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 server->io_stats = nfs_alloc_iostats();
285 if (server->io_stats == NULL)
288 root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
289 /* Did getting the root inode fail? */
290 if (IS_ERR(root_inode)) {
291 no_root_error = PTR_ERR(root_inode);
294 sb->s_root = d_alloc_root(root_inode);
296 no_root_error = -ENOMEM;
299 sb->s_root->d_op = server->rpc_ops->dentry_ops;
301 /* mount time stamp, in seconds */
302 server->mount_time = jiffies;
304 /* Get some general file system info */
305 if (server->namelen == 0 &&
306 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
307 server->namelen = pathinfo.max_namelen;
308 /* Work out a lot of parameters */
309 if (server->rsize == 0)
310 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
311 if (server->wsize == 0)
312 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
314 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
315 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
316 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
317 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
319 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
320 if (server->rsize > max_rpc_payload)
321 server->rsize = max_rpc_payload;
322 if (server->rsize > NFS_MAX_FILE_IO_SIZE)
323 server->rsize = NFS_MAX_FILE_IO_SIZE;
324 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
326 if (server->wsize > max_rpc_payload)
327 server->wsize = max_rpc_payload;
328 if (server->wsize > NFS_MAX_FILE_IO_SIZE)
329 server->wsize = NFS_MAX_FILE_IO_SIZE;
330 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
332 if (sb->s_blocksize == 0)
333 sb->s_blocksize = nfs_block_bits(server->wsize,
334 &sb->s_blocksize_bits);
335 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
337 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
338 if (server->dtsize > PAGE_CACHE_SIZE)
339 server->dtsize = PAGE_CACHE_SIZE;
340 if (server->dtsize > server->rsize)
341 server->dtsize = server->rsize;
343 if (server->flags & NFS_MOUNT_NOAC) {
344 server->acregmin = server->acregmax = 0;
345 server->acdirmin = server->acdirmax = 0;
346 sb->s_flags |= MS_SYNCHRONOUS;
348 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
350 sb->s_maxbytes = fsinfo.maxfilesize;
351 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
352 sb->s_maxbytes = MAX_LFS_FILESIZE;
354 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
355 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
357 /* We're airborne Set socket buffersize */
358 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
360 /* Yargs. It didn't work out. */
362 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
363 if (!IS_ERR(root_inode))
365 return no_root_error;
368 static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
370 to->to_initval = timeo * HZ / 10;
371 to->to_retries = retrans;
378 to->to_initval = 60 * HZ;
379 if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
380 to->to_initval = NFS_MAX_TCP_TIMEOUT;
381 to->to_increment = to->to_initval;
382 to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
383 to->to_exponential = 0;
388 to->to_initval = 11 * HZ / 10;
389 if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
390 to->to_initval = NFS_MAX_UDP_TIMEOUT;
391 to->to_maxval = NFS_MAX_UDP_TIMEOUT;
392 to->to_exponential = 1;
398 * Create an RPC client handle.
400 static struct rpc_clnt *
401 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
403 struct rpc_timeout timeparms;
404 struct rpc_xprt *xprt = NULL;
405 struct rpc_clnt *clnt = NULL;
406 int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
408 nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
410 server->retrans_timeo = timeparms.to_initval;
411 server->retrans_count = timeparms.to_retries;
413 /* create transport and client */
414 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
416 dprintk("%s: cannot create RPC transport. Error = %ld\n",
417 __FUNCTION__, PTR_ERR(xprt));
418 return (struct rpc_clnt *)xprt;
420 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
421 server->rpc_ops->version, data->pseudoflavor);
423 dprintk("%s: cannot create RPC client. Error = %ld\n",
424 __FUNCTION__, PTR_ERR(xprt));
429 clnt->cl_softrtry = 1;
438 * The way this works is that the mount process passes a structure
439 * in the data argument which contains the server's IP address
440 * and the root file handle obtained from the server's mount
441 * daemon. We stash these away in the private superblock fields.
444 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
446 struct nfs_server *server;
447 rpc_authflavor_t authflavor;
450 sb->s_blocksize_bits = 0;
453 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
455 server->rsize = nfs_block_size(data->rsize, NULL);
457 server->wsize = nfs_block_size(data->wsize, NULL);
458 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
460 server->acregmin = data->acregmin*HZ;
461 server->acregmax = data->acregmax*HZ;
462 server->acdirmin = data->acdirmin*HZ;
463 server->acdirmax = data->acdirmax*HZ;
465 /* Start lockd here, before we might error out */
466 if (!(server->flags & NFS_MOUNT_NONLM))
469 server->namelen = data->namlen;
470 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
471 if (!server->hostname)
473 strcpy(server->hostname, data->hostname);
475 /* Check NFS protocol revision and initialize RPC op vector
476 * and file handle pool. */
478 if (server->flags & NFS_MOUNT_VER3) {
479 server->rpc_ops = &nfs_v3_clientops;
480 server->caps |= NFS_CAP_READDIRPLUS;
482 server->rpc_ops = &nfs_v2_clientops;
485 server->rpc_ops = &nfs_v2_clientops;
488 /* Fill in pseudoflavor for mount version < 5 */
489 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
490 data->pseudoflavor = RPC_AUTH_UNIX;
491 authflavor = data->pseudoflavor; /* save for sb_init() */
492 /* XXX maybe we want to add a server->pseudoflavor field */
494 /* Create RPC client handles */
495 server->client = nfs_create_client(server, data);
496 if (IS_ERR(server->client))
497 return PTR_ERR(server->client);
498 /* RFC 2623, sec 2.3.2 */
499 if (authflavor != RPC_AUTH_UNIX) {
500 struct rpc_auth *auth;
502 server->client_sys = rpc_clone_client(server->client);
503 if (IS_ERR(server->client_sys))
504 return PTR_ERR(server->client_sys);
505 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
507 return PTR_ERR(auth);
509 atomic_inc(&server->client->cl_count);
510 server->client_sys = server->client;
512 if (server->flags & NFS_MOUNT_VER3) {
513 #ifdef CONFIG_NFS_V3_ACL
514 if (!(server->flags & NFS_MOUNT_NOACL)) {
515 server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
516 /* No errors! Assume that Sun nfsacls are supported */
517 if (!IS_ERR(server->client_acl))
518 server->caps |= NFS_CAP_ACLS;
521 server->flags &= ~NFS_MOUNT_NOACL;
522 #endif /* CONFIG_NFS_V3_ACL */
524 * The VFS shouldn't apply the umask to mode bits. We will
525 * do so ourselves when necessary.
527 sb->s_flags |= MS_POSIXACL;
528 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
529 server->namelen = NFS3_MAXNAMLEN;
532 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
533 server->namelen = NFS2_MAXNAMLEN;
536 sb->s_op = &nfs_sops;
537 return nfs_sb_init(sb, authflavor);
541 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
543 struct nfs_server *server = NFS_SB(sb);
544 unsigned char blockbits;
545 unsigned long blockres;
546 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
547 struct nfs_fattr fattr;
548 struct nfs_fsstat res = {
555 error = server->rpc_ops->statfs(server, rootfh, &res);
556 buf->f_type = NFS_SUPER_MAGIC;
561 * Current versions of glibc do not correctly handle the
562 * case where f_frsize != f_bsize. Eventually we want to
563 * report the value of wtmult in this field.
565 buf->f_frsize = sb->s_blocksize;
568 * On most *nix systems, f_blocks, f_bfree, and f_bavail
569 * are reported in units of f_frsize. Linux hasn't had
570 * an f_frsize field in its statfs struct until recently,
571 * thus historically Linux's sys_statfs reports these
572 * fields in units of f_bsize.
574 buf->f_bsize = sb->s_blocksize;
575 blockbits = sb->s_blocksize_bits;
576 blockres = (1 << blockbits) - 1;
577 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
578 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
579 buf->f_bavail = (res.abytes + blockres) >> blockbits;
581 buf->f_files = res.tfiles;
582 buf->f_ffree = res.afiles;
584 buf->f_namelen = server->namelen;
590 dprintk("%s: statfs error = %d\n", __FUNCTION__, -error);
591 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
596 static void nfs_show_mount_options(struct seq_file *m, struct nfs_server *nfss, int showdefaults)
598 static struct proc_nfs_info {
603 { NFS_MOUNT_SOFT, ",soft", ",hard" },
604 { NFS_MOUNT_INTR, ",intr", "" },
605 { NFS_MOUNT_NOCTO, ",nocto", "" },
606 { NFS_MOUNT_NOAC, ",noac", "" },
607 { NFS_MOUNT_NONLM, ",nolock", "" },
608 { NFS_MOUNT_NOACL, ",noacl", "" },
611 struct proc_nfs_info *nfs_infop;
615 seq_printf(m, ",vers=%d", nfss->rpc_ops->version);
616 seq_printf(m, ",rsize=%d", nfss->rsize);
617 seq_printf(m, ",wsize=%d", nfss->wsize);
618 if (nfss->acregmin != 3*HZ || showdefaults)
619 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
620 if (nfss->acregmax != 60*HZ || showdefaults)
621 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
622 if (nfss->acdirmin != 30*HZ || showdefaults)
623 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
624 if (nfss->acdirmax != 60*HZ || showdefaults)
625 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
626 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
627 if (nfss->flags & nfs_infop->flag)
628 seq_puts(m, nfs_infop->str);
630 seq_puts(m, nfs_infop->nostr);
632 switch (nfss->client->cl_xprt->prot) {
640 snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
643 seq_printf(m, ",proto=%s", proto);
644 seq_printf(m, ",timeo=%lu", 10U * nfss->retrans_timeo / HZ);
645 seq_printf(m, ",retrans=%u", nfss->retrans_count);
648 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
650 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
652 nfs_show_mount_options(m, nfss, 0);
654 seq_puts(m, ",addr=");
655 seq_escape(m, nfss->hostname, " \t\n\\");
660 static int nfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
663 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
664 struct rpc_auth *auth = nfss->client->cl_auth;
665 struct nfs_iostats totals = { };
667 seq_printf(m, "statvers=%s", NFS_IOSTAT_VERS);
670 * Display all mount option settings
672 seq_printf(m, "\n\topts:\t");
673 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? "ro" : "rw");
674 seq_puts(m, mnt->mnt_sb->s_flags & MS_SYNCHRONOUS ? ",sync" : "");
675 seq_puts(m, mnt->mnt_sb->s_flags & MS_NOATIME ? ",noatime" : "");
676 seq_puts(m, mnt->mnt_sb->s_flags & MS_NODIRATIME ? ",nodiratime" : "");
677 nfs_show_mount_options(m, nfss, 1);
679 seq_printf(m, "\n\tage:\t%lu", (jiffies - nfss->mount_time) / HZ);
681 seq_printf(m, "\n\tcaps:\t");
682 seq_printf(m, "caps=0x%x", nfss->caps);
683 seq_printf(m, ",wtmult=%d", nfss->wtmult);
684 seq_printf(m, ",dtsize=%d", nfss->dtsize);
685 seq_printf(m, ",bsize=%d", nfss->bsize);
686 seq_printf(m, ",namelen=%d", nfss->namelen);
689 if (nfss->rpc_ops->version == 4) {
690 seq_printf(m, "\n\tnfsv4:\t");
691 seq_printf(m, "bm0=0x%x", nfss->attr_bitmask[0]);
692 seq_printf(m, ",bm1=0x%x", nfss->attr_bitmask[1]);
693 seq_printf(m, ",acl=0x%x", nfss->acl_bitmask);
698 * Display security flavor in effect for this mount
700 seq_printf(m, "\n\tsec:\tflavor=%d", auth->au_ops->au_flavor);
702 seq_printf(m, ",pseudoflavor=%d", auth->au_flavor);
705 * Display superblock I/O counters
707 for (cpu = 0; cpu < NR_CPUS; cpu++) {
708 struct nfs_iostats *stats;
710 if (!cpu_possible(cpu))
714 stats = per_cpu_ptr(nfss->io_stats, cpu);
716 for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
717 totals.events[i] += stats->events[i];
718 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
719 totals.bytes[i] += stats->bytes[i];
724 seq_printf(m, "\n\tevents:\t");
725 for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
726 seq_printf(m, "%lu ", totals.events[i]);
727 seq_printf(m, "\n\tbytes:\t");
728 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
729 seq_printf(m, "%Lu ", totals.bytes[i]);
732 rpc_print_iostats(m, nfss->client);
738 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
740 int nfs_sync_mapping(struct address_space *mapping)
744 if (mapping->nrpages == 0)
746 unmap_mapping_range(mapping, 0, 0, 0);
747 ret = filemap_write_and_wait(mapping);
750 ret = nfs_wb_all(mapping->host);
756 * Invalidate the local caches
758 static void nfs_zap_caches_locked(struct inode *inode)
760 struct nfs_inode *nfsi = NFS_I(inode);
761 int mode = inode->i_mode;
763 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
765 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
766 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
768 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
769 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
770 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
772 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
775 void nfs_zap_caches(struct inode *inode)
777 spin_lock(&inode->i_lock);
778 nfs_zap_caches_locked(inode);
779 spin_unlock(&inode->i_lock);
782 static void nfs_zap_acl_cache(struct inode *inode)
784 void (*clear_acl_cache)(struct inode *);
786 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
787 if (clear_acl_cache != NULL)
788 clear_acl_cache(inode);
789 spin_lock(&inode->i_lock);
790 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
791 spin_unlock(&inode->i_lock);
795 * Invalidate, but do not unhash, the inode.
796 * NB: must be called with inode->i_lock held!
798 static void nfs_invalidate_inode(struct inode *inode)
800 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
801 nfs_zap_caches_locked(inode);
804 struct nfs_find_desc {
806 struct nfs_fattr *fattr;
810 * In NFSv3 we can have 64bit inode numbers. In order to support
811 * this, and re-exported directories (also seen in NFSv2)
812 * we are forced to allow 2 different inodes to have the same
816 nfs_find_actor(struct inode *inode, void *opaque)
818 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
819 struct nfs_fh *fh = desc->fh;
820 struct nfs_fattr *fattr = desc->fattr;
822 if (NFS_FILEID(inode) != fattr->fileid)
824 if (nfs_compare_fh(NFS_FH(inode), fh))
826 if (is_bad_inode(inode) || NFS_STALE(inode))
832 nfs_init_locked(struct inode *inode, void *opaque)
834 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
835 struct nfs_fattr *fattr = desc->fattr;
837 NFS_FILEID(inode) = fattr->fileid;
838 nfs_copy_fh(NFS_FH(inode), desc->fh);
842 /* Don't use READDIRPLUS on directories that we believe are too large */
843 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
846 * This is our front-end to iget that looks up inodes by file handle
847 * instead of inode number.
850 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
852 struct nfs_find_desc desc = {
856 struct inode *inode = NULL;
859 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
863 printk("NFS: Buggy server - nlink == 0!\n");
867 hash = nfs_fattr_to_ino_t(fattr);
869 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
872 if (inode->i_state & I_NEW) {
873 struct nfs_inode *nfsi = NFS_I(inode);
875 /* We set i_ino for the few things that still rely on it,
879 /* We can't support update_atime(), since the server will reset it */
880 inode->i_flags |= S_NOATIME|S_NOCMTIME;
881 inode->i_mode = fattr->mode;
882 /* Why so? Because we want revalidate for devices/FIFOs, and
883 * that's precisely what we have in nfs_file_inode_operations.
885 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
886 if (S_ISREG(inode->i_mode)) {
887 inode->i_fop = &nfs_file_operations;
888 inode->i_data.a_ops = &nfs_file_aops;
889 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
890 } else if (S_ISDIR(inode->i_mode)) {
891 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
892 inode->i_fop = &nfs_dir_operations;
893 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
894 && fattr->size <= NFS_LIMIT_READDIRPLUS)
895 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
896 } else if (S_ISLNK(inode->i_mode))
897 inode->i_op = &nfs_symlink_inode_operations;
899 init_special_inode(inode, inode->i_mode, fattr->rdev);
901 nfsi->read_cache_jiffies = fattr->time_start;
902 nfsi->last_updated = jiffies;
903 inode->i_atime = fattr->atime;
904 inode->i_mtime = fattr->mtime;
905 inode->i_ctime = fattr->ctime;
906 if (fattr->valid & NFS_ATTR_FATTR_V4)
907 nfsi->change_attr = fattr->change_attr;
908 inode->i_size = nfs_size_to_loff_t(fattr->size);
909 inode->i_nlink = fattr->nlink;
910 inode->i_uid = fattr->uid;
911 inode->i_gid = fattr->gid;
912 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
914 * report the blocks in 512byte units
916 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
917 inode->i_blksize = inode->i_sb->s_blocksize;
919 inode->i_blocks = fattr->du.nfs2.blocks;
920 inode->i_blksize = fattr->du.nfs2.blocksize;
922 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
923 nfsi->attrtimeo_timestamp = jiffies;
924 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
925 nfsi->cache_access.cred = NULL;
927 unlock_new_inode(inode);
929 nfs_refresh_inode(inode, fattr);
930 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
932 (long long)NFS_FILEID(inode),
933 atomic_read(&inode->i_count));
939 printk("nfs_fhget: iget failed\n");
943 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
946 nfs_setattr(struct dentry *dentry, struct iattr *attr)
948 struct inode *inode = dentry->d_inode;
949 struct nfs_fattr fattr;
952 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
954 if (attr->ia_valid & ATTR_SIZE) {
955 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
956 attr->ia_valid &= ~ATTR_SIZE;
959 /* Optimization: if the end result is no change, don't RPC */
960 attr->ia_valid &= NFS_VALID_ATTRS;
961 if (attr->ia_valid == 0)
965 nfs_begin_data_update(inode);
966 /* Write all dirty data */
967 filemap_write_and_wait(inode->i_mapping);
970 * Return any delegations if we're going to change ACLs
972 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
973 nfs_inode_return_delegation(inode);
974 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
976 nfs_refresh_inode(inode, &fattr);
977 nfs_end_data_update(inode);
983 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
984 * @inode: pointer to struct inode
985 * @attr: pointer to struct iattr
987 * Note: we do this in the *proc.c in order to ensure that
988 * it works for things like exclusive creates too.
990 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
992 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
993 if ((attr->ia_valid & ATTR_MODE) != 0) {
994 int mode = attr->ia_mode & S_IALLUGO;
995 mode |= inode->i_mode & ~S_IALLUGO;
996 inode->i_mode = mode;
998 if ((attr->ia_valid & ATTR_UID) != 0)
999 inode->i_uid = attr->ia_uid;
1000 if ((attr->ia_valid & ATTR_GID) != 0)
1001 inode->i_gid = attr->ia_gid;
1002 spin_lock(&inode->i_lock);
1003 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1004 spin_unlock(&inode->i_lock);
1006 if ((attr->ia_valid & ATTR_SIZE) != 0) {
1007 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
1008 inode->i_size = attr->ia_size;
1009 vmtruncate(inode, attr->ia_size);
1013 static int nfs_wait_schedule(void *word)
1015 if (signal_pending(current))
1016 return -ERESTARTSYS;
1022 * Wait for the inode to get unlocked.
1024 static int nfs_wait_on_inode(struct inode *inode)
1026 struct rpc_clnt *clnt = NFS_CLIENT(inode);
1027 struct nfs_inode *nfsi = NFS_I(inode);
1031 rpc_clnt_sigmask(clnt, &oldmask);
1032 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
1033 nfs_wait_schedule, TASK_INTERRUPTIBLE);
1034 rpc_clnt_sigunmask(clnt, &oldmask);
1039 static void nfs_wake_up_inode(struct inode *inode)
1041 struct nfs_inode *nfsi = NFS_I(inode);
1043 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
1044 smp_mb__after_clear_bit();
1045 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
1048 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1050 struct inode *inode = dentry->d_inode;
1051 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
1054 /* Flush out writes to the server in order to update c/mtime */
1055 nfs_sync_inode(inode, 0, 0, FLUSH_WAIT|FLUSH_NOCOMMIT);
1058 * We may force a getattr if the user cares about atime.
1060 * Note that we only have to check the vfsmount flags here:
1061 * - NFS always sets S_NOATIME by so checking it would give a
1063 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
1064 * no point in checking those.
1066 if ((mnt->mnt_flags & MNT_NOATIME) ||
1067 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
1071 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1073 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1075 generic_fillattr(inode, stat);
1079 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
1081 struct nfs_open_context *ctx;
1083 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
1085 atomic_set(&ctx->count, 1);
1086 ctx->dentry = dget(dentry);
1087 ctx->vfsmnt = mntget(mnt);
1088 ctx->cred = get_rpccred(cred);
1090 ctx->lockowner = current->files;
1092 ctx->dir_cookie = 0;
1097 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1100 atomic_inc(&ctx->count);
1104 void put_nfs_open_context(struct nfs_open_context *ctx)
1106 if (atomic_dec_and_test(&ctx->count)) {
1107 if (!list_empty(&ctx->list)) {
1108 struct inode *inode = ctx->dentry->d_inode;
1109 spin_lock(&inode->i_lock);
1110 list_del(&ctx->list);
1111 spin_unlock(&inode->i_lock);
1113 if (ctx->state != NULL)
1114 nfs4_close_state(ctx->state, ctx->mode);
1115 if (ctx->cred != NULL)
1116 put_rpccred(ctx->cred);
1118 mntput(ctx->vfsmnt);
1124 * Ensure that mmap has a recent RPC credential for use when writing out
1127 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1129 struct inode *inode = filp->f_dentry->d_inode;
1130 struct nfs_inode *nfsi = NFS_I(inode);
1132 filp->private_data = get_nfs_open_context(ctx);
1133 spin_lock(&inode->i_lock);
1134 list_add(&ctx->list, &nfsi->open_files);
1135 spin_unlock(&inode->i_lock);
1139 * Given an inode, search for an open context with the desired characteristics
1141 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
1143 struct nfs_inode *nfsi = NFS_I(inode);
1144 struct nfs_open_context *pos, *ctx = NULL;
1146 spin_lock(&inode->i_lock);
1147 list_for_each_entry(pos, &nfsi->open_files, list) {
1148 if (cred != NULL && pos->cred != cred)
1150 if ((pos->mode & mode) == mode) {
1151 ctx = get_nfs_open_context(pos);
1155 spin_unlock(&inode->i_lock);
1159 static void nfs_file_clear_open_context(struct file *filp)
1161 struct inode *inode = filp->f_dentry->d_inode;
1162 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
1165 filp->private_data = NULL;
1166 spin_lock(&inode->i_lock);
1167 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1168 spin_unlock(&inode->i_lock);
1169 put_nfs_open_context(ctx);
1174 * These allocate and release file read/write context information.
1176 int nfs_open(struct inode *inode, struct file *filp)
1178 struct nfs_open_context *ctx;
1179 struct rpc_cred *cred;
1181 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1183 return PTR_ERR(cred);
1184 ctx = alloc_nfs_open_context(filp->f_vfsmnt, filp->f_dentry, cred);
1188 ctx->mode = filp->f_mode;
1189 nfs_file_set_open_context(filp, ctx);
1190 put_nfs_open_context(ctx);
1194 int nfs_release(struct inode *inode, struct file *filp)
1196 nfs_file_clear_open_context(filp);
1201 * This function is called whenever some part of NFS notices that
1202 * the cached attributes have to be refreshed.
1205 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1207 int status = -ESTALE;
1208 struct nfs_fattr fattr;
1209 struct nfs_inode *nfsi = NFS_I(inode);
1211 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1212 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1215 if (!inode || is_bad_inode(inode))
1217 if (NFS_STALE(inode))
1220 status = nfs_wait_on_inode(inode);
1223 if (NFS_STALE(inode)) {
1225 /* Do we trust the cached ESTALE? */
1226 if (NFS_ATTRTIMEO(inode) != 0) {
1227 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1234 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1236 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1238 (long long)NFS_FILEID(inode), status);
1239 if (status == -ESTALE) {
1240 nfs_zap_caches(inode);
1241 if (!S_ISDIR(inode->i_mode))
1242 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1247 spin_lock(&inode->i_lock);
1248 status = nfs_update_inode(inode, &fattr);
1250 spin_unlock(&inode->i_lock);
1251 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1253 (long long)NFS_FILEID(inode), status);
1256 spin_unlock(&inode->i_lock);
1258 nfs_revalidate_mapping(inode, inode->i_mapping);
1260 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1261 nfs_zap_acl_cache(inode);
1263 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1265 (long long)NFS_FILEID(inode));
1268 nfs_wake_up_inode(inode);
1275 int nfs_attribute_timeout(struct inode *inode)
1277 struct nfs_inode *nfsi = NFS_I(inode);
1279 if (nfs_have_delegation(inode, FMODE_READ))
1281 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1285 * nfs_revalidate_inode - Revalidate the inode attributes
1286 * @server - pointer to nfs_server struct
1287 * @inode - pointer to inode struct
1289 * Updates inode attribute information by retrieving the data from the server.
1291 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1293 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1294 if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1295 && !nfs_attribute_timeout(inode))
1296 return NFS_STALE(inode) ? -ESTALE : 0;
1297 return __nfs_revalidate_inode(server, inode);
1301 * nfs_revalidate_mapping - Revalidate the pagecache
1302 * @inode - pointer to host inode
1303 * @mapping - pointer to mapping
1305 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1307 struct nfs_inode *nfsi = NFS_I(inode);
1309 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1310 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1311 if (S_ISREG(inode->i_mode))
1312 nfs_sync_mapping(mapping);
1313 invalidate_inode_pages2(mapping);
1315 spin_lock(&inode->i_lock);
1316 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1317 if (S_ISDIR(inode->i_mode)) {
1318 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1319 /* This ensures we revalidate child dentries */
1320 nfsi->cache_change_attribute = jiffies;
1322 spin_unlock(&inode->i_lock);
1324 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1326 (long long)NFS_FILEID(inode));
1331 * nfs_begin_data_update
1332 * @inode - pointer to inode
1333 * Declare that a set of operations will update file data on the server
1335 void nfs_begin_data_update(struct inode *inode)
1337 atomic_inc(&NFS_I(inode)->data_updates);
1341 * nfs_end_data_update
1342 * @inode - pointer to inode
1343 * Declare end of the operations that will update file data
1344 * This will mark the inode as immediately needing revalidation
1345 * of its attribute cache.
1347 void nfs_end_data_update(struct inode *inode)
1349 struct nfs_inode *nfsi = NFS_I(inode);
1351 if (!nfs_have_delegation(inode, FMODE_READ)) {
1352 /* Directories and symlinks: invalidate page cache */
1353 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
1354 spin_lock(&inode->i_lock);
1355 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1356 spin_unlock(&inode->i_lock);
1359 nfsi->cache_change_attribute = jiffies;
1360 atomic_dec(&nfsi->data_updates);
1363 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1365 struct nfs_inode *nfsi = NFS_I(inode);
1367 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1368 && nfsi->change_attr == fattr->pre_change_attr) {
1369 nfsi->change_attr = fattr->change_attr;
1370 nfsi->cache_change_attribute = jiffies;
1373 /* If we have atomic WCC data, we may update some attributes */
1374 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1375 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1376 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1377 nfsi->cache_change_attribute = jiffies;
1379 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1380 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1381 nfsi->cache_change_attribute = jiffies;
1383 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
1384 inode->i_size = fattr->size;
1385 nfsi->cache_change_attribute = jiffies;
1391 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1392 * @inode - pointer to inode
1393 * @fattr - updated attributes
1395 * Verifies the attribute cache. If we have just changed the attributes,
1396 * so that fattr carries weak cache consistency data, then it may
1397 * also update the ctime/mtime/change_attribute.
1399 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1401 struct nfs_inode *nfsi = NFS_I(inode);
1402 loff_t cur_size, new_isize;
1406 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1409 /* Has the inode gone and changed behind our back? */
1410 if (nfsi->fileid != fattr->fileid
1411 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1415 /* Are we in the process of updating data on the server? */
1416 data_unstable = nfs_caches_unstable(inode);
1418 /* Do atomic weak cache consistency updates */
1419 nfs_wcc_update_inode(inode, fattr);
1421 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0) {
1422 if (nfsi->change_attr == fattr->change_attr)
1424 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1426 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1429 /* Verify a few of the more important attributes */
1430 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1431 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1433 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1436 cur_size = i_size_read(inode);
1437 new_isize = nfs_size_to_loff_t(fattr->size);
1438 if (cur_size != new_isize && nfsi->npages == 0)
1439 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1441 /* Have any file permissions changed? */
1442 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1443 || inode->i_uid != fattr->uid
1444 || inode->i_gid != fattr->gid)
1445 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1447 /* Has the link count changed? */
1448 if (inode->i_nlink != fattr->nlink)
1449 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1452 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1453 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1455 nfsi->read_cache_jiffies = fattr->time_start;
1460 * nfs_refresh_inode - try to update the inode attribute cache
1461 * @inode - pointer to inode
1462 * @fattr - updated attributes
1464 * Check that an RPC call that returned attributes has not overlapped with
1465 * other recent updates of the inode metadata, then decide whether it is
1466 * safe to do a full update of the inode attributes, or whether just to
1467 * call nfs_check_inode_attributes.
1469 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1471 struct nfs_inode *nfsi = NFS_I(inode);
1474 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1476 spin_lock(&inode->i_lock);
1477 nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1478 if (time_after(fattr->time_start, nfsi->last_updated))
1479 status = nfs_update_inode(inode, fattr);
1481 status = nfs_check_inode_attributes(inode, fattr);
1483 spin_unlock(&inode->i_lock);
1488 * nfs_post_op_update_inode - try to update the inode attribute cache
1489 * @inode - pointer to inode
1490 * @fattr - updated attributes
1492 * After an operation that has changed the inode metadata, mark the
1493 * attribute cache as being invalid, then try to update it.
1495 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1497 struct nfs_inode *nfsi = NFS_I(inode);
1500 spin_lock(&inode->i_lock);
1501 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
1502 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1505 status = nfs_update_inode(inode, fattr);
1507 spin_unlock(&inode->i_lock);
1512 * Many nfs protocol calls return the new file attributes after
1513 * an operation. Here we update the inode to reflect the state
1514 * of the server's inode.
1516 * This is a bit tricky because we have to make sure all dirty pages
1517 * have been sent off to the server before calling invalidate_inode_pages.
1518 * To make sure no other process adds more write requests while we try
1519 * our best to flush them, we make them sleep during the attribute refresh.
1521 * A very similar scenario holds for the dir cache.
1523 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1525 struct nfs_inode *nfsi = NFS_I(inode);
1526 loff_t cur_isize, new_isize;
1527 unsigned int invalid = 0;
1530 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1531 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1532 atomic_read(&inode->i_count), fattr->valid);
1534 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1537 if (nfsi->fileid != fattr->fileid)
1541 * Make sure the inode's type hasn't changed.
1543 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1547 * Update the read time so we don't revalidate too often.
1549 nfsi->read_cache_jiffies = fattr->time_start;
1550 nfsi->last_updated = jiffies;
1552 /* Are we racing with known updates of the metadata on the server? */
1553 data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
1555 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1557 /* Do atomic weak cache consistency updates */
1558 nfs_wcc_update_inode(inode, fattr);
1560 /* Check if our cached file size is stale */
1561 new_isize = nfs_size_to_loff_t(fattr->size);
1562 cur_isize = i_size_read(inode);
1563 if (new_isize != cur_isize) {
1564 /* Do we perhaps have any outstanding writes? */
1565 if (nfsi->npages == 0) {
1566 /* No, but did we race with nfs_end_data_update()? */
1568 inode->i_size = new_isize;
1569 invalid |= NFS_INO_INVALID_DATA;
1571 invalid |= NFS_INO_INVALID_ATTR;
1572 } else if (new_isize > cur_isize) {
1573 inode->i_size = new_isize;
1574 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1576 nfsi->cache_change_attribute = jiffies;
1577 dprintk("NFS: isize change on server for file %s/%ld\n",
1578 inode->i_sb->s_id, inode->i_ino);
1581 /* Check if the mtime agrees */
1582 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1583 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1584 dprintk("NFS: mtime change on server for file %s/%ld\n",
1585 inode->i_sb->s_id, inode->i_ino);
1586 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1587 nfsi->cache_change_attribute = jiffies;
1590 /* If ctime has changed we should definitely clear access+acl caches */
1591 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1592 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1593 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1594 nfsi->cache_change_attribute = jiffies;
1596 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1598 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1599 inode->i_uid != fattr->uid ||
1600 inode->i_gid != fattr->gid)
1601 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1603 inode->i_mode = fattr->mode;
1604 inode->i_nlink = fattr->nlink;
1605 inode->i_uid = fattr->uid;
1606 inode->i_gid = fattr->gid;
1608 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1610 * report the blocks in 512byte units
1612 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1613 inode->i_blksize = inode->i_sb->s_blocksize;
1615 inode->i_blocks = fattr->du.nfs2.blocks;
1616 inode->i_blksize = fattr->du.nfs2.blocksize;
1619 if ((fattr->valid & NFS_ATTR_FATTR_V4)) {
1620 if (nfsi->change_attr != fattr->change_attr) {
1621 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1622 inode->i_sb->s_id, inode->i_ino);
1623 nfsi->change_attr = fattr->change_attr;
1624 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1625 nfsi->cache_change_attribute = jiffies;
1627 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA);
1630 /* Update attrtimeo value if we're out of the unstable period */
1631 if (invalid & NFS_INO_INVALID_ATTR) {
1632 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1633 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1634 nfsi->attrtimeo_timestamp = jiffies;
1635 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1636 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1637 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1638 nfsi->attrtimeo_timestamp = jiffies;
1640 /* Don't invalidate the data if we were to blame */
1641 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1642 || S_ISLNK(inode->i_mode)))
1643 invalid &= ~NFS_INO_INVALID_DATA;
1645 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1646 if (!nfs_have_delegation(inode, FMODE_READ))
1647 nfsi->cache_validity |= invalid;
1652 * Big trouble! The inode has become a different object.
1655 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1656 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1660 * No need to worry about unhashing the dentry, as the
1661 * lookup validation will know that the inode is bad.
1662 * (But we fall through to invalidate the caches.)
1664 nfs_invalidate_inode(inode);
1668 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1669 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1670 NFS_SERVER(inode)->hostname, inode->i_sb->s_id,
1671 (long long)nfsi->fileid, (long long)fattr->fileid);
1676 * File system information
1679 static int nfs_set_super(struct super_block *s, void *data)
1681 s->s_fs_info = data;
1682 return set_anon_super(s, data);
1685 static int nfs_compare_super(struct super_block *sb, void *data)
1687 struct nfs_server *server = data;
1688 struct nfs_server *old = NFS_SB(sb);
1690 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1692 if (old->addr.sin_port != server->addr.sin_port)
1694 return !nfs_compare_fh(&old->fh, &server->fh);
1697 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1698 int flags, const char *dev_name, void *raw_data)
1701 struct nfs_server *server = NULL;
1702 struct super_block *s;
1703 struct nfs_fh *root;
1704 struct nfs_mount_data *data = raw_data;
1706 s = ERR_PTR(-EINVAL);
1708 dprintk("%s: missing data argument\n", __FUNCTION__);
1711 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1712 dprintk("%s: bad mount version\n", __FUNCTION__);
1715 switch (data->version) {
1721 if (data->flags & NFS_MOUNT_VER3) {
1722 dprintk("%s: mount structure version %d does not support NFSv3\n",
1727 data->root.size = NFS2_FHSIZE;
1728 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1730 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1731 dprintk("%s: mount structure version %d does not support strong security\n",
1737 memset(data->context, 0, sizeof(data->context));
1739 #ifndef CONFIG_NFS_V3
1740 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1741 s = ERR_PTR(-EPROTONOSUPPORT);
1742 if (data->flags & NFS_MOUNT_VER3) {
1743 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1746 #endif /* CONFIG_NFS_V3 */
1748 s = ERR_PTR(-ENOMEM);
1749 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
1752 /* Zero out the NFS state stuff */
1753 init_nfsv4_state(server);
1754 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1757 if (data->flags & NFS_MOUNT_VER3)
1758 root->size = data->root.size;
1760 root->size = NFS2_FHSIZE;
1761 s = ERR_PTR(-EINVAL);
1762 if (root->size > sizeof(root->data)) {
1763 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1766 memcpy(root->data, data->root.data, root->size);
1768 /* We now require that the mount process passes the remote address */
1769 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1770 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1771 dprintk("%s: mount program didn't pass remote address!\n",
1776 /* Fire up rpciod if not yet running */
1777 s = ERR_PTR(rpciod_up());
1779 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1780 __FUNCTION__, PTR_ERR(s));
1784 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1785 if (IS_ERR(s) || s->s_root)
1786 goto out_rpciod_down;
1790 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1792 up_write(&s->s_umount);
1793 deactivate_super(s);
1794 return ERR_PTR(error);
1796 s->s_flags |= MS_ACTIVE;
1805 static void nfs_kill_super(struct super_block *s)
1807 struct nfs_server *server = NFS_SB(s);
1811 if (!IS_ERR(server->client))
1812 rpc_shutdown_client(server->client);
1813 if (!IS_ERR(server->client_sys))
1814 rpc_shutdown_client(server->client_sys);
1815 if (!IS_ERR(server->client_acl))
1816 rpc_shutdown_client(server->client_acl);
1818 if (!(server->flags & NFS_MOUNT_NONLM))
1819 lockd_down(); /* release rpc.lockd */
1821 rpciod_down(); /* release rpciod */
1823 nfs_free_iostats(server->io_stats);
1824 kfree(server->hostname);
1828 static struct file_system_type nfs_fs_type = {
1829 .owner = THIS_MODULE,
1831 .get_sb = nfs_get_sb,
1832 .kill_sb = nfs_kill_super,
1833 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1836 #ifdef CONFIG_NFS_V4
1838 static void nfs4_clear_inode(struct inode *);
1841 static struct super_operations nfs4_sops = {
1842 .alloc_inode = nfs_alloc_inode,
1843 .destroy_inode = nfs_destroy_inode,
1844 .write_inode = nfs_write_inode,
1845 .delete_inode = nfs_delete_inode,
1846 .statfs = nfs_statfs,
1847 .clear_inode = nfs4_clear_inode,
1848 .umount_begin = nfs_umount_begin,
1849 .show_options = nfs_show_options,
1850 .show_stats = nfs_show_stats,
1854 * Clean out any remaining NFSv4 state that might be left over due
1855 * to open() calls that passed nfs_atomic_lookup, but failed to call
1858 static void nfs4_clear_inode(struct inode *inode)
1860 struct nfs_inode *nfsi = NFS_I(inode);
1862 /* If we are holding a delegation, return it! */
1863 nfs_inode_return_delegation(inode);
1864 /* First call standard NFS clear_inode() code */
1865 nfs_clear_inode(inode);
1866 /* Now clear out any remaining state */
1867 while (!list_empty(&nfsi->open_states)) {
1868 struct nfs4_state *state;
1870 state = list_entry(nfsi->open_states.next,
1873 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1876 (long long)NFS_FILEID(inode),
1878 BUG_ON(atomic_read(&state->count) != 1);
1879 nfs4_close_state(state, state->state);
1884 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1886 struct nfs_server *server;
1887 struct nfs4_client *clp = NULL;
1888 struct rpc_xprt *xprt = NULL;
1889 struct rpc_clnt *clnt = NULL;
1890 struct rpc_timeout timeparms;
1891 rpc_authflavor_t authflavour;
1894 sb->s_blocksize_bits = 0;
1895 sb->s_blocksize = 0;
1896 server = NFS_SB(sb);
1897 if (data->rsize != 0)
1898 server->rsize = nfs_block_size(data->rsize, NULL);
1899 if (data->wsize != 0)
1900 server->wsize = nfs_block_size(data->wsize, NULL);
1901 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1902 server->caps = NFS_CAP_ATOMIC_OPEN;
1904 server->acregmin = data->acregmin*HZ;
1905 server->acregmax = data->acregmax*HZ;
1906 server->acdirmin = data->acdirmin*HZ;
1907 server->acdirmax = data->acdirmax*HZ;
1909 server->rpc_ops = &nfs_v4_clientops;
1911 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
1913 server->retrans_timeo = timeparms.to_initval;
1914 server->retrans_count = timeparms.to_retries;
1916 clp = nfs4_get_client(&server->addr.sin_addr);
1918 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1922 /* Now create transport and client */
1923 authflavour = RPC_AUTH_UNIX;
1924 if (data->auth_flavourlen != 0) {
1925 if (data->auth_flavourlen != 1) {
1926 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1927 __FUNCTION__, data->auth_flavourlen);
1931 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1937 down_write(&clp->cl_sem);
1938 if (IS_ERR(clp->cl_rpcclient)) {
1939 xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
1941 up_write(&clp->cl_sem);
1942 err = PTR_ERR(xprt);
1943 dprintk("%s: cannot create RPC transport. Error = %d\n",
1947 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1948 server->rpc_ops->version, authflavour);
1950 up_write(&clp->cl_sem);
1951 err = PTR_ERR(clnt);
1952 dprintk("%s: cannot create RPC client. Error = %d\n",
1957 clnt->cl_softrtry = 1;
1958 clp->cl_rpcclient = clnt;
1959 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1962 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1963 clnt = rpc_clone_client(clp->cl_rpcclient);
1965 server->nfs4_state = clp;
1966 up_write(&clp->cl_sem);
1970 err = PTR_ERR(clnt);
1971 dprintk("%s: cannot create RPC client. Error = %d\n",
1976 server->client = clnt;
1978 if (server->nfs4_state->cl_idmap == NULL) {
1979 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1983 if (clnt->cl_auth->au_flavor != authflavour) {
1984 struct rpc_auth *auth;
1986 auth = rpcauth_create(authflavour, clnt);
1988 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1989 return PTR_ERR(auth);
1993 sb->s_time_gran = 1;
1995 sb->s_op = &nfs4_sops;
1996 err = nfs_sb_init(sb, authflavour);
2001 nfs4_put_client(clp);
2005 static int nfs4_compare_super(struct super_block *sb, void *data)
2007 struct nfs_server *server = data;
2008 struct nfs_server *old = NFS_SB(sb);
2010 if (strcmp(server->hostname, old->hostname) != 0)
2012 if (strcmp(server->mnt_path, old->mnt_path) != 0)
2018 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
2023 return ERR_PTR(-EINVAL);
2024 if (src->len < maxlen)
2027 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
2029 return ERR_PTR(-ENOMEM);
2031 if (copy_from_user(dst, src->data, maxlen)) {
2033 return ERR_PTR(-EFAULT);
2039 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
2040 int flags, const char *dev_name, void *raw_data)
2043 struct nfs_server *server;
2044 struct super_block *s;
2045 struct nfs4_mount_data *data = raw_data;
2049 dprintk("%s: missing data argument\n", __FUNCTION__);
2050 return ERR_PTR(-EINVAL);
2052 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
2053 dprintk("%s: bad mount version\n", __FUNCTION__);
2054 return ERR_PTR(-EINVAL);
2057 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
2059 return ERR_PTR(-ENOMEM);
2060 /* Zero out the NFS state stuff */
2061 init_nfsv4_state(server);
2062 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
2064 p = nfs_copy_user_string(NULL, &data->hostname, 256);
2067 server->hostname = p;
2069 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
2072 server->mnt_path = p;
2074 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
2075 sizeof(server->ip_addr) - 1);
2079 /* We now require that the mount process passes the remote address */
2080 if (data->host_addrlen != sizeof(server->addr)) {
2081 s = ERR_PTR(-EINVAL);
2084 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
2085 s = ERR_PTR(-EFAULT);
2088 if (server->addr.sin_family != AF_INET ||
2089 server->addr.sin_addr.s_addr == INADDR_ANY) {
2090 dprintk("%s: mount program didn't pass remote IP address!\n",
2092 s = ERR_PTR(-EINVAL);
2096 /* Fire up rpciod if not yet running */
2097 s = ERR_PTR(rpciod_up());
2099 dprintk("%s: couldn't start rpciod! Error = %ld\n",
2100 __FUNCTION__, PTR_ERR(s));
2104 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
2106 if (IS_ERR(s) || s->s_root)
2111 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
2113 up_write(&s->s_umount);
2114 deactivate_super(s);
2115 return ERR_PTR(error);
2117 s->s_flags |= MS_ACTIVE;
2120 s = (struct super_block *)p;
2122 kfree(server->mnt_path);
2123 kfree(server->hostname);
2128 static void nfs4_kill_super(struct super_block *sb)
2130 struct nfs_server *server = NFS_SB(sb);
2132 nfs_return_all_delegations(sb);
2133 kill_anon_super(sb);
2135 nfs4_renewd_prepare_shutdown(server);
2137 if (server->client != NULL && !IS_ERR(server->client))
2138 rpc_shutdown_client(server->client);
2140 destroy_nfsv4_state(server);
2144 nfs_free_iostats(server->io_stats);
2145 kfree(server->hostname);
2149 static struct file_system_type nfs4_fs_type = {
2150 .owner = THIS_MODULE,
2152 .get_sb = nfs4_get_sb,
2153 .kill_sb = nfs4_kill_super,
2154 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2157 static const int nfs_set_port_min = 0;
2158 static const int nfs_set_port_max = 65535;
2159 static int param_set_port(const char *val, struct kernel_param *kp)
2162 int num = simple_strtol(val, &endp, 0);
2163 if (endp == val || *endp || num < nfs_set_port_min || num > nfs_set_port_max)
2165 *((int *)kp->arg) = num;
2169 module_param_call(callback_tcpport, param_set_port, param_get_int,
2170 &nfs_callback_set_tcpport, 0644);
2172 static int param_set_idmap_timeout(const char *val, struct kernel_param *kp)
2175 int num = simple_strtol(val, &endp, 0);
2177 if (endp == val || *endp || num < 0 || jif < num)
2179 *((int *)kp->arg) = jif;
2183 module_param_call(idmap_cache_timeout, param_set_idmap_timeout, param_get_int,
2184 &nfs_idmap_cache_timeout, 0644);
2186 #define nfs4_init_once(nfsi) \
2188 INIT_LIST_HEAD(&(nfsi)->open_states); \
2189 nfsi->delegation = NULL; \
2190 nfsi->delegation_state = 0; \
2191 init_rwsem(&nfsi->rwsem); \
2194 static inline int register_nfs4fs(void)
2198 ret = nfs_register_sysctl();
2201 ret = register_filesystem(&nfs4_fs_type);
2203 nfs_unregister_sysctl();
2207 static inline void unregister_nfs4fs(void)
2209 unregister_filesystem(&nfs4_fs_type);
2210 nfs_unregister_sysctl();
2213 #define nfs4_init_once(nfsi) \
2215 #define register_nfs4fs() (0)
2216 #define unregister_nfs4fs()
2219 extern int nfs_init_nfspagecache(void);
2220 extern void nfs_destroy_nfspagecache(void);
2221 extern int nfs_init_readpagecache(void);
2222 extern void nfs_destroy_readpagecache(void);
2223 extern int nfs_init_writepagecache(void);
2224 extern void nfs_destroy_writepagecache(void);
2225 #ifdef CONFIG_NFS_DIRECTIO
2226 extern int nfs_init_directcache(void);
2227 extern void nfs_destroy_directcache(void);
2230 static kmem_cache_t * nfs_inode_cachep;
2232 static struct inode *nfs_alloc_inode(struct super_block *sb)
2234 struct nfs_inode *nfsi;
2235 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2239 nfsi->cache_validity = 0UL;
2240 nfsi->cache_change_attribute = jiffies;
2241 #ifdef CONFIG_NFS_V3_ACL
2242 nfsi->acl_access = ERR_PTR(-EAGAIN);
2243 nfsi->acl_default = ERR_PTR(-EAGAIN);
2245 #ifdef CONFIG_NFS_V4
2246 nfsi->nfs4_acl = NULL;
2247 #endif /* CONFIG_NFS_V4 */
2248 return &nfsi->vfs_inode;
2251 static void nfs_destroy_inode(struct inode *inode)
2253 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2256 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2258 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2260 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2261 SLAB_CTOR_CONSTRUCTOR) {
2262 inode_init_once(&nfsi->vfs_inode);
2263 spin_lock_init(&nfsi->req_lock);
2264 INIT_LIST_HEAD(&nfsi->dirty);
2265 INIT_LIST_HEAD(&nfsi->commit);
2266 INIT_LIST_HEAD(&nfsi->open_files);
2267 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2268 atomic_set(&nfsi->data_updates, 0);
2272 nfs4_init_once(nfsi);
2276 static int nfs_init_inodecache(void)
2278 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2279 sizeof(struct nfs_inode),
2280 0, SLAB_RECLAIM_ACCOUNT,
2282 if (nfs_inode_cachep == NULL)
2288 static void nfs_destroy_inodecache(void)
2290 if (kmem_cache_destroy(nfs_inode_cachep))
2291 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2297 static int __init init_nfs_fs(void)
2301 err = nfs_init_nfspagecache();
2305 err = nfs_init_inodecache();
2309 err = nfs_init_readpagecache();
2313 err = nfs_init_writepagecache();
2317 #ifdef CONFIG_NFS_DIRECTIO
2318 err = nfs_init_directcache();
2323 #ifdef CONFIG_PROC_FS
2324 rpc_proc_register(&nfs_rpcstat);
2326 err = register_filesystem(&nfs_fs_type);
2329 if ((err = register_nfs4fs()) != 0)
2333 #ifdef CONFIG_PROC_FS
2334 rpc_proc_unregister("nfs");
2336 #ifdef CONFIG_NFS_DIRECTIO
2337 nfs_destroy_directcache();
2340 nfs_destroy_writepagecache();
2342 nfs_destroy_readpagecache();
2344 nfs_destroy_inodecache();
2346 nfs_destroy_nfspagecache();
2351 static void __exit exit_nfs_fs(void)
2353 #ifdef CONFIG_NFS_DIRECTIO
2354 nfs_destroy_directcache();
2356 nfs_destroy_writepagecache();
2357 nfs_destroy_readpagecache();
2358 nfs_destroy_inodecache();
2359 nfs_destroy_nfspagecache();
2360 #ifdef CONFIG_PROC_FS
2361 rpc_proc_unregister("nfs");
2363 unregister_filesystem(&nfs_fs_type);
2364 unregister_nfs4fs();
2367 /* Not quite true; I just maintain it */
2368 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2369 MODULE_LICENSE("GPL");
2371 module_init(init_nfs_fs)
2372 module_exit(exit_nfs_fs)