Whamcloud - gitweb
Lproc-snmp code drop
[fs/lustre-release.git] / lustre / llite / super.c
index 8ea85d5..40ae623 100644 (file)
@@ -6,37 +6,49 @@
  * This code is issued under the GNU General Public License.
  * See the file COPYING in this distribution
  *
- * Copryright (C) 1996 Peter J. Braam <braam@stelias.com>
- * Copryright (C) 1999 Stelias Computing Inc. <braam@stelias.com>
- * Copryright (C) 1999 Seagate Technology Inc.
- * Copryright (C) 2001 Mountain View Data, Inc.
  * Copryright (C) 2002 Cluster File Systems, Inc.
- *
  */
 
 #define DEBUG_SUBSYSTEM S_LLITE
 
 #include <linux/module.h>
+#include <linux/random.h>
+#include <linux/version.h>
 #include <linux/lustre_lite.h>
 #include <linux/lustre_ha.h>
+#include <linux/obd_lov.h>
 #include <linux/lustre_dlm.h>
+#include <linux/init.h>
+#include <linux/fs.h>
+#include <linux/lprocfs_status.h>
+
 
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 kmem_cache_t *ll_file_data_slab;
 extern struct address_space_operations ll_aops;
 extern struct address_space_operations ll_dir_aops;
 struct super_operations ll_super_operations;
 
-extern int ll_recover(struct ptlrpc_client *);
+extern int ll_recover(struct recovd_data *, int);
 extern int ll_commitcbd_setup(struct ll_sb_info *);
 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
 
+extern int rd_dev_uuid(char* page, char **start, off_t off,
+                           int count, int *eof, void *data);
+
+extern int rd_dev_name(char* page, char **start, off_t off,
+                    int count, int *eof, void *data);
+
+
+extern lprocfs_vars_t status_var_nm_1[];
+
 static char *ll_read_opt(const char *opt, char *data)
 {
         char *value;
         char *retval;
         ENTRY;
 
-        CDEBUG(D_INFO, "option: %s, data %s\n", opt, data);
+        CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
         if ( strncmp(opt, data, strlen(opt)) )
                 RETURN(NULL);
         if ( (value = strchr(data, '=')) == NULL )
@@ -48,13 +60,24 @@ static char *ll_read_opt(const char *opt, char *data)
                 CERROR("out of memory!\n");
                 RETURN(NULL);
         }
-        
+
         memcpy(retval, value, strlen(value)+1);
         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
         RETURN(retval);
 }
 
-static void ll_options(char *options, char **ost, char **mds)
+static int ll_set_opt(const char *opt, char *data, int fl)
+{
+        ENTRY;
+
+        CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
+        if ( strncmp(opt, data, strlen(opt)) )
+                RETURN(0);
+        else
+                RETURN(fl);
+}
+
+static void ll_options(char *options, char **ost, char **mds, int *flags)
 {
         char *this_char;
         ENTRY;
@@ -67,28 +90,43 @@ static void ll_options(char *options, char **ost, char **mds)
         for (this_char = strtok (options, ",");
              this_char != NULL;
              this_char = strtok (NULL, ",")) {
-                CDEBUG(D_INFO, "this_char %s\n", this_char);
-                if ( (!*ost && (*ost = ll_read_opt("ost", this_char)))||
-                     (!*mds && (*mds = ll_read_opt("mds", this_char))) )
+                CDEBUG(D_SUPER, "this_char %s\n", this_char);
+                if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
+                     (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
+                     (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
+                      ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
                         continue;
         }
         EXIT;
 }
 
+#ifndef log2
+#define log2(n) ffz(~(n))
+#endif
+
 static struct super_block * ll_read_super(struct super_block *sb,
                                           void *data, int silent)
 {
         struct inode *root = 0;
+        struct obd_device *obd;
         struct ll_sb_info *sbi;
-        char *ost = NULL;
-        char *mds = NULL;
-        int devno;
+        char *osc = NULL;
+        char *mdc = NULL;
         int err;
         struct ll_fid rootfid;
-        __u64 last_committed, last_rcvd;
-        __u32 last_xid;
+        struct obd_statfs osfs;
         struct ptlrpc_request *request = NULL;
-        struct ll_inode_md md;
+        struct ptlrpc_connection *mdc_conn;
+        struct ll_read_inode2_cookie lic;
+        class_uuid_t uuid;
+
+        
+        /* Lprocfs variables */
+        char mnt_name[100];
+        char uuid_name[100];
+        lprocfs_vars_t d_vars[3];
+        
+
 
         ENTRY;
         MOD_INC_USE_COUNT;
@@ -99,100 +137,96 @@ static struct super_block * ll_read_super(struct super_block *sb,
                 RETURN(NULL);
         }
 
+        INIT_LIST_HEAD(&sbi->ll_conn_chain);
+        generate_random_uuid(uuid);
+        class_uuid_unparse(uuid, sbi->ll_sb_uuid);
+
         sb->u.generic_sbp = sbi;
 
-        ll_options(data, &ost, &mds);
+        ll_options(data, &osc, &mdc, &sbi->ll_flags);
 
-        if (!ost) {
-                CERROR("no ost\n");
+        if (!osc) {
+                CERROR("no osc\n");
                 GOTO(out_free, sb = NULL);
         }
 
-        if (!mds) {
-                CERROR("no mds\n");
+        if (!mdc) {
+                CERROR("no mdc\n");
                 GOTO(out_free, sb = NULL);
         }
 
-        devno = simple_strtoul(ost, NULL, 0);
-        if (devno >= MAX_OBD_DEVICES) {
-                CERROR("devno of %s too high\n", ost);
+        obd = class_uuid2obd(mdc);
+        if (!obd) {
+                CERROR("MDC %s: not setup or attached\n", mdc);
                 GOTO(out_free, sb = NULL);
         }
 
-        sbi->ll_osc_conn.oc_dev = &obd_dev[devno];
-        err = obd_connect(&sbi->ll_osc_conn);
+        err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid,
+                          ptlrpc_recovd, ll_recover);
         if (err) {
-                CERROR("cannot connect to %s: rc = %d\n", ost, err);
+                CERROR("cannot connect to %s: rc = %d\n", mdc, err);
                 GOTO(out_free, sb = NULL);
         }
 
-        devno = simple_strtoul(mds, NULL, 0);
-        if (devno >= MAX_OBD_DEVICES) {
-                CERROR("devno of %s too high\n", mds);
-                GOTO(out_free, sb = NULL);
-        }
+#warning Peter: is this the right place to raise the connection level?
+        mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
+        mdc_conn->c_level = LUSTRE_CONN_FULL;
+        list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
 
-        sbi->ll_mdc_conn.oc_dev = &obd_dev[devno];
-        err = obd_connect(&sbi->ll_mdc_conn);
-        if (err) {
-                CERROR("cannot connect to %s: rc = %d\n", mds, err);
-                GOTO(out_free, sb = NULL);
+        obd = class_uuid2obd(osc);
+        if (!obd) {
+                CERROR("OSC %s: not setup or attached\n", osc);
+                GOTO(out_mdc, sb = NULL);
         }
 
-#if 0
-        err = connmgr_connect(ptlrpc_connmgr, sbi->ll_mds_conn);
+        err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid,
+                          ptlrpc_recovd, ll_recover);
         if (err) {
-                CERROR("cannot connect to MDS: rc = %d\n", err);
-                GOTO(out_rpc, sb = NULL);
+                CERROR("cannot connect to %s: rc = %d\n", osc, err);
+                GOTO(out_mdc, sb = NULL);
         }
-#endif 
-
-        sbi2mdc(sbi)->mdc_conn->c_level = LUSTRE_CONN_FULL;
 
-        /* XXX: need to store the last_* values somewhere */
-        err = mdc_getstatus(&sbi->ll_mdc_conn,
-                            &rootfid, &last_committed, 
-                            &last_rcvd, &last_xid,
-                            &request);
+        err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
         if (err) {
                 CERROR("cannot mds_connect: rc = %d\n", err);
-                GOTO(out_disc, sb = NULL);
+                GOTO(out_mdc, sb = NULL);
         }
-        CDEBUG(D_SUPER, "rootfid %ld\n", (unsigned long)rootfid.id);
+        CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
         sbi->ll_rootino = rootfid.id;
 
-        sb->s_maxbytes = 1ULL << 36;
-        /* XXX get this with a get_info call (like we have in OBDFS),
-           this info call should return the blocksize of the MDS */
-        sb->s_blocksize = 4096;
-        sb->s_blocksize_bits = 12;
+        memset(&osfs, 0, sizeof(osfs));
+        err = obd_statfs(&sbi->ll_mdc_conn, &osfs);
+        sb->s_blocksize = osfs.os_bsize;
+        sb->s_blocksize_bits = log2(osfs.os_bsize);
         sb->s_magic = LL_SUPER_MAGIC;
+        sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
+
         sb->s_op = &ll_super_operations;
 
         /* make root inode */
-        err = mdc_getattr(&sbi->ll_mdc_conn,
-                          sbi->ll_rootino, S_IFDIR,
+        err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
         if (err) {
                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
-                GOTO(out_mdc, sb = NULL);
+                GOTO(out_request, sb = NULL);
         }
 
         /* initialize committed transaction callback daemon */
-        spin_lock_init(&sbi->ll_commitcbd_lock); 
+        spin_lock_init(&sbi->ll_commitcbd_lock);
         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
         sbi->ll_commitcbd_flags = 0;
         err = ll_commitcbd_setup(sbi);
         if (err) {
                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
-                GOTO(out_mdc, sb = NULL);
+                GOTO(out_request, sb = NULL);
         }
 
-        md.body = lustre_msg_buf(request->rq_repmsg, 0);
-        md.obdo = NULL;
-        root = iget4(sb, sbi->ll_rootino, NULL, &md);
-                     
+        lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
+        lic.lic_lmm = NULL;
+        LASSERT(sbi->ll_rootino != 0);
+        root = iget4(sb, sbi->ll_rootino, NULL, &lic);
+
         if (root) {
                 sb->s_root = d_alloc_root(root);
         } else {
@@ -200,24 +234,83 @@ static struct super_block * ll_read_super(struct super_block *sb,
                 GOTO(out_cdb, sb = NULL);
         }
 
-        ptlrpc_free_req(request);
+        ptlrpc_req_finished(request);
+        request = NULL;
+
+        
+        /* Register this mount instance with LProcFS */
+        
+        snprintf(mnt_name, 100, "mount_%s", sbi->ll_sb_uuid);
+        sbi->ll_mnt_root=lprocfs_reg_mnt(mnt_name);
+        if(!sbi->ll_mnt_root)
+                goto out_dev;
+
+        /* Add the static configuration info */
+        lprocfs_add_vars(sbi->ll_mnt_root, (lprocfs_vars_t*)status_var_nm_1,
+                         (void*)sb);
+
+         /* Add the dynamic configuration stuff */
+        /* MDC */
+        obd = class_uuid2obd(mdc);
+        
+        /* Reuse mnt_name */
+        
+        sprintf(mnt_name, "status/%s/common_name", obd->obd_type->typ_name);
+
+        memset(d_vars, 0, sizeof(d_vars));
+        d_vars[0].read_fptr=rd_dev_name;
+        d_vars[0].write_fptr=0;
+        d_vars[0].name=(char*)mnt_name;
+         
+        memset(uuid_name, '\0', strlen(uuid_name));
+        sprintf(uuid_name, "status/%s/uuid", obd->obd_type->typ_name);
+        d_vars[1].read_fptr=rd_dev_uuid;
+        d_vars[1].write_fptr=0;
+        d_vars[1].name=(char*)uuid_name;
+        
+        err=lprocfs_add_vars(sbi->ll_mnt_root, (lprocfs_vars_t*)d_vars, 
+                             (void*)obd);
+        if (err) {
+                CDEBUG(D_OTHER, "Unable to add fs proc dynamic variables");
+        }
 
+        /* OSC or LOV*/
+        obd = class_uuid2obd(osc);
+        /* Reuse mnt_name */
+        memset(mnt_name, '\0', strlen(mnt_name));
+        sprintf(mnt_name, "status/%s/common_name", obd->obd_type->typ_name);
+
+        memset(d_vars, 0, sizeof(d_vars));
+        d_vars[0].read_fptr=rd_dev_name;
+        d_vars[0].write_fptr=0;
+        d_vars[0].name=(char*)mnt_name;
+         
+        memset(uuid_name, '\0', strlen(uuid_name));
+        sprintf(uuid_name, "status/%s/uuid", obd->obd_type->typ_name);
+        d_vars[1].read_fptr=rd_dev_uuid;
+        d_vars[1].write_fptr=0;
+        d_vars[1].name=(char*)uuid_name;
+        
+        err=lprocfs_add_vars(sbi->ll_mnt_root, (lprocfs_vars_t*)d_vars, 
+                             (void*)obd);        
+        if (err) {
+                CDEBUG(D_OTHER, "Unable to add fs proc dynamic variables");
+        }
 out_dev:
-        if (mds)
-                OBD_FREE(mds, strlen(mds) + 1);
-        if (ost)
-                OBD_FREE(ost, strlen(ost) + 1);
+        if (mdc)
+                OBD_FREE(mdc, strlen(mdc) + 1);
+        if (osc)
+                OBD_FREE(osc, strlen(osc) + 1);
 
         RETURN(sb);
 
 out_cdb:
         ll_commitcbd_cleanup(sbi);
+out_request:
+        ptlrpc_req_finished(request);
+        obd_disconnect(&sbi->ll_osc_conn);
 out_mdc:
         obd_disconnect(&sbi->ll_mdc_conn);
-        obd_disconnect(&sbi->ll_osc_conn);
-out_disc:
-        ptlrpc_free_req(request);
-        obd_disconnect(&sbi->ll_osc_conn);
 out_free:
         OBD_FREE(sbi, sizeof(*sbi));
 
@@ -228,9 +321,22 @@ out_free:
 static void ll_put_super(struct super_block *sb)
 {
         struct ll_sb_info *sbi = ll_s2sbi(sb);
+        struct ll_fid rootfid;
         ENTRY;
+
+        list_del(&sbi->ll_conn_chain);
         ll_commitcbd_cleanup(sbi);
         obd_disconnect(&sbi->ll_osc_conn);
+
+        /* NULL request to force sync on the MDS, and get the last_committed
+         * value to flush remaining RPCs from the sending queue on client.
+         *
+         * XXX This should be an mdc_sync() call to sync the whole MDS fs,
+         *     which we can call for other reasons as well.
+         */
+        mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
+
+        lprocfs_dereg_mnt(sbi->ll_mnt_root);
         obd_disconnect(&sbi->ll_mdc_conn);
         OBD_FREE(sbi, sizeof(*sbi));
 
@@ -240,54 +346,82 @@ static void ll_put_super(struct super_block *sb)
 
 static void ll_clear_inode(struct inode *inode)
 {
+        struct ll_sb_info *sbi = ll_i2sbi(inode);
+        int rc;
+        ENTRY;
+
+        rc = mdc_cancel_unused(&sbi->ll_mdc_conn, inode, LDLM_FL_NO_CALLBACK);
+        if (rc < 0) {
+                CERROR("obd_cancel_unused: %d\n", rc);
+                /* XXX FIXME do something dramatic */
+        }
+
         if (atomic_read(&inode->i_count) == 0) {
-                struct obdo *oa = ll_i2info(inode)->lli_obdo;
-                if (oa) {
-                        obdo_free(oa);
-                        ll_i2info(inode)->lli_obdo = NULL;
+                struct ll_inode_info *lli = ll_i2info(inode);
+                struct lov_stripe_md *lsm = lli->lli_smd;
+                char *symlink_name = lli->lli_symlink_name;
+
+                if (lsm) {
+                        OBD_FREE(lsm, ll_ost_easize(inode->i_sb));
+                        lli->lli_smd = NULL;
                 }
-                if (ll_i2info(inode)->lli_symlink_name) {
-                        OBD_FREE(ll_i2info(inode)->lli_symlink_name,
-                                 strlen(ll_i2info(inode)->lli_symlink_name)+ 1);
-                        ll_i2info(inode)->lli_symlink_name = NULL;
+                if (symlink_name) {
+                        OBD_FREE(symlink_name, strlen(symlink_name) + 1);
+                        lli->lli_symlink_name = NULL;
                 }
         }
+
+        EXIT;
 }
 
 static void ll_delete_inode(struct inode *inode)
 {
-        if (S_ISREG(inode->i_mode)) { 
-                int err; 
-                struct obdo *oa; 
-                oa = ll_i2info(inode)->lli_obdo;
+        ENTRY;
+        if (S_ISREG(inode->i_mode)) {
+                int err;
+                struct obdo *oa;
+                struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 
-                if (!oa) {
-                        CERROR("no memory\n");
-                        GOTO(out, -ENOMEM);
+                if (!lsm)
+                        GOTO(out, -EINVAL);
+
+                if (lsm->lsm_object_id == 0) {
+                        CERROR("This really happens\n");
+                        /* No obdo was ever created */
+                        GOTO(out, 0);
                 }
 
-                err = obd_destroy(ll_i2obdconn(inode), oa);
-                CDEBUG(D_INODE, "obd destroy of %Ld error %d\n",
-                       (unsigned long long)oa->o_id, err);
+                oa = obdo_alloc();
+                if (oa == NULL)
+                        GOTO(out, -ENOMEM);
+
+                oa->o_id = lsm->lsm_object_id;
+                oa->o_easize = ll_mds_easize(inode->i_sb);
+                oa->o_mode = inode->i_mode;
+                oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLTYPE;
+
+                err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
                 obdo_free(oa);
+                CDEBUG(D_SUPER, "obd destroy of objid "LPX64" error %d\n",
+                       lsm->lsm_object_id, err);
         }
 out:
         clear_inode(inode);
+        EXIT;
 }
 
-/* like inode_setattr, but doesn't mark the inode dirty */ 
+/* like inode_setattr, but doesn't mark the inode dirty */
 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 {
         unsigned int ia_valid = attr->ia_valid;
         int error = 0;
 
-        if ((ia_valid & ATTR_SIZE) && trunc ) {
+        if ((ia_valid & ATTR_SIZE) && trunc) {
                 error = vmtruncate(inode, attr->ia_size);
                 if (error)
                         goto out;
-        } else if (ia_valid & ATTR_SIZE) { 
+        } else if (ia_valid & ATTR_SIZE)
                 inode->i_size = attr->ia_size;
-        }               
 
         if (ia_valid & ATTR_UID)
                 inode->i_uid = attr->ia_uid;
@@ -319,8 +453,7 @@ int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
         /* change incore inode */
         ll_attr2inode(inode, attr, do_trunc);
 
-        err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr,
-                          &request);
+        err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
         if (err)
                 CERROR("mdc_setattr fails (%d)\n", err);
 
@@ -331,97 +464,139 @@ int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 
 int ll_setattr(struct dentry *de, struct iattr *attr)
 {
+        int rc = inode_change_ok(de->d_inode, attr);
+
+        if (rc)
+                return rc;
+
         return ll_inode_setattr(de->d_inode, attr, 1);
 }
 
-static int ll_statfs(struct super_block *sb, struct statfs *buf)
+static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 {
-        struct statfs tmp;
-        int err;
+        struct ll_sb_info *sbi = ll_s2sbi(sb);
+        struct obd_statfs osfs;
+        int rc;
         ENTRY;
 
-        err = obd_statfs(&ll_s2sbi(sb)->ll_osc_conn, &tmp);
-        if (err) {
-                CERROR("obd_statfs fails (%d)\n", err);
-                RETURN(err);
-        }
-        memcpy(buf, &tmp, sizeof(*buf));
-#if 0
-        err = mdc_statfs(&sbi->ll_mds_client, sbi->ll_mds_conn, &tmp,
-                          &request);
-        if (err) {
-                CERROR("obd_statfs fails (%d)\n", err);
-                RETURN(err);
+        memset(sfs, 0, sizeof(*sfs));
+        rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
+        statfs_unpack(sfs, &osfs);
+        if (rc)
+                CERROR("mdc_statfs fails: rc = %d\n", rc);
+        else
+                CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
+                       " objects "LPU64"/"LPU64"\n",
+                       osfs.os_bavail, osfs.os_blocks,
+                       osfs.os_ffree, osfs.os_files);
+
+        /* temporary until mds_statfs returns statfs info for all OSTs */
+        if (!rc) {
+                rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
+                if (rc) {
+                        CERROR("obd_statfs fails: rc = %d\n", rc);
+                        GOTO(out, rc);
+                }
+                CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
+                       " objects "LPU64"/"LPU64"\n",
+                       osfs.os_bavail, osfs.os_blocks,
+                       osfs.os_ffree, osfs.os_files);
+
+                while (osfs.os_blocks > ~0UL) {
+                        sfs->f_bsize <<= 1;
+
+                        osfs.os_blocks >>= 1;
+                        osfs.os_bfree >>= 1;
+                        osfs.os_bavail >>= 1;
+                }
+                sfs->f_blocks = osfs.os_blocks;
+                sfs->f_bfree = osfs.os_bfree;
+                sfs->f_bavail = osfs.os_bavail;
+                if (osfs.os_ffree < (__u64)sfs->f_ffree)
+                        sfs->f_ffree = osfs.os_ffree;
         }
-       if (tmp.f_files < buf->f_files)
-                buf->f_files = tmp.f_files;
-       if (tmp.f_ffree < buf->f_ffree)
-                buf->f_ffree = tmp.f_ffree;
-        buf->f_namelen = tmp.f_namelen;
-#endif
-        CDEBUG(D_SUPER, "statfs returns avail %ld\n", tmp.f_bavail);
 
-        RETURN(err);
+out:
+        RETURN(rc);
 }
 
-static void inline ll_to_inode(struct inode *dst, struct ll_inode_md *md)
+void ll_update_inode(struct inode *inode, struct mds_body *body)
 {
-        struct mds_body *body = md->body;
-        struct ll_inode_info *ii = ll_i2info(dst);
+        if (body->valid & OBD_MD_FLID)
+                inode->i_ino = body->ino;
+        if (body->valid & OBD_MD_FLATIME)
+                inode->i_atime = body->atime;
+        if (body->valid & OBD_MD_FLMTIME)
+                inode->i_mtime = body->mtime;
+        if (body->valid & OBD_MD_FLCTIME)
+                inode->i_ctime = body->ctime;
+        if (body->valid & OBD_MD_FLMODE)
+                inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
+        if (body->valid & OBD_MD_FLTYPE)
+                inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
+        if (body->valid & OBD_MD_FLUID)
+                inode->i_uid = body->uid;
+        if (body->valid & OBD_MD_FLGID)
+                inode->i_gid = body->gid;
+        if (body->valid & OBD_MD_FLFLAGS)
+                inode->i_flags = body->flags;
+        if (body->valid & OBD_MD_FLNLINK)
+                inode->i_nlink = body->nlink;
+        if (body->valid & OBD_MD_FLGENER)
+                inode->i_generation = body->generation;
+        if (body->valid & OBD_MD_FLRDEV)
+                inode->i_rdev = body->rdev;
+        if (body->valid & OBD_MD_FLSIZE)
+                inode->i_size = body->size;
+}
+
+static void ll_read_inode2(struct inode *inode, void *opaque)
+{
+        struct ll_read_inode2_cookie *lic = opaque;
+        struct mds_body *body = lic->lic_body;
+        struct ll_inode_info *lli = ll_i2info(inode);
+        ENTRY;
+
+        sema_init(&lli->lli_open_sem, 1);
 
         /* core attributes first */
-        if ( body->valid & OBD_MD_FLID )
-                dst->i_ino = body->ino;
-        if ( body->valid & OBD_MD_FLATIME ) 
-                dst->i_atime = body->atime;
-        if ( body->valid & OBD_MD_FLMTIME ) 
-                dst->i_mtime = body->mtime;
-        if ( body->valid & OBD_MD_FLCTIME ) 
-                dst->i_ctime = body->ctime;
-        if ( body->valid & OBD_MD_FLSIZE ) 
-                dst->i_size = body->size;
-        if ( body->valid & OBD_MD_FLMODE ) 
-                dst->i_mode = body->mode;
-        if ( body->valid & OBD_MD_FLUID ) 
-                dst->i_uid = body->uid;
-        if ( body->valid & OBD_MD_FLGID ) 
-                dst->i_gid = body->gid;
-        if ( body->valid & OBD_MD_FLFLAGS ) 
-                dst->i_flags = body->flags;
-        if ( body->valid & OBD_MD_FLNLINK )
-                dst->i_nlink = body->nlink;
-        if ( body->valid & OBD_MD_FLGENER )
-                dst->i_generation = body->generation;
-
-        /* this will become more elaborate for striping etc */ 
-        if (md->obdo != NULL) {
-                ii->lli_obdo = obdo_alloc();
-                memcpy(ii->lli_obdo, md->obdo, sizeof(*md->obdo));
-        }
-#if 0
-
-        if (obdo_has_inline(oa)) {
-                if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
-                    S_ISFIFO(inode->i_mode)) {
-                        obd_rdev rdev = *((obd_rdev *)oa->o_inline);
-                        CDEBUG(D_INODE,
-                               "copying device %x from obdo to inode\n", rdev);
-                        init_special_inode(inode, inode->i_mode, rdev);
-                } else {
-                        CDEBUG(D_INFO, "copying inline from obdo to inode\n");
-                        memcpy(oinfo->lli_inline, oa->o_inline, OBD_INLINESZ);
+        ll_update_inode(inode, body);
+
+        //if (body->valid & OBD_MD_FLEASIZE)
+        if (lic && lic->lic_lmm) {
+                struct lov_mds_md *lmm = lic->lic_lmm;
+                int size;
+
+                /* XXX This should probably not be an error in the future,
+                 *     when we allow LOV OSTs to be added.
+                 */
+                if (lmm->lmm_easize != ll_mds_easize(inode->i_sb)) {
+                        CERROR("Striping metadata size error %ld\n",
+                               inode->i_ino);
+                        LBUG();
+                }
+                size = ll_ost_easize(inode->i_sb);
+                OBD_ALLOC(lli->lli_smd, size);
+                if (!lli->lli_smd) {
+                        CERROR("No memory for %d\n", size);
+                        LBUG();
                 }
-                oinfo->lli_flags |= OBD_FL_INLINEDATA;
+                lov_unpackmd(lli->lli_smd, lmm);
+        } else {
+                lli->lli_smd = NULL;
         }
-#endif 
-} /* ll_to_inode */
 
-static inline void ll_read_inode2(struct inode *inode, void *opaque)
-{
-        struct ll_inode_md *md = opaque;
-        
-        ENTRY;
-        ll_to_inode(inode, md);
+        /* Get the authoritative file size */
+        if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
+                int rc;
+
+                rc = ll_file_size(inode, lli->lli_smd);
+                if (rc) {
+                        CERROR("ll_file_size: %d\n", rc);
+                        /* FIXME: need to somehow prevent inode creation */
+                        LBUG();
+                }
+        }
 
         /* OIDEBUG(inode); */
 
@@ -432,19 +607,54 @@ static inline void ll_read_inode2(struct inode *inode, void *opaque)
                 EXIT;
         } else if (S_ISDIR(inode->i_mode)) {
                 inode->i_op = &ll_dir_inode_operations;
-                inode->i_fop = &ll_dir_operations; 
+                inode->i_fop = &ll_dir_operations;
                 inode->i_mapping->a_ops = &ll_dir_aops;
                 EXIT;
         } else if (S_ISLNK(inode->i_mode)) {
                 inode->i_op = &ll_fast_symlink_inode_operations;
                 EXIT;
         } else {
-                init_special_inode(inode, inode->i_mode,
-                                   ((int *)ll_i2info(inode)->lli_inline)[0]);
+                init_special_inode(inode, inode->i_mode, inode->i_rdev);
                 EXIT;
         }
+}
 
-        return;
+static inline void invalidate_request_list(struct list_head *req_list)
+{
+        struct list_head *tmp, *n;
+        list_for_each_safe(tmp, n, req_list) {
+                struct ptlrpc_request *req =
+                        list_entry(tmp, struct ptlrpc_request, rq_list);
+                CERROR("invalidating req xid "LPD64" op %d to %s:%d\n",
+                       (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
+                       req->rq_connection->c_remote_uuid,
+                       req->rq_import->imp_client->cli_request_portal);
+                req->rq_flags |= PTL_RPC_FL_ERR;
+                wake_up(&req->rq_wait_for_rep);
+        }
+}
+
+void ll_umount_begin(struct super_block *sb)
+{
+        struct ll_sb_info *sbi = ll_s2sbi(sb);
+        struct list_head *ctmp;
+
+        ENTRY;
+
+        list_for_each(ctmp, &sbi->ll_conn_chain) {
+                struct ptlrpc_connection *conn;
+                conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
+
+                spin_lock(&conn->c_lock);
+                /* XXX should just be dealing with imports, probably through
+                 * XXX iocontrol, need next-gen recovery! */
+                conn->c_flags |= CONN_INVALID;
+                invalidate_request_list(&conn->c_sending_head);
+                invalidate_request_list(&conn->c_delayed_head);
+                spin_unlock(&conn->c_lock);
+        }
+
+        EXIT;
 }
 
 /* exported operations */
@@ -454,7 +664,8 @@ struct super_operations ll_super_operations =
         clear_inode: ll_clear_inode,
         delete_inode: ll_delete_inode,
         put_super: ll_put_super,
-        statfs: ll_statfs
+        statfs: ll_statfs,
+        umount_begin: ll_umount_begin
 };
 
 struct file_system_type lustre_lite_fs_type = {
@@ -463,10 +674,10 @@ struct file_system_type lustre_lite_fs_type = {
 
 static int __init init_lustre_lite(void)
 {
-        printk(KERN_INFO "Lustre Lite 0.0.1, braam@clusterfs.com\n");
+        printk(KERN_INFO "Lustre Lite 0.5.14, info@clusterfs.com\n");
         ll_file_data_slab = kmem_cache_create("ll_file_data",
                                               sizeof(struct ll_file_data), 0,
-                                               SLAB_HWCACHE_ALIGN, NULL, NULL);
+                                              SLAB_HWCACHE_ALIGN, NULL, NULL);
         if (ll_file_data_slab == NULL)
                 return -ENOMEM;
         return register_filesystem(&lustre_lite_fs_type);
@@ -478,9 +689,10 @@ static void __exit exit_lustre_lite(void)
         kmem_cache_destroy(ll_file_data_slab);
 }
 
-MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
+MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 MODULE_LICENSE("GPL");
 
 module_init(init_lustre_lite);
 module_exit(exit_lustre_lite);
+#endif