Whamcloud - gitweb
Commit OST AMD support to HEAD so we can being running with a common code base.
[fs/lustre-release.git] / lustre / llite / llite_lib.c
index 00fa891..906e978 100644 (file)
@@ -32,6 +32,7 @@
 #include <linux/lustre_ha.h>
 #include <linux/lustre_dlm.h>
 #include <linux/lprocfs_status.h>
+#include <linux/lustre_snap.h>
 #include "llite_internal.h"
 
 kmem_cache_t *ll_file_data_slab;
@@ -53,12 +54,22 @@ struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
         if (!sbi)
                 RETURN(NULL);
 
-        spin_lock_init(&sbi->ll_pglist_lock);
+        spin_lock_init(&sbi->ll_lock);
         INIT_LIST_HEAD(&sbi->ll_pglist);
         sbi->ll_pglist_gen = 0;
+        sbi->ll_max_read_ahead_pages = SBI_DEFAULT_RA_MAX;
         INIT_LIST_HEAD(&sbi->ll_conn_chain);
         INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
-        ll_s2sbi(sb) = sbi;
+        INIT_LIST_HEAD(&sbi->ll_mnt_list);
+        sema_init(&sbi->ll_gns_sem, 1);
+        init_completion(&sbi->ll_gns_completion);
+        sbi->ll_gns_state = LL_GNS_STATE_IDLE;
+        sbi->ll_gns_timer.data = (unsigned long)sbi;
+        sbi->ll_gns_timer.function = ll_gns_timer_callback;
+        init_timer(&sbi->ll_gns_timer);
+        INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
+
+        ll_set_sbi(sb, sbi);
 
         generate_random_uuid(uuid);
         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
@@ -70,12 +81,32 @@ void lustre_free_sbi(struct super_block *sb)
         struct ll_sb_info *sbi = ll_s2sbi(sb);
         ENTRY;
 
-        if (sbi != NULL)
+        if (sbi != NULL) {
+                list_del(&sbi->ll_gns_sbi_head);
+                del_timer(&sbi->ll_gns_timer);
                 OBD_FREE(sbi, sizeof(*sbi));
-        ll_s2sbi(sb) = NULL;
+        }
+        ll_set_sbi(sb, NULL);
         EXIT;
 }
 
+int lustre_init_ea_size(struct ll_sb_info *sbi)
+{
+        struct lov_desc desc;
+        int rc, valsize;
+        
+        /* OSC may have larger ea size */
+        valsize = sizeof(desc);
+        rc = obd_get_info(sbi->ll_osc_exp, strlen("lovdesc") + 1, "lovdesc", 
+                          &valsize, &desc);
+        if (rc)
+                RETURN(rc);
+        obd_init_ea_size(sbi->ll_mdc_exp, obd_size_diskmd(sbi->ll_osc_exp, NULL),
+                         desc.ld_tgt_count*sizeof(struct llog_cookie));
+
+        RETURN(rc);
+}
+
 int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
 {
         struct inode *root = 0;
@@ -103,9 +134,7 @@ int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
                         CERROR("could not register mount in /proc/lustre");
         }
 
-        mdc_init_ea_size(obd, osc);
-
-        err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid);
+        err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, 0);
         if (err == -EBUSY) {
                 CERROR("An MDS (mdc %s) is performing recovery, of which this"
                        " client is not a part.  Please wait for recovery to "
@@ -126,9 +155,10 @@ int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
         sb->s_blocksize_bits = log2(osfs.os_bsize);
         sb->s_magic = LL_SUPER_MAGIC;
         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
+       
+        devno = get_uuid2int(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid, 
+                             strlen(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid));
 
-        devno = get_uuid2int(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid,
-                             strlen(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid));
         sb->s_dev = devno;
 
         obd = class_name2obd(osc);
@@ -137,7 +167,7 @@ int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
                 GOTO(out_mdc, err);
         }
 
-        err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid);
+        err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, 0);
         if (err == -EBUSY) {
                 CERROR("An OST (osc %s) is performing recovery, of which this"
                        " client is not a part.  Please wait for recovery to "
@@ -149,7 +179,9 @@ int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
         }
         sbi->ll_osc_exp = class_conn2export(&osc_conn);
 
-        err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
+        lustre_init_ea_size(sbi);
+
+        err = md_getstatus(sbi->ll_mdc_exp, &rootfid);
         if (err) {
                 CERROR("cannot mds_connect: rc = %d\n", err);
                 GOTO(out_osc, err);
@@ -161,14 +193,15 @@ int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
 
         /* make root inode
          * XXX: move this to after cbd setup? */
-        err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
-                          OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
+        err = md_getattr(sbi->ll_mdc_exp, &rootfid,
+                         OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS, 0, &request);
         if (err) {
-                CERROR("mdc_getattr failed for root: rc = %d\n", err);
+                CERROR("md_getattr failed for root: rc = %d\n", err);
                 GOTO(out_osc, err);
         }
 
-        err = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
+        err = mdc_req2lustre_md(sbi->ll_mdc_exp, request, 0, 
+                                sbi->ll_osc_exp, &md);
         if (err) {
                 CERROR("failed to understand root inode md: rc = %d\n",err);
                 ptlrpc_req_finished (request);
@@ -192,15 +225,23 @@ int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
                 GOTO(out_root, err);
         }
 
-       /* making vm readahead 0 for 2.4.x. In the case of 2.6.x, 
-          backing dev info assigned to inode mapping is used for
-          determining maximal readahead. */
+        ll_gns_add_timer(sbi);
+
+        /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
+           backing dev info assigned to inode mapping is used for
+           determining maximal readahead. */
 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
         /* bug 2805 - set VM readahead to zero */
         vm_max_readahead = vm_min_readahead = 0;
 #endif
 
         sb->s_root = d_alloc_root(root);
+
+#ifdef S_PDIROPS
+        CWARN("Enabling PDIROPS\n");
+        sb->s_flags |= S_PDIROPS;
+#endif
+
         RETURN(err);
 
 out_root:
@@ -221,6 +262,8 @@ void lustre_common_put_super(struct super_block *sb)
         struct hlist_node *tmp, *next;
         ENTRY;
 
+        ll_gns_del_timer(sbi);
+
         ll_close_thread_shutdown(sbi->ll_lcq);
 
         list_del(&sbi->ll_conn_chain);
@@ -238,6 +281,8 @@ void lustre_common_put_super(struct super_block *sb)
         spin_lock(&dcache_lock);
         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
+                CWARN("orphan dentry %*s (%p) at unmount\n",
+                      dentry->d_name.len, dentry->d_name.name, dentry);
                 shrink_dcache_parent(dentry);
         }
         spin_unlock(&dcache_lock);
@@ -280,7 +325,8 @@ int ll_set_opt(const char *opt, char *data, int fl)
                 RETURN(fl);
 }
 
-void ll_options(char *options, char **ost, char **mdc, int *flags)
+void ll_options(char *options, char **ost, char **mdc, int *flags, 
+                char **clone_opts)
 {
         char *this_char;
 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
@@ -305,12 +351,16 @@ void ll_options(char *options, char **ost, char **mdc, int *flags)
                         continue;
                 if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
                         continue;
+                if (!*clone_opts && (*clone_opts = ll_read_opt("clone", 
+                                                                this_char))) 
+                        continue; 
                 if (!(*flags & LL_SBI_NOLCK) &&
                     ((*flags) = (*flags) |
                                 ll_set_opt("nolock", this_char,
                                            LL_SBI_NOLCK)))
                         continue;
         }
+        
         EXIT;
 }
 
@@ -321,6 +371,7 @@ void ll_lli_init(struct ll_inode_info *lli)
         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
         spin_lock_init(&lli->lli_lock);
         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
+        lli->lli_inode_magic = LLI_INODE_MAGIC;
 }
 
 int ll_fill_super(struct super_block *sb, void *data, int silent)
@@ -328,6 +379,7 @@ int ll_fill_super(struct super_block *sb, void *data, int silent)
         struct ll_sb_info *sbi;
         char *osc = NULL;
         char *mdc = NULL;
+        char *clone_opts = NULL;
         int err;
         ENTRY;
 
@@ -338,7 +390,7 @@ int ll_fill_super(struct super_block *sb, void *data, int silent)
                 RETURN(-ENOMEM);
 
         sbi->ll_flags |= LL_SBI_READAHEAD;
-        ll_options(data, &osc, &mdc, &sbi->ll_flags);
+        ll_options(data, &osc, &mdc, &sbi->ll_flags, &clone_opts);
 
         if (!osc) {
                 CERROR("no osc\n");
@@ -359,12 +411,14 @@ out:
                 OBD_FREE(mdc, strlen(mdc) + 1);
         if (osc)
                 OBD_FREE(osc, strlen(osc) + 1);
+        if (clone_opts)
+                OBD_FREE(clone_opts, strlen(clone_opts) + 1);
 
         RETURN(err);
 } /* ll_read_super */
 
-int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
-                       struct config_llog_instance *cfg, int allow_recov)
+static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
+                              struct config_llog_instance *cfg, int allow_recov)
 {
         struct lustre_cfg lcfg;
         struct portals_cfg pcfg;
@@ -446,7 +500,7 @@ int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
         if (err)
                 GOTO(out_cleanup, err);
 
-        err = obd_connect(&mdc_conn, obd, &mdc_uuid);
+        err = obd_connect(&mdc_conn, obd, &mdc_uuid, 0);
         if (err) {
                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
                 GOTO(out_cleanup, err);
@@ -454,18 +508,10 @@ int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
 
         exp = class_conn2export(&mdc_conn);
 
-        ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
-#if 1
-        rc = class_config_parse_llog(ctxt, profile, cfg);
-#else
-        /*
-         * For debugging, it's useful to just dump the log
-         */
-        rc = class_config_dump_llog(ctxt, profile, cfg);
-#endif
-        if (rc) {
-                CERROR("class_config_parse_llog failed: rc = %d\n", rc);
-        }
+        ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
+        rc = class_config_process_llog(ctxt, profile, cfg);
+        if (rc)
+                CERROR("class_config_process_llog failed: rc = %d\n", rc);
 
         err = obd_disconnect(exp, 0);
 
@@ -529,7 +575,7 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
                 struct config_llog_instance cfg;
                 int len;
 
-                if (!lmd->lmd_mds) {
+                if (lmd->lmd_mds[0] == '\0') {
                         CERROR("no mds name\n");
                         GOTO(out_free, err = -EINVAL);
                 }
@@ -553,7 +599,6 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
                 if (err < 0) {
                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
-
                         GOTO(out_free, err);
                 }
 
@@ -589,7 +634,7 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
 
         if (err)
                 GOTO(out_free, err);
-
+        
 out_dev:
         if (mdc)
                 OBD_FREE(mdc, strlen(mdc) + 1);
@@ -604,6 +649,7 @@ out_free:
                 int err;
 
                 if (sbi->ll_instance != NULL) {
+                        struct lustre_mount_data *lmd = sbi->ll_lmd;
                         char * cln_prof;
                         struct config_llog_instance cfg;
 
@@ -611,10 +657,9 @@ out_free:
                         cfg.cfg_uuid = sbi->ll_sb_uuid;
 
                         OBD_ALLOC(cln_prof, len);
-                        sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
+                        sprintf(cln_prof, "%s-clean", lmd->lmd_profile);
 
-                        err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg,
-                                                 0);
+                        err = lustre_process_log(lmd, cln_prof, &cfg, 0);
                         if (err < 0)
                                 CERROR("Unable to process log: %s\n", cln_prof);
                         OBD_FREE(cln_prof, len);
@@ -670,7 +715,6 @@ void lustre_put_super(struct super_block *sb)
         obd = NULL;
 
         lustre_common_put_super(sb);
-
         if (sbi->ll_lmd != NULL) {
                 char * cln_prof;
                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
@@ -707,26 +751,101 @@ void lustre_put_super(struct super_block *sb)
         EXIT;
 } /* lustre_put_super */
 
+int ll_process_config_update(struct ll_sb_info *sbi, int clean)
+{
+        struct obd_export *mdc_exp = sbi->ll_mdc_exp;
+        struct lustre_mount_data *lmd = sbi->ll_lmd;
+        struct llog_ctxt *ctxt;
+        struct config_llog_instance cfg;
+        char *profile = lmd->lmd_profile, *name = NULL;
+        int rc, namelen =  0, version;
+        ENTRY;
+
+        if (profile == NULL)
+                RETURN(0);
+        if (lmd == NULL) {
+                CERROR("Client not mounted with zero-conf; cannot process "
+                       "update log.\n");
+                RETURN(0);
+        }
+
+        rc = ldlm_cli_cancel_unused(mdc_exp->exp_obd->obd_namespace, NULL,
+                                    LDLM_FL_CONFIG_CHANGE, NULL);
+        if (rc != 0)
+                CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
+
+        rc = obd_cancel_unused(sbi->ll_osc_exp, NULL, LDLM_FL_CONFIG_CHANGE,
+                               NULL);
+        if (rc != 0)
+                CWARN("obd_cancel_unused(lov): %d\n", rc);
+
+        cfg.cfg_instance = sbi->ll_instance;
+        cfg.cfg_uuid = sbi->ll_sb_uuid;
+        cfg.cfg_local_nid = lmd->lmd_local_nid;
+
+        namelen = strlen(profile) + 20; /* -clean-######### */
+        OBD_ALLOC(name, namelen);
+        if (name == NULL)
+                RETURN(-ENOMEM);
+
+        if (clean) {
+                version = sbi->ll_config_version - 1;
+                sprintf(name, "%s-clean-%d", profile, version);
+        } else {
+                version = sbi->ll_config_version + 1;
+                sprintf(name, "%s-%d", profile, version);
+        }
+
+        CWARN("Applying configuration log %s\n", name);
+
+        ctxt = llog_get_context(&mdc_exp->exp_obd->obd_llogs,
+                                LLOG_CONFIG_REPL_CTXT);
+        rc = class_config_process_llog(ctxt, name, &cfg);
+        if (rc == 0)
+                sbi->ll_config_version = version;
+        CWARN("Finished applying configuration log %s: %d\n", name, rc);
+
+        if (rc == 0 && clean == 0) {
+                struct lov_desc desc;
+                int rc, valsize;
+                valsize = sizeof(desc);
+                rc = obd_get_info(sbi->ll_osc_exp, strlen("lovdesc") + 1,
+                                  "lovdesc", &valsize, &desc);
+
+                rc = obd_init_ea_size(mdc_exp,
+                                      obd_size_diskmd(sbi->ll_osc_exp, NULL),
+                                      (desc.ld_tgt_count *
+                                       sizeof(struct llog_cookie)));
+        }
+        OBD_FREE(name, namelen);
+        RETURN(rc);
+}
 
 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 {
-        struct inode *inode;
+        struct inode *inode = NULL;
         l_lock(&lock->l_resource->lr_namespace->ns_lock);
-        if (lock->l_ast_data)
-                inode = igrab(lock->l_ast_data);
-        else
-                inode = NULL;
+        if (lock->l_ast_data) {
+                struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
+                if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
+                        inode = igrab(lock->l_ast_data);
+                } else {
+                        CERROR("DEBUG: l_ast_data %p is bogus: magic %x\n",
+                               lock->l_ast_data, lli->lli_inode_magic);
+                }
+        }
         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
         return inode;
 }
 
-static int null_if_equal(struct ldlm_lock *lock, void *data)
+int null_if_equal(struct ldlm_lock *lock, void *data)
 {
-        if (data == lock->l_ast_data)
+        if (data == lock->l_ast_data) {
                 lock->l_ast_data = NULL;
 
-        if (lock->l_req_mode != lock->l_granted_mode)
-                return LDLM_ITER_STOP;
+                if (lock->l_req_mode != lock->l_granted_mode)
+                        LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
+        }
 
         return LDLM_ITER_CONTINUE;
 }
@@ -741,9 +860,10 @@ void ll_clear_inode(struct inode *inode)
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
                inode->i_generation, inode);
 
+        lli->lli_inode_magic = LLI_INODE_DEAD;
         ll_inode2fid(&fid, inode);
         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
-        mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
+        md_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
 
         if (lli->lli_smd)
                 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
@@ -754,6 +874,13 @@ void ll_clear_inode(struct inode *inode)
                 lli->lli_smd = NULL;
         }
 
+        if (lli->lli_mea) {
+                /* FIXME: change cbdata for mea here */
+                obd_free_memmd(sbi->ll_mdc_exp,
+                               (struct lov_stripe_md **) &lli->lli_mea);
+                lli->lli_mea = NULL;
+        }
+
         if (lli->lli_symlink_name) {
                 OBD_FREE(lli->lli_symlink_name,
                          strlen(lli->lli_symlink_name) + 1);
@@ -833,17 +960,17 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr)
                 struct lustre_md md;
                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
 
-                rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
-                                 attr, NULL, 0, NULL, 0, &request);
-
+                rc = md_setattr(sbi->ll_mdc_exp, &op_data,
+                                attr, NULL, 0, NULL, 0, &request);
                 if (rc) {
                         ptlrpc_req_finished(request);
                         if (rc != -EPERM && rc != -EACCES)
-                                CERROR("mdc_setattr fails: rc = %d\n", rc);
+                                CERROR("md_setattr fails: rc = %d\n", rc);
                         RETURN(rc);
                 }
 
-                rc = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
+                rc = mdc_req2lustre_md(sbi->ll_mdc_exp, request, 0, 
+                                       sbi->ll_osc_exp, &md);
                 if (rc) {
                         ptlrpc_req_finished(request);
                         RETURN(rc);
@@ -852,7 +979,7 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr)
                 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
                  * but needed to set timestamps backwards on utime. */
                 inode_setattr(inode, attr);
-                ll_update_inode(inode, md.body, md.lsm);
+                ll_update_inode(inode, &md);
                 ptlrpc_req_finished(request);
 
                 if (!lsm || !S_ISREG(inode->i_mode)) {
@@ -905,7 +1032,7 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr)
                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
                                     ast_flags);
                 down(&inode->i_sem);
-                if (rc != ELDLM_OK)
+                if (rc != 0)
                         RETURN(rc);
 
                 rc = vmtruncate(inode, attr->ia_size);
@@ -928,7 +1055,8 @@ int ll_setattr_raw(struct inode *inode, struct iattr *attr)
                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
                        inode->i_ino, LTIME_S(attr->ia_mtime));
                 oa.o_id = lsm->lsm_object_id;
-                oa.o_valid = OBD_MD_FLID;
+                oa.o_gr = lsm->lsm_object_gr;
+                oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
                 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME);
                 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
@@ -1024,27 +1152,25 @@ int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
         return 0;
 }
 
-void dump_lsm(int level, struct lov_stripe_md *lsm)
-{
-        CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic 0x%08X, "
-               "stripe_size %u, stripe_count %u\n",
-               lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
-               lsm->lsm_stripe_size, lsm->lsm_stripe_count);
-}
-
-void ll_update_inode(struct inode *inode, struct mds_body *body,
-                     struct lov_stripe_md *lsm)
+void ll_update_inode(struct inode *inode, struct lustre_md *md)
 {
         struct ll_inode_info *lli = ll_i2info(inode);
+        struct mds_body *body = md->body;
+        struct lov_stripe_md *lsm = md->lsm;
+        struct mea *mea = md->mea;
+        ENTRY;
 
-        LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
+        LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
+        LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
         if (lsm != NULL) {
+                LASSERT(lsm->lsm_object_gr > 0);
                 if (lli->lli_smd == NULL) {
                         lli->lli_smd = lsm;
                         lli->lli_maxbytes = lsm->lsm_maxbytes;
                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
                 } else {
+                        int i;
                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
                                 CERROR("lsm mismatch for inode %ld\n",
                                        inode->i_ino);
@@ -1054,6 +1180,18 @@ void ll_update_inode(struct inode *inode, struct mds_body *body,
                                 dump_lsm(D_ERROR, lsm);
                                 LBUG();
                         }
+                        /* XXX FIXME -- We should decide on a safer (atomic) and
+                         * more elegant way to update the lsm */
+                        for (i = 0; i < lsm->lsm_stripe_count; i++) {
+                                lli->lli_smd->lsm_oinfo[i].loi_id =
+                                        lsm->lsm_oinfo[i].loi_id;
+                                lli->lli_smd->lsm_oinfo[i].loi_gr =
+                                        lsm->lsm_oinfo[i].loi_gr;
+                                lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
+                                        lsm->lsm_oinfo[i].loi_ost_idx;
+                                lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
+                                        lsm->lsm_oinfo[i].loi_ost_gen;
+                        }
                 }
                 /* bug 2844 - limit i_blksize for broken user-space apps */
                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
@@ -1062,6 +1200,22 @@ void ll_update_inode(struct inode *inode, struct mds_body *body,
                         obd_free_memmd(ll_i2obdexp(inode), &lsm);
         }
 
+        if (mea != NULL) {
+                if (lli->lli_mea == NULL) {
+                        lli->lli_mea = mea;
+                } else {
+                        if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
+                                CERROR("mea mismatch for inode %lu\n",
+                                        inode->i_ino);
+                                LBUG();
+                        }
+                }
+                if (lli->lli_mea != mea)
+                        obd_free_memmd(ll_i2mdcexp(inode),
+                                        (struct lov_stripe_md **) &mea);
+                        
+        }
+
         if (body->valid & OBD_MD_FLID)
                 inode->i_ino = body->ino;
         if (body->valid & OBD_MD_FLATIME)
@@ -1102,12 +1256,16 @@ void ll_update_inode(struct inode *inode, struct mds_body *body,
 
         if (body->valid & OBD_MD_FLSIZE)
                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
+
+        lli->lli_mds = body->mds;
+        inode->i_dev = (kdev_t) body->mds;
+        LASSERT(body->mds < 1000);
 }
 
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0))
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 static struct backing_dev_info ll_backing_dev_info = {
-       .ra_pages       = 0,    /* No readahead */
-       .memory_backed  = 0,    /* Does contribute to dirty memory */
+        .ra_pages       = 0,    /* No readahead */
+        .memory_backed  = 0,    /* Does contribute to dirty memory */
 };
 #endif
 
@@ -1131,7 +1289,9 @@ void ll_read_inode2(struct inode *inode, void *opaque)
         LTIME_S(inode->i_mtime) = 0;
         LTIME_S(inode->i_atime) = 0;
         LTIME_S(inode->i_ctime) = 0;
-        ll_update_inode(inode, md->body, md->lsm);
+
+        inode->i_rdev = 0;
+        ll_update_inode(inode, md);
 
         /* OIDEBUG(inode); */
 
@@ -1154,15 +1314,15 @@ void ll_read_inode2(struct inode *inode, void *opaque)
 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
                 init_special_inode(inode, inode->i_mode,
                                    kdev_t_to_nr(inode->i_rdev));
-               
-               /* initializing backing dev info. */
-               inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
+
+                /* initializing backing dev info. */
+                inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
 #else
                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 #endif
                 lli->ll_save_ifop = inode->i_fop;
-                
-               if (S_ISCHR(inode->i_mode))
+
+                if (S_ISCHR(inode->i_mode))
                         inode->i_fop = &ll_special_chr_inode_fops;
                 else if (S_ISBLK(inode->i_mode))
                         inode->i_fop = &ll_special_blk_inode_fops;
@@ -1170,18 +1330,37 @@ void ll_read_inode2(struct inode *inode, void *opaque)
                         inode->i_fop = &ll_special_fifo_inode_fops;
                 else if (S_ISSOCK(inode->i_mode))
                         inode->i_fop = &ll_special_sock_inode_fops;
-        
-               CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
+
+                CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
                       inode->i_fop);
-                     
+
                 if (lli->ll_save_ifop->owner) {
                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
                               lli->ll_save_ifop->owner);
-               }
+                }
                 EXIT;
         }
 }
 
+void ll_delete_inode(struct inode *inode)
+{
+        int rc;
+        struct ll_fid fid;
+        struct ll_sb_info *sbi = ll_i2sbi(inode);
+        ENTRY;
+        
+        ll_inode2fid(&fid, inode);
+
+        rc = md_delete_object(sbi->ll_mdc_exp, &fid);
+        if (rc) {
+                CERROR("md_delete_object() failed, error %d.\n",
+                       rc);
+        }
+
+        clear_inode(inode);
+        EXIT;
+}
+
 int ll_iocontrol(struct inode *inode, struct file *file,
                  unsigned int cmd, unsigned long arg)
 {
@@ -1197,7 +1376,7 @@ int ll_iocontrol(struct inode *inode, struct file *file,
                 struct mds_body *body;
 
                 ll_inode2fid(&fid, inode);
-                rc = mdc_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
+                rc = md_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
                 if (rc) {
                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
                         RETURN(-abs(rc));
@@ -1235,26 +1414,27 @@ int ll_iocontrol(struct inode *inode, struct file *file,
                 attr.ia_attr_flags = flags;
                 attr.ia_valid |= ATTR_ATTR_FLAG;
 
-                rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
-                                 &attr, NULL, 0, NULL, 0, &req);
+                rc = md_setattr(sbi->ll_mdc_exp, &op_data,
+                                &attr, NULL, 0, NULL, 0, &req);
                 if (rc) {
                         ptlrpc_req_finished(req);
                         if (rc != -EPERM && rc != -EACCES)
-                                CERROR("mdc_setattr fails: rc = %d\n", rc);
+                                CERROR("md_setattr fails: rc = %d\n", rc);
                         obdo_free(oa);
                         RETURN(rc);
                 }
                 ptlrpc_req_finished(req);
 
                 oa->o_id = lsm->lsm_object_id;
+                oa->o_gr = lsm->lsm_object_gr;
                 oa->o_flags = flags;
-                oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
+                oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
 
                 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
                 obdo_free(oa);
                 if (rc) {
                         if (rc != -EPERM && rc != -EACCES)
-                                CERROR("mdc_setattr fails: rc = %d\n", rc);
+                                CERROR("md_setattr fails: rc = %d\n", rc);
                         RETURN(rc);
                 }
 
@@ -1320,25 +1500,29 @@ void ll_umount_begin(struct super_block *sb)
         EXIT;
 }
 
-int ll_prep_inode(struct obd_export *exp, struct inode **inode,
-                  struct ptlrpc_request *req, int offset,struct super_block *sb)
+int ll_prep_inode(struct obd_export *osc_exp, struct obd_export *mdc_exp,
+                  struct inode **inode, struct ptlrpc_request *req,
+                  int offset, struct super_block *sb)
 {
         struct lustre_md md;
         int rc = 0;
 
-        rc = mdc_req2lustre_md(req, offset, exp, &md);
+        rc = mdc_req2lustre_md(mdc_exp, req, offset, osc_exp, &md);
         if (rc)
                 RETURN(rc);
 
         if (*inode) {
-                ll_update_inode(*inode, md.body, md.lsm);
+                ll_update_inode(*inode, &md);
         } else {
                 LASSERT(sb);
                 *inode = ll_iget(sb, md.body->ino, &md);
                 if (*inode == NULL || is_bad_inode(*inode)) {
                         /* free the lsm if we allocated one above */
                         if (md.lsm != NULL)
-                                obd_free_memmd(exp, &md.lsm);
+                                obd_free_memmd(osc_exp, &md.lsm);
+                        if (md.mea != NULL)
+                                obd_free_memmd(mdc_exp,
+                                               (struct lov_stripe_md**)&md.mea);
                         rc = -ENOMEM;
                         CERROR("new_inode -fatal: rc %d\n", rc);
                 }