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 /* Get some general file system info */
303 if (server->namelen == 0 &&
304 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
305 server->namelen = pathinfo.max_namelen;
306 /* Work out a lot of parameters */
307 if (server->rsize == 0)
308 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
309 if (server->wsize == 0)
310 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
312 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
313 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
314 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
315 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
317 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
318 if (server->rsize > max_rpc_payload)
319 server->rsize = max_rpc_payload;
320 if (server->rsize > NFS_MAX_FILE_IO_SIZE)
321 server->rsize = NFS_MAX_FILE_IO_SIZE;
322 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
324 if (server->wsize > max_rpc_payload)
325 server->wsize = max_rpc_payload;
326 if (server->wsize > NFS_MAX_FILE_IO_SIZE)
327 server->wsize = NFS_MAX_FILE_IO_SIZE;
328 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
330 if (sb->s_blocksize == 0)
331 sb->s_blocksize = nfs_block_bits(server->wsize,
332 &sb->s_blocksize_bits);
333 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
335 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
336 if (server->dtsize > PAGE_CACHE_SIZE)
337 server->dtsize = PAGE_CACHE_SIZE;
338 if (server->dtsize > server->rsize)
339 server->dtsize = server->rsize;
341 if (server->flags & NFS_MOUNT_NOAC) {
342 server->acregmin = server->acregmax = 0;
343 server->acdirmin = server->acdirmax = 0;
344 sb->s_flags |= MS_SYNCHRONOUS;
346 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
348 sb->s_maxbytes = fsinfo.maxfilesize;
349 if (sb->s_maxbytes > MAX_LFS_FILESIZE)
350 sb->s_maxbytes = MAX_LFS_FILESIZE;
352 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
353 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
355 /* We're airborne Set socket buffersize */
356 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
358 /* Yargs. It didn't work out. */
360 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
361 if (!IS_ERR(root_inode))
363 return no_root_error;
366 static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
368 to->to_initval = timeo * HZ / 10;
369 to->to_retries = retrans;
376 to->to_initval = 60 * HZ;
377 if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
378 to->to_initval = NFS_MAX_TCP_TIMEOUT;
379 to->to_increment = to->to_initval;
380 to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
381 to->to_exponential = 0;
386 to->to_initval = 11 * HZ / 10;
387 if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
388 to->to_initval = NFS_MAX_UDP_TIMEOUT;
389 to->to_maxval = NFS_MAX_UDP_TIMEOUT;
390 to->to_exponential = 1;
396 * Create an RPC client handle.
398 static struct rpc_clnt *
399 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
401 struct rpc_timeout timeparms;
402 struct rpc_xprt *xprt = NULL;
403 struct rpc_clnt *clnt = NULL;
404 int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
406 nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
408 server->retrans_timeo = timeparms.to_initval;
409 server->retrans_count = timeparms.to_retries;
411 /* create transport and client */
412 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
414 dprintk("%s: cannot create RPC transport. Error = %ld\n",
415 __FUNCTION__, PTR_ERR(xprt));
416 return (struct rpc_clnt *)xprt;
418 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
419 server->rpc_ops->version, data->pseudoflavor);
421 dprintk("%s: cannot create RPC client. Error = %ld\n",
422 __FUNCTION__, PTR_ERR(xprt));
427 clnt->cl_softrtry = 1;
436 * The way this works is that the mount process passes a structure
437 * in the data argument which contains the server's IP address
438 * and the root file handle obtained from the server's mount
439 * daemon. We stash these away in the private superblock fields.
442 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
444 struct nfs_server *server;
445 rpc_authflavor_t authflavor;
448 sb->s_blocksize_bits = 0;
451 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
453 server->rsize = nfs_block_size(data->rsize, NULL);
455 server->wsize = nfs_block_size(data->wsize, NULL);
456 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
458 server->acregmin = data->acregmin*HZ;
459 server->acregmax = data->acregmax*HZ;
460 server->acdirmin = data->acdirmin*HZ;
461 server->acdirmax = data->acdirmax*HZ;
463 /* Start lockd here, before we might error out */
464 if (!(server->flags & NFS_MOUNT_NONLM))
467 server->namelen = data->namlen;
468 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
469 if (!server->hostname)
471 strcpy(server->hostname, data->hostname);
473 /* Check NFS protocol revision and initialize RPC op vector
474 * and file handle pool. */
476 if (server->flags & NFS_MOUNT_VER3) {
477 server->rpc_ops = &nfs_v3_clientops;
478 server->caps |= NFS_CAP_READDIRPLUS;
480 server->rpc_ops = &nfs_v2_clientops;
483 server->rpc_ops = &nfs_v2_clientops;
486 /* Fill in pseudoflavor for mount version < 5 */
487 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
488 data->pseudoflavor = RPC_AUTH_UNIX;
489 authflavor = data->pseudoflavor; /* save for sb_init() */
490 /* XXX maybe we want to add a server->pseudoflavor field */
492 /* Create RPC client handles */
493 server->client = nfs_create_client(server, data);
494 if (IS_ERR(server->client))
495 return PTR_ERR(server->client);
496 /* RFC 2623, sec 2.3.2 */
497 if (authflavor != RPC_AUTH_UNIX) {
498 struct rpc_auth *auth;
500 server->client_sys = rpc_clone_client(server->client);
501 if (IS_ERR(server->client_sys))
502 return PTR_ERR(server->client_sys);
503 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
505 return PTR_ERR(auth);
507 atomic_inc(&server->client->cl_count);
508 server->client_sys = server->client;
510 if (server->flags & NFS_MOUNT_VER3) {
511 #ifdef CONFIG_NFS_V3_ACL
512 if (!(server->flags & NFS_MOUNT_NOACL)) {
513 server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
514 /* No errors! Assume that Sun nfsacls are supported */
515 if (!IS_ERR(server->client_acl))
516 server->caps |= NFS_CAP_ACLS;
519 server->flags &= ~NFS_MOUNT_NOACL;
520 #endif /* CONFIG_NFS_V3_ACL */
522 * The VFS shouldn't apply the umask to mode bits. We will
523 * do so ourselves when necessary.
525 sb->s_flags |= MS_POSIXACL;
526 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
527 server->namelen = NFS3_MAXNAMLEN;
530 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
531 server->namelen = NFS2_MAXNAMLEN;
534 sb->s_op = &nfs_sops;
535 return nfs_sb_init(sb, authflavor);
539 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
541 struct nfs_server *server = NFS_SB(sb);
542 unsigned char blockbits;
543 unsigned long blockres;
544 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
545 struct nfs_fattr fattr;
546 struct nfs_fsstat res = {
553 error = server->rpc_ops->statfs(server, rootfh, &res);
554 buf->f_type = NFS_SUPER_MAGIC;
559 * Current versions of glibc do not correctly handle the
560 * case where f_frsize != f_bsize. Eventually we want to
561 * report the value of wtmult in this field.
563 buf->f_frsize = sb->s_blocksize;
566 * On most *nix systems, f_blocks, f_bfree, and f_bavail
567 * are reported in units of f_frsize. Linux hasn't had
568 * an f_frsize field in its statfs struct until recently,
569 * thus historically Linux's sys_statfs reports these
570 * fields in units of f_bsize.
572 buf->f_bsize = sb->s_blocksize;
573 blockbits = sb->s_blocksize_bits;
574 blockres = (1 << blockbits) - 1;
575 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
576 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
577 buf->f_bavail = (res.abytes + blockres) >> blockbits;
579 buf->f_files = res.tfiles;
580 buf->f_ffree = res.afiles;
582 buf->f_namelen = server->namelen;
588 dprintk("%s: statfs error = %d\n", __FUNCTION__, -error);
589 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
594 static void nfs_show_mount_options(struct seq_file *m, struct nfs_server *nfss, int showdefaults)
596 static struct proc_nfs_info {
601 { NFS_MOUNT_SOFT, ",soft", ",hard" },
602 { NFS_MOUNT_INTR, ",intr", "" },
603 { NFS_MOUNT_NOCTO, ",nocto", "" },
604 { NFS_MOUNT_NOAC, ",noac", "" },
605 { NFS_MOUNT_NONLM, ",nolock", "" },
606 { NFS_MOUNT_NOACL, ",noacl", "" },
609 struct proc_nfs_info *nfs_infop;
613 seq_printf(m, ",vers=%d", nfss->rpc_ops->version);
614 seq_printf(m, ",rsize=%d", nfss->rsize);
615 seq_printf(m, ",wsize=%d", nfss->wsize);
616 if (nfss->acregmin != 3*HZ || showdefaults)
617 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
618 if (nfss->acregmax != 60*HZ || showdefaults)
619 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
620 if (nfss->acdirmin != 30*HZ || showdefaults)
621 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
622 if (nfss->acdirmax != 60*HZ || showdefaults)
623 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
624 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
625 if (nfss->flags & nfs_infop->flag)
626 seq_puts(m, nfs_infop->str);
628 seq_puts(m, nfs_infop->nostr);
630 switch (nfss->client->cl_xprt->prot) {
638 snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
641 seq_printf(m, ",proto=%s", proto);
642 seq_printf(m, ",timeo=%lu", 10U * nfss->retrans_timeo / HZ);
643 seq_printf(m, ",retrans=%u", nfss->retrans_count);
646 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
648 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
650 nfs_show_mount_options(m, nfss, 0);
652 seq_puts(m, ",addr=");
653 seq_escape(m, nfss->hostname, " \t\n\\");
658 static int nfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
661 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
662 struct rpc_auth *auth = nfss->client->cl_auth;
663 struct nfs_iostats totals = { };
665 seq_printf(m, "statvers=%s", NFS_IOSTAT_VERS);
668 * Display all mount option settings
670 seq_printf(m, "\n\topts:\t");
671 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? "ro" : "rw");
672 seq_puts(m, mnt->mnt_sb->s_flags & MS_SYNCHRONOUS ? ",sync" : "");
673 seq_puts(m, mnt->mnt_sb->s_flags & MS_NOATIME ? ",noatime" : "");
674 seq_puts(m, mnt->mnt_sb->s_flags & MS_NODIRATIME ? ",nodiratime" : "");
675 nfs_show_mount_options(m, nfss, 1);
677 seq_printf(m, "\n\tcaps:\t");
678 seq_printf(m, "caps=0x%x", nfss->caps);
679 seq_printf(m, ",wtmult=%d", nfss->wtmult);
680 seq_printf(m, ",dtsize=%d", nfss->dtsize);
681 seq_printf(m, ",bsize=%d", nfss->bsize);
682 seq_printf(m, ",namelen=%d", nfss->namelen);
685 if (nfss->rpc_ops->version == 4) {
686 seq_printf(m, "\n\tnfsv4:\t");
687 seq_printf(m, "bm0=0x%x", nfss->attr_bitmask[0]);
688 seq_printf(m, ",bm1=0x%x", nfss->attr_bitmask[1]);
689 seq_printf(m, ",acl=0x%x", nfss->acl_bitmask);
694 * Display security flavor in effect for this mount
696 seq_printf(m, "\n\tsec:\tflavor=%d", auth->au_ops->au_flavor);
698 seq_printf(m, ",pseudoflavor=%d", auth->au_flavor);
701 * Display superblock I/O counters
703 for (cpu = 0; cpu < NR_CPUS; cpu++) {
704 struct nfs_iostats *stats;
706 if (!cpu_possible(cpu))
710 stats = per_cpu_ptr(nfss->io_stats, cpu);
712 for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
713 totals.events[i] += stats->events[i];
714 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
715 totals.bytes[i] += stats->bytes[i];
720 seq_printf(m, "\n\tevents:\t");
721 for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
722 seq_printf(m, "%lu ", totals.events[i]);
723 seq_printf(m, "\n\tbytes:\t");
724 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
725 seq_printf(m, "%Lu ", totals.bytes[i]);
731 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
733 int nfs_sync_mapping(struct address_space *mapping)
737 if (mapping->nrpages == 0)
739 unmap_mapping_range(mapping, 0, 0, 0);
740 ret = filemap_write_and_wait(mapping);
743 ret = nfs_wb_all(mapping->host);
749 * Invalidate the local caches
751 static void nfs_zap_caches_locked(struct inode *inode)
753 struct nfs_inode *nfsi = NFS_I(inode);
754 int mode = inode->i_mode;
756 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
758 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
759 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
761 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
762 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
763 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
765 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
768 void nfs_zap_caches(struct inode *inode)
770 spin_lock(&inode->i_lock);
771 nfs_zap_caches_locked(inode);
772 spin_unlock(&inode->i_lock);
775 static void nfs_zap_acl_cache(struct inode *inode)
777 void (*clear_acl_cache)(struct inode *);
779 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
780 if (clear_acl_cache != NULL)
781 clear_acl_cache(inode);
782 spin_lock(&inode->i_lock);
783 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
784 spin_unlock(&inode->i_lock);
788 * Invalidate, but do not unhash, the inode.
789 * NB: must be called with inode->i_lock held!
791 static void nfs_invalidate_inode(struct inode *inode)
793 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
794 nfs_zap_caches_locked(inode);
797 struct nfs_find_desc {
799 struct nfs_fattr *fattr;
803 * In NFSv3 we can have 64bit inode numbers. In order to support
804 * this, and re-exported directories (also seen in NFSv2)
805 * we are forced to allow 2 different inodes to have the same
809 nfs_find_actor(struct inode *inode, void *opaque)
811 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
812 struct nfs_fh *fh = desc->fh;
813 struct nfs_fattr *fattr = desc->fattr;
815 if (NFS_FILEID(inode) != fattr->fileid)
817 if (nfs_compare_fh(NFS_FH(inode), fh))
819 if (is_bad_inode(inode) || NFS_STALE(inode))
825 nfs_init_locked(struct inode *inode, void *opaque)
827 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
828 struct nfs_fattr *fattr = desc->fattr;
830 NFS_FILEID(inode) = fattr->fileid;
831 nfs_copy_fh(NFS_FH(inode), desc->fh);
835 /* Don't use READDIRPLUS on directories that we believe are too large */
836 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
839 * This is our front-end to iget that looks up inodes by file handle
840 * instead of inode number.
843 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
845 struct nfs_find_desc desc = {
849 struct inode *inode = NULL;
852 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
856 printk("NFS: Buggy server - nlink == 0!\n");
860 hash = nfs_fattr_to_ino_t(fattr);
862 if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
865 if (inode->i_state & I_NEW) {
866 struct nfs_inode *nfsi = NFS_I(inode);
868 /* We set i_ino for the few things that still rely on it,
872 /* We can't support update_atime(), since the server will reset it */
873 inode->i_flags |= S_NOATIME|S_NOCMTIME;
874 inode->i_mode = fattr->mode;
875 /* Why so? Because we want revalidate for devices/FIFOs, and
876 * that's precisely what we have in nfs_file_inode_operations.
878 inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
879 if (S_ISREG(inode->i_mode)) {
880 inode->i_fop = &nfs_file_operations;
881 inode->i_data.a_ops = &nfs_file_aops;
882 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
883 } else if (S_ISDIR(inode->i_mode)) {
884 inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
885 inode->i_fop = &nfs_dir_operations;
886 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
887 && fattr->size <= NFS_LIMIT_READDIRPLUS)
888 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
889 } else if (S_ISLNK(inode->i_mode))
890 inode->i_op = &nfs_symlink_inode_operations;
892 init_special_inode(inode, inode->i_mode, fattr->rdev);
894 nfsi->read_cache_jiffies = fattr->time_start;
895 nfsi->last_updated = jiffies;
896 inode->i_atime = fattr->atime;
897 inode->i_mtime = fattr->mtime;
898 inode->i_ctime = fattr->ctime;
899 if (fattr->valid & NFS_ATTR_FATTR_V4)
900 nfsi->change_attr = fattr->change_attr;
901 inode->i_size = nfs_size_to_loff_t(fattr->size);
902 inode->i_nlink = fattr->nlink;
903 inode->i_uid = fattr->uid;
904 inode->i_gid = fattr->gid;
905 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
907 * report the blocks in 512byte units
909 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
910 inode->i_blksize = inode->i_sb->s_blocksize;
912 inode->i_blocks = fattr->du.nfs2.blocks;
913 inode->i_blksize = fattr->du.nfs2.blocksize;
915 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
916 nfsi->attrtimeo_timestamp = jiffies;
917 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
918 nfsi->cache_access.cred = NULL;
920 unlock_new_inode(inode);
922 nfs_refresh_inode(inode, fattr);
923 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
925 (long long)NFS_FILEID(inode),
926 atomic_read(&inode->i_count));
932 printk("nfs_fhget: iget failed\n");
936 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
939 nfs_setattr(struct dentry *dentry, struct iattr *attr)
941 struct inode *inode = dentry->d_inode;
942 struct nfs_fattr fattr;
945 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
947 if (attr->ia_valid & ATTR_SIZE) {
948 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
949 attr->ia_valid &= ~ATTR_SIZE;
952 /* Optimization: if the end result is no change, don't RPC */
953 attr->ia_valid &= NFS_VALID_ATTRS;
954 if (attr->ia_valid == 0)
958 nfs_begin_data_update(inode);
959 /* Write all dirty data */
960 filemap_write_and_wait(inode->i_mapping);
963 * Return any delegations if we're going to change ACLs
965 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
966 nfs_inode_return_delegation(inode);
967 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
969 nfs_refresh_inode(inode, &fattr);
970 nfs_end_data_update(inode);
976 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
977 * @inode: pointer to struct inode
978 * @attr: pointer to struct iattr
980 * Note: we do this in the *proc.c in order to ensure that
981 * it works for things like exclusive creates too.
983 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
985 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
986 if ((attr->ia_valid & ATTR_MODE) != 0) {
987 int mode = attr->ia_mode & S_IALLUGO;
988 mode |= inode->i_mode & ~S_IALLUGO;
989 inode->i_mode = mode;
991 if ((attr->ia_valid & ATTR_UID) != 0)
992 inode->i_uid = attr->ia_uid;
993 if ((attr->ia_valid & ATTR_GID) != 0)
994 inode->i_gid = attr->ia_gid;
995 spin_lock(&inode->i_lock);
996 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
997 spin_unlock(&inode->i_lock);
999 if ((attr->ia_valid & ATTR_SIZE) != 0) {
1000 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
1001 inode->i_size = attr->ia_size;
1002 vmtruncate(inode, attr->ia_size);
1006 static int nfs_wait_schedule(void *word)
1008 if (signal_pending(current))
1009 return -ERESTARTSYS;
1015 * Wait for the inode to get unlocked.
1017 static int nfs_wait_on_inode(struct inode *inode)
1019 struct rpc_clnt *clnt = NFS_CLIENT(inode);
1020 struct nfs_inode *nfsi = NFS_I(inode);
1024 rpc_clnt_sigmask(clnt, &oldmask);
1025 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
1026 nfs_wait_schedule, TASK_INTERRUPTIBLE);
1027 rpc_clnt_sigunmask(clnt, &oldmask);
1032 static void nfs_wake_up_inode(struct inode *inode)
1034 struct nfs_inode *nfsi = NFS_I(inode);
1036 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
1037 smp_mb__after_clear_bit();
1038 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
1041 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1043 struct inode *inode = dentry->d_inode;
1044 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
1047 /* Flush out writes to the server in order to update c/mtime */
1048 nfs_sync_inode(inode, 0, 0, FLUSH_WAIT|FLUSH_NOCOMMIT);
1051 * We may force a getattr if the user cares about atime.
1053 * Note that we only have to check the vfsmount flags here:
1054 * - NFS always sets S_NOATIME by so checking it would give a
1056 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
1057 * no point in checking those.
1059 if ((mnt->mnt_flags & MNT_NOATIME) ||
1060 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
1064 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1066 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1068 generic_fillattr(inode, stat);
1072 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
1074 struct nfs_open_context *ctx;
1076 ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
1078 atomic_set(&ctx->count, 1);
1079 ctx->dentry = dget(dentry);
1080 ctx->vfsmnt = mntget(mnt);
1081 ctx->cred = get_rpccred(cred);
1083 ctx->lockowner = current->files;
1085 ctx->dir_cookie = 0;
1090 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1093 atomic_inc(&ctx->count);
1097 void put_nfs_open_context(struct nfs_open_context *ctx)
1099 if (atomic_dec_and_test(&ctx->count)) {
1100 if (!list_empty(&ctx->list)) {
1101 struct inode *inode = ctx->dentry->d_inode;
1102 spin_lock(&inode->i_lock);
1103 list_del(&ctx->list);
1104 spin_unlock(&inode->i_lock);
1106 if (ctx->state != NULL)
1107 nfs4_close_state(ctx->state, ctx->mode);
1108 if (ctx->cred != NULL)
1109 put_rpccred(ctx->cred);
1111 mntput(ctx->vfsmnt);
1117 * Ensure that mmap has a recent RPC credential for use when writing out
1120 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1122 struct inode *inode = filp->f_dentry->d_inode;
1123 struct nfs_inode *nfsi = NFS_I(inode);
1125 filp->private_data = get_nfs_open_context(ctx);
1126 spin_lock(&inode->i_lock);
1127 list_add(&ctx->list, &nfsi->open_files);
1128 spin_unlock(&inode->i_lock);
1132 * Given an inode, search for an open context with the desired characteristics
1134 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
1136 struct nfs_inode *nfsi = NFS_I(inode);
1137 struct nfs_open_context *pos, *ctx = NULL;
1139 spin_lock(&inode->i_lock);
1140 list_for_each_entry(pos, &nfsi->open_files, list) {
1141 if (cred != NULL && pos->cred != cred)
1143 if ((pos->mode & mode) == mode) {
1144 ctx = get_nfs_open_context(pos);
1148 spin_unlock(&inode->i_lock);
1152 static void nfs_file_clear_open_context(struct file *filp)
1154 struct inode *inode = filp->f_dentry->d_inode;
1155 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
1158 filp->private_data = NULL;
1159 spin_lock(&inode->i_lock);
1160 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
1161 spin_unlock(&inode->i_lock);
1162 put_nfs_open_context(ctx);
1167 * These allocate and release file read/write context information.
1169 int nfs_open(struct inode *inode, struct file *filp)
1171 struct nfs_open_context *ctx;
1172 struct rpc_cred *cred;
1174 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1176 return PTR_ERR(cred);
1177 ctx = alloc_nfs_open_context(filp->f_vfsmnt, filp->f_dentry, cred);
1181 ctx->mode = filp->f_mode;
1182 nfs_file_set_open_context(filp, ctx);
1183 put_nfs_open_context(ctx);
1187 int nfs_release(struct inode *inode, struct file *filp)
1189 nfs_file_clear_open_context(filp);
1194 * This function is called whenever some part of NFS notices that
1195 * the cached attributes have to be refreshed.
1198 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1200 int status = -ESTALE;
1201 struct nfs_fattr fattr;
1202 struct nfs_inode *nfsi = NFS_I(inode);
1204 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
1205 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
1208 if (!inode || is_bad_inode(inode))
1210 if (NFS_STALE(inode))
1213 status = nfs_wait_on_inode(inode);
1216 if (NFS_STALE(inode)) {
1218 /* Do we trust the cached ESTALE? */
1219 if (NFS_ATTRTIMEO(inode) != 0) {
1220 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
1227 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1229 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1231 (long long)NFS_FILEID(inode), status);
1232 if (status == -ESTALE) {
1233 nfs_zap_caches(inode);
1234 if (!S_ISDIR(inode->i_mode))
1235 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1240 spin_lock(&inode->i_lock);
1241 status = nfs_update_inode(inode, &fattr);
1243 spin_unlock(&inode->i_lock);
1244 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1246 (long long)NFS_FILEID(inode), status);
1249 spin_unlock(&inode->i_lock);
1251 nfs_revalidate_mapping(inode, inode->i_mapping);
1253 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1254 nfs_zap_acl_cache(inode);
1256 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1258 (long long)NFS_FILEID(inode));
1261 nfs_wake_up_inode(inode);
1268 int nfs_attribute_timeout(struct inode *inode)
1270 struct nfs_inode *nfsi = NFS_I(inode);
1272 if (nfs_have_delegation(inode, FMODE_READ))
1274 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1278 * nfs_revalidate_inode - Revalidate the inode attributes
1279 * @server - pointer to nfs_server struct
1280 * @inode - pointer to inode struct
1282 * Updates inode attribute information by retrieving the data from the server.
1284 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1286 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1287 if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
1288 && !nfs_attribute_timeout(inode))
1289 return NFS_STALE(inode) ? -ESTALE : 0;
1290 return __nfs_revalidate_inode(server, inode);
1294 * nfs_revalidate_mapping - Revalidate the pagecache
1295 * @inode - pointer to host inode
1296 * @mapping - pointer to mapping
1298 void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1300 struct nfs_inode *nfsi = NFS_I(inode);
1302 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1303 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1304 if (S_ISREG(inode->i_mode))
1305 nfs_sync_mapping(mapping);
1306 invalidate_inode_pages2(mapping);
1308 spin_lock(&inode->i_lock);
1309 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1310 if (S_ISDIR(inode->i_mode)) {
1311 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1312 /* This ensures we revalidate child dentries */
1313 nfsi->cache_change_attribute = jiffies;
1315 spin_unlock(&inode->i_lock);
1317 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1319 (long long)NFS_FILEID(inode));
1324 * nfs_begin_data_update
1325 * @inode - pointer to inode
1326 * Declare that a set of operations will update file data on the server
1328 void nfs_begin_data_update(struct inode *inode)
1330 atomic_inc(&NFS_I(inode)->data_updates);
1334 * nfs_end_data_update
1335 * @inode - pointer to inode
1336 * Declare end of the operations that will update file data
1337 * This will mark the inode as immediately needing revalidation
1338 * of its attribute cache.
1340 void nfs_end_data_update(struct inode *inode)
1342 struct nfs_inode *nfsi = NFS_I(inode);
1344 if (!nfs_have_delegation(inode, FMODE_READ)) {
1345 /* Directories and symlinks: invalidate page cache */
1346 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
1347 spin_lock(&inode->i_lock);
1348 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1349 spin_unlock(&inode->i_lock);
1352 nfsi->cache_change_attribute = jiffies;
1353 atomic_dec(&nfsi->data_updates);
1356 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1358 struct nfs_inode *nfsi = NFS_I(inode);
1360 if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
1361 && nfsi->change_attr == fattr->pre_change_attr) {
1362 nfsi->change_attr = fattr->change_attr;
1363 nfsi->cache_change_attribute = jiffies;
1366 /* If we have atomic WCC data, we may update some attributes */
1367 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
1368 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1369 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1370 nfsi->cache_change_attribute = jiffies;
1372 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1373 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1374 nfsi->cache_change_attribute = jiffies;
1376 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
1377 inode->i_size = fattr->size;
1378 nfsi->cache_change_attribute = jiffies;
1384 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1385 * @inode - pointer to inode
1386 * @fattr - updated attributes
1388 * Verifies the attribute cache. If we have just changed the attributes,
1389 * so that fattr carries weak cache consistency data, then it may
1390 * also update the ctime/mtime/change_attribute.
1392 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1394 struct nfs_inode *nfsi = NFS_I(inode);
1395 loff_t cur_size, new_isize;
1399 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1402 /* Has the inode gone and changed behind our back? */
1403 if (nfsi->fileid != fattr->fileid
1404 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1408 /* Are we in the process of updating data on the server? */
1409 data_unstable = nfs_caches_unstable(inode);
1411 /* Do atomic weak cache consistency updates */
1412 nfs_wcc_update_inode(inode, fattr);
1414 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0) {
1415 if (nfsi->change_attr == fattr->change_attr)
1417 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1419 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1422 /* Verify a few of the more important attributes */
1423 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1424 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1426 nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1429 cur_size = i_size_read(inode);
1430 new_isize = nfs_size_to_loff_t(fattr->size);
1431 if (cur_size != new_isize && nfsi->npages == 0)
1432 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1434 /* Have any file permissions changed? */
1435 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1436 || inode->i_uid != fattr->uid
1437 || inode->i_gid != fattr->gid)
1438 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1440 /* Has the link count changed? */
1441 if (inode->i_nlink != fattr->nlink)
1442 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1445 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1446 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1448 nfsi->read_cache_jiffies = fattr->time_start;
1453 * nfs_refresh_inode - try to update the inode attribute cache
1454 * @inode - pointer to inode
1455 * @fattr - updated attributes
1457 * Check that an RPC call that returned attributes has not overlapped with
1458 * other recent updates of the inode metadata, then decide whether it is
1459 * safe to do a full update of the inode attributes, or whether just to
1460 * call nfs_check_inode_attributes.
1462 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1464 struct nfs_inode *nfsi = NFS_I(inode);
1467 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1469 spin_lock(&inode->i_lock);
1470 nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
1471 if (time_after(fattr->time_start, nfsi->last_updated))
1472 status = nfs_update_inode(inode, fattr);
1474 status = nfs_check_inode_attributes(inode, fattr);
1476 spin_unlock(&inode->i_lock);
1481 * nfs_post_op_update_inode - try to update the inode attribute cache
1482 * @inode - pointer to inode
1483 * @fattr - updated attributes
1485 * After an operation that has changed the inode metadata, mark the
1486 * attribute cache as being invalid, then try to update it.
1488 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1490 struct nfs_inode *nfsi = NFS_I(inode);
1493 spin_lock(&inode->i_lock);
1494 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
1495 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
1498 status = nfs_update_inode(inode, fattr);
1500 spin_unlock(&inode->i_lock);
1505 * Many nfs protocol calls return the new file attributes after
1506 * an operation. Here we update the inode to reflect the state
1507 * of the server's inode.
1509 * This is a bit tricky because we have to make sure all dirty pages
1510 * have been sent off to the server before calling invalidate_inode_pages.
1511 * To make sure no other process adds more write requests while we try
1512 * our best to flush them, we make them sleep during the attribute refresh.
1514 * A very similar scenario holds for the dir cache.
1516 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1518 struct nfs_inode *nfsi = NFS_I(inode);
1519 loff_t cur_isize, new_isize;
1520 unsigned int invalid = 0;
1523 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1524 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1525 atomic_read(&inode->i_count), fattr->valid);
1527 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1530 if (nfsi->fileid != fattr->fileid)
1534 * Make sure the inode's type hasn't changed.
1536 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1540 * Update the read time so we don't revalidate too often.
1542 nfsi->read_cache_jiffies = fattr->time_start;
1543 nfsi->last_updated = jiffies;
1545 /* Are we racing with known updates of the metadata on the server? */
1546 data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
1548 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
1550 /* Do atomic weak cache consistency updates */
1551 nfs_wcc_update_inode(inode, fattr);
1553 /* Check if our cached file size is stale */
1554 new_isize = nfs_size_to_loff_t(fattr->size);
1555 cur_isize = i_size_read(inode);
1556 if (new_isize != cur_isize) {
1557 /* Do we perhaps have any outstanding writes? */
1558 if (nfsi->npages == 0) {
1559 /* No, but did we race with nfs_end_data_update()? */
1561 inode->i_size = new_isize;
1562 invalid |= NFS_INO_INVALID_DATA;
1564 invalid |= NFS_INO_INVALID_ATTR;
1565 } else if (new_isize > cur_isize) {
1566 inode->i_size = new_isize;
1567 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1569 nfsi->cache_change_attribute = jiffies;
1570 dprintk("NFS: isize change on server for file %s/%ld\n",
1571 inode->i_sb->s_id, inode->i_ino);
1574 /* Check if the mtime agrees */
1575 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1576 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1577 dprintk("NFS: mtime change on server for file %s/%ld\n",
1578 inode->i_sb->s_id, inode->i_ino);
1579 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1580 nfsi->cache_change_attribute = jiffies;
1583 /* If ctime has changed we should definitely clear access+acl caches */
1584 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1585 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1586 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1587 nfsi->cache_change_attribute = jiffies;
1589 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1591 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1592 inode->i_uid != fattr->uid ||
1593 inode->i_gid != fattr->gid)
1594 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1596 inode->i_mode = fattr->mode;
1597 inode->i_nlink = fattr->nlink;
1598 inode->i_uid = fattr->uid;
1599 inode->i_gid = fattr->gid;
1601 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1603 * report the blocks in 512byte units
1605 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1606 inode->i_blksize = inode->i_sb->s_blocksize;
1608 inode->i_blocks = fattr->du.nfs2.blocks;
1609 inode->i_blksize = fattr->du.nfs2.blocksize;
1612 if ((fattr->valid & NFS_ATTR_FATTR_V4)) {
1613 if (nfsi->change_attr != fattr->change_attr) {
1614 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1615 inode->i_sb->s_id, inode->i_ino);
1616 nfsi->change_attr = fattr->change_attr;
1617 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1618 nfsi->cache_change_attribute = jiffies;
1620 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA);
1623 /* Update attrtimeo value if we're out of the unstable period */
1624 if (invalid & NFS_INO_INVALID_ATTR) {
1625 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1626 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1627 nfsi->attrtimeo_timestamp = jiffies;
1628 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1629 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1630 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1631 nfsi->attrtimeo_timestamp = jiffies;
1633 /* Don't invalidate the data if we were to blame */
1634 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1635 || S_ISLNK(inode->i_mode)))
1636 invalid &= ~NFS_INO_INVALID_DATA;
1638 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1639 if (!nfs_have_delegation(inode, FMODE_READ))
1640 nfsi->cache_validity |= invalid;
1645 * Big trouble! The inode has become a different object.
1648 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1649 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1653 * No need to worry about unhashing the dentry, as the
1654 * lookup validation will know that the inode is bad.
1655 * (But we fall through to invalidate the caches.)
1657 nfs_invalidate_inode(inode);
1661 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1662 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1663 NFS_SERVER(inode)->hostname, inode->i_sb->s_id,
1664 (long long)nfsi->fileid, (long long)fattr->fileid);
1669 * File system information
1672 static int nfs_set_super(struct super_block *s, void *data)
1674 s->s_fs_info = data;
1675 return set_anon_super(s, data);
1678 static int nfs_compare_super(struct super_block *sb, void *data)
1680 struct nfs_server *server = data;
1681 struct nfs_server *old = NFS_SB(sb);
1683 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1685 if (old->addr.sin_port != server->addr.sin_port)
1687 return !nfs_compare_fh(&old->fh, &server->fh);
1690 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1691 int flags, const char *dev_name, void *raw_data)
1694 struct nfs_server *server = NULL;
1695 struct super_block *s;
1696 struct nfs_fh *root;
1697 struct nfs_mount_data *data = raw_data;
1699 s = ERR_PTR(-EINVAL);
1701 dprintk("%s: missing data argument\n", __FUNCTION__);
1704 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1705 dprintk("%s: bad mount version\n", __FUNCTION__);
1708 switch (data->version) {
1714 if (data->flags & NFS_MOUNT_VER3) {
1715 dprintk("%s: mount structure version %d does not support NFSv3\n",
1720 data->root.size = NFS2_FHSIZE;
1721 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1723 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1724 dprintk("%s: mount structure version %d does not support strong security\n",
1730 memset(data->context, 0, sizeof(data->context));
1732 #ifndef CONFIG_NFS_V3
1733 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1734 s = ERR_PTR(-EPROTONOSUPPORT);
1735 if (data->flags & NFS_MOUNT_VER3) {
1736 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1739 #endif /* CONFIG_NFS_V3 */
1741 s = ERR_PTR(-ENOMEM);
1742 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
1745 /* Zero out the NFS state stuff */
1746 init_nfsv4_state(server);
1747 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1750 if (data->flags & NFS_MOUNT_VER3)
1751 root->size = data->root.size;
1753 root->size = NFS2_FHSIZE;
1754 s = ERR_PTR(-EINVAL);
1755 if (root->size > sizeof(root->data)) {
1756 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1759 memcpy(root->data, data->root.data, root->size);
1761 /* We now require that the mount process passes the remote address */
1762 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1763 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1764 dprintk("%s: mount program didn't pass remote address!\n",
1769 /* Fire up rpciod if not yet running */
1770 s = ERR_PTR(rpciod_up());
1772 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1773 __FUNCTION__, PTR_ERR(s));
1777 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1778 if (IS_ERR(s) || s->s_root)
1779 goto out_rpciod_down;
1783 error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
1785 up_write(&s->s_umount);
1786 deactivate_super(s);
1787 return ERR_PTR(error);
1789 s->s_flags |= MS_ACTIVE;
1798 static void nfs_kill_super(struct super_block *s)
1800 struct nfs_server *server = NFS_SB(s);
1804 if (!IS_ERR(server->client))
1805 rpc_shutdown_client(server->client);
1806 if (!IS_ERR(server->client_sys))
1807 rpc_shutdown_client(server->client_sys);
1808 if (!IS_ERR(server->client_acl))
1809 rpc_shutdown_client(server->client_acl);
1811 if (!(server->flags & NFS_MOUNT_NONLM))
1812 lockd_down(); /* release rpc.lockd */
1814 rpciod_down(); /* release rpciod */
1816 kfree(server->hostname);
1820 static struct file_system_type nfs_fs_type = {
1821 .owner = THIS_MODULE,
1823 .get_sb = nfs_get_sb,
1824 .kill_sb = nfs_kill_super,
1825 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1828 #ifdef CONFIG_NFS_V4
1830 static void nfs4_clear_inode(struct inode *);
1833 static struct super_operations nfs4_sops = {
1834 .alloc_inode = nfs_alloc_inode,
1835 .destroy_inode = nfs_destroy_inode,
1836 .write_inode = nfs_write_inode,
1837 .delete_inode = nfs_delete_inode,
1838 .statfs = nfs_statfs,
1839 .clear_inode = nfs4_clear_inode,
1840 .umount_begin = nfs_umount_begin,
1841 .show_options = nfs_show_options,
1842 .show_stats = nfs_show_stats,
1846 * Clean out any remaining NFSv4 state that might be left over due
1847 * to open() calls that passed nfs_atomic_lookup, but failed to call
1850 static void nfs4_clear_inode(struct inode *inode)
1852 struct nfs_inode *nfsi = NFS_I(inode);
1854 /* If we are holding a delegation, return it! */
1855 nfs_inode_return_delegation(inode);
1856 /* First call standard NFS clear_inode() code */
1857 nfs_clear_inode(inode);
1858 /* Now clear out any remaining state */
1859 while (!list_empty(&nfsi->open_states)) {
1860 struct nfs4_state *state;
1862 state = list_entry(nfsi->open_states.next,
1865 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1868 (long long)NFS_FILEID(inode),
1870 BUG_ON(atomic_read(&state->count) != 1);
1871 nfs4_close_state(state, state->state);
1876 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
1878 struct nfs_server *server;
1879 struct nfs4_client *clp = NULL;
1880 struct rpc_xprt *xprt = NULL;
1881 struct rpc_clnt *clnt = NULL;
1882 struct rpc_timeout timeparms;
1883 rpc_authflavor_t authflavour;
1886 sb->s_blocksize_bits = 0;
1887 sb->s_blocksize = 0;
1888 server = NFS_SB(sb);
1889 if (data->rsize != 0)
1890 server->rsize = nfs_block_size(data->rsize, NULL);
1891 if (data->wsize != 0)
1892 server->wsize = nfs_block_size(data->wsize, NULL);
1893 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
1894 server->caps = NFS_CAP_ATOMIC_OPEN;
1896 server->acregmin = data->acregmin*HZ;
1897 server->acregmax = data->acregmax*HZ;
1898 server->acdirmin = data->acdirmin*HZ;
1899 server->acdirmax = data->acdirmax*HZ;
1901 server->rpc_ops = &nfs_v4_clientops;
1903 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
1905 server->retrans_timeo = timeparms.to_initval;
1906 server->retrans_count = timeparms.to_retries;
1908 clp = nfs4_get_client(&server->addr.sin_addr);
1910 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
1914 /* Now create transport and client */
1915 authflavour = RPC_AUTH_UNIX;
1916 if (data->auth_flavourlen != 0) {
1917 if (data->auth_flavourlen != 1) {
1918 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
1919 __FUNCTION__, data->auth_flavourlen);
1923 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
1929 down_write(&clp->cl_sem);
1930 if (IS_ERR(clp->cl_rpcclient)) {
1931 xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
1933 up_write(&clp->cl_sem);
1934 err = PTR_ERR(xprt);
1935 dprintk("%s: cannot create RPC transport. Error = %d\n",
1939 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
1940 server->rpc_ops->version, authflavour);
1942 up_write(&clp->cl_sem);
1943 err = PTR_ERR(clnt);
1944 dprintk("%s: cannot create RPC client. Error = %d\n",
1949 clnt->cl_softrtry = 1;
1950 clp->cl_rpcclient = clnt;
1951 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
1954 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
1955 clnt = rpc_clone_client(clp->cl_rpcclient);
1957 server->nfs4_state = clp;
1958 up_write(&clp->cl_sem);
1962 err = PTR_ERR(clnt);
1963 dprintk("%s: cannot create RPC client. Error = %d\n",
1968 server->client = clnt;
1970 if (server->nfs4_state->cl_idmap == NULL) {
1971 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
1975 if (clnt->cl_auth->au_flavor != authflavour) {
1976 struct rpc_auth *auth;
1978 auth = rpcauth_create(authflavour, clnt);
1980 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
1981 return PTR_ERR(auth);
1985 sb->s_time_gran = 1;
1987 sb->s_op = &nfs4_sops;
1988 err = nfs_sb_init(sb, authflavour);
1993 nfs4_put_client(clp);
1997 static int nfs4_compare_super(struct super_block *sb, void *data)
1999 struct nfs_server *server = data;
2000 struct nfs_server *old = NFS_SB(sb);
2002 if (strcmp(server->hostname, old->hostname) != 0)
2004 if (strcmp(server->mnt_path, old->mnt_path) != 0)
2010 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
2015 return ERR_PTR(-EINVAL);
2016 if (src->len < maxlen)
2019 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
2021 return ERR_PTR(-ENOMEM);
2023 if (copy_from_user(dst, src->data, maxlen)) {
2025 return ERR_PTR(-EFAULT);
2031 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
2032 int flags, const char *dev_name, void *raw_data)
2035 struct nfs_server *server;
2036 struct super_block *s;
2037 struct nfs4_mount_data *data = raw_data;
2041 dprintk("%s: missing data argument\n", __FUNCTION__);
2042 return ERR_PTR(-EINVAL);
2044 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
2045 dprintk("%s: bad mount version\n", __FUNCTION__);
2046 return ERR_PTR(-EINVAL);
2049 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
2051 return ERR_PTR(-ENOMEM);
2052 /* Zero out the NFS state stuff */
2053 init_nfsv4_state(server);
2054 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
2056 p = nfs_copy_user_string(NULL, &data->hostname, 256);
2059 server->hostname = p;
2061 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
2064 server->mnt_path = p;
2066 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
2067 sizeof(server->ip_addr) - 1);
2071 /* We now require that the mount process passes the remote address */
2072 if (data->host_addrlen != sizeof(server->addr)) {
2073 s = ERR_PTR(-EINVAL);
2076 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
2077 s = ERR_PTR(-EFAULT);
2080 if (server->addr.sin_family != AF_INET ||
2081 server->addr.sin_addr.s_addr == INADDR_ANY) {
2082 dprintk("%s: mount program didn't pass remote IP address!\n",
2084 s = ERR_PTR(-EINVAL);
2088 /* Fire up rpciod if not yet running */
2089 s = ERR_PTR(rpciod_up());
2091 dprintk("%s: couldn't start rpciod! Error = %ld\n",
2092 __FUNCTION__, PTR_ERR(s));
2096 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
2098 if (IS_ERR(s) || s->s_root)
2103 error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
2105 up_write(&s->s_umount);
2106 deactivate_super(s);
2107 return ERR_PTR(error);
2109 s->s_flags |= MS_ACTIVE;
2112 s = (struct super_block *)p;
2114 kfree(server->mnt_path);
2115 kfree(server->hostname);
2116 nfs_free_iostats(server->io_stats);
2121 static void nfs4_kill_super(struct super_block *sb)
2123 struct nfs_server *server = NFS_SB(sb);
2125 nfs_return_all_delegations(sb);
2126 kill_anon_super(sb);
2128 nfs4_renewd_prepare_shutdown(server);
2130 if (server->client != NULL && !IS_ERR(server->client))
2131 rpc_shutdown_client(server->client);
2133 destroy_nfsv4_state(server);
2137 kfree(server->hostname);
2141 static struct file_system_type nfs4_fs_type = {
2142 .owner = THIS_MODULE,
2144 .get_sb = nfs4_get_sb,
2145 .kill_sb = nfs4_kill_super,
2146 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2149 static const int nfs_set_port_min = 0;
2150 static const int nfs_set_port_max = 65535;
2151 static int param_set_port(const char *val, struct kernel_param *kp)
2154 int num = simple_strtol(val, &endp, 0);
2155 if (endp == val || *endp || num < nfs_set_port_min || num > nfs_set_port_max)
2157 *((int *)kp->arg) = num;
2161 module_param_call(callback_tcpport, param_set_port, param_get_int,
2162 &nfs_callback_set_tcpport, 0644);
2164 static int param_set_idmap_timeout(const char *val, struct kernel_param *kp)
2167 int num = simple_strtol(val, &endp, 0);
2169 if (endp == val || *endp || num < 0 || jif < num)
2171 *((int *)kp->arg) = jif;
2175 module_param_call(idmap_cache_timeout, param_set_idmap_timeout, param_get_int,
2176 &nfs_idmap_cache_timeout, 0644);
2178 #define nfs4_init_once(nfsi) \
2180 INIT_LIST_HEAD(&(nfsi)->open_states); \
2181 nfsi->delegation = NULL; \
2182 nfsi->delegation_state = 0; \
2183 init_rwsem(&nfsi->rwsem); \
2186 static inline int register_nfs4fs(void)
2190 ret = nfs_register_sysctl();
2193 ret = register_filesystem(&nfs4_fs_type);
2195 nfs_unregister_sysctl();
2199 static inline void unregister_nfs4fs(void)
2201 unregister_filesystem(&nfs4_fs_type);
2202 nfs_unregister_sysctl();
2205 #define nfs4_init_once(nfsi) \
2207 #define register_nfs4fs() (0)
2208 #define unregister_nfs4fs()
2211 extern int nfs_init_nfspagecache(void);
2212 extern void nfs_destroy_nfspagecache(void);
2213 extern int nfs_init_readpagecache(void);
2214 extern void nfs_destroy_readpagecache(void);
2215 extern int nfs_init_writepagecache(void);
2216 extern void nfs_destroy_writepagecache(void);
2217 #ifdef CONFIG_NFS_DIRECTIO
2218 extern int nfs_init_directcache(void);
2219 extern void nfs_destroy_directcache(void);
2222 static kmem_cache_t * nfs_inode_cachep;
2224 static struct inode *nfs_alloc_inode(struct super_block *sb)
2226 struct nfs_inode *nfsi;
2227 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2231 nfsi->cache_validity = 0UL;
2232 nfsi->cache_change_attribute = jiffies;
2233 #ifdef CONFIG_NFS_V3_ACL
2234 nfsi->acl_access = ERR_PTR(-EAGAIN);
2235 nfsi->acl_default = ERR_PTR(-EAGAIN);
2237 #ifdef CONFIG_NFS_V4
2238 nfsi->nfs4_acl = NULL;
2239 #endif /* CONFIG_NFS_V4 */
2240 return &nfsi->vfs_inode;
2243 static void nfs_destroy_inode(struct inode *inode)
2245 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2248 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2250 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2252 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2253 SLAB_CTOR_CONSTRUCTOR) {
2254 inode_init_once(&nfsi->vfs_inode);
2255 spin_lock_init(&nfsi->req_lock);
2256 INIT_LIST_HEAD(&nfsi->dirty);
2257 INIT_LIST_HEAD(&nfsi->commit);
2258 INIT_LIST_HEAD(&nfsi->open_files);
2259 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2260 atomic_set(&nfsi->data_updates, 0);
2264 nfs4_init_once(nfsi);
2268 static int nfs_init_inodecache(void)
2270 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2271 sizeof(struct nfs_inode),
2272 0, SLAB_RECLAIM_ACCOUNT,
2274 if (nfs_inode_cachep == NULL)
2280 static void nfs_destroy_inodecache(void)
2282 if (kmem_cache_destroy(nfs_inode_cachep))
2283 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2289 static int __init init_nfs_fs(void)
2293 err = nfs_init_nfspagecache();
2297 err = nfs_init_inodecache();
2301 err = nfs_init_readpagecache();
2305 err = nfs_init_writepagecache();
2309 #ifdef CONFIG_NFS_DIRECTIO
2310 err = nfs_init_directcache();
2315 #ifdef CONFIG_PROC_FS
2316 rpc_proc_register(&nfs_rpcstat);
2318 err = register_filesystem(&nfs_fs_type);
2321 if ((err = register_nfs4fs()) != 0)
2325 #ifdef CONFIG_PROC_FS
2326 rpc_proc_unregister("nfs");
2328 #ifdef CONFIG_NFS_DIRECTIO
2329 nfs_destroy_directcache();
2332 nfs_destroy_writepagecache();
2334 nfs_destroy_readpagecache();
2336 nfs_destroy_inodecache();
2338 nfs_destroy_nfspagecache();
2343 static void __exit exit_nfs_fs(void)
2345 #ifdef CONFIG_NFS_DIRECTIO
2346 nfs_destroy_directcache();
2348 nfs_destroy_writepagecache();
2349 nfs_destroy_readpagecache();
2350 nfs_destroy_inodecache();
2351 nfs_destroy_nfspagecache();
2352 #ifdef CONFIG_PROC_FS
2353 rpc_proc_unregister("nfs");
2355 unregister_filesystem(&nfs_fs_type);
2356 unregister_nfs4fs();
2359 /* Not quite true; I just maintain it */
2360 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2361 MODULE_LICENSE("GPL");
2363 module_init(init_nfs_fs)
2364 module_exit(exit_nfs_fs)