4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/config.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
39 #include <asm/system.h>
40 #include <asm/uaccess.h>
44 #include "delegation.h"
47 #define NFSDBG_FACILITY NFSDBG_VFS
48 #define NFS_PARANOIA 1
50 /* Maximum number of readahead requests
51 * FIXME: this should really be a sysctl so that users may tune it to suit
52 * their needs. People that do NFS over a slow network, might for
53 * instance want to reduce it to something closer to 1 for improved
54 * interactive response.
56 #define NFS_MAX_READAHEAD (RPC_DEF_SLOT_TABLE - 1)
58 static void nfs_invalidate_inode(struct inode *);
59 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
61 static struct inode *nfs_alloc_inode(struct super_block *sb);
62 static void nfs_destroy_inode(struct inode *);
63 static int nfs_write_inode(struct inode *,int);
64 static void nfs_delete_inode(struct inode *);
65 static void nfs_clear_inode(struct inode *);
66 static void nfs_umount_begin(struct super_block *);
67 static int nfs_statfs(struct super_block *, struct kstatfs *);
68 static int nfs_show_options(struct seq_file *, struct vfsmount *);
69 static int nfs_show_stats(struct seq_file *, struct vfsmount *);
70 static void nfs_zap_acl_cache(struct inode *);
72 static struct rpc_program nfs_program;
74 static struct super_operations nfs_sops = {
75 .alloc_inode = nfs_alloc_inode,
76 .destroy_inode = nfs_destroy_inode,
77 .write_inode = nfs_write_inode,
78 .delete_inode = nfs_delete_inode,
80 .clear_inode = nfs_clear_inode,
81 .umount_begin = nfs_umount_begin,
82 .show_options = nfs_show_options,
83 .show_stats = nfs_show_stats,
89 static struct rpc_stat nfs_rpcstat = {
90 .program = &nfs_program
92 static struct rpc_version * nfs_version[] = {
96 #if defined(CONFIG_NFS_V3)
98 #elif defined(CONFIG_NFS_V4)
101 #if defined(CONFIG_NFS_V4)
106 static struct rpc_program nfs_program = {
108 .number = NFS_PROGRAM,
109 .nrvers = sizeof(nfs_version) / sizeof(nfs_version[0]),
110 .version = nfs_version,
111 .stats = &nfs_rpcstat,
112 .pipe_dir_name = "/nfs",
115 #ifdef CONFIG_NFS_V3_ACL
116 static struct rpc_stat nfsacl_rpcstat = { &nfsacl_program };
117 static struct rpc_version * nfsacl_version[] = {
118 [3] = &nfsacl_version3,
121 struct rpc_program nfsacl_program = {
123 .number = NFS_ACL_PROGRAM,
124 .nrvers = sizeof(nfsacl_version) / sizeof(nfsacl_version[0]),
125 .version = nfsacl_version,
126 .stats = &nfsacl_rpcstat,
128 #endif /* CONFIG_NFS_V3_ACL */
130 static inline unsigned long
131 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
133 return nfs_fileid_to_ino_t(fattr->fileid);
137 nfs_write_inode(struct inode *inode, int sync)
139 int flags = sync ? FLUSH_WAIT : 0;
142 ret = nfs_commit_inode(inode, flags);
149 nfs_delete_inode(struct inode * inode)
151 dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
153 truncate_inode_pages(&inode->i_data, 0);
157 * The following should never happen...
159 if (nfs_have_writebacks(inode)) {
160 printk(KERN_ERR "nfs_delete_inode: inode %ld has pending RPC requests\n", inode->i_ino);
167 nfs_clear_inode(struct inode *inode)
169 struct nfs_inode *nfsi = NFS_I(inode);
170 struct rpc_cred *cred;
173 BUG_ON (!list_empty(&nfsi->open_files));
174 nfs_zap_acl_cache(inode);
175 cred = nfsi->cache_access.cred;
178 BUG_ON(atomic_read(&nfsi->data_updates) != 0);
182 nfs_umount_begin(struct super_block *sb)
184 struct rpc_clnt *rpc = NFS_SB(sb)->client;
186 /* -EIO all pending I/O */
188 rpc_killall_tasks(rpc);
189 rpc = NFS_SB(sb)->client_acl;
191 rpc_killall_tasks(rpc);
195 static inline unsigned long
196 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
198 /* make sure blocksize is a power of two */
199 if ((bsize & (bsize - 1)) || nrbitsp) {
200 unsigned char nrbits;
202 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
213 * Calculate the number of 512byte blocks used.
215 static inline unsigned long
216 nfs_calc_block_size(u64 tsize)
218 loff_t used = (tsize + 511) >> 9;
219 return (used > ULONG_MAX) ? ULONG_MAX : used;
223 * Compute and set NFS server blocksize
225 static inline unsigned long
226 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
228 if (bsize < NFS_MIN_FILE_IO_SIZE)
229 bsize = NFS_DEF_FILE_IO_SIZE;
230 else if (bsize >= NFS_MAX_FILE_IO_SIZE)
231 bsize = NFS_MAX_FILE_IO_SIZE;
233 return nfs_block_bits(bsize, nrbitsp);
237 * Obtain the root inode of the file system.
239 static struct inode *
240 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
242 struct nfs_server *server = NFS_SB(sb);
246 error = server->rpc_ops->getroot(server, rootfh, fsinfo);
248 dprintk("nfs_get_root: getattr error = %d\n", -error);
249 return ERR_PTR(error);
252 rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
254 return ERR_PTR(-ENOMEM);
259 * Do NFS version-independent mount processing, and sanity checking
262 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
264 struct nfs_server *server;
265 struct inode *root_inode;
266 struct nfs_fattr fattr;
267 struct nfs_fsinfo fsinfo = {
270 struct nfs_pathconf pathinfo = {
273 int no_root_error = 0;
274 unsigned long max_rpc_payload;
276 /* We probably want something more informative here */
277 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
281 sb->s_magic = NFS_SUPER_MAGIC;
283 root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
284 /* Did getting the root inode fail? */
285 if (IS_ERR(root_inode)) {
286 no_root_error = PTR_ERR(root_inode);
289 sb->s_root = d_alloc_root(root_inode);
291 no_root_error = -ENOMEM;
294 sb->s_root->d_op = server->rpc_ops->dentry_ops;
296 server->io_stats = nfs_alloc_iostats();
297 if (!server->io_stats) {
298 no_root_error = -ENOMEM;
302 /* mount time stamp, in seconds */
303 server->mount_time = jiffies;
305 /* Get some general file system info */
306 if (server->namelen == 0 &&
307 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
308 server->namelen = pathinfo.max_namelen;
309 /* Work out a lot of parameters */
310 if (server->rsize == 0)
311 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
312 if (server->wsize == 0)
313 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
315 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
316 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
317 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
318 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
320 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
321 if (server->rsize > max_rpc_payload)
322 server->rsize = max_rpc_payload;
323 if (server->rsize > NFS_MAX_FILE_IO_SIZE)
324 server->rsize = NFS_MAX_FILE_IO_SIZE;
325 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
327 if (server->wsize > max_rpc_payload)
328 server->wsize = max_rpc_payload;
329 if (server->wsize > NFS_MAX_FILE_IO_SIZE)
330 server->wsize = NFS_MAX_FILE_IO_SIZE;
331 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
333 if (sb->s_blocksize == 0)
334 sb->s_blocksize = nfs_block_bits(server->wsize,
335 &sb->s_blocksize_bits);
336 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
338 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
339 if (server->dtsize > PAGE_CACHE_SIZE)
340 server->dtsize = PAGE_CACHE_SIZE;
341 if (server->dtsize > server->rsize)
342 server->dtsize = server->rsize;
344 if (server->flags & NFS_MOUNT_NOAC) {
345 server->acregmin = server->acregmax = 0;
346 server->acdirmin = server->acdirmax = 0;
347 sb->s_flags |= MS_SYNCHRONOUS;
349 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
351 sb->s_maxbytes = fsinfo.maxfilesize;
352 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
353 sb->s_maxbytes = MAX_LFS_FILESIZE;
355 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
356 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
358 /* We're airborne Set socket buffersize */
359 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
361 /* Yargs. It didn't work out. */
363 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
364 if (!IS_ERR(root_inode))
366 return no_root_error;
369 static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
371 to->to_initval = timeo * HZ / 10;
372 to->to_retries = retrans;
379 to->to_initval = 60 * HZ;
380 if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
381 to->to_initval = NFS_MAX_TCP_TIMEOUT;
382 to->to_increment = to->to_initval;
383 to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
384 to->to_exponential = 0;
389 to->to_initval = 11 * HZ / 10;
390 if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
391 to->to_initval = NFS_MAX_UDP_TIMEOUT;
392 to->to_maxval = NFS_MAX_UDP_TIMEOUT;
393 to->to_exponential = 1;
399 * Create an RPC client handle.
401 static struct rpc_clnt *
402 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
404 struct rpc_timeout timeparms;
405 struct rpc_xprt *xprt = NULL;
406 struct rpc_clnt *clnt = NULL;
407 int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
409 nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
411 server->retrans_timeo = timeparms.to_initval;
412 server->retrans_count = timeparms.to_retries;
414 /* create transport and client */
415 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
417 dprintk("%s: cannot create RPC transport. Error = %ld\n",
418 __FUNCTION__, PTR_ERR(xprt));
419 return (struct rpc_clnt *)xprt;
421 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
422 server->rpc_ops->version, data->pseudoflavor);
424 dprintk("%s: cannot create RPC client. Error = %ld\n",
425 __FUNCTION__, PTR_ERR(xprt));
430 clnt->cl_softrtry = 1;
439 * The way this works is that the mount process passes a structure
440 * in the data argument which contains the server's IP address
441 * and the root file handle obtained from the server's mount
442 * daemon. We stash these away in the private superblock fields.
445 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
447 struct nfs_server *server;
448 rpc_authflavor_t authflavor;
451 sb->s_blocksize_bits = 0;
454 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
456 server->rsize = nfs_block_size(data->rsize, NULL);
458 server->wsize = nfs_block_size(data->wsize, NULL);
459 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
461 server->acregmin = data->acregmin*HZ;
462 server->acregmax = data->acregmax*HZ;
463 server->acdirmin = data->acdirmin*HZ;
464 server->acdirmax = data->acdirmax*HZ;
466 /* Start lockd here, before we might error out */
467 if (!(server->flags & NFS_MOUNT_NONLM))
470 server->namelen = data->namlen;
471 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
472 if (!server->hostname)
474 strcpy(server->hostname, data->hostname);
476 /* Check NFS protocol revision and initialize RPC op vector
477 * and file handle pool. */
479 if (server->flags & NFS_MOUNT_VER3) {
480 server->rpc_ops = &nfs_v3_clientops;
481 server->caps |= NFS_CAP_READDIRPLUS;
483 server->rpc_ops = &nfs_v2_clientops;
486 server->rpc_ops = &nfs_v2_clientops;
489 /* Fill in pseudoflavor for mount version < 5 */
490 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
491 data->pseudoflavor = RPC_AUTH_UNIX;
492 authflavor = data->pseudoflavor; /* save for sb_init() */
493 /* XXX maybe we want to add a server->pseudoflavor field */
495 /* Create RPC client handles */
496 server->client = nfs_create_client(server, data);
497 if (IS_ERR(server->client))
498 return PTR_ERR(server->client);
499 /* RFC 2623, sec 2.3.2 */
500 if (authflavor != RPC_AUTH_UNIX) {
501 struct rpc_auth *auth;
503 server->client_sys = rpc_clone_client(server->client);
504 if (IS_ERR(server->client_sys))
505 return PTR_ERR(server->client_sys);
506 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
508 return PTR_ERR(auth);
510 atomic_inc(&server->client->cl_count);
511 server->client_sys = server->client;
513 if (server->flags & NFS_MOUNT_VER3) {
514 #ifdef CONFIG_NFS_V3_ACL
515 if (!(server->flags & NFS_MOUNT_NOACL)) {
516 server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
517 /* No errors! Assume that Sun nfsacls are supported */
518 if (!IS_ERR(server->client_acl))
519 server->caps |= NFS_CAP_ACLS;
522 server->flags &= ~NFS_MOUNT_NOACL;
523 #endif /* CONFIG_NFS_V3_ACL */
525 * The VFS shouldn't apply the umask to mode bits. We will
526 * do so ourselves when necessary.
528 sb->s_flags |= MS_POSIXACL;
529 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
530 server->namelen = NFS3_MAXNAMLEN;
533 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
534 server->namelen = NFS2_MAXNAMLEN;
537 sb->s_op = &nfs_sops;
538 return nfs_sb_init(sb, authflavor);
542 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
544 struct nfs_server *server = NFS_SB(sb);
545 unsigned char blockbits;
546 unsigned long blockres;
547 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
548 struct nfs_fattr fattr;
549 struct nfs_fsstat res = {
556 error = server->rpc_ops->statfs(server, rootfh, &res);
557 buf->f_type = NFS_SUPER_MAGIC;
562 * Current versions of glibc do not correctly handle the
563 * case where f_frsize != f_bsize. Eventually we want to
564 * report the value of wtmult in this field.
566 buf->f_frsize = sb->s_blocksize;
569 * On most *nix systems, f_blocks, f_bfree, and f_bavail
570 * are reported in units of f_frsize. Linux hasn't had
571 * an f_frsize field in its statfs struct until recently,
572 * thus historically Linux's sys_statfs reports these
573 * fields in units of f_bsize.
575 buf->f_bsize = sb->s_blocksize;
576 blockbits = sb->s_blocksize_bits;
577 blockres = (1 << blockbits) - 1;
578 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
579 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
580 buf->f_bavail = (res.abytes + blockres) >> blockbits;
582 buf->f_files = res.tfiles;
583 buf->f_ffree = res.afiles;
585 buf->f_namelen = server->namelen;
591 dprintk("%s: statfs error = %d\n", __FUNCTION__, -error);
592 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
597 static void nfs_show_mount_options(struct seq_file *m, struct nfs_server *nfss, int showdefaults)
599 static struct proc_nfs_info {
604 { NFS_MOUNT_SOFT, ",soft", ",hard" },
605 { NFS_MOUNT_INTR, ",intr", "" },
606 { NFS_MOUNT_NOCTO, ",nocto", "" },
607 { NFS_MOUNT_NOAC, ",noac", "" },
608 { NFS_MOUNT_NONLM, ",nolock", "" },
609 { NFS_MOUNT_NOACL, ",noacl", "" },
612 struct proc_nfs_info *nfs_infop;
616 seq_printf(m, ",vers=%d", nfss->rpc_ops->version);
617 seq_printf(m, ",rsize=%d", nfss->rsize);
618 seq_printf(m, ",wsize=%d", nfss->wsize);
619 if (nfss->acregmin != 3*HZ || showdefaults)
620 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
621 if (nfss->acregmax != 60*HZ || showdefaults)
622 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
623 if (nfss->acdirmin != 30*HZ || showdefaults)
624 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
625 if (nfss->acdirmax != 60*HZ || showdefaults)
626 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
627 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
628 if (nfss->flags & nfs_infop->flag)
629 seq_puts(m, nfs_infop->str);
631 seq_puts(m, nfs_infop->nostr);
633 switch (nfss->client->cl_xprt->prot) {
641 snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
644 seq_printf(m, ",proto=%s", proto);
645 seq_printf(m, ",timeo=%lu", 10U * nfss->retrans_timeo / HZ);
646 seq_printf(m, ",retrans=%u", nfss->retrans_count);
649 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
651 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
653 nfs_show_mount_options(m, nfss, 0);
655 seq_puts(m, ",addr=");
656 seq_escape(m, nfss->hostname, " \t\n\\");
661 static int nfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
664 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
665 struct rpc_auth *auth = nfss->client->cl_auth;
666 struct nfs_iostats totals = { };
668 seq_printf(m, "statvers=%s", NFS_IOSTAT_VERS);
671 * Display all mount option settings
673 seq_printf(m, "\n\topts:\t");
674 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? "ro" : "rw");
675 seq_puts(m, mnt->mnt_sb->s_flags & MS_SYNCHRONOUS ? ",sync" : "");
676 seq_puts(m, mnt->mnt_sb->s_flags & MS_NOATIME ? ",noatime" : "");
677 seq_puts(m, mnt->mnt_sb->s_flags & MS_NODIRATIME ? ",nodiratime" : "");
678 nfs_show_mount_options(m, nfss, 1);
680 seq_printf(m, "\n\tage:\t%lu", (jiffies - nfss->mount_time) / HZ);
682 seq_printf(m, "\n\tcaps:\t");
683 seq_printf(m, "caps=0x%x", nfss->caps);
684 seq_printf(m, ",wtmult=%d", nfss->wtmult);
685 seq_printf(m, ",dtsize=%d", nfss->dtsize);
686 seq_printf(m, ",bsize=%d", nfss->bsize);
687 seq_printf(m, ",namelen=%d", nfss->namelen);
690 if (nfss->rpc_ops->version == 4) {
691 seq_printf(m, "\n\tnfsv4:\t");
692 seq_printf(m, "bm0=0x%x", nfss->attr_bitmask[0]);
693 seq_printf(m, ",bm1=0x%x", nfss->attr_bitmask[1]);
694 seq_printf(m, ",acl=0x%x", nfss->acl_bitmask);
699 * Display security flavor in effect for this mount
701 seq_printf(m, "\n\tsec:\tflavor=%d", auth->au_ops->au_flavor);
703 seq_printf(m, ",pseudoflavor=%d", auth->au_flavor);
706 * Display superblock I/O counters
708 for (cpu = 0; cpu < NR_CPUS; cpu++) {
709 struct nfs_iostats *stats;
711 if (!cpu_possible(cpu))
715 stats = per_cpu_ptr(nfss->io_stats, cpu);
717 for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
718 totals.events[i] += stats->events[i];
719 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
720 totals.bytes[i] += stats->bytes[i];
725 seq_printf(m, "\n\tevents:\t");
726 for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
727 seq_printf(m, "%lu ", totals.events[i]);
728 seq_printf(m, "\n\tbytes:\t");
729 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
730 seq_printf(m, "%Lu ", totals.bytes[i]);
736 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
738 int nfs_sync_mapping(struct address_space *mapping)
742 if (mapping->nrpages == 0)
744 unmap_mapping_range(mapping, 0, 0, 0);
745 ret = filemap_write_and_wait(mapping);
748 ret = nfs_wb_all(mapping->host);
754 * Invalidate the local caches
756 static void nfs_zap_caches_locked(struct inode *inode)
758 struct nfs_inode *nfsi = NFS_I(inode);
759 int mode = inode->i_mode;
761 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
763 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
764 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
766 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
767 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
768 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
770 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
773 void nfs_zap_caches(struct inode *inode)
775 spin_lock(&inode->i_lock);
776 nfs_zap_caches_locked(inode);
777 spin_unlock(&inode->i_lock);
780 static void nfs_zap_acl_cache(struct inode *inode)
782 void (*clear_acl_cache)(struct inode *);
784 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
785 if (clear_acl_cache != NULL)
786 clear_acl_cache(inode);
787 spin_lock(&inode->i_lock);
788 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
789 spin_unlock(&inode->i_lock);
793 * Invalidate, but do not unhash, the inode.
794 * NB: must be called with inode->i_lock held!
796 static void nfs_invalidate_inode(struct inode *inode)
798 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
799 nfs_zap_caches_locked(inode);
802 struct nfs_find_desc {
804 struct nfs_fattr *fattr;
808 * In NFSv3 we can have 64bit inode numbers. In order to support
809 * this, and re-exported directories (also seen in NFSv2)
810 * we are forced to allow 2 different inodes to have the same
814 nfs_find_actor(struct inode *inode, void *opaque)
816 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
817 struct nfs_fh *fh = desc->fh;
818 struct nfs_fattr *fattr = desc->fattr;
820 if (NFS_FILEID(inode) != fattr->fileid)
822 if (nfs_compare_fh(NFS_FH(inode), fh))
824 if (is_bad_inode(inode) || NFS_STALE(inode))
830 nfs_init_locked(struct inode *inode, void *opaque)
832 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
833 struct nfs_fattr *fattr = desc->fattr;
835 NFS_FILEID(inode) = fattr->fileid;
836 nfs_copy_fh(NFS_FH(inode), desc->fh);
840 /* Don't use READDIRPLUS on directories that we believe are too large */
841 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
844 * This is our front-end to iget that looks up inodes by file handle
845 * instead of inode number.
848 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
850 struct nfs_find_desc desc = {
854 struct inode *inode = NULL;
857 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
861 printk("NFS: Buggy server - nlink == 0!\n");
865 hash = nfs_fattr_to_ino_t(fattr);
867 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
870 if (inode->i_state & I_NEW) {
871 struct nfs_inode *nfsi = NFS_I(inode);
873 /* We set i_ino for the few things that still rely on it,
877 /* We can't support update_atime(), since the server will reset it */
878 inode->i_flags |= S_NOATIME|S_NOCMTIME;
879 inode->i_mode = fattr->mode;
880 /* Why so? Because we want revalidate for devices/FIFOs, and
881 * that's precisely what we have in nfs_file_inode_operations.
883 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
884 if (S_ISREG(inode->i_mode)) {
885 inode->i_fop = &nfs_file_operations;
886 inode->i_data.a_ops = &nfs_file_aops;
887 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
888 } else if (S_ISDIR(inode->i_mode)) {
889 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
890 inode->i_fop = &nfs_dir_operations;
891 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
892 && fattr->size <= NFS_LIMIT_READDIRPLUS)
893 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
894 } else if (S_ISLNK(inode->i_mode))
895 inode->i_op = &nfs_symlink_inode_operations;
897 init_special_inode(inode, inode->i_mode, fattr->rdev);
899 nfsi->read_cache_jiffies = fattr->time_start;
900 nfsi->last_updated = jiffies;
901 inode->i_atime = fattr->atime;
902 inode->i_mtime = fattr->mtime;
903 inode->i_ctime = fattr->ctime;
904 if (fattr->valid & NFS_ATTR_FATTR_V4)
905 nfsi->change_attr = fattr->change_attr;
906 inode->i_size = nfs_size_to_loff_t(fattr->size);
907 inode->i_nlink = fattr->nlink;
908 inode->i_uid = fattr->uid;
909 inode->i_gid = fattr->gid;
910 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
912 * report the blocks in 512byte units
914 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
915 inode->i_blksize = inode->i_sb->s_blocksize;
917 inode->i_blocks = fattr->du.nfs2.blocks;
918 inode->i_blksize = fattr->du.nfs2.blocksize;
920 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
921 nfsi->attrtimeo_timestamp = jiffies;
922 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
923 nfsi->cache_access.cred = NULL;
925 unlock_new_inode(inode);
927 nfs_refresh_inode(inode, fattr);
928 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
930 (long long)NFS_FILEID(inode),
931 atomic_read(&inode->i_count));
937 printk("nfs_fhget: iget failed\n");
941 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
944 nfs_setattr(struct dentry *dentry, struct iattr *attr)
946 struct inode *inode = dentry->d_inode;
947 struct nfs_fattr fattr;
950 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
952 if (attr->ia_valid & ATTR_SIZE) {
953 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
954 attr->ia_valid &= ~ATTR_SIZE;
957 /* Optimization: if the end result is no change, don't RPC */
958 attr->ia_valid &= NFS_VALID_ATTRS;
959 if (attr->ia_valid == 0)
963 nfs_begin_data_update(inode);
964 /* Write all dirty data */
965 filemap_write_and_wait(inode->i_mapping);
968 * Return any delegations if we're going to change ACLs
970 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
971 nfs_inode_return_delegation(inode);
972 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
974 nfs_refresh_inode(inode, &fattr);
975 nfs_end_data_update(inode);
981 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
982 * @inode: pointer to struct inode
983 * @attr: pointer to struct iattr
985 * Note: we do this in the *proc.c in order to ensure that
986 * it works for things like exclusive creates too.
988 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
990 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
991 if ((attr->ia_valid & ATTR_MODE) != 0) {
992 int mode = attr->ia_mode & S_IALLUGO;
993 mode |= inode->i_mode & ~S_IALLUGO;
994 inode->i_mode = mode;
996 if ((attr->ia_valid & ATTR_UID) != 0)
997 inode->i_uid = attr->ia_uid;
998 if ((attr->ia_valid & ATTR_GID) != 0)
999 inode->i_gid = attr->ia_gid;
1000 spin_lock(&inode->i_lock);
1001 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1002 spin_unlock(&inode->i_lock);
1004 if ((attr->ia_valid & ATTR_SIZE) != 0) {
1005 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
1006 inode->i_size = attr->ia_size;
1007 vmtruncate(inode, attr->ia_size);
1011 static int nfs_wait_schedule(void *word)
1013 if (signal_pending(current))
1014 return -ERESTARTSYS;
1020 * Wait for the inode to get unlocked.
1022 static int nfs_wait_on_inode(struct inode *inode)
1024 struct rpc_clnt *clnt = NFS_CLIENT(inode);
1025 struct nfs_inode *nfsi = NFS_I(inode);
1029 rpc_clnt_sigmask(clnt, &oldmask);
1030 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
1031 nfs_wait_schedule, TASK_INTERRUPTIBLE);
1032 rpc_clnt_sigunmask(clnt, &oldmask);
1037 static void nfs_wake_up_inode(struct inode *inode)
1039 struct nfs_inode *nfsi = NFS_I(inode);
1041 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
1042 smp_mb__after_clear_bit();
1043 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
1046 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1048 struct inode *inode = dentry->d_inode;
1049 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
1052 /* Flush out writes to the server in order to update c/mtime */
1053 nfs_sync_inode(inode, 0, 0, FLUSH_WAIT|FLUSH_NOCOMMIT);
1056 * We may force a getattr if the user cares about atime.
1058 * Note that we only have to check the vfsmount flags here:
1059 * - NFS always sets S_NOATIME by so checking it would give a
1061 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
1062 * no point in checking those.
1064 if ((mnt->mnt_flags & MNT_NOATIME) ||
1065 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
1069 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1071 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1073 generic_fillattr(inode, stat);
1077 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
1079 struct nfs_open_context *ctx;
1081 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
1083 atomic_set(&ctx->count, 1);
1084 ctx->dentry = dget(dentry);
1085 ctx->vfsmnt = mntget(mnt);
1086 ctx->cred = get_rpccred(cred);
1088 ctx->lockowner = current->files;
1090 ctx->dir_cookie = 0;
1095 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1098 atomic_inc(&ctx->count);
1102 void put_nfs_open_context(struct nfs_open_context *ctx)
1104 if (atomic_dec_and_test(&ctx->count)) {
1105 if (!list_empty(&ctx->list)) {
1106 struct inode *inode = ctx->dentry->d_inode;
1107 spin_lock(&inode->i_lock);
1108 list_del(&ctx->list);
1109 spin_unlock(&inode->i_lock);
1111 if (ctx->state != NULL)
1112 nfs4_close_state(ctx->state, ctx->mode);
1113 if (ctx->cred != NULL)
1114 put_rpccred(ctx->cred);
1116 mntput(ctx->vfsmnt);
1122 * Ensure that mmap has a recent RPC credential for use when writing out
1125 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1127 struct inode *inode = filp->f_dentry->d_inode;
1128 struct nfs_inode *nfsi = NFS_I(inode);
1130 filp->private_data = get_nfs_open_context(ctx);
1131 spin_lock(&inode->i_lock);
1132 list_add(&ctx->list, &nfsi->open_files);
1133 spin_unlock(&inode->i_lock);
1137 * Given an inode, search for an open context with the desired characteristics
1139 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
1141 struct nfs_inode *nfsi = NFS_I(inode);
1142 struct nfs_open_context *pos, *ctx = NULL;
1144 spin_lock(&inode->i_lock);
1145 list_for_each_entry(pos, &nfsi->open_files, list) {
1146 if (cred != NULL && pos->cred != cred)
1148 if ((pos->mode & mode) == mode) {
1149 ctx = get_nfs_open_context(pos);
1153 spin_unlock(&inode->i_lock);
1157 static void nfs_file_clear_open_context(struct file *filp)
1159 struct inode *inode = filp->f_dentry->d_inode;
1160 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
1163 filp->private_data = NULL;
1164 spin_lock(&inode->i_lock);
1165 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1166 spin_unlock(&inode->i_lock);
1167 put_nfs_open_context(ctx);
1172 * These allocate and release file read/write context information.
1174 int nfs_open(struct inode *inode, struct file *filp)
1176 struct nfs_open_context *ctx;
1177 struct rpc_cred *cred;
1179 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1181 return PTR_ERR(cred);
1182 ctx = alloc_nfs_open_context(filp->f_vfsmnt, filp->f_dentry, cred);
1186 ctx->mode = filp->f_mode;
1187 nfs_file_set_open_context(filp, ctx);
1188 put_nfs_open_context(ctx);
1192 int nfs_release(struct inode *inode, struct file *filp)
1194 nfs_file_clear_open_context(filp);
1199 * This function is called whenever some part of NFS notices that
1200 * the cached attributes have to be refreshed.
1203 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1205 int status = -ESTALE;
1206 struct nfs_fattr fattr;
1207 struct nfs_inode *nfsi = NFS_I(inode);
1209 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1210 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1213 if (!inode || is_bad_inode(inode))
1215 if (NFS_STALE(inode))
1218 status = nfs_wait_on_inode(inode);
1221 if (NFS_STALE(inode)) {
1223 /* Do we trust the cached ESTALE? */
1224 if (NFS_ATTRTIMEO(inode) != 0) {
1225 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1232 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1234 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1236 (long long)NFS_FILEID(inode), status);
1237 if (status == -ESTALE) {
1238 nfs_zap_caches(inode);
1239 if (!S_ISDIR(inode->i_mode))
1240 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1245 spin_lock(&inode->i_lock);
1246 status = nfs_update_inode(inode, &fattr);
1248 spin_unlock(&inode->i_lock);
1249 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1251 (long long)NFS_FILEID(inode), status);
1254 spin_unlock(&inode->i_lock);
1256 nfs_revalidate_mapping(inode, inode->i_mapping);
1258 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1259 nfs_zap_acl_cache(inode);
1261 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1263 (long long)NFS_FILEID(inode));
1266 nfs_wake_up_inode(inode);
1273 int nfs_attribute_timeout(struct inode *inode)
1275 struct nfs_inode *nfsi = NFS_I(inode);
1277 if (nfs_have_delegation(inode, FMODE_READ))
1279 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1283 * nfs_revalidate_inode - Revalidate the inode attributes
1284 * @server - pointer to nfs_server struct
1285 * @inode - pointer to inode struct
1287 * Updates inode attribute information by retrieving the data from the server.
1289 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1291 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1292 if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1293 && !nfs_attribute_timeout(inode))
1294 return NFS_STALE(inode) ? -ESTALE : 0;
1295 return __nfs_revalidate_inode(server, inode);
1299 * nfs_revalidate_mapping - Revalidate the pagecache
1300 * @inode - pointer to host inode
1301 * @mapping - pointer to mapping
1303 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1305 struct nfs_inode *nfsi = NFS_I(inode);
1307 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1308 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1309 if (S_ISREG(inode->i_mode))
1310 nfs_sync_mapping(mapping);
1311 invalidate_inode_pages2(mapping);
1313 spin_lock(&inode->i_lock);
1314 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1315 if (S_ISDIR(inode->i_mode)) {
1316 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1317 /* This ensures we revalidate child dentries */
1318 nfsi->cache_change_attribute = jiffies;
1320 spin_unlock(&inode->i_lock);
1322 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1324 (long long)NFS_FILEID(inode));
1329 * nfs_begin_data_update
1330 * @inode - pointer to inode
1331 * Declare that a set of operations will update file data on the server
1333 void nfs_begin_data_update(struct inode *inode)
1335 atomic_inc(&NFS_I(inode)->data_updates);
1339 * nfs_end_data_update
1340 * @inode - pointer to inode
1341 * Declare end of the operations that will update file data
1342 * This will mark the inode as immediately needing revalidation
1343 * of its attribute cache.
1345 void nfs_end_data_update(struct inode *inode)
1347 struct nfs_inode *nfsi = NFS_I(inode);
1349 if (!nfs_have_delegation(inode, FMODE_READ)) {
1350 /* Directories and symlinks: invalidate page cache */
1351 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
1352 spin_lock(&inode->i_lock);
1353 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1354 spin_unlock(&inode->i_lock);
1357 nfsi->cache_change_attribute = jiffies;
1358 atomic_dec(&nfsi->data_updates);
1361 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1363 struct nfs_inode *nfsi = NFS_I(inode);
1365 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1366 && nfsi->change_attr == fattr->pre_change_attr) {
1367 nfsi->change_attr = fattr->change_attr;
1368 nfsi->cache_change_attribute = jiffies;
1371 /* If we have atomic WCC data, we may update some attributes */
1372 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1373 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1374 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1375 nfsi->cache_change_attribute = jiffies;
1377 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1378 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1379 nfsi->cache_change_attribute = jiffies;
1381 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
1382 inode->i_size = fattr->size;
1383 nfsi->cache_change_attribute = jiffies;
1389 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1390 * @inode - pointer to inode
1391 * @fattr - updated attributes
1393 * Verifies the attribute cache. If we have just changed the attributes,
1394 * so that fattr carries weak cache consistency data, then it may
1395 * also update the ctime/mtime/change_attribute.
1397 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1399 struct nfs_inode *nfsi = NFS_I(inode);
1400 loff_t cur_size, new_isize;
1404 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1407 /* Has the inode gone and changed behind our back? */
1408 if (nfsi->fileid != fattr->fileid
1409 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1413 /* Are we in the process of updating data on the server? */
1414 data_unstable = nfs_caches_unstable(inode);
1416 /* Do atomic weak cache consistency updates */
1417 nfs_wcc_update_inode(inode, fattr);
1419 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0) {
1420 if (nfsi->change_attr == fattr->change_attr)
1422 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1424 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1427 /* Verify a few of the more important attributes */
1428 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1429 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1431 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1434 cur_size = i_size_read(inode);
1435 new_isize = nfs_size_to_loff_t(fattr->size);
1436 if (cur_size != new_isize && nfsi->npages == 0)
1437 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1439 /* Have any file permissions changed? */
1440 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1441 || inode->i_uid != fattr->uid
1442 || inode->i_gid != fattr->gid)
1443 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1445 /* Has the link count changed? */
1446 if (inode->i_nlink != fattr->nlink)
1447 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1450 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1451 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1453 nfsi->read_cache_jiffies = fattr->time_start;
1458 * nfs_refresh_inode - try to update the inode attribute cache
1459 * @inode - pointer to inode
1460 * @fattr - updated attributes
1462 * Check that an RPC call that returned attributes has not overlapped with
1463 * other recent updates of the inode metadata, then decide whether it is
1464 * safe to do a full update of the inode attributes, or whether just to
1465 * call nfs_check_inode_attributes.
1467 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1469 struct nfs_inode *nfsi = NFS_I(inode);
1472 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1474 spin_lock(&inode->i_lock);
1475 nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1476 if (time_after(fattr->time_start, nfsi->last_updated))
1477 status = nfs_update_inode(inode, fattr);
1479 status = nfs_check_inode_attributes(inode, fattr);
1481 spin_unlock(&inode->i_lock);
1486 * nfs_post_op_update_inode - try to update the inode attribute cache
1487 * @inode - pointer to inode
1488 * @fattr - updated attributes
1490 * After an operation that has changed the inode metadata, mark the
1491 * attribute cache as being invalid, then try to update it.
1493 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1495 struct nfs_inode *nfsi = NFS_I(inode);
1498 spin_lock(&inode->i_lock);
1499 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
1500 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1503 status = nfs_update_inode(inode, fattr);
1505 spin_unlock(&inode->i_lock);
1510 * Many nfs protocol calls return the new file attributes after
1511 * an operation. Here we update the inode to reflect the state
1512 * of the server's inode.
1514 * This is a bit tricky because we have to make sure all dirty pages
1515 * have been sent off to the server before calling invalidate_inode_pages.
1516 * To make sure no other process adds more write requests while we try
1517 * our best to flush them, we make them sleep during the attribute refresh.
1519 * A very similar scenario holds for the dir cache.
1521 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1523 struct nfs_inode *nfsi = NFS_I(inode);
1524 loff_t cur_isize, new_isize;
1525 unsigned int invalid = 0;
1528 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1529 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1530 atomic_read(&inode->i_count), fattr->valid);
1532 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1535 if (nfsi->fileid != fattr->fileid)
1539 * Make sure the inode's type hasn't changed.
1541 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1545 * Update the read time so we don't revalidate too often.
1547 nfsi->read_cache_jiffies = fattr->time_start;
1548 nfsi->last_updated = jiffies;
1550 /* Are we racing with known updates of the metadata on the server? */
1551 data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
1553 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1555 /* Do atomic weak cache consistency updates */
1556 nfs_wcc_update_inode(inode, fattr);
1558 /* Check if our cached file size is stale */
1559 new_isize = nfs_size_to_loff_t(fattr->size);
1560 cur_isize = i_size_read(inode);
1561 if (new_isize != cur_isize) {
1562 /* Do we perhaps have any outstanding writes? */
1563 if (nfsi->npages == 0) {
1564 /* No, but did we race with nfs_end_data_update()? */
1566 inode->i_size = new_isize;
1567 invalid |= NFS_INO_INVALID_DATA;
1569 invalid |= NFS_INO_INVALID_ATTR;
1570 } else if (new_isize > cur_isize) {
1571 inode->i_size = new_isize;
1572 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1574 nfsi->cache_change_attribute = jiffies;
1575 dprintk("NFS: isize change on server for file %s/%ld\n",
1576 inode->i_sb->s_id, inode->i_ino);
1579 /* Check if the mtime agrees */
1580 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1581 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1582 dprintk("NFS: mtime change on server for file %s/%ld\n",
1583 inode->i_sb->s_id, inode->i_ino);
1584 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1585 nfsi->cache_change_attribute = jiffies;
1588 /* If ctime has changed we should definitely clear access+acl caches */
1589 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1590 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1591 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1592 nfsi->cache_change_attribute = jiffies;
1594 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1596 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1597 inode->i_uid != fattr->uid ||
1598 inode->i_gid != fattr->gid)
1599 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1601 inode->i_mode = fattr->mode;
1602 inode->i_nlink = fattr->nlink;
1603 inode->i_uid = fattr->uid;
1604 inode->i_gid = fattr->gid;
1606 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1608 * report the blocks in 512byte units
1610 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1611 inode->i_blksize = inode->i_sb->s_blocksize;
1613 inode->i_blocks = fattr->du.nfs2.blocks;
1614 inode->i_blksize = fattr->du.nfs2.blocksize;
1617 if ((fattr->valid & NFS_ATTR_FATTR_V4)) {
1618 if (nfsi->change_attr != fattr->change_attr) {
1619 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1620 inode->i_sb->s_id, inode->i_ino);
1621 nfsi->change_attr = fattr->change_attr;
1622 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1623 nfsi->cache_change_attribute = jiffies;
1625 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA);
1628 /* Update attrtimeo value if we're out of the unstable period */
1629 if (invalid & NFS_INO_INVALID_ATTR) {
1630 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1631 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1632 nfsi->attrtimeo_timestamp = jiffies;
1633 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1634 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1635 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1636 nfsi->attrtimeo_timestamp = jiffies;
1638 /* Don't invalidate the data if we were to blame */
1639 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1640 || S_ISLNK(inode->i_mode)))
1641 invalid &= ~NFS_INO_INVALID_DATA;
1643 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1644 if (!nfs_have_delegation(inode, FMODE_READ))
1645 nfsi->cache_validity |= invalid;
1650 * Big trouble! The inode has become a different object.
1653 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1654 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1658 * No need to worry about unhashing the dentry, as the
1659 * lookup validation will know that the inode is bad.
1660 * (But we fall through to invalidate the caches.)
1662 nfs_invalidate_inode(inode);
1666 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1667 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1668 NFS_SERVER(inode)->hostname, inode->i_sb->s_id,
1669 (long long)nfsi->fileid, (long long)fattr->fileid);
1674 * File system information
1677 static int nfs_set_super(struct super_block *s, void *data)
1679 s->s_fs_info = data;
1680 return set_anon_super(s, data);
1683 static int nfs_compare_super(struct super_block *sb, void *data)
1685 struct nfs_server *server = data;
1686 struct nfs_server *old = NFS_SB(sb);
1688 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1690 if (old->addr.sin_port != server->addr.sin_port)
1692 return !nfs_compare_fh(&old->fh, &server->fh);
1695 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1696 int flags, const char *dev_name, void *raw_data)
1699 struct nfs_server *server = NULL;
1700 struct super_block *s;
1701 struct nfs_fh *root;
1702 struct nfs_mount_data *data = raw_data;
1704 s = ERR_PTR(-EINVAL);
1706 dprintk("%s: missing data argument\n", __FUNCTION__);
1709 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1710 dprintk("%s: bad mount version\n", __FUNCTION__);
1713 switch (data->version) {
1719 if (data->flags & NFS_MOUNT_VER3) {
1720 dprintk("%s: mount structure version %d does not support NFSv3\n",
1725 data->root.size = NFS2_FHSIZE;
1726 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1728 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1729 dprintk("%s: mount structure version %d does not support strong security\n",
1735 memset(data->context, 0, sizeof(data->context));
1737 #ifndef CONFIG_NFS_V3
1738 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1739 s = ERR_PTR(-EPROTONOSUPPORT);
1740 if (data->flags & NFS_MOUNT_VER3) {
1741 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1744 #endif /* CONFIG_NFS_V3 */
1746 s = ERR_PTR(-ENOMEM);
1747 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
1750 /* Zero out the NFS state stuff */
1751 init_nfsv4_state(server);
1752 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1755 if (data->flags & NFS_MOUNT_VER3)
1756 root->size = data->root.size;
1758 root->size = NFS2_FHSIZE;
1759 s = ERR_PTR(-EINVAL);
1760 if (root->size > sizeof(root->data)) {
1761 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1764 memcpy(root->data, data->root.data, root->size);
1766 /* We now require that the mount process passes the remote address */
1767 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1768 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1769 dprintk("%s: mount program didn't pass remote address!\n",
1774 /* Fire up rpciod if not yet running */
1775 s = ERR_PTR(rpciod_up());
1777 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1778 __FUNCTION__, PTR_ERR(s));
1782 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1783 if (IS_ERR(s) || s->s_root)
1784 goto out_rpciod_down;
1788 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1790 up_write(&s->s_umount);
1791 deactivate_super(s);
1792 return ERR_PTR(error);
1794 s->s_flags |= MS_ACTIVE;
1803 static void nfs_kill_super(struct super_block *s)
1805 struct nfs_server *server = NFS_SB(s);
1809 if (!IS_ERR(server->client))
1810 rpc_shutdown_client(server->client);
1811 if (!IS_ERR(server->client_sys))
1812 rpc_shutdown_client(server->client_sys);
1813 if (!IS_ERR(server->client_acl))
1814 rpc_shutdown_client(server->client_acl);
1816 if (!(server->flags & NFS_MOUNT_NONLM))
1817 lockd_down(); /* release rpc.lockd */
1819 rpciod_down(); /* release rpciod */
1821 kfree(server->hostname);
1825 static struct file_system_type nfs_fs_type = {
1826 .owner = THIS_MODULE,
1828 .get_sb = nfs_get_sb,
1829 .kill_sb = nfs_kill_super,
1830 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1833 #ifdef CONFIG_NFS_V4
1835 static void nfs4_clear_inode(struct inode *);
1838 static struct super_operations nfs4_sops = {
1839 .alloc_inode = nfs_alloc_inode,
1840 .destroy_inode = nfs_destroy_inode,
1841 .write_inode = nfs_write_inode,
1842 .delete_inode = nfs_delete_inode,
1843 .statfs = nfs_statfs,
1844 .clear_inode = nfs4_clear_inode,
1845 .umount_begin = nfs_umount_begin,
1846 .show_options = nfs_show_options,
1847 .show_stats = nfs_show_stats,
1851 * Clean out any remaining NFSv4 state that might be left over due
1852 * to open() calls that passed nfs_atomic_lookup, but failed to call
1855 static void nfs4_clear_inode(struct inode *inode)
1857 struct nfs_inode *nfsi = NFS_I(inode);
1859 /* If we are holding a delegation, return it! */
1860 nfs_inode_return_delegation(inode);
1861 /* First call standard NFS clear_inode() code */
1862 nfs_clear_inode(inode);
1863 /* Now clear out any remaining state */
1864 while (!list_empty(&nfsi->open_states)) {
1865 struct nfs4_state *state;
1867 state = list_entry(nfsi->open_states.next,
1870 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1873 (long long)NFS_FILEID(inode),
1875 BUG_ON(atomic_read(&state->count) != 1);
1876 nfs4_close_state(state, state->state);
1881 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1883 struct nfs_server *server;
1884 struct nfs4_client *clp = NULL;
1885 struct rpc_xprt *xprt = NULL;
1886 struct rpc_clnt *clnt = NULL;
1887 struct rpc_timeout timeparms;
1888 rpc_authflavor_t authflavour;
1891 sb->s_blocksize_bits = 0;
1892 sb->s_blocksize = 0;
1893 server = NFS_SB(sb);
1894 if (data->rsize != 0)
1895 server->rsize = nfs_block_size(data->rsize, NULL);
1896 if (data->wsize != 0)
1897 server->wsize = nfs_block_size(data->wsize, NULL);
1898 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1899 server->caps = NFS_CAP_ATOMIC_OPEN;
1901 server->acregmin = data->acregmin*HZ;
1902 server->acregmax = data->acregmax*HZ;
1903 server->acdirmin = data->acdirmin*HZ;
1904 server->acdirmax = data->acdirmax*HZ;
1906 server->rpc_ops = &nfs_v4_clientops;
1908 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
1910 server->retrans_timeo = timeparms.to_initval;
1911 server->retrans_count = timeparms.to_retries;
1913 clp = nfs4_get_client(&server->addr.sin_addr);
1915 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1919 /* Now create transport and client */
1920 authflavour = RPC_AUTH_UNIX;
1921 if (data->auth_flavourlen != 0) {
1922 if (data->auth_flavourlen != 1) {
1923 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1924 __FUNCTION__, data->auth_flavourlen);
1928 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1934 down_write(&clp->cl_sem);
1935 if (IS_ERR(clp->cl_rpcclient)) {
1936 xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
1938 up_write(&clp->cl_sem);
1939 err = PTR_ERR(xprt);
1940 dprintk("%s: cannot create RPC transport. Error = %d\n",
1944 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1945 server->rpc_ops->version, authflavour);
1947 up_write(&clp->cl_sem);
1948 err = PTR_ERR(clnt);
1949 dprintk("%s: cannot create RPC client. Error = %d\n",
1954 clnt->cl_softrtry = 1;
1955 clp->cl_rpcclient = clnt;
1956 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1959 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1960 clnt = rpc_clone_client(clp->cl_rpcclient);
1962 server->nfs4_state = clp;
1963 up_write(&clp->cl_sem);
1967 err = PTR_ERR(clnt);
1968 dprintk("%s: cannot create RPC client. Error = %d\n",
1973 server->client = clnt;
1975 if (server->nfs4_state->cl_idmap == NULL) {
1976 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1980 if (clnt->cl_auth->au_flavor != authflavour) {
1981 struct rpc_auth *auth;
1983 auth = rpcauth_create(authflavour, clnt);
1985 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1986 return PTR_ERR(auth);
1990 sb->s_time_gran = 1;
1992 sb->s_op = &nfs4_sops;
1993 err = nfs_sb_init(sb, authflavour);
1998 nfs4_put_client(clp);
2002 static int nfs4_compare_super(struct super_block *sb, void *data)
2004 struct nfs_server *server = data;
2005 struct nfs_server *old = NFS_SB(sb);
2007 if (strcmp(server->hostname, old->hostname) != 0)
2009 if (strcmp(server->mnt_path, old->mnt_path) != 0)
2015 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
2020 return ERR_PTR(-EINVAL);
2021 if (src->len < maxlen)
2024 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
2026 return ERR_PTR(-ENOMEM);
2028 if (copy_from_user(dst, src->data, maxlen)) {
2030 return ERR_PTR(-EFAULT);
2036 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
2037 int flags, const char *dev_name, void *raw_data)
2040 struct nfs_server *server;
2041 struct super_block *s;
2042 struct nfs4_mount_data *data = raw_data;
2046 dprintk("%s: missing data argument\n", __FUNCTION__);
2047 return ERR_PTR(-EINVAL);
2049 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
2050 dprintk("%s: bad mount version\n", __FUNCTION__);
2051 return ERR_PTR(-EINVAL);
2054 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
2056 return ERR_PTR(-ENOMEM);
2057 /* Zero out the NFS state stuff */
2058 init_nfsv4_state(server);
2059 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
2061 p = nfs_copy_user_string(NULL, &data->hostname, 256);
2064 server->hostname = p;
2066 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
2069 server->mnt_path = p;
2071 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
2072 sizeof(server->ip_addr) - 1);
2076 /* We now require that the mount process passes the remote address */
2077 if (data->host_addrlen != sizeof(server->addr)) {
2078 s = ERR_PTR(-EINVAL);
2081 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
2082 s = ERR_PTR(-EFAULT);
2085 if (server->addr.sin_family != AF_INET ||
2086 server->addr.sin_addr.s_addr == INADDR_ANY) {
2087 dprintk("%s: mount program didn't pass remote IP address!\n",
2089 s = ERR_PTR(-EINVAL);
2093 /* Fire up rpciod if not yet running */
2094 s = ERR_PTR(rpciod_up());
2096 dprintk("%s: couldn't start rpciod! Error = %ld\n",
2097 __FUNCTION__, PTR_ERR(s));
2101 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
2103 if (IS_ERR(s) || s->s_root)
2108 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
2110 up_write(&s->s_umount);
2111 deactivate_super(s);
2112 return ERR_PTR(error);
2114 s->s_flags |= MS_ACTIVE;
2117 s = (struct super_block *)p;
2119 kfree(server->mnt_path);
2120 kfree(server->hostname);
2121 nfs_free_iostats(server->io_stats);
2126 static void nfs4_kill_super(struct super_block *sb)
2128 struct nfs_server *server = NFS_SB(sb);
2130 nfs_return_all_delegations(sb);
2131 kill_anon_super(sb);
2133 nfs4_renewd_prepare_shutdown(server);
2135 if (server->client != NULL && !IS_ERR(server->client))
2136 rpc_shutdown_client(server->client);
2138 destroy_nfsv4_state(server);
2142 kfree(server->hostname);
2146 static struct file_system_type nfs4_fs_type = {
2147 .owner = THIS_MODULE,
2149 .get_sb = nfs4_get_sb,
2150 .kill_sb = nfs4_kill_super,
2151 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2154 static const int nfs_set_port_min = 0;
2155 static const int nfs_set_port_max = 65535;
2156 static int param_set_port(const char *val, struct kernel_param *kp)
2159 int num = simple_strtol(val, &endp, 0);
2160 if (endp == val || *endp || num < nfs_set_port_min || num > nfs_set_port_max)
2162 *((int *)kp->arg) = num;
2166 module_param_call(callback_tcpport, param_set_port, param_get_int,
2167 &nfs_callback_set_tcpport, 0644);
2169 static int param_set_idmap_timeout(const char *val, struct kernel_param *kp)
2172 int num = simple_strtol(val, &endp, 0);
2174 if (endp == val || *endp || num < 0 || jif < num)
2176 *((int *)kp->arg) = jif;
2180 module_param_call(idmap_cache_timeout, param_set_idmap_timeout, param_get_int,
2181 &nfs_idmap_cache_timeout, 0644);
2183 #define nfs4_init_once(nfsi) \
2185 INIT_LIST_HEAD(&(nfsi)->open_states); \
2186 nfsi->delegation = NULL; \
2187 nfsi->delegation_state = 0; \
2188 init_rwsem(&nfsi->rwsem); \
2191 static inline int register_nfs4fs(void)
2195 ret = nfs_register_sysctl();
2198 ret = register_filesystem(&nfs4_fs_type);
2200 nfs_unregister_sysctl();
2204 static inline void unregister_nfs4fs(void)
2206 unregister_filesystem(&nfs4_fs_type);
2207 nfs_unregister_sysctl();
2210 #define nfs4_init_once(nfsi) \
2212 #define register_nfs4fs() (0)
2213 #define unregister_nfs4fs()
2216 extern int nfs_init_nfspagecache(void);
2217 extern void nfs_destroy_nfspagecache(void);
2218 extern int nfs_init_readpagecache(void);
2219 extern void nfs_destroy_readpagecache(void);
2220 extern int nfs_init_writepagecache(void);
2221 extern void nfs_destroy_writepagecache(void);
2222 #ifdef CONFIG_NFS_DIRECTIO
2223 extern int nfs_init_directcache(void);
2224 extern void nfs_destroy_directcache(void);
2227 static kmem_cache_t * nfs_inode_cachep;
2229 static struct inode *nfs_alloc_inode(struct super_block *sb)
2231 struct nfs_inode *nfsi;
2232 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2236 nfsi->cache_validity = 0UL;
2237 nfsi->cache_change_attribute = jiffies;
2238 #ifdef CONFIG_NFS_V3_ACL
2239 nfsi->acl_access = ERR_PTR(-EAGAIN);
2240 nfsi->acl_default = ERR_PTR(-EAGAIN);
2242 #ifdef CONFIG_NFS_V4
2243 nfsi->nfs4_acl = NULL;
2244 #endif /* CONFIG_NFS_V4 */
2245 return &nfsi->vfs_inode;
2248 static void nfs_destroy_inode(struct inode *inode)
2250 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2253 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2255 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2257 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2258 SLAB_CTOR_CONSTRUCTOR) {
2259 inode_init_once(&nfsi->vfs_inode);
2260 spin_lock_init(&nfsi->req_lock);
2261 INIT_LIST_HEAD(&nfsi->dirty);
2262 INIT_LIST_HEAD(&nfsi->commit);
2263 INIT_LIST_HEAD(&nfsi->open_files);
2264 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2265 atomic_set(&nfsi->data_updates, 0);
2269 nfs4_init_once(nfsi);
2273 static int nfs_init_inodecache(void)
2275 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2276 sizeof(struct nfs_inode),
2277 0, SLAB_RECLAIM_ACCOUNT,
2279 if (nfs_inode_cachep == NULL)
2285 static void nfs_destroy_inodecache(void)
2287 if (kmem_cache_destroy(nfs_inode_cachep))
2288 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2294 static int __init init_nfs_fs(void)
2298 err = nfs_init_nfspagecache();
2302 err = nfs_init_inodecache();
2306 err = nfs_init_readpagecache();
2310 err = nfs_init_writepagecache();
2314 #ifdef CONFIG_NFS_DIRECTIO
2315 err = nfs_init_directcache();
2320 #ifdef CONFIG_PROC_FS
2321 rpc_proc_register(&nfs_rpcstat);
2323 err = register_filesystem(&nfs_fs_type);
2326 if ((err = register_nfs4fs()) != 0)
2330 #ifdef CONFIG_PROC_FS
2331 rpc_proc_unregister("nfs");
2333 #ifdef CONFIG_NFS_DIRECTIO
2334 nfs_destroy_directcache();
2337 nfs_destroy_writepagecache();
2339 nfs_destroy_readpagecache();
2341 nfs_destroy_inodecache();
2343 nfs_destroy_nfspagecache();
2348 static void __exit exit_nfs_fs(void)
2350 #ifdef CONFIG_NFS_DIRECTIO
2351 nfs_destroy_directcache();
2353 nfs_destroy_writepagecache();
2354 nfs_destroy_readpagecache();
2355 nfs_destroy_inodecache();
2356 nfs_destroy_nfspagecache();
2357 #ifdef CONFIG_PROC_FS
2358 rpc_proc_unregister("nfs");
2360 unregister_filesystem(&nfs_fs_type);
2361 unregister_nfs4fs();
2364 /* Not quite true; I just maintain it */
2365 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2366 MODULE_LICENSE("GPL");
2368 module_init(init_nfs_fs)
2369 module_exit(exit_nfs_fs)