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 *);
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)
224 if (bsize < NFS_MIN_FILE_IO_SIZE)
225 bsize = NFS_DEF_FILE_IO_SIZE;
226 else if (bsize >= NFS_MAX_FILE_IO_SIZE)
227 bsize = NFS_MAX_FILE_IO_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->rsize > NFS_MAX_FILE_IO_SIZE)
311 server->rsize = NFS_MAX_FILE_IO_SIZE;
312 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
314 if (server->wsize > max_rpc_payload)
315 server->wsize = max_rpc_payload;
316 if (server->wsize > NFS_MAX_FILE_IO_SIZE)
317 server->wsize = NFS_MAX_FILE_IO_SIZE;
318 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
320 if (sb->s_blocksize == 0)
321 sb->s_blocksize = nfs_block_bits(server->wsize,
322 &sb->s_blocksize_bits);
323 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
325 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
326 if (server->dtsize > PAGE_CACHE_SIZE)
327 server->dtsize = PAGE_CACHE_SIZE;
328 if (server->dtsize > server->rsize)
329 server->dtsize = server->rsize;
331 if (server->flags & NFS_MOUNT_NOAC) {
332 server->acregmin = server->acregmax = 0;
333 server->acdirmin = server->acdirmax = 0;
334 sb->s_flags |= MS_SYNCHRONOUS;
336 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
338 sb->s_maxbytes = fsinfo.maxfilesize;
339 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
340 sb->s_maxbytes = MAX_LFS_FILESIZE;
342 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
343 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
345 /* We're airborne Set socket buffersize */
346 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
348 /* Yargs. It didn't work out. */
350 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
351 if (!IS_ERR(root_inode))
353 return no_root_error;
356 static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
358 to->to_initval = timeo * HZ / 10;
359 to->to_retries = retrans;
366 to->to_initval = 60 * HZ;
367 if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
368 to->to_initval = NFS_MAX_TCP_TIMEOUT;
369 to->to_increment = to->to_initval;
370 to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
371 to->to_exponential = 0;
376 to->to_initval = 11 * HZ / 10;
377 if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
378 to->to_initval = NFS_MAX_UDP_TIMEOUT;
379 to->to_maxval = NFS_MAX_UDP_TIMEOUT;
380 to->to_exponential = 1;
386 * Create an RPC client handle.
388 static struct rpc_clnt *
389 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
391 struct rpc_timeout timeparms;
392 struct rpc_xprt *xprt = NULL;
393 struct rpc_clnt *clnt = NULL;
394 int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
396 nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
398 /* create transport and client */
399 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
401 dprintk("%s: cannot create RPC transport. Error = %ld\n",
402 __FUNCTION__, PTR_ERR(xprt));
403 return (struct rpc_clnt *)xprt;
405 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
406 server->rpc_ops->version, data->pseudoflavor);
408 dprintk("%s: cannot create RPC client. Error = %ld\n",
409 __FUNCTION__, PTR_ERR(xprt));
414 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_fdatawrite(mapping);
650 ret = filemap_fdatawait(mapping);
653 ret = nfs_wb_all(mapping->host);
659 * Invalidate the local caches
661 static void nfs_zap_caches_locked(struct inode *inode)
663 struct nfs_inode *nfsi = NFS_I(inode);
664 int mode = inode->i_mode;
666 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
667 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
669 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
670 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
671 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
673 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
676 void nfs_zap_caches(struct inode *inode)
678 spin_lock(&inode->i_lock);
679 nfs_zap_caches_locked(inode);
680 spin_unlock(&inode->i_lock);
683 static void nfs_zap_acl_cache(struct inode *inode)
685 void (*clear_acl_cache)(struct inode *);
687 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
688 if (clear_acl_cache != NULL)
689 clear_acl_cache(inode);
690 spin_lock(&inode->i_lock);
691 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
692 spin_unlock(&inode->i_lock);
696 * Invalidate, but do not unhash, the inode.
697 * NB: must be called with inode->i_lock held!
699 static void nfs_invalidate_inode(struct inode *inode)
701 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
702 nfs_zap_caches_locked(inode);
705 struct nfs_find_desc {
707 struct nfs_fattr *fattr;
711 * In NFSv3 we can have 64bit inode numbers. In order to support
712 * this, and re-exported directories (also seen in NFSv2)
713 * we are forced to allow 2 different inodes to have the same
717 nfs_find_actor(struct inode *inode, void *opaque)
719 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
720 struct nfs_fh *fh = desc->fh;
721 struct nfs_fattr *fattr = desc->fattr;
723 if (NFS_FILEID(inode) != fattr->fileid)
725 if (nfs_compare_fh(NFS_FH(inode), fh))
727 if (is_bad_inode(inode) || NFS_STALE(inode))
733 nfs_init_locked(struct inode *inode, void *opaque)
735 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
736 struct nfs_fattr *fattr = desc->fattr;
738 NFS_FILEID(inode) = fattr->fileid;
739 nfs_copy_fh(NFS_FH(inode), desc->fh);
743 /* Don't use READDIRPLUS on directories that we believe are too large */
744 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
747 * This is our front-end to iget that looks up inodes by file handle
748 * instead of inode number.
751 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
753 struct nfs_find_desc desc = {
757 struct inode *inode = NULL;
760 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
764 printk("NFS: Buggy server - nlink == 0!\n");
768 hash = nfs_fattr_to_ino_t(fattr);
770 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
773 if (inode->i_state & I_NEW) {
774 struct nfs_inode *nfsi = NFS_I(inode);
776 /* We set i_ino for the few things that still rely on it,
780 /* We can't support update_atime(), since the server will reset it */
781 inode->i_flags |= S_NOATIME|S_NOCMTIME;
782 inode->i_mode = fattr->mode;
783 /* Why so? Because we want revalidate for devices/FIFOs, and
784 * that's precisely what we have in nfs_file_inode_operations.
786 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
787 if (S_ISREG(inode->i_mode)) {
788 inode->i_fop = &nfs_file_operations;
789 inode->i_data.a_ops = &nfs_file_aops;
790 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
791 } else if (S_ISDIR(inode->i_mode)) {
792 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
793 inode->i_fop = &nfs_dir_operations;
794 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
795 && fattr->size <= NFS_LIMIT_READDIRPLUS)
796 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
797 } else if (S_ISLNK(inode->i_mode))
798 inode->i_op = &nfs_symlink_inode_operations;
800 init_special_inode(inode, inode->i_mode, fattr->rdev);
802 nfsi->read_cache_jiffies = fattr->time_start;
803 nfsi->last_updated = jiffies;
804 inode->i_atime = fattr->atime;
805 inode->i_mtime = fattr->mtime;
806 inode->i_ctime = fattr->ctime;
807 if (fattr->valid & NFS_ATTR_FATTR_V4)
808 nfsi->change_attr = fattr->change_attr;
809 inode->i_size = nfs_size_to_loff_t(fattr->size);
810 inode->i_nlink = fattr->nlink;
811 inode->i_uid = fattr->uid;
812 inode->i_gid = fattr->gid;
813 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
815 * report the blocks in 512byte units
817 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
818 inode->i_blksize = inode->i_sb->s_blocksize;
820 inode->i_blocks = fattr->du.nfs2.blocks;
821 inode->i_blksize = fattr->du.nfs2.blocksize;
823 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
824 nfsi->attrtimeo_timestamp = jiffies;
825 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
826 nfsi->cache_access.cred = NULL;
828 unlock_new_inode(inode);
830 nfs_refresh_inode(inode, fattr);
831 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
833 (long long)NFS_FILEID(inode),
834 atomic_read(&inode->i_count));
840 printk("nfs_fhget: iget failed\n");
844 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
847 nfs_setattr(struct dentry *dentry, struct iattr *attr)
849 struct inode *inode = dentry->d_inode;
850 struct nfs_fattr fattr;
853 if (attr->ia_valid & ATTR_SIZE) {
854 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
855 attr->ia_valid &= ~ATTR_SIZE;
858 /* Optimization: if the end result is no change, don't RPC */
859 attr->ia_valid &= NFS_VALID_ATTRS;
860 if (attr->ia_valid == 0)
864 nfs_begin_data_update(inode);
865 /* Write all dirty data if we're changing file permissions or size */
866 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
867 if (filemap_fdatawrite(inode->i_mapping) == 0)
868 filemap_fdatawait(inode->i_mapping);
872 * Return any delegations if we're going to change ACLs
874 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
875 nfs_inode_return_delegation(inode);
876 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
878 nfs_refresh_inode(inode, &fattr);
879 nfs_end_data_update(inode);
885 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
886 * @inode: pointer to struct inode
887 * @attr: pointer to struct iattr
889 * Note: we do this in the *proc.c in order to ensure that
890 * it works for things like exclusive creates too.
892 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
894 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
895 if ((attr->ia_valid & ATTR_MODE) != 0) {
896 int mode = attr->ia_mode & S_IALLUGO;
897 mode |= inode->i_mode & ~S_IALLUGO;
898 inode->i_mode = mode;
900 if ((attr->ia_valid & ATTR_UID) != 0)
901 inode->i_uid = attr->ia_uid;
902 if ((attr->ia_valid & ATTR_GID) != 0)
903 inode->i_gid = attr->ia_gid;
904 spin_lock(&inode->i_lock);
905 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
906 spin_unlock(&inode->i_lock);
908 if ((attr->ia_valid & ATTR_SIZE) != 0) {
909 inode->i_size = attr->ia_size;
910 vmtruncate(inode, attr->ia_size);
914 static int nfs_wait_schedule(void *word)
916 if (signal_pending(current))
923 * Wait for the inode to get unlocked.
925 static int nfs_wait_on_inode(struct inode *inode)
927 struct rpc_clnt *clnt = NFS_CLIENT(inode);
928 struct nfs_inode *nfsi = NFS_I(inode);
932 rpc_clnt_sigmask(clnt, &oldmask);
933 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
934 nfs_wait_schedule, TASK_INTERRUPTIBLE);
935 rpc_clnt_sigunmask(clnt, &oldmask);
940 static void nfs_wake_up_inode(struct inode *inode)
942 struct nfs_inode *nfsi = NFS_I(inode);
944 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
945 smp_mb__after_clear_bit();
946 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
949 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
951 struct inode *inode = dentry->d_inode;
952 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
955 /* Flush out writes to the server in order to update c/mtime */
956 nfs_sync_inode(inode, 0, 0, FLUSH_WAIT|FLUSH_NOCOMMIT);
957 if (__IS_FLG(inode, MS_NOATIME))
959 else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
961 /* We may force a getattr if the user cares about atime */
963 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
965 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
967 generic_fillattr(inode, stat);
971 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
973 struct nfs_open_context *ctx;
975 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
977 atomic_set(&ctx->count, 1);
978 ctx->dentry = dget(dentry);
979 ctx->cred = get_rpccred(cred);
981 ctx->lockowner = current->files;
988 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
991 atomic_inc(&ctx->count);
995 void put_nfs_open_context(struct nfs_open_context *ctx)
997 if (atomic_dec_and_test(&ctx->count)) {
998 if (!list_empty(&ctx->list)) {
999 struct inode *inode = ctx->dentry->d_inode;
1000 spin_lock(&inode->i_lock);
1001 list_del(&ctx->list);
1002 spin_unlock(&inode->i_lock);
1004 if (ctx->state != NULL)
1005 nfs4_close_state(ctx->state, ctx->mode);
1006 if (ctx->cred != NULL)
1007 put_rpccred(ctx->cred);
1014 * Ensure that mmap has a recent RPC credential for use when writing out
1017 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1019 struct inode *inode = filp->f_dentry->d_inode;
1020 struct nfs_inode *nfsi = NFS_I(inode);
1022 filp->private_data = get_nfs_open_context(ctx);
1023 spin_lock(&inode->i_lock);
1024 list_add(&ctx->list, &nfsi->open_files);
1025 spin_unlock(&inode->i_lock);
1029 * Given an inode, search for an open context with the desired characteristics
1031 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
1033 struct nfs_inode *nfsi = NFS_I(inode);
1034 struct nfs_open_context *pos, *ctx = NULL;
1036 spin_lock(&inode->i_lock);
1037 list_for_each_entry(pos, &nfsi->open_files, list) {
1038 if (cred != NULL && pos->cred != cred)
1040 if ((pos->mode & mode) == mode) {
1041 ctx = get_nfs_open_context(pos);
1045 spin_unlock(&inode->i_lock);
1049 void nfs_file_clear_open_context(struct file *filp)
1051 struct inode *inode = filp->f_dentry->d_inode;
1052 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
1055 filp->private_data = NULL;
1056 spin_lock(&inode->i_lock);
1057 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1058 spin_unlock(&inode->i_lock);
1059 put_nfs_open_context(ctx);
1064 * These allocate and release file read/write context information.
1066 int nfs_open(struct inode *inode, struct file *filp)
1068 struct nfs_open_context *ctx;
1069 struct rpc_cred *cred;
1071 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1073 return PTR_ERR(cred);
1074 ctx = alloc_nfs_open_context(filp->f_dentry, cred);
1078 ctx->mode = filp->f_mode;
1079 nfs_file_set_open_context(filp, ctx);
1080 put_nfs_open_context(ctx);
1084 int nfs_release(struct inode *inode, struct file *filp)
1086 nfs_file_clear_open_context(filp);
1091 * This function is called whenever some part of NFS notices that
1092 * the cached attributes have to be refreshed.
1095 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1097 int status = -ESTALE;
1098 struct nfs_fattr fattr;
1099 struct nfs_inode *nfsi = NFS_I(inode);
1101 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1102 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1105 if (!inode || is_bad_inode(inode))
1107 if (NFS_STALE(inode))
1110 status = nfs_wait_on_inode(inode);
1113 if (NFS_STALE(inode)) {
1115 /* Do we trust the cached ESTALE? */
1116 if (NFS_ATTRTIMEO(inode) != 0) {
1117 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1124 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1126 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1128 (long long)NFS_FILEID(inode), status);
1129 if (status == -ESTALE) {
1130 nfs_zap_caches(inode);
1131 if (!S_ISDIR(inode->i_mode))
1132 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1137 spin_lock(&inode->i_lock);
1138 status = nfs_update_inode(inode, &fattr);
1140 spin_unlock(&inode->i_lock);
1141 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1143 (long long)NFS_FILEID(inode), status);
1146 spin_unlock(&inode->i_lock);
1148 nfs_revalidate_mapping(inode, inode->i_mapping);
1150 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1151 nfs_zap_acl_cache(inode);
1153 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1155 (long long)NFS_FILEID(inode));
1158 nfs_wake_up_inode(inode);
1165 int nfs_attribute_timeout(struct inode *inode)
1167 struct nfs_inode *nfsi = NFS_I(inode);
1169 if (nfs_have_delegation(inode, FMODE_READ))
1171 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1175 * nfs_revalidate_inode - Revalidate the inode attributes
1176 * @server - pointer to nfs_server struct
1177 * @inode - pointer to inode struct
1179 * Updates inode attribute information by retrieving the data from the server.
1181 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1183 if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1184 && !nfs_attribute_timeout(inode))
1185 return NFS_STALE(inode) ? -ESTALE : 0;
1186 return __nfs_revalidate_inode(server, inode);
1190 * nfs_revalidate_mapping - Revalidate the pagecache
1191 * @inode - pointer to host inode
1192 * @mapping - pointer to mapping
1194 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1196 struct nfs_inode *nfsi = NFS_I(inode);
1198 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1199 if (S_ISREG(inode->i_mode))
1200 nfs_sync_mapping(mapping);
1201 invalidate_inode_pages2(mapping);
1203 spin_lock(&inode->i_lock);
1204 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1205 if (S_ISDIR(inode->i_mode)) {
1206 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1207 /* This ensures we revalidate child dentries */
1208 nfsi->cache_change_attribute = jiffies;
1210 spin_unlock(&inode->i_lock);
1212 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1214 (long long)NFS_FILEID(inode));
1219 * nfs_begin_data_update
1220 * @inode - pointer to inode
1221 * Declare that a set of operations will update file data on the server
1223 void nfs_begin_data_update(struct inode *inode)
1225 atomic_inc(&NFS_I(inode)->data_updates);
1229 * nfs_end_data_update
1230 * @inode - pointer to inode
1231 * Declare end of the operations that will update file data
1232 * This will mark the inode as immediately needing revalidation
1233 * of its attribute cache.
1235 void nfs_end_data_update(struct inode *inode)
1237 struct nfs_inode *nfsi = NFS_I(inode);
1239 if (!nfs_have_delegation(inode, FMODE_READ)) {
1240 /* Directories and symlinks: invalidate page cache */
1241 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
1242 spin_lock(&inode->i_lock);
1243 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1244 spin_unlock(&inode->i_lock);
1247 nfsi->cache_change_attribute = jiffies;
1248 atomic_dec(&nfsi->data_updates);
1251 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1253 struct nfs_inode *nfsi = NFS_I(inode);
1255 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1256 && nfsi->change_attr == fattr->pre_change_attr) {
1257 nfsi->change_attr = fattr->change_attr;
1258 nfsi->cache_change_attribute = jiffies;
1261 /* If we have atomic WCC data, we may update some attributes */
1262 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1263 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1264 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1265 nfsi->cache_change_attribute = jiffies;
1267 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1268 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1269 nfsi->cache_change_attribute = jiffies;
1271 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
1272 inode->i_size = fattr->size;
1273 nfsi->cache_change_attribute = jiffies;
1279 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1280 * @inode - pointer to inode
1281 * @fattr - updated attributes
1283 * Verifies the attribute cache. If we have just changed the attributes,
1284 * so that fattr carries weak cache consistency data, then it may
1285 * also update the ctime/mtime/change_attribute.
1287 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1289 struct nfs_inode *nfsi = NFS_I(inode);
1290 loff_t cur_size, new_isize;
1294 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1297 /* Are we in the process of updating data on the server? */
1298 data_unstable = nfs_caches_unstable(inode);
1300 /* Do atomic weak cache consistency updates */
1301 nfs_wcc_update_inode(inode, fattr);
1303 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1304 nfsi->change_attr != fattr->change_attr) {
1305 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1307 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1310 /* Has the inode gone and changed behind our back? */
1311 if (nfsi->fileid != fattr->fileid
1312 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1316 cur_size = i_size_read(inode);
1317 new_isize = nfs_size_to_loff_t(fattr->size);
1319 /* Verify a few of the more important attributes */
1320 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1321 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1323 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1325 if (cur_size != new_isize) {
1326 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1327 if (nfsi->npages == 0)
1328 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1331 /* Have any file permissions changed? */
1332 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1333 || inode->i_uid != fattr->uid
1334 || inode->i_gid != fattr->gid)
1335 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1337 /* Has the link count changed? */
1338 if (inode->i_nlink != fattr->nlink)
1339 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1341 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1342 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1344 nfsi->read_cache_jiffies = fattr->time_start;
1349 * nfs_refresh_inode - try to update the inode attribute cache
1350 * @inode - pointer to inode
1351 * @fattr - updated attributes
1353 * Check that an RPC call that returned attributes has not overlapped with
1354 * other recent updates of the inode metadata, then decide whether it is
1355 * safe to do a full update of the inode attributes, or whether just to
1356 * call nfs_check_inode_attributes.
1358 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1360 struct nfs_inode *nfsi = NFS_I(inode);
1363 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1365 spin_lock(&inode->i_lock);
1366 nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1367 if (time_after(fattr->time_start, nfsi->last_updated))
1368 status = nfs_update_inode(inode, fattr);
1370 status = nfs_check_inode_attributes(inode, fattr);
1372 spin_unlock(&inode->i_lock);
1377 * nfs_post_op_update_inode - try to update the inode attribute cache
1378 * @inode - pointer to inode
1379 * @fattr - updated attributes
1381 * After an operation that has changed the inode metadata, mark the
1382 * attribute cache as being invalid, then try to update it.
1384 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1386 struct nfs_inode *nfsi = NFS_I(inode);
1389 spin_lock(&inode->i_lock);
1390 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
1391 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1394 status = nfs_update_inode(inode, fattr);
1396 spin_unlock(&inode->i_lock);
1401 * Many nfs protocol calls return the new file attributes after
1402 * an operation. Here we update the inode to reflect the state
1403 * of the server's inode.
1405 * This is a bit tricky because we have to make sure all dirty pages
1406 * have been sent off to the server before calling invalidate_inode_pages.
1407 * To make sure no other process adds more write requests while we try
1408 * our best to flush them, we make them sleep during the attribute refresh.
1410 * A very similar scenario holds for the dir cache.
1412 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1414 struct nfs_inode *nfsi = NFS_I(inode);
1415 loff_t cur_isize, new_isize;
1416 unsigned int invalid = 0;
1419 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1420 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1421 atomic_read(&inode->i_count), fattr->valid);
1423 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1426 if (nfsi->fileid != fattr->fileid)
1430 * Make sure the inode's type hasn't changed.
1432 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1436 * Update the read time so we don't revalidate too often.
1438 nfsi->read_cache_jiffies = fattr->time_start;
1439 nfsi->last_updated = jiffies;
1441 /* Are we racing with known updates of the metadata on the server? */
1442 data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
1444 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1446 /* Do atomic weak cache consistency updates */
1447 nfs_wcc_update_inode(inode, fattr);
1449 /* Check if our cached file size is stale */
1450 new_isize = nfs_size_to_loff_t(fattr->size);
1451 cur_isize = i_size_read(inode);
1452 if (new_isize != cur_isize) {
1453 /* Do we perhaps have any outstanding writes? */
1454 if (nfsi->npages == 0) {
1455 /* No, but did we race with nfs_end_data_update()? */
1457 inode->i_size = new_isize;
1458 invalid |= NFS_INO_INVALID_DATA;
1460 invalid |= NFS_INO_INVALID_ATTR;
1461 } else if (new_isize > cur_isize) {
1462 inode->i_size = new_isize;
1463 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1465 nfsi->cache_change_attribute = jiffies;
1466 dprintk("NFS: isize change on server for file %s/%ld\n",
1467 inode->i_sb->s_id, inode->i_ino);
1470 /* Check if the mtime agrees */
1471 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1472 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1473 dprintk("NFS: mtime change on server for file %s/%ld\n",
1474 inode->i_sb->s_id, inode->i_ino);
1475 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1476 nfsi->cache_change_attribute = jiffies;
1479 if ((fattr->valid & NFS_ATTR_FATTR_V4)
1480 && nfsi->change_attr != fattr->change_attr) {
1481 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1482 inode->i_sb->s_id, inode->i_ino);
1483 nfsi->change_attr = fattr->change_attr;
1484 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1485 nfsi->cache_change_attribute = jiffies;
1488 /* If ctime has changed we should definitely clear access+acl caches */
1489 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1490 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1491 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1492 nfsi->cache_change_attribute = jiffies;
1494 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1496 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1497 inode->i_uid != fattr->uid ||
1498 inode->i_gid != fattr->gid)
1499 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1501 inode->i_mode = fattr->mode;
1502 inode->i_nlink = fattr->nlink;
1503 inode->i_uid = fattr->uid;
1504 inode->i_gid = fattr->gid;
1506 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1508 * report the blocks in 512byte units
1510 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1511 inode->i_blksize = inode->i_sb->s_blocksize;
1513 inode->i_blocks = fattr->du.nfs2.blocks;
1514 inode->i_blksize = fattr->du.nfs2.blocksize;
1517 /* Update attrtimeo value if we're out of the unstable period */
1518 if (invalid & NFS_INO_INVALID_ATTR) {
1519 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1520 nfsi->attrtimeo_timestamp = jiffies;
1521 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1522 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1523 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1524 nfsi->attrtimeo_timestamp = jiffies;
1526 /* Don't invalidate the data if we were to blame */
1527 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1528 || S_ISLNK(inode->i_mode)))
1529 invalid &= ~NFS_INO_INVALID_DATA;
1531 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1532 if (!nfs_have_delegation(inode, FMODE_READ))
1533 nfsi->cache_validity |= invalid;
1538 * Big trouble! The inode has become a different object.
1541 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1542 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1546 * No need to worry about unhashing the dentry, as the
1547 * lookup validation will know that the inode is bad.
1548 * (But we fall through to invalidate the caches.)
1550 nfs_invalidate_inode(inode);
1554 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1555 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1556 NFS_SERVER(inode)->hostname, inode->i_sb->s_id,
1557 (long long)nfsi->fileid, (long long)fattr->fileid);
1562 * File system information
1565 static int nfs_set_super(struct super_block *s, void *data)
1567 s->s_fs_info = data;
1568 return set_anon_super(s, data);
1571 static int nfs_compare_super(struct super_block *sb, void *data)
1573 struct nfs_server *server = data;
1574 struct nfs_server *old = NFS_SB(sb);
1576 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1578 if (old->addr.sin_port != server->addr.sin_port)
1580 return !nfs_compare_fh(&old->fh, &server->fh);
1583 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1584 int flags, const char *dev_name, void *raw_data)
1587 struct nfs_server *server = NULL;
1588 struct super_block *s;
1589 struct nfs_fh *root;
1590 struct nfs_mount_data *data = raw_data;
1592 s = ERR_PTR(-EINVAL);
1594 dprintk("%s: missing data argument\n", __FUNCTION__);
1597 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1598 dprintk("%s: bad mount version\n", __FUNCTION__);
1601 switch (data->version) {
1607 if (data->flags & NFS_MOUNT_VER3) {
1608 dprintk("%s: mount structure version %d does not support NFSv3\n",
1613 data->root.size = NFS2_FHSIZE;
1614 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1616 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1617 dprintk("%s: mount structure version %d does not support strong security\n",
1623 memset(data->context, 0, sizeof(data->context));
1625 #ifndef CONFIG_NFS_V3
1626 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1627 s = ERR_PTR(-EPROTONOSUPPORT);
1628 if (data->flags & NFS_MOUNT_VER3) {
1629 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1632 #endif /* CONFIG_NFS_V3 */
1634 s = ERR_PTR(-ENOMEM);
1635 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1638 memset(server, 0, sizeof(struct nfs_server));
1639 /* Zero out the NFS state stuff */
1640 init_nfsv4_state(server);
1641 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1644 if (data->flags & NFS_MOUNT_VER3)
1645 root->size = data->root.size;
1647 root->size = NFS2_FHSIZE;
1648 s = ERR_PTR(-EINVAL);
1649 if (root->size > sizeof(root->data)) {
1650 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1653 memcpy(root->data, data->root.data, root->size);
1655 /* We now require that the mount process passes the remote address */
1656 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1657 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1658 dprintk("%s: mount program didn't pass remote address!\n",
1663 /* Fire up rpciod if not yet running */
1664 s = ERR_PTR(rpciod_up());
1666 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1667 __FUNCTION__, PTR_ERR(s));
1671 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1672 if (IS_ERR(s) || s->s_root)
1673 goto out_rpciod_down;
1677 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1679 up_write(&s->s_umount);
1680 deactivate_super(s);
1681 return ERR_PTR(error);
1683 s->s_flags |= MS_ACTIVE;
1692 static void nfs_kill_super(struct super_block *s)
1694 struct nfs_server *server = NFS_SB(s);
1698 if (!IS_ERR(server->client))
1699 rpc_shutdown_client(server->client);
1700 if (!IS_ERR(server->client_sys))
1701 rpc_shutdown_client(server->client_sys);
1702 if (!IS_ERR(server->client_acl))
1703 rpc_shutdown_client(server->client_acl);
1705 if (!(server->flags & NFS_MOUNT_NONLM))
1706 lockd_down(); /* release rpc.lockd */
1708 rpciod_down(); /* release rpciod */
1710 kfree(server->hostname);
1714 static struct file_system_type nfs_fs_type = {
1715 .owner = THIS_MODULE,
1717 .get_sb = nfs_get_sb,
1718 .kill_sb = nfs_kill_super,
1719 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1722 #ifdef CONFIG_NFS_V4
1724 static void nfs4_clear_inode(struct inode *);
1727 static struct super_operations nfs4_sops = {
1728 .alloc_inode = nfs_alloc_inode,
1729 .destroy_inode = nfs_destroy_inode,
1730 .write_inode = nfs_write_inode,
1731 .delete_inode = nfs_delete_inode,
1732 .statfs = nfs_statfs,
1733 .clear_inode = nfs4_clear_inode,
1734 .umount_begin = nfs_umount_begin,
1735 .show_options = nfs_show_options,
1739 * Clean out any remaining NFSv4 state that might be left over due
1740 * to open() calls that passed nfs_atomic_lookup, but failed to call
1743 static void nfs4_clear_inode(struct inode *inode)
1745 struct nfs_inode *nfsi = NFS_I(inode);
1747 /* If we are holding a delegation, return it! */
1748 nfs_inode_return_delegation(inode);
1749 /* First call standard NFS clear_inode() code */
1750 nfs_clear_inode(inode);
1751 /* Now clear out any remaining state */
1752 while (!list_empty(&nfsi->open_states)) {
1753 struct nfs4_state *state;
1755 state = list_entry(nfsi->open_states.next,
1758 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1761 (long long)NFS_FILEID(inode),
1763 BUG_ON(atomic_read(&state->count) != 1);
1764 nfs4_close_state(state, state->state);
1769 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1771 struct nfs_server *server;
1772 struct nfs4_client *clp = NULL;
1773 struct rpc_xprt *xprt = NULL;
1774 struct rpc_clnt *clnt = NULL;
1775 struct rpc_timeout timeparms;
1776 rpc_authflavor_t authflavour;
1779 sb->s_blocksize_bits = 0;
1780 sb->s_blocksize = 0;
1781 server = NFS_SB(sb);
1782 if (data->rsize != 0)
1783 server->rsize = nfs_block_size(data->rsize, NULL);
1784 if (data->wsize != 0)
1785 server->wsize = nfs_block_size(data->wsize, NULL);
1786 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1787 server->caps = NFS_CAP_ATOMIC_OPEN;
1789 server->acregmin = data->acregmin*HZ;
1790 server->acregmax = data->acregmax*HZ;
1791 server->acdirmin = data->acdirmin*HZ;
1792 server->acdirmax = data->acdirmax*HZ;
1794 server->rpc_ops = &nfs_v4_clientops;
1796 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
1798 clp = nfs4_get_client(&server->addr.sin_addr);
1800 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1804 /* Now create transport and client */
1805 authflavour = RPC_AUTH_UNIX;
1806 if (data->auth_flavourlen != 0) {
1807 if (data->auth_flavourlen != 1) {
1808 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1809 __FUNCTION__, data->auth_flavourlen);
1813 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1819 down_write(&clp->cl_sem);
1820 if (IS_ERR(clp->cl_rpcclient)) {
1821 xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
1823 up_write(&clp->cl_sem);
1824 err = PTR_ERR(xprt);
1825 dprintk("%s: cannot create RPC transport. Error = %d\n",
1829 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1830 server->rpc_ops->version, authflavour);
1832 up_write(&clp->cl_sem);
1833 err = PTR_ERR(clnt);
1834 dprintk("%s: cannot create RPC client. Error = %d\n",
1839 clnt->cl_softrtry = 1;
1840 clnt->cl_chatty = 1;
1841 clp->cl_rpcclient = clnt;
1842 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1845 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1846 clnt = rpc_clone_client(clp->cl_rpcclient);
1848 server->nfs4_state = clp;
1849 up_write(&clp->cl_sem);
1853 err = PTR_ERR(clnt);
1854 dprintk("%s: cannot create RPC client. Error = %d\n",
1859 server->client = clnt;
1861 if (server->nfs4_state->cl_idmap == NULL) {
1862 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1866 if (clnt->cl_auth->au_flavor != authflavour) {
1867 struct rpc_auth *auth;
1869 auth = rpcauth_create(authflavour, clnt);
1871 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1872 return PTR_ERR(auth);
1876 sb->s_time_gran = 1;
1878 sb->s_op = &nfs4_sops;
1879 err = nfs_sb_init(sb, authflavour);
1884 nfs4_put_client(clp);
1888 static int nfs4_compare_super(struct super_block *sb, void *data)
1890 struct nfs_server *server = data;
1891 struct nfs_server *old = NFS_SB(sb);
1893 if (strcmp(server->hostname, old->hostname) != 0)
1895 if (strcmp(server->mnt_path, old->mnt_path) != 0)
1901 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
1906 return ERR_PTR(-EINVAL);
1907 if (src->len < maxlen)
1910 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
1912 return ERR_PTR(-ENOMEM);
1914 if (copy_from_user(dst, src->data, maxlen)) {
1916 return ERR_PTR(-EFAULT);
1922 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
1923 int flags, const char *dev_name, void *raw_data)
1926 struct nfs_server *server;
1927 struct super_block *s;
1928 struct nfs4_mount_data *data = raw_data;
1932 dprintk("%s: missing data argument\n", __FUNCTION__);
1933 return ERR_PTR(-EINVAL);
1935 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
1936 dprintk("%s: bad mount version\n", __FUNCTION__);
1937 return ERR_PTR(-EINVAL);
1940 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1942 return ERR_PTR(-ENOMEM);
1943 memset(server, 0, sizeof(struct nfs_server));
1944 /* Zero out the NFS state stuff */
1945 init_nfsv4_state(server);
1946 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1948 p = nfs_copy_user_string(NULL, &data->hostname, 256);
1951 server->hostname = p;
1953 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
1956 server->mnt_path = p;
1958 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
1959 sizeof(server->ip_addr) - 1);
1963 /* We now require that the mount process passes the remote address */
1964 if (data->host_addrlen != sizeof(server->addr)) {
1965 s = ERR_PTR(-EINVAL);
1968 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
1969 s = ERR_PTR(-EFAULT);
1972 if (server->addr.sin_family != AF_INET ||
1973 server->addr.sin_addr.s_addr == INADDR_ANY) {
1974 dprintk("%s: mount program didn't pass remote IP address!\n",
1976 s = ERR_PTR(-EINVAL);
1980 /* Fire up rpciod if not yet running */
1981 s = ERR_PTR(rpciod_up());
1983 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1984 __FUNCTION__, PTR_ERR(s));
1988 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
1990 if (IS_ERR(s) || s->s_root)
1995 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1997 up_write(&s->s_umount);
1998 deactivate_super(s);
1999 return ERR_PTR(error);
2001 s->s_flags |= MS_ACTIVE;
2004 s = (struct super_block *)p;
2006 kfree(server->mnt_path);
2007 kfree(server->hostname);
2012 static void nfs4_kill_super(struct super_block *sb)
2014 struct nfs_server *server = NFS_SB(sb);
2016 nfs_return_all_delegations(sb);
2017 kill_anon_super(sb);
2019 nfs4_renewd_prepare_shutdown(server);
2021 if (server->client != NULL && !IS_ERR(server->client))
2022 rpc_shutdown_client(server->client);
2023 rpciod_down(); /* release rpciod */
2025 destroy_nfsv4_state(server);
2027 kfree(server->hostname);
2031 static struct file_system_type nfs4_fs_type = {
2032 .owner = THIS_MODULE,
2034 .get_sb = nfs4_get_sb,
2035 .kill_sb = nfs4_kill_super,
2036 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2039 #define nfs4_init_once(nfsi) \
2041 INIT_LIST_HEAD(&(nfsi)->open_states); \
2042 nfsi->delegation = NULL; \
2043 nfsi->delegation_state = 0; \
2044 init_rwsem(&nfsi->rwsem); \
2046 #define register_nfs4fs() register_filesystem(&nfs4_fs_type)
2047 #define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
2049 #define nfs4_init_once(nfsi) \
2051 #define register_nfs4fs() (0)
2052 #define unregister_nfs4fs()
2055 extern int nfs_init_nfspagecache(void);
2056 extern void nfs_destroy_nfspagecache(void);
2057 extern int nfs_init_readpagecache(void);
2058 extern void nfs_destroy_readpagecache(void);
2059 extern int nfs_init_writepagecache(void);
2060 extern void nfs_destroy_writepagecache(void);
2061 #ifdef CONFIG_NFS_DIRECTIO
2062 extern int nfs_init_directcache(void);
2063 extern void nfs_destroy_directcache(void);
2066 static kmem_cache_t * nfs_inode_cachep;
2068 static struct inode *nfs_alloc_inode(struct super_block *sb)
2070 struct nfs_inode *nfsi;
2071 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2075 nfsi->cache_validity = 0UL;
2076 nfsi->cache_change_attribute = jiffies;
2077 #ifdef CONFIG_NFS_V3_ACL
2078 nfsi->acl_access = ERR_PTR(-EAGAIN);
2079 nfsi->acl_default = ERR_PTR(-EAGAIN);
2081 #ifdef CONFIG_NFS_V4
2082 nfsi->nfs4_acl = NULL;
2083 #endif /* CONFIG_NFS_V4 */
2084 return &nfsi->vfs_inode;
2087 static void nfs_destroy_inode(struct inode *inode)
2089 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2092 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2094 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2096 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2097 SLAB_CTOR_CONSTRUCTOR) {
2098 inode_init_once(&nfsi->vfs_inode);
2099 spin_lock_init(&nfsi->req_lock);
2100 INIT_LIST_HEAD(&nfsi->dirty);
2101 INIT_LIST_HEAD(&nfsi->commit);
2102 INIT_LIST_HEAD(&nfsi->open_files);
2103 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2104 atomic_set(&nfsi->data_updates, 0);
2108 nfs4_init_once(nfsi);
2112 static int nfs_init_inodecache(void)
2114 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2115 sizeof(struct nfs_inode),
2116 0, SLAB_RECLAIM_ACCOUNT,
2118 if (nfs_inode_cachep == NULL)
2124 static void nfs_destroy_inodecache(void)
2126 if (kmem_cache_destroy(nfs_inode_cachep))
2127 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2133 static int __init init_nfs_fs(void)
2137 err = nfs_init_nfspagecache();
2141 err = nfs_init_inodecache();
2145 err = nfs_init_readpagecache();
2149 err = nfs_init_writepagecache();
2153 #ifdef CONFIG_NFS_DIRECTIO
2154 err = nfs_init_directcache();
2159 #ifdef CONFIG_PROC_FS
2160 rpc_proc_register(&nfs_rpcstat);
2162 err = register_filesystem(&nfs_fs_type);
2165 if ((err = register_nfs4fs()) != 0)
2169 #ifdef CONFIG_PROC_FS
2170 rpc_proc_unregister("nfs");
2172 #ifdef CONFIG_NFS_DIRECTIO
2173 nfs_destroy_directcache();
2176 nfs_destroy_writepagecache();
2178 nfs_destroy_readpagecache();
2180 nfs_destroy_inodecache();
2182 nfs_destroy_nfspagecache();
2187 static void __exit exit_nfs_fs(void)
2189 #ifdef CONFIG_NFS_DIRECTIO
2190 nfs_destroy_directcache();
2192 nfs_destroy_writepagecache();
2193 nfs_destroy_readpagecache();
2194 nfs_destroy_inodecache();
2195 nfs_destroy_nfspagecache();
2196 #ifdef CONFIG_PROC_FS
2197 rpc_proc_unregister("nfs");
2199 unregister_filesystem(&nfs_fs_type);
2200 unregister_nfs4fs();
2203 /* Not quite true; I just maintain it */
2204 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2205 MODULE_LICENSE("GPL");
2207 module_init(init_nfs_fs)
2208 module_exit(exit_nfs_fs)