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/sunrpc/metrics.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/nfs4_mount.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/smp_lock.h>
35 #include <linux/seq_file.h>
36 #include <linux/mount.h>
37 #include <linux/nfs_idmap.h>
38 #include <linux/vfs.h>
40 #include <asm/system.h>
41 #include <asm/uaccess.h>
45 #include "delegation.h"
48 #define NFSDBG_FACILITY NFSDBG_VFS
49 #define NFS_PARANOIA 1
51 /* Maximum number of readahead requests
52 * FIXME: this should really be a sysctl so that users may tune it to suit
53 * their needs. People that do NFS over a slow network, might for
54 * instance want to reduce it to something closer to 1 for improved
55 * interactive response.
57 #define NFS_MAX_READAHEAD (RPC_DEF_SLOT_TABLE - 1)
59 static void nfs_invalidate_inode(struct inode *);
60 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
62 static struct inode *nfs_alloc_inode(struct super_block *sb);
63 static void nfs_destroy_inode(struct inode *);
64 static int nfs_write_inode(struct inode *,int);
65 static void nfs_delete_inode(struct inode *);
66 static void nfs_clear_inode(struct inode *);
67 static void nfs_umount_begin(struct super_block *);
68 static int nfs_statfs(struct super_block *, struct kstatfs *);
69 static int nfs_show_options(struct seq_file *, struct vfsmount *);
70 static int nfs_show_stats(struct seq_file *, struct vfsmount *);
71 static void nfs_zap_acl_cache(struct inode *);
73 static struct rpc_program nfs_program;
75 static struct super_operations nfs_sops = {
76 .alloc_inode = nfs_alloc_inode,
77 .destroy_inode = nfs_destroy_inode,
78 .write_inode = nfs_write_inode,
79 .delete_inode = nfs_delete_inode,
81 .clear_inode = nfs_clear_inode,
82 .umount_begin = nfs_umount_begin,
83 .show_options = nfs_show_options,
84 .show_stats = nfs_show_stats,
90 static struct rpc_stat nfs_rpcstat = {
91 .program = &nfs_program
93 static struct rpc_version * nfs_version[] = {
97 #if defined(CONFIG_NFS_V3)
99 #elif defined(CONFIG_NFS_V4)
102 #if defined(CONFIG_NFS_V4)
107 static struct rpc_program nfs_program = {
109 .number = NFS_PROGRAM,
110 .nrvers = sizeof(nfs_version) / sizeof(nfs_version[0]),
111 .version = nfs_version,
112 .stats = &nfs_rpcstat,
113 .pipe_dir_name = "/nfs",
116 #ifdef CONFIG_NFS_V3_ACL
117 static struct rpc_stat nfsacl_rpcstat = { &nfsacl_program };
118 static struct rpc_version * nfsacl_version[] = {
119 [3] = &nfsacl_version3,
122 struct rpc_program nfsacl_program = {
124 .number = NFS_ACL_PROGRAM,
125 .nrvers = sizeof(nfsacl_version) / sizeof(nfsacl_version[0]),
126 .version = nfsacl_version,
127 .stats = &nfsacl_rpcstat,
129 #endif /* CONFIG_NFS_V3_ACL */
131 static inline unsigned long
132 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
134 return nfs_fileid_to_ino_t(fattr->fileid);
138 nfs_write_inode(struct inode *inode, int sync)
140 int flags = sync ? FLUSH_WAIT : 0;
143 ret = nfs_commit_inode(inode, flags);
150 nfs_delete_inode(struct inode * inode)
152 dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
154 truncate_inode_pages(&inode->i_data, 0);
158 * The following should never happen...
160 if (nfs_have_writebacks(inode)) {
161 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
168 nfs_clear_inode(struct inode *inode)
170 struct nfs_inode *nfsi = NFS_I(inode);
171 struct rpc_cred *cred;
174 BUG_ON (!list_empty(&nfsi->open_files));
175 nfs_zap_acl_cache(inode);
176 cred = nfsi->cache_access.cred;
179 BUG_ON(atomic_read(&nfsi->data_updates) != 0);
183 nfs_umount_begin(struct super_block *sb)
185 struct rpc_clnt *rpc = NFS_SB(sb)->client;
187 /* -EIO all pending I/O */
189 rpc_killall_tasks(rpc);
190 rpc = NFS_SB(sb)->client_acl;
192 rpc_killall_tasks(rpc);
196 static inline unsigned long
197 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
199 /* make sure blocksize is a power of two */
200 if ((bsize & (bsize - 1)) || nrbitsp) {
201 unsigned char nrbits;
203 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
214 * Calculate the number of 512byte blocks used.
216 static inline unsigned long
217 nfs_calc_block_size(u64 tsize)
219 loff_t used = (tsize + 511) >> 9;
220 return (used > ULONG_MAX) ? ULONG_MAX : used;
224 * Compute and set NFS server blocksize
226 static inline unsigned long
227 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
229 if (bsize < NFS_MIN_FILE_IO_SIZE)
230 bsize = NFS_DEF_FILE_IO_SIZE;
231 else if (bsize >= NFS_MAX_FILE_IO_SIZE)
232 bsize = NFS_MAX_FILE_IO_SIZE;
234 return nfs_block_bits(bsize, nrbitsp);
238 * Obtain the root inode of the file system.
240 static struct inode *
241 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
243 struct nfs_server *server = NFS_SB(sb);
247 error = server->rpc_ops->getroot(server, rootfh, fsinfo);
249 dprintk("nfs_get_root: getattr error = %d\n", -error);
250 return ERR_PTR(error);
253 rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
255 return ERR_PTR(-ENOMEM);
260 * Do NFS version-independent mount processing, and sanity checking
263 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
265 struct nfs_server *server;
266 struct inode *root_inode;
267 struct nfs_fattr fattr;
268 struct nfs_fsinfo fsinfo = {
271 struct nfs_pathconf pathinfo = {
274 int no_root_error = 0;
275 unsigned long max_rpc_payload;
277 /* We probably want something more informative here */
278 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
282 sb->s_magic = NFS_SUPER_MAGIC;
284 root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
285 /* Did getting the root inode fail? */
286 if (IS_ERR(root_inode)) {
287 no_root_error = PTR_ERR(root_inode);
290 sb->s_root = d_alloc_root(root_inode);
292 no_root_error = -ENOMEM;
295 sb->s_root->d_op = server->rpc_ops->dentry_ops;
297 server->io_stats = nfs_alloc_iostats();
298 if (!server->io_stats) {
299 no_root_error = -ENOMEM;
303 /* mount time stamp, in seconds */
304 server->mount_time = jiffies;
306 /* Get some general file system info */
307 if (server->namelen == 0 &&
308 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
309 server->namelen = pathinfo.max_namelen;
310 /* Work out a lot of parameters */
311 if (server->rsize == 0)
312 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
313 if (server->wsize == 0)
314 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
316 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
317 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
318 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
319 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
321 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
322 if (server->rsize > max_rpc_payload)
323 server->rsize = max_rpc_payload;
324 if (server->rsize > NFS_MAX_FILE_IO_SIZE)
325 server->rsize = NFS_MAX_FILE_IO_SIZE;
326 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
328 if (server->wsize > max_rpc_payload)
329 server->wsize = max_rpc_payload;
330 if (server->wsize > NFS_MAX_FILE_IO_SIZE)
331 server->wsize = NFS_MAX_FILE_IO_SIZE;
332 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
334 if (sb->s_blocksize == 0)
335 sb->s_blocksize = nfs_block_bits(server->wsize,
336 &sb->s_blocksize_bits);
337 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
339 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
340 if (server->dtsize > PAGE_CACHE_SIZE)
341 server->dtsize = PAGE_CACHE_SIZE;
342 if (server->dtsize > server->rsize)
343 server->dtsize = server->rsize;
345 if (server->flags & NFS_MOUNT_NOAC) {
346 server->acregmin = server->acregmax = 0;
347 server->acdirmin = server->acdirmax = 0;
348 sb->s_flags |= MS_SYNCHRONOUS;
350 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
352 sb->s_maxbytes = fsinfo.maxfilesize;
353 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
354 sb->s_maxbytes = MAX_LFS_FILESIZE;
356 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
357 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
359 /* We're airborne Set socket buffersize */
360 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
362 /* Yargs. It didn't work out. */
364 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
365 if (!IS_ERR(root_inode))
367 return no_root_error;
370 static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
372 to->to_initval = timeo * HZ / 10;
373 to->to_retries = retrans;
380 to->to_initval = 60 * HZ;
381 if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
382 to->to_initval = NFS_MAX_TCP_TIMEOUT;
383 to->to_increment = to->to_initval;
384 to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
385 to->to_exponential = 0;
390 to->to_initval = 11 * HZ / 10;
391 if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
392 to->to_initval = NFS_MAX_UDP_TIMEOUT;
393 to->to_maxval = NFS_MAX_UDP_TIMEOUT;
394 to->to_exponential = 1;
400 * Create an RPC client handle.
402 static struct rpc_clnt *
403 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
405 struct rpc_timeout timeparms;
406 struct rpc_xprt *xprt = NULL;
407 struct rpc_clnt *clnt = NULL;
408 int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
410 nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
412 server->retrans_timeo = timeparms.to_initval;
413 server->retrans_count = timeparms.to_retries;
415 /* create transport and client */
416 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
418 dprintk("%s: cannot create RPC transport. Error = %ld\n",
419 __FUNCTION__, PTR_ERR(xprt));
420 return (struct rpc_clnt *)xprt;
422 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
423 server->rpc_ops->version, data->pseudoflavor);
425 dprintk("%s: cannot create RPC client. Error = %ld\n",
426 __FUNCTION__, PTR_ERR(xprt));
431 clnt->cl_softrtry = 1;
440 * The way this works is that the mount process passes a structure
441 * in the data argument which contains the server's IP address
442 * and the root file handle obtained from the server's mount
443 * daemon. We stash these away in the private superblock fields.
446 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
448 struct nfs_server *server;
449 rpc_authflavor_t authflavor;
452 sb->s_blocksize_bits = 0;
455 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
457 server->rsize = nfs_block_size(data->rsize, NULL);
459 server->wsize = nfs_block_size(data->wsize, NULL);
460 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
462 server->acregmin = data->acregmin*HZ;
463 server->acregmax = data->acregmax*HZ;
464 server->acdirmin = data->acdirmin*HZ;
465 server->acdirmax = data->acdirmax*HZ;
467 /* Start lockd here, before we might error out */
468 if (!(server->flags & NFS_MOUNT_NONLM))
471 server->namelen = data->namlen;
472 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
473 if (!server->hostname)
475 strcpy(server->hostname, data->hostname);
477 /* Check NFS protocol revision and initialize RPC op vector
478 * and file handle pool. */
480 if (server->flags & NFS_MOUNT_VER3) {
481 server->rpc_ops = &nfs_v3_clientops;
482 server->caps |= NFS_CAP_READDIRPLUS;
484 server->rpc_ops = &nfs_v2_clientops;
487 server->rpc_ops = &nfs_v2_clientops;
490 /* Fill in pseudoflavor for mount version < 5 */
491 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
492 data->pseudoflavor = RPC_AUTH_UNIX;
493 authflavor = data->pseudoflavor; /* save for sb_init() */
494 /* XXX maybe we want to add a server->pseudoflavor field */
496 /* Create RPC client handles */
497 server->client = nfs_create_client(server, data);
498 if (IS_ERR(server->client))
499 return PTR_ERR(server->client);
500 /* RFC 2623, sec 2.3.2 */
501 if (authflavor != RPC_AUTH_UNIX) {
502 struct rpc_auth *auth;
504 server->client_sys = rpc_clone_client(server->client);
505 if (IS_ERR(server->client_sys))
506 return PTR_ERR(server->client_sys);
507 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
509 return PTR_ERR(auth);
511 atomic_inc(&server->client->cl_count);
512 server->client_sys = server->client;
514 if (server->flags & NFS_MOUNT_VER3) {
515 #ifdef CONFIG_NFS_V3_ACL
516 if (!(server->flags & NFS_MOUNT_NOACL)) {
517 server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
518 /* No errors! Assume that Sun nfsacls are supported */
519 if (!IS_ERR(server->client_acl))
520 server->caps |= NFS_CAP_ACLS;
523 server->flags &= ~NFS_MOUNT_NOACL;
524 #endif /* CONFIG_NFS_V3_ACL */
526 * The VFS shouldn't apply the umask to mode bits. We will
527 * do so ourselves when necessary.
529 sb->s_flags |= MS_POSIXACL;
530 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
531 server->namelen = NFS3_MAXNAMLEN;
534 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
535 server->namelen = NFS2_MAXNAMLEN;
538 sb->s_op = &nfs_sops;
539 return nfs_sb_init(sb, authflavor);
543 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
545 struct nfs_server *server = NFS_SB(sb);
546 unsigned char blockbits;
547 unsigned long blockres;
548 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
549 struct nfs_fattr fattr;
550 struct nfs_fsstat res = {
557 error = server->rpc_ops->statfs(server, rootfh, &res);
558 buf->f_type = NFS_SUPER_MAGIC;
563 * Current versions of glibc do not correctly handle the
564 * case where f_frsize != f_bsize. Eventually we want to
565 * report the value of wtmult in this field.
567 buf->f_frsize = sb->s_blocksize;
570 * On most *nix systems, f_blocks, f_bfree, and f_bavail
571 * are reported in units of f_frsize. Linux hasn't had
572 * an f_frsize field in its statfs struct until recently,
573 * thus historically Linux's sys_statfs reports these
574 * fields in units of f_bsize.
576 buf->f_bsize = sb->s_blocksize;
577 blockbits = sb->s_blocksize_bits;
578 blockres = (1 << blockbits) - 1;
579 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
580 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
581 buf->f_bavail = (res.abytes + blockres) >> blockbits;
583 buf->f_files = res.tfiles;
584 buf->f_ffree = res.afiles;
586 buf->f_namelen = server->namelen;
592 dprintk("%s: statfs error = %d\n", __FUNCTION__, -error);
593 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
598 static void nfs_show_mount_options(struct seq_file *m, struct nfs_server *nfss, int showdefaults)
600 static struct proc_nfs_info {
605 { NFS_MOUNT_SOFT, ",soft", ",hard" },
606 { NFS_MOUNT_INTR, ",intr", "" },
607 { NFS_MOUNT_NOCTO, ",nocto", "" },
608 { NFS_MOUNT_NOAC, ",noac", "" },
609 { NFS_MOUNT_NONLM, ",nolock", "" },
610 { NFS_MOUNT_NOACL, ",noacl", "" },
613 struct proc_nfs_info *nfs_infop;
617 seq_printf(m, ",vers=%d", nfss->rpc_ops->version);
618 seq_printf(m, ",rsize=%d", nfss->rsize);
619 seq_printf(m, ",wsize=%d", nfss->wsize);
620 if (nfss->acregmin != 3*HZ || showdefaults)
621 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
622 if (nfss->acregmax != 60*HZ || showdefaults)
623 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
624 if (nfss->acdirmin != 30*HZ || showdefaults)
625 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
626 if (nfss->acdirmax != 60*HZ || showdefaults)
627 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
628 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
629 if (nfss->flags & nfs_infop->flag)
630 seq_puts(m, nfs_infop->str);
632 seq_puts(m, nfs_infop->nostr);
634 switch (nfss->client->cl_xprt->prot) {
642 snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
645 seq_printf(m, ",proto=%s", proto);
646 seq_printf(m, ",timeo=%lu", 10U * nfss->retrans_timeo / HZ);
647 seq_printf(m, ",retrans=%u", nfss->retrans_count);
650 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
652 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
654 nfs_show_mount_options(m, nfss, 0);
656 seq_puts(m, ",addr=");
657 seq_escape(m, nfss->hostname, " \t\n\\");
662 static int nfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
665 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
666 struct rpc_auth *auth = nfss->client->cl_auth;
667 struct nfs_iostats totals = { };
669 seq_printf(m, "statvers=%s", NFS_IOSTAT_VERS);
672 * Display all mount option settings
674 seq_printf(m, "\n\topts:\t");
675 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? "ro" : "rw");
676 seq_puts(m, mnt->mnt_sb->s_flags & MS_SYNCHRONOUS ? ",sync" : "");
677 seq_puts(m, mnt->mnt_sb->s_flags & MS_NOATIME ? ",noatime" : "");
678 seq_puts(m, mnt->mnt_sb->s_flags & MS_NODIRATIME ? ",nodiratime" : "");
679 nfs_show_mount_options(m, nfss, 1);
681 seq_printf(m, "\n\tage:\t%lu", (jiffies - nfss->mount_time) / HZ);
683 seq_printf(m, "\n\tcaps:\t");
684 seq_printf(m, "caps=0x%x", nfss->caps);
685 seq_printf(m, ",wtmult=%d", nfss->wtmult);
686 seq_printf(m, ",dtsize=%d", nfss->dtsize);
687 seq_printf(m, ",bsize=%d", nfss->bsize);
688 seq_printf(m, ",namelen=%d", nfss->namelen);
691 if (nfss->rpc_ops->version == 4) {
692 seq_printf(m, "\n\tnfsv4:\t");
693 seq_printf(m, "bm0=0x%x", nfss->attr_bitmask[0]);
694 seq_printf(m, ",bm1=0x%x", nfss->attr_bitmask[1]);
695 seq_printf(m, ",acl=0x%x", nfss->acl_bitmask);
700 * Display security flavor in effect for this mount
702 seq_printf(m, "\n\tsec:\tflavor=%d", auth->au_ops->au_flavor);
704 seq_printf(m, ",pseudoflavor=%d", auth->au_flavor);
707 * Display superblock I/O counters
709 for (cpu = 0; cpu < NR_CPUS; cpu++) {
710 struct nfs_iostats *stats;
712 if (!cpu_possible(cpu))
716 stats = per_cpu_ptr(nfss->io_stats, cpu);
718 for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
719 totals.events[i] += stats->events[i];
720 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
721 totals.bytes[i] += stats->bytes[i];
726 seq_printf(m, "\n\tevents:\t");
727 for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
728 seq_printf(m, "%lu ", totals.events[i]);
729 seq_printf(m, "\n\tbytes:\t");
730 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
731 seq_printf(m, "%Lu ", totals.bytes[i]);
734 rpc_print_iostats(m, nfss->client);
740 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
742 int nfs_sync_mapping(struct address_space *mapping)
746 if (mapping->nrpages == 0)
748 unmap_mapping_range(mapping, 0, 0, 0);
749 ret = filemap_write_and_wait(mapping);
752 ret = nfs_wb_all(mapping->host);
758 * Invalidate the local caches
760 static void nfs_zap_caches_locked(struct inode *inode)
762 struct nfs_inode *nfsi = NFS_I(inode);
763 int mode = inode->i_mode;
765 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
767 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
768 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
770 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
771 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
772 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
774 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
777 void nfs_zap_caches(struct inode *inode)
779 spin_lock(&inode->i_lock);
780 nfs_zap_caches_locked(inode);
781 spin_unlock(&inode->i_lock);
784 static void nfs_zap_acl_cache(struct inode *inode)
786 void (*clear_acl_cache)(struct inode *);
788 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
789 if (clear_acl_cache != NULL)
790 clear_acl_cache(inode);
791 spin_lock(&inode->i_lock);
792 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
793 spin_unlock(&inode->i_lock);
797 * Invalidate, but do not unhash, the inode.
798 * NB: must be called with inode->i_lock held!
800 static void nfs_invalidate_inode(struct inode *inode)
802 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
803 nfs_zap_caches_locked(inode);
806 struct nfs_find_desc {
808 struct nfs_fattr *fattr;
812 * In NFSv3 we can have 64bit inode numbers. In order to support
813 * this, and re-exported directories (also seen in NFSv2)
814 * we are forced to allow 2 different inodes to have the same
818 nfs_find_actor(struct inode *inode, void *opaque)
820 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
821 struct nfs_fh *fh = desc->fh;
822 struct nfs_fattr *fattr = desc->fattr;
824 if (NFS_FILEID(inode) != fattr->fileid)
826 if (nfs_compare_fh(NFS_FH(inode), fh))
828 if (is_bad_inode(inode) || NFS_STALE(inode))
834 nfs_init_locked(struct inode *inode, void *opaque)
836 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
837 struct nfs_fattr *fattr = desc->fattr;
839 NFS_FILEID(inode) = fattr->fileid;
840 nfs_copy_fh(NFS_FH(inode), desc->fh);
844 /* Don't use READDIRPLUS on directories that we believe are too large */
845 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
848 * This is our front-end to iget that looks up inodes by file handle
849 * instead of inode number.
852 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
854 struct nfs_find_desc desc = {
858 struct inode *inode = NULL;
861 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
865 printk("NFS: Buggy server - nlink == 0!\n");
869 hash = nfs_fattr_to_ino_t(fattr);
871 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
874 if (inode->i_state & I_NEW) {
875 struct nfs_inode *nfsi = NFS_I(inode);
877 /* We set i_ino for the few things that still rely on it,
881 /* We can't support update_atime(), since the server will reset it */
882 inode->i_flags |= S_NOATIME|S_NOCMTIME;
883 inode->i_mode = fattr->mode;
884 /* Why so? Because we want revalidate for devices/FIFOs, and
885 * that's precisely what we have in nfs_file_inode_operations.
887 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
888 if (S_ISREG(inode->i_mode)) {
889 inode->i_fop = &nfs_file_operations;
890 inode->i_data.a_ops = &nfs_file_aops;
891 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
892 } else if (S_ISDIR(inode->i_mode)) {
893 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
894 inode->i_fop = &nfs_dir_operations;
895 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
896 && fattr->size <= NFS_LIMIT_READDIRPLUS)
897 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
898 } else if (S_ISLNK(inode->i_mode))
899 inode->i_op = &nfs_symlink_inode_operations;
901 init_special_inode(inode, inode->i_mode, fattr->rdev);
903 nfsi->read_cache_jiffies = fattr->time_start;
904 nfsi->last_updated = jiffies;
905 inode->i_atime = fattr->atime;
906 inode->i_mtime = fattr->mtime;
907 inode->i_ctime = fattr->ctime;
908 if (fattr->valid & NFS_ATTR_FATTR_V4)
909 nfsi->change_attr = fattr->change_attr;
910 inode->i_size = nfs_size_to_loff_t(fattr->size);
911 inode->i_nlink = fattr->nlink;
912 inode->i_uid = fattr->uid;
913 inode->i_gid = fattr->gid;
914 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
916 * report the blocks in 512byte units
918 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
919 inode->i_blksize = inode->i_sb->s_blocksize;
921 inode->i_blocks = fattr->du.nfs2.blocks;
922 inode->i_blksize = fattr->du.nfs2.blocksize;
924 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
925 nfsi->attrtimeo_timestamp = jiffies;
926 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
927 nfsi->cache_access.cred = NULL;
929 unlock_new_inode(inode);
931 nfs_refresh_inode(inode, fattr);
932 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
934 (long long)NFS_FILEID(inode),
935 atomic_read(&inode->i_count));
941 printk("nfs_fhget: iget failed\n");
945 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
948 nfs_setattr(struct dentry *dentry, struct iattr *attr)
950 struct inode *inode = dentry->d_inode;
951 struct nfs_fattr fattr;
954 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
956 if (attr->ia_valid & ATTR_SIZE) {
957 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
958 attr->ia_valid &= ~ATTR_SIZE;
961 /* Optimization: if the end result is no change, don't RPC */
962 attr->ia_valid &= NFS_VALID_ATTRS;
963 if (attr->ia_valid == 0)
967 nfs_begin_data_update(inode);
968 /* Write all dirty data */
969 filemap_write_and_wait(inode->i_mapping);
972 * Return any delegations if we're going to change ACLs
974 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
975 nfs_inode_return_delegation(inode);
976 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
978 nfs_refresh_inode(inode, &fattr);
979 nfs_end_data_update(inode);
985 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
986 * @inode: pointer to struct inode
987 * @attr: pointer to struct iattr
989 * Note: we do this in the *proc.c in order to ensure that
990 * it works for things like exclusive creates too.
992 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
994 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
995 if ((attr->ia_valid & ATTR_MODE) != 0) {
996 int mode = attr->ia_mode & S_IALLUGO;
997 mode |= inode->i_mode & ~S_IALLUGO;
998 inode->i_mode = mode;
1000 if ((attr->ia_valid & ATTR_UID) != 0)
1001 inode->i_uid = attr->ia_uid;
1002 if ((attr->ia_valid & ATTR_GID) != 0)
1003 inode->i_gid = attr->ia_gid;
1004 spin_lock(&inode->i_lock);
1005 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1006 spin_unlock(&inode->i_lock);
1008 if ((attr->ia_valid & ATTR_SIZE) != 0) {
1009 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
1010 inode->i_size = attr->ia_size;
1011 vmtruncate(inode, attr->ia_size);
1015 static int nfs_wait_schedule(void *word)
1017 if (signal_pending(current))
1018 return -ERESTARTSYS;
1024 * Wait for the inode to get unlocked.
1026 static int nfs_wait_on_inode(struct inode *inode)
1028 struct rpc_clnt *clnt = NFS_CLIENT(inode);
1029 struct nfs_inode *nfsi = NFS_I(inode);
1033 rpc_clnt_sigmask(clnt, &oldmask);
1034 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
1035 nfs_wait_schedule, TASK_INTERRUPTIBLE);
1036 rpc_clnt_sigunmask(clnt, &oldmask);
1041 static void nfs_wake_up_inode(struct inode *inode)
1043 struct nfs_inode *nfsi = NFS_I(inode);
1045 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
1046 smp_mb__after_clear_bit();
1047 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
1050 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1052 struct inode *inode = dentry->d_inode;
1053 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
1056 /* Flush out writes to the server in order to update c/mtime */
1057 nfs_sync_inode(inode, 0, 0, FLUSH_WAIT|FLUSH_NOCOMMIT);
1060 * We may force a getattr if the user cares about atime.
1062 * Note that we only have to check the vfsmount flags here:
1063 * - NFS always sets S_NOATIME by so checking it would give a
1065 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
1066 * no point in checking those.
1068 if ((mnt->mnt_flags & MNT_NOATIME) ||
1069 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
1073 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1075 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1077 generic_fillattr(inode, stat);
1081 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
1083 struct nfs_open_context *ctx;
1085 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
1087 atomic_set(&ctx->count, 1);
1088 ctx->dentry = dget(dentry);
1089 ctx->vfsmnt = mntget(mnt);
1090 ctx->cred = get_rpccred(cred);
1092 ctx->lockowner = current->files;
1094 ctx->dir_cookie = 0;
1099 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1102 atomic_inc(&ctx->count);
1106 void put_nfs_open_context(struct nfs_open_context *ctx)
1108 if (atomic_dec_and_test(&ctx->count)) {
1109 if (!list_empty(&ctx->list)) {
1110 struct inode *inode = ctx->dentry->d_inode;
1111 spin_lock(&inode->i_lock);
1112 list_del(&ctx->list);
1113 spin_unlock(&inode->i_lock);
1115 if (ctx->state != NULL)
1116 nfs4_close_state(ctx->state, ctx->mode);
1117 if (ctx->cred != NULL)
1118 put_rpccred(ctx->cred);
1120 mntput(ctx->vfsmnt);
1126 * Ensure that mmap has a recent RPC credential for use when writing out
1129 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1131 struct inode *inode = filp->f_dentry->d_inode;
1132 struct nfs_inode *nfsi = NFS_I(inode);
1134 filp->private_data = get_nfs_open_context(ctx);
1135 spin_lock(&inode->i_lock);
1136 list_add(&ctx->list, &nfsi->open_files);
1137 spin_unlock(&inode->i_lock);
1141 * Given an inode, search for an open context with the desired characteristics
1143 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
1145 struct nfs_inode *nfsi = NFS_I(inode);
1146 struct nfs_open_context *pos, *ctx = NULL;
1148 spin_lock(&inode->i_lock);
1149 list_for_each_entry(pos, &nfsi->open_files, list) {
1150 if (cred != NULL && pos->cred != cred)
1152 if ((pos->mode & mode) == mode) {
1153 ctx = get_nfs_open_context(pos);
1157 spin_unlock(&inode->i_lock);
1161 static void nfs_file_clear_open_context(struct file *filp)
1163 struct inode *inode = filp->f_dentry->d_inode;
1164 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
1167 filp->private_data = NULL;
1168 spin_lock(&inode->i_lock);
1169 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1170 spin_unlock(&inode->i_lock);
1171 put_nfs_open_context(ctx);
1176 * These allocate and release file read/write context information.
1178 int nfs_open(struct inode *inode, struct file *filp)
1180 struct nfs_open_context *ctx;
1181 struct rpc_cred *cred;
1183 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1185 return PTR_ERR(cred);
1186 ctx = alloc_nfs_open_context(filp->f_vfsmnt, filp->f_dentry, cred);
1190 ctx->mode = filp->f_mode;
1191 nfs_file_set_open_context(filp, ctx);
1192 put_nfs_open_context(ctx);
1196 int nfs_release(struct inode *inode, struct file *filp)
1198 nfs_file_clear_open_context(filp);
1203 * This function is called whenever some part of NFS notices that
1204 * the cached attributes have to be refreshed.
1207 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1209 int status = -ESTALE;
1210 struct nfs_fattr fattr;
1211 struct nfs_inode *nfsi = NFS_I(inode);
1213 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1214 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1217 if (!inode || is_bad_inode(inode))
1219 if (NFS_STALE(inode))
1222 status = nfs_wait_on_inode(inode);
1225 if (NFS_STALE(inode)) {
1227 /* Do we trust the cached ESTALE? */
1228 if (NFS_ATTRTIMEO(inode) != 0) {
1229 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1236 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1238 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1240 (long long)NFS_FILEID(inode), status);
1241 if (status == -ESTALE) {
1242 nfs_zap_caches(inode);
1243 if (!S_ISDIR(inode->i_mode))
1244 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1249 spin_lock(&inode->i_lock);
1250 status = nfs_update_inode(inode, &fattr);
1252 spin_unlock(&inode->i_lock);
1253 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1255 (long long)NFS_FILEID(inode), status);
1258 spin_unlock(&inode->i_lock);
1260 nfs_revalidate_mapping(inode, inode->i_mapping);
1262 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1263 nfs_zap_acl_cache(inode);
1265 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1267 (long long)NFS_FILEID(inode));
1270 nfs_wake_up_inode(inode);
1277 int nfs_attribute_timeout(struct inode *inode)
1279 struct nfs_inode *nfsi = NFS_I(inode);
1281 if (nfs_have_delegation(inode, FMODE_READ))
1283 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1287 * nfs_revalidate_inode - Revalidate the inode attributes
1288 * @server - pointer to nfs_server struct
1289 * @inode - pointer to inode struct
1291 * Updates inode attribute information by retrieving the data from the server.
1293 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1295 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1296 if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1297 && !nfs_attribute_timeout(inode))
1298 return NFS_STALE(inode) ? -ESTALE : 0;
1299 return __nfs_revalidate_inode(server, inode);
1303 * nfs_revalidate_mapping - Revalidate the pagecache
1304 * @inode - pointer to host inode
1305 * @mapping - pointer to mapping
1307 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1309 struct nfs_inode *nfsi = NFS_I(inode);
1311 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1312 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1313 if (S_ISREG(inode->i_mode))
1314 nfs_sync_mapping(mapping);
1315 invalidate_inode_pages2(mapping);
1317 spin_lock(&inode->i_lock);
1318 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1319 if (S_ISDIR(inode->i_mode)) {
1320 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1321 /* This ensures we revalidate child dentries */
1322 nfsi->cache_change_attribute = jiffies;
1324 spin_unlock(&inode->i_lock);
1326 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1328 (long long)NFS_FILEID(inode));
1333 * nfs_begin_data_update
1334 * @inode - pointer to inode
1335 * Declare that a set of operations will update file data on the server
1337 void nfs_begin_data_update(struct inode *inode)
1339 atomic_inc(&NFS_I(inode)->data_updates);
1343 * nfs_end_data_update
1344 * @inode - pointer to inode
1345 * Declare end of the operations that will update file data
1346 * This will mark the inode as immediately needing revalidation
1347 * of its attribute cache.
1349 void nfs_end_data_update(struct inode *inode)
1351 struct nfs_inode *nfsi = NFS_I(inode);
1353 if (!nfs_have_delegation(inode, FMODE_READ)) {
1354 /* Directories and symlinks: invalidate page cache */
1355 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
1356 spin_lock(&inode->i_lock);
1357 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1358 spin_unlock(&inode->i_lock);
1361 nfsi->cache_change_attribute = jiffies;
1362 atomic_dec(&nfsi->data_updates);
1365 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1367 struct nfs_inode *nfsi = NFS_I(inode);
1369 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1370 && nfsi->change_attr == fattr->pre_change_attr) {
1371 nfsi->change_attr = fattr->change_attr;
1372 nfsi->cache_change_attribute = jiffies;
1375 /* If we have atomic WCC data, we may update some attributes */
1376 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1377 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1378 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1379 nfsi->cache_change_attribute = jiffies;
1381 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1382 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1383 nfsi->cache_change_attribute = jiffies;
1385 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
1386 inode->i_size = fattr->size;
1387 nfsi->cache_change_attribute = jiffies;
1393 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1394 * @inode - pointer to inode
1395 * @fattr - updated attributes
1397 * Verifies the attribute cache. If we have just changed the attributes,
1398 * so that fattr carries weak cache consistency data, then it may
1399 * also update the ctime/mtime/change_attribute.
1401 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1403 struct nfs_inode *nfsi = NFS_I(inode);
1404 loff_t cur_size, new_isize;
1408 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1411 /* Has the inode gone and changed behind our back? */
1412 if (nfsi->fileid != fattr->fileid
1413 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1417 /* Are we in the process of updating data on the server? */
1418 data_unstable = nfs_caches_unstable(inode);
1420 /* Do atomic weak cache consistency updates */
1421 nfs_wcc_update_inode(inode, fattr);
1423 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0) {
1424 if (nfsi->change_attr == fattr->change_attr)
1426 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1428 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1431 /* Verify a few of the more important attributes */
1432 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1433 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1435 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1438 cur_size = i_size_read(inode);
1439 new_isize = nfs_size_to_loff_t(fattr->size);
1440 if (cur_size != new_isize && nfsi->npages == 0)
1441 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1443 /* Have any file permissions changed? */
1444 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1445 || inode->i_uid != fattr->uid
1446 || inode->i_gid != fattr->gid)
1447 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1449 /* Has the link count changed? */
1450 if (inode->i_nlink != fattr->nlink)
1451 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1454 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1455 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1457 nfsi->read_cache_jiffies = fattr->time_start;
1462 * nfs_refresh_inode - try to update the inode attribute cache
1463 * @inode - pointer to inode
1464 * @fattr - updated attributes
1466 * Check that an RPC call that returned attributes has not overlapped with
1467 * other recent updates of the inode metadata, then decide whether it is
1468 * safe to do a full update of the inode attributes, or whether just to
1469 * call nfs_check_inode_attributes.
1471 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1473 struct nfs_inode *nfsi = NFS_I(inode);
1476 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1478 spin_lock(&inode->i_lock);
1479 nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1480 if (time_after(fattr->time_start, nfsi->last_updated))
1481 status = nfs_update_inode(inode, fattr);
1483 status = nfs_check_inode_attributes(inode, fattr);
1485 spin_unlock(&inode->i_lock);
1490 * nfs_post_op_update_inode - try to update the inode attribute cache
1491 * @inode - pointer to inode
1492 * @fattr - updated attributes
1494 * After an operation that has changed the inode metadata, mark the
1495 * attribute cache as being invalid, then try to update it.
1497 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1499 struct nfs_inode *nfsi = NFS_I(inode);
1502 spin_lock(&inode->i_lock);
1503 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
1504 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1507 status = nfs_update_inode(inode, fattr);
1509 spin_unlock(&inode->i_lock);
1514 * Many nfs protocol calls return the new file attributes after
1515 * an operation. Here we update the inode to reflect the state
1516 * of the server's inode.
1518 * This is a bit tricky because we have to make sure all dirty pages
1519 * have been sent off to the server before calling invalidate_inode_pages.
1520 * To make sure no other process adds more write requests while we try
1521 * our best to flush them, we make them sleep during the attribute refresh.
1523 * A very similar scenario holds for the dir cache.
1525 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1527 struct nfs_inode *nfsi = NFS_I(inode);
1528 loff_t cur_isize, new_isize;
1529 unsigned int invalid = 0;
1532 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1533 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1534 atomic_read(&inode->i_count), fattr->valid);
1536 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1539 if (nfsi->fileid != fattr->fileid)
1543 * Make sure the inode's type hasn't changed.
1545 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1549 * Update the read time so we don't revalidate too often.
1551 nfsi->read_cache_jiffies = fattr->time_start;
1552 nfsi->last_updated = jiffies;
1554 /* Are we racing with known updates of the metadata on the server? */
1555 data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
1557 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1559 /* Do atomic weak cache consistency updates */
1560 nfs_wcc_update_inode(inode, fattr);
1562 /* Check if our cached file size is stale */
1563 new_isize = nfs_size_to_loff_t(fattr->size);
1564 cur_isize = i_size_read(inode);
1565 if (new_isize != cur_isize) {
1566 /* Do we perhaps have any outstanding writes? */
1567 if (nfsi->npages == 0) {
1568 /* No, but did we race with nfs_end_data_update()? */
1570 inode->i_size = new_isize;
1571 invalid |= NFS_INO_INVALID_DATA;
1573 invalid |= NFS_INO_INVALID_ATTR;
1574 } else if (new_isize > cur_isize) {
1575 inode->i_size = new_isize;
1576 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1578 nfsi->cache_change_attribute = jiffies;
1579 dprintk("NFS: isize change on server for file %s/%ld\n",
1580 inode->i_sb->s_id, inode->i_ino);
1583 /* Check if the mtime agrees */
1584 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1585 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1586 dprintk("NFS: mtime change on server for file %s/%ld\n",
1587 inode->i_sb->s_id, inode->i_ino);
1588 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1589 nfsi->cache_change_attribute = jiffies;
1592 /* If ctime has changed we should definitely clear access+acl caches */
1593 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1594 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1595 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1596 nfsi->cache_change_attribute = jiffies;
1598 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1600 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1601 inode->i_uid != fattr->uid ||
1602 inode->i_gid != fattr->gid)
1603 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1605 inode->i_mode = fattr->mode;
1606 inode->i_nlink = fattr->nlink;
1607 inode->i_uid = fattr->uid;
1608 inode->i_gid = fattr->gid;
1610 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1612 * report the blocks in 512byte units
1614 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1615 inode->i_blksize = inode->i_sb->s_blocksize;
1617 inode->i_blocks = fattr->du.nfs2.blocks;
1618 inode->i_blksize = fattr->du.nfs2.blocksize;
1621 if ((fattr->valid & NFS_ATTR_FATTR_V4)) {
1622 if (nfsi->change_attr != fattr->change_attr) {
1623 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1624 inode->i_sb->s_id, inode->i_ino);
1625 nfsi->change_attr = fattr->change_attr;
1626 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1627 nfsi->cache_change_attribute = jiffies;
1629 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA);
1632 /* Update attrtimeo value if we're out of the unstable period */
1633 if (invalid & NFS_INO_INVALID_ATTR) {
1634 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1635 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1636 nfsi->attrtimeo_timestamp = jiffies;
1637 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1638 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1639 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1640 nfsi->attrtimeo_timestamp = jiffies;
1642 /* Don't invalidate the data if we were to blame */
1643 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1644 || S_ISLNK(inode->i_mode)))
1645 invalid &= ~NFS_INO_INVALID_DATA;
1647 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1648 if (!nfs_have_delegation(inode, FMODE_READ))
1649 nfsi->cache_validity |= invalid;
1654 * Big trouble! The inode has become a different object.
1657 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1658 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1662 * No need to worry about unhashing the dentry, as the
1663 * lookup validation will know that the inode is bad.
1664 * (But we fall through to invalidate the caches.)
1666 nfs_invalidate_inode(inode);
1670 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1671 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1672 NFS_SERVER(inode)->hostname, inode->i_sb->s_id,
1673 (long long)nfsi->fileid, (long long)fattr->fileid);
1678 * File system information
1681 static int nfs_set_super(struct super_block *s, void *data)
1683 s->s_fs_info = data;
1684 return set_anon_super(s, data);
1687 static int nfs_compare_super(struct super_block *sb, void *data)
1689 struct nfs_server *server = data;
1690 struct nfs_server *old = NFS_SB(sb);
1692 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1694 if (old->addr.sin_port != server->addr.sin_port)
1696 return !nfs_compare_fh(&old->fh, &server->fh);
1699 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1700 int flags, const char *dev_name, void *raw_data)
1703 struct nfs_server *server = NULL;
1704 struct super_block *s;
1705 struct nfs_fh *root;
1706 struct nfs_mount_data *data = raw_data;
1708 s = ERR_PTR(-EINVAL);
1710 dprintk("%s: missing data argument\n", __FUNCTION__);
1713 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1714 dprintk("%s: bad mount version\n", __FUNCTION__);
1717 switch (data->version) {
1723 if (data->flags & NFS_MOUNT_VER3) {
1724 dprintk("%s: mount structure version %d does not support NFSv3\n",
1729 data->root.size = NFS2_FHSIZE;
1730 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1732 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1733 dprintk("%s: mount structure version %d does not support strong security\n",
1739 memset(data->context, 0, sizeof(data->context));
1741 #ifndef CONFIG_NFS_V3
1742 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1743 s = ERR_PTR(-EPROTONOSUPPORT);
1744 if (data->flags & NFS_MOUNT_VER3) {
1745 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1748 #endif /* CONFIG_NFS_V3 */
1750 s = ERR_PTR(-ENOMEM);
1751 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
1754 /* Zero out the NFS state stuff */
1755 init_nfsv4_state(server);
1756 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1759 if (data->flags & NFS_MOUNT_VER3)
1760 root->size = data->root.size;
1762 root->size = NFS2_FHSIZE;
1763 s = ERR_PTR(-EINVAL);
1764 if (root->size > sizeof(root->data)) {
1765 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1768 memcpy(root->data, data->root.data, root->size);
1770 /* We now require that the mount process passes the remote address */
1771 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1772 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1773 dprintk("%s: mount program didn't pass remote address!\n",
1778 /* Fire up rpciod if not yet running */
1779 s = ERR_PTR(rpciod_up());
1781 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1782 __FUNCTION__, PTR_ERR(s));
1786 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1787 if (IS_ERR(s) || s->s_root)
1788 goto out_rpciod_down;
1792 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1794 up_write(&s->s_umount);
1795 deactivate_super(s);
1796 return ERR_PTR(error);
1798 s->s_flags |= MS_ACTIVE;
1807 static void nfs_kill_super(struct super_block *s)
1809 struct nfs_server *server = NFS_SB(s);
1813 if (!IS_ERR(server->client))
1814 rpc_shutdown_client(server->client);
1815 if (!IS_ERR(server->client_sys))
1816 rpc_shutdown_client(server->client_sys);
1817 if (!IS_ERR(server->client_acl))
1818 rpc_shutdown_client(server->client_acl);
1820 if (!(server->flags & NFS_MOUNT_NONLM))
1821 lockd_down(); /* release rpc.lockd */
1823 rpciod_down(); /* release rpciod */
1825 kfree(server->hostname);
1829 static struct file_system_type nfs_fs_type = {
1830 .owner = THIS_MODULE,
1832 .get_sb = nfs_get_sb,
1833 .kill_sb = nfs_kill_super,
1834 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1837 #ifdef CONFIG_NFS_V4
1839 static void nfs4_clear_inode(struct inode *);
1842 static struct super_operations nfs4_sops = {
1843 .alloc_inode = nfs_alloc_inode,
1844 .destroy_inode = nfs_destroy_inode,
1845 .write_inode = nfs_write_inode,
1846 .delete_inode = nfs_delete_inode,
1847 .statfs = nfs_statfs,
1848 .clear_inode = nfs4_clear_inode,
1849 .umount_begin = nfs_umount_begin,
1850 .show_options = nfs_show_options,
1851 .show_stats = nfs_show_stats,
1855 * Clean out any remaining NFSv4 state that might be left over due
1856 * to open() calls that passed nfs_atomic_lookup, but failed to call
1859 static void nfs4_clear_inode(struct inode *inode)
1861 struct nfs_inode *nfsi = NFS_I(inode);
1863 /* If we are holding a delegation, return it! */
1864 nfs_inode_return_delegation(inode);
1865 /* First call standard NFS clear_inode() code */
1866 nfs_clear_inode(inode);
1867 /* Now clear out any remaining state */
1868 while (!list_empty(&nfsi->open_states)) {
1869 struct nfs4_state *state;
1871 state = list_entry(nfsi->open_states.next,
1874 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1877 (long long)NFS_FILEID(inode),
1879 BUG_ON(atomic_read(&state->count) != 1);
1880 nfs4_close_state(state, state->state);
1885 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1887 struct nfs_server *server;
1888 struct nfs4_client *clp = NULL;
1889 struct rpc_xprt *xprt = NULL;
1890 struct rpc_clnt *clnt = NULL;
1891 struct rpc_timeout timeparms;
1892 rpc_authflavor_t authflavour;
1895 sb->s_blocksize_bits = 0;
1896 sb->s_blocksize = 0;
1897 server = NFS_SB(sb);
1898 if (data->rsize != 0)
1899 server->rsize = nfs_block_size(data->rsize, NULL);
1900 if (data->wsize != 0)
1901 server->wsize = nfs_block_size(data->wsize, NULL);
1902 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1903 server->caps = NFS_CAP_ATOMIC_OPEN;
1905 server->acregmin = data->acregmin*HZ;
1906 server->acregmax = data->acregmax*HZ;
1907 server->acdirmin = data->acdirmin*HZ;
1908 server->acdirmax = data->acdirmax*HZ;
1910 server->rpc_ops = &nfs_v4_clientops;
1912 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
1914 server->retrans_timeo = timeparms.to_initval;
1915 server->retrans_count = timeparms.to_retries;
1917 clp = nfs4_get_client(&server->addr.sin_addr);
1919 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1923 /* Now create transport and client */
1924 authflavour = RPC_AUTH_UNIX;
1925 if (data->auth_flavourlen != 0) {
1926 if (data->auth_flavourlen != 1) {
1927 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1928 __FUNCTION__, data->auth_flavourlen);
1932 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1938 down_write(&clp->cl_sem);
1939 if (IS_ERR(clp->cl_rpcclient)) {
1940 xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
1942 up_write(&clp->cl_sem);
1943 err = PTR_ERR(xprt);
1944 dprintk("%s: cannot create RPC transport. Error = %d\n",
1948 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1949 server->rpc_ops->version, authflavour);
1951 up_write(&clp->cl_sem);
1952 err = PTR_ERR(clnt);
1953 dprintk("%s: cannot create RPC client. Error = %d\n",
1958 clnt->cl_softrtry = 1;
1959 clp->cl_rpcclient = clnt;
1960 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1963 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1964 clnt = rpc_clone_client(clp->cl_rpcclient);
1966 server->nfs4_state = clp;
1967 up_write(&clp->cl_sem);
1971 err = PTR_ERR(clnt);
1972 dprintk("%s: cannot create RPC client. Error = %d\n",
1977 server->client = clnt;
1979 if (server->nfs4_state->cl_idmap == NULL) {
1980 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1984 if (clnt->cl_auth->au_flavor != authflavour) {
1985 struct rpc_auth *auth;
1987 auth = rpcauth_create(authflavour, clnt);
1989 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1990 return PTR_ERR(auth);
1994 sb->s_time_gran = 1;
1996 sb->s_op = &nfs4_sops;
1997 err = nfs_sb_init(sb, authflavour);
2002 nfs4_put_client(clp);
2006 static int nfs4_compare_super(struct super_block *sb, void *data)
2008 struct nfs_server *server = data;
2009 struct nfs_server *old = NFS_SB(sb);
2011 if (strcmp(server->hostname, old->hostname) != 0)
2013 if (strcmp(server->mnt_path, old->mnt_path) != 0)
2019 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
2024 return ERR_PTR(-EINVAL);
2025 if (src->len < maxlen)
2028 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
2030 return ERR_PTR(-ENOMEM);
2032 if (copy_from_user(dst, src->data, maxlen)) {
2034 return ERR_PTR(-EFAULT);
2040 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
2041 int flags, const char *dev_name, void *raw_data)
2044 struct nfs_server *server;
2045 struct super_block *s;
2046 struct nfs4_mount_data *data = raw_data;
2050 dprintk("%s: missing data argument\n", __FUNCTION__);
2051 return ERR_PTR(-EINVAL);
2053 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
2054 dprintk("%s: bad mount version\n", __FUNCTION__);
2055 return ERR_PTR(-EINVAL);
2058 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
2060 return ERR_PTR(-ENOMEM);
2061 /* Zero out the NFS state stuff */
2062 init_nfsv4_state(server);
2063 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
2065 p = nfs_copy_user_string(NULL, &data->hostname, 256);
2068 server->hostname = p;
2070 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
2073 server->mnt_path = p;
2075 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
2076 sizeof(server->ip_addr) - 1);
2080 /* We now require that the mount process passes the remote address */
2081 if (data->host_addrlen != sizeof(server->addr)) {
2082 s = ERR_PTR(-EINVAL);
2085 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
2086 s = ERR_PTR(-EFAULT);
2089 if (server->addr.sin_family != AF_INET ||
2090 server->addr.sin_addr.s_addr == INADDR_ANY) {
2091 dprintk("%s: mount program didn't pass remote IP address!\n",
2093 s = ERR_PTR(-EINVAL);
2097 /* Fire up rpciod if not yet running */
2098 s = ERR_PTR(rpciod_up());
2100 dprintk("%s: couldn't start rpciod! Error = %ld\n",
2101 __FUNCTION__, PTR_ERR(s));
2105 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
2107 if (IS_ERR(s) || s->s_root)
2112 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
2114 up_write(&s->s_umount);
2115 deactivate_super(s);
2116 return ERR_PTR(error);
2118 s->s_flags |= MS_ACTIVE;
2121 s = (struct super_block *)p;
2123 kfree(server->mnt_path);
2124 kfree(server->hostname);
2125 nfs_free_iostats(server->io_stats);
2130 static void nfs4_kill_super(struct super_block *sb)
2132 struct nfs_server *server = NFS_SB(sb);
2134 nfs_return_all_delegations(sb);
2135 kill_anon_super(sb);
2137 nfs4_renewd_prepare_shutdown(server);
2139 if (server->client != NULL && !IS_ERR(server->client))
2140 rpc_shutdown_client(server->client);
2142 destroy_nfsv4_state(server);
2146 kfree(server->hostname);
2150 static struct file_system_type nfs4_fs_type = {
2151 .owner = THIS_MODULE,
2153 .get_sb = nfs4_get_sb,
2154 .kill_sb = nfs4_kill_super,
2155 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2158 static const int nfs_set_port_min = 0;
2159 static const int nfs_set_port_max = 65535;
2160 static int param_set_port(const char *val, struct kernel_param *kp)
2163 int num = simple_strtol(val, &endp, 0);
2164 if (endp == val || *endp || num < nfs_set_port_min || num > nfs_set_port_max)
2166 *((int *)kp->arg) = num;
2170 module_param_call(callback_tcpport, param_set_port, param_get_int,
2171 &nfs_callback_set_tcpport, 0644);
2173 static int param_set_idmap_timeout(const char *val, struct kernel_param *kp)
2176 int num = simple_strtol(val, &endp, 0);
2178 if (endp == val || *endp || num < 0 || jif < num)
2180 *((int *)kp->arg) = jif;
2184 module_param_call(idmap_cache_timeout, param_set_idmap_timeout, param_get_int,
2185 &nfs_idmap_cache_timeout, 0644);
2187 #define nfs4_init_once(nfsi) \
2189 INIT_LIST_HEAD(&(nfsi)->open_states); \
2190 nfsi->delegation = NULL; \
2191 nfsi->delegation_state = 0; \
2192 init_rwsem(&nfsi->rwsem); \
2195 static inline int register_nfs4fs(void)
2199 ret = nfs_register_sysctl();
2202 ret = register_filesystem(&nfs4_fs_type);
2204 nfs_unregister_sysctl();
2208 static inline void unregister_nfs4fs(void)
2210 unregister_filesystem(&nfs4_fs_type);
2211 nfs_unregister_sysctl();
2214 #define nfs4_init_once(nfsi) \
2216 #define register_nfs4fs() (0)
2217 #define unregister_nfs4fs()
2220 extern int nfs_init_nfspagecache(void);
2221 extern void nfs_destroy_nfspagecache(void);
2222 extern int nfs_init_readpagecache(void);
2223 extern void nfs_destroy_readpagecache(void);
2224 extern int nfs_init_writepagecache(void);
2225 extern void nfs_destroy_writepagecache(void);
2226 #ifdef CONFIG_NFS_DIRECTIO
2227 extern int nfs_init_directcache(void);
2228 extern void nfs_destroy_directcache(void);
2231 static kmem_cache_t * nfs_inode_cachep;
2233 static struct inode *nfs_alloc_inode(struct super_block *sb)
2235 struct nfs_inode *nfsi;
2236 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2240 nfsi->cache_validity = 0UL;
2241 nfsi->cache_change_attribute = jiffies;
2242 #ifdef CONFIG_NFS_V3_ACL
2243 nfsi->acl_access = ERR_PTR(-EAGAIN);
2244 nfsi->acl_default = ERR_PTR(-EAGAIN);
2246 #ifdef CONFIG_NFS_V4
2247 nfsi->nfs4_acl = NULL;
2248 #endif /* CONFIG_NFS_V4 */
2249 return &nfsi->vfs_inode;
2252 static void nfs_destroy_inode(struct inode *inode)
2254 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2257 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2259 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2261 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2262 SLAB_CTOR_CONSTRUCTOR) {
2263 inode_init_once(&nfsi->vfs_inode);
2264 spin_lock_init(&nfsi->req_lock);
2265 INIT_LIST_HEAD(&nfsi->dirty);
2266 INIT_LIST_HEAD(&nfsi->commit);
2267 INIT_LIST_HEAD(&nfsi->open_files);
2268 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2269 atomic_set(&nfsi->data_updates, 0);
2273 nfs4_init_once(nfsi);
2277 static int nfs_init_inodecache(void)
2279 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2280 sizeof(struct nfs_inode),
2281 0, SLAB_RECLAIM_ACCOUNT,
2283 if (nfs_inode_cachep == NULL)
2289 static void nfs_destroy_inodecache(void)
2291 if (kmem_cache_destroy(nfs_inode_cachep))
2292 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2298 static int __init init_nfs_fs(void)
2302 err = nfs_init_nfspagecache();
2306 err = nfs_init_inodecache();
2310 err = nfs_init_readpagecache();
2314 err = nfs_init_writepagecache();
2318 #ifdef CONFIG_NFS_DIRECTIO
2319 err = nfs_init_directcache();
2324 #ifdef CONFIG_PROC_FS
2325 rpc_proc_register(&nfs_rpcstat);
2327 err = register_filesystem(&nfs_fs_type);
2330 if ((err = register_nfs4fs()) != 0)
2334 #ifdef CONFIG_PROC_FS
2335 rpc_proc_unregister("nfs");
2337 #ifdef CONFIG_NFS_DIRECTIO
2338 nfs_destroy_directcache();
2341 nfs_destroy_writepagecache();
2343 nfs_destroy_readpagecache();
2345 nfs_destroy_inodecache();
2347 nfs_destroy_nfspagecache();
2352 static void __exit exit_nfs_fs(void)
2354 #ifdef CONFIG_NFS_DIRECTIO
2355 nfs_destroy_directcache();
2357 nfs_destroy_writepagecache();
2358 nfs_destroy_readpagecache();
2359 nfs_destroy_inodecache();
2360 nfs_destroy_nfspagecache();
2361 #ifdef CONFIG_PROC_FS
2362 rpc_proc_unregister("nfs");
2364 unregister_filesystem(&nfs_fs_type);
2365 unregister_nfs4fs();
2368 /* Not quite true; I just maintain it */
2369 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2370 MODULE_LICENSE("GPL");
2372 module_init(init_nfs_fs)
2373 module_exit(exit_nfs_fs)