857 err = PTR_ERR(clnt);
1858 dprintk("%s: cannot create RPC client. Error = %d\n",
1863 server->client = clnt;
1865 if (server->nfs4_state->cl_idmap == NULL) {
1866 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1870 if (clnt->cl_auth->au_flavor != authflavour) {
1871 struct rpc_auth *auth;
1873 auth = rpcauth_create(authflavour, clnt);
1875 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1876 return PTR_ERR(auth);
1880 sb->s_time_gran = 1;
1882 sb->s_op = &nfs4_sops;
1883 err = nfs_sb_init(sb, authflavour);
1888 nfs4_put_client(clp);
1892 static int nfs4_compare_super(struct super_block *sb, void *data)
1894 struct nfs_server *server = data;
1895 struct nfs_server *old = NFS_SB(sb);
1897 if (strcmp(server->hostname, old->hostname) != 0)
1899 if (strcmp(server->mnt_path, old->mnt_path) != 0)
1905 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1910 return ERR_PTR(-EINVAL);
1911 if (src->len < maxlen)
1914 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1916 return ERR_PTR(-ENOMEM);
1918 if (copy_from_user(dst, src->data, maxlen)) {
1920 return ERR_PTR(-EFAULT);
1926 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1927 int flags, const char *dev_name, void *raw_data)
1930 struct nfs_server *server;
1931 struct super_block *s;
1932 struct nfs4_mount_data *data = raw_data;
1936 dprintk("%s: missing data argument\n", __FUNCTION__);
1937 return ERR_PTR(-EINVAL);
1939 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1940 dprintk("%s: bad mount version\n", __FUNCTION__);
1941 return ERR_PTR(-EINVAL);
1944 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1946 return ERR_PTR(-ENOMEM);
1947 memset(server, 0, sizeof(struct nfs_server));
1948 /* Zero out the NFS state stuff */
1949 init_nfsv4_state(server);
1950 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1952 p = nfs_copy_user_string(NULL, &data->hostname, 256);
1955 server->hostname = p;
1957 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1960 server->mnt_path = p;
1962 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1963 sizeof(server->ip_addr) - 1);
1967 /* We now require that the mount process passes the remote address */
1968 if (data->host_addrlen != sizeof(server->addr)) {
1969 s = ERR_PTR(-EINVAL);
1972 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1973 s = ERR_PTR(-EFAULT);
1976 if (server->addr.sin_family != AF_INET ||
1977 server->addr.sin_addr.s_addr == INADDR_ANY) {
1978 dprintk("%s: mount program didn't pass remote IP address!\n",
1980 s = ERR_PTR(-EINVAL);
1984 /* Fire up rpciod if not yet running */
1985 s = ERR_PTR(rpciod_up());
1987 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1988 __FUNCTION__, PTR_ERR(s));
1992 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1994 if (IS_ERR(s) || s->s_root)
1999 error = nfs4_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
2001 up_write(&s->s_umount);
2002 deactivate_super(s);
2003 return ERR_PTR(error);
2005 s->s_flags |= MS_ACTIVE;
2008 s = (struct super_block *)p;
2010 kfree(server->mnt_path);
2011 kfree(server->hostname);
2016 static void nfs4_kill_super(struct super_block *sb)
2018 struct nfs_server *server = NFS_SB(sb);
2020 nfs_return_all_delegations(sb);
2021 kill_anon_super(sb);
2023 nfs4_renewd_prepare_shutdown(server);
2025 if (server->client != NULL && !IS_ERR(server->client))
2026 rpc_shutdown_client(server->client);
2027 rpciod_down(); /* release rpciod */
2029 destroy_nfsv4_state(server);
2031 kfree(server->hostname);
2035 static struct file_system_type nfs4_fs_type = {
2036 .owner = THIS_MODULE,
2038 .get_sb = nfs4_get_sb,
2039 .kill_sb = nfs4_kill_super,
2040 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2043 static const int nfs_set_port_min = 0;
2044 static const int nfs_set_port_max = 65535;
2045 static int param_set_port(const char *val, struct kernel_param *kp)
2048 int num = simple_strtol(val, &endp, 0);
2049 if (endp == val || *endp || num < nfs_set_port_min || num > nfs_set_port_max)
2051 *((int *)kp->arg) = num;
2055 module_param_call(callback_tcpport, param_set_port, param_get_int,
2056 &nfs_callback_set_tcpport, 0644);
2058 static int param_set_idmap_timeout(const char *val, struct kernel_param *kp)
2061 int num = simple_strtol(val, &endp, 0);
2063 if (endp == val || *endp || num < 0 || jif < num)
2065 *((int *)kp->arg) = jif;
2069 module_param_call(idmap_cache_timeout, param_set_idmap_timeout, param_get_int,
2070 &nfs_idmap_cache_timeout, 0644);
2072 #define nfs4_init_once(nfsi) \
2074 INIT_LIST_HEAD(&(nfsi)->open_states); \
2075 nfsi->delegation = NULL; \
2076 nfsi->delegation_state = 0; \
2077 init_rwsem(&nfsi->rwsem); \
2080 static inline int register_nfs4fs(void)
2084 ret = nfs_register_sysctl();
2087 ret = register_filesystem(&nfs4_fs_type);
2089 nfs_unregister_sysctl();
2093 static inline void unregister_nfs4fs(void)
2095 unregister_filesystem(&nfs4_fs_type);
2096 nfs_unregister_sysctl();
2099 #define nfs4_init_once(nfsi) \
2101 #define register_nfs4fs() (0)
2102 #define unregister_nfs4fs()
2105 extern int nfs_init_nfspagecache(void);
2106 extern void nfs_destroy_nfspagecache(void);
2107 extern int nfs_init_readpagecache(void);
2108 extern void nfs_destroy_readpagecache(void);
2109 extern int nfs_init_writepagecache(void);
2110 extern void nfs_destroy_writepagecache(void);
2111 #ifdef CONFIG_NFS_DIRECTIO
2112 extern int nfs_init_directcache(void);
2113 extern void nfs_destroy_directcache(void);
2116 static kmem_cache_t * nfs_inode_cachep;
2118 static struct inode *nfs_alloc_inode(struct super_block *sb)
2120 struct nfs_inode *nfsi;
2121 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2125 nfsi->cache_validity = 0UL;
2126 nfsi->cache_change_attribute = jiffies;
2127 #ifdef CONFIG_NFS_V3_ACL
2128 nfsi->acl_access = ERR_PTR(-EAGAIN);
2129 nfsi->acl_default = ERR_PTR(-EAGAIN);
2131 #ifdef CONFIG_NFS_V4
2132 nfsi->nfs4_acl = NULL;
2133 #endif /* CONFIG_NFS_V4 */
2134 return &nfsi->vfs_inode;
2137 static void nfs_destroy_inode(struct inode *inode)
2139 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2142 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2144 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2146 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2147 SLAB_CTOR_CONSTRUCTOR) {
2148 inode_init_once(&nfsi->vfs_inode);
2149 spin_lock_init(&nfsi->req_lock);
2150 INIT_LIST_HEAD(&nfsi->dirty);
2151 INIT_LIST_HEAD(&nfsi->commit);
2152 INIT_LIST_HEAD(&nfsi->open_files);
2153 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2154 atomic_set(&nfsi->data_updates, 0);
2158 nfs4_init_once(nfsi);
2162 static int nfs_init_inodecache(void)
2164 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2165 sizeof(struct nfs_inode),
2166 0, (SLAB_RECLAIM_ACCOUNT|
2169 if (nfs_inode_cachep == NULL)
2175 static void nfs_destroy_inodecache(void)
2177 if (kmem_cache_destroy(nfs_inode_cachep))
2178 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2184 static int __init init_nfs_fs(void)
2188 err = nfs_init_nfspagecache();
2192 err = nfs_init_inodecache();
2196 err = nfs_init_readpagecache();
2200 err = nfs_init_writepagecache();
2204 #ifdef CONFIG_NFS_DIRECTIO
2205 err = nfs_init_directcache();
2210 #ifdef CONFIG_PROC_FS
2211 rpc_proc_register(&nfs_rpcstat);
2213 err = register_filesystem(&nfs_fs_type);
2216 if ((err = register_nfs4fs()) != 0)
2220 #ifdef CONFIG_PROC_FS
2221 rpc_proc_unregister("nfs");
2223 #ifdef CONFIG_NFS_DIRECTIO
2224 nfs_destroy_directcache();
2227 nfs_destroy_writepagecache();
2229 nfs_destroy_readpagecache();
2231 nfs_destroy_inodecache();
2233 nfs_destroy_nfspagecache();
2238 static void __exit exit_nfs_fs(void)
2240 #ifdef CONFIG_NFS_DIRECTIO
2241 nfs_destroy_directcache();
2243 nfs_destroy_writepagecache();
2244 nfs_destroy_readpagecache();
2245 nfs_destroy_inodecache();
2246 nfs_destroy_nfspagecache();
2247 #ifdef CONFIG_PROC_FS
2248 rpc_proc_unregister("nfs");
2250 unregister_filesystem(&nfs_fs_type);
2251 unregister_nfs4fs();
2254 /* Not quite true; I just maintain it */
2255 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2256 MODULE_LICENSE("GPL");
2258 module_init(init_nfs_fs)
2259 module_exit(exit_nfs_fs)