4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/config.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/sunrpc/metrics.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/nfs4_mount.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/smp_lock.h>
35 #include <linux/seq_file.h>
36 #include <linux/mount.h>
37 #include <linux/nfs_idmap.h>
38 #include <linux/vfs.h>
39 #include <linux/inet.h>
40 #include <linux/nfs_xdr.h>
42 #include <asm/system.h>
43 #include <asm/uaccess.h>
47 #include "delegation.h"
50 #define NFSDBG_FACILITY NFSDBG_VFS
51 #define NFS_PARANOIA 1
53 /* Maximum number of readahead requests
54 * FIXME: this should really be a sysctl so that users may tune it to suit
55 * their needs. People that do NFS over a slow network, might for
56 * instance want to reduce it to something closer to 1 for improved
57 * interactive response.
59 #define NFS_MAX_READAHEAD (RPC_DEF_SLOT_TABLE - 1)
61 static void nfs_invalidate_inode(struct inode *);
62 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
64 static struct inode *nfs_alloc_inode(struct super_block *sb);
65 static void nfs_destroy_inode(struct inode *);
66 static int nfs_write_inode(struct inode *,int);
67 static void nfs_clear_inode(struct inode *);
68 static void nfs_umount_begin(struct vfsmount *, int);
69 static int nfs_statfs(struct super_block *, struct kstatfs *);
70 static int nfs_show_options(struct seq_file *, struct vfsmount *);
71 static int nfs_show_stats(struct seq_file *, struct vfsmount *);
72 static void nfs_zap_acl_cache(struct inode *);
74 static struct rpc_program nfs_program;
76 static struct super_operations nfs_sops = {
77 .alloc_inode = nfs_alloc_inode,
78 .destroy_inode = nfs_destroy_inode,
79 .write_inode = nfs_write_inode,
81 .clear_inode = nfs_clear_inode,
82 .umount_begin = nfs_umount_begin,
83 .show_options = nfs_show_options,
84 .show_stats = nfs_show_stats,
90 static struct rpc_stat nfs_rpcstat = {
91 .program = &nfs_program
93 static struct rpc_version * nfs_version[] = {
97 #if defined(CONFIG_NFS_V3)
99 #elif defined(CONFIG_NFS_V4)
102 #if defined(CONFIG_NFS_V4)
107 static struct rpc_program nfs_program = {
109 .number = NFS_PROGRAM,
110 .nrvers = ARRAY_SIZE(nfs_version),
111 .version = nfs_version,
112 .stats = &nfs_rpcstat,
113 .pipe_dir_name = "/nfs",
116 #ifdef CONFIG_NFS_V3_ACL
117 static struct rpc_stat nfsacl_rpcstat = { &nfsacl_program };
118 static struct rpc_version * nfsacl_version[] = {
119 [3] = &nfsacl_version3,
122 struct rpc_program nfsacl_program = {
124 .number = NFS_ACL_PROGRAM,
125 .nrvers = ARRAY_SIZE(nfsacl_version),
126 .version = nfsacl_version,
127 .stats = &nfsacl_rpcstat,
129 #endif /* CONFIG_NFS_V3_ACL */
131 static inline unsigned long
132 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
134 return nfs_fileid_to_ino_t(fattr->fileid);
138 nfs_write_inode(struct inode *inode, int sync)
140 int flags = sync ? FLUSH_SYNC : 0;
143 ret = nfs_commit_inode(inode, flags);
150 nfs_clear_inode(struct inode *inode)
152 struct nfs_inode *nfsi = NFS_I(inode);
153 struct rpc_cred *cred;
156 * The following should never happen...
158 BUG_ON(nfs_have_writebacks(inode));
159 BUG_ON (!list_empty(&nfsi->open_files));
160 nfs_zap_acl_cache(inode);
161 cred = nfsi->cache_access.cred;
164 BUG_ON(atomic_read(&nfsi->data_updates) != 0);
167 static void nfs_umount_begin(struct vfsmount *vfsmnt, int flags)
169 struct nfs_server *server;
170 struct rpc_clnt *rpc;
172 shrink_submounts(vfsmnt, &nfs_automount_list);
173 if (!(flags & MNT_FORCE))
175 /* -EIO all pending I/O */
176 server = NFS_SB(vfsmnt->mnt_sb);
177 rpc = server->client;
179 rpc_killall_tasks(rpc);
180 rpc = server->client_acl;
182 rpc_killall_tasks(rpc);
186 static inline unsigned long
187 nfs_block_bits(unsigned long bsize, unsigned char *nrbitsp)
189 /* make sure blocksize is a power of two */
190 if ((bsize & (bsize - 1)) || nrbitsp) {
191 unsigned char nrbits;
193 for (nrbits = 31; nrbits && !(bsize & (1 << nrbits)); nrbits--)
204 * Calculate the number of 512byte blocks used.
206 static inline unsigned long
207 nfs_calc_block_size(u64 tsize)
209 loff_t used = (tsize + 511) >> 9;
210 return (used > ULONG_MAX) ? ULONG_MAX : used;
214 * Compute and set NFS server blocksize
216 static inline unsigned long
217 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
219 if (bsize < NFS_MIN_FILE_IO_SIZE)
220 bsize = NFS_DEF_FILE_IO_SIZE;
221 else if (bsize >= NFS_MAX_FILE_IO_SIZE)
222 bsize = NFS_MAX_FILE_IO_SIZE;
224 return nfs_block_bits(bsize, nrbitsp);
228 nfs_super_set_maxbytes(struct super_block *sb, __u64 maxfilesize)
230 sb->s_maxbytes = (loff_t)maxfilesize;
231 if (sb->s_maxbytes > MAX_LFS_FILESIZE || sb->s_maxbytes <= 0)
232 sb->s_maxbytes = MAX_LFS_FILESIZE;
236 * Obtain the root inode of the file system.
238 static struct inode *
239 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
241 struct nfs_server *server = NFS_SB(sb);
244 error = server->rpc_ops->getroot(server, rootfh, fsinfo);
246 dprintk("nfs_get_root: getattr error = %d\n", -error);
247 return ERR_PTR(error);
250 server->fsid = fsinfo->fattr->fsid;
251 return nfs_fhget(sb, rootfh, fsinfo->fattr);
255 * Do NFS version-independent mount processing, and sanity checking
258 nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
260 struct nfs_server *server;
261 struct inode *root_inode;
262 struct nfs_fattr fattr;
263 struct nfs_fsinfo fsinfo = {
266 struct nfs_pathconf pathinfo = {
269 int no_root_error = 0;
270 unsigned long max_rpc_payload;
272 /* We probably want something more informative here */
273 snprintf(sb->s_id, sizeof(sb->s_id), "%x:%x", MAJOR(sb->s_dev), MINOR(sb->s_dev));
277 sb->s_magic = NFS_SUPER_MAGIC;
279 server->io_stats = nfs_alloc_iostats();
280 if (server->io_stats == NULL)
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 /* mount time stamp, in seconds */
297 server->mount_time = jiffies;
299 /* Get some general file system info */
300 if (server->namelen == 0 &&
301 server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
302 server->namelen = pathinfo.max_namelen;
303 /* Work out a lot of parameters */
304 if (server->rsize == 0)
305 server->rsize = nfs_block_size(fsinfo.rtpref, NULL);
306 if (server->wsize == 0)
307 server->wsize = nfs_block_size(fsinfo.wtpref, NULL);
309 if (fsinfo.rtmax >= 512 && server->rsize > fsinfo.rtmax)
310 server->rsize = nfs_block_size(fsinfo.rtmax, NULL);
311 if (fsinfo.wtmax >= 512 && server->wsize > fsinfo.wtmax)
312 server->wsize = nfs_block_size(fsinfo.wtmax, NULL);
314 max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
315 if (server->rsize > max_rpc_payload)
316 server->rsize = max_rpc_payload;
317 if (server->rsize > NFS_MAX_FILE_IO_SIZE)
318 server->rsize = NFS_MAX_FILE_IO_SIZE;
319 server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
321 if (server->wsize > max_rpc_payload)
322 server->wsize = max_rpc_payload;
323 if (server->wsize > NFS_MAX_FILE_IO_SIZE)
324 server->wsize = NFS_MAX_FILE_IO_SIZE;
325 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
327 if (sb->s_blocksize == 0)
328 sb->s_blocksize = nfs_block_bits(server->wsize,
329 &sb->s_blocksize_bits);
330 server->wtmult = nfs_block_bits(fsinfo.wtmult, NULL);
332 server->dtsize = nfs_block_size(fsinfo.dtpref, NULL);
333 if (server->dtsize > PAGE_CACHE_SIZE)
334 server->dtsize = PAGE_CACHE_SIZE;
335 if (server->dtsize > server->rsize)
336 server->dtsize = server->rsize;
338 if (server->flags & NFS_MOUNT_NOAC) {
339 server->acregmin = server->acregmax = 0;
340 server->acdirmin = server->acdirmax = 0;
341 sb->s_flags |= MS_SYNCHRONOUS;
343 server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
345 nfs_super_set_maxbytes(sb, fsinfo.maxfilesize);
347 server->client->cl_intr = (server->flags & NFS_MOUNT_INTR) ? 1 : 0;
348 server->client->cl_softrtry = (server->flags & NFS_MOUNT_SOFT) ? 1 : 0;
350 /* We're airborne Set socket buffersize */
351 rpc_setbufsize(server->client, server->wsize + 100, server->rsize + 100);
353 /* Yargs. It didn't work out. */
355 dprintk("nfs_sb_init: get root inode failed: errno %d\n", -no_root_error);
356 if (!IS_ERR(root_inode))
358 return no_root_error;
361 static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
363 to->to_initval = timeo * HZ / 10;
364 to->to_retries = retrans;
371 to->to_initval = 60 * HZ;
372 if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
373 to->to_initval = NFS_MAX_TCP_TIMEOUT;
374 to->to_increment = to->to_initval;
375 to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
376 to->to_exponential = 0;
381 to->to_initval = 11 * HZ / 10;
382 if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
383 to->to_initval = NFS_MAX_UDP_TIMEOUT;
384 to->to_maxval = NFS_MAX_UDP_TIMEOUT;
385 to->to_exponential = 1;
391 * Create an RPC client handle.
393 static struct rpc_clnt *
394 nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
396 struct rpc_timeout timeparms;
397 struct rpc_xprt *xprt = NULL;
398 struct rpc_clnt *clnt = NULL;
399 int proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
401 nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
403 server->retrans_timeo = timeparms.to_initval;
404 server->retrans_count = timeparms.to_retries;
406 /* create transport and client */
407 xprt = xprt_create_proto(proto, &server->addr, &timeparms);
409 dprintk("%s: cannot create RPC transport. Error = %ld\n",
410 __FUNCTION__, PTR_ERR(xprt));
411 return (struct rpc_clnt *)xprt;
413 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
414 server->rpc_ops->version, data->pseudoflavor);
416 dprintk("%s: cannot create RPC client. Error = %ld\n",
417 __FUNCTION__, PTR_ERR(xprt));
422 clnt->cl_softrtry = 1;
431 * The way this works is that the mount process passes a structure
432 * in the data argument which contains the server's IP address
433 * and the root file handle obtained from the server's mount
434 * daemon. We stash these away in the private superblock fields.
437 nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
439 struct nfs_server *server;
440 rpc_authflavor_t authflavor;
443 sb->s_blocksize_bits = 0;
446 sb->s_blocksize = nfs_block_size(data->bsize, &sb->s_blocksize_bits);
448 server->rsize = nfs_block_size(data->rsize, NULL);
450 server->wsize = nfs_block_size(data->wsize, NULL);
451 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
453 server->acregmin = data->acregmin*HZ;
454 server->acregmax = data->acregmax*HZ;
455 server->acdirmin = data->acdirmin*HZ;
456 server->acdirmax = data->acdirmax*HZ;
458 /* Start lockd here, before we might error out */
459 if (!(server->flags & NFS_MOUNT_NONLM))
462 server->namelen = data->namlen;
463 server->hostname = kmalloc(strlen(data->hostname) + 1, GFP_KERNEL);
464 if (!server->hostname)
466 strcpy(server->hostname, data->hostname);
468 /* Check NFS protocol revision and initialize RPC op vector
469 * and file handle pool. */
471 if (server->flags & NFS_MOUNT_VER3) {
472 server->rpc_ops = &nfs_v3_clientops;
473 server->caps |= NFS_CAP_READDIRPLUS;
475 server->rpc_ops = &nfs_v2_clientops;
478 server->rpc_ops = &nfs_v2_clientops;
481 /* Fill in pseudoflavor for mount version < 5 */
482 if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
483 data->pseudoflavor = RPC_AUTH_UNIX;
484 authflavor = data->pseudoflavor; /* save for sb_init() */
485 /* XXX maybe we want to add a server->pseudoflavor field */
487 /* Create RPC client handles */
488 server->client = nfs_create_client(server, data);
489 if (IS_ERR(server->client))
490 return PTR_ERR(server->client);
491 /* RFC 2623, sec 2.3.2 */
492 if (authflavor != RPC_AUTH_UNIX) {
493 struct rpc_auth *auth;
495 server->client_sys = rpc_clone_client(server->client);
496 if (IS_ERR(server->client_sys))
497 return PTR_ERR(server->client_sys);
498 auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
500 return PTR_ERR(auth);
502 atomic_inc(&server->client->cl_count);
503 server->client_sys = server->client;
505 if (server->flags & NFS_MOUNT_VER3) {
506 #ifdef CONFIG_NFS_V3_ACL
507 if (!(server->flags & NFS_MOUNT_NOACL)) {
508 server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
509 /* No errors! Assume that Sun nfsacls are supported */
510 if (!IS_ERR(server->client_acl))
511 server->caps |= NFS_CAP_ACLS;
514 server->flags &= ~NFS_MOUNT_NOACL;
515 #endif /* CONFIG_NFS_V3_ACL */
517 * The VFS shouldn't apply the umask to mode bits. We will
518 * do so ourselves when necessary.
520 sb->s_flags |= MS_POSIXACL;
521 if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
522 server->namelen = NFS3_MAXNAMLEN;
525 if (server->namelen == 0 || server->namelen > NFS2_MAXNAMLEN)
526 server->namelen = NFS2_MAXNAMLEN;
529 sb->s_op = &nfs_sops;
530 return nfs_sb_init(sb, authflavor);
534 nfs_statfs(struct super_block *sb, struct kstatfs *buf)
536 struct nfs_server *server = NFS_SB(sb);
537 unsigned char blockbits;
538 unsigned long blockres;
539 struct nfs_fh *rootfh = NFS_FH(sb->s_root->d_inode);
540 struct nfs_fattr fattr;
541 struct nfs_fsstat res = {
548 error = server->rpc_ops->statfs(server, rootfh, &res);
549 buf->f_type = NFS_SUPER_MAGIC;
554 * Current versions of glibc do not correctly handle the
555 * case where f_frsize != f_bsize. Eventually we want to
556 * report the value of wtmult in this field.
558 buf->f_frsize = sb->s_blocksize;
561 * On most *nix systems, f_blocks, f_bfree, and f_bavail
562 * are reported in units of f_frsize. Linux hasn't had
563 * an f_frsize field in its statfs struct until recently,
564 * thus historically Linux's sys_statfs reports these
565 * fields in units of f_bsize.
567 buf->f_bsize = sb->s_blocksize;
568 blockbits = sb->s_blocksize_bits;
569 blockres = (1 << blockbits) - 1;
570 buf->f_blocks = (res.tbytes + blockres) >> blockbits;
571 buf->f_bfree = (res.fbytes + blockres) >> blockbits;
572 buf->f_bavail = (res.abytes + blockres) >> blockbits;
574 buf->f_files = res.tfiles;
575 buf->f_ffree = res.afiles;
577 buf->f_namelen = server->namelen;
583 dprintk("%s: statfs error = %d\n", __FUNCTION__, -error);
584 buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
589 static void nfs_show_mount_options(struct seq_file *m, struct nfs_server *nfss, int showdefaults)
591 static struct proc_nfs_info {
596 { NFS_MOUNT_SOFT, ",soft", ",hard" },
597 { NFS_MOUNT_INTR, ",intr", "" },
598 { NFS_MOUNT_NOCTO, ",nocto", "" },
599 { NFS_MOUNT_NOAC, ",noac", "" },
600 { NFS_MOUNT_NONLM, ",nolock", "" },
601 { NFS_MOUNT_NOACL, ",noacl", "" },
604 struct proc_nfs_info *nfs_infop;
608 seq_printf(m, ",vers=%d", nfss->rpc_ops->version);
609 seq_printf(m, ",rsize=%d", nfss->rsize);
610 seq_printf(m, ",wsize=%d", nfss->wsize);
611 if (nfss->acregmin != 3*HZ || showdefaults)
612 seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
613 if (nfss->acregmax != 60*HZ || showdefaults)
614 seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
615 if (nfss->acdirmin != 30*HZ || showdefaults)
616 seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
617 if (nfss->acdirmax != 60*HZ || showdefaults)
618 seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
619 for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
620 if (nfss->flags & nfs_infop->flag)
621 seq_puts(m, nfs_infop->str);
623 seq_puts(m, nfs_infop->nostr);
625 switch (nfss->client->cl_xprt->prot) {
633 snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
636 seq_printf(m, ",proto=%s", proto);
637 seq_printf(m, ",timeo=%lu", 10U * nfss->retrans_timeo / HZ);
638 seq_printf(m, ",retrans=%u", nfss->retrans_count);
641 static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
643 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
645 nfs_show_mount_options(m, nfss, 0);
647 seq_puts(m, ",addr=");
648 seq_escape(m, nfss->hostname, " \t\n\\");
653 static int nfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
656 struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
657 struct rpc_auth *auth = nfss->client->cl_auth;
658 struct nfs_iostats totals = { };
660 seq_printf(m, "statvers=%s", NFS_IOSTAT_VERS);
663 * Display all mount option settings
665 seq_printf(m, "\n\topts:\t");
666 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? "ro" : "rw");
667 seq_puts(m, mnt->mnt_sb->s_flags & MS_SYNCHRONOUS ? ",sync" : "");
668 seq_puts(m, mnt->mnt_sb->s_flags & MS_NOATIME ? ",noatime" : "");
669 seq_puts(m, mnt->mnt_sb->s_flags & MS_NODIRATIME ? ",nodiratime" : "");
670 nfs_show_mount_options(m, nfss, 1);
672 seq_printf(m, "\n\tage:\t%lu", (jiffies - nfss->mount_time) / HZ);
674 seq_printf(m, "\n\tcaps:\t");
675 seq_printf(m, "caps=0x%x", nfss->caps);
676 seq_printf(m, ",wtmult=%d", nfss->wtmult);
677 seq_printf(m, ",dtsize=%d", nfss->dtsize);
678 seq_printf(m, ",bsize=%d", nfss->bsize);
679 seq_printf(m, ",namelen=%d", nfss->namelen);
682 if (nfss->rpc_ops->version == 4) {
683 seq_printf(m, "\n\tnfsv4:\t");
684 seq_printf(m, "bm0=0x%x", nfss->attr_bitmask[0]);
685 seq_printf(m, ",bm1=0x%x", nfss->attr_bitmask[1]);
686 seq_printf(m, ",acl=0x%x", nfss->acl_bitmask);
691 * Display security flavor in effect for this mount
693 seq_printf(m, "\n\tsec:\tflavor=%d", auth->au_ops->au_flavor);
695 seq_printf(m, ",pseudoflavor=%d", auth->au_flavor);
698 * Display superblock I/O counters
700 for_each_possible_cpu(cpu) {
701 struct nfs_iostats *stats;
704 stats = per_cpu_ptr(nfss->io_stats, cpu);
706 for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
707 totals.events[i] += stats->events[i];
708 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
709 totals.bytes[i] += stats->bytes[i];
714 seq_printf(m, "\n\tevents:\t");
715 for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
716 seq_printf(m, "%lu ", totals.events[i]);
717 seq_printf(m, "\n\tbytes:\t");
718 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
719 seq_printf(m, "%Lu ", totals.bytes[i]);
722 rpc_print_iostats(m, nfss->client);
728 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
730 int nfs_sync_mapping(struct address_space *mapping)
734 if (mapping->nrpages == 0)
736 unmap_mapping_range(mapping, 0, 0, 0);
737 ret = filemap_write_and_wait(mapping);
740 ret = nfs_wb_all(mapping->host);
746 * Invalidate the local caches
748 static void nfs_zap_caches_locked(struct inode *inode)
750 struct nfs_inode *nfsi = NFS_I(inode);
751 int mode = inode->i_mode;
753 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
755 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
756 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
758 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
759 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
760 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
762 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
765 void nfs_zap_caches(struct inode *inode)
767 spin_lock(&inode->i_lock);
768 nfs_zap_caches_locked(inode);
769 spin_unlock(&inode->i_lock);
772 static void nfs_zap_acl_cache(struct inode *inode)
774 void (*clear_acl_cache)(struct inode *);
776 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
777 if (clear_acl_cache != NULL)
778 clear_acl_cache(inode);
779 spin_lock(&inode->i_lock);
780 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
781 spin_unlock(&inode->i_lock);
785 * Invalidate, but do not unhash, the inode.
786 * NB: must be called with inode->i_lock held!
788 static void nfs_invalidate_inode(struct inode *inode)
790 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
791 nfs_zap_caches_locked(inode);
794 struct nfs_find_desc {
796 struct nfs_fattr *fattr;
800 * In NFSv3 we can have 64bit inode numbers. In order to support
801 * this, and re-exported directories (also seen in NFSv2)
802 * we are forced to allow 2 different inodes to have the same
806 nfs_find_actor(struct inode *inode, void *opaque)
808 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
809 struct nfs_fh *fh = desc->fh;
810 struct nfs_fattr *fattr = desc->fattr;
812 if (NFS_FILEID(inode) != fattr->fileid)
814 if (nfs_compare_fh(NFS_FH(inode), fh))
816 if (is_bad_inode(inode) || NFS_STALE(inode))
822 nfs_init_locked(struct inode *inode, void *opaque)
824 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
825 struct nfs_fattr *fattr = desc->fattr;
827 NFS_FILEID(inode) = fattr->fileid;
828 nfs_copy_fh(NFS_FH(inode), desc->fh);
832 /* Don't use READDIRPLUS on directories that we believe are too large */
833 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
836 * This is our front-end to iget that looks up inodes by file handle
837 * instead of inode number.
840 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
842 struct nfs_find_desc desc = {
846 struct inode *inode = ERR_PTR(-ENOENT);
849 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
853 printk("NFS: Buggy server - nlink == 0!\n");
857 hash = nfs_fattr_to_ino_t(fattr);
859 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
861 inode = ERR_PTR(-ENOMEM);
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 /* Deal with crossing mountpoints */
890 if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
891 inode->i_op = &nfs_mountpoint_inode_operations;
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 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
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_wait(inode, 0, 0, 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));
1212 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1214 if (!inode || is_bad_inode(inode))
1216 if (NFS_STALE(inode))
1219 status = nfs_wait_on_inode(inode);
1222 if (NFS_STALE(inode)) {
1224 /* Do we trust the cached ESTALE? */
1225 if (NFS_ATTRTIMEO(inode) != 0) {
1226 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME)) {
1233 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
1235 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
1237 (long long)NFS_FILEID(inode), status);
1238 if (status == -ESTALE) {
1239 nfs_zap_caches(inode);
1240 if (!S_ISDIR(inode->i_mode))
1241 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
1246 spin_lock(&inode->i_lock);
1247 status = nfs_update_inode(inode, &fattr);
1249 spin_unlock(&inode->i_lock);
1250 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
1252 (long long)NFS_FILEID(inode), status);
1255 spin_unlock(&inode->i_lock);
1257 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1258 nfs_zap_acl_cache(inode);
1260 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
1262 (long long)NFS_FILEID(inode));
1265 nfs_wake_up_inode(inode);
1272 int nfs_attribute_timeout(struct inode *inode)
1274 struct nfs_inode *nfsi = NFS_I(inode);
1276 if (nfs_have_delegation(inode, FMODE_READ))
1278 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
1282 * nfs_revalidate_inode - Revalidate the inode attributes
1283 * @server - pointer to nfs_server struct
1284 * @inode - pointer to inode struct
1286 * Updates inode attribute information by retrieving the data from the server.
1288 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1290 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
1291 && !nfs_attribute_timeout(inode))
1292 return NFS_STALE(inode) ? -ESTALE : 0;
1293 return __nfs_revalidate_inode(server, inode);
1297 * nfs_revalidate_mapping - Revalidate the pagecache
1298 * @inode - pointer to host inode
1299 * @mapping - pointer to mapping
1301 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1303 struct nfs_inode *nfsi = NFS_I(inode);
1306 if (NFS_STALE(inode))
1308 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
1309 || nfs_attribute_timeout(inode))
1310 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1312 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
1313 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1314 if (S_ISREG(inode->i_mode))
1315 nfs_sync_mapping(mapping);
1316 invalidate_inode_pages2(mapping);
1318 spin_lock(&inode->i_lock);
1319 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1320 if (S_ISDIR(inode->i_mode)) {
1321 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1322 /* This ensures we revalidate child dentries */
1323 nfsi->cache_change_attribute = jiffies;
1325 spin_unlock(&inode->i_lock);
1327 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
1329 (long long)NFS_FILEID(inode));
1335 * nfs_begin_data_update
1336 * @inode - pointer to inode
1337 * Declare that a set of operations will update file data on the server
1339 void nfs_begin_data_update(struct inode *inode)
1341 atomic_inc(&NFS_I(inode)->data_updates);
1345 * nfs_end_data_update
1346 * @inode - pointer to inode
1347 * Declare end of the operations that will update file data
1348 * This will mark the inode as immediately needing revalidation
1349 * of its attribute cache.
1351 void nfs_end_data_update(struct inode *inode)
1353 struct nfs_inode *nfsi = NFS_I(inode);
1355 if (!nfs_have_delegation(inode, FMODE_READ)) {
1356 /* Directories and symlinks: invalidate page cache */
1357 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
1358 spin_lock(&inode->i_lock);
1359 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1360 spin_unlock(&inode->i_lock);
1363 nfsi->cache_change_attribute = jiffies;
1364 atomic_dec(&nfsi->data_updates);
1367 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1369 struct nfs_inode *nfsi = NFS_I(inode);
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 /* Has the inode gone and changed behind our back? */
1405 if (nfsi->fileid != fattr->fileid
1406 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1410 /* Are we in the process of updating data on the server? */
1411 data_unstable = nfs_caches_unstable(inode);
1413 /* Do atomic weak cache consistency updates */
1414 nfs_wcc_update_inode(inode, fattr);
1416 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1417 nfsi->change_attr != fattr->change_attr)
1418 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1420 /* Verify a few of the more important attributes */
1421 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
1422 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1424 cur_size = i_size_read(inode);
1425 new_isize = nfs_size_to_loff_t(fattr->size);
1426 if (cur_size != new_isize && nfsi->npages == 0)
1427 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1429 /* Have any file permissions changed? */
1430 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
1431 || inode->i_uid != fattr->uid
1432 || inode->i_gid != fattr->gid)
1433 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1435 /* Has the link count changed? */
1436 if (inode->i_nlink != fattr->nlink)
1437 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
1439 if (!timespec_equal(&inode->i_atime, &fattr->atime))
1440 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
1442 nfsi->read_cache_jiffies = fattr->time_start;
1447 * nfs_refresh_inode - try to update the inode attribute cache
1448 * @inode - pointer to inode
1449 * @fattr - updated attributes
1451 * Check that an RPC call that returned attributes has not overlapped with
1452 * other recent updates of the inode metadata, then decide whether it is
1453 * safe to do a full update of the inode attributes, or whether just to
1454 * call nfs_check_inode_attributes.
1456 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1458 struct nfs_inode *nfsi = NFS_I(inode);
1461 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1463 spin_lock(&inode->i_lock);
1464 if (time_after(fattr->time_start, nfsi->last_updated))
1465 status = nfs_update_inode(inode, fattr);
1467 status = nfs_check_inode_attributes(inode, fattr);
1469 spin_unlock(&inode->i_lock);
1474 * nfs_post_op_update_inode - try to update the inode attribute cache
1475 * @inode - pointer to inode
1476 * @fattr - updated attributes
1478 * After an operation that has changed the inode metadata, mark the
1479 * attribute cache as being invalid, then try to update it.
1481 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1483 struct nfs_inode *nfsi = NFS_I(inode);
1486 spin_lock(&inode->i_lock);
1487 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
1488 nfsi->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1491 status = nfs_update_inode(inode, fattr);
1493 spin_unlock(&inode->i_lock);
1498 * Many nfs protocol calls return the new file attributes after
1499 * an operation. Here we update the inode to reflect the state
1500 * of the server's inode.
1502 * This is a bit tricky because we have to make sure all dirty pages
1503 * have been sent off to the server before calling invalidate_inode_pages.
1504 * To make sure no other process adds more write requests while we try
1505 * our best to flush them, we make them sleep during the attribute refresh.
1507 * A very similar scenario holds for the dir cache.
1509 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1511 struct nfs_server *server;
1512 struct nfs_inode *nfsi = NFS_I(inode);
1513 loff_t cur_isize, new_isize;
1514 unsigned int invalid = 0;
1517 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1518 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
1519 atomic_read(&inode->i_count), fattr->valid);
1521 if (nfsi->fileid != fattr->fileid)
1525 * Make sure the inode's type hasn't changed.
1527 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1530 server = NFS_SERVER(inode);
1531 /* Update the fsid if and only if this is the root directory */
1532 if (inode == inode->i_sb->s_root->d_inode
1533 && !nfs_fsid_equal(&server->fsid, &fattr->fsid))
1534 server->fsid = fattr->fsid;
1537 * Update the read time so we don't revalidate too often.
1539 nfsi->read_cache_jiffies = fattr->time_start;
1540 nfsi->last_updated = jiffies;
1542 /* Are we racing with known updates of the metadata on the server? */
1543 data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
1545 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATIME);
1547 /* Do atomic weak cache consistency updates */
1548 nfs_wcc_update_inode(inode, fattr);
1550 /* Check if our cached file size is stale */
1551 new_isize = nfs_size_to_loff_t(fattr->size);
1552 cur_isize = i_size_read(inode);
1553 if (new_isize != cur_isize) {
1554 /* Do we perhaps have any outstanding writes? */
1555 if (nfsi->npages == 0) {
1556 /* No, but did we race with nfs_end_data_update()? */
1558 inode->i_size = new_isize;
1559 invalid |= NFS_INO_INVALID_DATA;
1561 invalid |= NFS_INO_INVALID_ATTR;
1562 } else if (new_isize > cur_isize) {
1563 inode->i_size = new_isize;
1564 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1566 nfsi->cache_change_attribute = jiffies;
1567 dprintk("NFS: isize change on server for file %s/%ld\n",
1568 inode->i_sb->s_id, inode->i_ino);
1571 /* Check if the mtime agrees */
1572 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1573 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1574 dprintk("NFS: mtime change on server for file %s/%ld\n",
1575 inode->i_sb->s_id, inode->i_ino);
1576 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1577 nfsi->cache_change_attribute = jiffies;
1580 /* If ctime has changed we should definitely clear access+acl caches */
1581 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1582 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1583 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1584 nfsi->cache_change_attribute = jiffies;
1586 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1588 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1589 inode->i_uid != fattr->uid ||
1590 inode->i_gid != fattr->gid)
1591 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1593 inode->i_mode = fattr->mode;
1594 inode->i_nlink = fattr->nlink;
1595 inode->i_uid = fattr->uid;
1596 inode->i_gid = fattr->gid;
1598 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1600 * report the blocks in 512byte units
1602 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1603 inode->i_blksize = inode->i_sb->s_blocksize;
1605 inode->i_blocks = fattr->du.nfs2.blocks;
1606 inode->i_blksize = fattr->du.nfs2.blocksize;
1609 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1610 nfsi->change_attr != fattr->change_attr) {
1611 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1612 inode->i_sb->s_id, inode->i_ino);
1613 nfsi->change_attr = fattr->change_attr;
1614 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1615 nfsi->cache_change_attribute = jiffies;
1618 /* Update attrtimeo value if we're out of the unstable period */
1619 if (invalid & NFS_INO_INVALID_ATTR) {
1620 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1621 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1622 nfsi->attrtimeo_timestamp = jiffies;
1623 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1624 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1625 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1626 nfsi->attrtimeo_timestamp = jiffies;
1628 /* Don't invalidate the data if we were to blame */
1629 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1630 || S_ISLNK(inode->i_mode)))
1631 invalid &= ~NFS_INO_INVALID_DATA;
1633 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1634 if (!nfs_have_delegation(inode, FMODE_READ))
1635 nfsi->cache_validity |= invalid;
1640 * Big trouble! The inode has become a different object.
1643 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1644 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1648 * No need to worry about unhashing the dentry, as the
1649 * lookup validation will know that the inode is bad.
1650 * (But we fall through to invalidate the caches.)
1652 nfs_invalidate_inode(inode);
1656 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1657 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1658 NFS_SERVER(inode)->hostname, inode->i_sb->s_id,
1659 (long long)nfsi->fileid, (long long)fattr->fileid);
1664 * File system information
1668 * nfs_path - reconstruct the path given an arbitrary dentry
1669 * @base - arbitrary string to prepend to the path
1670 * @dentry - pointer to dentry
1671 * @buffer - result buffer
1672 * @buflen - length of buffer
1674 * Helper function for constructing the path from the
1675 * root dentry to an arbitrary hashed dentry.
1677 * This is mainly for use in figuring out the path on the
1678 * server side when automounting on top of an existing partition.
1680 static char *nfs_path(const char *base, const struct dentry *dentry,
1681 char *buffer, ssize_t buflen)
1683 char *end = buffer+buflen;
1688 spin_lock(&dcache_lock);
1689 while (!IS_ROOT(dentry)) {
1690 namelen = dentry->d_name.len;
1691 buflen -= namelen + 1;
1695 memcpy(end, dentry->d_name.name, namelen);
1697 dentry = dentry->d_parent;
1699 spin_unlock(&dcache_lock);
1700 namelen = strlen(base);
1701 /* Strip off excess slashes in base string */
1702 while (namelen > 0 && base[namelen - 1] == '/')
1708 memcpy(end, base, namelen);
1711 return ERR_PTR(-ENAMETOOLONG);
1714 struct nfs_clone_mount {
1715 const struct super_block *sb;
1716 const struct dentry *dentry;
1718 struct nfs_fattr *fattr;
1721 struct sockaddr_in *addr;
1722 rpc_authflavor_t authflavor;
1725 static struct super_block *nfs_clone_generic_sb(struct nfs_clone_mount *data,
1726 struct super_block *(*fill_sb)(struct nfs_server *, struct nfs_clone_mount *),
1727 struct nfs_server *(*fill_server)(struct super_block *, struct nfs_clone_mount *))
1729 struct nfs_server *server;
1730 struct nfs_server *parent = NFS_SB(data->sb);
1731 struct super_block *sb = ERR_PTR(-EINVAL);
1732 void *err = ERR_PTR(-ENOMEM);
1736 server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
1739 memcpy(server, parent, sizeof(*server));
1740 hostname = (data->hostname != NULL) ? data->hostname : parent->hostname;
1741 len = strlen(hostname) + 1;
1742 server->hostname = kmalloc(len, GFP_KERNEL);
1743 if (server->hostname == NULL)
1745 memcpy(server->hostname, hostname, len);
1746 if (rpciod_up() != 0)
1749 sb = fill_sb(server, data);
1750 if (IS_ERR((err = sb)) || sb->s_root)
1753 server = fill_server(sb, data);
1754 if (IS_ERR((err = server)))
1755 goto out_deactivate;
1758 up_write(&sb->s_umount);
1759 deactivate_super(sb);
1760 return (struct super_block *)err;
1764 kfree(server->hostname);
1768 return (struct super_block *)err;
1771 static int nfs_set_super(struct super_block *s, void *data)
1773 s->s_fs_info = data;
1774 return set_anon_super(s, data);
1777 static int nfs_compare_super(struct super_block *sb, void *data)
1779 struct nfs_server *server = data;
1780 struct nfs_server *old = NFS_SB(sb);
1782 if (old->addr.sin_addr.s_addr != server->addr.sin_addr.s_addr)
1784 if (old->addr.sin_port != server->addr.sin_port)
1786 return !nfs_compare_fh(&old->fh, &server->fh);
1789 static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
1790 int flags, const char *dev_name, void *raw_data)
1793 struct nfs_server *server = NULL;
1794 struct super_block *s;
1795 struct nfs_fh *root;
1796 struct nfs_mount_data *data = raw_data;
1798 s = ERR_PTR(-EINVAL);
1800 dprintk("%s: missing data argument\n", __FUNCTION__);
1803 if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
1804 dprintk("%s: bad mount version\n", __FUNCTION__);
1807 switch (data->version) {
1813 if (data->flags & NFS_MOUNT_VER3) {
1814 dprintk("%s: mount structure version %d does not support NFSv3\n",
1819 data->root.size = NFS2_FHSIZE;
1820 memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
1822 if (data->flags & NFS_MOUNT_SECFLAVOUR) {
1823 dprintk("%s: mount structure version %d does not support strong security\n",
1829 memset(data->context, 0, sizeof(data->context));
1831 #ifndef CONFIG_NFS_V3
1832 /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
1833 s = ERR_PTR(-EPROTONOSUPPORT);
1834 if (data->flags & NFS_MOUNT_VER3) {
1835 dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
1838 #endif /* CONFIG_NFS_V3 */
1840 s = ERR_PTR(-ENOMEM);
1841 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
1844 /* Zero out the NFS state stuff */
1845 init_nfsv4_state(server);
1846 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1849 if (data->flags & NFS_MOUNT_VER3)
1850 root->size = data->root.size;
1852 root->size = NFS2_FHSIZE;
1853 s = ERR_PTR(-EINVAL);
1854 if (root->size > sizeof(root->data)) {
1855 dprintk("%s: invalid root filehandle\n", __FUNCTION__);
1858 memcpy(root->data, data->root.data, root->size);
1860 /* We now require that the mount process passes the remote address */
1861 memcpy(&server->addr, &data->addr, sizeof(server->addr));
1862 if (server->addr.sin_addr.s_addr == INADDR_ANY) {
1863 dprintk("%s: mount program didn't pass remote address!\n",
1868 /* Fire up rpciod if not yet running */
1869 s = ERR_PTR(rpciod_up());
1871 dprintk("%s: couldn't start rpciod! Error = %ld\n",
1872 __FUNCTION__, PTR_ERR(s));
1876 s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
1877 if (IS_ERR(s) || s->s_root)
1878 goto out_rpciod_down;
1882 error = nfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1884 up_write(&s->s_umount);
1885 deactivate_super(s);
1886 return ERR_PTR(error);
1888 s->s_flags |= MS_ACTIVE;
1897 static void nfs_kill_super(struct super_block *s)
1899 struct nfs_server *server = NFS_SB(s);
1903 if (!IS_ERR(server->client))
1904 rpc_shutdown_client(server->client);
1905 if (!IS_ERR(server->client_sys))
1906 rpc_shutdown_client(server->client_sys);
1907 if (!IS_ERR(server->client_acl))
1908 rpc_shutdown_client(server->client_acl);
1910 if (!(server->flags & NFS_MOUNT_NONLM))
1911 lockd_down(); /* release rpc.lockd */
1913 rpciod_down(); /* release rpciod */
1915 nfs_free_iostats(server->io_stats);
1916 kfree(server->hostname);
1918 nfs_release_automount_timer();
1921 static struct file_system_type nfs_fs_type = {
1922 .owner = THIS_MODULE,
1924 .get_sb = nfs_get_sb,
1925 .kill_sb = nfs_kill_super,
1926 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
1929 static struct super_block *nfs_clone_sb(struct nfs_server *server, struct nfs_clone_mount *data)
1931 struct super_block *sb;
1933 server->fsid = data->fattr->fsid;
1934 nfs_copy_fh(&server->fh, data->fh);
1935 sb = sget(&nfs_fs_type, nfs_compare_super, nfs_set_super, server);
1936 if (!IS_ERR(sb) && sb->s_root == NULL && !(server->flags & NFS_MOUNT_NONLM))
1941 static struct nfs_server *nfs_clone_server(struct super_block *sb, struct nfs_clone_mount *data)
1943 struct nfs_server *server = NFS_SB(sb);
1944 struct nfs_server *parent = NFS_SB(data->sb);
1945 struct inode *root_inode;
1946 struct nfs_fsinfo fsinfo;
1947 void *err = ERR_PTR(-ENOMEM);
1949 sb->s_op = data->sb->s_op;
1950 sb->s_blocksize = data->sb->s_blocksize;
1951 sb->s_blocksize_bits = data->sb->s_blocksize_bits;
1952 sb->s_maxbytes = data->sb->s_maxbytes;
1954 server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
1955 server->io_stats = nfs_alloc_iostats();
1956 if (server->io_stats == NULL)
1959 server->client = rpc_clone_client(parent->client);
1960 if (IS_ERR((err = server->client)))
1963 if (!IS_ERR(parent->client_sys)) {
1964 server->client_sys = rpc_clone_client(parent->client_sys);
1965 if (IS_ERR((err = server->client_sys)))
1968 if (!IS_ERR(parent->client_acl)) {
1969 server->client_acl = rpc_clone_client(parent->client_acl);
1970 if (IS_ERR((err = server->client_acl)))
1973 root_inode = nfs_fhget(sb, data->fh, data->fattr);
1976 sb->s_root = d_alloc_root(root_inode);
1979 fsinfo.fattr = data->fattr;
1980 if (NFS_PROTO(root_inode)->fsinfo(server, data->fh, &fsinfo) == 0)
1981 nfs_super_set_maxbytes(sb, fsinfo.maxfilesize);
1982 sb->s_root->d_op = server->rpc_ops->dentry_ops;
1983 sb->s_flags |= MS_ACTIVE;
1991 static struct super_block *nfs_clone_nfs_sb(struct file_system_type *fs_type,
1992 int flags, const char *dev_name, void *raw_data)
1994 struct nfs_clone_mount *data = raw_data;
1995 return nfs_clone_generic_sb(data, nfs_clone_sb, nfs_clone_server);
1998 static struct file_system_type clone_nfs_fs_type = {
1999 .owner = THIS_MODULE,
2001 .get_sb = nfs_clone_nfs_sb,
2002 .kill_sb = nfs_kill_super,
2003 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2006 #ifdef CONFIG_NFS_V4
2008 static void nfs4_clear_inode(struct inode *);
2011 static struct super_operations nfs4_sops = {
2012 .alloc_inode = nfs_alloc_inode,
2013 .destroy_inode = nfs_destroy_inode,
2014 .write_inode = nfs_write_inode,
2015 .statfs = nfs_statfs,
2016 .clear_inode = nfs4_clear_inode,
2017 .umount_begin = nfs_umount_begin,
2018 .show_options = nfs_show_options,
2019 .show_stats = nfs_show_stats,
2023 * Clean out any remaining NFSv4 state that might be left over due
2024 * to open() calls that passed nfs_atomic_lookup, but failed to call
2027 static void nfs4_clear_inode(struct inode *inode)
2029 struct nfs_inode *nfsi = NFS_I(inode);
2031 /* If we are holding a delegation, return it! */
2032 nfs_inode_return_delegation(inode);
2033 /* First call standard NFS clear_inode() code */
2034 nfs_clear_inode(inode);
2035 /* Now clear out any remaining state */
2036 while (!list_empty(&nfsi->open_states)) {
2037 struct nfs4_state *state;
2039 state = list_entry(nfsi->open_states.next,
2042 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
2045 (long long)NFS_FILEID(inode),
2047 BUG_ON(atomic_read(&state->count) != 1);
2048 nfs4_close_state(state, state->state);
2053 static struct rpc_clnt *nfs4_create_client(struct nfs_server *server,
2054 struct rpc_timeout *timeparms, int proto, rpc_authflavor_t flavor)
2056 struct nfs4_client *clp;
2057 struct rpc_xprt *xprt = NULL;
2058 struct rpc_clnt *clnt = NULL;
2061 clp = nfs4_get_client(&server->addr.sin_addr);
2063 dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
2064 return ERR_PTR(err);
2067 /* Now create transport and client */
2068 down_write(&clp->cl_sem);
2069 if (IS_ERR(clp->cl_rpcclient)) {
2070 xprt = xprt_create_proto(proto, &server->addr, timeparms);
2072 up_write(&clp->cl_sem);
2073 err = PTR_ERR(xprt);
2074 dprintk("%s: cannot create RPC transport. Error = %d\n",
2078 clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
2079 server->rpc_ops->version, flavor);
2081 up_write(&clp->cl_sem);
2082 err = PTR_ERR(clnt);
2083 dprintk("%s: cannot create RPC client. Error = %d\n",
2088 clnt->cl_softrtry = 1;
2089 clp->cl_rpcclient = clnt;
2090 memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
2093 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
2094 clnt = rpc_clone_client(clp->cl_rpcclient);
2096 server->nfs4_state = clp;
2097 up_write(&clp->cl_sem);
2101 dprintk("%s: cannot create RPC client. Error = %d\n",
2106 if (server->nfs4_state->cl_idmap == NULL) {
2107 dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
2108 return ERR_PTR(-ENOMEM);
2111 if (clnt->cl_auth->au_flavor != flavor) {
2112 struct rpc_auth *auth;
2114 auth = rpcauth_create(flavor, clnt);
2116 dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
2117 return (struct rpc_clnt *)auth;
2124 nfs4_put_client(clp);
2125 return ERR_PTR(err);
2128 static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data, int silent)
2130 struct nfs_server *server;
2131 struct rpc_timeout timeparms;
2132 rpc_authflavor_t authflavour;
2135 sb->s_blocksize_bits = 0;
2136 sb->s_blocksize = 0;
2137 server = NFS_SB(sb);
2138 if (data->rsize != 0)
2139 server->rsize = nfs_block_size(data->rsize, NULL);
2140 if (data->wsize != 0)
2141 server->wsize = nfs_block_size(data->wsize, NULL);
2142 server->flags = data->flags & NFS_MOUNT_FLAGMASK;
2143 server->caps = NFS_CAP_ATOMIC_OPEN;
2145 server->acregmin = data->acregmin*HZ;
2146 server->acregmax = data->acregmax*HZ;
2147 server->acdirmin = data->acdirmin*HZ;
2148 server->acdirmax = data->acdirmax*HZ;
2150 server->rpc_ops = &nfs_v4_clientops;
2152 nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
2154 server->retrans_timeo = timeparms.to_initval;
2155 server->retrans_count = timeparms.to_retries;
2157 /* Now create transport and client */
2158 authflavour = RPC_AUTH_UNIX;
2159 if (data->auth_flavourlen != 0) {
2160 if (data->auth_flavourlen != 1) {
2161 dprintk("%s: Invalid number of RPC auth flavours %d.\n",
2162 __FUNCTION__, data->auth_flavourlen);
2166 if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
2172 server->client = nfs4_create_client(server, &timeparms, data->proto, authflavour);
2173 if (IS_ERR(server->client)) {
2174 err = PTR_ERR(server->client);
2175 dprintk("%s: cannot create RPC client. Error = %d\n",
2180 sb->s_time_gran = 1;
2182 sb->s_op = &nfs4_sops;
2183 err = nfs_sb_init(sb, authflavour);
2189 static int nfs4_compare_super(struct super_block *sb, void *data)
2191 struct nfs_server *server = data;
2192 struct nfs_server *old = NFS_SB(sb);
2194 if (strcmp(server->hostname, old->hostname) != 0)
2196 if (strcmp(server->mnt_path, old->mnt_path) != 0)
2202 nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
2207 return ERR_PTR(-EINVAL);
2208 if (src->len < maxlen)
2211 p = dst = kmalloc(maxlen + 1, GFP_KERNEL);
2213 return ERR_PTR(-ENOMEM);
2215 if (copy_from_user(dst, src->data, maxlen)) {
2217 return ERR_PTR(-EFAULT);
2223 static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
2224 int flags, const char *dev_name, void *raw_data)
2227 struct nfs_server *server;
2228 struct super_block *s;
2229 struct nfs4_mount_data *data = raw_data;
2233 dprintk("%s: missing data argument\n", __FUNCTION__);
2234 return ERR_PTR(-EINVAL);
2236 if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
2237 dprintk("%s: bad mount version\n", __FUNCTION__);
2238 return ERR_PTR(-EINVAL);
2241 server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
2243 return ERR_PTR(-ENOMEM);
2244 /* Zero out the NFS state stuff */
2245 init_nfsv4_state(server);
2246 server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
2248 p = nfs_copy_user_string(NULL, &data->hostname, 256);
2251 server->hostname = p;
2253 p = nfs_copy_user_string(NULL, &data->mnt_path, 1024);
2256 server->mnt_path = p;
2258 p = nfs_copy_user_string(server->ip_addr, &data->client_addr,
2259 sizeof(server->ip_addr) - 1);
2263 /* We now require that the mount process passes the remote address */
2264 if (data->host_addrlen != sizeof(server->addr)) {
2265 s = ERR_PTR(-EINVAL);
2268 if (copy_from_user(&server->addr, data->host_addr, sizeof(server->addr))) {
2269 s = ERR_PTR(-EFAULT);
2272 if (server->addr.sin_family != AF_INET ||
2273 server->addr.sin_addr.s_addr == INADDR_ANY) {
2274 dprintk("%s: mount program didn't pass remote IP address!\n",
2276 s = ERR_PTR(-EINVAL);
2280 /* Fire up rpciod if not yet running */
2281 s = ERR_PTR(rpciod_up());
2283 dprintk("%s: couldn't start rpciod! Error = %ld\n",
2284 __FUNCTION__, PTR_ERR(s));
2288 s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
2290 if (IS_ERR(s) || s->s_root)
2295 error = nfs4_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
2297 up_write(&s->s_umount);
2298 deactivate_super(s);
2299 return ERR_PTR(error);
2301 s->s_flags |= MS_ACTIVE;
2304 s = (struct super_block *)p;
2306 kfree(server->mnt_path);
2307 kfree(server->hostname);
2312 static void nfs4_kill_super(struct super_block *sb)
2314 struct nfs_server *server = NFS_SB(sb);
2316 nfs_return_all_delegations(sb);
2317 kill_anon_super(sb);
2319 nfs4_renewd_prepare_shutdown(server);
2321 if (server->client != NULL && !IS_ERR(server->client))
2322 rpc_shutdown_client(server->client);
2324 destroy_nfsv4_state(server);
2328 nfs_free_iostats(server->io_stats);
2329 kfree(server->hostname);
2331 nfs_release_automount_timer();
2334 static struct file_system_type nfs4_fs_type = {
2335 .owner = THIS_MODULE,
2337 .get_sb = nfs4_get_sb,
2338 .kill_sb = nfs4_kill_super,
2339 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2342 static const int nfs_set_port_min = 0;
2343 static const int nfs_set_port_max = 65535;
2344 static int param_set_port(const char *val, struct kernel_param *kp)
2347 int num = simple_strtol(val, &endp, 0);
2348 if (endp == val || *endp || num < nfs_set_port_min || num > nfs_set_port_max)
2350 *((int *)kp->arg) = num;
2354 module_param_call(callback_tcpport, param_set_port, param_get_int,
2355 &nfs_callback_set_tcpport, 0644);
2357 static int param_set_idmap_timeout(const char *val, struct kernel_param *kp)
2360 int num = simple_strtol(val, &endp, 0);
2362 if (endp == val || *endp || num < 0 || jif < num)
2364 *((int *)kp->arg) = jif;
2368 module_param_call(idmap_cache_timeout, param_set_idmap_timeout, param_get_int,
2369 &nfs_idmap_cache_timeout, 0644);
2371 /* Constructs the SERVER-side path */
2372 static inline char *nfs4_path(const struct dentry *dentry, char *buffer, ssize_t buflen)
2374 return nfs_path(NFS_SB(dentry->d_sb)->mnt_path, dentry, buffer, buflen);
2377 static inline char *nfs4_dup_path(const struct dentry *dentry)
2379 char *page = (char *) __get_free_page(GFP_USER);
2382 path = nfs4_path(dentry, page, PAGE_SIZE);
2383 if (!IS_ERR(path)) {
2384 int len = PAGE_SIZE + page - path;
2387 path = kmalloc(len, GFP_KERNEL);
2389 memcpy(path, tmp, len);
2391 path = ERR_PTR(-ENOMEM);
2393 free_page((unsigned long)page);
2397 static struct super_block *nfs4_clone_sb(struct nfs_server *server, struct nfs_clone_mount *data)
2399 const struct dentry *dentry = data->dentry;
2400 struct nfs4_client *clp = server->nfs4_state;
2401 struct super_block *sb;
2403 server->fsid = data->fattr->fsid;
2404 nfs_copy_fh(&server->fh, data->fh);
2405 server->mnt_path = nfs4_dup_path(dentry);
2406 if (IS_ERR(server->mnt_path)) {
2407 sb = (struct super_block *)server->mnt_path;
2410 sb = sget(&nfs4_fs_type, nfs4_compare_super, nfs_set_super, server);
2411 if (IS_ERR(sb) || sb->s_root)
2413 nfs4_server_capabilities(server, &server->fh);
2415 down_write(&clp->cl_sem);
2416 atomic_inc(&clp->cl_count);
2417 list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
2418 up_write(&clp->cl_sem);
2421 kfree(server->mnt_path);
2423 server->mnt_path = NULL;
2427 static struct super_block *nfs_clone_nfs4_sb(struct file_system_type *fs_type,
2428 int flags, const char *dev_name, void *raw_data)
2430 struct nfs_clone_mount *data = raw_data;
2431 return nfs_clone_generic_sb(data, nfs4_clone_sb, nfs_clone_server);
2434 static struct file_system_type clone_nfs4_fs_type = {
2435 .owner = THIS_MODULE,
2437 .get_sb = nfs_clone_nfs4_sb,
2438 .kill_sb = nfs4_kill_super,
2439 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2442 #define nfs4_init_once(nfsi) \
2444 INIT_LIST_HEAD(&(nfsi)->open_states); \
2445 nfsi->delegation = NULL; \
2446 nfsi->delegation_state = 0; \
2447 init_rwsem(&nfsi->rwsem); \
2450 static inline int register_nfs4fs(void)
2454 ret = nfs_register_sysctl();
2457 ret = register_filesystem(&nfs4_fs_type);
2459 nfs_unregister_sysctl();
2463 static inline void unregister_nfs4fs(void)
2465 unregister_filesystem(&nfs4_fs_type);
2466 nfs_unregister_sysctl();
2469 #define nfs4_fill_sb(a,b) ERR_PTR(-EINVAL)
2470 #define nfs4_fill_super(a,b) ERR_PTR(-EINVAL)
2471 #define nfs4_init_once(nfsi) \
2473 #define register_nfs4fs() (0)
2474 #define unregister_nfs4fs()
2477 static inline char *nfs_devname(const struct vfsmount *mnt_parent,
2478 const struct dentry *dentry,
2479 char *buffer, ssize_t buflen)
2481 return nfs_path(mnt_parent->mnt_devname, dentry, buffer, buflen);
2485 * nfs_do_submount - set up mountpoint when crossing a filesystem boundary
2486 * @mnt_parent - mountpoint of parent directory
2487 * @dentry - parent directory
2488 * @fh - filehandle for new root dentry
2489 * @fattr - attributes for new root inode
2492 struct vfsmount *nfs_do_submount(const struct vfsmount *mnt_parent,
2493 const struct dentry *dentry, struct nfs_fh *fh,
2494 struct nfs_fattr *fattr)
2496 struct nfs_clone_mount mountdata = {
2497 .sb = mnt_parent->mnt_sb,
2502 struct vfsmount *mnt = ERR_PTR(-ENOMEM);
2503 char *page = (char *) __get_free_page(GFP_USER);
2506 dprintk("%s: submounting on %s/%s\n", __FUNCTION__,
2507 dentry->d_parent->d_name.name,
2508 dentry->d_name.name);
2511 devname = nfs_devname(mnt_parent, dentry, page, PAGE_SIZE);
2512 mnt = (struct vfsmount *)devname;
2513 if (IS_ERR(devname))
2515 switch (NFS_SB(mnt_parent->mnt_sb)->rpc_ops->version) {
2518 mnt = vfs_kern_mount(&clone_nfs_fs_type, 0, devname, &mountdata);
2521 mnt = vfs_kern_mount(&clone_nfs4_fs_type, 0, devname, &mountdata);
2527 free_page((unsigned long)page);
2529 dprintk("%s: done\n", __FUNCTION__);
2533 /* Check if fs_root is valid */
2534 static inline char *nfs4_pathname_string(struct nfs4_pathname *pathname, char *buffer, ssize_t buflen)
2536 char *end = buffer + buflen;
2542 n = pathname->ncomponents;
2544 struct nfs4_string *component = &pathname->components[n];
2545 buflen -= component->len + 1;
2548 end -= component->len;
2549 memcpy(end, component->data, component->len);
2554 return ERR_PTR(-ENAMETOOLONG);
2557 /* Check if the string represents a "valid" IPv4 address */
2558 static inline int valid_ipaddr4(const char *buf)
2560 int rc, count, in[4];
2562 rc = sscanf(buf, "%d.%d.%d.%d", &in[0], &in[1], &in[2], &in[3]);
2565 for (count = 0; count < 4; count++) {
2566 if (in[count] > 255)
2572 static struct super_block *nfs4_referral_sb(struct nfs_server *server, struct nfs_clone_mount *data)
2574 struct super_block *sb = ERR_PTR(-ENOMEM);
2577 len = strlen(data->mnt_path) + 1;
2578 server->mnt_path = kmalloc(len, GFP_KERNEL);
2579 if (server->mnt_path == NULL)
2581 memcpy(server->mnt_path, data->mnt_path, len);
2582 memcpy(&server->addr, data->addr, sizeof(struct sockaddr_in));
2584 sb = sget(&nfs4_fs_type, nfs4_compare_super, nfs_set_super, server);
2585 if (IS_ERR(sb) || sb->s_root)
2589 kfree(server->mnt_path);
2591 server->mnt_path = NULL;
2595 static struct nfs_server *nfs4_referral_server(struct super_block *sb, struct nfs_clone_mount *data)
2597 struct nfs_server *server = NFS_SB(sb);
2598 struct rpc_timeout timeparms;
2599 int proto, timeo, retrans;
2602 proto = IPPROTO_TCP;
2603 /* Since we are following a referral and there may be alternatives,
2604 set the timeouts and retries to low values */
2607 nfs_init_timeout_values(&timeparms, proto, timeo, retrans);
2609 server->client = nfs4_create_client(server, &timeparms, proto, data->authflavor);
2610 if (IS_ERR((err = server->client)))
2613 sb->s_time_gran = 1;
2614 sb->s_op = &nfs4_sops;
2615 err = ERR_PTR(nfs_sb_init(sb, data->authflavor));
2619 return (struct nfs_server *)err;
2622 static struct super_block *nfs_referral_nfs4_sb(struct file_system_type *fs_type,
2623 int flags, const char *dev_name, void *raw_data)
2625 struct nfs_clone_mount *data = raw_data;
2626 return nfs_clone_generic_sb(data, nfs4_referral_sb, nfs4_referral_server);
2629 static struct file_system_type nfs_referral_nfs4_fs_type = {
2630 .owner = THIS_MODULE,
2632 .get_sb = nfs_referral_nfs4_sb,
2633 .kill_sb = nfs4_kill_super,
2634 .fs_flags = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
2638 * nfs_follow_referral - set up mountpoint when hitting a referral on moved error
2639 * @mnt_parent - mountpoint of parent directory
2640 * @dentry - parent directory
2641 * @fspath - fs path returned in fs_locations
2642 * @mntpath - mount path to new server
2643 * @hostname - hostname of new server
2644 * @addr - host addr of new server
2647 struct vfsmount *nfs_follow_referral(const struct vfsmount *mnt_parent,
2648 const struct dentry *dentry, struct nfs4_fs_locations *locations)
2650 struct vfsmount *mnt = ERR_PTR(-ENOENT);
2651 struct nfs_clone_mount mountdata = {
2652 .sb = mnt_parent->mnt_sb,
2654 .authflavor = NFS_SB(mnt_parent->mnt_sb)->client->cl_auth->au_flavor,
2657 char *path, *fs_path;
2661 if (locations == NULL || locations->nlocations <= 0)
2664 dprintk("%s: referral at %s/%s\n", __FUNCTION__,
2665 dentry->d_parent->d_name.name, dentry->d_name.name);
2667 /* Ensure fs path is a prefix of current dentry path */
2668 page = (char *) __get_free_page(GFP_USER);
2671 page2 = (char *) __get_free_page(GFP_USER);
2675 path = nfs4_path(dentry, page, PAGE_SIZE);
2679 fs_path = nfs4_pathname_string(&locations->fs_path, page2, PAGE_SIZE);
2680 if (IS_ERR(fs_path))
2683 if (strncmp(path, fs_path, strlen(fs_path)) != 0) {
2684 dprintk("%s: path %s does not begin with fsroot %s\n", __FUNCTION__, path, fs_path);
2688 devname = nfs_devname(mnt_parent, dentry, page, PAGE_SIZE);
2689 if (IS_ERR(devname)) {
2690 mnt = (struct vfsmount *)devname;
2695 while (loc < locations->nlocations && IS_ERR(mnt)) {
2696 struct nfs4_fs_location *location = &locations->locations[loc];
2699 if (location == NULL || location->nservers <= 0 ||
2700 location->rootpath.ncomponents == 0) {
2705 mnt_path = nfs4_pathname_string(&location->rootpath, page2, PAGE_SIZE);
2706 if (IS_ERR(mnt_path)) {
2710 mountdata.mnt_path = mnt_path;
2713 while (s < location->nservers) {
2714 struct sockaddr_in addr = {};
2716 if (location->servers[s].len <= 0 ||
2717 valid_ipaddr4(location->servers[s].data) < 0) {
2722 mountdata.hostname = location->servers[s].data;
2723 addr.sin_addr.s_addr = in_aton(mountdata.hostname);
2724 addr.sin_family = AF_INET;
2725 addr.sin_port = htons(NFS_PORT);
2726 mountdata.addr = &addr;
2728 mnt = vfs_kern_mount(&nfs_referral_nfs4_fs_type, 0, devname, &mountdata);
2738 free_page((unsigned long)page);
2739 free_page((unsigned long)page2);
2741 dprintk("%s: done\n", __FUNCTION__);
2746 * nfs_do_refmount - handle crossing a referral on server
2747 * @dentry - dentry of referral
2748 * @nd - nameidata info
2751 struct vfsmount *nfs_do_refmount(const struct vfsmount *mnt_parent, struct dentry *dentry)
2753 struct vfsmount *mnt = ERR_PTR(-ENOENT);
2754 struct dentry *parent;
2755 struct nfs4_fs_locations *fs_locations = NULL;
2759 /* BUG_ON(IS_ROOT(dentry)); */
2760 dprintk("%s: enter\n", __FUNCTION__);
2762 page = alloc_page(GFP_KERNEL);
2766 fs_locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2767 if (fs_locations == NULL)
2771 parent = dget_parent(dentry);
2772 dprintk("%s: getting locations for %s/%s\n", __FUNCTION__, parent->d_name.name, dentry->d_name.name);
2773 err = nfs4_proc_fs_locations(parent->d_inode, dentry, fs_locations, page);
2775 if (err != 0 || fs_locations->nlocations <= 0 ||
2776 fs_locations->fs_path.ncomponents <= 0)
2779 mnt = nfs_follow_referral(mnt_parent, dentry, fs_locations);
2782 kfree(fs_locations);
2784 dprintk("%s: done\n", __FUNCTION__);
2788 extern int nfs_init_nfspagecache(void);
2789 extern void nfs_destroy_nfspagecache(void);
2790 extern int nfs_init_readpagecache(void);
2791 extern void nfs_destroy_readpagecache(void);
2792 extern int nfs_init_writepagecache(void);
2793 extern void nfs_destroy_writepagecache(void);
2794 #ifdef CONFIG_NFS_DIRECTIO
2795 extern int nfs_init_directcache(void);
2796 extern void nfs_destroy_directcache(void);
2799 static kmem_cache_t * nfs_inode_cachep;
2801 static struct inode *nfs_alloc_inode(struct super_block *sb)
2803 struct nfs_inode *nfsi;
2804 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
2808 nfsi->cache_validity = 0UL;
2809 nfsi->cache_change_attribute = jiffies;
2810 #ifdef CONFIG_NFS_V3_ACL
2811 nfsi->acl_access = ERR_PTR(-EAGAIN);
2812 nfsi->acl_default = ERR_PTR(-EAGAIN);
2814 #ifdef CONFIG_NFS_V4
2815 nfsi->nfs4_acl = NULL;
2816 #endif /* CONFIG_NFS_V4 */
2817 return &nfsi->vfs_inode;
2820 static void nfs_destroy_inode(struct inode *inode)
2822 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2825 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
2827 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2829 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2830 SLAB_CTOR_CONSTRUCTOR) {
2831 inode_init_once(&nfsi->vfs_inode);
2832 spin_lock_init(&nfsi->req_lock);
2833 INIT_LIST_HEAD(&nfsi->dirty);
2834 INIT_LIST_HEAD(&nfsi->commit);
2835 INIT_LIST_HEAD(&nfsi->open_files);
2836 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
2837 atomic_set(&nfsi->data_updates, 0);
2841 nfs4_init_once(nfsi);
2845 static int nfs_init_inodecache(void)
2847 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2848 sizeof(struct nfs_inode),
2849 0, (SLAB_RECLAIM_ACCOUNT|
2852 if (nfs_inode_cachep == NULL)
2858 static void nfs_destroy_inodecache(void)
2860 if (kmem_cache_destroy(nfs_inode_cachep))
2861 printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
2867 static int __init init_nfs_fs(void)
2871 err = nfs_init_nfspagecache();
2875 err = nfs_init_inodecache();
2879 err = nfs_init_readpagecache();
2883 err = nfs_init_writepagecache();
2887 #ifdef CONFIG_NFS_DIRECTIO
2888 err = nfs_init_directcache();
2893 #ifdef CONFIG_PROC_FS
2894 rpc_proc_register(&nfs_rpcstat);
2896 err = register_filesystem(&nfs_fs_type);
2899 if ((err = register_nfs4fs()) != 0)
2903 #ifdef CONFIG_PROC_FS
2904 rpc_proc_unregister("nfs");
2906 #ifdef CONFIG_NFS_DIRECTIO
2907 nfs_destroy_directcache();
2910 nfs_destroy_writepagecache();
2912 nfs_destroy_readpagecache();
2914 nfs_destroy_inodecache();
2916 nfs_destroy_nfspagecache();
2921 static void __exit exit_nfs_fs(void)
2923 #ifdef CONFIG_NFS_DIRECTIO
2924 nfs_destroy_directcache();
2926 nfs_destroy_writepagecache();
2927 nfs_destroy_readpagecache();
2928 nfs_destroy_inodecache();
2929 nfs_destroy_nfspagecache();
2930 #ifdef CONFIG_PROC_FS
2931 rpc_proc_unregister("nfs");
2933 unregister_filesystem(&nfs_fs_type);
2934 unregister_nfs4fs();
2937 /* Not quite true; I just maintain it */
2938 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2939 MODULE_LICENSE("GPL");
2941 module_init(init_nfs_fs)
2942 module_exit(exit_nfs_fs)