Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.4-vs2.0.2-rc26.diff
[linux-2.6.git] / fs / nfs / inode.c
index 717a411..64bae22 100644 (file)
@@ -26,6 +26,7 @@
 #include <linux/unistd.h>
 #include <linux/sunrpc/clnt.h>
 #include <linux/sunrpc/stats.h>
+#include <linux/sunrpc/metrics.h>
 #include <linux/nfs_fs.h>
 #include <linux/nfs_mount.h>
 #include <linux/nfs4_mount.h>
 #include <asm/system.h>
 #include <asm/uaccess.h>
 
+#include "nfs4_fs.h"
+#include "callback.h"
 #include "delegation.h"
+#include "iostat.h"
 
 #define NFSDBG_FACILITY                NFSDBG_VFS
 #define NFS_PARANOIA 1
@@ -54,7 +58,7 @@
 #define NFS_MAX_READAHEAD      (RPC_DEF_SLOT_TABLE - 1)
 
 static void nfs_invalidate_inode(struct inode *);
-static int nfs_update_inode(struct inode *, struct nfs_fattr *, unsigned long);
+static int nfs_update_inode(struct inode *, struct nfs_fattr *);
 
 static struct inode *nfs_alloc_inode(struct super_block *sb);
 static void nfs_destroy_inode(struct inode *);
@@ -64,6 +68,8 @@ static void nfs_clear_inode(struct inode *);
 static void nfs_umount_begin(struct super_block *);
 static int  nfs_statfs(struct super_block *, struct kstatfs *);
 static int  nfs_show_options(struct seq_file *, struct vfsmount *);
+static int  nfs_show_stats(struct seq_file *, struct vfsmount *);
+static void nfs_zap_acl_cache(struct inode *);
 
 static struct rpc_program      nfs_program;
 
@@ -76,6 +82,7 @@ static struct super_operations nfs_sops = {
        .clear_inode    = nfs_clear_inode,
        .umount_begin   = nfs_umount_begin,
        .show_options   = nfs_show_options,
+       .show_stats     = nfs_show_stats,
 };
 
 /*
@@ -101,12 +108,27 @@ static struct rpc_version *       nfs_version[] = {
 static struct rpc_program      nfs_program = {
        .name                   = "nfs",
        .number                 = NFS_PROGRAM,
-       .nrvers                 = sizeof(nfs_version) / sizeof(nfs_version[0]),
+       .nrvers                 = ARRAY_SIZE(nfs_version),
        .version                = nfs_version,
        .stats                  = &nfs_rpcstat,
        .pipe_dir_name          = "/nfs",
 };
 
+#ifdef CONFIG_NFS_V3_ACL
+static struct rpc_stat         nfsacl_rpcstat = { &nfsacl_program };
+static struct rpc_version *    nfsacl_version[] = {
+       [3]                     = &nfsacl_version3,
+};
+
+struct rpc_program             nfsacl_program = {
+       .name =                 "nfsacl",
+       .number =               NFS_ACL_PROGRAM,
+       .nrvers =               ARRAY_SIZE(nfsacl_version),
+       .version =              nfsacl_version,
+       .stats =                &nfsacl_rpcstat,
+};
+#endif  /* CONFIG_NFS_V3_ACL */
+
 static inline unsigned long
 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
 {
@@ -116,10 +138,10 @@ nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
 static int
 nfs_write_inode(struct inode *inode, int sync)
 {
-       int flags = sync ? FLUSH_WAIT : 0;
+       int flags = sync ? FLUSH_SYNC : 0;
        int ret;
 
-       ret = nfs_commit_inode(inode, 0, 0, flags);
+       ret = nfs_commit_inode(inode, flags);
        if (ret < 0)
                return ret;
        return 0;
@@ -130,6 +152,8 @@ nfs_delete_inode(struct inode * inode)
 {
        dprintk("NFS: delete_inode(%s/%ld)\n", inode->i_sb->s_id, inode->i_ino);
 
+       truncate_inode_pages(&inode->i_data, 0);
+
        nfs_wb_all(inode);
        /*
         * The following should never happen...
@@ -141,10 +165,6 @@ nfs_delete_inode(struct inode * inode)
        clear_inode(inode);
 }
 
-/*
- * For the moment, the only task for the NFS clear_inode method is to
- * release the mmap credential
- */
 static void
 nfs_clear_inode(struct inode *inode)
 {
@@ -153,6 +173,7 @@ nfs_clear_inode(struct inode *inode)
 
        nfs_wb_all(inode);
        BUG_ON (!list_empty(&nfsi->open_files));
+       nfs_zap_acl_cache(inode);
        cred = nfsi->cache_access.cred;
        if (cred)
                put_rpccred(cred);
@@ -162,11 +183,13 @@ nfs_clear_inode(struct inode *inode)
 void
 nfs_umount_begin(struct super_block *sb)
 {
-       struct nfs_server *server = NFS_SB(sb);
-       struct rpc_clnt *rpc;
+       struct rpc_clnt *rpc = NFS_SB(sb)->client;
 
        /* -EIO all pending I/O */
-       if ((rpc = server->client) != NULL)
+       if (!IS_ERR(rpc))
+               rpc_killall_tasks(rpc);
+       rpc = NFS_SB(sb)->client_acl;
+       if (!IS_ERR(rpc))
                rpc_killall_tasks(rpc);
 }
 
@@ -204,10 +227,10 @@ nfs_calc_block_size(u64 tsize)
 static inline unsigned long
 nfs_block_size(unsigned long bsize, unsigned char *nrbitsp)
 {
-       if (bsize < 1024)
-               bsize = NFS_DEF_FILE_IO_BUFFER_SIZE;
-       else if (bsize >= NFS_MAX_FILE_IO_BUFFER_SIZE)
-               bsize = NFS_MAX_FILE_IO_BUFFER_SIZE;
+       if (bsize < NFS_MIN_FILE_IO_SIZE)
+               bsize = NFS_DEF_FILE_IO_SIZE;
+       else if (bsize >= NFS_MAX_FILE_IO_SIZE)
+               bsize = NFS_MAX_FILE_IO_SIZE;
 
        return nfs_block_bits(bsize, nrbitsp);
 }
@@ -219,7 +242,6 @@ static struct inode *
 nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *fsinfo)
 {
        struct nfs_server       *server = NFS_SB(sb);
-       struct inode *rooti;
        int                     error;
 
        error = server->rpc_ops->getroot(server, rootfh, fsinfo);
@@ -228,10 +250,7 @@ nfs_get_root(struct super_block *sb, struct nfs_fh *rootfh, struct nfs_fsinfo *f
                return ERR_PTR(error);
        }
 
-       rooti = nfs_fhget(sb, rootfh, fsinfo->fattr);
-       if (!rooti)
-               return ERR_PTR(-ENOMEM);
-       return rooti;
+       return nfs_fhget(sb, rootfh, fsinfo->fattr);
 }
 
 /*
@@ -259,6 +278,10 @@ nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
 
        sb->s_magic      = NFS_SUPER_MAGIC;
 
+       server->io_stats = nfs_alloc_iostats();
+       if (server->io_stats == NULL)
+               return -ENOMEM;
+
        root_inode = nfs_get_root(sb, &server->fh, &fsinfo);
        /* Did getting the root inode fail? */
        if (IS_ERR(root_inode)) {
@@ -272,6 +295,9 @@ nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
        }
        sb->s_root->d_op = server->rpc_ops->dentry_ops;
 
+       /* mount time stamp, in seconds */
+       server->mount_time = jiffies;
+
        /* Get some general file system info */
        if (server->namelen == 0 &&
            server->rpc_ops->pathconf(server, &server->fh, &pathinfo) >= 0)
@@ -290,20 +316,15 @@ nfs_sb_init(struct super_block *sb, rpc_authflavor_t authflavor)
        max_rpc_payload = nfs_block_size(rpc_max_payload(server->client), NULL);
        if (server->rsize > max_rpc_payload)
                server->rsize = max_rpc_payload;
-       if (server->wsize > max_rpc_payload)
-               server->wsize = max_rpc_payload;
-
+       if (server->rsize > NFS_MAX_FILE_IO_SIZE)
+               server->rsize = NFS_MAX_FILE_IO_SIZE;
        server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
-       if (server->rpages > NFS_READ_MAXIOV) {
-               server->rpages = NFS_READ_MAXIOV;
-               server->rsize = server->rpages << PAGE_CACHE_SHIFT;
-       }
 
+       if (server->wsize > max_rpc_payload)
+               server->wsize = max_rpc_payload;
+       if (server->wsize > NFS_MAX_FILE_IO_SIZE)
+               server->wsize = NFS_MAX_FILE_IO_SIZE;
        server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
-        if (server->wpages > NFS_WRITE_MAXIOV) {
-               server->wpages = NFS_WRITE_MAXIOV;
-                server->wsize = server->wpages << PAGE_CACHE_SHIFT;
-       }
 
        if (sb->s_blocksize == 0)
                sb->s_blocksize = nfs_block_bits(server->wsize,
@@ -345,6 +366,35 @@ out_no_root:
        return no_root_error;
 }
 
+static void nfs_init_timeout_values(struct rpc_timeout *to, int proto, unsigned int timeo, unsigned int retrans)
+{
+       to->to_initval = timeo * HZ / 10;
+       to->to_retries = retrans;
+       if (!to->to_retries)
+               to->to_retries = 2;
+
+       switch (proto) {
+       case IPPROTO_TCP:
+               if (!to->to_initval)
+                       to->to_initval = 60 * HZ;
+               if (to->to_initval > NFS_MAX_TCP_TIMEOUT)
+                       to->to_initval = NFS_MAX_TCP_TIMEOUT;
+               to->to_increment = to->to_initval;
+               to->to_maxval = to->to_initval + (to->to_increment * to->to_retries);
+               to->to_exponential = 0;
+               break;
+       case IPPROTO_UDP:
+       default:
+               if (!to->to_initval)
+                       to->to_initval = 11 * HZ / 10;
+               if (to->to_initval > NFS_MAX_UDP_TIMEOUT)
+                       to->to_initval = NFS_MAX_UDP_TIMEOUT;
+               to->to_maxval = NFS_MAX_UDP_TIMEOUT;
+               to->to_exponential = 1;
+               break;
+       }
+}
+
 /*
  * Create an RPC client handle.
  */
@@ -354,42 +404,35 @@ nfs_create_client(struct nfs_server *server, const struct nfs_mount_data *data)
        struct rpc_timeout      timeparms;
        struct rpc_xprt         *xprt = NULL;
        struct rpc_clnt         *clnt = NULL;
-       int                     tcp   = (data->flags & NFS_MOUNT_TCP);
+       int                     proto = (data->flags & NFS_MOUNT_TCP) ? IPPROTO_TCP : IPPROTO_UDP;
 
-       /* Initialize timeout values */
-       timeparms.to_initval = data->timeo * HZ / 10;
-       timeparms.to_retries = data->retrans;
-       timeparms.to_maxval  = tcp ? RPC_MAX_TCP_TIMEOUT : RPC_MAX_UDP_TIMEOUT;
-       timeparms.to_exponential = 1;
+       nfs_init_timeout_values(&timeparms, proto, data->timeo, data->retrans);
 
-       if (!timeparms.to_initval)
-               timeparms.to_initval = (tcp ? 600 : 11) * HZ / 10;
-       if (!timeparms.to_retries)
-               timeparms.to_retries = 5;
+       server->retrans_timeo = timeparms.to_initval;
+       server->retrans_count = timeparms.to_retries;
 
        /* create transport and client */
-       xprt = xprt_create_proto(tcp ? IPPROTO_TCP : IPPROTO_UDP,
-                                &server->addr, &timeparms);
+       xprt = xprt_create_proto(proto, &server->addr, &timeparms);
        if (IS_ERR(xprt)) {
-               printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
+               dprintk("%s: cannot create RPC transport. Error = %ld\n",
+                               __FUNCTION__, PTR_ERR(xprt));
                return (struct rpc_clnt *)xprt;
        }
        clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
                                 server->rpc_ops->version, data->pseudoflavor);
        if (IS_ERR(clnt)) {
-               printk(KERN_WARNING "NFS: cannot create RPC client.\n");
+               dprintk("%s: cannot create RPC client. Error = %ld\n",
+                               __FUNCTION__, PTR_ERR(xprt));
                goto out_fail;
        }
 
        clnt->cl_intr     = 1;
        clnt->cl_softrtry = 1;
        clnt->cl_tagxid   = 1;
-       clnt->cl_chatty   = 1;
 
        return clnt;
 
 out_fail:
-       xprt_destroy(xprt);
        return clnt;
 }
 
@@ -433,21 +476,16 @@ nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
 
        /* Check NFS protocol revision and initialize RPC op vector
         * and file handle pool. */
-       if (server->flags & NFS_MOUNT_VER3) {
 #ifdef CONFIG_NFS_V3
+       if (server->flags & NFS_MOUNT_VER3) {
                server->rpc_ops = &nfs_v3_clientops;
                server->caps |= NFS_CAP_READDIRPLUS;
-               if (data->version < 4) {
-                       printk(KERN_NOTICE "NFS: NFSv3 not supported by mount program.\n");
-                       return -EIO;
-               }
-#else
-               printk(KERN_NOTICE "NFS: NFSv3 not supported.\n");
-               return -EIO;
-#endif
        } else {
                server->rpc_ops = &nfs_v2_clientops;
        }
+#else
+       server->rpc_ops = &nfs_v2_clientops;
+#endif
 
        /* Fill in pseudoflavor for mount version < 5 */
        if (!(data->flags & NFS_MOUNT_SECFLAVOUR))
@@ -461,17 +499,34 @@ nfs_fill_super(struct super_block *sb, struct nfs_mount_data *data, int silent)
                return PTR_ERR(server->client);
        /* RFC 2623, sec 2.3.2 */
        if (authflavor != RPC_AUTH_UNIX) {
+               struct rpc_auth *auth;
+
                server->client_sys = rpc_clone_client(server->client);
                if (IS_ERR(server->client_sys))
                        return PTR_ERR(server->client_sys);
-               if (!rpcauth_create(RPC_AUTH_UNIX, server->client_sys))
-                       return -ENOMEM;
+               auth = rpcauth_create(RPC_AUTH_UNIX, server->client_sys);
+               if (IS_ERR(auth))
+                       return PTR_ERR(auth);
        } else {
                atomic_inc(&server->client->cl_count);
                server->client_sys = server->client;
        }
-
        if (server->flags & NFS_MOUNT_VER3) {
+#ifdef CONFIG_NFS_V3_ACL
+               if (!(server->flags & NFS_MOUNT_NOACL)) {
+                       server->client_acl = rpc_bind_new_program(server->client, &nfsacl_program, 3);
+                       /* No errors! Assume that Sun nfsacls are supported */
+                       if (!IS_ERR(server->client_acl))
+                               server->caps |= NFS_CAP_ACLS;
+               }
+#else
+               server->flags &= ~NFS_MOUNT_NOACL;
+#endif /* CONFIG_NFS_V3_ACL */
+               /*
+                * The VFS shouldn't apply the umask to mode bits. We will
+                * do so ourselves when necessary.
+                */
+               sb->s_flags |= MS_POSIXACL;
                if (server->namelen == 0 || server->namelen > NFS3_MAXNAMLEN)
                        server->namelen = NFS3_MAXNAMLEN;
                sb->s_time_gran = 1;
@@ -531,17 +586,16 @@ nfs_statfs(struct super_block *sb, struct kstatfs *buf)
        buf->f_namelen = server->namelen;
  out:
        unlock_kernel();
-
        return 0;
 
  out_err:
-       printk(KERN_WARNING "nfs_statfs: statfs error = %d\n", -error);
+       dprintk("%s: statfs error = %d\n", __FUNCTION__, -error);
        buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1;
        goto out;
 
 }
 
-static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
+static void nfs_show_mount_options(struct seq_file *m, struct nfs_server *nfss, int showdefaults)
 {
        static struct proc_nfs_info {
                int flag;
@@ -550,27 +604,27 @@ static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
        } nfs_info[] = {
                { NFS_MOUNT_SOFT, ",soft", ",hard" },
                { NFS_MOUNT_INTR, ",intr", "" },
-               { NFS_MOUNT_POSIX, ",posix", "" },
-               { NFS_MOUNT_TCP, ",tcp", ",udp" },
                { NFS_MOUNT_NOCTO, ",nocto", "" },
                { NFS_MOUNT_NOAC, ",noac", "" },
-               { NFS_MOUNT_NONLM, ",nolock", ",lock" },
+               { NFS_MOUNT_NONLM, ",nolock", "" },
+               { NFS_MOUNT_NOACL, ",noacl", "" },
                { NFS_MOUNT_TAGXID, ",tagxid", "" },
                { 0, NULL, NULL }
        };
        struct proc_nfs_info *nfs_infop;
-       struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
+       char buf[12];
+       char *proto;
 
-       seq_printf(m, ",v%d", nfss->rpc_ops->version);
+       seq_printf(m, ",vers=%d", nfss->rpc_ops->version);
        seq_printf(m, ",rsize=%d", nfss->rsize);
        seq_printf(m, ",wsize=%d", nfss->wsize);
-       if (nfss->acregmin != 3*HZ)
+       if (nfss->acregmin != 3*HZ || showdefaults)
                seq_printf(m, ",acregmin=%d", nfss->acregmin/HZ);
-       if (nfss->acregmax != 60*HZ)
+       if (nfss->acregmax != 60*HZ || showdefaults)
                seq_printf(m, ",acregmax=%d", nfss->acregmax/HZ);
-       if (nfss->acdirmin != 30*HZ)
+       if (nfss->acdirmin != 30*HZ || showdefaults)
                seq_printf(m, ",acdirmin=%d", nfss->acdirmin/HZ);
-       if (nfss->acdirmax != 60*HZ)
+       if (nfss->acdirmax != 60*HZ || showdefaults)
                seq_printf(m, ",acdirmax=%d", nfss->acdirmax/HZ);
        for (nfs_infop = nfs_info; nfs_infop->flag; nfs_infop++) {
                if (nfss->flags & nfs_infop->flag)
@@ -578,41 +632,173 @@ static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
                else
                        seq_puts(m, nfs_infop->nostr);
        }
+       switch (nfss->client->cl_xprt->prot) {
+               case IPPROTO_TCP:
+                       proto = "tcp";
+                       break;
+               case IPPROTO_UDP:
+                       proto = "udp";
+                       break;
+               default:
+                       snprintf(buf, sizeof(buf), "%u", nfss->client->cl_xprt->prot);
+                       proto = buf;
+       }
+       seq_printf(m, ",proto=%s", proto);
+       seq_printf(m, ",timeo=%lu", 10U * nfss->retrans_timeo / HZ);
+       seq_printf(m, ",retrans=%u", nfss->retrans_count);
+}
+
+static int nfs_show_options(struct seq_file *m, struct vfsmount *mnt)
+{
+       struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
+
+       nfs_show_mount_options(m, nfss, 0);
+
        seq_puts(m, ",addr=");
        seq_escape(m, nfss->hostname, " \t\n\\");
+
+       return 0;
+}
+
+static int nfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
+{
+       int i, cpu;
+       struct nfs_server *nfss = NFS_SB(mnt->mnt_sb);
+       struct rpc_auth *auth = nfss->client->cl_auth;
+       struct nfs_iostats totals = { };
+
+       seq_printf(m, "statvers=%s", NFS_IOSTAT_VERS);
+
+       /*
+        * Display all mount option settings
+        */
+       seq_printf(m, "\n\topts:\t");
+       seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? "ro" : "rw");
+       seq_puts(m, mnt->mnt_sb->s_flags & MS_SYNCHRONOUS ? ",sync" : "");
+       seq_puts(m, mnt->mnt_sb->s_flags & MS_NOATIME ? ",noatime" : "");
+       seq_puts(m, mnt->mnt_sb->s_flags & MS_NODIRATIME ? ",nodiratime" : "");
+       nfs_show_mount_options(m, nfss, 1);
+
+       seq_printf(m, "\n\tage:\t%lu", (jiffies - nfss->mount_time) / HZ);
+
+       seq_printf(m, "\n\tcaps:\t");
+       seq_printf(m, "caps=0x%x", nfss->caps);
+       seq_printf(m, ",wtmult=%d", nfss->wtmult);
+       seq_printf(m, ",dtsize=%d", nfss->dtsize);
+       seq_printf(m, ",bsize=%d", nfss->bsize);
+       seq_printf(m, ",namelen=%d", nfss->namelen);
+
+#ifdef CONFIG_NFS_V4
+       if (nfss->rpc_ops->version == 4) {
+               seq_printf(m, "\n\tnfsv4:\t");
+               seq_printf(m, "bm0=0x%x", nfss->attr_bitmask[0]);
+               seq_printf(m, ",bm1=0x%x", nfss->attr_bitmask[1]);
+               seq_printf(m, ",acl=0x%x", nfss->acl_bitmask);
+       }
+#endif
+
+       /*
+        * Display security flavor in effect for this mount
+        */
+       seq_printf(m, "\n\tsec:\tflavor=%d", auth->au_ops->au_flavor);
+       if (auth->au_flavor)
+               seq_printf(m, ",pseudoflavor=%d", auth->au_flavor);
+
+       /*
+        * Display superblock I/O counters
+        */
+       for_each_possible_cpu(cpu) {
+               struct nfs_iostats *stats;
+
+               preempt_disable();
+               stats = per_cpu_ptr(nfss->io_stats, cpu);
+
+               for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
+                       totals.events[i] += stats->events[i];
+               for (i = 0; i < __NFSIOS_BYTESMAX; i++)
+                       totals.bytes[i] += stats->bytes[i];
+
+               preempt_enable();
+       }
+
+       seq_printf(m, "\n\tevents:\t");
+       for (i = 0; i < __NFSIOS_COUNTSMAX; i++)
+               seq_printf(m, "%lu ", totals.events[i]);
+       seq_printf(m, "\n\tbytes:\t");
+       for (i = 0; i < __NFSIOS_BYTESMAX; i++)
+               seq_printf(m, "%Lu ", totals.bytes[i]);
+       seq_printf(m, "\n");
+
+       rpc_print_iostats(m, nfss->client);
+
        return 0;
 }
 
+/**
+ * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
+ */
+int nfs_sync_mapping(struct address_space *mapping)
+{
+       int ret;
+
+       if (mapping->nrpages == 0)
+               return 0;
+       unmap_mapping_range(mapping, 0, 0, 0);
+       ret = filemap_write_and_wait(mapping);
+       if (ret != 0)
+               goto out;
+       ret = nfs_wb_all(mapping->host);
+out:
+       return ret;
+}
+
 /*
  * Invalidate the local caches
  */
-void
-nfs_zap_caches(struct inode *inode)
+static void nfs_zap_caches_locked(struct inode *inode)
 {
        struct nfs_inode *nfsi = NFS_I(inode);
        int mode = inode->i_mode;
 
+       nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
+
        NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
        NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
 
        memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
        if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
-               nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS;
+               nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
        else
-               nfsi->flags |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS;
+               nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
+}
+
+void nfs_zap_caches(struct inode *inode)
+{
+       spin_lock(&inode->i_lock);
+       nfs_zap_caches_locked(inode);
+       spin_unlock(&inode->i_lock);
+}
+
+static void nfs_zap_acl_cache(struct inode *inode)
+{
+       void (*clear_acl_cache)(struct inode *);
+
+       clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
+       if (clear_acl_cache != NULL)
+               clear_acl_cache(inode);
+       spin_lock(&inode->i_lock);
+       NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
+       spin_unlock(&inode->i_lock);
 }
 
 /*
- * Invalidate, but do not unhash, the inode
+ * Invalidate, but do not unhash, the inode.
+ * NB: must be called with inode->i_lock held!
  */
-static void
-nfs_invalidate_inode(struct inode *inode)
+static void nfs_invalidate_inode(struct inode *inode)
 {
-       umode_t save_mode = inode->i_mode;
-
-       make_bad_inode(inode);
-       inode->i_mode = save_mode;
-       nfs_zap_caches(inode);
+       set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
+       nfs_zap_caches_locked(inode);
 }
 
 struct nfs_find_desc {
@@ -667,7 +853,7 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
                .fh     = fh,
                .fattr  = fattr
        };
-       struct inode *inode = NULL;
+       struct inode *inode = ERR_PTR(-ENOENT);
        unsigned long hash;
 
        if ((fattr->valid & NFS_ATTR_FATTR) == 0)
@@ -680,8 +866,11 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
 
        hash = nfs_fattr_to_ino_t(fattr);
 
-       if (!(inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc)))
+       inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
+       if (inode == NULL) {
+               inode = ERR_PTR(-ENOMEM);
                goto out_no_inode;
+       }
 
        if (inode->i_state & I_NEW) {
                struct nfs_inode *nfsi = NFS_I(inode);
@@ -696,7 +885,7 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
                /* Why so? Because we want revalidate for devices/FIFOs, and
                 * that's precisely what we have in nfs_file_inode_operations.
                 */
-               inode->i_op = &nfs_file_inode_operations;
+               inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
                if (S_ISREG(inode->i_mode)) {
                        inode->i_fop = &nfs_file_operations;
                        inode->i_data.a_ops = &nfs_file_aops;
@@ -706,13 +895,14 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
                        inode->i_fop = &nfs_dir_operations;
                        if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
                            && fattr->size <= NFS_LIMIT_READDIRPLUS)
-                               NFS_FLAGS(inode) |= NFS_INO_ADVISE_RDPLUS;
+                               set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
                } else if (S_ISLNK(inode->i_mode))
                        inode->i_op = &nfs_symlink_inode_operations;
                else
                        init_special_inode(inode, inode->i_mode, fattr->rdev);
 
-               nfsi->read_cache_jiffies = fattr->timestamp;
+               nfsi->read_cache_jiffies = fattr->time_start;
+               nfsi->last_updated = jiffies;
                inode->i_atime = fattr->atime;
                inode->i_mtime = fattr->mtime;
                inode->i_ctime = fattr->ctime;
@@ -751,7 +941,7 @@ out:
        return inode;
 
 out_no_inode:
-       printk("nfs_fhget: iget failed\n");
+       dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
        goto out;
 }
 
@@ -764,6 +954,8 @@ nfs_setattr(struct dentry *dentry, struct iattr *attr)
        struct nfs_fattr fattr;
        int error;
 
+       nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
+
        if (attr->ia_valid & ATTR_SIZE) {
                if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
                        attr->ia_valid &= ~ATTR_SIZE;
@@ -776,71 +968,112 @@ nfs_setattr(struct dentry *dentry, struct iattr *attr)
 
        lock_kernel();
        nfs_begin_data_update(inode);
-       /* Write all dirty data if we're changing file permissions or size */
-       if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE)) != 0) {
-               if (filemap_fdatawrite(inode->i_mapping) == 0)
-                       filemap_fdatawait(inode->i_mapping);
-               nfs_wb_all(inode);
-       }
+       /* Write all dirty data */
+       filemap_write_and_wait(inode->i_mapping);
+       nfs_wb_all(inode);
+       /*
+        * Return any delegations if we're going to change ACLs
+        */
+       if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
+               nfs_inode_return_delegation(inode);
        error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
-       if (error == 0) {
+       if (error == 0)
                nfs_refresh_inode(inode, &fattr);
+       nfs_end_data_update(inode);
+       unlock_kernel();
+       return error;
+}
+
+/**
+ * nfs_setattr_update_inode - Update inode metadata after a setattr call.
+ * @inode: pointer to struct inode
+ * @attr: pointer to struct iattr
+ *
+ * Note: we do this in the *proc.c in order to ensure that
+ *       it works for things like exclusive creates too.
+ */
+void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
+{
+       if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
                if ((attr->ia_valid & ATTR_MODE) != 0) {
-                       int mode;
-                       mode = inode->i_mode & ~S_IALLUGO;
-                       mode |= attr->ia_mode & S_IALLUGO;
+                       int mode = attr->ia_mode & S_IALLUGO;
+                       mode |= inode->i_mode & ~S_IALLUGO;
                        inode->i_mode = mode;
                }
                if ((attr->ia_valid & ATTR_UID) != 0)
                        inode->i_uid = attr->ia_uid;
                if ((attr->ia_valid & ATTR_GID) != 0)
                        inode->i_gid = attr->ia_gid;
-               if ((attr->ia_valid & ATTR_XID) != 0)
+               if ((attr->ia_valid & ATTR_XID) && IS_TAGXID(inode))
                        inode->i_xid = attr->ia_xid;
-               if ((attr->ia_valid & ATTR_SIZE) != 0) {
-                       inode->i_size = attr->ia_size;
-                       vmtruncate(inode, attr->ia_size);
-               }
+               spin_lock(&inode->i_lock);
+               NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
+               spin_unlock(&inode->i_lock);
+       }
+       if ((attr->ia_valid & ATTR_SIZE) != 0) {
+               nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
+               inode->i_size = attr->ia_size;
+               vmtruncate(inode, attr->ia_size);
        }
-       if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
-               NFS_FLAGS(inode) |= NFS_INO_INVALID_ACCESS;
-       nfs_end_data_update(inode);
-       unlock_kernel();
-       return error;
+}
+
+static int nfs_wait_schedule(void *word)
+{
+       if (signal_pending(current))
+               return -ERESTARTSYS;
+       schedule();
+       return 0;
 }
 
 /*
  * Wait for the inode to get unlocked.
- * (Used for NFS_INO_LOCKED and NFS_INO_REVALIDATING).
  */
-static int
-nfs_wait_on_inode(struct inode *inode, int flag)
+static int nfs_wait_on_inode(struct inode *inode)
 {
        struct rpc_clnt *clnt = NFS_CLIENT(inode);
        struct nfs_inode *nfsi = NFS_I(inode);
-
+       sigset_t oldmask;
        int error;
-       if (!(NFS_FLAGS(inode) & flag))
-               return 0;
-       atomic_inc(&inode->i_count);
-       error = nfs_wait_event(clnt, nfsi->nfs_i_wait,
-                               !(NFS_FLAGS(inode) & flag));
-       iput(inode);
+
+       rpc_clnt_sigmask(clnt, &oldmask);
+       error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
+                                       nfs_wait_schedule, TASK_INTERRUPTIBLE);
+       rpc_clnt_sigunmask(clnt, &oldmask);
+
        return error;
 }
 
+static void nfs_wake_up_inode(struct inode *inode)
+{
+       struct nfs_inode *nfsi = NFS_I(inode);
+
+       clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
+       smp_mb__after_clear_bit();
+       wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
+}
+
 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
 {
        struct inode *inode = dentry->d_inode;
-       struct nfs_inode *nfsi = NFS_I(inode);
-       int need_atime = nfsi->flags & NFS_INO_INVALID_ATIME;
+       int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
        int err;
 
-       if (__IS_FLG(inode, MS_NOATIME))
-               need_atime = 0;
-       else if (__IS_FLG(inode, MS_NODIRATIME) && S_ISDIR(inode->i_mode))
+       /* Flush out writes to the server in order to update c/mtime */
+       nfs_sync_inode_wait(inode, 0, 0, FLUSH_NOCOMMIT);
+
+       /*
+        * We may force a getattr if the user cares about atime.
+        *
+        * Note that we only have to check the vfsmount flags here:
+        *  - NFS always sets S_NOATIME by so checking it would give a
+        *    bogus result
+        *  - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
+        *    no point in checking those.
+        */
+       if ((mnt->mnt_flags & MNT_NOATIME) ||
+           ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
                need_atime = 0;
-       /* We may force a getattr if the user cares about atime */
+
        if (need_atime)
                err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
        else
@@ -850,7 +1083,7 @@ int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
        return err;
 }
 
-struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rpc_cred *cred)
+static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
 {
        struct nfs_open_context *ctx;
 
@@ -858,11 +1091,12 @@ struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, struct rp
        if (ctx != NULL) {
                atomic_set(&ctx->count, 1);
                ctx->dentry = dget(dentry);
+               ctx->vfsmnt = mntget(mnt);
                ctx->cred = get_rpccred(cred);
                ctx->state = NULL;
                ctx->lockowner = current->files;
                ctx->error = 0;
-               init_waitqueue_head(&ctx->waitq);
+               ctx->dir_cookie = 0;
        }
        return ctx;
 }
@@ -888,6 +1122,7 @@ void put_nfs_open_context(struct nfs_open_context *ctx)
                if (ctx->cred != NULL)
                        put_rpccred(ctx->cred);
                dput(ctx->dentry);
+               mntput(ctx->vfsmnt);
                kfree(ctx);
        }
 }
@@ -896,7 +1131,7 @@ void put_nfs_open_context(struct nfs_open_context *ctx)
  * Ensure that mmap has a recent RPC credential for use when writing out
  * shared pages
  */
-void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
+static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
 {
        struct inode *inode = filp->f_dentry->d_inode;
        struct nfs_inode *nfsi = NFS_I(inode);
@@ -907,13 +1142,18 @@ void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
        spin_unlock(&inode->i_lock);
 }
 
-struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
+/*
+ * Given an inode, search for an open context with the desired characteristics
+ */
+struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
 {
        struct nfs_inode *nfsi = NFS_I(inode);
        struct nfs_open_context *pos, *ctx = NULL;
 
        spin_lock(&inode->i_lock);
        list_for_each_entry(pos, &nfsi->open_files, list) {
+               if (cred != NULL && pos->cred != cred)
+                       continue;
                if ((pos->mode & mode) == mode) {
                        ctx = get_nfs_open_context(pos);
                        break;
@@ -923,7 +1163,7 @@ struct nfs_open_context *nfs_find_open_context(struct inode *inode, int mode)
        return ctx;
 }
 
-void nfs_file_clear_open_context(struct file *filp)
+static void nfs_file_clear_open_context(struct file *filp)
 {
        struct inode *inode = filp->f_dentry->d_inode;
        struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
@@ -948,22 +1188,18 @@ int nfs_open(struct inode *inode, struct file *filp)
        cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
        if (IS_ERR(cred))
                return PTR_ERR(cred);
-       ctx = alloc_nfs_open_context(filp->f_dentry, cred);
+       ctx = alloc_nfs_open_context(filp->f_vfsmnt, filp->f_dentry, cred);
        put_rpccred(cred);
        if (ctx == NULL)
                return -ENOMEM;
        ctx->mode = filp->f_mode;
        nfs_file_set_open_context(filp, ctx);
        put_nfs_open_context(ctx);
-       if ((filp->f_mode & FMODE_WRITE) != 0)
-               nfs_begin_data_update(inode);
        return 0;
 }
 
 int nfs_release(struct inode *inode, struct file *filp)
 {
-       if ((filp->f_mode & FMODE_WRITE) != 0)
-               nfs_end_data_update(inode);
        nfs_file_clear_open_context(filp);
        return 0;
 }
@@ -978,8 +1214,6 @@ __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
        int              status = -ESTALE;
        struct nfs_fattr fattr;
        struct nfs_inode *nfsi = NFS_I(inode);
-       unsigned long verifier;
-       unsigned int flags;
 
        dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
                inode->i_sb->s_id, (long long)NFS_FILEID(inode));
@@ -990,21 +1224,20 @@ __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
        if (NFS_STALE(inode))
                goto out_nowait;
 
-       while (NFS_REVALIDATING(inode)) {
-               status = nfs_wait_on_inode(inode, NFS_INO_REVALIDATING);
-               if (status < 0)
-                       goto out_nowait;
-               if (NFS_ATTRTIMEO(inode) == 0)
-                       continue;
-               if (NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME))
-                       continue;
-               status = NFS_STALE(inode) ? -ESTALE : 0;
-               goto out_nowait;
+       status = nfs_wait_on_inode(inode);
+       if (status < 0)
+               goto out;
+       if (NFS_STALE(inode)) {
+               status = -ESTALE;
+               /* Do we trust the cached ESTALE? */
+               if (NFS_ATTRTIMEO(inode) != 0) {
+                       if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ATIME)) {
+                               /* no */
+                       } else
+                               goto out;
+               }
        }
-       NFS_FLAGS(inode) |= NFS_INO_REVALIDATING;
 
-       /* Protect against RPC races by saving the change attribute */
-       verifier = nfs_save_change_attribute(inode);
        status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
        if (status != 0) {
                dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
@@ -1013,48 +1246,34 @@ __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
                if (status == -ESTALE) {
                        nfs_zap_caches(inode);
                        if (!S_ISDIR(inode->i_mode))
-                               NFS_FLAGS(inode) |= NFS_INO_STALE;
+                               set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
                }
                goto out;
        }
 
-       status = nfs_update_inode(inode, &fattr, verifier);
+       spin_lock(&inode->i_lock);
+       status = nfs_update_inode(inode, &fattr);
        if (status) {
+               spin_unlock(&inode->i_lock);
                dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
                         inode->i_sb->s_id,
                         (long long)NFS_FILEID(inode), status);
                goto out;
        }
-       flags = nfsi->flags;
-       /*
-        * We may need to keep the attributes marked as invalid if
-        * we raced with nfs_end_attr_update().
-        */
-       if (verifier == nfsi->cache_change_attribute)
-               nfsi->flags &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
-       /* Do the page cache invalidation */
-       if (flags & NFS_INO_INVALID_DATA) {
-               if (S_ISREG(inode->i_mode)) {
-                       if (filemap_fdatawrite(inode->i_mapping) == 0)
-                               filemap_fdatawait(inode->i_mapping);
-                       nfs_wb_all(inode);
-               }
-               nfsi->flags &= ~NFS_INO_INVALID_DATA;
-               invalidate_inode_pages2(inode->i_mapping);
-               memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
-               dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
-                               inode->i_sb->s_id,
-                               (long long)NFS_FILEID(inode));
-               /* This ensures we revalidate dentries */
-               nfsi->cache_change_attribute++;
-       }
+       spin_unlock(&inode->i_lock);
+
+       nfs_revalidate_mapping(inode, inode->i_mapping);
+
+       if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
+               nfs_zap_acl_cache(inode);
+
        dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
                inode->i_sb->s_id,
                (long long)NFS_FILEID(inode));
 
-out:
-       NFS_FLAGS(inode) &= ~NFS_INO_REVALIDATING;
-       wake_up(&nfsi->nfs_i_wait);
+ out:
+       nfs_wake_up_inode(inode);
+
  out_nowait:
        unlock_kernel();
        return status;
@@ -1078,12 +1297,43 @@ int nfs_attribute_timeout(struct inode *inode)
  */
 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
 {
-       if (!(NFS_FLAGS(inode) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
+       nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
+       if (!(NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
                        && !nfs_attribute_timeout(inode))
                return NFS_STALE(inode) ? -ESTALE : 0;
        return __nfs_revalidate_inode(server, inode);
 }
 
+/**
+ * nfs_revalidate_mapping - Revalidate the pagecache
+ * @inode - pointer to host inode
+ * @mapping - pointer to mapping
+ */
+void nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
+{
+       struct nfs_inode *nfsi = NFS_I(inode);
+
+       if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
+               nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
+               if (S_ISREG(inode->i_mode))
+                       nfs_sync_mapping(mapping);
+               invalidate_inode_pages2(mapping);
+
+               spin_lock(&inode->i_lock);
+               nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
+               if (S_ISDIR(inode->i_mode)) {
+                       memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
+                       /* This ensures we revalidate child dentries */
+                       nfsi->cache_change_attribute = jiffies;
+               }
+               spin_unlock(&inode->i_lock);
+
+               dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
+                               inode->i_sb->s_id,
+                               (long long)NFS_FILEID(inode));
+       }
+}
+
 /**
  * nfs_begin_data_update
  * @inode - pointer to inode
@@ -1106,39 +1356,46 @@ void nfs_end_data_update(struct inode *inode)
        struct nfs_inode *nfsi = NFS_I(inode);
 
        if (!nfs_have_delegation(inode, FMODE_READ)) {
-               /* Mark the attribute cache for revalidation */
-               nfsi->flags |= NFS_INO_INVALID_ATTR;
-               /* Directories and symlinks: invalidate page cache too */
-               if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
-                       nfsi->flags |= NFS_INO_INVALID_DATA;
+               /* Directories and symlinks: invalidate page cache */
+               if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
+                       spin_lock(&inode->i_lock);
+                       nfsi->cache_validity |= NFS_INO_INVALID_DATA;
+                       spin_unlock(&inode->i_lock);
+               }
        }
-       nfsi->cache_change_attribute ++;
+       nfsi->cache_change_attribute = jiffies;
        atomic_dec(&nfsi->data_updates);
 }
 
-/**
- * nfs_end_data_update_defer
- * @inode - pointer to inode
- * Declare end of the operations that will update file data
- * This will defer marking the inode as needing revalidation
- * unless there are no other pending updates.
- */
-void nfs_end_data_update_defer(struct inode *inode)
+static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 {
        struct nfs_inode *nfsi = NFS_I(inode);
 
-       if (atomic_dec_and_test(&nfsi->data_updates)) {
-               /* Mark the attribute cache for revalidation */
-               nfsi->flags |= NFS_INO_INVALID_ATTR;
-               /* Directories and symlinks: invalidate page cache too */
-               if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
-                       nfsi->flags |= NFS_INO_INVALID_DATA;
-               nfsi->cache_change_attribute ++;
+       if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
+                       && nfsi->change_attr == fattr->pre_change_attr) {
+               nfsi->change_attr = fattr->change_attr;
+               nfsi->cache_change_attribute = jiffies;
+       }
+
+       /* If we have atomic WCC data, we may update some attributes */
+       if ((fattr->valid & NFS_ATTR_WCC) != 0) {
+               if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
+                       memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
+                       nfsi->cache_change_attribute = jiffies;
+               }
+               if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
+                       memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
+                       nfsi->cache_change_attribute = jiffies;
+               }
+               if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
+                       inode->i_size = fattr->size;
+                       nfsi->cache_change_attribute = jiffies;
+               }
        }
 }
 
 /**
- * nfs_refresh_inode - verify consistency of the inode attribute cache
+ * nfs_check_inode_attributes - verify consistency of the inode attribute cache
  * @inode - pointer to inode
  * @fattr - updated attributes
  *
@@ -1146,56 +1403,50 @@ void nfs_end_data_update_defer(struct inode *inode)
  * so that fattr carries weak cache consistency data, then it may
  * also update the ctime/mtime/change_attribute.
  */
-int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
+static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
 {
        struct nfs_inode *nfsi = NFS_I(inode);
        loff_t cur_size, new_isize;
        int data_unstable;
        uid_t uid;
        gid_t gid;
-       xid_t xid = 0;
-
-       /* Do we hold a delegation? */
-       if (nfs_have_delegation(inode, FMODE_READ))
-               return 0;
-
-       /* Are we in the process of updating data on the server? */
-       data_unstable = nfs_caches_unstable(inode);
+       xid_t xid;
 
-       if (fattr->valid & NFS_ATTR_FATTR_V4) {
-               if ((fattr->valid & NFS_ATTR_PRE_CHANGE) != 0
-                               && nfsi->change_attr == fattr->pre_change_attr)
-                       nfsi->change_attr = fattr->change_attr;
-               if (!data_unstable && nfsi->change_attr != fattr->change_attr)
-                       nfsi->flags |= NFS_INO_INVALID_ATTR;
-       }
 
        if ((fattr->valid & NFS_ATTR_FATTR) == 0)
                return 0;
 
        /* Has the inode gone and changed behind our back? */
        if (nfsi->fileid != fattr->fileid
-                       || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
+                       || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
                return -EIO;
+       }
 
-       cur_size = i_size_read(inode);
-       new_isize = nfs_size_to_loff_t(fattr->size);
+       /* Are we in the process of updating data on the server? */
+       data_unstable = nfs_caches_unstable(inode);
 
-       /* If we have atomic WCC data, we may update some attributes */
-       if ((fattr->valid & NFS_ATTR_WCC) != 0) {
-               if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
-                       memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
-               if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime))
-                       memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
+       /* Do atomic weak cache consistency updates */
+       nfs_wcc_update_inode(inode, fattr);
+
+       if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0) {
+               if (nfsi->change_attr == fattr->change_attr)
+                       goto out;
+               nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
+               if (!data_unstable)
+                       nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
        }
 
        /* Verify a few of the more important attributes */
-       if (!data_unstable) {
-               if (!timespec_equal(&inode->i_mtime, &fattr->mtime)
-                               || cur_size != new_isize)
-                       nfsi->flags |= NFS_INO_INVALID_ATTR;
-       } else if (S_ISREG(inode->i_mode) && new_isize > cur_size)
-                       nfsi->flags |= NFS_INO_INVALID_ATTR;
+       if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
+               nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
+               if (!data_unstable)
+                       nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
+       }
+
+       cur_size = i_size_read(inode);
+       new_isize = nfs_size_to_loff_t(fattr->size);
+       if (cur_size != new_isize && nfsi->npages == 0)
+               nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 
        uid = INOXID_UID(XID_TAG(inode), fattr->uid, fattr->gid);
        gid = INOXID_GID(XID_TAG(inode), fattr->uid, fattr->gid);
@@ -1206,19 +1457,72 @@ int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
                        || inode->i_uid != uid
                        || inode->i_gid != gid
                        || inode->i_xid != xid)
-               nfsi->flags |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
+               nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
 
        /* Has the link count changed? */
        if (inode->i_nlink != fattr->nlink)
-               nfsi->flags |= NFS_INO_INVALID_ATTR;
+               nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
 
+out:
        if (!timespec_equal(&inode->i_atime, &fattr->atime))
-               nfsi->flags |= NFS_INO_INVALID_ATIME;
+               nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
 
-       nfsi->read_cache_jiffies = fattr->timestamp;
+       nfsi->read_cache_jiffies = fattr->time_start;
        return 0;
 }
 
+/**
+ * nfs_refresh_inode - try to update the inode attribute cache
+ * @inode - pointer to inode
+ * @fattr - updated attributes
+ *
+ * Check that an RPC call that returned attributes has not overlapped with
+ * other recent updates of the inode metadata, then decide whether it is
+ * safe to do a full update of the inode attributes, or whether just to
+ * call nfs_check_inode_attributes.
+ */
+int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
+{
+       struct nfs_inode *nfsi = NFS_I(inode);
+       int status;
+
+       if ((fattr->valid & NFS_ATTR_FATTR) == 0)
+               return 0;
+       spin_lock(&inode->i_lock);
+       nfsi->cache_validity &= ~NFS_INO_REVAL_PAGECACHE;
+       if (time_after(fattr->time_start, nfsi->last_updated))
+               status = nfs_update_inode(inode, fattr);
+       else
+               status = nfs_check_inode_attributes(inode, fattr);
+
+       spin_unlock(&inode->i_lock);
+       return status;
+}
+
+/**
+ * nfs_post_op_update_inode - try to update the inode attribute cache
+ * @inode - pointer to inode
+ * @fattr - updated attributes
+ *
+ * After an operation that has changed the inode metadata, mark the
+ * attribute cache as being invalid, then try to update it.
+ */
+int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
+{
+       struct nfs_inode *nfsi = NFS_I(inode);
+       int status = 0;
+
+       spin_lock(&inode->i_lock);
+       if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
+               nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS;
+               goto out;
+       }
+       status = nfs_update_inode(inode, fattr);
+out:
+       spin_unlock(&inode->i_lock);
+       return status;
+}
+
 /*
  * Many nfs protocol calls return the new file attributes after
  * an operation.  Here we update the inode to reflect the state
@@ -1231,17 +1535,15 @@ int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
  *
  * A very similar scenario holds for the dir cache.
  */
-static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsigned long verifier)
+static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 {
        struct nfs_inode *nfsi = NFS_I(inode);
-       __u64           new_size;
-       loff_t          new_isize;
+       loff_t cur_isize, new_isize;
        unsigned int    invalid = 0;
-       loff_t          cur_isize;
-       int data_unstable;
-       uid_t           uid;
-       gid_t           gid;
-       xid_t           xid = 0;
+       int data_stable;
+       uid_t uid;
+       gid_t gid;
+       xid_t xid;
 
        dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
                        __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
@@ -1250,14 +1552,8 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsign
        if ((fattr->valid & NFS_ATTR_FATTR) == 0)
                return 0;
 
-       if (nfsi->fileid != fattr->fileid) {
-               printk(KERN_ERR "%s: inode number mismatch\n"
-                      "expected (%s/0x%Lx), got (%s/0x%Lx)\n",
-                      __FUNCTION__,
-                      inode->i_sb->s_id, (long long)nfsi->fileid,
-                      inode->i_sb->s_id, (long long)fattr->fileid);
-               goto out_err;
-       }
+       if (nfsi->fileid != fattr->fileid)
+               goto out_fileid;
 
        /*
         * Make sure the inode's type hasn't changed.
@@ -1268,60 +1564,53 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsign
        /*
         * Update the read time so we don't revalidate too often.
         */
-       nfsi->read_cache_jiffies = fattr->timestamp;
+       nfsi->read_cache_jiffies = fattr->time_start;
+       nfsi->last_updated = jiffies;
 
        /* Are we racing with known updates of the metadata on the server? */
-       data_unstable = ! nfs_verify_change_attribute(inode, verifier);
+       data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
+       if (data_stable)
+               nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME);
+
+       /* Do atomic weak cache consistency updates */
+       nfs_wcc_update_inode(inode, fattr);
 
-       /* Check if the file size agrees */
-       new_size = fattr->size;
+       /* Check if our cached file size is stale */
        new_isize = nfs_size_to_loff_t(fattr->size);
        cur_isize = i_size_read(inode);
-       if (cur_isize != new_size) {
-#ifdef NFS_DEBUG_VERBOSE
-               printk(KERN_DEBUG "NFS: isize change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
-#endif
-               /*
-                * If we have pending writebacks, things can get
-                * messy.
-                */
-               if (S_ISREG(inode->i_mode) && data_unstable) {
-                       if (new_isize > cur_isize) {
+       if (new_isize != cur_isize) {
+               /* Do we perhaps have any outstanding writes? */
+               if (nfsi->npages == 0) {
+                       /* No, but did we race with nfs_end_data_update()? */
+                       if (data_stable) {
                                inode->i_size = new_isize;
-                               invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
+                               invalid |= NFS_INO_INVALID_DATA;
                        }
-               } else {
+                       invalid |= NFS_INO_INVALID_ATTR;
+               } else if (new_isize > cur_isize) {
                        inode->i_size = new_isize;
                        invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
                }
+               nfsi->cache_change_attribute = jiffies;
+               dprintk("NFS: isize change on server for file %s/%ld\n",
+                               inode->i_sb->s_id, inode->i_ino);
        }
 
-       /*
-        * Note: we don't check inode->i_mtime since pipes etc.
-        *       can change this value in VFS without requiring a
-        *       cache revalidation.
-        */
+       /* Check if the mtime agrees */
        if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
                memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
-#ifdef NFS_DEBUG_VERBOSE
-               printk(KERN_DEBUG "NFS: mtime change on %s/%ld\n", inode->i_sb->s_id, inode->i_ino);
-#endif
-               if (!data_unstable)
-                       invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
+               dprintk("NFS: mtime change on server for file %s/%ld\n",
+                               inode->i_sb->s_id, inode->i_ino);
+               invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
+               nfsi->cache_change_attribute = jiffies;
        }
 
-       if ((fattr->valid & NFS_ATTR_FATTR_V4)
-           && nfsi->change_attr != fattr->change_attr) {
-#ifdef NFS_DEBUG_VERBOSE
-               printk(KERN_DEBUG "NFS: change_attr change on %s/%ld\n",
-                      inode->i_sb->s_id, inode->i_ino);
-#endif
-               nfsi->change_attr = fattr->change_attr;
-               if (!data_unstable)
-                       invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS;
+       /* If ctime has changed we should definitely clear access+acl caches */
+       if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
+               invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
+               memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
+               nfsi->cache_change_attribute = jiffies;
        }
-
-       memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
        memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
 
        uid = INOXID_UID(XID_TAG(inode), fattr->uid, fattr->gid);
@@ -1332,7 +1621,7 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsign
            inode->i_uid != uid ||
            inode->i_gid != gid ||
            inode->i_xid != xid)
-               invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS;
+               invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
 
        inode->i_mode = fattr->mode;
        inode->i_nlink = fattr->nlink;
@@ -1351,8 +1640,20 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsign
                inode->i_blksize = fattr->du.nfs2.blocksize;
        }
 
+       if ((fattr->valid & NFS_ATTR_FATTR_V4)) {
+               if (nfsi->change_attr != fattr->change_attr) {
+                       dprintk("NFS: change_attr change on server for file %s/%ld\n",
+                                       inode->i_sb->s_id, inode->i_ino);
+                       nfsi->change_attr = fattr->change_attr;
+                       invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
+                       nfsi->cache_change_attribute = jiffies;
+               } else
+                       invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA);
+       }
+
        /* Update attrtimeo value if we're out of the unstable period */
        if (invalid & NFS_INO_INVALID_ATTR) {
+               nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
                nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
                nfsi->attrtimeo_timestamp = jiffies;
        } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
@@ -1364,8 +1665,10 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsign
        if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
                                || S_ISLNK(inode->i_mode)))
                invalid &= ~NFS_INO_INVALID_DATA;
+       if (data_stable)
+               invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
        if (!nfs_have_delegation(inode, FMODE_READ))
-               nfsi->flags |= invalid;
+               nfsi->cache_validity |= invalid;
 
        return 0;
  out_changed:
@@ -1376,15 +1679,21 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr, unsign
        printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
                        __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
 #endif
+ out_err:
        /*
         * No need to worry about unhashing the dentry, as the
         * lookup validation will know that the inode is bad.
         * (But we fall through to invalidate the caches.)
         */
        nfs_invalidate_inode(inode);
- out_err:
-       NFS_FLAGS(inode) |= NFS_INO_STALE;
        return -ESTALE;
+
+ out_fileid:
+       printk(KERN_ERR "NFS: server %s error: fileid changed\n"
+               "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
+               NFS_SERVER(inode)->hostname, inode->i_sb->s_id,
+               (long long)nfsi->fileid, (long long)fattr->fileid);
+       goto out_err;
 }
 
 /*
@@ -1413,76 +1722,96 @@ static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
        int flags, const char *dev_name, void *raw_data)
 {
        int error;
-       struct nfs_server *server;
+       struct nfs_server *server = NULL;
        struct super_block *s;
        struct nfs_fh *root;
        struct nfs_mount_data *data = raw_data;
 
-       if (!data) {
-               printk("nfs_read_super: missing data argument\n");
-               return ERR_PTR(-EINVAL);
+       s = ERR_PTR(-EINVAL);
+       if (data == NULL) {
+               dprintk("%s: missing data argument\n", __FUNCTION__);
+               goto out_err;
        }
-
-       server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
-       if (!server)
-               return ERR_PTR(-ENOMEM);
-       memset(server, 0, sizeof(struct nfs_server));
-       /* Zero out the NFS state stuff */
-       init_nfsv4_state(server);
-
-       if (data->version != NFS_MOUNT_VERSION) {
-               printk("nfs warning: mount version %s than kernel\n",
-                       data->version < NFS_MOUNT_VERSION ? "older" : "newer");
-               if (data->version < 2)
+       if (data->version <= 0 || data->version > NFS_MOUNT_VERSION) {
+               dprintk("%s: bad mount version\n", __FUNCTION__);
+               goto out_err;
+       }
+       switch (data->version) {
+               case 1:
                        data->namlen = 0;
-               if (data->version < 3)
+               case 2:
                        data->bsize  = 0;
-               if (data->version < 4) {
-                       data->flags &= ~NFS_MOUNT_VER3;
+               case 3:
+                       if (data->flags & NFS_MOUNT_VER3) {
+                               dprintk("%s: mount structure version %d does not support NFSv3\n",
+                                               __FUNCTION__,
+                                               data->version);
+                               goto out_err;
+                       }
                        data->root.size = NFS2_FHSIZE;
                        memcpy(data->root.data, data->old_root.data, NFS2_FHSIZE);
-               }
-               if (data->version < 5)
-                       data->flags &= ~NFS_MOUNT_SECFLAVOUR;
+               case 4:
+                       if (data->flags & NFS_MOUNT_SECFLAVOUR) {
+                               dprintk("%s: mount structure version %d does not support strong security\n",
+                                               __FUNCTION__,
+                                               data->version);
+                               goto out_err;
+                       }
+               case 5:
+                       memset(data->context, 0, sizeof(data->context));
+       }
+#ifndef CONFIG_NFS_V3
+       /* If NFSv3 is not compiled in, return -EPROTONOSUPPORT */
+       s = ERR_PTR(-EPROTONOSUPPORT);
+       if (data->flags & NFS_MOUNT_VER3) {
+               dprintk("%s: NFSv3 not compiled into kernel\n", __FUNCTION__);
+               goto out_err;
        }
+#endif /* CONFIG_NFS_V3 */
+
+       s = ERR_PTR(-ENOMEM);
+       server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
+       if (!server)
+               goto out_err;
+       /* Zero out the NFS state stuff */
+       init_nfsv4_state(server);
+       server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
 
        root = &server->fh;
        if (data->flags & NFS_MOUNT_VER3)
                root->size = data->root.size;
        else
                root->size = NFS2_FHSIZE;
+       s = ERR_PTR(-EINVAL);
        if (root->size > sizeof(root->data)) {
-               printk("nfs_get_sb: invalid root filehandle\n");
-               kfree(server);
-               return ERR_PTR(-EINVAL);
+               dprintk("%s: invalid root filehandle\n", __FUNCTION__);
+               goto out_err;
        }
        memcpy(root->data, data->root.data, root->size);
 
        /* We now require that the mount process passes the remote address */
        memcpy(&server->addr, &data->addr, sizeof(server->addr));
        if (server->addr.sin_addr.s_addr == INADDR_ANY) {
-               printk("NFS: mount program didn't pass remote address!\n");
-               kfree(server);
-               return ERR_PTR(-EINVAL);
+               dprintk("%s: mount program didn't pass remote address!\n",
+                               __FUNCTION__);
+               goto out_err;
        }
 
-       s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
-
-       if (IS_ERR(s) || s->s_root) {
-               kfree(server);
-               return s;
+       /* Fire up rpciod if not yet running */
+       s = ERR_PTR(rpciod_up());
+       if (IS_ERR(s)) {
+               dprintk("%s: couldn't start rpciod! Error = %ld\n",
+                               __FUNCTION__, PTR_ERR(s));
+               goto out_err;
        }
 
-       s->s_flags = flags;
+       s = sget(fs_type, nfs_compare_super, nfs_set_super, server);
+       if (IS_ERR(s) || s->s_root)
+               goto out_rpciod_down;
 
-       /* Fire up rpciod if not yet running */
-       if (rpciod_up() != 0) {
-               printk(KERN_WARNING "NFS: couldn't start rpciod!\n");
-               kfree(server);
-               return ERR_PTR(-EIO);
-       }
+       s->s_flags = flags;
 
-       error = nfs_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
+       error = nfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
        if (error) {
                up_write(&s->s_umount);
                deactivate_super(s);
@@ -1490,6 +1819,11 @@ static struct super_block *nfs_get_sb(struct file_system_type *fs_type,
        }
        s->s_flags |= MS_ACTIVE;
        return s;
+out_rpciod_down:
+       rpciod_down();
+out_err:
+       kfree(server);
+       return s;
 }
 
 static void nfs_kill_super(struct super_block *s)
@@ -1498,18 +1832,20 @@ static void nfs_kill_super(struct super_block *s)
 
        kill_anon_super(s);
 
-       if (server->client != NULL && !IS_ERR(server->client))
+       if (!IS_ERR(server->client))
                rpc_shutdown_client(server->client);
-       if (server->client_sys != NULL && !IS_ERR(server->client_sys))
+       if (!IS_ERR(server->client_sys))
                rpc_shutdown_client(server->client_sys);
+       if (!IS_ERR(server->client_acl))
+               rpc_shutdown_client(server->client_acl);
 
        if (!(server->flags & NFS_MOUNT_NONLM))
                lockd_down();   /* release rpc.lockd */
 
        rpciod_down();          /* release rpciod */
 
-       if (server->hostname != NULL)
-               kfree(server->hostname);
+       nfs_free_iostats(server->io_stats);
+       kfree(server->hostname);
        kfree(server);
 }
 
@@ -1535,6 +1871,7 @@ static struct super_operations nfs4_sops = {
        .clear_inode    = nfs4_clear_inode,
        .umount_begin   = nfs_umount_begin,
        .show_options   = nfs_show_options,
+       .show_stats     = nfs_show_stats,
 };
 
 /*
@@ -1547,8 +1884,7 @@ static void nfs4_clear_inode(struct inode *inode)
        struct nfs_inode *nfsi = NFS_I(inode);
 
        /* If we are holding a delegation, return it! */
-       if (nfsi->delegation != NULL)
-               nfs_inode_return_delegation(inode);
+       nfs_inode_return_delegation(inode);
        /* First call standard NFS clear_inode() code */
        nfs_clear_inode(inode);
        /* Now clear out any remaining state */
@@ -1577,7 +1913,7 @@ static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data,
        struct rpc_clnt *clnt = NULL;
        struct rpc_timeout timeparms;
        rpc_authflavor_t authflavour;
-       int proto, err = -EIO;
+       int err = -EIO;
 
        sb->s_blocksize_bits = 0;
        sb->s_blocksize = 0;
@@ -1595,42 +1931,27 @@ static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data,
        server->acdirmax = data->acdirmax*HZ;
 
        server->rpc_ops = &nfs_v4_clientops;
-       /* Initialize timeout values */
 
-       timeparms.to_initval = data->timeo * HZ / 10;
-       timeparms.to_retries = data->retrans;
-       timeparms.to_exponential = 1;
-       if (!timeparms.to_retries)
-               timeparms.to_retries = 5;
+       nfs_init_timeout_values(&timeparms, data->proto, data->timeo, data->retrans);
 
-       proto = data->proto;
-       /* Which IP protocol do we use? */
-       switch (proto) {
-       case IPPROTO_TCP:
-               timeparms.to_maxval  = RPC_MAX_TCP_TIMEOUT;
-               if (!timeparms.to_initval)
-                       timeparms.to_initval = 600 * HZ / 10;
-               break;
-       case IPPROTO_UDP:
-               timeparms.to_maxval  = RPC_MAX_UDP_TIMEOUT;
-               if (!timeparms.to_initval)
-                       timeparms.to_initval = 11 * HZ / 10;
-               break;
-       default:
-               return -EINVAL;
-       }
+       server->retrans_timeo = timeparms.to_initval;
+       server->retrans_count = timeparms.to_retries;
 
        clp = nfs4_get_client(&server->addr.sin_addr);
        if (!clp) {
-               printk(KERN_WARNING "NFS: failed to create NFS4 client.\n");
+               dprintk("%s: failed to create NFS4 client.\n", __FUNCTION__);
                return -EIO;
        }
 
        /* Now create transport and client */
        authflavour = RPC_AUTH_UNIX;
        if (data->auth_flavourlen != 0) {
-               if (data->auth_flavourlen > 1)
-                       printk(KERN_INFO "NFS: cannot yet deal with multiple auth flavours.\n");
+               if (data->auth_flavourlen != 1) {
+                       dprintk("%s: Invalid number of RPC auth flavours %d.\n",
+                                       __FUNCTION__, data->auth_flavourlen);
+                       err = -EINVAL;
+                       goto out_fail;
+               }
                if (copy_from_user(&authflavour, data->auth_flavours, sizeof(authflavour))) {
                        err = -EFAULT;
                        goto out_fail;
@@ -1638,44 +1959,30 @@ static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data,
        }
 
        down_write(&clp->cl_sem);
-       if (clp->cl_rpcclient == NULL) {
-               xprt = xprt_create_proto(proto, &server->addr, &timeparms);
+       if (IS_ERR(clp->cl_rpcclient)) {
+               xprt = xprt_create_proto(data->proto, &server->addr, &timeparms);
                if (IS_ERR(xprt)) {
                        up_write(&clp->cl_sem);
-                       printk(KERN_WARNING "NFS: cannot create RPC transport.\n");
                        err = PTR_ERR(xprt);
+                       dprintk("%s: cannot create RPC transport. Error = %d\n",
+                                       __FUNCTION__, err);
                        goto out_fail;
                }
                clnt = rpc_create_client(xprt, server->hostname, &nfs_program,
                                server->rpc_ops->version, authflavour);
                if (IS_ERR(clnt)) {
                        up_write(&clp->cl_sem);
-                       printk(KERN_WARNING "NFS: cannot create RPC client.\n");
-                       xprt_destroy(xprt);
                        err = PTR_ERR(clnt);
+                       dprintk("%s: cannot create RPC client. Error = %d\n",
+                                       __FUNCTION__, err);
                        goto out_fail;
                }
                clnt->cl_intr     = 1;
                clnt->cl_softrtry = 1;
-               clnt->cl_chatty   = 1;
                clp->cl_rpcclient = clnt;
-               clp->cl_cred = rpcauth_lookupcred(clnt->cl_auth, 0);
-               if (IS_ERR(clp->cl_cred)) {
-                       up_write(&clp->cl_sem);
-                       err = PTR_ERR(clp->cl_cred);
-                       clp->cl_cred = NULL;
-                       goto out_fail;
-               }
                memcpy(clp->cl_ipaddr, server->ip_addr, sizeof(clp->cl_ipaddr));
                nfs_idmap_new(clp);
        }
-       if (list_empty(&clp->cl_superblocks)) {
-               err = nfs4_init_client(clp);
-               if (err != 0) {
-                       up_write(&clp->cl_sem);
-                       goto out_fail;
-               }
-       }
        list_add_tail(&server->nfs4_siblings, &clp->cl_superblocks);
        clnt = rpc_clone_client(clp->cl_rpcclient);
        if (!IS_ERR(clnt))
@@ -1684,21 +1991,26 @@ static int nfs4_fill_super(struct super_block *sb, struct nfs4_mount_data *data,
        clp = NULL;
 
        if (IS_ERR(clnt)) {
-               printk(KERN_WARNING "NFS: cannot create RPC client.\n");
-               return PTR_ERR(clnt);
+               err = PTR_ERR(clnt);
+               dprintk("%s: cannot create RPC client. Error = %d\n",
+                               __FUNCTION__, err);
+               return err;
        }
 
        server->client    = clnt;
 
        if (server->nfs4_state->cl_idmap == NULL) {
-               printk(KERN_WARNING "NFS: failed to create idmapper.\n");
+               dprintk("%s: failed to create idmapper.\n", __FUNCTION__);
                return -ENOMEM;
        }
 
        if (clnt->cl_auth->au_flavor != authflavour) {
-               if (rpcauth_create(authflavour, clnt) == NULL) {
-                       printk(KERN_WARNING "NFS: couldn't create credcache!\n");
-                       return -ENOMEM;
+               struct rpc_auth *auth;
+
+               auth = rpcauth_create(authflavour, clnt);
+               if (IS_ERR(auth)) {
+                       dprintk("%s: couldn't create credcache!\n", __FUNCTION__);
+                       return PTR_ERR(auth);
                }
        }
 
@@ -1741,8 +2053,7 @@ nfs_copy_user_string(char *dst, struct nfs_string *src, int maxlen)
                        return ERR_PTR(-ENOMEM);
        }
        if (copy_from_user(dst, src->data, maxlen)) {
-               if (p != NULL)
-                       kfree(p);
+               kfree(p);
                return ERR_PTR(-EFAULT);
        }
        dst[maxlen] = '\0';
@@ -1758,22 +2069,21 @@ static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
        struct nfs4_mount_data *data = raw_data;
        void *p;
 
-       if (!data) {
-               printk("nfs_read_super: missing data argument\n");
+       if (data == NULL) {
+               dprintk("%s: missing data argument\n", __FUNCTION__);
+               return ERR_PTR(-EINVAL);
+       }
+       if (data->version <= 0 || data->version > NFS4_MOUNT_VERSION) {
+               dprintk("%s: bad mount version\n", __FUNCTION__);
                return ERR_PTR(-EINVAL);
        }
 
-       server = kmalloc(sizeof(struct nfs_server), GFP_KERNEL);
+       server = kzalloc(sizeof(struct nfs_server), GFP_KERNEL);
        if (!server)
                return ERR_PTR(-ENOMEM);
-       memset(server, 0, sizeof(struct nfs_server));
        /* Zero out the NFS state stuff */
        init_nfsv4_state(server);
-
-       if (data->version != NFS4_MOUNT_VERSION) {
-               printk("nfs warning: mount version %s than kernel\n",
-                       data->version < NFS4_MOUNT_VERSION ? "older" : "newer");
-       }
+       server->client = server->client_sys = server->client_acl = ERR_PTR(-EINVAL);
 
        p = nfs_copy_user_string(NULL, &data->hostname, 256);
        if (IS_ERR(p))
@@ -1801,11 +2111,20 @@ static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
        }
        if (server->addr.sin_family != AF_INET ||
            server->addr.sin_addr.s_addr == INADDR_ANY) {
-               printk("NFS: mount program didn't pass remote IP address!\n");
+               dprintk("%s: mount program didn't pass remote IP address!\n",
+                               __FUNCTION__);
                s = ERR_PTR(-EINVAL);
                goto out_free;
        }
 
+       /* Fire up rpciod if not yet running */
+       s = ERR_PTR(rpciod_up());
+       if (IS_ERR(s)) {
+               dprintk("%s: couldn't start rpciod! Error = %ld\n",
+                               __FUNCTION__, PTR_ERR(s));
+               goto out_free;
+       }
+
        s = sget(fs_type, nfs4_compare_super, nfs_set_super, server);
 
        if (IS_ERR(s) || s->s_root)
@@ -1813,14 +2132,7 @@ static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
 
        s->s_flags = flags;
 
-       /* Fire up rpciod if not yet running */
-       if (rpciod_up() != 0) {
-               printk(KERN_WARNING "NFS: couldn't start rpciod!\n");
-               s = ERR_PTR(-EIO);
-               goto out_free;
-       }
-
-       error = nfs4_fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
+       error = nfs4_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
        if (error) {
                up_write(&s->s_umount);
                deactivate_super(s);
@@ -1831,10 +2143,8 @@ static struct super_block *nfs4_get_sb(struct file_system_type *fs_type,
 out_err:
        s = (struct super_block *)p;
 out_free:
-       if (server->mnt_path)
-               kfree(server->mnt_path);
-       if (server->hostname)
-               kfree(server->hostname);
+       kfree(server->mnt_path);
+       kfree(server->hostname);
        kfree(server);
        return s;
 }
@@ -1850,12 +2160,13 @@ static void nfs4_kill_super(struct super_block *sb)
 
        if (server->client != NULL && !IS_ERR(server->client))
                rpc_shutdown_client(server->client);
-       rpciod_down();          /* release rpciod */
 
        destroy_nfsv4_state(server);
 
-       if (server->hostname != NULL)
-               kfree(server->hostname);
+       rpciod_down();
+
+       nfs_free_iostats(server->io_stats);
+       kfree(server->hostname);
        kfree(server);
 }
 
@@ -1867,6 +2178,35 @@ static struct file_system_type nfs4_fs_type = {
        .fs_flags       = FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
 };
 
+static const int nfs_set_port_min = 0;
+static const int nfs_set_port_max = 65535;
+static int param_set_port(const char *val, struct kernel_param *kp)
+{
+       char *endp;
+       int num = simple_strtol(val, &endp, 0);
+       if (endp == val || *endp || num < nfs_set_port_min || num > nfs_set_port_max)
+               return -EINVAL;
+       *((int *)kp->arg) = num;
+       return 0;
+}
+
+module_param_call(callback_tcpport, param_set_port, param_get_int,
+                &nfs_callback_set_tcpport, 0644);
+
+static int param_set_idmap_timeout(const char *val, struct kernel_param *kp)
+{
+       char *endp;
+       int num = simple_strtol(val, &endp, 0);
+       int jif = num * HZ;
+       if (endp == val || *endp || num < 0 || jif < num)
+               return -EINVAL;
+       *((int *)kp->arg) = jif;
+       return 0;
+}
+
+module_param_call(idmap_cache_timeout, param_set_idmap_timeout, param_get_int,
+                &nfs_idmap_cache_timeout, 0644);
+
 #define nfs4_init_once(nfsi) \
        do { \
                INIT_LIST_HEAD(&(nfsi)->open_states); \
@@ -1874,8 +2214,25 @@ static struct file_system_type nfs4_fs_type = {
                nfsi->delegation_state = 0; \
                init_rwsem(&nfsi->rwsem); \
        } while(0)
-#define register_nfs4fs() register_filesystem(&nfs4_fs_type)
-#define unregister_nfs4fs() unregister_filesystem(&nfs4_fs_type)
+
+static inline int register_nfs4fs(void)
+{
+       int ret;
+
+       ret = nfs_register_sysctl();
+       if (ret != 0)
+               return ret;
+       ret = register_filesystem(&nfs4_fs_type);
+       if (ret != 0)
+               nfs_unregister_sysctl();
+       return ret;
+}
+
+static inline void unregister_nfs4fs(void)
+{
+       unregister_filesystem(&nfs4_fs_type);
+       nfs_unregister_sysctl();
+}
 #else
 #define nfs4_init_once(nfsi) \
        do { } while (0)
@@ -1902,7 +2259,16 @@ static struct inode *nfs_alloc_inode(struct super_block *sb)
        nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
        if (!nfsi)
                return NULL;
-       nfsi->flags = 0;
+       nfsi->flags = 0UL;
+       nfsi->cache_validity = 0UL;
+       nfsi->cache_change_attribute = jiffies;
+#ifdef CONFIG_NFS_V3_ACL
+       nfsi->acl_access = ERR_PTR(-EAGAIN);
+       nfsi->acl_default = ERR_PTR(-EAGAIN);
+#endif
+#ifdef CONFIG_NFS_V4
+       nfsi->nfs4_acl = NULL;
+#endif /* CONFIG_NFS_V4 */
        return &nfsi->vfs_inode;
 }
 
@@ -1927,7 +2293,6 @@ static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
                nfsi->ndirty = 0;
                nfsi->ncommit = 0;
                nfsi->npages = 0;
-               init_waitqueue_head(&nfsi->nfs_i_wait);
                nfs4_init_once(nfsi);
        }
 }
@@ -1936,7 +2301,8 @@ static int nfs_init_inodecache(void)
 {
        nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
                                             sizeof(struct nfs_inode),
-                                            0, SLAB_RECLAIM_ACCOUNT,
+                                            0, (SLAB_RECLAIM_ACCOUNT|
+                                               SLAB_MEM_SPREAD),
                                             init_once, NULL);
        if (nfs_inode_cachep == NULL)
                return -ENOMEM;
@@ -1992,11 +2358,11 @@ out:
 #ifdef CONFIG_PROC_FS
        rpc_proc_unregister("nfs");
 #endif
-       nfs_destroy_writepagecache();
 #ifdef CONFIG_NFS_DIRECTIO
-out0:
        nfs_destroy_directcache();
+out0:
 #endif
+       nfs_destroy_writepagecache();
 out1:
        nfs_destroy_readpagecache();
 out2: